PROBLEMS; “RIEMANNIAN GEOMETRY”

ANDREAS STRÖMBERGSSON

This is a collection of problems for the course “Riemannian Geometry”,

1MA196, fall 2017, at Uppsala University.
(http://www.math.uu.se/∼astrombe/riemanngeometri2017/rg2017.html)
I remark that the purpose of many of the problems below is mainly to fill in
or explain some (pedantic) facts or details which I felt were appropriate to
mention in my lectures, and which I couldn’t find in Jost’s book. In a later
version, I will probably move the content of these problems into some kind
of appendices in the lecture notes.

Contents
1. Problems 1
2. Solution suggestions 45
References 244

1. Problems

Problem 1. [Manifolds are path-connected] Prove that if M is a topo-

logical manifold (in the sense defined in the course, in particular M is con-
nected) then M is path-connected, i.e. for any two points p, q ∈ M there is
a curve γ : [0, 1] → M with γ(0) = p and γ(1) = q.
Solution:
Problem 2. [A criterion for paracompactness.] p. 45.
(a). Let M be any topological space which is locally Euclidean. Prove that
M is second countable iff M has a countable atlas.
[Pedantically, in Lecture #1 we only defined the notation of an “atlas” when
M is connected and Hausdorff; however the same defnition applies to any
locally Euclidean topological space.]
(b). Let M be a connected Hausdorff space which is locally Euclidean. Prove
that M is paracompact iff M has a countable atlas.
Solution:
p. 46.

1
 2 ANDREAS STRÖMBERGSSON

Problem 3. [Invariance of dimension.] Brouwer’s Theorem on invari-

ance of dimension states: If nonempty open sets U ⊂ Rd1 and V ⊂ Rd1 are
homeomorphic, then d1 = d2 . (Cf., e.g., Hatcher, [7, Thm. 2.26].) Using this
result, prove the following: If M is a connected Hausdorff space for which
every point has an open neighborhood U which is homeomorphic to an open
subset Ω of Rd for some d ∈ Z≥1 (which apriori may depend on U ), then in
fact all the dimensions d appearing must be one and the same.
(Thus, in Def. 1 in Lecture #1, we would not obtain any new objects if we
modify the definition so that the dimension is allowed to depend on U .)
Solution:
p. 47. Problem 4. [Every C ∞ atlas determines a unique C ∞ structure.]
Prove the following statement from Lecture #1 (here made slightly more pre-
cise): “Any C ∞ atlas on a topological manifold M is contained in a unique
C ∞ structure on M , namely the family of all charts which are compatible
with every chart in the given atlas.”
Solution:
p. 47. Problem 5. [Basic property of C ∞ structures.]
(a). Let Br (0) be the open ball in Rd of radius r > 0, centered at the origin.
Prove that there exists an uncountable family H of homeomorphisms of
B1 (0) onto itself, with each h ∈ H satisfying h(x) = x for all x ∈ / B1/2 (0),
−1
such that for any two h1 6= h2 ∈ H, the function h1 ◦ h2 is not C ∞ .
[Hint. One can e.g. take each h to be of the form h(x) = f (kxk)kxk−1 x (for x 6= 0) where
f is a piecewise linear function on (0, 1).]

(b). Let M be a topological manifold. Prove that if M has one C ∞ structure

then there exists an uncountable family F of distinct C ∞ structures on M
such that for any two structures in F, the corresponding C ∞ manifolds are
diffeomorphic.
[Hint. One approach is as follows. Let H be as in part (a) and fix a chart (U, x) on M
with x(U ) = B1 (0) (prove that such a chart exists). Now for each h ∈ H we can define a
homeomorphism ϕh : M → M by letting ϕh be “given by h inside U and the identity map
everywhere else”. Now we get a new C ∞ structure by “composing the given C ∞ structure
with ϕh ”. (These things have to be made precise.)]
Remark: The problem shows why it is much more interesting to ask for the
number of diffeomorphism classes of C ∞ structures on a given topological
manifold M . (Cf. the end of Lecture #1.)
Solution:
p. 49.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 3

Problem 6. [Open submanifolds] Let M be a C ∞ manifold and let U

be an open subset of M .
(a). Prove that U inherits from M a natural structure of a (not necessarily
connected) C ∞ manifold. This C ∞ manifold is called an open submanifold
of M .
(b). Prove that the inclusion map i : U → M is C ∞ .
(c). Let N be another C ∞ manifold and f a map from M to N . Prove that if
f is C ∞ , then so is the map f|U : U → N for every open subset U ⊂ M (with
its inherited C ∞ manifold structure). Prove also the following converse: If
{Uα } is a family of open sets covering M and f|Uα is C ∞ for every α, then
f itself is C ∞ .
Solution:
Problem 7. [Existence of C ∞ functions with desired properties.] p. 54.
Let M be a C ∞ manifold.
(a). Let f be a function from M to R and let U be an open subset of M such
that f|U ∈ C ∞ (U ) and supp(f ) ⊂ U . (Recall that supp(f ) is the closure in
M of the set {p ∈ M : f (p) 6= 0}.) Prove that f ∈ C ∞ (M ).
(b). Let U be an open subset of M and let f : U → R be a C ∞ function
with compact support. Prove that the function
(
e e f (p) if p ∈ U
f : M → R, f (p) =
0 if p ∈
/U
is C ∞ .
(c). Prove that for every p ∈ M and every open subset U ⊂ M with p ∈ U ,
there exists a C ∞ function f : M → [0, 1] which has compact support
contained in U and which satisfies f (p) = 1.
(d). (A strengthening of (c).) Prove that if K is compact and U is open
with K ⊂ U ⊂ M , then there exists a C ∞ function f : M → [0, 1] which
has compact support contained in U , and which satisfies f|K ≡ 1.
[Hint: When M = Rd the claim is a well-known fact of analysis; cf., e.g.,
[10, Thm. 1.4.1]. Thus it remains to reduce to this Euclidean setting...]
(e). (A simple consequence of (d) and (a).) Prove that if K is compact and
U is open with K ⊂ U ⊂ M , and if f : U → R is a C ∞ is a function, then
there exists a C ∞ function f1 : M → R which satisfies f1|K ≡ f|K .
Solution:
p. 55.
 4 ANDREAS STRÖMBERGSSON

Problem 8. [Basic facts about product manifolds]

(a) Prove that if M and N are C ∞ manifolds then the Cartesian product
M × N also naturally carries the structure of a C ∞ manifold. (Cf. [12, p. 4
(Ex. 4)].)
(b) Prove that the projection maps pr1 : M ×N → M and pr2 : M ×N → N
are C ∞ .
(c) Prove that if f : M → N1 and g : M → N2 are C ∞ maps of manifolds
then also the map (f, g) : M → N1 × N2 , defined by
(f, g)(p) := (f (p), g(p)),
is C ∞.
(d) Prove that if f : M1 → N1 and g : M2 → N2 are C ∞ maps of manifolds
then also the map
M1 × M2 → N1 × N2 , (p, q) 7→ (f (p), g(q)),
is C ∞ .
Solution:
p. 57.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 5

Problem 9. [Definition of quotient manifold.] Let M be a topological

manifold, and let Homeo(M ) be the group of all homeomorphisms of M
onto itself (the group operation is composition). Let Γ be a subgroup of
Homeo(M ). We assume that Γ acts freely on M , meaning that for any
γ ∈ Γ and p ∈ M , if γ(p) = p then γ = Id. We also assume that Γ acts
properly discontinuously on M , meaning that for any compact set K ⊂ M ,
the set {γ ∈ Γ : γ(K) ∩ K 6= ∅} is finite. Let us define the relation ∼ on M
by [p ∼ q iff ∃γ ∈ Γ s.t. γ(p) = q].
(a) Prove that ∼ is an equivalence relation.
(b) Let us write [p] for the ∼ equivalence class of a point p ∈ M ; let Γ\M :=
{[p] : p ∈ M } be the set of all equivalence classes, and let π : M → Γ\M ,
π(p) := [p], be the corresponding projection map. Define a topology on Γ\M
by declaring U ⊂ Γ\M to be open iff π −1 (U ) is open in M . Prove that this
indeed is a topology; it is called the quotient topology. Prove also that Γ\M
with this topology is a topological manifold of the same dimension as M .
(c) Now on top of the previous assumptions we assume that M is a C ∞
manifold, and that every γ ∈ Γ is a diffeomorphism of M . (In other words,
Γ ⊂ Diff(M ).) Prove that Γ\M inherits from M a natural C ∞ structure,
and that π is a C ∞ map.
Solution:
Problem 10. [Constructing a C ∞ manifold without requiring from p. 60.
start that it is a topological space.]
(a) Prove that if {(Uα , xα )} is an atlas on a (topological) manifold M , and
V is any subset of M , then V is open iff V ∩ Uα is open in Uα for every α.
(b) Let us define a “(d-dimensional) C ∞ fold” to be a set M together with
a family {(Uα , xα )}α∈A where for each α ∈ A, Uα is a subset of M and xα is
a bijection from Uα onto an open subset of Rd , such that M = ∪α∈A Uα and
for any α, β ∈ A, xα (Uα ∩ Uβ ) is an open subset of xα (Uα ), and the map
xβ ◦ x−1 ∞
α on xα (Uα ∩ Uβ ) is C .

Given a “C ∞ fold” M , let us call a subset V ⊂ M “open” if xα (V ∩ Uα ) is

open in Rd for every α ∈ A. Prove that this defines a topology on M . Prove
also – by giving an example – that this topology is not always Hausdorff.
(c) Prove that a sufficient criterion for the topology defined in part (b) to
be Hausdorff is that for any two points p, q ∈ M there is α ∈ A such that
p, q ∈ Uα . (You may also like to prove the following partial converse: If M is
a C ∞ manifold then for any two points p, q ∈ M there is a C ∞ chart (U, x)
on M such that p, q ∈ U .)
(d) Let M be a “C ∞ fold” and assume that the topology defined above is
Hausdorff, and also connected and paracompact. Prove that then M is a
C ∞ manifold, with {(Uα , xα )} being a C ∞ atlas.
Solution:
p. 64.
 6 ANDREAS STRÖMBERGSSON

Problem 11. [Partition of unity: Some variants.]

(a). Prove the following variation of [12, Lemma 1.1.1]: Let M be a C ∞
manifold and let U = (Uα )α∈A be an open cover of M . Then there exist C ∞
P ϕα : M → [0, 1] (α ∈ A) such that supp ϕα ⊂ Uα for every α ∈ A,
functions
and α∈A ϕα (x) = 1 for all x ∈ M .
(Remark: Note that in the above statement it is not always possible to make
each ϕα have compact support; consider e.g. the case U = {M }; then we
are forced to choose the single ϕ-function to be ϕ ≡ 1.)
(Hint: The above statement can e.g. be deduced as a consequence of [12,
Lemma 1.1.1].)
(b). Prove that both in [12, Lemma 1.1.1], and in the statement of (a) above,
√
we can further require that all functions ϕα are such that also ϕα is C ∞ .
Solution:
p. 67. Problem 12. [Extending a function from a curve to a manifold.]
Let M be a C ∞ manifold, let c : [a, b] → M be a C ∞ curve, let s ∈ (a, b),
and assume ċ(s) 6= 0.
(a) Prove that there is ε > 0 and a (C ∞ ) chart (U, x) for M such that
a < s − ε < s + ε < b and
c(t) ∈ U and x(c(t)) = (t − s, 0, . . . , 0), ∀t ∈ (s − ε, s + ε).

(b) Prove that given any C ∞ function f : [a, b] → R, there is ε > 0 and a
C ∞ function g : M → R such that a < s − ε < s + ε < b and g(c(t)) = f (t)
for all t ∈ (s − ε, s + ε).
Solution:
p. 69. Problem 13. [Details in the definition of tangent space.]
In the following all references are to Lecture #2:
(a). In Definition 3, verify that ∼ is an equivalence relation.
(b). On p. 4 (below Definition 3): Prove that for any fixed chart (U, x) with
p ∈ U , the map u 7→ [(U, x, u)] is indeed a bijection from Rd onto Tp M .
(c). On p. 5: Verify the claim that if M is a ((connected)) open subset of a
finite dimensional vector space V over R, then there is a natural identifica-
tion “Tp M = V ”, for every p ∈ M .
(d). On p. 7: Verify that dfp is well-defined.
(e). On p. 7: Verify the chain rule d(g◦f )p = dgf (p) ◦dfp , when f : M1 → M2
and g : M2 → M3 are C ∞ maps between C ∞ manifolds.
(f). On p. 8–9: Verify the three facts stated here!
Solution:
p. 70.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 7

Problem 14. [Tangent vector of a curve.] Let M be a C ∞ manifold of

dimension d, let c : I → M be a C ∞ curve, and let (U, x) be a chart on M .
For t ∈ I with c(t) ∈ U , we define c1 (t), . . . , cd (t) ∈ R by
x(c(t)) = (c1 (t), . . . , cd (t)).
Then prove that
∂
ċ(t) = ċj (t)
.
∂xj
Also explain how this formula shows that the two definitions of “tangent
vector of a curve” in Lecture #2 (p. 2 and 8) are consistent with each other.
Solution:
Problem 15. [Alternative definition of tangent space.] p. 76.
(a). Let M be a C ∞ manifold and let p ∈ M . By definition, a derivation at
p is an R-linear map D : C ∞ (M ) → R that satisfies the Leibniz identity
D(f g) = D(f ) · g(p) + f (p) · D(g), ∀f, g ∈ C ∞ (M ).
Prove that there is a natural bijection between the set of all derivations at
p and the tangent space Tp (M ).
(b). A vector field X on M is by definition a C ∞ map X : M → T M
satisfying π ◦ X = 1M . (Thus using notation from Lecture #7, a vector field
on M is the same as a section in Γ(T M ).) Also by definition, a derivation of
C ∞ (M ) is an R-linear map D : C ∞ (M ) → C ∞ (M ) which satisfies D(f g) =
D(f )g + f D(g) for all f, g ∈ C ∞ (M ). Prove that there is a natural bijection
between the set of vector fields on M and the set of derivations of C ∞ (M ).
Solution:
Problem 16. [The definition of the tangent bundle T M .] Prove that p. 78.
the construction in Lecture #2, p. 10, leads to a well-defined C ∞ manifold
T M , and that the projection map π : T M → M is C ∞ .
[Hint: Use Problem 10.]
Solution:
Problem 17. [Some facts about df .] p. 79.
Let M, N be C ∞ manifolds and let f : M → N be a C ∞ . Let π : T M → M
and π ′ : T N → N be the standard projection maps.
(a). Prove that df : T M → T N is a C ∞ map and π ′ ◦ df = f ◦ π. (Facts
from Lecture #2.)
(b). Prove that for any C ∞ map ϕ : N → R and any X ∈ T M ,
df (X)(ϕ) = X(ϕ ◦ f ).

(c). Prove that if f : M1 → M2 and g : M2 → M3 are C ∞ maps between C ∞

manifolds, then d(g ◦ f ) = dg ◦ df (equality between maps T M1 → T M3 ).
Solution:
p. 81.
 8 ANDREAS STRÖMBERGSSON

Problem 18. [Riemannian structure on a submanifold of a Rie-

mannian manifold.] Let f : M → N be a C ∞ immersion of C ∞ mani-
folds, and assume that N is equipped with a Riemannian metric.
(a). Prove that then also M gets naturally equipped with a Riemannian
metric, by setting, for any p ∈ M and v, w ∈ Tp M :
hv, wi := hdfp (v), dfp (w)i.
(In particular this means that any immersed submanifold of a Riemannian
manifold gets naturally equipped with a Riemannian metric.)
(b). Prove also that for any piecewise C ∞ curve γ : [a, b] → M we have
L(γ) = L(f ◦ γ) and E(γ) = E(f ◦ γ).
(c). Prove that d(p, q) ≥ d(f (p), f (q)) for all p, q ∈ M , and give an example
where strict inequality holds.
Solution:
p. 81. Problem 19. [Existence of a C ∞ curve between any two points.]
Let M be a C ∞ manifold.
(a). Prove that for any two points p, q ∈ M there exists a piecewise C ∞
curve γ : [0, 1] → M with γ(0) = p and γ(1) = q.
(b). Show that “piecewise C ∞ ” can be sharpened to “C ∞ ” in the previous
statement.
Solution:
p. 83.
Problem 20. [Basic properties of the hyperbolic space H n .]
Go through the discussion in [12, Sec. 5.4], and verify all claims up until the
computation of the curvature using Jacobi fields! In particular:
(a). Verify that if p ∈ H n then Tp H n is orthogonal to p wrt the form h·, ·i,
and the restriction of I to Tp H n is positive definite, so that we obtain a
Riemannian metric on H n .
(b). Prove that O(1, n) 1 is a group, and that O(1, n) has a normal subgroup
of index 2, which we call O+ (1, n), such that each T ∈ O + (1, n) acts on H n
by isometries.
(c). Prove that for any p ∈ H n and v ∈ Tp H n , v 6= 0, there is a transforma-
tion R ∈ O + (1, n) whose set of fixed points in Rn+1 equals the 2-dimensional
plane spanned by p and v. (Hint: The map can be constructed as the “h·, ·i-
reflection” in said plane.)
(d). Conclude by proving the formula which Jost states for a geodesic with
an arbitrary starting condition.
Solution:
p. 85.

1I think the group which Jost calls “O(n, 1)” is more appropriately called “O(1, n)”, in
view of the definition of h·, ·, i.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 9

Problem 21. [The maximal domain for exp, and the geodesic flow.]
The goal of this problem is to prove Theorem 2 in Lecture #4. Note that
the proof basically just consists in squeezing as much information as possible
out of the local ODE existence and uniqueness result (Theorem 1).
(a). For each p ∈ M and v ∈ Tp M there is a uniquely determined open
interval Iv ⊂ R containing 0 such that (i) there exists a geodesic cv : Iv → M
with cv (0) = p, ċv (0) = v and (ii) given any open interval J ⊂ R containing
0 and any geodesic γ : J → M with γ(0) = p, γ̇(0) = v, then J ⊂ Iv and
γ ≡ cv|J .
We call the above curve cv the (unique) maximal geodesic starting at
v ∈ Tp M .
(b). Set W = {(t, v) ∈ R×T M : t ∈ Iv } and define the map θ : W → T M
by θ(t, v) := ċv (t). Prove that for all v ∈ T M and s ∈ Iv we have θ(0, v) = v,
Iθ(s,v) = Iv − s, 2 and θ(θ(s, v), t) = θ(t + s, v) (∀t ∈ Iθ(s,v) ).
(c). There exist an open subset D ⊂ T M and a C ∞ map exp : D → M
such that for each p ∈ M and v ∈ Tp M , Iv := {t ∈ R : tv ∈ D} is an
an open interval containing 0, and the curve t 7→ exp(tv), Iv → M , is the
unique maximal geodesic starting at v. (Note that it is obvious that D and
exp are uniquely determined by the required properties.)
(d). Note that by (c), the set W in part (b) equals
W = {(t, v) ∈ R × T M : tv ∈ D},
and that this is an open subset of R × T M . For t ∈ R, set
Wt := {v ∈ T M : (t, v) ∈ W }
Prove that for each t ∈ R, Wt is an open subset of T M , and the map θ(t, ·)
is a C ∞ diffeomorphism of Wt onto W−t with inverse θ(−t, ·).
(The map θ : W → T M is called the geodesic flow on T M .)
Solution:
p. 90.

2Here we use the natural notation I − s := {x − s : x ∈ I }.

v v
 10 ANDREAS STRÖMBERGSSON

Problem 22. [Varying the center of normal coordinates.]

(a). Prove Theorem 3’ in Lecture #4.
[Hint: One approach is as follows. First prove that the differential of the map
(π, exp) : D → M × M at 0p is non-singular; hence by the Inverse Function
Theorem there is a neighborhood of 0p in which (π, exp) is a diffeomorphism.]
(b). Let r > 0 and let U be an open subset of a Riemannian manifold M , and
assume that for every p ∈ U , Br (0p ) ⊂ D and expp|Br (0p ) is a diffeomorphism
onto an open subset of M ; let us agree to write simply exp−1 p for the inverse
map. Set
V := {(p, expp (v)) : p ∈ U, v ∈ Br (0p )} ⊂ M × M.
Prove that V is an open subset of M × M , and that the map V → T M ,
(p, q) 7→ exp−1 ∞
p (q) is C . (More generally one may let r be a continuous
function of p.)
Solution:
p. 94. Problem 23. [The Riemannian metric wrt polar coordinates.]
(The point of this problem is to go through the details in the proof of Jost’s
[12, Thm. 1.4.5].)
Let M be a Riemannian manifold, p ∈ M , and take r > 0 so that expp
restricted to Br (0) ⊂ Tp (M ) is a diffeomorphism onto an open subset U ⊂
M . Let (U, x) be the corresponding normal coordinates. Let also (V, ϕ) be
a chart on S d−1 , and define the (“polar coordinates”) chart (R+ V, y) on Rd
by
R+ V := {rv : r ∈ R+ , v ∈ V }
(an open cone) and
  x 
(y 1 , . . . , y d ) = kxk, ϕ .
kxk
Set U ′ = x−1 (R+ V ∩ Br (0)); then (U ′ , y ◦ x) is a chart on M . Prove that in
the coordinates defined by this chart, the Riemannian metric satisfies
 
1 0 ··· 0
0 h22 (y) · · · h2d (y)
 
(hij (y)) =  .. .. ..  , ∀y ∈ (0, r) × ϕ(V ).
. . . 
0 hd2 (y) · · · hdd (y)
Solution:
p. 96.
Problem 24. [Any (pw C ∞ ) curve realizing d(p, q) is a geodesic.]
Prove Theorem 2 in Lecture #5: Let M be a Riemannian manifold and
let γ : [a, b] → M be a pw C ∞ curve which is parametrized by arc length.
Assume that L(γ) = d(γ(a), γ(b)). Then γ is a geodesic.
Solution:
p. 98.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 11

Problem 25. [Completeness.]

Let M = Rd with its standard C ∞ manifold structure. Give an example
of a complete Riemannian metric on M , and also an example of one non-
complete Riemannian metric on M .
(Thus, the parenthesis in Jost’s [12, Thm. 1.7.1(i)] is misleading; complete-
ness is not a property of the topology, but depends on the choice of metric.)
Solution:
p. 99.
Problem 26. [A closed embedded submanifold is complete.]
(a). Let N be a complete Riemannian manifold and let M be an embedded
submanifold of N which is closed. Prove that M is complete.
(b). Prove that if we replace “embedded submanifold” by “immersed sub-
manifold” in (a), then the conclusion is no longer valid, in general!
Solution:
Problem 27. [Spheres and distances.] p. 100.
The following properties play a role in the proof of the Hopf-Rinow Theorem.
Let (X, d) be an arbitrary metric space. Recall that for p ∈ X and r > 0 we
write Br (p) for the open ball Br (p) := {q ∈ X : d(p, q) < r}.
(a). Prove that d is a continuous function (X × X → R≥0 ).
(b). Prove that for any p ∈ X, r > 0,
∂Br (p) ⊂ {q ∈ X : d(p, q) = r},
and both these sets are closed. Furthermore if (X, d) is a Riemannian man-
ifold3 then equality holds: ∂Br (p) = {q ∈ X : d(p, q) = r}.
(c). Continue to assume that (X, d) is a Riemannian manifold. Let p, q ∈ X,
r > 0, and assume d(p, q) > r. Assume that p0 is a point on ∂Br (p) where
d(·, q)∂Br (p) is minimal. Prove that d(p, q) = d(p, p0 ) + d(p0 , q).
Solution:
Problem 28. [Consequences of Br (0p ) ⊂ Dp .] p. 101.
Let M be a Riemannian manifold, let p ∈ M and R > 0, and assume
BR (0p ) ⊂ Dp . Prove that then for every point q ∈ BR (p), the distance
d(p, q) is realized by a geodesic, and hence BR (p) = expp (BR (0p )).
Solution:
Problem 29. [Existence of geodesics in homotopy classes.] p. 102.
Prove that Theorem 1 in Lecture #5 remains true for any complete (instead
of compact) Riemannian manifold.
Solution:
p. 103.

3By this we mean: X is a Riemannian manifold and d is the metric on X which comes
from the Riemannian structure.
 12 ANDREAS STRÖMBERGSSON

Problem 30. [Injectivity radius on a surface of revolution.]

(The following problem is a slight variation of [12, Ch. 1, Problem 11].)
Consider the surface of revolution
S := {(x, ex cos α, ex sin α) : x, α ∈ R}.
(a). Prove that S is a closed differentiable submanifold of R3 (cf. the notes
to Lecture #2).
(b). Equip S with the Riemannian metric induced by the standard Riemann-
ian metric on R3 (cf. Problem 18; note that S is complete by Problem 26).
Fix x0 ∈ R and let p0 = (x0 , ex0 , 0) ∈ S. Prove that the injectivity radius of
p0 satisfies i(p0 ) ≤ πex0 .
Solution:
p. 104.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 13

Problem 31. [The fundamental group of the n-punctured plane.]

Let p1 , . . . , pn be n distinct points in R2 . Compute π1 (R2 \ {p1 , . . . , pn }).
Solution:
Problem 32. [Covering space; lifting of structure.] p. 107.
e together
A covering space of a topological space X is a topological space X
with a continuous map π : X e → X satisfying the following condition: Each
point x ∈ X has an open neighborhood U in X such that π −1 (U ) is a union
e each of which is mapped homeomorphically onto
of disjoint open sets in X,
U by π.
(a). Let M be a topological manifold of dimension d and let π : M f→M
be a covering space of M which is connected and second countable. Prove
that then also M f is a topological manifold of dimension d. (In fact the
assumption that M f is second countable is redundant; see the remark at the
end of the solution.)
(b). Let M be a C ∞ manifold of dimension d and let π : M f → M be a
covering space of M which is connected and second countable. Prove that
then M f has a unique structure as a C ∞ manifold such that π is C ∞ and
each point p ∈ M has an open neighborhood U in M such that π −1 (U ) is a
union of disjoint open sets in M f, each of which is mapped diffeomorphically
onto U by π.
(c). Let M be a Riemannian manifold of dimension d and let π : M f→M
be a covering space of M which is connected and second countable. Prove
that then M f has a unique structure as a Riemannian manifold such that π
∞
is C and each point p ∈ M has an open neighborhood U in M such that
f, each of which is mapped (C ∞ )
π −1 (U ) is a union of disjoint open sets in M
isometrically onto U by π.
(d). Prove that for any topological manifold M and any subgroup Γ <
Homeo(M ) acting freely and properly discontinuously on M , if Γ\M and
π : M → Γ\M are as in Problem 9, then π : M → Γ\M is a covering space
of Γ\M .
Solution:
p. 108.
 14 ANDREAS STRÖMBERGSSON

Problem 33. [Trivial vector bundle; basis of sections.]

Let (E, π, M ) be a vector bundle of rank n and let U be an open subset of
M . Prove that the following statements are equivalent:
(a) E|U is trivial;
(b) there is some ϕ such that (U, ϕ) is a bundle chart for E;
(c) there is a basis of sections in ΓE|U , i.e. sections s1 , . . . , sn ∈ ΓE|U such
that s1 (p), . . . , sn (p) is a basis of Ep for every p ∈ U .
Solution:
p. 113. Problem 34. [Trivial vector bundle; one more (very!) basic fact.]
Let (E, π, M ) be a vector bundle of rank n, let U be an open subset of M ,
and let s1 , . . . , sn ∈ ΓE|U be a basis of sections in ΓE|U (cf. Problem 33(c)).
Prove that for every section s ∈ ΓE|U there exists a unique n-tuple of func-
tions α1 , . . . , αn ∈ C ∞ (U ) such that s = αj sj .
Solution:
p. 115. Problem 35. [About sections: restrictions and surjectivity to fibers.]
Let (E, π, M ) be a vector bundle over a C ∞ manifold M .
(a) Prove that for every open set U ⊂ M , every section s ∈ Γ(E|U ), and
every point p ∈ U , there exists a section s′ ∈ Γ(E) such that s′|V = s|V for
some open set V ⊂ U containing p.
(b) Prove that for every point p ∈ M there exist an open set V ⊂ M with
p ∈ V and sections b1 , . . . , bn ∈ Γ(E) such that b1|V , . . . , bn|V form a basis
of sections of E|V .
(c) Prove that for every p ∈ M and every v ∈ Ep , there is some s ∈ ΓE such
that s(p) = v.
Solution:
p. 115. Problem 36. [Defining a vector bundle without requiring from
start that it is a manifold.] Let M be a C ∞ manifold, let E be a
set and let π : E → M be a surjective map. Assume that for every p ∈ M ,
Ep := π −1 (p) carries the structure of an n-dimensional real vector space.
Also let {(Uα , ϕα )}α∈A be a family such that for each α ∈ A, Uα is an open
subset of M and ϕα is a bijection of π −1 (Uα ) onto Uα × Rn such that for
every p ∈ Uα , the map (ϕα )p := (ϕα )|Ep is a linear isomorphism of Ep onto
{p} × Rn . Assume that M = ∪α∈A Uα , and that for any α, β ∈ A, the map
ϕβ ◦ϕ−1 n ∞
α from (Uα ∩Uβ )×R to itself is C . Prove that then E has a unique
C ∞ manifold structure such that (E, π, M ) is a vector bundle of rank n, and
(Uα , ϕα ) is a bundle chart for every α ∈ A.
Solution:
p. 116.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 15

Problem 37. [Classifying all vector bundles over S 1 .]

(a). Prove that the Möbius bundle over S 1 (cf. Lecture #7, p. 2) is not
trivial.
(b). Classify all vector bundles over S 1 up to isomorphism.
Solution:
Problem 38. [Finite cover of trivializing sets.] p. 119.
Let M be a C ∞ manifold of dimension d and let E be a vector bundle over
M . Prove that then there exists an open cover U1 , . . . , Ud+1 of M such that
E|Uj is trivial for each j = 1, . . . , d + 1.
[Remark: We will need to make use of this result a few times later. Then
what will matter for us is the fact that U1 , . . . , Ud+1 is a finite open cover;
the exact number of open sets used will not be of importance.
[Hint: You may make use of the following theorem from dimension theory:
Let M be a topological manifold of dimension d. Then every open cover U
of M has a refinement W such that for any d + 2 tuple of distinct open sets
W1 , . . . , Wd+2 ∈ W, one has W1 ∩ · · · ∩ Wd+2 = ∅.
Cf. [9, Thm. V.8 and p. 25 (Ex. III.4)].]
Solution:
Problem 39. [Definitions of E1 ⊗ E2 , Hom(E1 , E2 ), E ∗ .] p. 124.
Let (E1 , π1 , M ) and (E2 , π2 , M ) be vector bundles over a C ∞ manifold M .
(a) Verify that E1 ⊗ E2 , as defined in Lecture #7, is indeed a vector bundle
over M .
(b) Similarly define the vector bundle Hom(E1 , E2 ).
(c) Similarly define the vector bundle E1∗ .
Hint for parts (a)-(c): See Problem 36!
Solution:
Problem 40. [Γ(Hom(E1 , E2 )) = bundle homomorphisms E1 → E2 .] p. 124.
Let (E1 , π1 , M ) and (E2 , π2 , M ) be vector bundles over a C ∞ manifold M .
Prove that there is a natural bijection between Γ(Hom(E1 , E2 )) and the set
of bundle homomorphisms E1 → E2 .
[Remarks: (1) From now on we will often identify these two sets, i.e. a
bundle homomorphism f : E1 → E2 is automatically viewed as an element in
Γ(Hom(E1 , E2 )), and vice versa. (2) See also Problem 43 below for another
important property of Γ(Hom(E1 , E2 )).]
Solution:
p. 128.
 16 ANDREAS STRÖMBERGSSON

Problem 41. [Definition of subbundle.] Let (E, π, M ) be a vector bun-

dle of rank n. Recall that in Lecture #7 we defined a subbundle of E to
be a subset E ′ ⊂ E such that for every p ∈ M there exists a bundle chart
(U, ϕ) for E such that p ∈ U and
(1) ϕ(E ′ ∩ π −1 (U )) = U × Rm
for some m ≤ n, where we view Rm ⊂ Rn through
Rm = {(x1 , . . . , xn ) ∈ Rn : xm+1 = · · · = xn = 0}.
In this situation, prove that
(a) m is independent of p and (U, ϕ);
(b) (E ′ , π|E ′ , M ) is a vector bundle of rank m, and for every bundle chart
(U, ϕ) satisfying (1), (U, ϕ|E ′ ∩π−1 (U ) ) is a bundle chart for E ′ .
[Hint: cf. Problem 36.]
(c) E ′ is a differentiable submanifold of E.
Solution:
p. 130. Problem 42. [Basic facts about the pulled back bundle f ∗ E.]
Let f : M → N be a C ∞ map and let (E, π, N ) be a vector bundle.
(a). Prove that the pulled back bundle, f ∗ E, defined in Lecture #7 as a
subset of M × E with extra structure, really is a vector bundle over M .
[Hint: cf. Problem 36.]
(b). Prove that f ∗ E is a differentiable submanifold of M × E.
Solution:
p. 133. Problem 43. [Properties of the functor Γ.]
Let (E1 , π1 , M ) and (E2 , π2 , M ) be vector bundles over a C ∞ manifold M .
Prove that there exist natural identifications (isomorphisms of C ∞ (M )-
modules) as follows:
(a). Γ(E1 ⊕ E2 ) = Γ(E1 ) ⊕ Γ(E2 ).
(b). Γ(E1∗ ) = (ΓE1 )∗ .
(c). Γ(Hom(E1 , E2 )) = Hom(ΓE1 , ΓE2 ).
(d). Γ(E1 ⊗ E2 ) = Γ(E1 ) ⊗ Γ(E2 ).
[Remarks: As we stressed in the lecture, any space of sections ΓE is a
C ∞ M -module, and when applying dual, “Hom” or “⊗” to spaces of sec-
tions, it should always be viewed as operations on C ∞ M -modules! Thus
(ΓE1 )∗ is the C ∞ M -module of C ∞ M -linear maps from ΓE1 to C ∞ M ,
“Hom(ΓE1 , ΓE2 )” is the C ∞ M -modules of C ∞ M -linear maps from ΓE1
to ΓE2 , and “Γ(E1 ) ⊗ Γ(E2 )” is the C ∞ M -module which in a more precise
notation would be denoted Γ(E1 ) ⊗C ∞ (M ) Γ(E2 ).]
Solution:
p. 135.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 17

Problem 44. [Sections along a function; Γf E.]

Let f : M → N be a C ∞ map and let (E, π, N ) be a vector bundle.
(a). A section of E along f (or “a lift of f to E”) is a C ∞ map σ : M → E
such that π ◦ σ = f . The set of sections of E along f is denoted Γf E. Prove
that Γf E has a structure as a C ∞ M -module, and that there is a natural
isomorphism of C ∞ M -modules Γf ∗ E ∼ = Γf E.
[Remark: From now on we will often use the above isomorphism to identify
Γf ∗ E and Γf E. As will be seen, to view a section s ∈ Γf ∗ E as an element
in Γf E simply means considering pr2 ◦s : M → E, i.e. “forgetting the first
component of s, which anyway contains redundant information about the
base point”. On the other hand, one can not in any reasonable way define
f ∗ E directly as a subset of E, unless f is injective; indeed, for any two
points p 6= q in M with f (p) = f (q) we want (f ∗ E)p and (f ∗ E)q to be two
disjoint copies of Ef (p) .]
(b). Note that for any s ∈ ΓE we have s ◦ f ∈ Γf E = Γf ∗ E; we call s ◦ f
the (f -)pullback of s. Prove that if V is an open set in N and U is an open
set in M with f (U ) ⊂ V , and if s1 , . . . , sn is a basis of sections in ΓE|V ,
then s1 ◦ f, . . . , sn ◦ f is a basis of sections in Γ(f ∗ E)|U .
(c). Prove any section of f ∗ E can be expressed as a function-linear combina-
of E. (In other words: Any σ ∈ Γf ∗ E can be
tion of f -pullbacks of sections P
expressed as a finite sum σ = m ∞
j=1 αj · (sj ◦ f ) where α1 , . . . , αm ∈ C (M )
and s1 , . . . , sm ∈ ΓE.) [Hint: Problems 11 and 38 may be useful.]
Solution:
Problem 45. [Interpreting Γ(Hom(E1 , f ∗ E2 )).] p. 141.
Let f : M → N be a C ∞ map and let (E1 , π1 , M ) and (E2 , π2 , N ) be vector
bundles. We say that a map h : E1 → E2 is a bundle homomorphism
along f if h is C ∞ , π2 ◦ h = f ◦ π1 , and for each x ∈ M the fiber map
hx := h|E1,x : E1,x → E2,f (x) is linear.
(a). Prove that there is a natural bijection between Γ(Hom(E1 , f ∗ E2 )) and
the set of bundle homomorphisms E1 → E2 along f .
(b). Explain how the result in (a) can be seen to generalize both Problem 40
and Problem 44(a).
Solution:
Problem 46. [Extending a section from a curve to the whole space.] p. 144.
Let (E, π, M ) be a vector bundle, let c : (a, b) → M be a C ∞ curve, let
s ∈ Γc E (cf. Problem 44(a)), let t0 ∈ (a, b), and assume ċ(t0 ) 6= 0. Prove
that there exist ε > 0 and a section s1 ∈ ΓE such that a < t0 −ε < t0 +ε < b
and s1 (c(t)) = s(t) for all t ∈ (t0 − ε, t0 + ε).
(Hint: cf. Problems 12 and 35.)
Solution:
p. 146.
 18 ANDREAS STRÖMBERGSSON

Problem 47. [Lie product of vector fields.]

Let M be a C ∞ manifold.
(a). For any vector fields X, Y on M , prove that there exists a unique vector
field Z on M satisfying Z(f ) = X(Y (f )) − Y (X(f )) for all f ∈ C ∞ (M ). By
definition, this vector field Z is denoted “[X, Y ]”, and called the Lie product
of X and Y .
[Hint: Use Problem 15(b).]
(b). Prove that our definition in part a is equivalent with Jost, [12, Def.
2.2.4].
(c). Prove the Jacobi identity:
[[X, Y ], Z] + [[Y, Z], X] + [[Z, X], Y ] = 0, ∀X, Y, Z ∈ Γ(T M ).

(d). Prove that for any X, Y ∈ Γ(T M ) and f ∈ C ∞ (M ),

[X, f Y ] = (Xf ) · Y + f · [X, Y ]
and
[f X, Y ] = −(Y f ) · X + f · [X, Y ].
Solution:
p. 147.

Problem 48. [Basic properties of the exterior derivative.]

Let M be a C ∞ manifold.
(a) Following Jost, [12, Def. 2.1.15], we define d : Ωr (M ) → Ωr+1 (M ) by
the requirement
P that for any ω ∈ Ωr (M ) and any C ∞
Pchart (U, x) on M , if
ω|U = I ωI dx (with ωI ∈ C ∞ (U )) then (dω)|U = I dωI ∧ dxI . 4 Prove
I

that this indeed gives a well-defined, R-linear map d : Ωr (M ) → Ωr+1 (M ).

(In other words, explain in detail what happens in [12, Cor. 2.1.2].)
(b) Prove that if f : M → N is a C ∞ map then d(f ∗ (ω)) = f ∗ (dω) for all
ω ∈ Ωr (N ). (In other words, provide more details for [12, Lemma 2.1.3].)
(c) Prove that for any ω ∈ Ωr (M ) and X0 , . . . , Xr ∈ Γ(T M ),
r
X

[dω](X0 , . . . , Xr ) = (−1)j Xj ω(X0 , . . . , X̂j , . . . , Xr )
j=0
X
+ (−1)j+k ω([Xj , Xk ], X0 , . . . , X̂j , . . . , X̂k , . . . , Xr ).
0≤j<k≤r

[Explanation of notation: “X0 , . . . , X̂j , . . . , Xr ” denotes “X0 , X1 , X2 , . . . , Xr but with the

term Xj removed”. Similarly “[Xj , Xk ], X0 , . . . , X̂j , . . . , X̂k , . . . , Xr ” denotes
“[Xj , Xk ], X0 , X1 , X2 , . . . , Xr but with both Xj and Xk removed”. Also, the sum in
the second line runs through all pairs hj, ki ∈ Z2 satisfying 0 ≤ j < k ≤ r.]
Solution:
p. 148.
4Note that dω = ∂ωI dxj ; hence our definition indeed agrees with [12, Def. 2.1.15].
I j ∂x
 PROBLEMS; “RIEMANNIAN GEOMETRY” 19

Problem 49. [Wedge product of vector valued forms]

(a). Let E1 and E2 be vector bundles over a C ∞ manifold M . We define
the wedge product ∧ : Ωr (E1 ) × Ωs (E2 ) → Ωr+s (E1 ⊗ E2 ), for any r, s ≥ 0,
to be the unique C ∞ (M )-bilinear map satisfying
(µ1 ⊗ ω1 ) ∧ (µ2 ⊗ ω2 ) = (µ1 ⊗ µ2 ) ⊗ (ω1 ∧ ω2 ),
∀µ1 ∈ Γ(E1 ), ω1 ∈ Ωr (M ), µ2 ∈ Γ(E2 ), ω2 ∈ Ωs (M ).
Prove that this indeed makes ∧ a well-defined C ∞ (M )-bilinear map . Note
also that in the special case E1 = E2 = M × R, this gives back the standard
wedge product Ωr (M ) × Ωs (M ) → Ωr+s (M ).
(b). [Associativity and “commutativity”.] Let E1 , E2 , E3 be vector bundles
over M and let r, s, t ≥ 0. Prove that
s1 ∧ (s2 ∧ s3 ) = (s1 ∧ s2 ) ∧ s3 , ∀s1 ∈ Ωr (E1 ), s2 ∈ Ωs (E2 ), s3 ∈ Ωt (E3 ).
e where E
(Here both expressions lie in Ωr+s+t (E), e = E1 ⊗ E2 ⊗ E3 = E1 ⊗
(E2 ⊗ E3 ) = (E1 ⊗ E2 ) ⊗ E3 .) Prove also that
s1 ∧ s2 = (−1)rs · J(s2 ∧ s1 ), ∀s1 ∈ Ωr (E1 ), s2 ∈ Ωs (E2 ),
where J is the isomorphism of C ∞ (M )-modules
∼
J : Ωr+s (E2 ⊗ E1 ) −
→ Ωr+s (E1 ⊗ E2 )
which maps J(µ2 ⊗ µ1 ⊗ ω) = µ1 ⊗ µ2 ⊗ ω for all µ1 ∈ ΓE1 , µ2 ∈ ΓE2 ,
ω ∈ Ωr+s (M ).
(c). [“vector-wedge-product”; extending commutativity.] Let E1 , E2 , E e be
vector bundles over M and assume given a “multiplication rule” from E1 , E2
e i.e. a C ∞ (M )-linear map m : Γ(E1 ⊗ E2 ) → Γ(E).
to E, e By extending with
the identity map on Ω (M ), this defines for each r ≥ 0 a C ∞ (M )-linear map
r
e which we also call m. Let m′ be the multiplication
Ωr (E1 ⊗ E2 ) → Ωr (E),
′
rule m : Γ(E2 ⊗ E1 ) → Γ(E) e defined by m′ (s2 ⊗ s1 ) = m(s1 ⊗ s2 ) for all
s1 ∈ ΓE1 , s2 ∈ ΓE2 , and call m′ also the corresponding map Ωr (E2 ⊗ E1 ) →
e Prove that
Ωr (E).
(2) m(s1 ∧ s2 ) = (−1)rs m′ (s2 ∧ s1 ), ∀s1 ∈ Ωr (E1 ), s2 ∈ Ωs (E2 ).

[Comments: In many cases we will write simply “m(s1 , s2 )” or “s1 ∧ s2 ”

to denote the combined vector-wedge-product m(s1 ∧ s2 )! For example this
appears in [12, (4.1.26)]; “A ∧ A”, wherein E1 = E2 = E e = End E and m
is – of course – composition. Other examples appear in the computation of
DF a bit further down on [12, p. 139]; e.g. “[A, F ]”; here again E1 = E2 =
Ee = End E but m is Lie bracket. Another example, in a slightly generalized
setting, is in [12, p. 154]; “Pe(F, . . . , F )”. A main example where the relation
(2) applies is when E := E1 = E2 = E e is a commutative (weak) algebra
bundle over M (with m being the multiplication rule). In this case m′ = m,
 20 ANDREAS STRÖMBERGSSON

and so (2) shows how the commutativity of E extends to Ω(E). On the

other hand, a natural example with E1 6= E2 is when E1 = E (an arbitrary
vector bundle over M ), E2 = E ∗ and Ee = M × R, with the multiplication
rule m (as well as m ) being the standard contraction from Γ(E ⊗ E ∗ ) (or
′

Γ(E ∗ ⊗ E)) to C ∞ (M ).]

(d). [extension of associativity.] Let E1 , E2 , E3 , E12 , E23 , E123 be vector
bundles over M and assume given multiplication rules
Γ(E1 ⊗ E2 ) → Γ(E12 ); Γ(E12 ⊗ E3 ) → Γ(E123 );
Γ(E2 ⊗ E3 ) → Γ(E23 ); Γ(E1 ⊗ E23 ) → Γ(E123 ).
For each of these, we denote the image of s ⊗ s′ simply by “s · s′ ”. Assume
that these multiplication rules satisfy the associativity relation
(s1 · s2 ) · s3 = s1 · (s2 · s3 ), ∀s1 ∈ ΓE1 , s2 ∈ ΓE2 , s3 ∈ ΓE3 .
In line with the above comments, let us write s1 ∧ s2 ∈ Ωr+s (E12 ) for the
combined vector-wedge-product of any s1 ∈ Ωr (E1 ) and s2 ∈ Ωs (E2 ); and
similarly for the other three product rules. Then prove that
(s1 ∧ s2 ) ∧ s3 = s1 ∧ (s2 ∧ s3 ), ∀s1 ∈ Ωr (E1 ), s2 ∈ Ωs (E2 ), s3 ∈ Ωt (E3 ).

[Comments: A main example of the above situation is of course when E :=

E1 = E2 = E3 = E12 = E23 = E123 is an associative (weak) algebra bundle
over M . A general example where E1 , E2 , E3 may be distinct vector bundles
is when Ej := Hom(Fj+1 , Fj ) for j = 1, 2, 3, where F1 , F2 , F3 , F4 are four
arbitrary vector bundles over M , and all multiplication rules are composition
(thus E12 = Hom(F3 , F1 ), etc., and the associativity relation holds).]
Solution:
p. 155.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 21

Problem 50. [Wedge-product of matrix valued forms made ex-

plicit.]
Let E1 , E2 , E3 be vector bundles over M ; then we have a standard multipli-
cation rule “◦” (composition of homomorphisms)
Γ(Hom(E2 , E3 )) × Γ(Hom(E1 , E2 )) → Γ(Hom(E1 , E3 )).
Let us write “◦” also for the corresponding vector-wedge-product
Ωr (Hom(E2 , E3 )) × Ωs (Hom(E1 , E2 )) → Ωr+s (Hom(E1 , E3 ))
(cf. Problem 49(c)). Let U be an open subset of M such that there exist
bases of sections
α1 , . . . , αn1 ∈ ΓE1|U and β1 , . . . , βn2 ∈ ΓE2|U and γ1 , . . . , γn3 ∈ ΓE3|U
(here nℓ = rank Eℓ ). Let
α1∗ , . . . , αn1 ∗ ∈ ΓE1|U
∗
and β 1∗ , . . . , β n2 ∗ ∈ ΓE2|U
∗

and γ 1∗ , . . . , γ n3 ∗ ∈ ΓE3|U
∗

be the dual bases.

Then for each µ ∈ Ωr (Hom(E2 , E3 )) there exist unique r-forms µkj ∈
Ωr (U ) such that µ|U = β j∗ ⊗γk ⊗µkj , and similarly for each η ∈ Ωs (Hom(E1 , E2 ))
there exist unique s-forms ηjk ∈ Ωs (U ) such that η|U = αj∗ ⊗ βk ⊗ ηjk . Prove
that in terms of this representation,
(µ ◦ η)|U = αi∗ ⊗ γk ⊗ (µkℓ ∧ ηiℓ ).

[Comment: Note that “µ|U = β j∗ ⊗ γk ⊗ µkj ” means that if we use the given
bases to identify E2|U with U × Rn2 and E3|U with U × Rn3 , then µ|U is
represented by the matrix
 1 
µ1 · · · µ1n2
 .. 
(µkj ) =  ... . 
µn1 3 · · · µnn32
(wherein each entry is an r-form). Similarly “η|U = αj∗ ⊗βk ⊗ηjk ” means that
η|U is represented by the matrix (ηjk ) and “(µ ◦ η)|U = αi∗ ⊗ γk ⊗ (µkℓ ∧ ηiℓ )”
means that (µ ◦ η)|U is represented by the matrix (µkℓ ∧ ηiℓ )k,i . Hence when
r = s = 0, the formula gives back the usual formula for matrix product;
(µ ◦ η)|U = (µkℓ · ηiℓ )k,i , as it should.]
Solution:
p. 158.
 22 ANDREAS STRÖMBERGSSON

Problem 51. [Wedge product; alternative definition]

(a). Let (E, π, M ) be a vector bundle. Prove that there is a natural identi-
fication of Ωr (E) with the space of alternating C ∞ (M )-multilinear maps
Γ(T M )(r) = Γ(T M ) × · · · × Γ(T M ) −−→ ΓE.
| {z }
r times

(b). Let E1 and E2 be vector bundles over M . Prove that using the iden-
tification in part (a), the wedge product s1 ∧ s2 (cf. Problem 49(a)) of any
s1 ∈ Ωr (E1 ) and s2 ∈ Ωs (E2 ) is given by
(s1 ∧ s2 )(X1 , . . . , Xr+s )
1 X
= sgn(σ) s1 (Xσ(1) , . . . , Xσ(r) ) ⊗ s2 (Xσ(r+1) , . . . , Xσ(r+s) ),
r!s!
σ∈Sr+s
∀X1 , . . . , Xr+s ∈ Γ(T M ),
where Sr+s is the group of all permutations of {1, . . . , r + s}. Prove also a
e
similar formula for the product “s1 · s2 ∈ Ωr+s (E)”, in the case when there
e (cf. Problem 49(c)).
is given a multiplication rule from E1 , E2 to E
Solution:
p. 159.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 23

Problem 52. [Restricting a connection to open sets.]

Complete the proof of Lemma 1 in Lecture #9; that is, prove the following:
Let (E, π, M ) be a vector bundle. For (a) and (b), let D be a connection on
E and let U ⊂ M be open.
(a) ∀s1 , s2 ∈ ΓE : s1|U = s2|U ⇒ (Ds1 )|U = (Ds2 )|U .
(b) There is a unique connection “D|U ” on E|U satisfying (Ds)|U = D|U (s|U )
for all s ∈ Γ(E).
(c) Let (Uα )α∈A be an open covering of M , and for each α ∈ A let Dα be a
connection on E|Uα . Assume that for any two α, β ∈ A, if V := Uα ∩ Uβ 6= ∅
then (Dα )|V = (Dβ )|V . Then there exists a unique connection D on E
satisfying D|Uα = Dα for every α ∈ A.
Solution:
Problem 53. . p. 162.
[Dv s depends only on the values of s along a curve with ċ(0) = v.]
Let (E, π, M ) be a vector bundle and let D be a connection on E. Let
v ∈ T M and let s1 , s2 ∈ ΓE. Assume that there exists a C ∞ curve c :
(−ε, ε) → M such that ċ(0) = v and s1 (c(t)) = s2 (c(t)) for all t ∈ (−ε, ε).
Prove that then Dv s1 = Dv s2 .
Solution:
Problem 54. [The connection “d” (for given local coordinates).] p. 164.

Let (U, ϕ) be a bundle chart of a vector bundle (E, π, M ), let s1 , . . . , sn ∈

Γ(E|U ) be the corresponding basis of sections and define the map d : Γ(E|U ) →
Γ((E ⊗ T ∗ M )|U ) by d(ak sk ) = sk ⊗ dak for any a1 , . . . , an ∈ C ∞ U . Prove
that this is a connection on E|U . (Cf. p. 6 in Lecture #9.)
Solution:
Problem 55. [Restriction of a connection to a subbundle.] p. 165.
Let D be a connection on a vector bundle (E, π, M ) and let E ′ be a vector
subbundle of E. Then also E ′ ⊗ T ∗ M is a vector subbundle of E ⊗ T ∗ M .
Assume that Ds ∈ Γ(E ′ ⊗ T ∗ M ) for all s ∈ ΓE ′ . Prove that then the
restriction of D to ΓE ′ is a connection on E ′ . Also give an example to show
that the given condition is not always satisfied.
Solution:
p. 165.
 24 ANDREAS STRÖMBERGSSON

Problem 56. [Alternative definition of the dual of a connection.]

Let D be a connection on a vector bundle (E, π, M ), and let D ∗ be the
dual connection on E ∗ . (Cf. Lecture #10.) Recall that given any C ∞ -curve
γ : (−ε, ε) → M we have a linear isomorphism
Pγ(0)→γ(h) : Eγ(0) → Eγ(h)
γ

for each h ∈ (−ε, ε); let us write P∗γ,h for the dual of that map; this is a
∗
linear isomorphism Eγ(h) ∗ . Prove that for any µ ∈ ΓE ∗ ,
→ Eγ(0)

∗
P∗γ,h (µ(γ(h))) − µ(γ(0))
Dγ̇(0) (µ) = lim in Eγ(0) .
h→0 h

Problem 57. [Defining the pullback of a connection.]

(a). Let f : M → N be a C ∞ map and let D be a connection on a vector
bundle (E, π, N ). Prove that there exists a unique connection f ∗ D on f ∗ E
such that for any s ∈ ΓE,
(f ∗ D)(s ◦ f ) = Ddf (·) (s) ∈ Γ(Hom(T M, f ∗ E)) = Γ(f ∗ E ⊗ T ∗ M ).
[Explanation: “Ddf (·) (s)” stands for the map
T M → E, [v 7→ Ddf (v) (s)],
which is a bundle homomorphism along f , and hence can be viewed as an
element of Γ(Hom(T M, f ∗ E)) by Problem 45.]
(b). (Comparing with Jost’s definition of f ∗ D, [12, p. 205].) Prove that f ∗ D
in part (a) is the unique connection on f ∗ E such that the following holds: For
any s ∈ Γf ∗ E = Γf E and any C ∞ curve c : (−ε, ε) → M , if s1 ∈ ΓE satisfies
s1 (f (c(t))) = s(c(t)) for all t ∈ (−ε, ε), then (f ∗ D)ċ(0) (s) = Ddf (ċ(0)) (s1 ).
(c). Let c : (−ε, ε) → M be any C ∞ curve such that f ◦ c is a constant point
q ∈ N . Prove that for any s ∈ Γf ∗ E,
d 
(f ∗ D)ċ(0) (s) = (s ◦ c)(t) ∈ Eq ,
dt |t=0
d
where dt (s ◦ c)(t) ∈ Ts(c(t)) (Eq ) = Eq stands for the tangent vector of the
curve s ◦ c in Eq .
(Comments: In the situation in (c), if s is not constant along c, the formula in (b) cannot
be used directly to compute (f ∗ D)ċ(0) (s), since there cannot exist any s1 ∈ ΓE satisfying
s1 (f (c(t))) = s(c(t)), ∀t ∈ (−ε, ε). Note also that the tangent vector of the curve s ◦ c,
d
dt
(s ◦ c)(t), is always a well-defined vector in Ts(c(t)) (E); however in the situation in (c)
d
we can view s ◦ c as a curve in the fiber Eq ; hence dt (s ◦ c)(t) ∈ Ts(c(t)) (Eq ), and this last
tangent space can naturally be identified with Eq by Problem 13(c).)
Solution:
p. 165.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 25

Problem 58. [The tensor product of two connections.]

Prove Proposition 2 in Lecture #10, i.e. the following: Let E1 , E2 be vector
bundles over M with connections D1 , D2 , respectively. Then there is a
unique connection D on E1 ⊗ E2 such that
D(µ ⊗ ν) = (D1 µ) ⊗ ν + µ ⊗ (D2 ν), ∀µ ∈ ΓE1 , ν ∈ ΓE2 .
Solution:
p. 171.
Problem 59. [A Leibniz rule for general connections.]
Let E1 , E2 , E3 be vector bundles over M , each equipped with a connection
“D”. Let us write “D” also for the corresponding connections on E1∗ and
E1 ⊗ E2 and Hom(E1 , E2 ) = E1∗ ⊗ E2 , etc.
(a). Given any
α ∈ Γ(E1 ⊗ E2 ) and β ∈ Γ(E1∗ ⊗ E3 ),
let us write “(α, β)” for the section in Γ(E2 ⊗ E3 ) obtained by contracting
the E1 -part of α against the E1∗ -part of β. Prove that then
D(α, β) = (Dα, β) + (α, Dβ) in Ω1 (E2 ⊗ E3 ).
(Here “(Dα, β)” is again defined by contracting the E1 -part of Dα against the E1∗ -part
of β, and similarly for “(α, Dβ)”; note that these can be viewed as vector-wedge-products
à la Problem 49(c), from Ωr (E1 ⊗ E2 ) × Ωs (E1∗ ⊗ E3 ) to Ωr+s (E2 ⊗ E3 ), coming from the
given product (·, ·) from Γ(E1 ⊗ E2 ) × Γ(E1∗ ⊗ E3 ) to Γ(E2 ⊗ E3 ).)

(b). Prove that for any α ∈ Γ(Hom(E2 , E3 )) and β ∈ Γ(Hom(E1 , E2 )),

D(α ◦ β) = (Dα) ◦ β + α ◦ (Dβ) in Ω1 (Hom(E1 , E3 )).

(c). Prove that for any α ∈ Γ(Hom(E1 , E2 )) and β ∈ ΓE1 ,

D(α(β)) = (Dα)(β) + α(Dβ) in Ω1 (E2 ).
Solution:
p. 174.
 26 ANDREAS STRÖMBERGSSON

Problem 60. [The exterior covariant derivative.]

Let D be a connection on a vector bundle (E, π, M ).
(a). Prove Proposition 4 in Lecture #10, i.e. the following: Then for any
p ≥ 0 there exists a unique R-linear map D : Ωp (E) → Ωp+1 (E) satisfying
D(µ ⊗ ω) = (Dµ) ∧ ω + µ ⊗ dω, ∀µ ∈ ΓE, ω ∈ Ωp (M ).

(b). Let (U, ϕ) be a fixed bundle chart for E; let d be the corresponding
naive connection on E|U and set A = D − d ∈ Ω1 (End E|U ), as usual. Prove
that for any µ ∈ Ωp (E|U ),
Dµ = dµ + A ∧ µ in Ωp+1 (E|U ),
where d is the naive exterior covariant derivative Ωp (E|U ) → Ωp+1 (E|U )
coming from the given bundle chart, and A ∧ µ is the image of A and µ un-
der the combined vector-wedge-product (cf. Problem 49(c)) Ω1 (End E|U ) ×
Ωp (E|U ) → Ωp+1 (E|U ) coming from the standard contraction (“evaluation”)
Γ(End E|U ) × ΓE|U → ΓE|U .
e be vector bundles over M , each equipped with a con-
(c). Let E1 , E2 , E
nection “D”. Assume given a multiplication rule from E1 , E2 to E,e i.e. a
∞
C (M )-linear map Γ(E1 ⊗ E2 ) → Γ(E). e We write s1 · s2 ∈ ΓEe for the
product of s1 ∈ ΓE1 , s2 ∈ ΓE2 , and we write “∧” for the corresponding
vector-wedge-product as in Problem 49(c). Assume that the connections
respect the multiplication rule, in the sense that
D(s1 · s2 ) = (Ds1 ) ∧ s2 + s1 ∧ (Ds2 ), ∀s1 ∈ ΓE1 , s2 ∈ ΓE2 .
Prove that then for any r, s ≥ 0,
D(µ1 ∧ µ2 ) = (Dµ1 ) ∧ µ2 + (−1)r µ1 ∧ Dµ2 , ∀µ1 ∈ Ωr (E1 ), µ2 ∈ Ωs (E2 ),
where “∧” is the vector-wedge-product as in Problem 49(c).
(d). Addendum to (c): Let m be the multiplication rule in (c), i.e. a C ∞ (M )-
linear map Γ(E1 ⊗ E2 ) → Γ(E). e By Problem 43(c), the multiplication rule
can be identified with a section m ∈ Γ(Hom(E1 ⊗ E2 , E)). e Prove that the
e
given connections on E1 , E2 , E respect the multiplication rule iff
Dm = 0.
e induced by the given con-
(Here D is the connection on Hom(E1 ⊗ E2 , E)
e
nections on E1 , E2 , E.)
Solution:
p. 175.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 27

Problem 61. [Explicit formula for exterior covariant derivative (us-

ing Lie product of vector fields).]
Let D be a connection on a vector bundle (E, π, M ); let r ≥ 0, and write “D”
also for the corresponding exterior covariant derivative Ωr (E) → Ωr+1 (E).
Prove that for any s ∈ Ωr (E) and X0 , . . . , Xr ∈ Γ(T M ),
r
X

[Ds](X0 , . . . , Xr ) = (−1)j DXj s(X0 , . . . , X̂j , . . . , Xr )
j=0
X
+ (−1)j+k s([Xj , Xk ], X0 , . . . , X̂j , . . . , X̂k , . . . , Xr ).
0≤j<k≤r

[Here “[Ds](X0 , . . . , Xr )” stands for the contraction of the form part of Ds ∈

Ωr+1 (E) against X0 , . . . , Xr ; and similarly for all “s(· · · )” in the right hand
side. For the rest of the notation, cf. Problem 48(c).]
Solution:
p. 180.
Problem 62. [One more explicit formula for exterior covariant de-
rivative.]
Let D be a connection on a vector bundle (E, π, M ); let r ≥ 0, and write “D”
r r+1
Vr Ω (E) → Ω (E).
also for the corresponding exterior covariant derivative
Recall from the solution of Problem 51 that E ⊗ M is in a natural way
a subbundle of E ⊗ Tr0 (M ); accordingly for any section s ∈ Ωr (E) let us
write “ se ” for s viewed as a section in Γ(E ⊗ Tr0 (M )). Furthermore let ∇
be an arbitrary torsion free connection on T M , and let us write “[ D ∇ ]” for
0
the connection on E ⊗ Tr (M ) induced by D and ∇. Then prove that for
any s ∈ Ωr (E) and X0 , . . . , Xr ∈ Γ(T M ),
r
X  
[Ds](X0 , . . . , Xr ) = (−1)j [ D
∇] se (X0 , . . . , X̂j , . . . , Xr ).
Xj
j=0
Solution:
p. 181.
Problem 63. [Basic facts about AdE (for E with a bundle metric).]
Let (E, π, M ) be a vector bundle equipped with a bundle metric. Recall that
AdE (as a subset of End E) was defined in Lecture #11, p. 11.
(a). Prove that AdE is a vector subbundle of End E.
(b). Prove that if D is any metric connection on E, and if we write D
also for the corresponding connection on End E, then Ds ∈ Ω1 (AdE) for
all s ∈ Γ(AdE) ⊂ Γ(End E). (Hence by Problem 55, the connection D on
End E descends to give a connection on AdE.)
Solution:
p. 183.
 28 ANDREAS STRÖMBERGSSON
Vr
Problem 64. [Some facts about (V ) for V a vector space]
V
(See Sec. 7.2 for the definition and some basic properties of r (V ).)
Let V be a finite dimensional vector space over R and let r ≥ 1.
(a). ForVany v1 , . . . , vr , w1 , . . . , wr ∈ V , the following statement about vec-
tors in r (V ):
h i
v1 ∧ · · · ∧ vr = c · w1 ∧ · · · ∧ wr for some c ∈ R, and v1 ∧ · · · ∧ vr 6= 0
holds if and only if v1 , . . . , vr are linearly independent and v1 , . . . , vr and
w1 , . . . , wr span the same r-dimensional linear subspace of V .
(b). Prove that if V is equipped with aVscalar product h·, ·i then there is
a corresponding scalar product h·, ·i on r (V ) which has the following two
properties:
V
(i) If e1 , . . . , en is any ON-basis for V then (eI ) is an ON-basis for r (V ),
where I runs through all r-tuples I = (i1 , . . . , ir ) ∈ {1, . . . , n}r with i1 <
· · · < ir , and eI := ei1 ∧ · · · ∧ eir .
(ii) For any v1 , . . . , vr , w1 , . . . , wr ∈ V ,
hv1 ∧ · · · ∧ vr , w1 ∧ · · · ∧ wr i = det(hvi , βj i)i,j .
V
Prove also that this scalar product on r (V ) is uniquely determined by the
requirement that either (i) or (ii) hold.
(c). With notation as in (b), for any v1 , . . . , vr ∈ V , the “length”
p
kv1 ∧ · · · ∧ vr k := hv1 ∧ · · · ∧ vr , v1 ∧ · · · ∧ vr i
equals the volume of the r-dimensional parallelotope spanned by v1 , . . . , vr
(wrt. the natural r-dimensional volume measure induced by the the scalar
product h·, ·i on V ).
Solution:
p. 186.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 29

Problem 65. [Equivalent criteria for a manifold being orientable.]

(a). Let M be a C ∞ manifold of dimension d. Prove that the following
three statements are equivalent:
(i) M possesses an oriented C ∞ atlas, i.e. an atlas such that all chart tran-
sition maps have everywhere positive Jacobian determinant.
(ii) There exists an atlas of bundle charts for the vector bundle (T M, π, M )
which makes it an oriented vector bundle (⇔ makes it have structure group
GL+ d (R); cf. Lecture #12, Def. 4).
(iii) There exists a nowhere vanishing d-form ω ∈ Ωd (M ).
[Comment: M is said to be orientable if one and hence all of the conditions
(i)–(iii) hold. Note that (i) is the definition given in Jost, [12, Def. 1.1.3].]
(b). Prove that T M is always an orientable manifold, regardless of whether
M is orientable or not.
Solution:
p. 188.
Problem 66. [Total covariant derivative of a tensor field.]
(a). Let M be a C ∞ manifold of dimension d and let ∇ be a connection
on T M . Write also ∇ for the corresponding connection on Tsr M , for any
r, s ≥ 0. Let A be a tensor field in Γ(T11 M ), and let Aji be the coefficients
of A wrt a given C ∞ chart (U, x) on M . (Thus: Aji ∈ C ∞ (U ) for all
i, j ∈ {1, . . . , d} and A|U = Aji · dxi ⊗ ∂x∂ j .) Also for each k ∈ {1, . . . , d} let
Aji;k be the coefficients of ∇ ∂ A. Prove that
∂xk

∂ j
Aji;k = A − Γℓki · Ajℓ + Γjkℓ · Aℓi in U.
∂xk i
(b). Generalize the above to the case of a tensor field A ∈ Γ(Tsr M ), for any
r, s ≥ 0.
Solution:
p. 191.
 30 ANDREAS STRÖMBERGSSON

Problem 67. [Some explicit computations in vector bundles over S d .]

Consider the sphere
S d = {x = (x1 , . . . , xd+1 ) ∈ Rd+1 : (x1 )2 + · · · + (xd+1 )2 = 1}
with its standard C ∞ manifold structure (cf. [12, p. 3, Ex. 1]), and let (U, y)
be the chart on S d given by
 x1 xd 
U = S d \ {(0, . . . , 0, −1)}; y(x) = , . . . , .
1 + xd+1 1 + xd+1

∂
(a). For d = 2, prove that the vector field y 1 on U can not be extended
∂y 1
to a (C ∞ ) vector field on S 2 .
(b). For d = 3, prove that the vector field
∂ 1 ∂ 1 − (y 1 )2 − (y 2 )2 + (y 3 )2 ∂
(y 1 y 3 − y 2 ) + (y 2 3
y + y ) +
∂y 1 ∂y 2 2 ∂y 3
on U can be extended to a (C ∞ ) vector field on S 3 .
(c). Prove that for d = 2, the section
1

1 2 2 2 4
dy 1 ⊗ dy 1 + dy 2 ⊗ dy 2
(1 + (y ) + (y ) )
of T20 (U ) has a unique extension to a (C ∞ ) section of T20 (S 2 ). Prove also
that the above section defines a Riemannian metric on U , but its extension
to T20 (S 2 ) does not define a Riemannian metric on S 2 .
(d). Let d = 2 and set V = S 2 \ {(0, 0, 1)} (recall U = S 2 \ {(0, 0, −1)}). Fix
an integer m, and define the function µ : U ∩ V → GL2 (R) by
 
cos(m α(y)) − sin(m α(y))
µ(y) = , where α(y) := arg(y 1 + iy 2 ).
sin(m α(y)) cos(m α(y))
(Thus α(y) is the argument of the complex number y 1 + iy 2 ; note that this
number is non-zero for all points in U ∩ V .) Prove that there exists a vector
bundle E of rank 2 over S 2 which has bundle charts (U, φ) and (V, ψ) with
transition function µ, that is, so that ψp = µ(p) · φp : Ep → R2 for every
p∈U ∩V. (Hint: Problem 36 may be useful.)
Solution:
p. 192.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 31

Problem 68. [More about the pullback of a connection.]

Let f : M → N be a C ∞ map; let (E, π, N ) be a vector bundle of rank n,
and let D be a connection on E.
(a). Let (U, x) be a chart for N and let s1 , . . . , sn be a basis of sections in
Γ(E|U ). Also let (V, y) be a chart for M with V ⊂ f −1 (U ), and recall that
then s1 ◦ f, . . . , sn ◦ f is a basis of sections in Γ((f ∗ E)|V ); cf. Problem 44(b).
Set d = dim N and d′ = dim M . Let Γkij ∈ C ∞ (U ) be the Christoffel
symbols of D with respect to the bases s1 , . . . , sn and ∂x∂ 1 , . . . , ∂x∂ d , and let
e k ∈ C ∞ (V ) be the Christoffel symbols of f ∗ D with respect to the bases
Γ ij
e k in terms of Γk !
s1 ◦ f, . . . , sn ◦ f and ∂y∂ 1 , . . . , ∂y∂d′ . Give a formula for Γ ij ij

(b). Let D be a connection on E. For clarity in this problem let us write

dD : Ωr (E) → Ωr+1 (E) (instead of just “D”) for the exterior covariant
∗
derivative corresponding to D; then also write df D : Ωr (f ∗ E) → Ωr+1 (f ∗ E)
for the exterior covariant derivative corresponding to the connection f ∗ D on
f ∗ E (cf. Problem 57). Prove that for every r ≥ 0 there is a unique R-linear
map f ∗ : Ωr (E) → Ωr (f ∗ E) satisfying
f ∗ (µ ⊗ ω) = (µ ◦ f ) ⊗ f ∗ (ω) for all µ ∈ ΓE and ω ∈ Ωr (N ).
Next prove that for any s ∈ Ωr (E),
∗D
(df )(f ∗ (s)) = f ∗ (dD s).

[Comment: In particular for r = 0 we have f ∗ (µ) = µ ◦ f for all µ ∈ ΓE,

∗
and dD = D : Ω0 (E) → Ω1 (E) and df D = f ∗ D : Ω0 (f ∗ E) → Ω1 (f ∗ E). In
this case the above formula says:
(f ∗ D)(f ∗ (s)) = f ∗ (D(s)),
which can be viewed as a (nicer!) reformulation of the formula in Prob-
lem 57(a)!]
Solution:
p. 199.
 32 ANDREAS STRÖMBERGSSON

Problem 69. [Basic about sectional curvature.]

In Lecture #15, Def. 1, prove that K(X ∧ Y ) indeed only depends on the
2-dimensional plane spanned by X, Y in Tp M .
Problem 70. [Scaling a Riemannian metric.]
Let M be a C ∞ manifold equipped with a Riemannian metric h·, ·i, and
let c > 0 be a constant. Let [·, ·] be the Riemannian metric on M defined
by [·, ·] := c h·, ·i (that is, [v, s] = c hv, wi for any p ∈ M , v, w ∈ Tp M ).
Prove that the two Riemannian manifolds (M, h·, ·i) and (M, [·, ·]) have the
same Levi-Civita connection and curvature tensor, but that the sectional
curvatures K on (M, h·, ·i) and K e on (M, [·, ·]) are related by
e
K(X ∧ Y ) = c−1 K(X ∧ Y )
for any p ∈ M and any linearly independent X, Y ∈ Tp M .
Solution:
p. 203.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 33

Problem 71. [Ricci curvature as average of sectional curvatures.]

(a). Let M be a Riemannian manifold of dimension d. Prove that there is
a constant Cd > 0 which only depends on d such that for any p ∈ M and
X ∈ Tp M with kXk = 1, the Ricci curvature in direction X, Ric(X, X),
equals Cd times the uniform average of the sectional curvatures of all planes
in Tp M containing X. Also determine the constant Cd explicitly.
(b). Similarly, prove that there is a constant Cd′ > 0 such that the scalar
curvature at any point p ∈ M equals Cd′ times the uniform average of the
Ricci curvatures of all unit vectors in Tp M .
Solution:
Problem 72. [Explicit formula for the curvature tensor in terms of p. 204.
sectional curvature.] Let V be a vector space over R and let
R: V ×V ×V ×V →R
be a multilinear form having the same symmetries as the curvature tensor
field Rm (cf. Lemma 1 in Lecture #14); that is, for all X, Y, Z, W ∈ V :
R(X, Y, Z, W ) = −R(Y, X, Z, W ) = −R(X, Y, W, Z) = R(Z, W, X, Y )
and
R(X, Y, Z, W ) + R(Y, Z, X, W ) + R(Z, X, Y, W ) = 0.
Set
K(X, Y ) := R(X, Y, Y, X).
Find an explicit formula expressing R(X, Y, Z, W ) in terms of the function
K. Note that this gives a proof of a corrected version of [12, Lemma 4.3.3].
Solution:
[Hint: One way to obtain this is by appropriately working through the p. 206.
steps in proof of the uniqueness Lemma 1 in lecture #15.]
Problem 73. [Analogue of Schur’s Theorem for Ricci curvature.]
Prove the second part of Theorem 1 in Lecture #15 (=[12, Thm. 4.3.2]);
“if dim M ≥ 3 and the Ricci curvature is constant at each point then M is
Einstein”.
Solution:
p. 208.
 34 ANDREAS STRÖMBERGSSON

Problem 74. [The pullback of a metric connection is metric.]

Let F : H → M be a C ∞ map of manifolds, let (E, π, M ) be a vector bundle
equipped with a bundle metric h·, ·i, and let D be a metric connection on E
preserving the bundle metric. Prove that h·, ·i in a natural way gives rise to
a bundle metric on F ∗ (E) (which we may also denote h·, ·i), and that the
pullbacked connection F ∗ (D) is metric with respect to this bundle metric.
(Comment: This fact is used in the proof of Lemma 1 in Lecture #16, and
also in Jost, [12, p. 206, lines -5 to -4].)

Problem 75. [Pullback and torsion.]

Let F : H → M be a C ∞ map of manifolds and let ∇ be a connection on
T M.
(a). Prove that the map
S : Γ(T H) × Γ(T H) → Γ(F ∗ (T M ));
S(X, Y ) = (F ∗ ∇)X (dF ◦ Y ) − (F ∗ ∇)Y (dF ◦ X) − dF ◦ [X, Y ],
is well-defined and C ∞ (H)-bilinear. Conclude that S can be identified with
a section in Γ(T ∗ H ⊗ T ∗ H ⊗ F ∗ (T M )).
(b). Prove that if ∇ is torsion free then S = 0.
(c). Use the above to give a detailed justification of the identity
∇ ∂ ċ = ∇ ∂ c′
∂s ∂t

appearing in the proof of Lemma 1 in Lecture #16 (and also in Jost, [12, p.
206 (line -4 to -3)]).

Problem 76. [Pullback of curvature.]

(a). Let f : M → N be a C ∞ map, and let D be a connection on a
vector bundle (E, π, N ), with curvature tensor R ∈ Ω2 (End E). Also let
e ∈ Ω2 (End(f ∗ E)) be the curvature tensor of the connection f ∗ D on f ∗ E.
R
Prove that for any p ∈ M and X, Y ∈ Tp (M ),
e
R(X, Y ) = R(df (X), df (Y )) in End(f ∗ E)p = End(Ef (p) ).

(b). Use the above to give a detailed justification of the identity

∂c ∂c  ∂c ∂c  ∂c
∇∂ ∇∂ (t, s) = ∇ ∂ ∇ ∂ (t, s) + R ,
∂s ∂t ∂s ∂t ∂s ∂s ∂t ∂s ∂s
appearing in the proof of Theorem 1 in Lecture #16 (and also in Jost, [12,
p. 208 (lines 4,7,8)]).
Solution:
p. 209.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 35

Problem 77. [Interpretation of curvature in terms of parallel trans-

port around a ’square’]
Let D be a connection on a vector bundle (E, π, M ) and let F = FD be
its curvature. Given p ∈ M , X, Y ∈ Tp M and v ∈ Ep , prove the following
formula for F (X, Y )(v): Let η > 0 and let f be a C ∞ function from
(−η, η)2 = {(x, y) ∈ R2 : −η < x, y < η}
∂ ∂
to M satisfying f (0, 0) = p, df(0,0) ( ∂x ) = X and df(0,0) ( ∂y ) = Y . For
0 < ε < η, let Pε : Ep → Ep denote parallel transport around the (“square”)
curve

f (t, 0)
 if 0 ≤ t ≤ ε

f (ε, t − ε) if ε ≤ t ≤ 2ε
c(t) =

 f (3ε − t, ε) if 2ε ≤ t ≤ 3ε


f (0, 4ε − t) if 3ε ≤ t ≤ 4ε.
Then
1

F (X, Y )(v) = − lim P ε (v) − v .
ε→0+ ε2

Problem 78. [Constant curvature metrics in normal coordinates.]

Let M be a Riemannian manifold with constant sectional curvature ρ. Let
p ∈ M and let (U, x) be normal coordinates with center p, and let (gij (x))
represent the Riemannian metric with respect to (U, x). Prove that for any
x ∈ x(U ) \ {0}:


 xi xj sin2 (ρ1/2 kxk)  xi xj 

 kxk2 + δij − if ρ > 0

 ρ kxk2 kxk2


gij (x) = δij if ρ = 0



 2  

 xi xj sinh (|ρ|1/2 kxk) xi xj

 + δij − if ρ < 0.
kxk 2 |ρ| kxk2 kxk2
(Verify also that the above expression extends to a C ∞ function on all of
x(U ), as it should.)
Solution:
p. 211.
 36 ANDREAS STRÖMBERGSSON

Problem 79. [Some relations for (gij ) in normal coordinates.]

Let M be a Riemannian manifold, let p ⊂ M , and let (U, x) be a chart on M
which gives normal coordinates centered at p. Let the Riemannian metric
be represented by (gij (x)) with respect to (U, x). Prove that for every i,
gii,ii (0) = 0
and for any i 6= j,
gii,jj (0) = gjj,ii (0) = −2gij,ij (0).

∂
Here gij,kℓ(x) := gij (x).
∂xk ∂xℓ
[Some hints/suggestions: For symmetry reasons we may assume i, j ∈ {1, 2}
and then it suffices to study gij (x) for x = (x1 , x2 , 0, . . . , 0). One can show
that Jost’s [12, Thm. 1.4.5] (⇔ Problem 23) implies thatpat any point
x = (x1 , x2 , 0, . . . , 0) the vector x1 ∂x∂ 1 + x2 ∂x∂ 2 has length x21 + x22 , and
is orthogonal to the vector −x2 ∂x∂ 1 + x1 ∂x∂ 2 . Now investigate carefully what
these facts imply for the functions gij (x1 , x2 , 0, . . . , 0) for i, j ∈ {1, 2}.]
Solution:
p. 214.
Problem 80. [A formula for sectional curvature.]
Let M be a Riemannian manifold, let p ⊂ M , and let Π be a plane in
Tp M (viz., a 2-dimensional linear subspace of Tp M ). Let Dr ⊂ Tp M be
the open disc of radius r in the plane Π, centered at 0. For r sufficiently
small, we know (by Theorem 3 in Lecture #4) that expp (Dr ) is an embedded
2-dimensional submanifold of M ; call its area Ar . Prove that
πr 2 − Ar
K(Π) = lim 12 .
r→0+ πr 4
(The Riemannian metric on expp (Dr ) is the one induced from M ; cf. Prob-
lem 18. Also the “area” of expp (Dr ) is the same as its “volume”; cf. p. 1 in
Lecture #12.) [Hint: The results from Problem 79 may be useful.]
Solution:
p. 216.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 37

Problem 81. [On a surface of revolution: geodesics, parallel trans-

port and sectional curvature.]
Let f be a C ∞ function from R to R>0 , and consider a surface of revolution
S := {(x, f (x) cos α, f (x) sin α) : x, α ∈ R}.
We take it as known5 that S is a closed differentiable submanifold of R3 , and
that for any real interval J = (a, b) with b < a + 2π the inverse of the map
(x, α) 7→ (x, f (x) cos α, f (x) sin α) from R × J to S is a chart on S. Equip S
with the Riemannian metric induced from the standard Riemannian metric
on R3 .
(a). Make explicit the ode describing an arbitrary geodesic on S, ∇γ̇ γ̇ = 0
(cf. p. 9 in Lecture #13), in the (x, α) coordinates. Your answer should be
of the form

ẍ + * ẋẋ + * ẋα̇ + * α̇α̇ = 0
α̈ + * ẋẋ + * ẋα̇ + * α̇α̇ = 0,

with each “ * ” being an explicit expression in x, α, f . Prove also from

this equation that f (x)2 · α̇ remains constant along any geodesic. Finally,
describe all geodesics which have x ≡ constant or α ≡ constant.
(b). Given x ∈ R, consider the closed curve c(t) = (x, f (x) cos t, f (x) sin t),
t ∈ [0, 2π], in S. Describe explicitly the parallel transport of an arbitrary
tangent vector v ∈ Tc(0) S along c.
(c). Compute the sectional curvature of S at an arbitrary point
(x, f (x) cos α, f (x) sin α).
(In particular, where is this sectional curvature positive/negative? Also, as
a consistency
√ check, verify that you get back the known answer for the case
f (x) = r 2 − x2 , |x| < r.)
Solution:
p. 218.

5(cf. Problem 30(a))

 38 ANDREAS STRÖMBERGSSON

Problem 82. [A formula involving ∇2 of a 1-form.]

Let M be a Riemannian manifold and let ∇ be the Levi-Civita connection
on T M . By the standard definitions of dual and tensor product connections
(cf. Propositions 1,2 in Lecture #10) ∇ gives rise to a connection on any
tensor bundle
Tsr (M ) = T
| M ⊗ ·{z
· · ⊗ T M} ⊗ T ∗
· · ⊗ T ∗ M},
| M ⊗ ·{z
r times s times
which we r
also call ∇. This ∇ is a map from ΓTs (M ) to
(3) Ω1 (Tsr (M )) = Γ(Tsr (M ) ⊗ T ∗ M ) = Γ(Ts+1r
(M )).
Prove that for any η ∈ Γ(T10 (M )), the tensor field
∇2 η := ∇(∇η) in Γ(T30 (M ))
satisfies
 
(∇2 η)(X, Y, Z) − (∇2 η)(X, Z, Y ) = η R(Y, Z)X ,
for all vector fields X, Y, Z ∈ Γ(T M ).
(Remark: We stress that the “new” T ∗ M -factor is put last in (3); thus for
any F ∈ Γ(Tsr (M )) and any ω 1 , . . . , ω r ∈ Γ(T ∗ M ), Y1 , . . . , Ys ∈ Γ(T M ),
X ∈ Γ(T M ),



(∇F ) ω 1 , . . . , ω r , Y1 , . . . , Ys , X = (∇X F ) ω 1 , . . . , ω r , Y1 , . . . , Ys .
Note also that the connections ∇ : Γ(Tsr (M )) → Γ(Ts+1
r (M )) considered

here should not be confused with the exterior covariant derivative defined
in Proposition 4 in Lecture #10.)
(Hint: The formula can be proved either by expressing everything in local
coordinates using Christoffel symbols, or by working through the definitions
expressing all “∇” appearing in terms of the original Levi-Civita connection
∇ : Γ(T M ) → Ω1 (T M ).)
Solution:
p. 223.
Problem 83. [Basic fact on existence of variations of a curve.]
Let M be a C ∞ manifold, let c : [0, 1] → M be a C ∞ curve, and let Y be a
vector field along c.
(a). Prove that there exists a variation of c with c′ = Y , and that if Y (0) =
0 = Y (1) then this variation can be taken to be proper.
(b). Prove that if γ0 , γ1 : (−ε′ , ε′ ) → M are C ∞ curves with γ0 (0) = c(0),
γ̇0 (0) = Y (0), γ1 (0) = c(1), γ̇1 (0) = Y (1), then there exists a variation of c
with c′ = Y such that c(0, s) = γ0 (s) and c(1, s) = γ1 (s) for all small s.
Solution:
p. 225.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 39

Problem 84. [On proper variations through geodesics.]

Prove that if c(t, s) is a proper variation of a geodesic c through geodesics
(viz., cs is a geodesic for every s), then E(s) = E(cs ) and L(s) = L(cs ) are
constant functions of s.
(Comment: This means that Jost’s sentence in [12, p. 216 (lines 13–14)] is
somewhat misleading; namely the length is always constant on the whole
family, for a proper variation through geodesics.)
Solution:
Problem 85. [Around Cor. 3 in Lecture #17 ≈ Jost’s Cor. 5.2.4.] p. 226.

(a). Let M = S d with its standard Riemannian metric, and let p ∈ M .

Give an example of a piecewise smooth curve γ : [0, 1] → Tp M such that
L(expp ◦γ) = kγ(1)k but γ is not a reparametrization of the curve t 7→ t·γ(1)
(t ∈ [0, 1]).
(Comment: This shows that the last statement in Jost’s [12, Cor. 5.2.4], i.e.
the criterion for when equality holds, is incorrect.)
(b). Use “Gauss Lemma” (= Cor. 2 in Lecture #17 = Jost’s [12, Cor. 5.2.3])
to derive the following strengthening of a result from Problem 23: Let M be
a Riemannian manifold, let p ∈ M , and let Dp = Tp M ∩ D be the maximal
domain of expp (cf. Problem 21). Let (W, y) be a C ∞ chart on Tp M with
W ⊂ Dp , which we assume is “polar coordinates” in the sense that
y 1 (w) = kwk, ∀w ∈ W,
and
(y 2 (cw), . . . , y d (cw)) = (y 2 (w), . . . , y d (w)) whenever c > 0, w ∈ W , cw ∈ W .
Prove that at every point ỹ ∈ y(W ), the matrix representing the symmetric
bilinear form


(v, w) 7→ d(expp ◦y −1 )ỹ (v), d(expp ◦y −1 )ỹ (w) , v, w ∈ Rd ,
is of the form
 
1 0 ··· 0
0 h22 (ỹ) · · · h2d (ỹ)
 
(hij (ỹ)) =  .. .. ..  .
. . . 
0 hd2 (ỹ) · · · hdd (ỹ)

(Comment: As explained in Lecture #17, the above fact can be used to prove
Cor. 3 in Lecture #17, which is [12, Cor. 5.2.4] with a modified criterion for
equality.)
(c). Prove the following alternative criterion for equality in #17, Cor. 3:
“If equality holds, and there does not exist a point conjugate to c(0) along
c, then γ must be a reparametrization of the curve t 7→ tv (t ∈ [0, 1]).”
Solution:
p. 226.
 40 ANDREAS STRÖMBERGSSON

Problem 86. [Remark 2 in Lecture #18]

Let c : [a, b] → M be a geodesic and let t0 6= t1 ∈ [a, b]. Prove that c(t0 ) and
c(t1 ) are conjugate along c iff the differential


d expc(t0 ) (t1 −t0 )·ċ(t0 ) : Tc(t0 ) M → Tc(t1 ) M
is singular.
Solution:
p. 229. Problem 87. [On the metric space CM of C ∞ curves on M .]
Let M be a Riemannian manifold. Introduce the space CM with its metric
d as on p. 3 in Lecture #18.
(a). Prove that d is well-defined, and is indeed a metric on CM .
(b). Prove that the metric space (CM , d) is not complete.
(c). Prove that neither E nor L are continuous on (CM , d); in fact for every
c ∈ CM and δ > 0, both E and L are unbounded on the open ball Bδ (c).
(d). As a small consolation, prove that both E and L are lower semicontin-
uous on (CM , d).
Problem 88. [Approximating a non-C ∞ vector field along a curve.]
Let c : [a, b] → M be a geodesic and let Y be a “pw C ∞ vector field along c”,
i.e. Y is a continuous function Y : [a, b] → T M such that Y (t) ∈ Tc(t) (M )
for all t ∈ [a, b], and such that there exist a finite number of ’break-points’
a = t0 < t1 < · · · < tm = b such that the restricted function Y|[tj−1 ,tj ] is C ∞
for each j = 1, 2, . . . , m. Prove that then for every ε > 0 there exists some
C ∞ vector field Z along c such that
m−1
[
Z(t) = Y (t) ∀t ∈ [a, b] \ (tj − ε, tj + ε)
j=1
and
 
I(Z, Z) − I(Y, Y ) < ε.
(Of
Pm course here “I(Y, Y )” is well-defined, for example it can be defined as
j=1 I(Y[tj−1 ,tj ] , Y[tj−1 ,tj ] ).)
Solution:
p. 230.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 41

Problem 89. [Equivalence of definitions of injectivity radius.]

Let M be a Riemannian manifold and let p ∈ M . Let r > 0 be such that
expp is defined and injective on the open ball Br (0) in Tp (M ). Prove that
then expp|Br (0) is a diffeomorphism of Br (0) onto an open subset of M .
(Comment: This proves that the injectivity radius of p can be defined either
as the supremum of all r > 0 for which expp is defined and injective on
Br (0) ⊂ Tp (M ), as in Jost [12, Def. 1.4.6], or as the supremum of all r > 0
for which expp|Br (0) is a diffeomorphism.)
Solution:
p. 231.
 42 ANDREAS STRÖMBERGSSON

Problem 90. [Vanishing derivatives up to order k.]

Let M be a C ∞ manifold and let f ∈ C ∞ (M ), p ∈ M and k ∈ Z≥0 . We say
that f has vanishing derivatives up to order k at p if for some chart (U, x) for
M with p ∈ U , any 1 ≤ r ≤ k and any j1 , . . . , jr ∈ {1, . . . , d} (d = dim M ),
∂ ∂
j
· · · jr f = 0 at p.
∂x 1 ∂x
Prove that when this holds, it follows that every chart (U, x) with p ∈ U
has the same property.
Solution:
p. 232.
Problem 91. [Geodesics and conjugate points on a perturbed sphere.]
Let S d be unit sphere equipped with its standard Riemannian metric, which
we denote by h·, ·i. For any function f ∈ C ∞ (M ) which is everywhere
positive, we write Sfd for S d equipped with the Riemannian metric
[v, w] := f (p) · hv, wi, ∀p ∈ S d , v, w ∈ Tp S d .
Fix a geodesic c : [0, π] → S d parametrized by arc length (thus the endpoints
c(0) and c(π) are antipodal points). For k ∈ Z≥0 , let Fk be the family of all
positive functions f ∈ C ∞ (M ) such that for every point p along c we have
f (p) = 1 and f has vanishing derivatives up to order k at p (cf. Problem 90).
(a). Prove that c is a geodesic in Sfd for every f ∈ F1 .
(b). Prove that for every f ∈ F2 , the following holds in Sfd : c is a geodesic,
c(0) and c(π) are conjugate along c, and there is no point before c(π) con-
jugate to c(0) along c.
(c). Let U ⊂ S d be an open set which has nonempty intersection with the
geodesic c, and let f be any function in F1 which satisfies f ≥ 1 on all S d
and f (p) > 1 for all p ∈ U \ c([0, π]). Prove that then c is a strict local
minimum for L in Sfd among pw C ∞ curves with fixed endpoints.
(d). Take U as in part (c), and let f be any function in F1 which satisfies
f ≤ 1 on all S d and f (p) < 1 for all p ∈ U \ c([0, π]). Prove that then c is not
a local minimum for L in Sfd among pw C ∞ curves with fixed endpoints.
[Comment: It is a standard fact from analysis that there exist functions f
as in (c) and (d), also in Fk with k arbitrarily large. It then follows from
(b), (c), (d) that in the situation described in the remark immediately be-
low Theorem 1 in Lecture #18 – i.e. when the endpoints of c are conjugate
but there is no previous point along c conjugate to the starting point – one
cannot make any general statement about c being or not being a (strict or
non-strict) local minimum for L!]
Solution:
p. 233.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 43

Problem 92. [A comparison result for lengths of curves.]

Let M0 and M be d-dimensional complete Riemannian manifolds such that
M0 has constant sectional curvature µ and the sectional curvature of M
is everywhere ≤ µ. Fix points p ∈ M and p0 ∈ M0 , and identify both
Tp M and Tp0 M with Rd in a way carrying the respective Riemannian scalar
products to the standard scalar product in Rd . Take r > 0 so small that
expp0 restricted to the open ball Br (0) ⊂ Rd is a diffeomorphism onto an
open subset of M0 . Prove that for any pw C ∞ curve c : [a, b] → Br (0),
L(expp ◦ c) ≥ L(expp0 ◦ c).
(Here expp ◦ c is a curve on M while expp0 ◦ c is a curve on M0 .)
[Hint: Try to prove a stronger statement comparing the norms of d(expp )x (v)
and d(expp0 )x (v) for any x ∈ Br (0) and v ∈ Rd . Here use can be made of
Corollaries 1 and 2 in Lecture #17 and Theorem 1 in Lecture #19 (the
Rauch Comparison Theorem).]
Solution:
p. 237.
Problem 93. [Focal points (special case).]
Let γ : [−η, η] → M and c : [a, b] → M be geodesics on the Riemannian
manifold M , satisfying c(a) = γ(0), ċ(a) 6= 0, and hċ(a), γ̇(0)i = 0. For
τ ∈ (a, b], c(τ ) is called a focal point of γ along c if there exists a nontrivial
Jacobi field X along c such that X(τ ) = 0, and
X(a) ∈ Span(γ̇(0)) and Ẋ(a) ⊥ γ̇(0) in Tc(a) (M ).
Prove that if there is some τ ∈ (a, b) such that c(τ ) is a focal point of γ
along c, then there exists a variation c : [a, b] × (−ε, ε) → M of the curve
c such that c(a, s) ∈ γ([−η, η]) and c(b, s) = c(b) for all s ∈ (−ε, ε) and
L(s) < L(0) for all s ∈ (−ε, ε) \ {0} (with L(s) := L(c(·, s)), as usual).
[Hint: If γ is a constant point then the result follows from Theorem 1 in
Lecture #18; thus try to extend the proof of that theorem to the present
situation. See also Problem 83(b).]
[Comment: More generally one can define the notion of “focal point” for
any submanifold of M (in the place of γ above); cf., e.g., [2, p. 23TM].
The general definition looks different from our definition above, however in
the special case which we consider, i.e. that of a submanifold which is a
geodesic, the two formulations can be shown to be equivalent. Note also
that the definition given in Jost, [12, Exc. 5.2], is completely incorrect.]
Solution:
p. 239.
 44 ANDREAS STRÖMBERGSSON

Problem 94. [Local isometry ⇒ covering map.]

f and M be Riemannian manifolds with M
Let M f complete, and let
π:Mf→M
be a local isometry. Prove that then M is complete and π is a covering map.
Solution:
p. 241. Problem 95. [The Killing-Hopf Theorem.]
Prove the Killing-Hopf Theorem: Let M be an n-dimensional complete,
simply connected Riemannian manifold with constant sectional curvature.
Then M is isometric to Rn (with its standard Riemannian metric) or a
sphere of radius r > 0 in Rn+1 (with its standard Riemannian metric) or
the hyperbolic space H n (ρ) introduced in [12, Sec. 5.4] (cf. Problem 20).
Solution:
p. 241.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 45

2. Solution suggestions

Problem 1: For any p ∈ M , let Up be the set of points q ∈ M for which

there exists a curve from p to q. Using the fact that M is locally Euclidean
one verifies that
 
(4) ∀p ∈ M : Up is open .

Next let us note:

 
(5) ∀p, q ∈ M : Up ∩ Uq 6= ∅ ⇒ q ∈ Up .
[Proof: Assume Up ∩ Uq 6= ∅; then there is a point q ′ ∈ Up ∩ Uq . Now q ′ ∈ Up
means that there is a curve γ1 in M from p to q ′ , and q ′ ∈ Uq means that
there is a curve γ2 in M from q to q ′ . Then the “product path” of γ1 and
the “inverse path” of γ2 6 is a curve in M from p to q. Hence q ∈ Up .]
Now for any p ∈ M , if q ∈ ∁Up (complement wrt M ) then also Uq ⊂ ∁Up ,
by (5), and Uq is open (by (4)), and q ∈ Uq (immediate from the definition of
Uq ). Hence every point in ∁Up has an open neighborhood which is contained
in ∁Up . Hence ∁Up is open (viz., Up is closed).
Hence for every p ∈ M , both Up and ∁Up are open. Furthermore M equals
the disjoint union of these two sets. Hence since M is connected, either Up
or ∁Up must be empty. But p ∈ Up ; hence ∁Up = ∅, i.e. Up = M . By the
definition of Up , this means that for every q ∈ M there exists a curve from
p to q. 

6we will discuss these notions in Lecture #6, and the product path in question will be
denoted “γ1 · γ 2 ”; however it should hopefully be clear already at this point how the curve
in question is constructed; just draw a picture!
46 ANDREAS STRÖMBERGSSON

Problem 2:
(a). First assume that M has a countable atlas A. For each chart (U, x) ∈
A, since x(U ) (an open subset of Rd ) is second countable, we can choose a
base U = U(U,x) for the topology of x(U ). Then set
U ′ = U(U,x)
′
:= {x−1 (V ) : V ∈ U(U,x) }.
This is a countable family of open subsets of U . Next let U ′′ be the union
′
of all families U(U,x) as (U, x) runs through A. This is a countable family of
open subsets of M . We claim that U ′′ is a base for the topology of M . In
order to prove this, let Ω be an arbitrary open set in M , and let p ∈ Ω. Take
a chart (U, x) ∈ A with p ∈ U . Then Ω ∩ U is an open set in U containing
p, and so x(Ω ∩ U ) is an open subset of x(U ) and x(p) ∈ x(Ω ∩ U ). Hence,
since U(U,x) is a base for x(U ), there is V ∈ U(U,x) such that
x(p) ∈ V ⊂ x(Ω ∩ U ).
Then x−1 (V ) ∈ U ′ ⊂ U ′′ and
p ∈ x−1 (V ) ⊂ Ω ∩ U ⊂ Ω.
This proves that U ′′ is a base for the topology of M . Done!
We now prove the opposite implication. Thus assue that M is second
countable; let U be a countable base for the topology of M . Also let A be
the family of all charts on M ; this is an atlas for M . Set
U ′ := {U ∈ U : there is some x : U → Rd s.t. (U, x) ∈ A}.
We claim that U ′ covers M , i.e. ∪U ∈U ′ U = M . To prove this, take an
arbitrary point p ∈ M . Then there is some chart (U, x) ∈ A with p ∈ U ,
and since U is a base for M there is V ∈ U such that p ∈ V ⊂ U . Now
(V, x|V ) is also a chart for M (since the restriction of a homeomorphism to
an open subset is itself a homeomorphism onto its image), i.e. (V, x|V ) ∈ A,
and thus V ∈ U ′ . Hence U ′ indeed covers M . It follows that if for each
U ∈ U ′ we choose one map xU : U → Rd such that (U, xU ) ∈ A, then
{(U, xU ) : U ∈ U ′ }
is an atlas for M . This atlas is countable since U ′ is countable (since U ′ ⊂ U ).
Done! 
(b). By the notes to Lecture #1, this is clear from part (a). 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 47

Problem 3: Let M be a connected topological space for which every

point has an open neighborhood U which is homeomorphic to an open subset
Ω of Rd for some d ∈ Z≥1 (which apriori may depend on U ). Note that we
actually don’t need to assume that M is Hausdorff for the following argument
to work.
For each d ∈ Z≥1 , let Fd be the family of all open sets U ⊂ M which are
homeomorphic to an open subset of Rd . Then the assumption on M implies
that
∞  [
[ 
(6) M= U .
d=1 U ∈Fd

Using Brouwer’s Theorem on invariance of dimension, we now have:

(7) ∀d 6= d′ ∈ Z≥1 : ∀U ∈ Fd , V ∈ Fd′ : U ∩ V = ∅.
[Detailed proof: Take such U, V and set W := U ∩V . Note that W is an open
subset of both U and V . Now U ∈ Fd implies that U is homomorphic to an
open subset of Rd ; this homeomorphism then restricts to a homeomorphism
of W to a (smaller) open subset of Rd . Similarly V ∈ Fd′ implies that W is
′
also homeomorphic to an open subset of Rd . Hence by Brouwer’s Theorem
on invariance of dimension, using d = d′ , we must have W = ∅, qed.]
The property (7) implies that the unions ∪U ∈Fd U are pairwise disjoint
for d = 1, 2, . . .. Also each such union is an open set, since it is a union of
open sets. Hence (6) expresses M as a union of disjoint open sets. But M
is connected; therefore ∪U ∈Fd U must be empty for all except (at most) one
d, say d0 . This means that Fd = {∅} for all d 6= d0 , and this implies the
desired result. 

Problem 4: Let the dimension of M be d. Let A be the given C ∞ atlas,

and let A′ be the family of all charts which are compatible with every chart
in A. Let us start by proving that A′ is a C ∞ atlas. Clearly A ⊂ A′ and
thus the charts in A′ cover M . Thus it remains to prove that any two charts
in A′ are C ∞ compatible. Thus consider any two charts (U, x), (V, y) ∈ A′ ;
we need to prove that the map
(8) y ◦ x−1 : x(U ∩ V ) → y(U ∩ V ) ⊂ Rd
is C ∞ . (It is clear that that map in (8) is a homeomorphism, since (U, x)
and (V, y) are charts.) Take p ∈ U ∩ V ; it suffices to prove that there is some
open neighborhood Ω ⊂ x(U ∩ V ) of x(p) such that (y ◦ x−1 )|Ω is C ∞ . Since
A is an atlas, there is some chart (W, z) ∈ A with p ∈ W . By assumption
both (U, x) and (V, y) are compatible with (W, z); hence both the maps
(9) z ◦ x−1 : x(U ∩ W ) → z(U ∩ W )
48 ANDREAS STRÖMBERGSSON

and
(10) y ◦ z −1 : z(V ∩ W ) → y(V ∩ W )
are diffeomorphisms. Now set
Ω := x(U ∩ V ∩ W ).
This is an open subset of x(U ), since U ∩ V ∩ W is an open subset of U and x
is a homeomorphism. Restricting the diffeomorphisms in (9) and (10) to the
open subsets Ω and z(U ∩ V ∩ W ), respectively, we obtain diffeomorphisms
(11) z ◦ x−1 : Ω → z(U ∩ V ∩ W )
and
(12) y ◦ z −1 : z(U ∩ V ∩ W ) → y(U ∩ V ∩ W ).
It follows that the composition of these two maps is also a diffeomorphism,
from Ω onto y(U ∩ V ∩ W ). But this composition equals (y ◦ x−1 )|Ω . Hence
we have proved, in particular, that (y ◦ x−1 )|Ω is C ∞ . This completes the
proof that A′ is a C ∞ atlas.
It is immediate from the construction of A′ that A′ is C ∞ structure, i.e. a
maximal C ∞ atlas. (Indeed, suppose that A′′ is any C ∞ atlas with A′′ ⊃ A′ .
Let (U, x) ∈ A′′ . By definition of “atlas”, (U, x) is compatible with every
chart in A′′ ; and A ⊂ A′ ⊂ A′′ ; hence (U, x) is compatible with every chart
in A, and therefore (U, x) ∈ A′ , by the definition of A′ . Hence we have
proved that A′′ ⊂ A′ , and so in fact A′′ = A′ .)
It remains to prove that A′ is the only C ∞ structure on M with A ⊂ A′ .
Thus assume that A′′ is an arbitrary C ∞ structure on M with A ⊂ A′′ .
Since A′′ is a C ∞ atlas, every chart (U, x) ∈ A′′ is compatible with every
chart in A′′ ; in particular (U, x) is compatible with every chart in A, and
thus (U, x) ∈ A′ , by the definition of A′ . Hence A′′ ⊂ A′ . But this implies
that A′′ = A′ , since A′′ is a maximal C ∞ atlas. This completes the proof.

 PROBLEMS; “RIEMANNIAN GEOMETRY” 49

Problem 5:
(a) One simple way to construct such a set H is as follows. Given any
real number 0 < t < 2, let


tr if r ∈ (0, 41 ]
ft : (0, 1) → (0, 1), ft (r) := 12 (t − 1) + (2 − t)r if r ∈ ( 41 , 12 ]


r if r ∈ ( 21 , 1).
One verifies that ft is continuous, strictly increasing, and bijective, with
inverse

 −1
t r if r ∈ (0, 4t ]
−1 −1
ft : (0, 1) → (0, 1), ft (r) := (2 − t)−1 (r + 12 (1 − t)) if r ∈ ( 4t , 21 ]


r if r ∈ ( 12 , 1)
which is also continuous and strictly increasing. (These facts are most easily
verified by simply drawing the graph of ft ; this graph is a union of three
line segments: one from (0, 0) to ( 41 , 14 t), one from ( 41 , 14 t) to ( 21 , 12 ), and one
from ( 12 , 12 ) to (1, 1).) Hence ft is a homeomorphism of (0, 1) onto itself.
Next define ht : B1 (0) → B1 (0) through

0 if x = 0
ht (x) = ft (kxk)
 x if x 6= 0.
kxk
Note that
kht (x)k = ft (kxk), ∀x ∈ B1 (0) \ {0};
and recall ft (r) ∈ (0, 1) for all r ∈ (0, 1); hence ht is indeed a map into
B1 (0). Clearly ht is continuous in B1 (0) \ {0}; but we also have kht (x)k → 0
(i.e. ht (x) tends to the origin) as x → 0, since ft (r) → 0 as r → 0+ ; therefore
ht is continuous in all B1 (0). Similarly one verifies that

0 if x = 0
e −1
ht (x) = ft (kxk)
 x if x 6= 0.
kxk
defines a continuous map e ht : B1 (0) → B1 (0) satisfying ke
ht (x)k = ft−1 (kxk)
for all x ∈ B1 (0) \ {0}. Now we have
ht (e
ht (x)) = e
ht (ht (x)) = x, ∀x ∈ B1 (0).
(Proof: This is immediate for x = 0. Now assume x 6= 0. Then
ft (ke
ht (x)k) e ft (ft−1 (kxk)) e kxk e
ht (e
ht (x)) = ht (x) = −1 ht (x) = −1 ht (x) = x.
e
kht (x)k f t (kxk) f t (kxk)

The proof of eht (ht (x)) = x is completely similar.) Hence ht and e

ht are both
bijections of B1 (0) onto B1 (0), and they are each other inverses. Since they
50 ANDREAS STRÖMBERGSSON

are both continuous, it follows that ht is a homeomorphism of B1 (0) onto

itself.
Note also that ht satisfies ht (x) = x for all x ∈ B1 (0) \ B1/2 (0), since
ft (r) = r for r ∈ [ 21 , 1).
We now let H be the family of all these homeomorphisms ht :
H := {ht : t ∈ (0, 2)}.
This family clearly satisfies all the requirements, if we can only prove that for
any two t1 6= t2 ∈ (0, 2), the homeomorphism ht1 ◦ h−1 ∞
t2 is not C . (Indeed,
this will in particular imply that ht1 6≡ ht2 for all t1 6= t2 ∈ (0, 2), and so the
family H is uncountable.)
Thus let t1 6= t2 ∈ (0, 2) be given. Now for all x 6= 0 we have:
ft1 (ft−1 (kxk)) ft−1 (kxk) ft1 (ft−1 (kxk))
(13) ht1 (h−1
t2 (x)) = −1
2
· 2
x = 2
x.
ft2 (kxk) kxk kxk
Using the explicit formulas for ft and ft−1 given above, we compute

−1

 t1 t2 r if r ∈ (0, 14 t2 ]
ft1 (ft−1 (r)) = 21 (t1 − 1) + 2−t 1−t2 1 1
2−t2 (r + 2 ) if r ∈ ( 4 t2 , 2 ]
1
2


r if r ∈ ( 21 , 1).
From this we see that the (continuous) function ft1 ◦ ft2 : (0, 1) → (0, 1) is
2−t1
not C ∞ ; for example at r = 12 the function has left derivative 2−t 2
and right
derivative 1, and these are not equal, since t1 6= t2 . (Similarly the function
has different left and right derivatives at r = 41 t2 .) From this it follows that
the function ht1 ◦ h−1 ∞ −1
t2 : B1 (0) → B1 (0) is not C . (Indeed, if ht1 ◦ ht2 were
C ∞ then, writing e1 for the standard unit vector (1, 0, . . . , 0) ∈ Rd , it would
follow that the function
(−1, 1) → (−1, 1), r 7→ ht1 (h−1
t2 (re1 )) · e1 ,

were C ∞ ; but it follows from (13) that ht1 (h−1 −1

t2 (re1 )) · e1 = ft1 (ft2 (r)) for
r ∈ (0, 1), and so we would have a contradiction against the fact that ft1 ◦ft2
is not C ∞ .) 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 51

(b). Let A be a fixed C ∞ structure on M (it exists by assumption). Fix

some chart (V, y) ∈ A. Let y0 be a point in y(V ); then since y(V ) is open,
there is some r > 0 such that Br (y0 ) ⊂ y(V ). Set U := y −1 (Br (y0 )); this
is an open subset of V , and (U, y|U ) ∈ A, since A is a maximal C ∞ atlas.
Define the map
1
x : U → Rd , x(p) := (y(p) − y0 ).
r
Then (U, x) ∈ A, since x equals y|U composed with a diffeomorphism of
Br (y0 ) onto B1 (0). Note that x(U ) = B1 (0). From now on we keep this
chart (U, x) fixed.
Now for any given homeomorphism h of B1 (0) onto B1 (0) satisfying
h(q) = q for all q ∈ B1 (0) \ B1/2 (0), we define a function ϕh : M → M
as follows:
(
p if p ∈
/U
ϕh (p) = −1
x (h(x(p))) if p ∈ U.
We claim that ϕh is continuous. It is immediate from the definition of ϕh
that the restrictions of ϕh to U and to M \ U are both continuous. Hence
since U is open (and so M \ U is closed) it now suffices to verify that if
p1 , p2 , . . . is any sequence of points in U such that pj → p ∈ M \U as j → ∞,
then ϕh (pj ) → ϕh (p). However the fact that (pj ) tends to a point outside U
implies that (pj ) has only finitely many points in any fixed compact subset
of U ; in particular for all sufficiently large j we have pj ∈ / x−1 (B1/2 (0)),
and thus h(x(pj )) = x(pj ) and ϕh (pj ) = pj . Also ϕh (p) = p since p ∈ / U,
and it follows that ϕh (pj ) → ϕh (p). This completes the proof that ϕh is
continuous.
Furthermore, one verifies immediately that ϕh is a bijection with inverse
map equal to ϕh−1 : M → M , and the above argument applies also to ϕh−1 ,
showing that ϕh−1 is continuous. Hence ϕh is a homeomorphism of M onto
itself.
Next we prove:
Lemma 1. If A is a C ∞ structure on M , and ϕ is a homeomorphism of
M onto itself, then also
Aϕ := {(ϕ−1 (V ), y ◦ ϕ) : (V, y) ∈ A}
is a C ∞ structure on M . Let us write (M, A) for the C ∞ manifold given by
A, and (M, Aϕ ) for the C ∞ manifold given by Aϕ . Then ϕ is a diffeomor-
phism of (M, Aϕ ) onto (M, A).

(Remark: The whole lemma can be seen as obvious. Namely, (M, Aϕ ) can
be seen as “what one gets from (M, A) after changing names on all points
according to ϕ”. Viewed in this way, ϕ “is identity map”!)
52 ANDREAS STRÖMBERGSSON

Proof. Let T be the family of all charts on the topological manifold M . Note
that for any (V, y) ∈ T we have (ϕ−1 (V ), y ◦ ϕ) ∈ T ; hence we have a map
Φ:T →T, Φ(V, y) := (ϕ−1 (V ), y ◦ ϕ).
In fact Φ is a bijection, with Φ−1 (V, y) = (ϕ(V ), y ◦ ϕ−1 ). Note that
(14) Aϕ = {Φ(V, y) : (V, y) ∈ A}.

Next we note that for any two charts (V, y), (W, z) ∈ T , we have the
equivalence
[(V, y) and (W, z) are C ∞ compatible]
(15) ⇔ [Φ(V, y) and Φ(W, z) are C ∞ compatible].
Indeed, by definition (V, y) and (W, z) are C ∞ compatible iff the map
(16) z ◦ y −1 : y(V ∩ W ) → z(V ∩ W )
is a diffeomorphism (it is always a homeomorphism), and similarly Φ(V, y)
and Φ(W, z) are C ∞ compatible iff the map
(17) z ◦ ϕ ◦ (y ◦ ϕ)−1 : (y ◦ ϕ)(ϕ−1 (V ) ∩ ϕ−1 (W ))
→ (z ◦ ϕ)(ϕ−1 (V ) ∩ ϕ−1 (W ))
is a diffeomorphism. However, ϕ−1 (V ) ∩ ϕ−1 (W ) = ϕ−1 (V ∩ W ), and now
by inspection one verifies that the two maps in (16) and (17) are the same.
Hence the equivalence in (15) holds.
Clearly the charts in Aϕ cover M , since the charts in A cover M . Note
also that any two charts in Aϕ are C ∞ compatible; this follows from (14)
and (15) and the fact that A is a C ∞ atlas. Hence Aϕ is a C ∞ atlas. In fact
(14) and (15) show that for any chart (V, y) ∈ T , if (V, y) is C ∞ compatible
with Aϕ then Φ−1 (V, y) is C ∞ compatible with A; hence Φ−1 (V, y) ∈ A
since A is a maximal C ∞ atlas, and so (V, y) ∈ Aϕ . Hence Aϕ is a maximal
C ∞ atlas on M , i.e. Aϕ is a C ∞ structure on M .
It remains to prove that ϕ is a diffeomorphism of (M, Aϕ ) onto (M, A).
For this, our task is to verify that for any (V, y) ∈ Aϕ and any (W, z) ∈ A,
the map
z ◦ ϕ ◦ y −1 : y(V ∩ ϕ−1 (W )) → z(V ∩ ϕ−1 (W ))
is a diffeomorphism. However (V, y) ∈ Aϕ means that (V, y) = Φ(Ve , ye) =
(ϕ−1 (Ve ), ye ◦ ϕ) for some (Ve , ye) ∈ A, and so
z ◦ ϕ ◦ y −1 = z ◦ ϕ ◦ (e
y ◦ ϕ)−1 = z ◦ ye−1
on the set
y(V ∩ ϕ−1 (W )) = ye(ϕ(V ∩ ϕ−1 (W ))) = ye(ϕ(V ) ∩ W ) = ye(Ve ∩ W ).
Thus, our task is to verify that z ◦ϕ◦y −1 is a diffeomorphism from ye(Ve ∩ W )
onto z(Ve ∩ W ), and this holds since (Ve , y) and (W, z) are charts in A. 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 53

Now take H as in part (a), and form the family

F := {Aϕh : h ∈ H}.
By Lemma 1, each Aϕh in F is a C ∞ structure on M , and all C ∞ manifolds
defined by these C ∞ structures are diffeomorphic. Hence it now only remains
to prove that Aϕh1 6= Aϕh2 for any two h1 6= h2 ∈ H. Let us write A1 = Aϕh1
and A2 = Aϕh2 for short.
Recall that (U, x) ∈ A; hence (ϕ−1 −1
h1 (U ), x ◦ ϕh1 ) ∈ A1 and (ϕh2 (U ), x ◦
ϕh2 ) ∈ A2 . Note that the map
(x ◦ ϕh1 ) ◦ (x ◦ ϕh2 )−1 : (x ◦ ϕh2 )(ϕ−1 −1
h2 (U )) → (x ◦ ϕh1 )(ϕh1 (U ))
is the same as
x ◦ ϕh1 ◦ ϕ−1
h2 ◦ x
−1
: B1 (0) → B1 (0),
and by the definition of ϕh , this is the same as
h1 ◦ h−1
2 : B1 (0) → B1 (0),
which is not C ∞. Hence the two charts (ϕ−1 −1
h1 (U ), x ◦ ϕh1 ) and (ϕh2 (U ), x ◦
ϕh2 ) are not C ∞ compatible, and therefore A1 6= A2 . 
54 ANDREAS STRÖMBERGSSON

Problem 6: All this is “completely obvious” once one understands the

basic machinery with (C ∞ ) atlases. Let us go through the details:
(a). Here we are talking about a new type of object: “a (not necessarily
connected) C ∞ manifold”. The definition should hopefully be obvious7 ...
Namely: A “(not necessarily connected) C ∞ manifold” is a topological space
M such that every connected component of M is a C ∞ manifold!
We now solve the given problem. Note that every connected component
of U is also an open subset of M . Hence if we can prove that every connected
open subset of M has a natural structure of a C ∞ manifold, then it follows
that U has a natural structure of a (not necessarily connected) C ∞ manifold,
and so we will be done.
Thus from now on assume that U is a connected open subset of M . Let
the dimension of M be d. We endow U with the restricted topology; then
U is a connected Hausdorff space.
Let A be the C ∞ structure of M . Thus A is a maximal C ∞ atlas on M .
Set
A|U := {(V, x) : (V, x) ∈ A, V ⊂ U }.
We wish to prove that A|U is a C ∞ atlas on U . Clearly every (V, x) ∈ A|U
is a chart on U and these charts are pairwise C ∞ compatible, since A is
a C ∞ atlas. Hence it remains to prove that the charts in A|U cover U .
Take p ∈ U . Then there is a chart (V, x) ∈ A with p ∈ V . Now note that
also (V ∩ U, x|V ∩U ) is a chart on M , and (V ∩ U, x|V ∩U ) is C ∞ compatible
with every chart in A since (V, x) is C ∞ compatible with every chart in
A. Hence, since A is maximal, we have (V ∩ U, x|V ∩U ) ∈ A. Thus also
(V ∩ U, x|V ∩U ) ∈ A|U , since V ∩ U ⊂ U . Also of course p ∈ V ∩ U . Hence
A|U contains a chart which contains p. Since this is true for every p ∈ U , we
conclude that the charts in A|U indeed cover U . Hence A|U is a C ∞ atlas
on U , and so determines a unique C ∞ structure on U (cf. Problem 4). 8
It remains to prove that U is paracompact. This is equivalent to proving
that U is second countable (cf. the notes to Lecture #1). However this is
clear from the fact that M is second countable; indeed it is easy to prove that
any open subset of a second countable topological space is second countable.


(b), (c) ... we leave this to the reader ... (Note that the first part of (c)
is immediate from (b), since f|U = f ◦ i.)

7In the literature it varies whether one defines a “manifold” to always be connected.
However recall that in our course, we do require every manifold to be connected!
8(In fact one can show that A is itself a C ∞ structure on U .)
|U
 PROBLEMS; “RIEMANNIAN GEOMETRY” 55

Problem 7:
(a) Let W = M \ supp(f ); this is an open subset of M , and W ∪ U = M .
We have f|U ∈ C ∞ (U ) by assumption. also f|W ≡ 0; hence f|W ∈ C ∞ (W ).
(Cf. Problem 6 regarding the fact that U and W are C ∞ manifolds; hence
the function spaces “C ∞ (U )” and “C ∞ (W )” are defined.) Hence every point
in M has an open neighbourhood in which f is C ∞ . Hence by Problem 6(c),
f ∈ C ∞ (M ).
(b) Let K = supp(f ); by assumption this is a compact subset of U .
We claim that supp(fe) = K; if we prove this then the desired statement
fe ∈ C ∞ (M ) follows from part (a). Note that K is a compact subset of M
(since “compactness is an absolute property”; for example, use the fact that
the inclusion map i : U → M is continuous, and the image of any compact
set under a continuous map is compact). Hence K is a closed subset of M .
Also note that {p ∈ M : fe(p) 6= 0} ⊂ K, by the definitions of fe and K.
Hence supp(fe), being the closure of {p ∈ M : fe(p) 6= 0} in M , is contained
in K. The opposite inclusion is obvious; hence supp(fe) = K. Done! 
(c). This is a special case of part (d).
(d). Let {(Uα , xα )} be a C ∞ atlas on M . Let (Vβ )β∈B together with
(ϕβ )β∈B be a partition of unity subordinate to (Uα ), as in [12, Lemma 1.1.1].
For each β ∈ F we write Kβ := supp ϕβ ; this is a compact set contained in
Vβ .
Let us start by noticing that for each β ∈ B, there exists a C ∞ function
fβ : Vβ → [0, 1] which has compact support contained in Vβ ∩ U and which
satisfies fβ|K∩Kβ ≡ 1. Indeed, since (Vβ )β∈B is a refinement of (Uα ), for our
given β ∈ F there exists some α such that Vβ ⊂ Uα . Using now the chart
(Uα , xα ) to translate the problem into Euclidean coordinates, we are reduced
to proving that for any compact set K e and open set U e with K e ⊂U e ⊂ Rd ,
there is a C ∞ function f : Rd → [0, 1] with compact support contained in U e
and which satisfies f|Ke ≡ 1. For this, cf., e.g., [10, Thm. 1.4.1] (one considers
the convolution of the characteristic function of K e and a “bump” function
with sufficiently small support).
For any β ∈ B such that K ∩ Vβ = ∅ we may of course choose the above
function fβ to be identically zero; from now on we require this to hold.
Next for each β ∈ F we define feβ : M → R by feβ ≡ fβ in Vβ and feβ = 0
outside Vβ ; by part (a) we then have feβ ∈ C ∞ (M ). We next set
X
f := ϕβ feβ .
β∈B

Clearly
P f ∈ C ∞ (M ). Also for every p ∈ M we have f (p) ≥ 0 and f (p) ≤
β∈F ϕβ (p) ≤ 1.
56 ANDREAS STRÖMBERGSSON
P
Fix an arbitrary point p ∈ K. We have β∈B ϕβ (p) = 1 by the defining
property ot (ϕβ ). Also feβ (p) = 1 for every β ∈ B with p ∈ Kβ (since we
are assuming p ∈ K), and thus feβ (p) = 1 for every β ∈ B with ϕβ (p) 6= 0.
P P
Hence f (p) = β∈B ϕβ (p)feβ (p) = β∈B ϕβ (p) = 1. We have thus proved
that f|K ≡ 1.
In order to prove that f has compact support, we will use the requirement
from above that fβ ≡ 0 (and so feβ ≡ 0) whenever K ∩ Vβ = ∅. This means
that the sum defining f may just as well be restricted to the following subset
of B:
F := {β ∈ B : K ∩ Vβ 6= ∅}.
Now since (Vβ ) is locally finite, F is a finite set. (This is a standard fact; here
is a detailed proof: Since (Vβ ) is locally finite, for every p ∈ M we can choose
– using the axiom of choice – an open set Up ⊂ M such that p ∈ Up and
#{β ∈ B : Vβ ∩Up 6= ∅} < ∞. Since K is compact, there exists a finite subset
F ′ ⊂ K such that K ⊂ ∪p∈F ′ Up . NowPfor every β ∈ F there is some p ∈ F ′
such that Vβ ∩ Up 6= ∅; hence #F ≤ p∈F ′ #{β ∈ B : Vβ ∩ Up 6= ∅} < ∞.
Done!)
P P
It follows from f (p) = β∈B ϕβ (p)feβ (p) = β∈F ϕβ (p)feβ (p) that
[ [
supp(f ) ⊂ supp(feβ ) = supp(fβ )
β∈F β∈F

(for the inclusion one uses the fact that ∪β∈F supp(feβ ) is a closed subset of
M ; for the equality see part (b)). The last set is a finite union of compact
sets, hence itself compact. Also by construction, supp(fβ ) ⊂ U for each
β ∈ F . Hence supp(f ) ⊂ U , and also since supp(f ) is closed and contained
in a compact set, supp(f ) is itself compact.
Hence the function f has all the desired properties. 

(e) Let g : M → [0, 1] be a function as in part (d), i.e. g is C ∞ , has

compact support contained in U , and satisfies g|K ≡ 1. Set
(
g(p)f (p) if p ∈ U
f1 (p) :=
0 if p ∈
/ U.
Then clearly f1|K ≡ f|K . Also the function p 7→ g(p)f (p) is a C ∞ function
U → R with compact support, and f1 is the same as “fe in part (b), but
starting from the function p 7→ g(p)f (p) on U ”. Hence f1 is C ∞ , by part
(b). 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 57

Problem 8.
(a). We endow M × N with the product topology (viz., a subset of M × N
is open iff it can be written as a union of sets of the form U × V with U ⊂ M
and V ⊂ N ). Then M ×N is Hausdorff and connected. (We leave the details
to the reader...) We will verify at the end that M × N is also paracompact
(according to Wikipedia the product of to general paracompact topological
spaces need not be paracompact; thus we need to make use of the fact that
M, N have more structure).
Let the dimensions of M and N be d and d′ , respectively. Let A be a
C∞ structure on M and let B be a C ∞ structure on N . For any charts
(U, x) ∈ A and (V, y) ∈ B, U × V is an open set in M × N , and we write
(x, y) 9 for the map
′ ′
(x, y) : U × V → Rd × Rd = Rd+d , (x, y)(p, q) := (x(p), y(q)).
This map (x, y) is in fact a homeomorphism from U × V onto x(U ) × y(V )
′
(which is an open subset of Rd × Rd ).
[Outline of proof: (x, y) is clearly a bijection from U ×V onto x(U )×y(V ).
We leave it to the reader to verify – or recall from basic point set topology
– that (x, y) is continuous. Similarly the inverse map, (x, y)−1 = (x−1 , y −1 )
is continuous since x−1 and y −1 are continuous.]
Hence for any charts (U, x) ∈ A and (V, y) ∈ B, we have that (U ×V, (x, y))
is a chart on M × N . Now set
C := {(U × V, (x, y)) : (U, x) ∈ A, (V, y) ∈ B}.
This is clearly an atlas on M × N .
Let us now verify that M × N is paracompact. By the notes to Lecture
#1, with reference to math.stackexchange.com/questions/527642, it suffices
to prove that M ×N is second countable, and this is a simple consequence of
the fact that M and N are second countable. Indeed, let UM be a countable
base for M and let UN be a countable base for N , and set
U = {U × V : U ∈ UM , V ∈ UN }.

9Note that there is a glitch between this notation “(x, y)” and the notation “(f, g)”
used in part (c). Let us discuss this carefully in the abstract setting of maps between
sets: Thus if A, B1 , B2 are three sets and α : A → B1 and β : A → B2 are maps, then we
define the map “(α, β) : A → B1 × B2 ” by (α, β)(p) := (α(p), β(p))”. This is the notation
which we use in part (c), and it is also the standard notation for “category theoretical
product”; cf. wikipedia. On the other hand if A1 , A2 , B1 , B2 are sets and γ : A1 → B1
and δ : A2 → B2 are maps then we define the map [γ, δ] : A1 × A2 → B1 × B2 by
[γ, δ](p, q) := (γ(p), δ(q)). The “(x, y)” which we use here in our solution to part (a) is
this constructions; we use the notation “[·, ·]” in this footnote for clarity, but in practice
there is no problem to use “(·, ·)” for both, and it is also quite standard. Note that the
two constructions are related by the simple relation [γ, δ] = (γ ◦ pr1 , δ ◦ pr2 ); indeed see
part (d) of the present problem.
58 ANDREAS STRÖMBERGSSON

Then U is countable. We claim that U is a base for M × N . To prove this,

let W be an open set in M × N , and let (p, q) be a point in W . Then, by the
definition of the product topology on M × N , there exist open sets U ′ ⊂ M
and V ′ ⊂ N such that (p, q) ∈ U ′ × V ′ ⊂ W . Next, since UM and UN are
bases, there exists U ∈ UM with p ∈ U ⊂ U ′ and there exists V ∈ UN with
q ∈ V ⊂ V ′ . Then U × V ∈ U and (p, q) ∈ U × V ⊂ W . The fact that such
a set exists in U for any given W, p as above proves that U is indeed a base
for M × N . Hence M × N is second countable.
Finally, we claim that C is a C ∞ atlas. To prove this we consider an
arbitrary pair of charts in C, say (U × V, (x, y)) and (W × Ω, (r, s)), where
(U, x), (W, r)) ∈ A and (V, y), (Ω, s) ∈ B. We have to prove that the map
′ ′
(r, s) ◦ (x, y)−1 : (x, y)(U × V ) → Rd × Rd = Rd+d
is C ∞ . However this map equals (r ◦ x−1 , s ◦ y −1 ), and this map is C ∞ since
both r ◦ x−1 and s ◦ y −1 are C ∞ . (Indeed, recall that by definition a map
f from an open subset D ⊂ Rm to Rn is C ∞ if and only if, when writing
f (z) = (f1 (z), . . . , fn (z)) for z ∈ D, each “component” map fj : D → R is
C ∞ . When applying this to f = (r ◦ x−1 , s ◦ y −1 ), each component fj is in
fact a component of either r ◦ x−1 or s ◦ y −1 , hence C ∞ .) Hence C is a C ∞
atlas on M × N , and so determines a unique C ∞ structure on M × N (cf.
Problem 4).
Hence M × N is a C ∞ manifold.

(b). Let the C ∞ atlases A, B, C be as in part (a). In order to prove that pr1
is C ∞ we have to prove that for any charts (W, z) ∈ A and (U ×V, (x, y)) ∈ C
(with (U, x) ∈ A and (V, y) ∈ B), the map
 
z ◦ pr1 ◦(x, y)−1 : (x, y) (U × V ) ∩ pr−1 1 (W ) → z(W ) ⊂ Rd

is C ∞ . We may here note that (U × V ) ∩ pr−1

1 (W ) = (U ∩ W ) × V . But the
above map equals:
(18) z ◦ pr1 ◦(x, y)−1 = (z ◦ x−1 ) ◦ p1 ,
′ ′
where p1 is the projection map Rd+d = Rd × Rd → Rd . The map z ◦ x−1
(from x(U ∩ W ) to z(W )) is C ∞ since (U, x), (W, z) ∈ A and A is a C ∞
chart. The map p1 is obviously C ∞ . Hence the composed map in (18) is
C ∞ , and we are done.
The proof that pr2 is C ∞ is completely similar. 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 59

(c). Let the dimensions of M, N1 , N2 be d, d1 , d2 , respectively. Let A, A1 , A2

be C ∞ structures on M, N1 , N2 , respectively, and set
Ae := {(U × V, (x, y)) : (U, x) ∈ A1 , (V, y) ∈ A2 }.
By part (a), Ae is a C ∞ atlas on N1 × N2 . Our task is to prove that for any
charts (W, z) ∈ A, (U, x) ∈ A1 and (V, y) ∈ A2 , setting
W ′ := W ∩ (f, g)−1 (U × V ),
the map
(19) (x, y) ◦ (f, g) ◦ z −1 : z(W ′ ) → (x, y)(U × V ) ⊂ Rd1 +d2
is C ∞ . Now we compute:
(20) (x, y) ◦ (f, g) ◦ z −1 = (x ◦ f ◦ z −1 , y ◦ f ◦ z −1 ),
or, in other words, for all α ∈ z(W ′ ) ⊂ Rd :
 
(x, y) ◦ (f, g) ◦ z −1 (α) = x(f (z −1 (α))), y(f (z −1 (α)))

in Rd1 × Rd2 = Rd1 +d2 . However the two maps x ◦ f ◦ z −1 and y ◦ f ◦ z −1

are C ∞ since f and g are C ∞ and A, A1 , A2 are C ∞ atlases; hence also
the map in (19), (20) is C ∞ (indeed, this is a basic fact about C ∞ maps
between open subsets of Euclidean spaces; cf. the argument at the end of
our solution to part (a)). This completes the proof. 

(d). This is immediate from parts (b) and (c) since the map in question
equals
(21) (f ◦ pr1 , g ◦ pr2 ),
where we use the “(·, ·)” notation from part (c), and pr1 , pr2 are the projec-
tion maps pr1 : M1 × M2 → M1 and pr2 : M1 × M2 → M2 .
[Detailed explanation (cf. also footnote 9 above): Write M := M1 × M2 ;
then f ◦ pr1 is a C ∞ map M → N1 , by part (b) and since any composition
of C ∞ maps is C ∞ . Similarly g ◦ pr2 is a C ∞ map M → N2 . Hence by part
(c), (f ◦ pr1 , g ◦ pr2 ) is a C ∞ map M → N1 × N2 . And this is indeed the
map which we are interested in, since for every (p, q) ∈ M1 × M2 = M we
have:
(f ◦ pr1 , g ◦ pr2 )(p, q) = (f ◦ pr1 (p, q), g ◦ pr2 (p, q)) = (f (p), g(q)).
Done!] 
60 ANDREAS STRÖMBERGSSON

Problem 9.
(a). ∼ is reflexive since Id ∈ Γ. To see that ∼ is symmetric, assume
p ∼ q; then there is γ ∈ Γ s.t. γ(p) = q; but then γ −1 ∈ Γ and γ −1 (q) = p;
hence q ∼ p. Finally let us prove that ∼ is transitive. Assume p ∼ q and
q ∼ r. Then there exist γ, γ ′ ∈ Γ such that γ(p) = q and γ ′ (q) = r. Then
γ ′ γ(p) = r, and γ ′ γ ∈ Γ. Hence ∼ is transitive. Done! 
(b). The fact that the definition gives a topology on Γ\M is immediate
if we note that “π −1 respects intersections and unions of sets”, i.e. for any
family {Uα } of subsets Uα ⊂ Γ\M we have π −1 (∪α Uα ) = ∪α π −1 (Uα ) and
π −1 (∩α Uα ) = ∩α π −1 (Uα ). (This is in fact a property of the inverse of any
map. Here we use it for finite intersections and arbitrary unions.) We also
use the fact that π −1 (∅) = ∅ and π −1 (Γ\M ) = M , both of which are open
in Γ\M .
Note also that it is immediate from the definition of the topology on Γ\M
that the projection map π : M → Γ\M is continuous.
We next prove that Γ\M is Hausdorff. This is considerably more difficult.
Thus consider two distinct, arbitrary points in Γ\M , say [p] and [q], where
p, q ∈ M . Since M is locally Euclidean we can choose open sets U, V ⊂ M
such that p ∈ U , q ∈ V , and U and V are compact. Now since Γ acts
properly discontinuously, the set

F := {γ ∈ Γ : γ(U ) ∩ V 6= ∅}

is finite. (Indeed, F is contained in {γ ∈ Γ : γ(K) ∩ K 6= ∅} for K := U ∪ V ,

and K is compact.) Now for each γ ∈ F we have γ(p) 6= q (since [p] 6= [q]);
hence since M is Hausdorff, there exist open sets Vγ , Wγ such that q ∈ Vγ ,
γ(p) ∈ Wγ , and Vγ ∩ Wγ = ∅. Set Uγ := γ −1 (Wγ ); then p ∈ Uγ , and Uγ is
open. Set
\  \ 
U1 := U ∩ Uγ ; V1 := V ∩ Vγ .
γ∈F γ∈F

Then U1 and V1 are open sets in M (since F is finite), and p ∈ U1 and

q ∈ V1 . We claim that

(22) ∀γ ∈ Γ : γ(U1 ) ∩ V1 = ∅.

To prove this, assume the opposite, i.e. γ(U1 ) ∩ V1 6= ∅ for some γ ∈ Γ.

Using U1 ⊂ U , V1 ⊂ V , and the definition of F , it follows that γ ∈ F . Using
U1 ⊂ Uγ , V1 ⊂ Vγ it then follows that γ(Uγ ) ∩ Vγ 6= ∅, i.e. Wγ ∩ Vγ 6= ∅,
contradicting our choice of Vγ , Wγ . Hence (22) is proved.
Now set

(23) U2 := π(U1 ); V2 := π(V1 ).

 PROBLEMS; “RIEMANNIAN GEOMETRY” 61

These are open subsets of Γ\M ! (Proof: One verifies that π −1 (U2 ) =
∪γ∈Γ γ(U2 ); and this is a union of open sets, hence open (in M ). There-
fore U2 is open in Γ\M . Similarly U2 is open in Γ\M .) Furthermore, (22)
implies that U2 ∩ V2 = ∅. Hence we have proved that Γ\M is Hausdorff.
Next we prove that Γ\M is connected; in fact we prove that Γ\M is path-
connected (this trivially implies connectedness). Consider any two points
in Γ\M , say [p] and [q] with p, q ∈ M . Since M is path-connected (cf.
Problem 1), there is a curve γ : [0, 1] → M with γ(0) = p and γ(1) = q. But
then π ◦ γ : [0, 1] → Γ\M is a curve from π(p) = [p] to π(q) = [q]. Hence
Γ\M is path-connected.
Next we will prove that Γ\M is locally Euclidean. We say that a subset
U ⊂ M is injectively embedded in Γ\M if the restriction π|U is injective.
Let I be the family of open sets in M which are injectively embedded in
Γ\M . Let us prove:
(24)
 
∀U ∈ I : π(U ) is open and π|U is a homeomorphism from U onto π(U ) .
Take U ∈ I. Clearly π|U is a bijection of U onto π(U ). We have also noted
that π is continuous. Furthermore π is an open map, i.e. maps any open
subset of M to an open subset of Γ\M ; this is shown by the argument
below (23). Using these facts it follows that π(U ) is open and π|U is a
homeomorphism from U onto π(U ), i.e. (24) is proved. Next we claim:
[
(25) I cover M ; that is, U = M.
U ∈I

[Proof: Let p ∈ M . By a slight modification of the construction we used

when proving that Γ\M is Hausdorff, we are going to construct an open
neighborhood of p in M which is injectively embedded in Γ\M . Choose an
open set U ⊂ M containing p such that U is compact. Then the set
F := {γ ∈ Γ : γ(U ) ∩ U 6= ∅}
is finite, since Γ acts properly discontinuously on M . Take any γ ∈ F \ {Id}.
Then γ(p) 6= p, since Γ acts freely on M . Hence there exist open sets Uγ , Vγ
such that p ∈ Uγ , γ(p) ∈ Vγ , and Uγ ∩ Vγ = ∅. Set
 \ 
U1 := U ∩ (Uγ ∩ γ −1 (Vγ )) .
γ∈F \{Id}

Then U1 is an open set in M (since F is finite) and p ∈ U1 . We claim that U1

is injectively embedded in Γ\M , i.e. U1 ∈ I. Indeed, assume the opposite.
Then there exist two points q 6= q ′ ∈ U1 with [q] = [q ′ ], i.e. γ(q) = q ′ for
some γ ∈ Γ. We have γ 6= Id since q ′ 6= q. Also q ′ ∈ γ(U1 ) ∩ U1 , thus
γ(U1 ) ∩ U1 6= ∅ and so (using U1 ⊂ U ) γ ∈ F . But now by the definition
of U1 , q ∈ U1 implies q ∈ Uγ , and γ(q) = q ′ ∈ U1 implies q ∈ Vγ . Hence
Uγ ∩ Vγ = ∅, contradicting our choice of Uγ , Vγ . This proves that U1 ∈ I.]
62 ANDREAS STRÖMBERGSSON

Now to prove that Γ\M is locally Euclidean, consider an arbitrary point

in Γ\M , say [p] with p ∈ M . By (25) there is a set U ∈ I with p ∈ U . Also,
since M is a topological manifold, there is a chart (V, x) with p ∈ V . Then
also (U ∩ V, x|U ∩V ) is a chart on M . It follows from (24) that π|U ∩V is a
homeomorphism from U ∩ V onto the open set π(U ∩ V ) ⊂ Γ\M . Hence
x ◦ (π|U ∩V )−1 is a homeomorphism from π(U ∩ V ) onto an open subset of
Rd . The fact that every point [p] in Γ\M has such an open neighborhood
which is homeomorphic to an open subset of Rd proves that Γ\M is locally
Euclidean.
It now only remains to prove that Γ\M is paracompact. Since we have
proved that Γ\M is Hausdorff and locally Euclidean, it actually suffices to
prove that Γ\M is second countable. (Indeed, cf. the notes to Lecture #1,
with reference to math.stackexchange.com/questions/527642.) However this
is quite trivial, using the fact that M is second countable, and the fact that
π : M → Γ\M is open and continuous. Indeed, let U be a countable base of
M (as a topological space). Set
U ′ = {π(U ) : U ∈ U }.
This is a countable family of open sets in Γ\M , since π is open. We claim
that U ′ is a base for Γ\M . To prove this, take an arbitrary open set V ⊂
Γ\M . Then π −1 (V ) is an open set in M , and since U is a base for M there
is a subfamily V ⊂ U such that π −1 (V ) = ∪U ∈V U . Applying π to each point
in this set identity we obtain π(π −1 (V )) = ∪U ∈V π(U ). But π(π −1 (V )) = V
(since π is surjective). Hence V = ∪U ∈V π(U ), and this means that V is a
union of certain sets in U ′ . Hence we have proved that U ′ is a (countable)
base for Γ\M , and hence Γ\M is second countable. 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 63

(c). Let A be the C ∞ structure on M . Set

A′ := {(π(U ), x ◦ (π|U )−1 ) : (U, x) ∈ A, U ∈ I}.
Using (24) we see that each element in A′ is a chart on Γ\M . We wish to
prove that A′ is an atlas on Γ\M , i.e. that the charts in A′ cover Γ\M .
For this consider an arbitrary point in Γ\M , say [p] with p ∈ M . By (25)
there is some U1 ∈ I with p ∈ U1 . Take any chart (U2 , x) ∈ A with p ∈ U2 .
Then also (U1 ∩ U2 , x|U1 ∩U2 ) ∈ A, since A is a maximal C ∞ atlas. But
U1 ∩ U2 ⊂ U1 and U1 ∈ I implies that U1 ∩ U2 ∈ I. Hence
(π(U1 ∩ U2 ), x|U1 ∩U2 ◦ (π|U1 ∩U2 )−1 ) ∈ A′ ,
and we have [p] ∈ π(U1 ∩ U2 ) since p ∈ U1 ∩ U2 . This completes the proof
that A′ is an atlas on Γ\M .
Next we prove that A′ is in fact a C ∞ atlas. Consider any two charts in
A′ , say
(26) (π(U ), x ◦ (π|U )−1 ) and (π(V ), y ◦ (π|V )−1 ),
for some (U, x), (V, y) ∈ A, U, V ∈ I. We have to prove that the two charts
in (26) are compatible, i.e. that the map
y ◦ (π|V )−1 ◦ π|U ◦ (x|U ∩V )−1 : x(U ∩ V ) → y(U ∩ V )
is C ∞ . However this map is equal to y ◦ (x|U ∩V )−1 , which we know is C ∞
since (U, x), (V, y) ∈ A and A is a C ∞ atlas. Hence A′ is indeed a C ∞ atlas
on Γ\M , and so determines a unique C ∞ structure on Γ\M (cf. Problem
4).
Finally note that for any (U, x) ∈ A with U ∈ I, the map π is represented
by
(x ◦ (π|U )−1 ) ◦ π ◦ x−1 : x(U ) → x(U )
with respect to the charts (U, x) ∈ A and (π(U ), x ◦ (π|U )−1 ) ∈ A′ . But the
above map is simply the identity map on x(U ), which of course is C ∞ . This
proves that the map π : M → Γ\M is C ∞ . 
64 ANDREAS STRÖMBERGSSON

Problem 10:
(a) If V is open (in M ) then V ∩ Uα is open in Uα for every α by definition
of the subspace topology of Uα .
Conversely, assume that V ∩ Uα is open in Uα for every α. By definition
of the subspace topology of Uα , this means that for each α there exists an
open set W ⊂ M such that V ∩ Uα = W ∩ Uα , and hence V ∩ Uα is open as
a subset of M . Now V = ∪α (V ∩ Uα ) since M = ∪α Uα ; thus V is a union
of open subsets of M and therefore V is itself an open subset of M . 
(b) Let T be the family of all “open” sets in a given “C ∞ fold” M . We
have to prove that (i) ∅ ∈ T , (ii) M ∈ T , and that T is closed under (iii)
arbitrary unions and under (iv) finite intersections:
(i) For every α ∈ A we have xα (∅ ∩ Uα ) = ∅ and this is an open subset of
Rd . Hence ∅ ∈ T .
(ii) For every α ∈ A we have xα (M ∩ Uα ) = xα (Uα ), which is an open
subset of Rd by our assumptions. Hence M ∈ T .
(iii) Let {Vβ }β∈B be an arbitrary family of sets in T . Then for every
α ∈ A,






xα ∪β∈B Vβ ∩ Uα = xα ∪β∈B Vβ ∩ Uα = ∪β∈B xα Vβ ∩ Uα ,


and here xα Vβ ∩ Uα
is an
open subset in Rd for every β ∈ B, since Vβ ∈ T .
Hence xα ∪β∈B Vβ ∩ Uα , being a union of open subsets of Rd , is itself an
open subset of Rd . This is true for every α ∈ A; hence ∪β∈B Vβ ∈ T .
(iv) Let {Vβ }β∈B be a finite family of sets in T . Then for every α ∈ A,






xα ∩β∈B Vβ ∩ Uα = xα ∩β∈B Vβ ∩ Uα = ∩β∈B xα Vβ ∩ Uα .


(The last equality holds since xα is injective.) Here xα Vβ ∩ Uα is an open



subset in Rd for every β ∈ B, since Vβ ∈ T . Hence xα ∩β∈B Vβ ∩ Uα ,
being a finite intersection of open subsets of Rd , is itself an open subset of
Rd . This is true for every α ∈ A; hence ∩β∈B Vβ ∈ T .
This completes the proof that T is a topology. 

Next we give examples showing that T is not always Hausdorff: Let

U′ ( U be non-empty open subsets of Rd and let M be the set
M := U ′ ⊔ ((U \ U ′ ) × {1, 2}).
M can be thought of as two copies of the set U , glued together along the
set U ′ .
For j = 1, 2 we define the subset Uj ⊂ M by
Uj := U ′ ⊔ ((U \ U ′ ) × {j}),
 PROBLEMS; “RIEMANNIAN GEOMETRY” 65

and let xj : Uj → U be the map defined by xj (p) = p for p ∈ U ′ , and

xj ((p, j)) = p for p ∈ U \ U ′ . Then xj is a bijection from Uj onto U , and
M = U1 ∪ U2 . Furthermore x1 (U1 ∩ U2 ) = x2 (U1 ∩ U2 ) = U ′ , an open subset
of U , and both the maps x2 ◦ x−1 −1
1 and x1 ◦ x2 are equal to the identity map
on U , which is C . Hence M with the family {(U1 , x1 ), (U2 , x2 )} is a C ∞
′ ∞

fold.
Now fix any point p ∈ U \ U ′ not lying in the interior of U \ U ′ ; such a
point certainly exists. (Indeed we can find such a point on any line segment
between a point in U ′ and a point in U \ U ′ .) Then every open subset V of
U containing p has nonempty intersection with U ′ . Now consider the two
points (p, 1) and (p, 2) in M . Let W1 , W2 be any two open subsets of M
such that (p, 1) ∈ W1 and (p, 2) ∈ W2 . Then for both j = 1, 2 we have
that xj (Wj ∩ Uj ) is an open subset of xj (Uj ) = U containing xj ((p, j)) = p.
Hence also x1 (W1 ∩ U1 ) ∩ x2 (W2 ∩ U2 ) is an open subset of U containing p,
and as we noted above this implies that this set has nonempty intersection
with U ′ , i.e. there exists a point
q ∈ U ′ ∩ x1 (W1 ∩ U1 ) ∩ x2 (W2 ∩ U2 ).
By the definitions of x1 , x2 it then follows that q ∈ W1 ∩ W2 . Thus we have
proved that for any two open subsets W1 , W2 of M subject to (p, 1) ∈ W1
and (p, 2) ∈ W2 , it holds that W1 ∩ W2 6= ∅. Hence M is not Hausdorff. 
(Compare Boothby [1, p. 59, Exc. 5]; note that the above shows that the
answer to that question is NO.)

(c) Assume that the stated criterion holds. Let p, q be two distinct points
in M . By assumption there is α ∈ A such that p, q ∈ Uα . Now xα (p) 6= xα (q)
since xα is a bijection, and hence (since Rd is Hausdorff) there exist two
disjoint open subsets W, W ′ ⊂ xα (Uα ) such that xα (p) ∈ W , xα (q) ∈ W ′ .
Then x−1 −1 ′ −1 −1
α (W ) and xα (W ) are disjoint, and p ∈ xα (W ), q ∈ xα (W ).
′

Hence if we can prove that x−1 −1 ′

α (W ) and xα (W ) are open in M then we
are done. By definition xα (W ) is open in M iff xβ (x−1
−1
α (W ) ∩ Uβ ) is open
in Rd for every β ∈ A. But note that

xβ (x−1 −1 −1
α (W ) ∩ Uβ ) = p ∈ xβ (Uα ∩ Uβ ) : xα ◦ xβ (p) ∈ W = ϕ (W ),

where ϕ := xα ◦ x−1
β : xβ (Uα ∩ Uβ ) → Rd . (We have that ϕ is a bijection
of xβ (Uα ∩ Uβ ) onto xα (Uα ∩ Uβ ).) By assumption ϕ is C ∞ , in particu-
lar continuous; hence since W is open also ϕ−1 (W ) is open, and we have
thus completed the proof that x−1 α (W ) is open in M . Of course the same
argument shows that xα−1 (W ′ ) is open in M . Done!

(Next we prove that the “partial converse”. Thus let M be a C ∞ manifold

and let p, q ∈ M . If p = q then the desired statement is trivial; hence from
now on we assume p 6= q. Then, since M is Hausdorff, there exist open sets
U1 , V1 ⊂ M with p ∈ U1 , q ∈ V1 and U1 ∩ V1 = ∅. Let (U, x) and (V, y) be
66 ANDREAS STRÖMBERGSSON

C ∞ charts on M with p ∈ U and q ∈ V . Then also (U ∩ U1 , x|U ∩U1 ) and

(V ∩ V1 , y|V ∩V1 ) are C ∞ charts on M , and after replacing (U, x) and (V, y)
with these, we have:
U ∩ V = ∅.
We may assume that x(p) 6= y(q); indeed otherwise replace y by the map
r 7→ v + y(r), V → Rd , where v is a fixed non-zero vector in Rd . Then we
can choose open sets (e.g. open balls) U ′ and V ′ in Rd such that x(p) ∈ U ′ ,
y(q) ∈ V ′ and U ′ ∩ V ′ = ∅. Now (x−1 (U ′ ), x|x−1 (U ′ ) ) and (y −1 (V ′ ), y|y−1 (V ′ ) )
are C ∞ charts on M , and after replacing (U, x) and (V, y) with these, we
have both
U ∩V =∅ and x(U ) ∩ y(V ) = ∅.
Now define the map z : U ∪ V → Rd by:
(
x(p) if p ∈ U
z(p) :=
y(p) if p ∈ V.
Using the fact that x(U ) and y(V ) are disjoint open subsets of Rd (and the
fact that x : U → x(U ) and y : V → y(V ) are homeomorphisms) it follows
that z is a homeomorphism from U ∪ V onto z(U ∪ V ) = x(U ) ∪ y(V ). Hence
(U ∪ V, z) is a chart on M , and one easily verifies that it is a C ∞ chart. This
C ∞ chart has the desired property, namely p, q ∈ U ∪ V !) 

(d) It is immediate from the definitions that Uα is open in M for every

α ∈ A. Now the only thing that has to be verified is that for every α ∈ A,
the map xα is a homeomorphism of Uα onto xα (Uα ) ⊂ Rd . Recall that
xα (Uα ) is open in Rd by assumption, and also xα is a bijection from Uα
onto xα (Uα ) ⊂ Rd . First let V be an arbitrary open subset of Uα ; then
by the definition of the topology on M , xα (V ) = xα (V ∩ Uα ) is an open
subset of xα (Uα ). This proves that xα is open. In order to prove that xα is
continuous, let W be an arbitrary open subset of xα (Uα ). Then we have to
prove that x−1
α (W ) is open in M . This is done by the argument in part (c).

 PROBLEMS; “RIEMANNIAN GEOMETRY” 67

Problem 11.
(a). By [12, Lemma 1.1.1] there exists a locally finite refinement V =
∞
P of U and C0 functions ψβ : M → [0, 1] with supp ψβ ∈ Vβ (∀β ∈ B)
(Vβ )β∈B
and β∈B ψβ (x) = 1 (∀x ∈ M ). Now since V is a local refinement of U ,
we can choose (using the axiom of choice, in general), for each β ∈ B, some
α(β) ∈ A so that Vβ ⊂ Uα(β) . Having made such a choice, we define, for
each α ∈ A:
X
ϕα := ψβ .
β∈B
(α(β)=α)

(The sum is taken over all β ∈ B which satisfy α(β) = α.) We claim that
these functions ϕα satisfy all the requirements in the problem formulation!
To prove this, let p be an arbitrary point in M . Then there is an open
neighborhood Ω ⊂ M of p such that the set
BΩ := {β ∈ B : Vβ ∩ Ω 6= ∅}
is finite. Now for p ∈ Ω we have
X
ϕα (p) = ψβ (p) (p ∈ Ω).
β∈BΩ
(α(β)=α)

In other words:
X
(27) ϕα|Ω = ψβ|Ω
β∈BΩ,α

where BΩ,α = {β ∈ BΩ : α(β) = α}. This says that, for every α ∈ A, ϕα|Ω
is a finite sum of C ∞ functions; hence ϕα|Ω is itself a C ∞ function. Since
every point p ∈ M has such a neighborhood Ω, we conclude that ϕα is C ∞
(∀α ∈ A).
Furthermore, from P the definition of ϕα , and the fact that each ψβ takes
values in [0, 1] and β∈B ψβ ≡ 1, it follows that ϕα (p) ∈ [0, 1] for all p ∈ M .
We also note that for every p ∈ M we have
X X X  X
(28) ϕα (p) = ψβ (p) = ψβ (p) = 1.
α∈A α∈A β∈B β∈B
(α(β)=α)

(The second equality follows by simply changing the order of summation;

this is permitted since all the
P terms are nonnegative, and the total sum is
convergent. In fact the sum α∈A ϕα (p) has only finitely many nonvanishing
terms; indeed for Ω as above we can have ϕα (p) > 0 only if α is in the finite
set {α(β) : β ∈ BΩ }.)
Now it only remains to prove that supp ϕα ⊂ Uα , ∀α ∈ A. To prove this,
fix α ∈ A, and fix an arbitrary point p ∈ M \ Uα . Take a neighborhood Ω
68 ANDREAS STRÖMBERGSSON

of p as above, i.e. so that the set BΩ is finite. Recall the formula (27). For
each β ∈ BΩ,α we have supp ψβ ⊂ Vβ ⊂ Uα(β) = Uα . Hence also
[
F := supp ψβ ⊂ Uα .
β∈BΩ,α

Also supp ψβ is a closed (even compact) subset of M for every β; hence since
BΩ,α is finite, the set F is also a closed subset of M . Hence
Ω′ := Ω \ F
is an open subset of M . Note that p ∈ Ω′ since p ∈ Ω, F ⊂ Uα and
p∈ / Uα . Also, by (27) and our definition of F , we have ϕα (q) = 0 for all
q ∈ Ω′ . Hence, since Ω′ is open, Ω′ is disjoint from supp ϕα , and in particular
p∈ / supp ϕα . To sum up, we have proved that every point p ∈ M \ Uα lies
outside supp ϕα . Hence supp ϕα ⊂ Uα , and we are done! 

(b). (We take the proof from [4, Lemma 9.5.2].)

Let us start from a partition of unity (ϕα )α∈A either as in [12, Lemma
1.1.1] or as in part (a). This Pmeans in particular that each ϕα is a C ∞
function M → [0, 1] and that α∈A ϕα (p) = 1 for all α. Also every point
p ∈ M has an open neighborhood Ω in M such that ϕα|Ω ≡ 0 for all except
finitely many α ∈ A (in the case of [12, Lemma 1.1.1] this is clear from the
statement, and in the case of part (a) it is a fact we noted in the proof; see
the text below (28)). Now set
X
(29) Φ(p) = ϕα (p)2 (p ∈ M ).
α∈A
P
Note that the “local finiteness” of the sum α∈A ϕα mentioned above im-
plies a similar local finiteness for the sum in (29), and in particular Φ ∈
C ∞ (M ) (i.e. Φ is a C ∞ Pfunction M → R). Furthermore for every p ∈ M
we have Φ(p) > 0, since α∈A ϕα (p) = 1 implies that there is at least one
α ∈ A with ϕα (p) > 0. Hence also p 7→ Φ(p)−1 is a C ∞ function on M , and
so the functions
ηα := Φ−1 · ϕ2α
are C ∞ , for every α ∈ A. It is also clear from the definition that each
function ηα takes values in R≥0 , and that supp ηα = supp ϕα . Furthermore,
for every p ∈ M :
X X
ηα (p) = Φ(p)−1 ϕα (p)2 = 1.
α∈A α∈A
(Hence also ηα (p) ∈ [0, 1] for all α ∈ A.) Hence the functions (ηα )α∈A sat-
isfy all the requirements which were imposed on (ϕα )α∈A , and furthermore
√
ηα = Φ−1/2 ϕα is a C ∞ function for every α ∈ A. 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 69

Problem 12.
(a) [We leave it to the reader to sort out certain details in the proof below,
hidden in phrases such as “passing to local coordinates”; “translation and
rotation”; etc; what we are doing there is creating a new C ∞ chart by
composing by appropriate diffeomorphism(s)...]
Passing to local coordinates we may assume M = Rn . After a rotation
and a scaling we may also assume ċ(s) = e1 := (1, 0, . . . , 0). Let us write
c(t) = (c1 (t), . . . , cn (t)); then c′1 (s) = 1 and c′j (s) = 0 for j ≥ 2. It follows
that there is ε > 0 such that c1 restricted to (s − ε, s + ε) is a diffeomorphism
onto an open interval I ⊂ R. Let α1 : I → (s − ε, s + ε) be the inverse
diffeomorphism. Then α1 (c1 (t)) = t for all t ∈ (s − ε, s + ε). Define
α : I × Rn−1 → (s − ε, s + ε) × Rn−1 ;
α(x1 , . . . , xn ) := (α1 (x1 ) − s, x2 , . . . , xn ).
Then α is a diffeomorphism of I × Rn−1 onto (−ε, +ε) × Rn−1 , and α(c(t)) =
(t−s, ∗, . . . , ∗) for all t ∈ (s−ε, s+ε). Hence after composing our coordinate
chart with α, we have c1 (t) = t−s for all t ∈ (s−ε, s+ε). Finally we consider
the map
β : (−ε, ε) × Rn−1 → (−ε, ε) × Rn−1
β(x1 , . . . , xn ) = (x1 , x2 − c2 (s + x1 ), . . . , xn − cn (s + x1 )).
Note that β is a C ∞ diffeomorphism of (−ε, ε) × Rn−1 onto (−ε, ε) × Rn−1 ;
indeed β is C ∞ and the inverse map is
(x1 , . . . , xn ) 7→ (x1 , x2 + c2 (s + x1 ), . . . , xn + cn (s + x1 )),
which is also C ∞ . Then
β(c(t)) = (t − s, 0, . . . , 0), ∀t ∈ (−ε, ε).
Hence by composing our coordinate chart with β, we obtain a coordinate
chart with the desired property! 

(b) Take ε > 0 and a chart (U, x) as in part (a). After possibly shrinking
U , we may assume that s + x1 (p) ∈ (a, b) for all p ∈ U . Define
h : U → R, h(p) := f (s + x1 (p)).
Then h is a C ∞ function and h(c(t)) = f (s + x1 (c(t))) = f (t) for all t ∈
(s − ε, s + ε). Now fix any open neighborhood U1 ⊂ U of c(s) having
compact closure U 1 in U . Then by Problem 7(d), there exists a C ∞ function
g : M → R which satisfies g|U1 ≡ h|U1 . By shrinking ε, we may assume that
c(t) ∈ U1 for all t ∈ (s − ε, s + ε). Then g(c(t)) = h(c(t)) = f (t) for all
t ∈ (s − ε, s + ε), and we are done. 
70 ANDREAS STRÖMBERGSSON

Problem 13:
(a). Recall that for a given C ∞ manifold M and a point p ∈ M , we
consider the set
(30)
S := {(U, x, u) : (U, x) is a chart on M with p ∈ U , and u ∈ Tx(p) (x(U ))},
(where Tx(p) (x(U )) := Rd ) and define the relation ∼ on S by
def
(U, x, u) ∼ (V, y, v) ⇔ u = d(x ◦ y −1 )y(p) (v).

We now prove that ∼ is an equivalence relation. For any (U, x, u) ∈ S we

have that x ◦ x−1 equals the identity map on x(U ) ⊂ Rd , thus the Jacobian
d(x◦x−1 ) is the identity map on Tx(p) (x(U )) = Rd , and so d(x◦x−1 )(u) = u.
Hence ∼ is reflexive.
Next to prove that ∼ is symmetric, assume (U, x, u) ∼ (V, y, v), i.e. u =
d(x ◦ y −1 )y(p) (v). Then
d(y ◦ x−1 )x(p) (u) = d(y ◦ x−1 )x(p) ◦ d(x ◦ y −1 )y(p) (v)
= d(y ◦ x−1 ◦ x ◦ y −1 )y(p) (v)
= d(1y(V ) )y(p) (v) = v.
(Explanation: For the second equality we used the chain rule, cf. p. 3 of
Lecture #2. In the last line, “1y(V ) ” is the identity map on the set y(V );
its differential at y(p) is of course the identity map on Ty(p) (y(V )) = Rd .)
Hence (V, y, v) ∼ (U, x, u). This proves that ∼ is symmetric.
Finally we prove that ∼ is transitive. Assume (U, x, u) ∼ (V, y, v) and
(V, y, v) ∼ (W, z, w), i.e. u = d(x ◦ y −1 )y(p) (v) and v = d(y ◦ z −1 )z(p) (w).
Then
u = d(x ◦ y −1 )y(p) ◦ d(y ◦ z −1 )z(p) (w)
= d(x ◦ y −1 ◦ y ◦ z −1 )z(p) (w)
= d(x ◦ z −1 )z(p) (w)
(here we again used the chain rule), and thus (U, x, u) ∼ (W, z, w). This
proves that ∼ is transitive.
Hence ∼ is an equivalence relation. 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 71

(b). Injectivity: Assume that u, v ∈ Rd give [(U, x, u)] = [(U, x, v)]. This
means that (U, x, u) ∼ (U, x, v), i.e. u = d(x ◦ x−1 )x(p) (v). But d(x ◦ x−1 )x(p)
is the identity map on Rd ; hence u = v. This proves that the given map is
injective.
Surjectivity: Consider an arbitrary element in Tp M ; we can always rep-
resent it as [(V, y, v)] for some (V, y, v) ∈ S (cf. (30)). Set
u = d(x ◦ y −1 )y(p) (v) ∈ Tx(p) (x(U )) = Rd .
Then
d(y ◦ x−1 )x(p) (u) = d(y ◦ x−1 )x(p) ◦ d(x ◦ y −1 )y(p) (v) = d1y(p) (v) = v,
and thus (U, x, u) ∼ (V, y, v), i.e. [(U, x, u)] = [(V, y, v)]. In other words, the
image of u under the given map equals [(V, y, v)]. This proves that the given
map is surjective. 
72 ANDREAS STRÖMBERGSSON

(c). Fix p ∈ M . Let T be a bijective linear map V → Rd . Then (M, T|M )

is a C ∞ chart on M . Consider the map
J : V → Tp M ; J(v) = [(M, T|M , T (v))].
It follows from part (b) that J is a bijection of V onto Tp M . We claim that
J is independent of the choice of T . To prove this, assume that also S is a
bijective linear map V → Rd . Then we need to prove that for every v ∈ V
we have [(M, T|M , T (v))] = [(M, S|M , S(v))] in Tp M . In other words (cf.
Def. 3 in Lecture #2), we need to prove
(31) S(v) = d(S ◦ T −1 )T (p) (T (v)), ∀v ∈ V.
Now we note the following very basic fact: “The differential of a linear map
is equal to the map itself”. More precisely: For any linear map L : Rd → Rn ,
and any x ∈ Rd , the differential dLx : Rd → Rn is equal to the map L itself.
We leave it to the reader to verify this fact; it is of course just a matter of
checking that the Jacobian matrix of L, exaluated at any point x, is equal
to the matrix of L itself.
Applying the fact just mentioned, with L = S ◦ T −1 : Rd → Rd , we
conclude that
d(S ◦ T −1 )T (p) (T (v)) = S ◦ T −1 (T (v)) = S(v),
i.e. we have proved (31)! Hence we have proved that our bijection J : V →
Tp M is independent of the choice of linear bijection V → Rd , i.e. the map J
is “canonically defined”. Therefore we can use this map J to identify Tp M
with V . 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 73

(d). Recall from Definition 4 (in Lecture #2) that we assume that f :
M → N is a C ∞ map between C ∞ manifolds, and p ∈ M . (Set d = dim M
and d′ = dim N .) Then dfp is defined to be the linear map from Tp M to
Tf (p) N which wrt any chart (U, x) on M with p ∈ U and (V, y) on N with
f (p) ∈ V is given by
dfp = d(y ◦ f ◦ x−1 )x(p) : Tx(p) (x(U )) → Ty(f (p)) (y(V )).
For this to make sense, recall that once the chart (U, x) is given, we can
identify Tp M with Tx(p) (x(U )) = Rd via the bijection v 7→ [(U, x, v)] from
Tx(p) (x(U )) onto Tp M (cf. p. 4 in Lecture #2 and part b of this problem);
′
similarly we can identify Tf (p) N with Ty(f (p)) (y(V )) = Rd . Thus the above
definition of dfp can be reformulated as saying that


dfp [(U, x, v)] := [(V, y, d(y ◦ f ◦ x−1 )x(p) (v))], ∀v ∈ Tx(p) (x(U )) = Rd .
This certainly makes dfp (α) defined for every vector α ∈ Tp M since every
α ∈ Tp M can be expressed as α = [(U, x, v)] for some v ∈ Tx(p) (x(U )). The
key issue is now to verify that dfp (α) does not depend on the above choice
of the charts (U, x) and (V, y)!
Thus assume that (Û , x̂) is also a chart on M with p ∈ Û and that (V̂ , ŷ)
is a chart on N with f (p) ∈ V̂ . Consider a fixed vector α ∈ Tp (M ); assume
that α is represented by v ∈ Rd wrt (U, x), and by d(x̂ ◦x−1 )x(p) (v) ∈ Rd wrt
(Û , x̂). Now the above definition says that dfp (α) is the vector in Tf (p) (N )
which is represented by
(32) d(y ◦ f ◦ x−1 )x(p) (v) ∈ Rd
wrt the chart (V, y), but also that dfp (α) is the vector in Tf (p) (N ) which is
represented by
(33) d(ŷ ◦ f ◦ x̂−1 )x̂(p) ◦ d(x̂ ◦ x−1 )x(p) (v) ∈ Rd
wrt the chart (V̂ , ŷ). Thus we have to prove that (32) and (33) represent
the same vector in Tf (p) (N ), i.e. that
d(y ◦ ŷ −1 )ŷ(f (p)) ◦ d(ŷ ◦ f ◦ x̂−1 )x̂(p) ◦ d(x̂ ◦ x−1 )x(p) (v)
= d(y ◦ f ◦ x−1 )x(p) (v).
However this is clear by the chain rule for the differential (for C ∞ maps
between vector spaces over R), using the fact that
(y ◦ ŷ −1 ) ◦ (ŷ ◦ f ◦ x̂−1 ) ◦ (x̂ ◦ x−1 ) = y ◦ f ◦ x−1 .
Done! 
74 ANDREAS STRÖMBERGSSON

(e). Fix a point p ∈ M1 ; then our task is to prove

(34) d(g ◦ f )p = dgf (p) ◦ dfp : Tp (M1 ) → Tg(f (p)) (M3 ).
Fix charts (U, x) on M1 , (V, y) on M2 , and (W, z) on M3 , satisfying p ∈ U ,
f (p) ∈ V , g(f (p)) ∈ W . With respect to these charts, d(g◦f )p is represented
by the map
d(z ◦ (g ◦ f ) ◦ x−1 )x(p) : Tx(p) (x(U )) → Tz(g(f (p))) (z(W )),
and dfp is represented by the map
d(y ◦ f ◦ x−1 )x(p) : Tx(p) (x(U )) → Ty(f (p)) (y(V )),
and dgf (p) is represented by the map
d(z ◦ g ◦ y −1 )y(f (p)) : Ty(f (p)) (y(V )) → Tz(g(f (p))) (z(W )).
Hence we will have proved (34) if we can prove
d(z ◦ (g ◦ f ) ◦ x−1 )x(p) = d(z ◦ g ◦ y −1 )y(f (p)) ◦ d(y ◦ f ◦ x−1 )x(p) .
However this is clear by the chain rule (for maps on Rd -spaces), since
z ◦ (g ◦ f ) ◦ x−1 = (z ◦ g ◦ y −1 ) ◦ (y ◦ f ◦ x−1 ).

 PROBLEMS; “RIEMANNIAN GEOMETRY” 75

(f). Fact #1: This is proved as follows:

 1
  v
−1 j ∂ ∂f ∂f  .  ∂f
v(f ) = dfp (v) = d(1R ◦ f ◦ x )x(p) (v )= · · · d  ..  = v j j .
∂xj ∂x1 ∂x ∂x
vd
(Explanation: In the first equality we use our definition of the directional
derivative; “v(f )”. In the second equality we use the definition of differen-
tial (Def. 4 in Lecture #2). In the third equality we use the definition of
differential for maps between Rd -spaces (Def. 2). The last equality is just
matrix multiplication. Note that in the last two expressions “f ” in fact
stands for the function f ◦ x−1 : x(U ) → R; this is in accordance with the
principle that we may identify U with x(U ) so long as the notation cannot
be misunderstood. Also in the last two expressions it is understood that the
partial derivatives are evaluated at the point x = x(p).)
Fact #2: This is proved as follows:
d
(f ◦ c)(t).
ċ(t)(f ) = dfc(t) (ċ(t)) = dfc(t) (dct (1)) = d(f ◦ c)t (1) =
dt
(In the first equality we use the definition of directional derivative; in the
second equality we use the definition of “tangent vector of a curve”; and
in the third equality we use the chain rule, cf. part (e) of this problem.
Finally the fourth equality could also be said to hold by the definition of
“tangent vector of a curve”; however since f ◦ c is a function from I ⊂ R
d
to R, “ dt (f ◦ c)(t)” has a more basic meaning as derivative of a real-valued
function on R, and of course these two interpretations are really the same
and give the same answer – as is easily verified by using the trivial “identity
map charts” on I and R.)
Fact #3: Using Fact #1 we have
∂(f g) ∂g ∂f
v(f g) = v j · j
= v j · f (p) · j + v j · g(p) · j
∂x |x=x(p) ∂x |x=x(p) ∂x |x=x(p)
= f (p) · v(g) + g(p) · v(f ),
proving the first formula. Next note that by the definition of directional
derivative the same formula can be written:
d(f g)p (v) = f (p) · dgp (v) + g(p) · dfp (v).
The fact that this holds for all v ∈ Tp M means that
d(f g)p = g(p) · dfp + f (p) · dgp
(equality of linear maps Tp M → R). 
76 ANDREAS STRÖMBERGSSON

Problem 14: Once one has gotten used to the machinery which we
have introduced, this problem is “completely obvious”. However, as a step
towards reaching such familiarity, it may be useful to work out a solution in
pedantic detail.
We have defined ċ(t) := dct (1). Furthermore, for any t ∈ I with c(t) ∈ U ,
the differential dct is, by definition, the map from Tt (I) = R to Tc(t) M which
with respect to the trivial chart (I, 1I ) on I and the chart (U, x) on M , is
represented by the linear map
d(x ◦ c ◦ 1−1 d
I )1I (t) : R → R .

This map equals d(x ◦ c)t , and so we get that ċ(t) := dct (1) is the vector in
Tc(t) M which with respect to the chart (U, x) is represented by
(35) d(x ◦ c)t (1) ∈ Rd .
But by the definition in the problem formulation,
x ◦ c(t) = (c1 (t), . . . , cd (t))
for all t ∈ I with c(t) ∈ U , and hence d(x ◦ c)t is the linear map given by
the (Jacobi) matrix
 1
∂c
 ∂t 
 . 
 ..  .
 
 ∂cd 
∂t
Of course “∂” can just as well be written “d” since each cj depends on
d j
only one variable; i.e. the entries of the above matrix are dt c (t) = ċj (t)
for j = 1, . . . , d. Applying the above linear map to the vector 1 ∈ R (so as
to evaluate the expression in (35)) we find10 that with respect to the chart
(U, x), ċ(t) is represented by
(36) (ċ1 (t), . . . , ċd (t)) ∈ Rd

∂
Next we turn to the right hand side of the desired formula, i.e. “ċj (t) j ”.
∂x
∂
Recall that by definition, at any point p ∈ U , is the tangent vector in
∂xj
Tp M which with respect to the chart (U, x) is represented by the standard
unit vector ej = (0, . . . , 1, . . . , 0) ∈ Rd (where the “1” is in the jth position).
∂
Hence, with respect to the chart (U, x), the tangent vector ċj (t) j ∈ Tc(t) M
∂x
is represented by
ċj (t)ej (t) = (ċ1 (t), 0, . . . , 0) + · · · + (0, . . . , 0, ċd (t)) = (ċ1 (t), . . . , ċd (t)).

10also recalling our convention that column matrices are identified with vectors
 PROBLEMS; “RIEMANNIAN GEOMETRY” 77

∂
This agrees with (36), i.e. we have proved that ċ(t) and ċj (t)
are repre-
∂xj
sented by the same vector in Rd (wrt (U, x)); hence they are equal, i.e. we
have proved the desired formula,
∂
ċ(t) = ċj (t) j ∈ Tp M.
∂x
Finally in order to verify that the two definitions of “tangent vector of a
curve” in Lecture #2 are consistent with each other, let us apply the above
to the special case M = an open subset of Rd . In this case we have the con-
vention that Tp M is identified with Rd for every p ∈ Rd , namely through the
representation of tangent vectors via the identity chart, (M, 1M ). Applying
the formula which we have proved above (in the form (36)) we conclude that
ċ(t) = (ċ1 (t), . . . , ċd (t)) in Tc(t) M = Rd .
Note that we have proved this formula starting from the general definition of
“tangent vector of a curve on a manifold”, and we now see that the formula
agrees with the concrete definition of “tangent vector of a curve in Rd ” which
we gave in Lecture #2 (p. 2). 
78 ANDREAS STRÖMBERGSSON

Problem 15:
(a). Cf., e.g., Boothby [1, Ch. 4.1] or Fieseler [5, Sec. 3]...
(b). Cf., e.g., Helgason, [8, Ch. 1.2.1]...
We here only give the easy part of the solution of part b: For any vector
field X ∈ Γ(T M ) and any f ∈ C ∞ (M ) we define Xf ∈ C ∞ (M ) by
(Xf )(p) = X(p)f, ∀p ∈ M.
In other words, by definition of directional derivative:
(Xf )(p) = dfp (X(p)) ∈ Tp R = R.
By definition of the differential df : T M → T R = R × R, the above formula
can also be expressed:
Xf = pr2 ◦ df ◦ X : M → R.
and this shows (via Problem 17(a)) that we indeed have Xf ∈ C ∞ (M ).
Let us note that for any X ∈ Γ(T M ), the map which we have now defined,
f 7→ Xf, C ∞ (M ) → C ∞ (M ),
is a derivation. (This is immediate from “Fact #3” on p. 9 in Lecture #2;
we prove this fact in Problem 13(f), and this fact also plays a crucial role in
part a of the present problem.)
Now it remains to prove that every derivation of C ∞ (M ) is obtained in
this way from some X ∈ Γ(T M ), and that any two distinct vector fields
yield distinct derivations....
 PROBLEMS; “RIEMANNIAN GEOMETRY” 79

Problem 16: As in the lecture, we define T M as a set to be the disjoint

union of all tangent spaces Tp M (p ∈ M ), and we let π : T M → M be
the projection map; π(w) = p for any w ∈ Tp M . Also, as a “proposed C ∞
atlas” on T M we take the set
A := {(T U, ϕx ) : (U, x) any C ∞ chart on M }
where T U = π −1 (U ) = ⊔p∈U Tp M and ϕx is the map
ϕx : T U → R2d = Rd × Rd ,
 
ϕx (w) = x(π(w)), dxπ(w) (w) .

Clearly for any C ∞ chart (U, x) on M , ϕx is a bijection from T U onto

x(U ) × Rd , which is an open subset of R2d ; and if also (V, y) is a C ∞ chart
on M then ϕx (T U ∩ T V ) = x(U ∩ V ) × Rd , which is also an open subset of
R2d , and as we verify in the lecture the map ϕy ◦ ϕ−1x : x(U ∩ V ) → R
2d is
∞
C . Hence all the conditions in Problem 10(b) are fulfilled, i.e. T M with the
family A is a “C ∞ fold”. In particular T M is now provided with a structure
of a topological space, namely a subset V ⊂ M is open iff ϕx (V ∩ T U ) is
open in R2d for every C ∞ chart (U, x) on M .
Now it suffices to prove that T M is Hausdorff, connected and paracom-
pact; for then it follows from Problem 10(d) that T M is a well-defined C ∞
manifold with A as a C ∞ atlas!
In order to prove that T M is Hausdorff, take two arbitrary points v, w ∈
T M . Then by the “partial converse” in Problem 10(c) (applied for our C ∞
manifold M ) there exists a C ∞ chart (U, x) on M such that π(v), π(w) ∈ U .
But then v, w ∈ T U , and so (T U, ϕx ) is a ’chart’ in A with v, w ∈ T U . The
fact that A contains such a chart for any pair of points v, w ∈ T M implies,
by Problem 10(c), that T M is Hausdorff!
Next we prove that T M is connected: Take any v, w ∈ T M . Let p = π(v)
and q = π(w). Let c1 : I → Tp M (I = [0, 1]) be any curve in the vector space
Tp M starting at v ∈ Tp M and ending at 0 ∈ Tp M . Note that the inclusion
map Tp M → T M is continuous. (We leave it as an exercise to verify this
fact; note that once T M has been proved to be a C ∞ manifold, the inclusion
map Tp M → T M can be seen to be C ∞ .) Hence c1 is continuous also as
a map from I to T M , i.e. c1 is a curve in T M . Similarly let c3 be any
curve in Tq M ⊂ T M going from 0 ∈ Tq M to w ∈ Tq M . Next, since
M is path-connected (cf. Problem 1), there is a curve e c2 : I → M going
from p to q. Note that the map f : M → T M taking any p ∈ M to the
vector 0 ∈ Tp M is continuous. (Again we leave this as an exercise.) Hence
c2 := f ◦ ec2 : I → T M is a curve in T M , going from the 0-vector in Tp M
to the 0-vector in Tq M . Now the “product curve” of c1 , c2 , c3 11 is a curve
11We will discuss this notion in Lecture #6; however it should hopefully be clear here
how the curve in question is constructed; just draw a picutre!
80 ANDREAS STRÖMBERGSSON

in T M going from v to w. The fact that such a curve exists for any two
v, w ∈ T M implies that T M is path-connected, and hence connected!
Finally we prove that T M is paracompact. Note that by what we have
already proved, T M is locally Euclidean and Hausdorff (cf. the solution to
Problem 10(d)), and the set A above is an atlas on T M . By Problem 2
it suffices to prove that T M has a countable atlas. Now fix any countable
atlas A′ on M (this exists by Problem 2). Then the following subset of A is
a countable atlas on T M :
{(T U, ϕx ) : (U, x) ∈ A′ }.

This completes the proof that T M is a C ∞ manifold with A as a C ∞

atlas.

We now turn to the last part of the problem, i.e. to prove that the map
π : T M → M is C ∞ . For this it suffices to prove that for any C ∞ chart
(V, y) on M and any (T U, ϕx ) ∈ A (thus: (U, x) is a C ∞ chart on M ), the
map
(37) y ◦ π ◦ ϕ−1
x : ϕx (T (U ∩ V )) → Rd
is C ∞ . However, the definition of ϕx says that, for any p ∈ U and w ∈
Tp U ⊂ T U :
ϕx (w) = (x(p), dxp (w)),
and thus
 
π ◦ ϕ−1
x x(p), dxp (w) = π(w) = p.
Hence
π ◦ ϕ−1 −1
x (z, v) = x (z), ∀(z, v) ∈ ϕx (T U ) = x(U ) × Rd ,
or, equivalently,
π ◦ ϕ−1
x =x
−1
◦ pr : x(U ) × Rd → M,
where pr is the projection pr : x(U ) × Rd → x(U ). Hence the map in (37)
equals y ◦ x−1 ◦ pr, and here y ◦ x−1 : x(U ∩ V ) → Rd is C ∞ since (U, x) and
(V, y) are C ∞ charts on M , and pr is obviously C ∞ . Hence the map in (37)
is C ∞ , and we are done. 

(Remark: Problem 36 gives a more general result.)

 PROBLEMS; “RIEMANNIAN GEOMETRY” 81

Problem 17:
(a). (When showing that df is C ∞ the key step is to verify that if U ⊂ Rd
′
is open and g : U → Rd is a C ∞ map, then the map (x, y) 7→ (g(x), dgx (y)),
′ ′
from U × Rd to Rd × Rd , is C ∞ .)
(b). Let p = π(X) ∈ M so that X ∈ Tp M ; then df (X) ∈ Tf (p) N . By our
definition of directional derivative,

df (X)(ϕ) = dϕ(df (X)) in Tϕ(f (p)) (R) = R.

Also by the same definition,

X(ϕ ◦ f ) = d(ϕ ◦ f )(X) in Tϕ(f (p)) (R) = R.

But d(ϕ ◦ f ) = dϕ ◦ df by the chain rule, and so the two expressions are
equal. 
(c). This is immediate from Problem 13(e). (Indeed, take v ∈ T M1 . Set
p := π(v); then v ∈ Tp M1 and now

d(g ◦ f )(v) = d(g ◦ f )p (v) = dgf (p) ◦ dfp (v) = dgf (p) (df (v)) = dg(df (v)),

where we used the formula from Problem 13(e) in the second equality.) 

Problem 18:
(a). Using the fact that (in the right hand side) h·, ·i is a scalar product
(viz., a positive definite symmetric bilinear form) on Tf (p) N , and dfp is a
linear map from Tp M to Tf (p) N , it follows that h·, ·i is a symmetric bilinear
form on Tp M which is positive semidefinite (viz., hv, vi ≥ 0 for all v ∈ Tp M ).
But using also the assumption that f is an immersion, i.e. dfp is injective for
each p, it follows that h·, ·i is in fact positive definite, i.e. a scalar product
on Tp M .
It remains to prove that h·, ·i depends smoothly on M . We leave the
details of this to the reader. 
(b). By definition
Z b
(38) L(f ◦ γ) = k(f ◦ γ)′ (t)k dt
a

(with the understanding that the integral has to be “splitted at each point
where γ is not C ∞ ). But here (f ◦ γ)′ (t) = d(f ◦ γ)t (1) = dfγ(t) ◦ dγt (1)
(where we used the def of directional derivative of a curve, and then the
82 ANDREAS STRÖMBERGSSON

chain rule), and so

q

′
k(f ◦ γ) (t)k = dfγ(t) ◦ dγt (1), dfγ(t) ◦ dγt (1)
q

(39) = dγt (1), dγt (1)
= kγ ′ (t)k,
where the second equality holds by our definition of h·, ·i on Tγ(t) M . Com-
bining (38) and (39) we obtain L(f ◦γ) = L(γ). The proof of E(f ◦γ) = E(γ)
is completely similar. 

(c). [Remark: In the inequality which we are going to prove,

d(p, q) ≥ d(f (p), f (q)),
of course “d” in the left hand side denotes the metric on M induced by the
Riemannian structure on M , and “d” in the right hand side denotes the
metric on N induced by the Riemannian structure on N . In the special
case when f is an inclusion map, so that M is as a set is a subset of N ,
one should give different names to these two metrics, e.g. “dM ” and “dN ”,
otherwise “d(p, q)” for p, q ∈ M is ambiguous!]
Let p, q ∈ M . By definition,

d(p, q) = inf L(γ) : γ : [a, b] → M is a piecewise C ∞ curve with

(40) γ(a) = p, γ(b) = q .
and

d(f (p), f (q)) = inf L(c) : c : [a, b] → N is a piecewise C ∞ curve with

(41) c(a) = f (p), c(b) = f (q) .
However for any curve γ satisfying the conditions in the right hand side of
(40), c := f ◦ γ is a piecewise C ∞ curve with c(a) = f (γ(a)) = f (p) and
c(b) = f (γ(b)) = f (q); thus c satisfies the conditions in the right hand side
of (41). Also L(c) = L(γ), by part (b). Hence every number L(γ) appearing
in the set in the right hand side of (40) also appears in the set in the right
hand side of (41); therefore the infinimum in (40) is ≥ the infinimum in
(41), i.e. d(p, q) ≥ d(f (p), f (q)), qed.
Example with strict inequality: Note that this is the usual situation! For
example take M = S d−1 , N = Rd and let f be the inclusion map. Then
dM (p, q) > dN (f (p), f (q)) for any points p 6= q ∈ M . 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 83

Problem 19:
(a). Let p, q ∈ M be given. By Problem 1 there exists a (continuous)
curve γ : [0, 1] → M with γ(0) = p and γ(1) = q. Let F be the family of
open subintervals I ⊂ [0, 1] 12 such that c(I) is contained in some C ∞ chart
on M . Note that F covers I. Hence since I is compact, there is a finite
subfamily F1 ⊂ F which covers I. In particular some I ∈ F1 must contain 0;
among all such intervals I ∈ F1 we pick the one which has the largest right
end-point; it is either [0, 1] or [0, t1 ) for some t1 ∈ (0, 1). In the latter case,
the point t1 must be contained in some interval in F1 not yet considered;
among all intervals in F1 containing t1 we pick the one which has the largest
right end-point; this interval is either of the form (t′1 , 1] or (t′1 , t2 ), for some
t′1 ∈ (0, t1 ) and t2 ∈ (t1 , 1). If it is of the form (t′1 , t2 ) then we consider all
intervals in F1 which contain t2 , etc. This process must eventually finish,
since F1 is finite, and this means that we have found a set of n ≥ 1 intervals
[0, t1 ), (t′1 , t2 ), (t′2 , t3 ), . . . , (t′n−1 , 1] in F1 ,
where 0 < t1 < t2 < . . . < tn−1 and 0 < t′j < tj for each j ∈ {1, . . . , n − 1}.
(If n = 1 then [0, 1] is in F1 and our set consists of this single interval.) By
the construction of F1 , there exist C ∞ charts (Uj , xj ) on M such that
γ([0, t1 )) ⊂ U1 , γ((t′1 , t2 )) ⊂ U2 , . . . , γ((t′n−1 , 1]) ⊂ Un .
Now for ε > 0 sufficiently small, if we set e t0 = 0, etj = tj − ε for j ∈
{1, . . . , n − 1}, and e
tn = 1, then
0=et0 < et1 < · · · < e
tn−1 < e
tn = 1
and t′j < e
tj < tj for j ∈ {1, . . . , n − 1}, so that
γ([e
tj−1 , e
tj ]) ⊂ Uj for j ∈ {1, . . . , n}.
Now we can define c : [0, 1] → M by letting, for each j ∈ {1, . . . , n}, c|[etj−1 ,etj ]
be the curve from γ(e tj−1 ) to γ(e
tj ) which in the chart (Uj , xj ) is represented
by a straight line segment from xj (γ(e tj−1 )) to xj (γ(e
tj )) (parametrized by
a constant times arc length, say). Then c is a continuous curve, and each
restriction c|[etj−1 ,etj ] is C ∞ ; thus c is a piecewise continuous curve, and it has
c(0) = p and c(1) = q. Done! 

12Here by “open” we mean wrt the topology of [0, 1] induced by the topology of R; in
particular [0, x) and (x, 1] are open subintervals of [0, 1] for any x ∈ (0, 1), and also [0, 1]
itself is an open subinterval of [0, 1].
84 ANDREAS STRÖMBERGSSON

(b). Outline: With e t0 , . . . , e

tn and charts (Uj , xj ) as above, we can con-
struct c : [0, 1] → M by letting c|[et0 ,et1 ] be an arbitrary C ∞ curve from γ(e t0 )
e
to γ(t1 ) (e.g., the line segment used in (a)). Using “Borel’s Lemma” (cf.
wikipedia) and working in the chart (U2 , x2 ), one can then construct a C ∞
curve c|[et1 ,et2 ] from γ(e
t1 ) to γ(e t2 ) which has the properties that “all deriva-
tives at et1 match up; i.e. c is in fact C ∞ on all [e t0 , e
t2 ]. Then just repeat.

 PROBLEMS; “RIEMANNIAN GEOMETRY” 85

Problem 20:
(a). Note that we can cover the whole of H n with one natural C ∞ chart
(H n , y), namely by letting y(x) := (x1 , . . . , xn ) for x = (x0 , x1 , . . . , xn ) ∈
H n . Then y(H n ) = Rn and the inverse map is
p 
y −1 (x1 , . . . , xn ) = 1 + (x1 )2 + · · · + (xn )2 , x1 , . . . , xn ,
∀x = (x1 , . . . , xn ) ∈ Rn .

Note that this map y −1 gives the embedding map i : H n → Rn+1 , expressed
wrt our selected chart on H n and the standard chart on Rn+1 . Hence wrt
these charts, for each p = (x0 , x1 , . . . , xn ) ∈ H n , dip : Tp H n → Tp Rn+1 is
the linear map with matrix
 
∂ p 1 2 n 2
∂ p 1 2 n 2
 ∂x1 1 + (x ) + · · · + (x ) · · · ∂xn
1 + (x ) + · · · + (x ) 
 ∂ 1 ∂ 1 
 x ··· x 
 ∂x1 ∂x n 
 
 .. .. 
 . . 
 ∂ n ∂ n 
1
x ··· x
 1 0 ∂x  ∂xn
x /x x2 /x0 · · · xn /x0
 1 0 ··· 0 
 
 ··· 0 
= 0 1 .
 .. .. 
 . . 
0 0 ··· 1

P j
This map takes an arbitrary vector ξ = (ξ1 , . . . , ξn ) in Rn to ( nj=1 xx0 ξj , ξ1 , . . . , ξn )
Pn ∂
in Rn+1 . (In other words, dip maps n
j=1 ξj ∂xj in Tp H to the vector
Pn x j P
( j=1 x0 ξj ) ∂x∂ 0 + nj=1 ξj ∂x∂ j in Tp Rn+1 .) Hence our first task is to prove
j
that for any ξ ∈ Rn , ( xx0 ξj , ξ1 , . . . , ξn ) is orthogonal to p wrt the form h·, ·i,
i.e. that
n
X xj
− ξj · x0 + ξ1 · x1 + · · · + ξn · xn = 0.
x0
j=1

This is clear by inspection!

The next task is to prove that the restriction of the form I (from [12, p.
228(top)]) to Tp H n (or perhaps more accurately; to dip (Tp H n )) is positive
definite. (At present I do not understand Jost’s claim that this follows from
Sylvester’s theorem. Exactly which theorem is this?) Thus we have to prove
that the following expression is positive, for any p = (x0 , x1 , . . . , xn ) ∈ H n
86 ANDREAS STRÖMBERGSSON

and ξ ∈ Rn \ {0}:
X 
xj  ∂ ∂ X xj  ∂
n n
X n n
X ∂
I ξj + ξj j , ξj + ξj j
x0 ∂x0 ∂x x0 ∂x0 ∂x
j=1 j=1 j=1 j=1
Xn
xj 2 n
X
(42) =− ξj + ξj2 .
x0
j=1 j=1

However by Cauchy-Schwarz we have

X n
xj 2 X xj 2 X 2 (x0 )2 − 1 X 2 X 2
n n n n
ξ j ≤ ξ j = ξ j < ξj ,
x0 x0 (x0 )2
j=1 j=1 j=1 j=1 j=1

where we used the fact that p = ∈(x0 , x1 , . . . , xn ) H n, and then used ξ 6= 0.

This shows that the expression in (42) is positive! 

(b) By definition, O(1, n) is the set of linear maps R : Rn+1 → Rn+1

which leave h·, ·i invariant, i.e. which satisfy
(43) hRx, Ryi = hx, yi, ∀x, y ∈ Rn+1 .
In particular if R ∈ O(1, n) and Rx = 0 for some x ∈ Rn+1 then hx, yi = 0 for
all y ∈ Rn+1 and this implies x = 0. Hence every R ∈ O(1, n) is invertible.
Setting now x = R−1 x′ and y = R−1 y ′ (with arbitrary x′ , y ′ ∈ Rn+1 ) in
the relation (43) we get hx′ , y ′ i = hR−1 x′ , R−1 y ′ i; hence R−1 ∈ O(1, n).
Hence O(1, n) is closed under taking inverse. The rest of the verification
that O(1, n) is a group is immediate.
Let us put
e n := {x ∈ Rn+1 : hx, xi = −1, x0 < 0},
H
so that the set
(44) {x ∈ Rn+1 : hx, xi = −1}
equals the disjoint union of H n and He n . It follows directly from that defi-
nition of O(1, n) that every R ∈ O(1, n) maps the set (44) onto itself; hence
since R is linear and invertible, R must map every connected component
of the set (44) onto the same or another connected component, in such a
way that the connected components are permuted. In other words: Every
R ∈ O(1, n) satisfies either
“(+)”: e n) = H
[R(H n ) = H n and R(H e n]
or
“(−)”: e n and R(H
[R(H n ) = H e n ) = H n ].

Let O+ (1, n) be the set of those R ∈ O(1, n) satisfying “(+)”. One verifies
immediately that O + (1, n) is closed under multiplication and inverses; thus
O + (1, n) is a subgroup of O(1, n). Next note that there exists R ∈ O(1, n)
 PROBLEMS; “RIEMANNIAN GEOMETRY” 87

which satisfies “(−)”; for example R = R0 := the diagonal matrix with

diagonal entries −1, 1, 1, . . . , 1. We now see that O(1, n) is the disjoint union
of the two cosets O + (1, n) and R0 · O+ (1, n) (where the latter coset consists
exactly of all R ∈ O(1, n) satisfying “(−)”). Hence O+ (1, n) is a subgroup
of O(1, n) of index 2 (and hence normal).
It now only remains to prove that each R ∈ O+ (1, n) acts by an isometry
on H n . Thus fix R ∈ O+ (1, n). Since H n is an embedded submanifold of
Rn+1 , and R is a linear (hence C ∞ ) map Rn+1 → Rn+1 preserving H n , it
follows that R|H n is a C ∞ map H n → H n . Considering also R−1 ∈ O + (1, n)
we see that R|H n is in fact a C ∞ diffeomorphism of H n onto H n , with inverse
= (R−1 )|H n . Note that for any x ∈ Rn+1 , if we identify Tx Rn+1 with Rn+1
in the standard way, then the bilinear form I on Tx Rn+1 equals the form
h·, ·i on Rn+1 . Now since R preserves the latter form, and dR = R (since R
is linear), it follows that dR preserves I, i.e.



I (dR)x (v), (dR)x (w) = I v, w , ∀x ∈ Rn+1 , v, w ∈ Tx Rn+1 .
In particular this holds for all x ∈ H n and v, w ∈ Tx H n ⊂ Tx Rn+1 ; and
this shows that R|H n preserves the Riemannian metric on H n . Hence R is
indeed an isometry of H n onto itself! 

(c). Take p ∈ H n and v ∈ Tp H n , v 6= 0. (Here we view Tp H n as a linear

subspace of Rn+1 ; cf. part (a).) As we proved in part (a), we then have
(45) hp, vi = 0.
In fact p, v are linearly independent. [Proof: p 6= 0 since p ∈ H n ; hence
we only need to prove that we cannot have v = tp for some t ∈ R. But
hp, pi = −1; hence v = tp would imply hp, vi = −t, and so t = 0 by (45),
contradicting v 6= 0.]
Let Π ⊂ Rn+1 be the 2-dimensional plane spanned by p, v. Because of
(45), if h·, ·i were a scalar product on Rn+1 , then
hx, pi hx, vi
(46) P : Rn+1 → Rn+1 , P (x) := p+ v
hp, pi hv, vi
would be the orthogonal projection from Rn+1 onto Π, and
(47) R : Rn+1 → Rn+1 , R(x) := 2P (x) − x
would be orthogonal reflection across Π. Now h·, ·i is NOT a scalar product
on Rn+1 (since it is not positive definite); however we still see that P and
R, as defined in (46) and (47), are well-defined linear maps on Rn+1 (indeed
recall that hp, pi = −1 and hv, vi > 0 by part (a)). Furthermore P (x) ∈ Π
for all x ∈ Rn+1 and P (x) = x for every x ∈ Π (for the last claim it suffices
to verify P (p) = p and P (v) = v). Hence also R(x) = x for all x ∈ Π, while
R(x) 6= x for x ∈ / Π (indeed R(x) = x ⇒ x = P (x) ∈ Π). In other words,
the set of fixed points of R equals Π.
88 ANDREAS STRÖMBERGSSON

It remains to prove R ∈ O+ (1, n). Let us first note that for any x ∈ Rn+1 ,
using (46) and (45) we have
hx, pi
hP (x) − x, pi = hp, pi + 0 − hx, pi = 0
hp, pi
and similarly hP (x) − x, vi; hence P (x) − x is orthogonal to p and v and
hence to all Π = SpanR {p, v}. In particular hP (x) − x, P (x)i = 0, since
P (x) ∈ Π. Therefore,
hRx, Rxi = hP (x) + (P (x) − x), P (x) + (P (x) − x)i
= hP (x), P (x)i + hP (x) − x, P (x) − xi
= hP (x) − (P (x) − x), P (x) − (P (x) − x)i
= hx, xi.
This holds for all x ∈ Rn+1 ; hence R ∈ O(1, n). Finally note that R(p) = p,
since p ∈ Π, and p ∈ H n ; hence in the notation of part (c) R cannot satisfy
“(−)” and so it must satisfy “(+)”, i.e. R ∈ O + (1, n). 

(d). Consider arbitrary p, v as in (c), but now also assume kvk = 1 (i.e.
hv, vi = 1). Let c : R → H n be the geodesic with c(0) = p, ċ(0) = v. (The
fact that c is defined on all R follows from the Theorem of Hopf-Rinow, since
H n is complete.) Take R ∈ O + (1, n) as in part (c); this is an isometry of
H n onto H n by part (b); hence also R ◦ c : R → H n is a geodesic on H n .
d
But R preserves p and v; hence R(c(0)) = p and dt R(c(t))|t=0 = v, and so
by [12, Thm. 1.4.2], R(c(t)) = c(t) for all t ∈ R. Also the set of fixed points
of R is Π; hence
c(t) ∈ Π, ∀t ∈ R.
Hence there are (uniquely determined) C ∞ functions x : R → R and y :
R → R such that
c(t) = x(t)p + y(t)v, ∀t ∈ R.
It follows from c(0) = p and ċ(0) = v that
x(0) = 1, ẋ(0) = 0; y(0) = 0, ẏ(0) = 1.
We also have c(t) ∈ H n and thus hc(t), c(t)i = −1, for all t ∈ R. Using
hp, pi = −1, hp, vi = 0 and hv, vi = 1, this translates into:
(48) −x(t)2 + y(t)2 = −1, ∀t ∈ R.
We also have kċ(t)k = 1, i.e. hċ(t), ċ(t)i = 1 for all t ∈ R, by [12, Lemma
1.4.5]; and this similarly translates into:
(49) −ẋ(t)2 + ẏ(t)2 = 1, ∀t ∈ R.
Equation (48) implies |x(t)| ≥ 1, and since
p x(0) = 1 and x is continu-
ous, it follows that x(t) ≥ 1 and x(t) = y(t)2 + 1 for all t ∈ R. Hence
 PROBLEMS; “RIEMANNIAN GEOMETRY” 89

y(t)ẏ(t)
ẋ(t) = p , and inserting this in (49) and simplifying we get
y(t)2 + 1
p
ẏ(t) = 1 + y(t)2 ∀t ∈ R.
Separating variables etc, this implies y(t) = sinh(C + t), ∀t ∈ R, where C is
a fixed real constant,
p and in fact C = 0 since y(0) = 0. Hence y(t) = sinh t
2
and so x(t) = y(t) + 1 = cosh t, i.e.
c(t) = (cosh t)p + (sinh t)v, ∀t ∈ R.

90 ANDREAS STRÖMBERGSSON

Problem 21:
Remark: The results which we prove here are special cases of correspond-
ing results on (maximal) integral curves of a vector field; cf. [1, Thm. IV.4.5].
Indeed, the geodesics are simply projections of the integral curves of a cer-
tain vector field on T M ; cf. [1, Thm. 7.1] as well as [12, Thm. 2.2.3 and Def.
2.2.3].
(a). Let p ∈ M and v ∈ Tp M be given. Let I, J ⊂ R be any two open
intervals containing 0 and let f : I → M and g : J → M be two geodesics
both satisfying f (0) = p, f˙(0) = v, g(0) = p, ġ(0) = v. Let
I ∗ = {t ∈ I ∩ J : f (t) = g(t) and f˙(t) = ġ(t)}.
We claim that I ∗ = I ∩ J. Note that this implies that f and g together
define a geodesic on the interval I ∪ J!
[Proof of I ∗ = I ∩ J: Take any s ∈ I ∗ . Using I ∗ ⊂ I ∩ J and the
fact that I ∩ J is open, it follows that there exists some δ > 0 such that
(s − δ, s + δ) ⊂ I ∩ J, and so we have two well-defined geodesics
f1 , g1 : Iδ → M ; f1 (t) := f (s + t), g1 (t) := g(s + t).
(Here Iδ := (−δ, δ).) These satisfy f1 (0) = f (s) = g(s) = g1 (0) and f˙1 (0) =
f˙(s) = ġ(s) = ġ1 (0). Hence by the local uniqueness theorem for geodesics
(Theorem 1’ on p. 2 in Lecture #4), there is some δ′ ∈ (0, δ] such that
f1 (t) = g1 (t) for all t ∈ Iδ′ , and so f (t) = g(t) and f˙(t) = ġ(t) for all
t ∈ (s − δ′ , s + δ′ ), i.e. (s − δ′ , s + δ′ ) ⊂ I ∗ . The fact that I ∗ contains such
a neighborhood around every point s ∈ I ∗ implies that I ∗ is open. But also
I ∗ is a closed subset of I ∩ J; this follows from the definition of I ∗ and the
fact that f, f˙, g, ġ are continuous. Hence I ∗ is either empty or a connected
component of I ∩ J, i.e. I ∗ = ∅ or I ∗ = I ∩ J. However 0 ∈ I ∗ , i.e. I ∗ is
non-empty. Therefore I ∗ = I ∩ J.]
Using the property just proved, it follows that if we let I be the union of
the domains of all geodesic curves such as f and g above, then there is a
well-defined geodesic cv : I → M with cv (0) = p, ċv (0) = v, and it has the
desired property! 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 91

(b). Let v ∈ T M and s ∈ Iv . First note that

(50) θ(0, v) = ċv (0) = v.
Set
w := θ(s, v) ∈ T M.
Since cv : Iv → M is a geodesic with ċv (s) = ċw (0), it follows that the curve
γ : Iv − s → M , γ(t) := cv (s + t) is a geodesic with γ̇(0) = ċw (0); hence by
the defining property of the maximal geodesic cw we have Iv − s ⊂ Iw and
γ(t) = cw (t) for all t ∈ Iv − s. In other words:
(51) cv (s + t) = cw (t), ∀t ∈ Iv − s.
Hence also
(52) ċv (s + t) = ċw (t), ∀t ∈ Iv − s,
and applying this for t = −s we get ċv (0) = ċw (−s). Hence the curve
η : Iw + s → M , η(t) := cw (s − t) is a geodesic with η̇(0) = ċv (0); and so
by the defining property of the maximal geodesic cv we have Iw + s ⊂ Iv
(and η(t) = cv (t), ∀t ∈ Iw + s). Now Iw + s ⊂ Iv and Iv − s ⊂ Iw together
imply Iw = Iv − s. Hence we have proved all desired relations; indeed cf.
(50) and (52), and note that (52) can be expressed as θ(s + t, v) = θ(w, t),
∀t ∈ Iv − s = Iw . 
(c). For any v ∈ T M , let cv : Iv → M be the unique maximal geodesic
starting at v. Now define D as follows:
D := {v ∈ T M : 1 ∈ Iv }.
Then define the map exp : D → M by
exp(v) := cv (1).
Note that exp(v) is well-defined, since v ∈ D implies 1 ∈ Iv .
In order to prove that D and exp have the desired properties, let us first
note a basic scaling property. As we noted in the lecture, if t 7→ c(t) is any
d
geodesic then so is t 7→ c(λt) for any constant λ ∈ R, and dt c(λt) = λċ(λt)
everywhere. Using this fact one easily derives the following scaling formula
for the maximal geodesics: For any v ∈ T M and λ ∈ R,
(53) Iλv = λ−1 Iv and cλv (t) = cv (λt), ∀t ∈ Iλv .
(Explanation of notation: “λ−1 Iv denotes the open interval {λ−1 t : t ∈ Iv };
in the special case λ = 0 the formula should of course be interpreted to say
I0v = R.)
Now for any v ∈ T M and t ∈ R, note that t ∈ Iv holds iff 1 ∈ t−1 Iv , and
by (53) this holds iff 1 ∈ Itv , i.e. iff tv ∈ D (note that with the appropriate
interpretation this discussion is correct also when t = 0; in particular note
92 ANDREAS STRÖMBERGSSON

that D contains the zero vector from every tangent space Tp M ). We have
thus proved that for every v ∈ T M ,
(54) Iv = {t ∈ R : tv ∈ D}.
Also by (53),
cv (t) = ctv (1) = exp(tv), ∀t ∈ Iv .
Hence it “only” remains to prove that D is open and that our map exp is
C ∞.
A crucial ingredient for the remaining part of the proof is to translate the
formula “θ(θ(s, v), t) = θ(t + s, v)” from part (b) into a composition formula
for exp. Thus take any v ∈ T M and s ∈ Iv , and write q = cv (s) = exp(sv)
and w := θ(s, v) = ċv (s) ∈ Tq (v). By (54), the formula Iw = Iv − s proved
in part (b) can be equivalently expressed as
 
(55) ∀v ∈ T M : ∀s ∈ Iv : ∀t ∈ R : t · ċv (s) ∈ D ⇔ (t + s)v ∈ D .
For any s, t satisfying the condition in (55), by part (b) we have θ(w, t) =
θ(t + s, v), i.e. ċw (t) = ċv (t + s). Applying the projection T M → M this
implies cw (t) = cv (t + s), i.e. exp(tw) = exp((t + s)v). Hence:
(56) ∀v ∈ T M : ∀s ∈ Iv : ∀t ∈ Iv − s : exp(t · ċv (s)) = exp((t + s)v).

Let D ′ be the set of all v ∈ T M with the property that v has an open
neighborhood Ω ⊂ D and exp is C ∞ on Ω. Clearly D ′ is an open subset of
the interior of D, and exp|D′ is C ∞ . Our task is to prove that D ′ = D! The
local existence theorem for geodesics (Theorem 1 in Lecture #4) implies
that D ′ contains the zero section in T M , i.e. D ′ contains the zero vector
from every tangent space Tp M (p ∈ M ).
Next we claim that, as a consequence of (56), D ′ has the following prop-
erty:
∀v ∈ T M : ∀s ∈ Iv : ∀t ∈ Iv − s :
 
(57) If sv ∈ D ′ and t · ċv (s) ∈ D ′ then (t + s)v ∈ D ′ .
[Proof: Fix s ∈ Iv , t ∈ Iv − s and assume sv ∈ D ′ and t · ċv (s) ∈ D ′ . Since
exp is C ∞ on D ′ , the function
d 
u 7→ ċu (s) = exp(tu)
dt |t=s

is C ∞ on the set U := {u ∈ T M : su ∈ D ′ } (verify this claim as an exercise!),

and U is an open subset of T M ; also v ∈ U . In particular u 7→ t · ċu (s) is
continuous on U , and so
U ′ := {u ∈ U : ċu (s) ∈ D ′ } = {u ∈ T M : su ∈ D ′ and t · ċu (s) ∈ D ′ }
is also an open subset of T M . Note that v ∈ U ′ by our assumptions. Now
for any u ∈ U ′ we have su ∈ D ′ ⊂ D, thus s ∈ Iu (cf. (54)) and also
 PROBLEMS; “RIEMANNIAN GEOMETRY” 93

t · ċu (s) ∈ D ′ ⊂ D, which by (55) implies (t + s)u ∈ D, i.e. t ∈ Iu − s; hence

by (56) we conclude:
∀u ∈ U ′ : exp((t + s)u) = exp(t · ċu (s)).
Here the right hand side is the composition of the function u 7→ t · ċu (s) and
exp, and both these functions are C ∞ when u ∈ U ′ (by the discussion above
regarding u 7→ ċu (s), and since t · ċu (s) ∈ D ′ , and by the definition of D ′ ).
Hence u 7→ exp((t + s)u) is C ∞ on U ′ , and since U ′ is an open set containing
v, this proves that (t + s)v ∈ D ′ .]
Assume now that there is some v ∈ D \D ′ (we will prove that this leads to
a contradiction). Let p = π(v), so that v ∈ Tp M . Then {t ∈ [0, 1] : tv ∈ / D′ }
is a closed subset of [0, 1] which contains 1 but not 0; this implies that there
is a minimal t1 ∈ (0, 1] with t1 v ∈/ D ′ . Note that t1 ∈ Iv , since 1 ∈ Iv . Set
q = exp(t1 v), and let 0q be the zero vector in Tq M . Then 0q ∈ D ′ , since D ′
contains the zero section; also D ′ is open and ε · ċv (t1 − ε) tends to 0q in
T M as ε → 0; hence there is some ε ∈ (0, t1 ) such that
ε · ċv (t1 − ε) ∈ D ′ .
Now set s = t1 − ε ∈ (0, t1 ). Note that sv ∈ D ′ , by our choice of t1 . Hence
(57) applies with our s and t = ε, and implies that t1 v ∈ D ′ . This is a
contradiction, since we constructed t1 so that t1 v ∈ / D ′ ! The conclusion is
that the does not exist any v ∈ D \ D ; in other words D ′ = D, and we are
′

done! 
(d). (This is now more or less straightforward; cf., e.g., [1, Thm. 3.12].)
94 ANDREAS STRÖMBERGSSON

Problem 22:
Fix a chart (U, x) on M containing p; then (T U, dx) is a C ∞ chart on T U
(cf. Problem 16 and the end of Lecture #2); dx maps T U onto x(U ) × Rd .
Set DU := T U ∩ D and V = dx(DU ); this is an open subset of x(U ) × Rd .
Of course, (U × U, (x, x)) is a chart on M × M ; cf. Problem 8.
Let us write x0 := x(p); then 0p is represented by (x0 , 0) ∈ V ⊂ Rd × Rd
in our chart (T U, dx).
In the local coordinates described above, the function
G := (π, exp) : DU → M × M
takes the form
(58) V → Rd × Rd ; (x, v) 7→ (x, F (x, v)),

where F : V → Rd is the function exp composed with the appropriate chart

maps. Let us write F (x, v) = (F 1 (x, v), . . . , F d (x, v)). Then the Jacobian
matrix of the map in (58) is given by
 
1 0 ··· 0 0 ··· 0
 0 1 ··· 0 0 ··· 0 
 
 
 
 0 0 ··· 1 0 ··· 0 
 1 1 1 1 1
 ∂F ∂F ∂F ∂F ∂F .
 
 ∂x1 ∂x2 · · · ∂xd ∂v 1 · · · ∂v d 
 . .. .. .. .. 
 .. . . . . 
 
 ∂F d ∂F d ∂F d ∂F d ∂F d 
··· ···
∂x1 ∂x2 ∂xd ∂v 1 ∂v d
Note that this matrix has the structure of a 2 × 2 block matrix where each
block is a d × d matrix. It follows from the basic formula (d expp )0 = 1Tp (M )
(cf. the proof of Theorem 3 in Lecture #4 = Jost [12, Thm. 1.4.3]) that at
(x0 , 0), the lower bottom block is the d × d identity matrix. Hence at (x0 , 0)
the above 2d × 2d matrix is non-singular (in fact the determinant equals 1).
Hence we have proved that the differential dG0p : T0p (T M ) → Tp M ×Tp M
is non-singular. Therefore, by the Inverse Function Theorem, there is a open
neigborhood Ω ⊂ D of 0p such that G restricted to Ω is a diffeomorphism
onto an open subset G(Ω) of M × M . After shrinking Ω if necessary, we
may assume that Ω has the following form, for some r > 0 and some open
neighborhood U of p in M :
(59) Ω = ⊔q∈U Br (0q ) = {v ∈ T M : π(v) ∈ U and kvk < r}.
(Here we used the facts that any set of the form (59) is open in T M , and
these sets form a neighborhood basis of 0p ; we leave the verification of these
as an exercise.)
 PROBLEMS; “RIEMANNIAN GEOMETRY” 95

Now G−1 is a C ∞ map from G(Ω) onto Ω. For each q ∈ U , set

Wq := {u ∈ M : (q, u) ∈ G(Ω)};
this is an open subset of M . Also for each q ∈ U define the map
Hq : Wq → T M, Hq (u) = G−1 (q, u).
Then Hq is a C ∞ map, and using G ◦ G−1 = 1 and G = (π, exp) we have
π(Hq (u)) = q and exp(Hq (u)) = u for all u ∈ Wq , i.e. Hq in fact maps Wq
onto Tq M ∩ Ω = Br (0q ), and
exp ◦Hq = 1Wq and Hq ◦ exp|Br (0q ) = 1Br (0q ) .
Hence for every q ∈ U , exp|Br (0q ) is a diffeomorphism onto an open set
(namely Wq ) in M , i.e. we have proved Theorem 3’ in Lecture #4! 

(b). Set
G
U ′ := Br (0p ) ⊂ T M ;
p∈U

this is an open subset of T M (as is easy to verify from the definition of the
topology of T M ; cf. Problem 16), and by assumption we have U ′ ⊂ D. As
in part (a) let us consider the C ∞ map G := (π, exp), but this time with U ′
as domain of definition:
G := (π, exp) : U ′ → M × M.
Note that G(U ′ ) = V and it follows from our assumptions that G is a
bijection from U ′ to V and G−1 : V → U ′ is exactly the map (p, q) 7→
exp−1
p (q) which we are interested in. Hence if we can prove that G is a
diffeomorphism onto an open subset of M × M then we are done; and in
fact it suffices to prove that every point in U ′ has an open neighbourhood in
U ′ on which G restricts to a diffeomorphism onto an open subset of M × M .
By the Inverse Function Theorem, this will be ensured if we can prove that
dG is non-singular at every point in U ′ .
Thus consider an arbitrary point v ∈ U ′ ; set q := π(v) ∈ U so that
v ∈ Br (0q ) ⊂ Tq M . Working with local coordinates of the same type as in
part (a)13, dGv is expressed by a 2d × 2d matrix which again is naturally
viewed as a 2 × 2 block matrix where each block is a d × d matrix: The
upper left block is the d × d identity matrix and the upper right block is the
d × d zero matrix; also the lower right block is a non-singular d × d matrix,
since (d expq )v : Tv Tq M = Tq M → Texp(v) M is non-singular (this holds since
v ∈ Br (0q ) and expq|Br (0q ) is a diffeomorphism by assumption). This implies
that dGv is non-singular, and we are done. 

13we now leave some details to the reader...

96 ANDREAS STRÖMBERGSSON

Problem 23: By definition of the exponential map, for any vector v ∈

Tp M the curve
r
(60) x(t) = tv (for |t| < )
kvk
represents a geodesic in M wrt the chart (U, x). Assuming now kvk = 1 and
v ∈ V , the (t > 0 part of the) curve (60) takes the following form in the
chart (U ′ , y ◦ x):
(61) y(t) = (t, ϕ(v)) (0 < t < r).
Now, wrt the chart (U ′ , y ◦ x), let Γijk (y) be the Christoffel symbols and
(hij (y)) be the matrix representing the Riemannian metric; then by [11,
Lemma 1.4.4] we have
(62) ÿ i (t) + Γijk (y(t))ẏ j (t)ẏ k (t) = 0, ∀t ∈ (0, r), i ∈ {1, . . . , d},
and
1

(63) Γijk (y) = hiℓ (y) hjℓ,k (y) + hkℓ,j (y) − hjk,ℓ (y)
2
for all y in the coordinate range. We now follow the discussion in [11, p.
22(mid)–23(top)]. Inserting (61) in (62) gives
0 + Γijk (y(t))δj,1 δk,1 = 0,
i.e.
Γi11 (y(t)) = 0, ∀t ∈ (0, r), i ∈ {1, . . . , d}.
Let us write Ω for the coordinate range for y, i.e. Ω = (0, r) × ϕ(V ) ⊂ Rd .
Note that the previous argument applies to any fixed v ∈ V ; this means
that y(t) = (t, ϕ(v)) can take any value in Ω. Hence
Γi11 (y) = 0, ∀y ∈ Ω, i ∈ {1, . . . , d}.
By (63), this means that


hiℓ (y) 2h1ℓ,1 (y) − h11,ℓ (y) = 0, ∀y ∈ Ω, i ∈ {1, . . . , d}.
P
Multiplying this by hki (y) and adding over i (using di=1 hki (y)hiℓ (y) = δkℓ ),
we get:
(64) 2h1k,1 (y) − h11,k (y) = 0, ∀y ∈ Ω, k ∈ {1, . . . , d}.
In particular for k = 1 this implies h11,1 (y) ≡ 0, i.e.
∂
(65) h11 (y) = 0, ∀y ∈ Ω.
∂y 1
However, by the transformation rule for the Riemannian metric expressed
wrt the two charts (U ′ , y ◦ x) and (U, x), we have
∂xk ∂xℓ
(66) h11 (y) = gkℓ (x) (∀y ∈ Ω),
∂y 1 ∂y 1
 PROBLEMS; “RIEMANNIAN GEOMETRY” 97

and recalling
  x 
(67) (y 1 , . . . , y d ) = kxk, ϕ ; thus x = y 1 · ϕ−1 (y 2 , . . . , y d ),
kxk
and writing
z = ϕ−1 (y 2 , . . . , y d ) ∈ S 1 ⊂ Rd ,
we get:
(68) h11 (y) = z k z ℓ gkℓ (x) (∀y ∈ Ω).

Now for any fixed (y 2 , . . . , y d ) ∈ ϕ(V ), if we let y 1 → 0+ then x → 0

in Rd and thus gkℓ (x) → δkℓ by part (a) of this problem; meanwhile z =
ϕ−1 (y 2 , . . . , y d ) is fixed; hence from (68) we get
d
X
(69) lim h11 (y) = z k z ℓ lim gkℓ (x) = (z k )2 = 1.
y 1 →0+ x→0
k=1

But (65) implies that h11 (y) equals a constant as y 1 varies in (0, r) while
(y 2 , . . . , y d ) is kept fixed; now (69) says that this constant must be 1, and
so we have proved
(70) h11 (y) = 1, ∀y ∈ Ω.
Inserting this in (64), for k = j ≥ 2, we get
∂
(71) h1j (y) = 0, ∀y ∈ Ω, j ∈ {2, . . . , d}.
∂y 1
Next, using (67) and the analogue of (66) for h1j (y) (j ≥ 2), we have
∂ −1 2
h1j (y) = z k y 1 wℓ · gkℓ (x), with w = ϕ (y , . . . , y d ) ∈ Rd .
∂y j
If we fix (y 2 , . . . , y d ) ∈ ϕ(V ) and let y 1 → 0+ then z, w are fixed while
gkℓ (x) → δkℓ ; hence
lim h1j (y) = 0, ∀(y 2 , . . . , y d ) ∈ ϕ(V ) (fixed), j ∈ {2, . . . , d}.
y 1 →0+

Combining this with (71) gives

(72) h1j (y) = 0, ∀y ∈ Ω, j ∈ {2, . . . , d}.
From (70), (72) and the symmetry hkℓ ≡ hℓk , we see that
 
1 0 ··· 0
0 h22 (y) · · · h2d (y)
 
(hij (y)) =  .. .. ..  , ∀y ∈ Ω.
. . . 
0 hd2 (y) · · · hdd (y)
98 ANDREAS STRÖMBERGSSON

Note also that since we know that (hij (y)) is positive definite for every y ∈ Ω,
it follows that the (d − 1) × (d − 1) matrix
 
h22 (y) · · · h2d (y)
 .. .. 
 . . 
hd2 (y) · · · hdd (y)
is positive definite for every y ∈ Ω. 

Problem 24: Assume that γ is not a geodesic. Then there is some

t0 ∈ (a, b) such that either γ(t) is not C ∞ at t = t0 or γ does not satisfy
the geodesic ODE at t = t0 . Then for any t1 ∈ [a, t0 ) and t2 ∈ (t0 , b] the
restricted curve γ[t1 ,t2 ] fails to be a geodesic. Set
p = γ(t0 ).
By Theorem 4:3’ (viz., Theorem 3’ in Lecture #4; cf. Problem 22(a)) there
exists r > 0 such that for every q ∈ Br (p) we have Br (0q ) ⊂ Dq and
expq|Br (0q ) is a diffeomorphism onto an open set in M . Then by Theorem
4:4 we have, for every q ∈ Br (p):
(73) expq (Br (0q )) = Bq (r),
and for every v ∈ Br (0q ), any pw C ∞ curve in M from q to expq (v) which
is not a reparametrization of the curve c(t) = exp(tv), t ∈ [0, 1], has length
strictly larger than kvk = d(q, expq (v)).
Now choose t1 ∈ [a, t0 ) and t2 ∈ (t0 , b] so that |t1 −t0 | < r/2 and |t2 −t0 | <
r/2. Then d(γ(t1 ), γ(t2 )) ≤ L(γ[t1 ,t2 ] ) = t2 − t1 < r, since γ is parametrized
by arc length. Similarly d(γ(t1 ), p) < r/2, so that γ(t1 ) ∈ Br (p). Set
q = γ(t1 ); we have just noted that d(q, γ(t2 )) < r; hence by (73) there is a
(unique) v ∈ Br (0q ) such that γ(t2 ) = expq (v). Also note that γ[t1 ,t2 ] cannot
be a reparametrization of the geodesic c(t) = expq (tv), t ∈ [0, 1], since we
have from above that γ[t1 ,t2 ] is not a geodesic (also γ is parametrized by
arc length). Hence by the property mentioned just below (73), L(γ[t1 ,t2 ] ) >
d(γ(t1 ), γ(t2 )), and so we get a shorter curve from γ(a) to γ(b) by forming the
product path of γ[a,t1 ] and c and γ[t2 ,b] . This contradicts L(γ) = d(γ(a), γ(b)).
Hence γ is a geodesic. 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 99

Problem 25: The standard Riemannian metric on Rd is an example of

a complete metric.
In order to give a non-complete metric, consider e.g. any non-surjective
embedding of Rd in Rd , that is, an injective (C ∞ ) immersion i : Rd → Rd
which is not surjective. (Such immersions certainly exist; one example is
i(x) = (1 + kxk2 )−1/2 x.) Let U = i(Rd ); the Inverse Function Theorem
implies that U is an open subset of Rd . We provide U with its C ∞ manifold
structure as an open submanifold of Rd ; also equip U with the Riemannian
metric induced by the standard Riemannian metric on Rd ; we denote this
by h·, ·i as usual. The Inverse Function Theorem also implies that i is a C ∞
diffeomorphism of Rd onto U . Now equip Rd with the Riemannian metric
which makes i an isometry; let us denote this metric by [·, ·]. Thus for any
p ∈ Rd and v, w ∈ Rd ,
[v, w] := hdip (v), dip (w)i.
(In other words, [·, ·] is the Riemannian metric on Rd coming from identifying
Rd with the (open) submanifold U of Rd , cf. Problem 18; here the latter “Rd ”
is equipped with the standard Riemannian metric h·, ·i.) We know that U
with the Riemannian metric h·, ·i is not complete; hence since i is a surjective
isometry, Rd with the Riemannian metric [·, ·] is not complete. Done! 

Alternative (for the non-complete example): Equip Rd with any explicit,

sufficiently rapidly decaying Riemannian metric, for example (gij (x)) with
2
(74) gij (x) = δij e−2kxk (with kxk2 = (x1 )2 + · · · + (xd )2 ).
Note that each matrix entry gij (x) is a C ∞ function of x ∈ Rd and also
the matrix (gij (x)) is positive definite for every x ∈ Rd (since it is a posi-
tive multiple of the identity matrix); hence the formula indeed gives a Rie-
mannian metric on Rd . To prove that this metric is not complete, consider
the sequence of points p1 , p2 , . . . with pj := (j, 0, · · · , 0). For any integers
1 ≤ j ≤ k we have
(75) d(pj , pk ) ≤ e−j .
Indeed, consider the C ∞ curve c : [j, k] → M , c(t) := (t, 0, . . . , 0). This is
a curve from pj to pk and its length with respect to the Riemannian metric
(74) is
Z kp Z kp Z k Z ∞
−t2
e−t dt = e−j .
−2t2
L(c) = hċ(t), ċ(t)i dt = e dt = e dt <
j j j j

(Here we used the fact that t2 > t for all t > j ≥ 1.) This proves that
(75) holds, and (75) in turn implies that p1 , p2 , . . . is a Cauchy sequence in
Rd equipped with the Riemannian metric in (74). On the other hand the
sequence p1 , p2 , . . . does not converge to any point in Rd . (Recall here that
“convergence” is a topological notion, i.e. it depends only on the topology of
Rd and not on the metric metrizing it; cf. here also Lemma 2 in Lecture #3.)
Hence Rd equipped with the Riemannian metric in (74) is not complete. 
100 ANDREAS STRÖMBERGSSON

Problem 26:
(a). Let us first note that this is not an immediate consequence of the fol-
lowing fact from basic point set topology: “Every closed subset of a complete
metric space is itself a complete metric space”. Namely, in that statement
it is understood that the subset is endowed with the metric which is simply
the restriction of the metric on the larger space. This is not the case in our
situation; we typically have dM (p, q) 6= dN (p, q) for p, q ∈ M ; cf. Problem
18(c)!
However, as we’ll see, the completeness of M is still easy to prove...
Let p1 , p2 , . . . be a Cauchy sequence in (M, dM ). By Problem 18(c) we
have dN (pj , pk ) ≤ dM (pj , pk ) for any j, k; hence p1 , p2 , . . . is also a Cauchy
sequence in (N, dN ). Therefore, since N is complete there exists a (unique)
limit point p := limj→∞ pj in N . Recall that the last limit relation by
definition means that
(76) lim dN (pj , p) = 0.
j→∞
Since M is closed in N we have p ∈ M . But since M is an embedded
submanifold of N , the topology of M equals the subspace topology of M as
a subset of N ; therefore limj→∞ dN (pj , p) = 0 implies limj→∞ dM (pj , p) = 0,
14
i.e. p = limj→∞ pj in M .
This proves that M is complete. 

(b). ((E.g. there is an appropriate “isometric immersion” of (0, ∞) into

R2 with closed image; easy to draw a picture...))

14Some further explanation: Here we are using the fact that the relation “p =
limj→∞ pj ” (⇔ “pj → p”) only depends on the topology of the space which we are working
in, and not on the choice of metric metrizing that topology. For example, pj → p in N
holds iff for every open neighborhood U ⊂ N of p, there exists J ∈ Z+ such that pj ∈ U for
all j ≥ J (and this is equivalent to (76) holding for any metric metrizing N ’s topology).
Now in our situation we wish to prove that limj→∞ dM (pj , p) = 0, or equivalently that for
every open set U in M with p ∈ U , there exists J ∈ Z+ such that pj ∈ U for all j ≥ J.
Let such an open set U ⊂ M be given. Since M has the subspace topology as a subset
of N , there is an open set V in N such that U = M ∩ V . Next, since pj → p in N there
exists J ≥ Z+ such that pj ∈ V for all j ≥ J. But the points p1 , p2 , . . . all lie in M ; hence
pj ∈ V ∩ M = U for all j ≥ J. Therefore pj → p in M .
 PROBLEMS; “RIEMANNIAN GEOMETRY” 101

Problem 27:
(a). This continuity is clear from
(77) |d(p, q) − d(p′ , q ′ )| ≤ d(p, p′ ) + d(q, q ′ ), ∀p, p′ , q, q ′ ∈ X.
[Proof of (77): By the triangle inequality we have
|d(p, q) − d(p′ , q)| ≤ d(p, p′ ) and |d(p′ , q) − d(p′ , q ′ )| ≤ d(q, q ′ ).
Hence
|d(p, q) − d(p′ , q ′ )| ≤ |d(p, q) − d(p′ , q)| + |d(p′ , q) − d(p′ , q ′ )| ≤ d(p, p′ ) + d(q, q ′ ).
Done!]
(b). Take any q ∈ X. If d(p, q) < r then setting s = r−d(p, q) > 0 we have
Bs (q) ⊂ Br (p) (by the triangle inequality) and so q ∈/ ∂Br (p). If d(p, q) > r
then setting s = d(p, q) − r > 0 we have Bs (q) ∩ Br (p) = ∅ (again by the
triangle inequality) and so q ∈ / ∂Br (p). This proves the stated inclusion.
The set ∂Br (p) is closed since the boundary of any set (in any topological
space) is closed. The fact that the set {q ∈ X : d(p, q) = r} is closed is an
immediate consequence of the fact that the metric d is continuous.
Next assume that (X, d) is a Riemannian manifold. Take any q ∈ X
with d(p, q) = r. By the definition of “d”, there exists a sequence of pw
C ∞ curves γ1 , γ2 , . . . on X such that each γj starts at p and ends at q,
and ℓj := L(γj ) < r + j −1 ; also ℓj ≥ r. We may assume that each γj is
parametrized by arc length, and has domain [0, ℓj ] where ℓj = L(γj ). Take
J ∈ Z+ so large that J −1 < r, and for each j ≥ J set
qj := γj (r − j −1 ).
Note that γj|[0,r−j −1] is a curve of length r−j −1 from p to qj ; hence d(p, qj ) ≤
r − j −1 and qj ∈ Br (p). On the other hand γj|[r−j −1,ℓj ] is a curve of length
≤ ℓj − (r − j −1 ) < (r + j −1 ) − (r − j −1 ) = 2j −1
from qj to q; hence d(qj , q) < 2j −1 . Hence qj → q as j → ∞. This shows
that q is in the closure of the set Br (p). Also q ∈
/ Br (p) since d(p, q) = r.
Hence q ∈ ∂Br (p). We have thus proved that {q ∈ X : d(p, q) = r} ⊂ Br (p),
and we are done. 
(Remark: If r is so small that there exists some r ′ > r such that Br′ (0p ) ⊂
Dp and exp|Br′ (0p ) is a diffeomorphism onto an open set, then the identity
∂Br (p) = {q ∈ X : d(p, q) = r} is a trivial consequence of Theorem 4
in Lecture #4. This is the only situation which occurs in the proof of the
Hopf-Rinow Theorem.)
102 ANDREAS STRÖMBERGSSON

(c). (The same argument appears on [12, p. 36].) We have

d(p, q) ≤ d(p, p0 ) + d(p0 , q)
by the triangle inequality; hence it now suffices to prove the opposite in-
equality. Let γ : [a, b] → X be any pw C ∞ curve with γ(a) = p and
γ(b) = q. Since t → d(p, γ(t)) is a continuous function of t, and d(p, γ(a)) =
0, d(p, γ(b)) = d(p, q) > r, there must exist some t0 ∈ (a, b) such that
d(p, γ(t0 )) = r. By part (b) we then have γ(t0 ) ∈ ∂Br (p), and hence be-
cause of the way p0 was chosen,
d(γ(t0 ), q) ≥ d(p0 , q).
Therefore,
L(γ) = L(γ[a,t0 ] ) + L(γ[t0 ,b] ) ≥ d(p, γ(t0 )) + d(γ(t0 ), q) ≥ r + d(p0 , q).
Since this is true for every pw C ∞ curve from p to q, we have
d(p, q) ≥ r + d(p0 , q) = d(p, p0 ) + d(p0 , q),
and the proof is complete. 

Problem 28: The fact that every distance d(p, q) < R is realized by a
geodesic is proved by more or less exactly the same proof as the “key fact”
in the proof of the Hopf-Rinow Theorem; cf. Lecture #5, pp. 9–11. Indeed,
assume q ∈ M and r := d(p, q) < R. Now the proof in Lecture #5, pp. 9–11
applies to our situation, word by word. The only difference is that now the
geodesic
c(t) := expp (tV )
(introduced in the last line of p. 9) is not guaranteed to be defined for all
t ∈ R, but it is certainly defined for all t with |t| < R, since BR (p) ⊂ Dp ;
in particular c(t) is defined for all t ∈ [0, r], and these are the only t-values
which are ever considered in the proof.
Finally, BR (p) = expp (BR (0p )) is indeed an immediate consequence of
the above fact. Indeed, the above fact implies BR (p) ⊂ expp (BR (0p )). On
the other hand for every v ∈ BR (0p ), the geodesic t 7→ exp(tv), t ∈ [0, 1], is
a curve of length kvk from p to expp (v), so that d(p, expp (v)) ≤ kvk < R, i.e.
expp (v) ∈ BR (p). Hence also the opposite inclusion, expp (BR (0p )) ⊂ BR (p),
holds. Done! 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 103

Problem 29:
(Note that Jost mentions this generalization in the beginning of his proof
of [12, Thm. 5.8.1].)
The proof in the two cases (fixed endpoints versus closed curves) is very
similar, and we treat here only the first case. Thus let c : I → M be a curve.
Set p = c(0) and q = c(1), and let F be the family of all pw C ∞ curves
homotopic to c. Then pick a minimizing sequence (γn ) for arc length in F
Thus each γn is a pw C ∞ curve, and
lim L(γn ) = L0 := inf L(c0 ).
n→∞ c0 ∈F
Wlog, assume also
L(γn ) ≥ L0 + 1, ∀n.

Set R := L0 + 2 and
K := BR (p).
Note that by construction, all the curves γ1 , γ2 , . . . are contained in K. By
the Hopf-Rinow Theorem (Theorem 5:3, (1) ⇒ (2)), K is compact. Hence
the proof of Cor. 4.1 (an immediate application of Thm. 4.3’) extends to show
that there exists some r0 > 0 such that for every point p′ ∈ K, expp′ |Br0 (0p′ )
is a diffeomorphism onto an open set in M . By shrinking r0 if necessary, we
may assume r0 < 1.
We have already remarked that all curves γ1 , γ2 , . . . are contained in K;
in fact they are even contained in the smaller ball BL0 +1 (p), and hence since
r0 < 1 and R = L0 + 2, it follows that the whole neighborhood Br0 (p′ ) is
contained in K, for all p′ ∈ γn , any n. Hence all points ever considered in
the proof of Theorem 1 in Lecture #5 lie in K, and so the proof carries over
to our situation! 
104 ANDREAS STRÖMBERGSSON

Problem 30:
(a). Consider the C ∞ map
f : R3 → R, f (x, y, z) = y 2 + z 2 − e2x .
One verifies immediately that S = f −1 (0). Furthermore
 
df(x,y,z) = −2e2x 2y 2z ,
which has rank 1 for all (x, y, z) ∈ R3 . Hence by [12, Lemma 1.3.2] (with
the slightly more precise formulation in the notes to Lecture #2; note that
this is what Jost’s proof actually gives), every connected component of S in
R3 is a closed differentiable submanifold of R3 . Hence if we prove that S is
connected then it follows that S itself is a differentiable submanifold of R3 .
However the connectedness is clear from the parametrization of S given in
the statement of the problem. Indeed, set
g(x, α) := (x, ex cos α, ex sin α).
Then g is a continuous (even C ∞ ) function from R2 to S. 15 Consider two
arbitrary points in S; these can be expressed as g(x, α) and g(x′ , β) for some
x, x′ , α, β ∈ R. Then
c : [0, 1] → S, c(t) = g((1 − t)x + tx′ , (1 − t)α + tβ)
is a curve in S from g(x, α) to g(x′ , β). This proves that S is path-connected,
and thus connected. This completes the proof that S is a differentiable
submanifold of R3 .
Note also that S is closed; for example this follows from S = f −1 (0) and
the fact that f is continuous. 

15Indeed, by inspection g is a continuous function from R2 to R3 , and g(x, α) ∈ S for

all (x, α) ∈ R2 . Hence since S is a (disconnected) union of differentiable submanifolds of
R3 and thus the topology of S agrees with the subset topology from S ⊂ R3 , it follows
that g is continuous also as a function from R2 to S.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 105

(b). Set p1 := (x0 , −ex0 , 0) ∈ S, and let γ be the following C ∞ curve (half
circle) on S:
γ : [0, π] → S, γ(t) = (x0 , ex0 cos t, ex0 sin t).
This is a curve from p0 to p1 , and L(γ) = πex0 (by Problem 18(b)). Hence
d(p0 , p1 ) ≤ πex0 . By the Hopf-Rinow Theorem there exists a geodesic c from
p0 to p1 with
L(c) = d(p0 , p1 ) ≤ πex0 .
We can take c to be parametrized by arc length; thus kċk ≡ 1 and the domain
of c is the interval [0, L(c)]. Now let R be the reflection map (x, y, z) 7→
(x, y, −z). This is an isometry of R3 onto itself, and R(S) = S; hence R is
also an isometry of S onto itself. Therefore R maps any geodesic in S to a
geodesic in S, and in particular the curve
c := R ◦ c : [0, L(c)] → S
e
is a geodesic in S, with kėck ≡ 1 and L(ec) = L(c). We have R(p0 ) = p0 and
c is a geodesic from p0 to p1 , just like c. Take v, ve ∈ Tp0 S
R(p1 ) = p1 ; hence e
so that c(t) = expp0 (tv) and e v ) for t ∈ [0, L(c)]. Note that
c(t) = expp0 (te
kvk = ke
v k = 1,
since kck ≡ kė
ck ≡ 1. Furthermore,
v 6= ve,
since c 6≡ e
c. [Proof of c 6≡ e c: The second coordinate of c(t) is a continuous
function of t starting at ex0 and ending at e−x0 ; hence for some t ∈ (0, L(c))
this coordinate must equal 0. For this t we have c(t) = (x, 0, z) ∈ S for
some x, z ∈ R, and from the definition of S it follows that z 6= 0 and so
R(c(t)) 6= c(t), i.e. e
c(t) 6= c(t) for this t.] Now
expp0 (L(c)v) = c(L(c)) = p1 = e
c(L(c)) = expp0 (L(c)e
v ),
and L(c)v 6= L(c)e v , kL(c)vk = kL(c)e v k = L(c) ≤ πex0 . This proves that
x
for every r > πe , the function expp0 is non-injective on the open ball
0

Br (0p ) ⊂ Tp (S). Hence i(p0 ) ≤ πex0 . 

(See also alternative solution on the next page!)

106 ANDREAS STRÖMBERGSSON

Alternative: Set p1 := (x0 , −ex0 , 0) ∈ S. Let γ1 , γ2 be the following two

curves (half circles) on S:
γ1 , γ2 : [0, π] → S, γ1 (t) = (x0 , ex0 cos t, ex0 sin t);
γ2 (t) = (x0 , ex0 cos t, −ex0 sin t).
Both these are curves from p0 to p1 , and
L(γ1 ) = L(γ2 ) = πex0
(by Problem 18(b)). (Hence d(p0 , p1 ) ≤ πex0 .)
Now by Theorem 1 in Lecture #5, generalized to complete manifolds (cf.
Problem 29), there exist geodesics c1 and c2 homotopic to γ1 and to γ2 ,
respectively, and from the proof of that theorem we see that we can take
L(cj ) to be smaller than or equal to the length of any pw C ∞ curve in the
homotopy class of γj ; in particular
L(cj ) ≤ L(γj ) = πex0 (j = 1, 2).
We can take c1 , c2 to be parametrized by arc length; then there exist two
unit vectors v1 , v2 ∈ Tp0 S such that
cj (t) = expp0 (tvj ), t ∈ [0, L(cj )].
In particular
expp0 (L(cj )vj ) = p1 for j = 1, 2,
It follows that if we can only prove that v1 6= v2 , then
i(p0 ) ≤ max(L(c1 ), L(c2 )) ≤ πex0 ,
and the proof will be complete.
In order to prove v1 6= v2 , let us assume the opposite, v1 = v2 . This means
that c1 ≡ c2 , and so γ1 and γ2 are homotopic. However this is “obviously”
not the case! (Details: γ1 ≃ γ2 would imply γ1 · γ 2 ≃ γ2 · γ 2 ≃ p0 , the
constant curve at p0 . Now note that the map
F : S → S 1, (x, ex cos α, ex sin α) 7→ (cos α, sin α)
is well-defined and continuous, and it maps the loop γ1 · γ 2 to the loop
t 7→ (cos t, sin t), [0, 2π] → S 1 . Hence, composing any homotopy showing
γ1 · γ 2 ≃ p0 with F , we obtain that the loop t 7→ (cos t, sin t) in S 1 represents
the identity element in π1 (S 1 ). However this is not the case, as we discussed
in Lecture #6, and as is carefully proved in Hatcher, [7, Thm. 1.7].) 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 107

Problem 31:
The following solution is sketchy and leaves out several details.
By [7, Prop. 1.5] we are free to choose the basepoint x0 . Let us choose x0
so that it does not lie on any line between two points in {p1 , . . . , pn }. Let
rej be the ray starting at x0 and going through pj ; it follows from our choice
of x0 that the n rays re1 , . . . , ren are distinct. After renaming the points
p1 , . . . , pn we may assume that the rays re1 , . . . , ren are ordered in positive
direction. Now choose rays r1 , . . . , rn with startpoint x0 such that r1 lies
en
between ren and re1 , and rj for j ∈ {2, . . . , n} lies between rej−1 and rej . Let A
be the infinite open wedge between rn and r1 coontaining ren \{x0 }; similarly
ej be the infinite open wedge between rj and rj+1
for j ∈ {1, . . . , n − 1} let A
containing rej . Take ε > 0 small. (Specifically, ε should be smaller than the
distance between x0 and rj for each j.) Let Aj be the open ε-neighborhood
of A ej (viz., the set of points in R2 which have distance < ε to some point in
Aj ), but with the point pj removed. The reader is adviced to draw a picture
of the situation!
Now A1 , . . . , An are open and path-connected subsets of
X := R2 \ {p1 , . . . , pn }
with X = ∪nj=1 Aj , and Aj ∩ Ak is path-connected for all j, k ∈ {1, . . . , n};
hence van Kampen’s theorem can be applied with A1 , . . . , An ; in particular
the natural homomorphism
Φ : π1 (A1 , x0 ) ∗ · · · ∗ π1 (An , x0 ) → π1 (X, x0 )
is surjective. Note also that for any j 6= k ∈ {1, . . . , n}, the set Aj ∩ Ak
is simply connected (proof?) i.e. π1 (Aj ∩ Ak ) = {e}. Hence van Kampen’s
theorem implies that Φ is an isomorphism. Finally each Aj is homotopy
equivalent with S 1 (proof?); hence π1 (Aj ) ∼
= Z, and so we conclude that
π1 (X, x0 ) is a free group with n generators.
Generators: [γ1 ], . . . , [γn ], where γj is a loop that is contained in Aj and
goes one time around pj . 
108 ANDREAS STRÖMBERGSSON

Problem 32:
(a). Let us first prove that Mf is Hausdorff. Let p and q be two distinct
points in Mf. Then π(p), π(q) ∈ M . If π(p) 6= π(q), then since M is Haus-
dorff, there exist disjoint open sets U, V ⊂ M with π(p) ∈ U and π(q) ∈ V .
f, and p ∈ π −1 (U ) and
Then π −1 (U ) and π −1 (V ) are disjoint open sets in M
q ∈ π −1 (V ). On the other hand if π(p) = π(q), then by the definition of
“covering space” there is an open neighborhood U of π(p) = π(q) in M such
that π −1 (U ) can be written as a union π −1 (U ) = ⊔j∈J Uj , where for each
j ∈ J, Uj is an open set in M f and π|U is a homeomorphism of Uj onto
j
U , and the sets Uj (j ∈ J) are pairwise disjoint. Now p, q ∈ π −1 (U ) and
hence there are unique i, j ∈ J such that p ∈ Ui and q ∈ Uj . If i = j then
π(p) = π(q) and the fact that π|Ui is injective imply p 6= q, contrary to our
assumption. Therefore i 6= j, and now Ui and Uj are two disjoint open sets
in Mf with p ∈ Ui and q ∈ Uj . This proves that M f is Hausdorff.

Next we prove that M f is locally Euclidean (of dimension d). Let p be an

f
arbitrary point in M . Then π(p) ∈ M , and since π : M f → M is a covering
space, π(p) has an open neighborhood U in M such that π −1 (U ) is a union
f, each of which is mapped homeomorphically onto
of disjoint open sets in M
U by π. Exactly one of these open sets in M f contains p; call this open set
e f e
U ⊂ M . Thus π|Ue is a homeomorphism of U onto U . Furthermore, since M
is a d-dimensional topological manifold, π(p) has an open neighborhood V
in M which is homeomorphic to an open subset of Rd . It follows that also
W := U ∩ V is homeomorphic to an open subset of Rd ; let ϕ : W → Rd be
one such homeomorphism. Now W f := (π e )−1 (W ) is an open subset of U
e
|U
containing p, and π f is a homeomorphism of W f onto W . It follows that
|W
f d
f is a homeomorphism of W onto an open subset of R . The fact that
ϕ ◦ π| W
every point p ∈ Mf has such an open neighborhood W
f in M f proves that Mf
is locally Euclidean.
f is connected and second countable by assumption; hence also para-
M
compact (cf. the notes to Lecture #1).
f is a topological manifold of dimension d.
Hence M 

Remark: In fact the assumption that M f is second countable is redundant;

f
any connected covering space M of a topological manifold is automatically
second countable. This is a consequence of the fact that the fundamental
group π1 (M ) of any topological manifold M is countable; cf., e.g., Lee, [15,
Prop. 1.16]. (Once we know that π1 (M ) is countable, the fact that M f
is second countable is proved by fairly simple arguments using the theory
developed in [7, Ch. 1.3]; cf. also Problem 2(a) above.)
 PROBLEMS; “RIEMANNIAN GEOMETRY” 109

(In this connection, here’s an issue which for a moment had me confused: One might
think that the “obvious” map π from the Long Line L (cf. wikipedia) to the circle S 1 ≃ R/Z
makes L a covering space of S 1 ; but L is not second countable! The resolution to this
seeming paradox is that the map π : L → S 1 is in fact not continuous, and hence not a
covering map; this is discussed here.)
110 ANDREAS STRÖMBERGSSON

f is a topological manifold and dim M

(b). By part (a), M f = dim M = d,
say. Let A be the C ∞ structure on M , and set
Ae := {(U
e , x ◦ π) : (U, x) ∈ A and U
e is an open subset of M
f which is
mapped homeomorphically onto U by π}

Note that for every (U e , x ◦ π) ∈ A,

e x ◦ π is a homeomorphism of U e onto an
open subset of Rd . Furthermore for every p ∈ M f there is some (U
e , x◦π) ∈ Ae
such that p ∈ U e (this is proved by the same argument which we used to prove
that M is locally Euclidean in part (a)). Hence Ae is a topological atlas on
f
M . We claim that Ae is in fact a C ∞ atlas on M f. To show this it remains
∞
to prove C compatibility between the charts in A. e Thus let (Ue , x ◦ π) and
(Ve , y ◦ π) be two arbitrary elements in A; e set U := π(Ue ) and V := π(Ve ) so
that (U, x), (V, y) ∈ A. We have to prove that the map
e ∩ Ve ) → y ◦ π(U
(y ◦ π) ◦ (x ◦ π)−1 : x ◦ π(U e ∩ Ve )

e ∩ Ve ) = U ∩ V and the above map equals

is C ∞ . But note that π(U
y ◦ x−1 : x(U ∩ V ) → y(U ∩ V ),

which is C ∞ since (U, x), (V, y) ∈ A. Hence we have proved that Ae is a C ∞

f.
atlas on M
By Problem 4, Ae determines a (unique) C ∞ structure on M f. Let us prove
that Mf equipped with this C ∞ structure has the desired properties. First
we prove that π is C ∞ . Given p ∈ M f, take (U
e , x ◦ π) ∈ Ae with p ∈ U
e ; also
e
set U := π(U ), so that (U, x) ∈ A. Then wrt the charts (U , x ◦ π) and (U, x),
the map π is represented by
e ) → x(U ).
x ◦ π ◦ (x ◦ π)−1 : x ◦ π(U

But π(Ue ) = U , and we see that the last map is simply the identity map on
x(U ) ⊂ Rd , which of course is a C ∞ map. Hence π is C ∞ locally near p,
f, the map π is C ∞ .
and since this is true for all p ∈ M
Next we prove that every point p ∈ M has an open neighborhood with
the stated property. Let p ∈ M be given. We know that p has an open
neighborhood U in M such that π −1 (U ) is a union of disjoint open sets in
f, each of which is mapped homeomorphically onto U by π. We will prove
M
e be any one of the open sets
that any such U is in fact ok for us. Thus let U
f with the property that π e is a homeomorphism of U
in M e onto U . We claim
|U
e onto U . We proved above that π
that π e is in fact a diffeomorphism of U
|U
is C ∞ ; hence π|Ue is C ∞ , and it remains to prove that (π|Ue )−1 : U → U e is
e . Take (Ve , y ◦ π) ∈ Ae
C ∞ . Take any point q ∈ U and set qe := (π e )−1 (q) ∈ U
|U
with qe ∈ Ve , and set V := π(Ve ) so that (V, y) ∈ A. Set W := U ∩ V ; then
 PROBLEMS; “RIEMANNIAN GEOMETRY” 111

(W, y|W ) ∈ A (since A is a maximal C ∞ atlas on M ); also Wf := (π e )−1 (W )

|U
f
is mapped homeomorphically onto W by π and so (W , y|W ◦ π) ∈ A. e Now
f , y|W ◦ π), the map (π e )−1 is represented by
wrt the charts (W, y|W ) and (W |U

(y ◦ π) ◦ (π|Ue )−1 −1
◦ (y|W ) f ).
: y(W ) → y ◦ π(W
One verifies that this is simply the identity map on y(W ) ⊂ Rd , which of
course is a C ∞ map. Note also that q ∈ W . The fact that every point
q ∈ U has such an open neighborhood W in which (π|Ue )−1 is C ∞ , implies
f
that (π e )−1 is C ∞ . This completes the proof that our C ∞ structure on M
|U
has all the desired properties.
Finally we prove that the above C ∞ structure on M f is uniquely deter-
mined by the stated requirements. (This is more or less obvious, but it
becomes somewhat technical to write out the details – at least in the way
I’ve done it. I think it is the least important part of this problem...) Thus
let B be any C ∞ structure on the topological manifold M f which satisfies the stated re-
quirements; our task is then to prove that B is compatible with the C ∞ atlas M f. Let
e e e
(U , ϕ) be any chart in B and let (V , y ◦ π) be any chart in A; then our task is to prove
that the map
e ∩ Ve ) → y ◦ π(U
(y ◦ π) ◦ ϕ−1 : ϕ(U e ∩ Ve )
is a diffeomorphism,
Set V := π(Ve ) so that (V, y) ∈ A and π|Ve is a homeomorphism of Ve onto V . Set
f
W =U e ∩ Ve and W = π(W f ); then W
f is an open subset of Ve , W is an open subset of V ,
and π|W f
f is a homeomorphism of W onto W . (For nontriviality, assume W 6= ∅.) Also
f f , π◦y|W ) ∈ A.
(W , ϕ f ) ∈ B and (W, y|W ) ∈ A (since B and A are maximal); hence also (W e
|W
Our task is to prove that the map
(78) f ) → y(W )
(y ◦ π) ◦ ϕ−1 : ϕ(W
is a diffeomorphism, i.e. that both the map (78) and its inverse,
(79) f ),
ϕ ◦ (y ◦ π)−1 : y(W ) → ϕ(W
are C ∞ .
−1
Let p ∈ W and set pe := (π|W f) (p). By the requirement which we have imposed on
B, there is an open neighborhood Ω′ of p in M such that π −1 (Ω′ ) is a union of disjoint
f, each of which is mapped diffeomorphically (wrt B) onto Ω′ by π. Among
open sets in M
f, let Ω
these open sets in M e ′ be the one which contains pe. Then set Ω
e := Ω
e′ ∩ W
f ; it follows
e f
that Ω is an open subset of W which contains pe and which is mapped diffeomorphically
(wrt B) by π onto Ω := π(Ω e′ ∩ W
f ) which is an open subset of W containing p. The
last statement (together with (W f , ϕ f ) ∈ B and (W, y|W ) ∈ A) implies that both the
|W
maps y ◦ π ◦ ϕ −1 e
: ϕ(Ω) → y(Ω) and ϕ ◦ π −1 ◦ y −1 : y(Ω) → ϕ(Ω) e are C ∞ . Those
maps are restrictions of the maps (78) and (79), and the fact that any point p has such a
neighborhood Ω in W now implies that the two maps (78) and (79) are C ∞ , and we are
done. 
112 ANDREAS STRÖMBERGSSON

f has a C ∞ manifold structure which is uniquely deter-

(c). By part (b), M
mined by the stated requirements (since any isometry is a diffeomorphism).
f → M is an immersion (since locally it is a diffeomorphism);
Note that π : M
hence by Problem 18(a) there is a unique Riemannian structure on M f such
f and v, w ∈ Tp M
that hv, wi = hdπ(v), dπ(w)i for any p ∈ M f. It is clear from
part (b) that this Riemannian structure has the desired properties. 

(d). This is easily deduced by inspecting the solution to Problem 9(b).

Indeed, there we saw that Γ\M is a topological manifold and π : M → Γ\M
is a continuous map. Now consider an arbitrary point in Γ\M , say [p] with
p ∈ M . By (25) there is some U ∈ I (viz., an open set in M which is
injectively embedded in Γ\M ) with p ∈ U . By (24), π(U ) is an open set in
Γ\M and π|U is a homeomorphism from U onto π(U ). Now for every point
q ∈ M , the equivalence class [q] = π −1 (π(q)) consists exactly of the points
γ(q) (γ ∈ Γ), and these are pairwise distinct (since Γ acts freely on M ).
Hence π −1 (π(U )) is a disjoint union of the sets γ(U ) (γ ∈ Γ):
G
(80) π −1 (π(U )) = γ(U ).
γ∈Γ

Here for each γ ∈ Γ, γ(U ) is open in M , since γ is a homeomorphism;

furthermore π|γ(U ) is a homeomorphism of γ(U ) onto π(U ), since π|γ(U ) =
π|U ◦ (γ|U )−1 , i.e. a composition of two homeomorphisms. Hence (80) ex-
presses π −1 (π(U )) as a union of disjoint open sets in M , each of which is
mapped homeomorphically onto π(U ) by π. The fact that each point [p] in
Γ\M has such an open neighborhood π(U ) proves that π : M → Γ\M is a
covering space of Γ\M . 
Comments to part (d): Hatcher in [7, Prop. 1.40] proves a stronger re-
sult under a weaker assumption. Indeed, note that our assumption that Γ
acts freely and properly discontinuously on M implies that the action is a
“covering space action” in the terminology of [7, p. 72]; this implication is
seen in the solution to Problem 9(b); indeed it is equivalent to (25). (This
is also the content of [7, p. 81, Problem 23].) It is worth pointing out that
the condition that Γ < Homeo(M ) acts by a “covering space action” on M
does not guarantee Γ\M to be Hausdorff; cf. [7, p. 81, Problem 25].
 PROBLEMS; “RIEMANNIAN GEOMETRY” 113

Problem 33:
WLOG we assume U = M . (To see that this is really no loss of generality,
note that if we prove (a) ⇔ (b) ⇔ (c) in the special case U = M , then
the general case follows by applying that statement to the vector bundle
(E|U , π, U ).)
Thus our task is to prove that the following statements are equivalent:
(a) E is trivial;
(b) there is some ϕ such that (M, ϕ) is a bundle chart for E;
(c) there is a basis of sections in ΓE, i.e. sections s1 , . . . , sn ∈ ΓE such that
s1 (p), . . . , sn (p) is a basis of Ep for every p ∈ M .
Here (a) ⇔ (b) is immediate by inspecting the definitions. Indeed, by
definition E is trivial iff there is a bundle isomorphism ϕ : E → M × Rn , i.e.
a C ∞ diffeomorphism ϕ : E → M × Rn with pr1 ◦ϕ = π such that ϕx = ϕ|Ex
is a vector space isomorphism Ex → {x} × Rn for each x ∈ M . But this is
the same as saying that (M, ϕ) is a bundle chart for E.
(b) ⇒ (c): Let ϕ be such that (M, ϕ) is a bundle chart for E. Let
e1 , . . . , en be the standard basis of Rn . For each j ∈ {1, . . . , n} we define the
function sj : M → E by sj (x) = ϕ−1 (x, ej ); then sj is C ∞ and π ◦ sj = 1M ;
hence sj ∈ Γ(E). Now for each x ∈ M , since e1 , . . . , en is a basis for Rn
and ϕ−1 n
x is a vector space isomorphism {x} × R → Ex , it follows that
s1 (x), . . . , sn (x) is a basis of Ex . Hence s1 , . . . , sn form a basis of sections of
E.
(c) ⇒ (b): Assume that s1 , . . . , sn is a basis of sections of E. Let us
define the map ψ : M × Rn → E by
n
X
ψ(x, (c1 , . . . , cn )) := cj · sj (x) ∈ Ex .
j=1

Clearly ψ is C ∞ and π ◦ ψ = 1M . Furthermore ψ(x, ·) is a vector space

isomorphism Rn → Ex for each x ∈ M , since s1 (x), . . . , sn (x) is a basis of
Ex . It follows that ψ is a bijection of M × Rn onto E. Let
ϕ = ψ −1 : E → M × Rn .
It follows that ϕx := ϕ|Ex is a vector space isomorphism Ex → {x} × Rn for
each x ∈ M . It remains to prove that ϕ is a diffeomorphism. We already
know that ϕ−1 = ψ is C ∞ , so it suffices to prove that ϕ is C ∞ , and for this
it suffices to prove that every point in E has an open neighborhood in E in
which ϕ is C ∞ .
Thus let p0 ∈ E be given. Set x0 = π(p0 ). Choose a bundle chart (U, ϕ) e
for E with x0 ∈ U and also a chart (V, α) for M with x0 ∈ V . In fact we may
assume V = U , since otherwise we may replace (U, ϕ) e with (U ∩ V, ϕ e|U ∩V )
114 ANDREAS STRÖMBERGSSON

and replace (V, α) with (U ∩ V, α|U ∩V ). Thus from now on (U, ϕ)

e is a bundle
chart for E, (U, α) is a chart for M , and x0 ∈ U .
Let us write “1” for the identity map on Rn ; then (U × Rn , (α, 1)) is a
chart on M × Rn 16 and (π −1 (U ), (α, 1) ◦ ϕ)
e is a chart on E. With respect
to these two charts, the map ψ is represented by the map
e ◦ ψ ◦ (α, 1)−1 : α(U ) × Rn → α(U ) × Rn ,
(α, 1) ◦ ϕ
and we compute that this map equals
 X n 
(81) (y, (c1 , . . . , cn )) 7→ y, e j (α−1 (y)))) ,
cj · pr2 (ϕ(s
j=1

where pr2 is the projection map U × Rn → Rn . Now for each y ∈ α(U ), the
vectors
(82) e 1 (α−1 (y)))), . . . , pr2 (ϕ(s
pr2 (ϕ(s e n (α−1 (y))))
form a basis of Rn , since s1 (α−1 (y)), . . . , sn (α−1 (y)) form a basis of Eα−1 (y) .
Let T (y) be the real n × n matrix formed by the columns of the vectors in
(82). It follows that this matrix is invertible for every y ∈ α(U ), and that
the map in (81) is given by


(83) (y, c) 7→ y, T (y) · c .
(Remember that we represent vectors in Rn as column matrices.) It follows
that ϕ, which is the inverse of ψ, with respect to the two charts above is
represented by the inverse of (83), i.e. by the map
α(U ) × Rn → α(U ) × Rn , (y, c) 7→ (y, T (y)−1 · c).
Our task is to prove that this map is C ∞ (it is a map from an open subset
of Rd × Rn = Rd+n to Rd+n ). However this is clear from the formula of
the inverse matrix T (y)−1 in terms of the adjunct of T (y); cf. here. (In-
deed, every entry of T (y) is a C ∞ function of y since it is a composition of
C ∞ functions, and using the formula for T (y)−1 we see that each of the n
coordinates of
T (y)−1 · c
equals a certain polynomial in the entries of T (y) and the coordinates of c,
divided by det(T (y)), and det(T (y)) is a nowhere vanishing, C ∞ function of
y ∈ α(U ).)
This completes the proof that (U, ϕ) is a bundle chart for E, and thus of
the implication (c) ⇒ (b). 

16Of course, “(α, 1)” here stands for the map U × Rn → α(U ) × Rn , (x, v) 7→ (α(x), v).
Cf. footnote 9 above – in the (pedantic) language of that footnote, we would write “[α, 1]”
in place of “(α, 1)”.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 115

Problem 34:
The fact that there exist unique functions α1 , . . . , αn ∈ C ∞ (U ) satisfying
s = αj sj is clear from the fact that s1 (p), . . . , sn (p) is a basis of Ep for every
p ∈ U . The fact that each function αj is C ∞ is clear from the proof of “(c)
⇒ (b)” in Problem 33. 

Problem 35:
(a) In fact we can achieve this for any open set V ⊂ U whose closure in U
is compact. (This clearly suffices for us, since every point p ∈ U is contained
in such a set V .) Indeed, take any such set V . Let K be the closure of V
in U ; thus K is compact by our assumption. Now by Problem 7(d), there
exists a C ∞ function f : M → [0, 1] which has compact support contained
in U , and which satisfies f|K ≡ 1. Let us define the function s′ : M → E by
(
′ f (p)s(p) (∈ Ep ) if p ∈ U
s (p) =
0 (∈ Ep ) if p ∈
/ U.
Then π ◦ s′ = 1M by construction. Also s′ is C ∞ . [Proof: Let C be the
support of f ; this is a compact set contained in U . Let U ′ = M \C; this is an
open set. Now s′|U = f|U · s is C ∞ , and s′|U ′ is C ∞ , since it is identically zero.
Also U ∪ U ′ = M . Hence every point p ∈ M has an open neighbourhood in
which s′ is C ∞ ; this implies that s′ is C ∞ throughout M .] Hence s′ ∈ Γ(E).
Also for every p ∈ V we have s′ (p) = f (p)s(p) = s(p); hence s′|V = s|V .
Done! 
(b) Let p ∈ M be given. Let (U, ϕ) be a bundle chart with p ∈ U , and
let eb1 , . . . , ebn be the corresponding basis of sections of E|U (cf. Problem 33;
thus ebj (y) = ϕ−1 (y, ej ), ∀y ∈ U ). Now by part (a) there exist an open
subset V ⊂ U with p ∈ V and global sections b1 , . . . , bn ∈ Γ(E) such that
bj|V = ebj|V for j = 1, . . . , n. In other words bj (y) = ebj (y) for all y ∈ V ,
j = 1, . . . , n, and thus b1 (y), . . . , bn (y) is a basis of Ey for each y ∈ V .
Hence b1|V , . . . , bn|V form a basis of sections of E|V . 
(c) Given p, we choose V, b1 , . . . , bn as in part (b). Now also let v ∈ Ep be
given. Since b1 (p), . . . , bn (p) is a basis of Ep , there exist (unique) c1 , . . . , cn ∈
R such that v = cj · bj (p). Set s = cj bj ∈ ΓE. Then s(p) = v. 
116 ANDREAS STRÖMBERGSSON

Problem 36:
Cf., e.g., Lee, [15, Lemma 10.6].
If (V, x) is any chart for M and α ∈ A is such that Uα ∩ V 6= ∅, then we
let σx,α be the map
σx,α : π −1 (Uα ∩ V ) → Rd × Rn ,
σx,α(v) = (xα ◦ pr1 ◦ϕα (v), pr2 ◦ϕα (v))
= (xα ◦ π(v), pr2 ◦ϕα (v))
This is a bijection from π −1 (Uα ∩ V ) onto x(Uα ∩ V ) × Rn , which is an open
subset of Rd × Rn . Clearly E can be covered by sets of the form π −1 (Uα ∩ V )
as above. Hence we see from Problem 10 (parts b and d) that the family of
all (π −1 (Uα ∩ V ), σx,α ) as above generate a (unique!) C ∞ manifold structure
on E, 17 provided that we can only prove (1) C ∞ compatibility and (2) that
the topology generated by the family of all (π −1 (Uα ∩ V ), σx,α ) is Hausdorff,
connected and paracompact.
We first prove C ∞ compatibility. Specifically, we have to prove that for
any charts (V, x) and (W, y) for M and any α, β ∈ A subject to Uα ∩ Uβ ∩
V ∩ W 6= ∅,
(84) σx,α (π −1 (Uα ∩ Uβ ∩ V ∩ W ))
is an open subset of Rd × Rn , and the map σy,β ◦ σx,α
−1 from the set (84) to
d n ∞
R × R is C . However by parsing the definitions we see that the set in
(84) equals
xα (Uα ∩ Uβ ∩ V ∩ W ) × Rn ,
which is indeed an open subset of Rd × Rn . Also the map σx,α
−1 on this set is

given by
−1
σx,α (z, w) = ϕ−1 −1
α (xα (z), w),
and hence
 
−1
σy,β ◦ σx,α (z, w) = xβ ◦ pr1 ◦ϕβ ◦ ϕ−1
α (x −1
α (z), w), pr 2 ◦ϕβ ◦ ϕ−1 −1
α (x α (z), w)
 
= xβ ◦ x−1 −1 −1
α (z), pr2 ◦ϕβ ◦ ϕα (xα (z), w) ,

which is C ∞ by inspection (in particular using our assumption that ϕβ ◦ϕ−1

α
is C ∞ ).
Now it is easy to prove that the induced topology on E is Hausdorff. (See
Problem 10(b) for the definition of the topology on E.) Indeed, let p, q ∈ E,
p 6= q. Note that π −1 (U ) is open in E for every open set U ⊂ M ; hence if
π(p) 6= π(q) then we can use the fact that M is Hausdorff to find disjoint
17And this choice of C ∞ manifold structure is clearly forced on us, from the require-
ments that (E, π, M ) is a vector bundle of rank n, and (Uα , ϕα ) is a bundle chart for every
α ∈ A.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 117

open sets U1 , U2 ⊂ M with p ∈ U1 , q ∈ U2 ; then π −1 (U1 ) and π −1 (U2 ) are

disjoint open sets in E, containing p resp. q, and we are done. It remains
to treat the case π(p) = π(q). Then choose a chart (V, x) for M and α ∈ A
such that π(p) = π(q) ∈ Uα ∩ V . Now σx,α(p) 6= σx,α (q) and hence there are
disjoint open subsets U1 , U2 ⊂ x(Uα ∩ V ) × Rn which contain σx,α (p) resp.
σx,α(q). By the argument in the solution to Problem 10(c), σx,α −1 (U ) and
1
−1 (U ) are open subsets of E; they are clearly disjoint and contain p resp.
σx,α 2
q. Done!
Next we verify that E is connected. Let A be any subset of E which is
both open and closed. This means that for any chart (V, x) for M and any
α ∈ A, σx,α (A ∩ π −1 (Uα ∩ V )) is both open and closed in xα (Uα ∩ V ) × Rn .
This implies that for every p ∈ Uα ∩ V , the set
(85) {w ∈ Rn : (xα (p), w) ∈ σx,α (A ∩ π −1 (Uα ∩ V ))}
is both open and closed in Rn , and since Rn is connected, the set in (85)
equals either ∅ or Rn . In view of the definition of σx,α and the fact that
(ϕα )|Ep is a bijection from Ep onto {p} × Rn , this implies that
(86) A ∩ Ep = ∅ or Ep ⊂ A.
Since every point p ∈ M is contained in some set of the form Uα ∩ V , the
dichotomy (86) holds for every p ∈ M . Set
AM := {p ∈ M : Ep ⊂ A} = {p ∈ M : 0p ∈ A}.
Here 0p denotes the zero vector in Ep , and the last equality holds because
of (86). Note that for any (V, x) and α as above, the fact that σx,α (A ∩
π −1 (Uα ∩ V )) is both open and closed in xα (Uα ∩ V ) × Rn implies that the
set
{z ∈ xα (Uα ∩ V ) : (z, 0) ∈ σx,α (A ∩ π −1 (Uα ∩ V ))}
is both open and closed in xα (Uα ∩ V ). Using here the fact that xα is a
homeomorphism, and the definition of σx,α and the fact that ϕα (0p ) = (p, 0)
∀p ∈ Uα ∩ V (and ϕα is a bijection), it follows that the set
{p ∈ Uα ∩ V : 0p ∈ A}
is both open and closed in Uα ∩ V . But that set equals AM ∩ Uα ∩ V , and the
fact that this set is both open and closed in Uα ∩ V , for any (V, x) and α as
above, implies that AM is both open and closed in M . But M is connected,
hence AM = ∅ or AM = M . In view of (86) this implies that either A = M
or A = ∅. Hence we have proved that E is connected.
Next we verify that E is paracompact. Note that by what we have already
verified, E is connected, Hausdorff, and locally Euclidean, and hence by
Problem 2(b) it suffices to prove that E a countable (topological) atlas. Let
U be a countable base for the topology of M . Let U ′ be the subset of those
Ω ∈ U for which there exists a chart (V, x) for M and some α ∈ A such that
Ω ⊂ Uα ∩ V . Then U ′ covers M (by the same argument as in the second
118 ANDREAS STRÖMBERGSSON

half of the solution to Problem 2(a)). Now for each Ω ∈ U ′ we choose one
chart (V, x) for M and one α ∈ A such that Ω ⊂ Uα ∩ V , and then set
ψΩ := (σx,α )|π−1 (Ω) . It follows from what we have proved above, and (the
solution to) Problem 10, that σx,α is a homeomorphism of π −1 (Uα ∩ V ) onto
x(U ∩ V ) × Rn , and hence also ψΩ is a homeomorphism of π −1 (Ω) onto an
open subset of Rd × Rn . Hence
{(π −1 (Ω), ψΩ ) : Ω ∈ U ′ }
is a (topological) atlas for E. This atlas is countable, since U ′ ⊂ U and U is
countable. Hence we have proved that E is paracompact!
Now we have verified all conditions necessary for Problem 10 (parts b
and d) to apply. Hence we have now provided E with a structure of a C ∞
manifold.
Now the map π : E → M is immediately verified to be C ∞ . Indeed, for
any (V, x) and α as above, using the chart (π −1 (Uα ∩ V ), σx,α ) for E and
the chart (V, x) for M , the map π is represented by the identity map on
xα (Uα ∩ V ).
Now it only remains to verify that for every α ∈ A, the bijection ϕα from
π −1 (Uα ) onto Uα × Rn is in fact a diffeomorphism. For this, it suffices to
verify that for every chart (V, x) for M (with Uα ∩ V 6= ∅), the restriction
of ϕα to π −1 (Uα ∩ V ) is a diffeomorphism onto (Uα ∩ V ) × Rn . However
this is clear since ϕα|π−1 (Uα ∩V ) equals the composition of σx,α with the dif-
feomorphism (z, v) 7→ (x−1 (z), v) from x(Uα ∩ V ) × Rn onto (Uα ∩ V ) × Rn ,
and σx,α is a diffeomorphism since (π −1 (Uα ∩ V ), σx,α ) is a chart in our C ∞
atlas for E (by Problem 10(d)). Done! 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 119

Problem 37:
(a). Let E be the Möbius bundle over S 1 , defined as in Lecture #7,
p. 2. Assume that E is trivial, i.e. there is a bundle chart (E, ϕ). (This
will lead to a contradiction.) Then by Problem 33 there is a global basis
of sections s ∈ ΓE, in other words a section s ∈ ΓE which is everywhere
non-zero. Let us identify S 1 with [0, 1]/≈ (where ≈ stands for identifying
the points 0 and 1 in [0, 1]) in the standard way, i.e. by mapping x ∈ [0, 1]/≈
to (cos(2πx), sin(2πx)). By definition s is a C ∞ function from [0, 1]/≈ to
E = [0, 1] × R/∼ such that pr1 (s(x)) = x for all x ∈ [0, 1]/≈ . In particular
there is some y ∈ R such that
s(0) = (0, y) = (1, −y) in E.
Note that y 6= 0, since s is everywhere non-zero. Now x 7→ pr2 (s(x))
is a C ∞ function from (0, 1) to R satisfying limx→0+ pr2 (s(x)) = y and
limx→1− pr2 (s(x)) = −y; hence by the intermediate value theorem there is
some x ∈ (0, 1) for which pr2 (s(x)) = 0, contradicting the fact that s is
everywhere non-zero.
Hence E is not trivial. 

(b). Let us write En = S 1 × Rn , the trivial vector bundle over S 1 of

e1 := E
rank n. Also let E e be the Möbius bundle over S 1 , and set for n ≥ 2:
Een := E
e ⊕ En−1 . We claim that the desired classification is as follows: The
vector bundles
(87) e1 , E2 , E
E1 , E e2 , E3 , E
e3 , . . . ,

are pairwise non-isomorphic, and every vector bundle over S 1 is isomorphic

to one of these!
We start by proving that the vector bundles in (87) are pairwise non-
isomorphic. Since isomorphisms of vector bundles preserve the rank, we
only need to prove that for each n the two vector bundles En and Een are
e
non-isomorphic, or equivalently that En is not trivial. We have already
proved this for n = 1 in part (a); hence we may here assume n ≥ 2. Thus
assume that Een is trivial. (This will lead to a contradiction.)
It seems convenient to use a slightly different model for the Möbius bundle
e
E1 than that used in part (a): We view E e1 as R2 / ∼ , where ∼ is the
equivalence relation
def  ′ 
(x, y) ∼ (x′ , y ′ ) ⇐⇒ x′ − x ∈ Z and y ′ = (−1)x −x y .
(For a precise description of the C ∞ manifold structure, see Problem 9(c),
and note that this quotient R2 /∼ is the same as Γ\R2 , where Γ is the group
of diffeomorphisms of R2 of the form (x, y) 7→ (x + n, (−1)n y), for n ∈ Z.)
def
We also view S 1 as R/∼ where x ∼ x′ ⇐⇒ x′ − x ∈ Z; then the projection
120 ANDREAS STRÖMBERGSSON

map π : E e → S 1 is given simply by projection onto the first coordinate;

[(x, y)] 7→ [x]. In a similar vein we also represent Een as the quotient space
n
(R × R )/∼ where
def  ′ 
(x, y) ∼ (x′ , y ′ ) ⇐⇒ x′ − x ∈ Z and y ′ = (Jn )x −x · y ,
where
 
−1 0 0 ··· 0
0 1 0 0
 
 0
Jn :=  0 0 1  ∈ GLn (R).
 .. .. .. 
 . . .
0 0 0 1
(Note that (Jn )m = I for all even m and (Jn )m = Jn for all odd m.) The
en → S 1 is again given by [(x, y)] 7→ [x].
projection π : E
Recall that we are assuming that E en is trivial. Then by Problem 33
there is a global basis of sections s1 , . . . , sn ∈ ΓE. Each sj is a C ∞ map
from S 1 = R/∼ to E en = (R × Rn )/∼ ; composing sj with the projection
R → R/∼ we obtain a C ∞ map sej : R → (R × Rn )/∼ such that for every
x ∈ R we have sej (x) = [(x, fj (x))] for some (unique) fj (x) ∈ Rn . One
verifies that fj is a C ∞ map R → Rn ; furthermore for each x ∈ R we have
[x] = [x + 1] in S 1 ; hence sej (x) = sej+1 (x), i.e.
(88) fj (x + 1) = Jn · fj (x), ∀x ∈ R.
Let F (x) be the real n × n matrix whose columns equal f1 (x), . . . , fn (x), in
this order. Then F (x) ∈ GLn (R) for each x ∈ R, since s1 ([x]), . . . , sn ([x]) is
en,[x] . Hence F is a C ∞ map R → Mn (R). Furthermore
a basis of the fiber E
(88) implies
F (x + 1) = Jn · F (x), ∀x ∈ R.
However det Jn = −1; hence the above implies det F (x + 1) = − det F (x).
By continuity this implies that for any fixed x ∈ R there is some x′ ∈ [x, x+1]
such that det F (x′ ) = 0, contradicting the fact that F (x′ ) ∈ GLn (R).
en is trivial leads to a contradic-
We have seen that the assumption that E
e
tion. Hence En is not trivial.

It remains to prove that every vector bundle over S 1 is isomorphic to one

of the vector bundles in (87). Thus let E be an arbitrary vector bundle over
S 1 . Set n = rank E. By definition of vector bundle, every point in S 1 is
contained in some bundle chart (U, ϕ) for E, and by shrinking U if necessary
we can assume U to be an open arc on S 1 . Hence since S 1 is compact, there
is a finite family of bundle charts (Uj , ϕj ) for E, j = 1, . . . , k, such that
S 1 = U1 ∪ · · · ∪ Uk and each Uj is an open arc.
Now we have:
 PROBLEMS; “RIEMANNIAN GEOMETRY” 121

Lemma 2. If (U, ϕ) and (V, ψ) are bundle charts for E such that U and
V are open arcs on S 1 and also U ∪ V is an open arc, then there exists a
bundle chart for E of the form (U ∪ V, η).

Proof. After a rotation we may assume U = (0, u) for some 0 < u ≤ 1. 18 If

V ⊂ U or U ⊂ V then there is nothing to prove; hence let us assume that
V 6⊂ U and U 6⊂ V . Then the assumptions imply that V contains either [0]
or [u]; after a reflection we may assume that V contains [u], and then we
must have V = (v1 , v2 ) for some 0 < v1 < u < v2 ≤ 1. Then U ∩ V = (v1 , u).
Let T : (v1 , u) → GLn (R) be the transition map between (U, ϕ) and (V, ψ)
(cf. [12, p. 42]), so that
(89) ϕ ◦ ψ −1 (p, x) = (p, T (p) · x), ∀p ∈ (v1 , u), x ∈ Rn .
Note that T is a C ∞ curve on the manifold GLn (R). Fix any real number
u′ ∈ (v1 , u); thus we now have
0 < v1 < u′ < u < v2 ≤ 1.
Using the same technique as in Problem 19(b), one shows that there exists
a C ∞ curve Te : (v1 , v2 ) → GLn (R) such that
(90) Te(p) = T (p), ∀p ∈ (v1 , u′ ).
Note that
U ∪ V = (0, v2 ).
Let us now define the map
η : π −1 (U ∪ V ) → (U ∪ V ) × Rn
as follows:

ϕ(w) if π(w) ∈ (0, u′ )
η(w) :=  
 π(w), Te(π(w)) · pr (ψ(w)) if π(w) ∈ (v1 , v2 ).
2

Note that this map is “over-defined”, since both options apply whenever
π(w) ∈ (v1 , u′ ); however using (89) and (90) one verifies that in this case
both options give the same value for η(w). It follows from this that η is C ∞ ,
since ϕ is C ∞ on π −1 ((0, u′ )) and w 7→ (π(w), Te(π(w)) · pr2 (ψ(w))) is C ∞
on π −1 ((v1 , v2 )). Also, by inspection, pr1 ◦ η = π, and for every p ∈ (0, v2 ),
ηp := pr2 ◦ η|Ep is a linear bijection from Ep onto Rn . Furthermore one
verifies that η is a bijection from π −1 (U ∪ V ) onto (U ∪ V ) × Rn , with
inverse given by:
(
−1 ϕ−1 (p, x) if p ∈ (0, u′ )
η (p, x) =

ψ −1 p, Te(p)−1 · x if p ∈ (v1 , v2 ).

18Of course, “(0, u)” here stands for the arc {[x] : x ∈ (0, u)} in S 1 = R/∼ . We will
employ this type of mild abuse of notation several times in the following...
122 ANDREAS STRÖMBERGSSON

As above one verifies that this map is well-defined although it is “over-

defined”, and that it is C ∞ . Hence η is a C ∞ diffeomorphism from π −1 (U ∪
V ) onto (U ∪ V ) × Rn , and therefore (U ∪ V, η) is a bundle chart for E. 

Applying the above lemma a finite number of times to our family of arcs
U1 , . . . , Uk , 19 we reduce to the case k = 2! Let us then write (U, ϕ) :=
(U1 , ϕ1 ) and (V, ψ) := (U2 , ϕ2 ). Thus now U, V are open arcs which cover
S 1 , and (U, ϕ) and (V, ψ) are bundle charts for E. With notation as in the
proof of Lemma 2, we may now assume U = (0, u) and V = (v1 , v2 ) where
0 < v2 − 1 < v1 < u < 1 < v2 < v1 + 1.
Thus U ∩ V is the union of the two disjoint open arcs (v1 , u) and (1, v2 ). As
in the proof of Lemma 2, let T : U ∩ V → GLn (R) be the transition map
between (U, ϕ) and (V, ψ), so that
(91)
ϕ ◦ ψ −1 (p, x) = (p, T (p) · x), ∀p ∈ U ∩ V = (v1 , u) ∪ (1, v2 ), x ∈ Rn .
Note that both T|(v1 ,u) and T|(1,v2 ) are C ∞ curves on the manifold GLn (R).
Note that GLn (R) has two connected components, namely
GL+
n (R) := {B ∈ GLn (R) : det B > 0}

and
GL−
n (R) := {B ∈ GLn (R) : det B < 0}.

Of course each of T|(v1 ,u) and T|(1,v2 ) is contained in a single connected com-
ponent.
Case I: T|(v1 ,u) and T|(1,v2 ) lie in the same connected component. Then
using the same technique as in Problem 19(b), one shows that, given any
ε > 0 so small that
0 < ε < v2 − 1 < v1 < u − ε < u < 1 < 1 + ε < v2 < v1 + 1,

there exists a C ∞ curve Te : (v1 , v2 ) → GLn (R) such that

Te(p) = T (p), ∀p ∈ (v1 , u − ε) ∪ (1 + ε, v2 ).
We can then define a map
η : E → S 1 × Rn
through:
(
ϕ(w) if π(w) ∈ (ε, u − ε)
η(w) :=  
π(w), Te(π(w)) · pr2 (ψ(w)) if π(w) ∈ (v1 , v2 ).

19each application reduces the number of arcs by one, and we stop whenever we find
two arcs which together cover S 1
 PROBLEMS; “RIEMANNIAN GEOMETRY” 123

As in the proof of Lemma 2 one verifies that this map is well-defined, and
is an isomorphism of vector bundles over S 1 . Hence we conclude that E is
isomorphic to the trivial vector bundle En = S 1 × Rn !
Case II: T|(v1 ,u) and T|(1,v2 ) lie in the different connected component. Then
the curve p 7→ Jn · T (p) for p ∈ (1, v2 ) lies in the same connected component
as T|(v1 ,u) , and hence, using the same technique as in Problem 19(b), one
shows that, given any ε > 0 so small that
0 < ε < v2 − 1 < v1 < u − ε < u < 1 < 1 + ε < v2 < v1 + 1,
there exists a C ∞ curve Te : (v1 , v2 ) → GLn (R) such that
(
T (p) ∀p ∈ (v1 , u − ε)
Te(p) =
Jn · T (p) ∀p ∈ (1 + ε, v2 ).
∼ e
Then Te can be used to define an isomorphism of vector bundles E −
→E n.
We leave out the details.

124 ANDREAS STRÖMBERGSSON

Problem 38: See Conlon, [3, Thm. 7.5.16].

Problem 39:
(a) Recall that as a set, we defined E1 ⊗ E2 to be
G

E1 ⊗ E2 = E1,p ⊗ E2,p ,
p∈M

with projection map π : E1 ⊗ E2 → M defined by π(v) = p if v ∈ E1,p ⊗ E2,p

(for any p ∈ M ). Also, if (U, ϕ1 ) is a bundle chart for E1 and (U, ϕ2 ) is a
bundle chart for E2 20 then we postulated that if we define
τ : π −1 (U ) → U × (Rn1 ⊗ Rn2 )


(92) τ (v) := p, (ϕ1,p ⊗ ϕ2,p )(v) , ∀p ∈ U, v ∈ E1,p ⊗ E2,p ,
then (U, τ ) is a bundle chart for E1 ⊗ E2 . In other words, τp = ϕ1,p ⊗ ϕ2,p
for each p ∈ M .
In order to verify that the above indeed gives a vector bundle
(E1 ⊗ E2 , π, M ),
we apply Problem 36, with the family of proposed bundle charts taken to
be the family of all (U, τ ) constructed as above, as h(U, ϕ1 ), (U, ϕ2 )i varies
through all pairs of bundle charts for E1 , E2 with “same U ”. Most of the
conditions in Problem 36 are immediately verified to hold. For example,
τp = ϕ1,p ⊗ ϕ2,p is a linear isomorphism of (E1 ⊗ E2 )p = E1,p ⊗ E2,p onto
{p} × (Rn1 ⊗ Rn2 ) – which we identify with Rn1 ⊗ Rn2 , since ϕj,p is a linear
isomorphism for Ej,p onto Rnj for j = 1, 2. Furthermore the sets U cover
M ; cf. footnote 20. The only condition which is not (completely) immediate
is the C ∞ compatibility of the proposed bundle charts.
Thus we need to verify that if both (U, ϕj ) (j = 1, 2) and (V, ψj ) (j = 1, 2),
are bundle charts for E1 and E2 respectively, and if τ is defined as in (92)
and τe : π −1 (V ) → V × (Rn1 ⊗ Rn2 ) is similarly defined using (V, ψ1 ) and
(V, ψ2 ) (that is, τe(v) := (p, (ψ1,p ⊗ψ2,p )(v)) for all p ∈ V and v ∈ E1,p ⊗E2,p ),
then (if also U ∩ V 6= ∅) the map τe ◦ τ −1 from (U ∩ V ) × (Rn1 ⊗ Rn2 ) to
itself is C ∞ . Now for any (p, v) ∈ (U ∩ V ) × (Rn1 ⊗ Rn2 ) we have
τe ◦ τ −1 (p, v) = (p, τep (τp−1 (v)));
hence (using Problem 8(c)) it suffices to verify that the map
(p, v) 7→ τep (τp−1 (v)), (U ∩ V ) × (Rn1 ⊗ Rn2 ) → Rn1 ⊗ Rn2

20– with the same U ! Note that the family of such open sets U certainly cover M , i.e.
for each p ∈ M there exist U, ϕ1 , ϕ2 such that p ∈ U and (U, ϕj ) is a bundle chart for Ej
for j = 1, 2! (Proof?)
 PROBLEMS; “RIEMANNIAN GEOMETRY” 125

is C ∞ . But
(93)
τep ◦ τp−1 = (ψ1,p ⊗ ψ2,p ) ◦ (ϕ1,p ⊗ ϕ2,p )−1 = (ψ1,p ◦ ϕ−1 −1
1,p ) ⊗ (ψ2,p ◦ ϕ2,p ),
where the last equality holds since ⊗ is a bifunctor which is covariant in
both arguments (cf. Sec. 7.2 of the lecture notes). Furthermore we know
that the two maps
(p, v) 7→ ψ1,p ◦ ϕ−1
1,p (v), (U ∩ V ) × Rm → Rm
and
(p, v) 7→ ψ2,p ◦ ϕ−1
2,p (v), (U ∩ V ) × Rn → Rn
are C ∞ . By using charts on U ∩ V we see that we will be done if we can
prove the following: Given any open set Ω ⊂ Rd and maps α : Ω → Mm (R)
and β : Ω → Mn (R) 21 such that the two maps
(x, v) 7→ α(x) · v, Ω × Rm → Rm
and
(x, v) 7→ β(x) · v, Ω × Rn → Rn
are C ∞ , then also the map
(94) (x, v) 7→ (α(x) ⊗ β(x)) · v, Ω × (Rm ⊗ Rn ) → (Rm ⊗ Rn )
is C ∞ . However the assumption about α and β is easily seen to be equivalent
to the statement that each matrix entry of α(x) is a C ∞ function of x ∈ Ω,
and similarly for β. Now in (94), by a (hopefully) obvious abuse of notation,
α(x) ⊗ β(x) stands for the matrix of the linear map from Rm ⊗ Rn = Rmn to
itself which is the “tensor product” of the two linear maps v 7→ α(x) · v and
v 7→ β(x) · v. This matrix is the Kronecker product of the matrices α(x) and
β(x); cf. wikipedia, and from the explicit formula for the Kronecker product
we immediately see that each of the (nm)2 matrix entries of α(x) ⊗ β(x) is
a C ∞ function of x ∈ Ω; hence it follows that the map in (94) is C ∞ , and
we are done!

(b) This is very similar to part (a) and we here only describe the set-up:
We define
Hom(E1 , E2 ) := ⊔p∈M Hom(E1,p ⊗ E2,p ),
and if (U, ϕ1 ) is a bundle chart for E1 and (U, ϕ2 ) is a bundle chart for
E2 then we postulate a corresponding bundle chart for Hom(E1 , E2 ) to be
(U, τ ), where τ is given by (analogue of (92)):
τ : π −1 (U ) → U × Hom(Rn1 , Rn2 )


(95) τ (v) := p, Hom(ϕ−1 1,p , ϕ2,p )(v) , ∀p ∈ U, v ∈ Hom(E1,p , E2,p ),

21Here M (R) is the space of real m × m matrices.

m
126 ANDREAS STRÖMBERGSSON

Cf. (96) below regarding the def of “Hom(ϕ−1 1,p , ϕ2,p )”; thus for each p ∈ M
−1
we get τp = Hom(ϕ1,p , ϕ2,p ); this is an R-linear map from Hom(E1 , E2 )p =
Hom(E1,p , E2,p ) to Hom(Rn1 , Rn2 ). The reason that we have to use “ϕ−1 1,p ”
is that Hom is contravariant in its first argument; cf. the discussion below,
especially footnote 23.
Instead of giving further details, we discuss the problem from a more
general point of view. The key property that is used in parts (a), (b), (c)
is that both “⊗” and “Hom” and “dual” are smooth (=C ∞ ) functors on
the category C of finite dimensional vector spaces over R. More specifically,
⊗ is a bifunctor covariant in both arguments (cf. Sec. 7.2 of the lecture
notes); Hom is a bifunctor which is contravariant in the first argument and
covariant in the second argument, and “dual” is a contravariant functor of
one variable. It is a general fact that given any smooth functor F of k
variables on C 22 then for any vector bundles E1 , . . . , Ek over M one can
define in a natural way a vector bundle “F(E1 , . . . , Ek )” over M . Cf. [16,
1.34 – 1.39]. Each of (a), (b), (c) is a special case of this fact.
Let us explain in some detail what it means to say that Hom is a “smooth
bifunctor on C”. Recall that C is the category of finite dimensional vector
spaces over R; thus “A ∈ ob(C)” means that A is a finite dimensional vector
space over R. For any two A, B ∈ ob(C), Hom(A, B) ∈ ob(C) is the vector
space of R-linear maps A → B. Furthermore given any A, A′ , B, B ′ ∈ ob(C)
and R-linear maps h : A′ → A and f : B → B ′ we define an R-linear map
“Hom(h, f )”:
(96)
 
Hom(h, f ) : Hom(A, B) → Hom(A′ , B ′ ); Hom(h, f ) (g) := f ◦ g ◦ h.
One immediately verifies that Hom(1A , 1B ) = 1Hom(A,B) for all A, B ∈ ob(C),
and that for any A, A′ , A′′ , B, B ′ , B ′′ ∈ ob(C) and any R-linear maps
h′ h f f′
A′′ −→ A′ −
→ A and → B ′ −→ B ′′
B−
we have:
(97) Hom(h′ , f ′ ) ◦ Hom(h, f ) = Hom(h ◦ h′ , f ′ ◦ f ).
The relations which we have here pointed out, mean exactly that Hom is a
bifunctor C × C → C, contravariant in the first argument23 and covariant in
the second argument.
Next, the smoothness of the bifunctor Hom consists in the following: Note
that for any A, A′ , B, B ′ ∈ ob(C), the operation of taking any pair of lin-
ear maps h : A′ → A and f : B → B ′ to the linear map Hom(f, g) :
22Thus for “⊗” and “Hom” we have k = 2, and for “dual” we have k = 1.
23contravariant – because of the switch of order between A and A′ in (96) versus in
“h : A′ → A”, and the corresponding switch of order between h′ and h in the left versus
the right hand side of (97).
 PROBLEMS; “RIEMANNIAN GEOMETRY” 127

Hom(A, B) → Hom(A′ , B ′ ), is itself a map

(98) Hom(A′ , A) × Hom(B, B ′ ) → Hom(Hom(A, B), Hom(A′ , B ′ ));
(h, f ) 7→ Hom(h, f ).
Here both Hom(A′ , A)×Hom(B, B ′ )and Hom(Hom(A, B), Hom(A′ , B ′ )) are
∞
C manifolds (since each “Hom” space is a finite dimensional vector space
over R), and hence it makes sense to claim that the map in (98) is C ∞ .
This is exactly what we mean by saying that the bifunctor Hom is smooth.
(Exercise: Prove this smoothness!)
(The corresponding smoothness of the bifunctor ⊗ is the statement that
for any A, A′ , B, B ′ ∈ ob(C), the map
Hom(A, A′ ) × Hom(B, B ′ ) → Hom(A ⊗ B, A′ , ⊗B ′ )
(f, g) 7→ f ⊗ g
isC ∞. This is proved by choosing bases for A, A′ , B, B ′ ; then Hom(A, A′ )
and Hom(B, B ′ ) and Hom(A ⊗ B, A′ , ⊗B ′ ) become spaces of (real) matrices,
and f ⊗ g is given by the Kronecker product of the matrices f and g, and the
smoothness is clear by inspection in the explicit formula for the Kronecker
product. Cf. the discussion at the end of the solution of part (a).)

(c) This is also covered by the general discussion above; viz., it is a special
case of [16, 1.38].
(In fact (c) can be obtained as a special case of (b); namely we have
E1∗ = Hom(E1 , E2 ) when E2 is the trivial vector bundle E2 = M × R. But
alternatively one could also deduce (b) as a consequence of (a) and (c),
namley for any vector bundles E1 and E2 we can identify Hom(E1 , E2 ) =
E1∗ ⊗ E2 .)
128 ANDREAS STRÖMBERGSSON

Problem 40: Let us write nj = rank Ej (j = 1, 2) and let π be the

projection map π : Hom(E1 , E2 ) → M .
Let s ∈ Γ(Hom(E1 , E2 )). Then for each p ∈ M , s(p) ∈ Hom(E1 , E2 )p =
Hom(E1,p , E2,p ), and so s gives rise to a map
f : E1 → E2 , f (x) := s(π1 (x))(x) (x ∈ E1 ).
By construction this map f satisfies π2 ◦ f = π1 , and furthermore for each
p ∈ M,
fp := f|E1,p = s(p) ∈ Hom(E1,p , E2,p ).
Hence if we can only prove that f is C ∞ then f is a bundle homomorphism
E1 → E2 .
To prove that f is C ∞ is a local problem: Thus we may pass to bundle
charts for E1 , E2 and a chart for M (suitably adapted), after which the
problem becomes(∗) : Given any open set Ω ⊂ Rd and any C ∞ function
(99) T : Ω → Hom(Rn1 , Rn2 ),
show that the map
(100) Ω × R n1 → Ω × R n2 , (x, v) 7→ (x, T (x)(v))
is C ∞ . This, however, is trivial: Recall that Hom(Rn1 , Rn2 ) can be identified
with the space of real n2 × n1 matrices, and to say that T is C ∞ means that
each matrix entry of T (x) is a smooth function of x ∈ Ω; then the smoothness
of the map (100) is clear from the explicit formula for the matrix product
T (x)·v. This completes the proof that f is C ∞ , and hence that f is a bundle
homomorphism E1 → E2 .
[(*) Let us give a few more details on the reduction to the Euclidean version of the
problem stated in (99), (100). It is remarkable how much more complicated this is to
actually spell out than it is to just “think it through in ones head”!24
Given x ∈ E1 it suffices to prove that there exists some open set V ⊂ E1 containing
x such that f|V is C ∞ . Let us choose V = π1−1 (U ) where U is an open set in M with
π1 (x) ∈ U for which there exist ϕ1 , ϕ2 such that (U, ϕ1 ) is a bundle chart for E1 and
(U, ϕ2 ) is a bundle chart for E2 ; thus our task is to prove that f|π−1 (U ) is C ∞ . Recall from
1
Problem 39(b) that (U, ϕ1 ) and (U, ϕ2 ) give rise to a bundle chart (U, τ ) for Hom(E1 , E2 )
such that
τp = Hom(ϕ−1
1,p , ϕ2,p ) : Hom(E1,p , E2,p ) → Hom(R
n1
, Rn2 ),
for all p ∈ U , and that, by the definition of the bifunctor “Hom”, this means that
τp (α) = ϕ2,p ◦ α ◦ ϕ−1
1,p ∈ Hom(R
n1
, Rn2 ), ∀α ∈ Hom(E1,p , E2,p ).

24In many situations in mathematics, such a phenomenon is a clear warning sign that
one does not really have a complete proof – and so there is good reason to carefully work
out the details. But for the task at hands it seems that there is not so much to worry
about...
 PROBLEMS; “RIEMANNIAN GEOMETRY” 129

Of course, τ itself is the map

τ :Hom(E1 , E2 )|U → U × Hom(Rn1 , Rn2 );
τ (α) = (p, τp (α)), ∀p ∈ U, α ∈ Hom(E1,p , E2,p ).

We have f (π1−1 (U )) ⊂ π2−1 (U ) (since π2 ◦ f = π1 ) and hence, since ϕ1 and ϕ2 are

diffeomorphisms, in order to prove that f|π−1 (U ) is C ∞ it suffices to prove that the map
1
ϕ2 ◦ f ◦ ϕ−1
1 : U ×R
n1
→ U × Rn2 is C ∞ . However at each p ∈ U we have
(pr2 ◦ϕ2 ◦ f ◦ ϕ−1 −1
1 )|{p}×Rn1 = ϕ2,p ◦ fp ◦ ϕ1,p = τp (fp ) = pr2 (τ (s(p))),

and hence for all (p, v) ∈ U × Rn1 :

   
(ϕ2 ◦ f ◦ ϕ−1
1 )(p, v) = p, pr2 (τ (s(p))) (v) .

Here pr2 ◦τ ◦ s is a C ∞ map from M to Hom(Rn1 , Rn2 ). Therefore, after passing to C ∞

charts on U , 25 we are reduced to the task stated in (99), (100)!]

We now continue with the solution. Let H be the set of bundle homo-
morphisms E1 → E2 . Then above we have constructed a map
(101) Γ(Hom(E1 , E2 )) → H, “s 7→ f ”.
We next construct the inverse map. Thus let f be a bundle homomorphism
E1 → E2 . Then by definition, for each p ∈ M , fp = f|E1,p is an R-linear map
from E1,p to E2,p , i.e. fp ∈ Hom(E1,p , E2,p ) = Hom(E1 , E2 )p ⊂ Hom(E1 , E2 ).
Let us define the map s : M → Hom(E1 , E2 ) by s(p) := fp . Clearly π ◦ s =
1M , and one verifies that s is C ∞ using bundle charts in a manner very
similar to what we did above. Hence s ∈ Γ(Hom(E1 , E2 )), and so we have
constructed a map
(102) H → Γ(Hom(E1 , E2 )), “f 7→ s”.
It is immediate from our definitions (in particular using “s(p) = fp ”) that
the two maps (101) and (102) are inverses to each other. Hence we the two
maps are in fact bijections. 

25
Here one could expand and give many more details! :-)
130 ANDREAS STRÖMBERGSSON

Problem 41:
(a) If (U, ϕ) is any bundle chart for E such that ϕ(π −1 (U )∩E ′ ) = U ×Rm ,
then Ep ∩E ′ is an m-dimensional subspace of Ep for every p ∈ U . [Proof: for
any p ∈ U the restriction of ϕ to Ep = π −1 ({p}) is a linear isomorphism from
Ep onto {p} × Rn , and ϕ(π −1 (U ) ∩ E ′ ) = U × Rm implies that ϕ(Ep ∩ E ′ ) =
{p} × Rm ; hence Ep ∩ E ′ is indeed an m-dimensional subspace of Ep .]
By assumption the bundle charts for E with the special property above
cover M ; hence for every p ∈ M the intersection Ep ∩ E ′ is a linear subspace
of Ep . Set µ(p) := dim(Ep ∩E ′ ); then µ is a function from M to {0, 1, . . . , n}.
It is clear from the previous discussion that µ is locally constant. (Indeed,
given p ∈ M , let (U, ϕ) be a bundle chart for E such that p ∈ U and
ϕ(π −1 (U ) ∩ E ′ ) = U × Rm for some m ≤ n; then we saw in the previous dis-
cussion that µ(q) = m for all q ∈ U .) Hence µ−1 ({m}) is an open subset of M
for each m ∈ {0, 1, . . . , n}. But the sets µ−1 ({0}), . . . , µ−1 ({n}) form a par-
tition of M ; furthermore M is connected (since M is a manifold), and thus
M cannot be represented as a union of two or more disjoint nonempty open
subsets. It follows that all except one of the sets µ−1 ({0}), . . . , µ−1 ({n}) are
empty. In other words there is m ∈ {0, . . . , n} such that µ−1 ({m}) = M ,
i.e. dim(Ep ∩ E ′ ) = m for all p ∈ M . Done! 
(b) Let F be the family of all bundle charts for E satisfying (1), i.e.
ϕ(E ′ ∩ π −1 (U )) = U × Rm . (By part (a) we know that m is a fixed integer,
0 ≤ m ≤ n, independent of (U, ϕ) ∈ F.) For each (U, ϕ) ∈ F we set
ϕ
e := ϕ|E ′ ∩π−1 (U ) .
We wish to prove that (E ′ , π|E ′ , M ) together with the family
{(U, ϕ)
e : (U, ϕ) ∈ F}
satisfy all the conditions required in Problem 36. For each p ∈ M we set
Ep′ = E ′ ∩ Ep ; we noted in part (a) that Ep′ is an m-dimensional subspace
of Ep . Also for each (U, ϕ) ∈ F, it holds by our assumptions that ϕ e is
′ −1 m
a bijection from E ∩ π (U ) onto U × R , and for each p ∈ U we have
e|Ep′ = ϕ|Ep′ and this is a linear isomorphism of Ep′ onto {p} × Rm . Also
ϕ
the sets U cover M as (U, ϕ) runs through F, by assumption. Hence it only
remains to prove that if both (U, ϕ), (V, ψ) ∈ F, then the map ψe ◦ ϕe−1 from
(U ∩ V ) × Rm onto itself is C ∞ . However that map is equal to the restriction
of the map
ψ ◦ ϕ−1 : (U ∩ V ) × Rn → (U ∩ V ) × Rn
to the set (U ∩ V ) × Rm , and from this the desired smoothness is clear.
[Let us discuss the very last step in some detail: By Problem 8(c) it
suffices to prove that both the maps pr1 ◦ψe ◦ ϕ
e−1 and pr2 ◦ψe ◦ ϕ
e−1 ; however
the first of these equals pr1 : (U ∩ V ) × R → U ∩ V which we know is C ∞ ;
m
 PROBLEMS; “RIEMANNIAN GEOMETRY” 131

hence it suffices to prove that the map

pr2 ◦ψe ◦ ϕ
e−1 : (U ∩ V ) × Rm → Rm
is C ∞ . That map is the restriction of the map
pr2 ◦ψ ◦ ϕ−1 : (U ∩ V ) × Rn → Rn
to (U ∩ V ) × Rm. Hence after passing to charts on (U ∩ V ) × Rm , we see that
it suffices to prove the following general fact: Given an open set Ω ⊂ Rk
(some k ∈ Z+ ) and a C ∞ map f : Ω → Rn , if it happens that f (x) ∈ Rm for
all x ∈ Ω then f is C ∞ also as a map Ω → Rm . This is of course completely
trivial from the definition of what it means for a map between Rd -spaces to
be C ∞ .26]
We have proved above that all the conditions of Problem 36 are fulfilled
and so by that problem, (E ′ , π|E ′ , M ) is a vector bundle of rank m, and
e is a bundle chart for E ′ for every (U, ϕ) ∈ F.
(U, ϕ) 

(c) This is more or less immediate from our assumptions about existence
of bundle charts satisfying (1), together with the following criterion for being
a differentiable submanifold which we pointed out in the notes to lecture #2
(here formulated with notation adapted to our setting): If E is any d + n
dimensional C ∞ manifold and E ′ is an arbitrary subset of E, then E ′ has
a (uniquely determined) structure of a differentiable submanifold of E of
dimension d + m if and only if for every x ∈ E ′ there is a C ∞ chart (V, ψ)
of E such that x ∈ V , ψ(x) = 0, ψ(V ) is an open cube (−ε, ε)d+n , and
(103) ψ(V ∩ E ′ ) = (−ε, ε)d+m × {0}n−m .
(Cf., e.g., [1, Sec. III.5, esp. Lemma 5.2].)
Details: Let x ∈ E ′ be given. Set p = π(x) ∈ M . Then by assumption
there is a bundle chart (U, ϕ) for E such that p ∈ U and (1) holds, i.e.
ϕ(E ′ ∩ π −1 (U )) = U × Rm . Now choose also any C ∞ chart (W, τ ) for M
with p ∈ W . Of course we may assume W ⊂ U (otherwise just replace W by
W ∩ U ) and τ (p) = 0 (otherwise just compose τ with a translation of Rd ).
Then τ (W ) is an open set in Rd containing 0; hence there is some ε > 0
such that (−ε, ε)d ⊂ τ (W ). Now we may replace W by the smaller open set
τ −1 ((−ε, ε)d ); after doing this we have τ (W ) = (−ε, ε)d . Now the map
ψ := (τ, 1Rn ) ◦ ϕ|π−1 (W )

26The conclusion here has the following generalization: Let S, N be C ∞ manifolds, let
f : S → N be a C ∞ map, and let M ⊂ N be a differentiable submanifold of N . Assume
that f (S) ⊂ M . Then f is C ∞ also as a map S → M . Cf., e.g., [15, Cor. 5.30]. The proof
of this fact basically reduces to what we have already done, if one uses charts as in [12,
Lemma 1.3.1]. Note that if we merely assume that M is an immersed submanifold of N ,
then the corresponding statement is false in general! (Can you give an example?)
132 ANDREAS STRÖMBERGSSON

is a diffeomorphism of π −1 (W ) onto (−ε, ε)d × Rn ⊂ Rd+n (since it is a com-

position of a diffeomorphism of π −1 (W ) onto W × Rn and a diffeomorphism
of W × Rn onto (−ε, ε)d × Rn ), and so
(π −1 (W ), ψ)
is a C ∞ chart for E ′ . It follows from ϕ(E ′ ∩ π −1 (U )) = U × Rm (where
“Rm ” really stands for Rm × {0}n−m ) that we have
ψ(E ′ ∩ π −1 (W )) = (−ε, ε)d × Rm × {0}n−m .
Hence if we set V := ψ −1 ((−ε, ε)d+n ) (an open subset of ψ −1 (W )) then also
(V, ψ|V )
is a C∞ chart for E′, and this chart satisfies x ∈ V , ψ(x) = 0, ψ(V ) =
(−ε, ε)d+n , and
ψ(E ′ ∩ V ) = (−ε, ε)d+m × {0}n−m ,
i.e. the condition (103). Done! 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 133

Problem 42:
(a). Recall that as a set, f ∗ E is defined to be
(104) f ∗ E := {(p, v) : p ∈ M, v ∈ Ef (p) } ⊂ M × E,
and we define the projection map π e : f ∗ E → M to be simply π
e := pr1 ,
∗
e(p, v) := p for all (p, v) ∈ f E. Also for any bundle chart (U, ϕ) for
i.e. π
(E, π, N ) we have specified that if ϕe is the map
(105) ϕ e−1 (f −1 (U )) → f −1 (U ) × Rn
e :π (n := rank E);
ϕ(p,
e v) := (p, pr2 (ϕ(v)))

then (f −1 (U ), ϕ)
e is a bundle chart for f ∗ E.
In order to prove that (f ∗ E, π e, M ) is a vector bundle, we now verify
e : f ∗ E → M with the above proposed bundle charts satisfy all the
that π
conditions required in Problem 36. Firstly, π e is obviously surjective, and for
every p ∈ M , the set (f ∗ E)p := π
e−1 (p) is seen to be
(f ∗ E)p := π
e−1 (p) = {p} × Ef (p) “= Ef (p) ”
(the last equality is our usual identification), and this set carries the struc-
ture of an n-dimensional real vector space since (E, π, N ) is a vector bundle.
Next let (U, ϕ) be any bundle chart for (E, π, N ). Then f −1 (U ) of course is
an open subset of M . Also
e−1 (f −1 (U )) = {(p, v) : p ∈ f −1 (U ), v ∈ Ef (p) },
π

and for any fixed p ∈ f −1 (U ), we know that pr2 ◦ϕ|Ef (p) is a linear isomor-
phism of Ef (p) onto Rn ; hence from the definition of ϕ,e (105), it follows that
∗
e to the set (f E)p = {p} × Ef (p) is a linear isomorphism
the restriction of ϕ
onto {p} × Rn . In particular ϕ e restricts to a bijection of {p} × Ef (p) onto
{p} × R , and using this fact for every p ∈ f −1 (U ) it follows that
n
S ϕ e is a
e−1 (f −1 (U )) onto f −1 (U ) × Rn . Furthermore, M = f −1 (U )
bijection of π
when the union is taken over all bundle charts (U, ϕ) for (E, π, N ), since the
family of such sets U cover N .
Now the only condition from Problem 36 which remains to be verified
is that if (U, ϕ) and (V, ψ) are any two bundle charts for (E, π, N ), then
ψe ◦ ϕ
e−1 is a C ∞ map from (f −1 (U ) ∩ f −1 (V )) × Rn to itself. To prove this,
first note that f −1 (U ) ∩ f −1 (V ) = f −1 (U ∩ V ). Next, by parsing the maps
one finds that


(106) ψe ◦ ϕ
e−1 (p, w) = p, pr2 (ψ(ϕ−1 (f (p), w))) ,
∀(p, w) ∈ f −1 (U ∩ V ) × Rn .

[Details: Let (p, w) ∈ f −1 (U ∩ V ) × Rn . Then f (p) ∈ U ∩ V and so

ϕ−1 (f (p), w) is defined and lies in Ef (p) ⊂ π −1 (U ∩V ). Hence (p, ϕ−1 (f (p), w)) ∈
134 ANDREAS STRÖMBERGSSON

(f ∗ E)p , and using (105) we have ϕ(p,

e ϕ−1 (f (p), w)) = (p, w); hence
(p, ϕ−1 (f (p), w)) = ϕ
e−1 (p, w);

and applying ψe to this relation and again using (105), we obtain (106)!]
Clearly (106) implies that ψe ◦ ϕ
e−1 is a C ∞ map from f −1 (U ∩ V ) × Rn
to itself (using the fact that the maps ψ, ϕ−1 , f are C ∞ , and also using
Problem 8(b),(c)).
Hence all the conditions from Problem 36 are satisfied, and so Problem 36
implies that (f ∗ E, π
e, M ) is a vector bundle, and that for any bundle chart
(U, ϕ) for (E, π, N ), (f −1 (U ), ϕ)
e is a bundle chart for f ∗ E. 

(b) Passing to local coordinates (viz., choosing appropriate charts and

bundle charts) one reduces to the case when M is an open subset of Rd , N
′
is an open subset of Rd , and E = N × Rn . 27 Then the definition of f ∗ ,
(104), becomes:
f ∗ E = {(p, f (p), v) : p ∈ M, v ∈ Rn } ⊂ M × E = M × N × Rn .
Now note that the map
′
ϕ : M × N × Rn → Rd × Rd × Rn ,
ϕ(p, q, v) = (p, q − f (p), v)
is a diffeomorphism of M × N × Rn onto
′
Ω = {(p, q ′ , v) ∈ M × Rd × Rn : q ′ + f (p) ∈ N },
′
which is an open subset of Rd × Rd × Rn (inverse map: ϕ−1 (p, q ′ , v) =
(p, q ′ + f (p), v)). Hence (M × N × Rn , ϕ) is a C ∞ chart on M × N × Rn ,
and we note that
ϕ(f ∗ E) = M × {0} × Rn ⊂ Ω.
The existence of such a chart immediately implies that for every p ∈ f ∗ E
there is a C ∞ chart (V, ψ) on M × N × Rn such that p ∈ V , ψ(p) = 0,
′ ′
ψ(V ) = (−ε, ε)d+d +n and ψ(V ∩ f ∗ E) = (−ε, ε)d+n × {0}d . (Indeed, simply
compose ϕ with a translation and a suitable permutation of the coordinates;
then restrict the domain appropriately – cf. the solution to Problem 41(c)).
Hence (by a result stated in the notes to Lecture #2; cf., e.g., [1, Sec. III.5]),
f ∗ E is a differentiable submanifold of M × E = M × N × Rn .
(One should also verify that the C ∞ manifold structure of f ∗ E as a differ-
entiable submanifold of M × E = M × N × Rn agrees with the C ∞ manifold
structure on f ∗ E defined in part (a) via Problem 36. This is “immediate”

27Here we are sweeping a lot of details under the carpet; however we have discussed
similar things many times previously...
 PROBLEMS; “RIEMANNIAN GEOMETRY” 135

by comparing the C ∞ charts provided in each case. – but would take some
effort to write out.) 

Problem 43:
(a). This is very direct: We map any (s1 , s2 ) ∈ Γ(E1 ) ⊕ Γ(E2 ) to the
section s ∈ Γ(E1 ⊕ E2 ) defined by
s(p) := (s1,p , s2,p ) ∈ E1,p ⊕ E2,p = (E1 ⊕ E2 )p .
One verifies that this map (s1 , s2 ) 7→ s is an isomorphism of C ∞ M -modules.
We leave out the details...

(b). This is a special case of (c). (Namely, take E2 = M × R in (c); then

Hom(E1 , E2 ) = E1∗ and Γ(E2 ) = C ∞ (M ), so that (c) gives the result that
we want.)
(Cf. also [3, Prop. 6.2.11 and Prop. 7.5.4].)

(c). (Cf., e.g., [16, Prop. 1.53].)

Given h ∈ Γ(Hom(E1 , E2 )) and s ∈ ΓE1 , let us define the map
(107) Φh,s : M → E2 , Φh,s(p) := h(p)(s(p)).
Clearly π2 ◦ Φh,s = 1M2 ; also Φh,s is a C ∞ map since h and s are C ∞
maps (passing to local coordinates this reduces to the basic fact pointed out
around (99), (100)). Hence Φh,s ∈ ΓE2 .
Next, given h ∈ Γ(Hom(E1 , E2 )), we consider the map s 7→ Φh,s . Actually
let us change notation by setting
Φh (s) := Φh,s (s ∈ ΓE1 ).
Our previous paragraph shows that Φh is a map
Φh : ΓE1 → ΓE2 .
It is immediate from (107) that Φh is C ∞ (M )-linear. Hence
Φh ∈ Hom(ΓE1 , ΓE2 ).

We have thus defined a map

(108) Γ(Hom(E1 , E2 )) → Hom(ΓE1 , ΓE2 ), h 7→ Φh .
It is again immediate from (107) that this map is C ∞ (M )-linear, i.e. a
homomorphism of C ∞ (M )-modules. We are going to prove that the map
(108) is a bijection. This will imply that it is an isomorphism of C ∞ (M )-
modules, as desired!
The proof of injectivity is easy: It suffices to prove that the kernel of
the map (108) is {0}. Thus let h ∈ Γ(Hom(E1 , E2 )) be given and assume
136 ANDREAS STRÖMBERGSSON

Φh = 0. Then Φh,s = 0 for all s ∈ ΓE1 , and so h(p)(s(p)) = 0 for all s ∈ ΓE1
and p ∈ M . Hence by Problem 35(c), for every p ∈ M we have h(p)(v) = 0,
∀v ∈ E1,p , i.e. h(p) = 0 in Hom(E1,p , E2,p ) = Hom(E1 , E2 )p . Since this is
true for every p ∈ M , we conclude that h = 0, as desired. This completes
the proof that the map in (108) is injective.
It remains to prove surjectivity. Thus take an arbitrary element Φ ∈
Hom(ΓE1 , ΓE2 ), i.e. a C ∞ (M )-linear map Φ : ΓE1 → ΓE2 . Let us start by
proving that Φ is “local” in the following sense:
Lemma 3. For any open set U ⊂ M and any s1 , s2 ∈ ΓE1 , if s1|U = s2|U
then Φ(s1 )|U = Φ(s2 )|U .

Proof. Assume s1|U = s2|U . Now our task is to prove that Φ(s1 )(p) =
Φ(s2 )(p) for every p ∈ U . Thus fix a point p ∈ U . By Problem 7(c) there is
a function f ∈ C ∞ (M ) which has compact support contained in U and which
satisfy f (p) = 1. Using s1|U = s2|U and f (p) = 0 for all p ∈ M \ U it follows
that f s1 = f s2 in ΓE1 ; thus Φ(f s1 ) = Φ(f s2 ). But Φ is C ∞ (M )-linear and
thus f Φ(s1 ) = f Φ(s2 ), and in particular f (p)Φ(s1 )(p) = f (p)Φ(s2 )(p), and
since f (p) 6= 0 this implies Φ(s1 )(p) = Φ(s2 )(p). Done! 

Let V be the family of open sets V ⊂ M such that there exist sections
b1 , . . . , bn ∈ ΓE1 and c1 , . . . , cm ∈ ΓE2 , such that b1|V , . . . , bn|V form a basis
of sections of E1|V and c1|V , . . . , cm|V form a basis of sections of E2|V . By
Problem 35(b), V covers M . Now take any V ∈ V, and choose sections
b1 , . . . , bn ∈ ΓE1 and c1 , . . . , cm ∈ ΓE2 with the property just mentioned.
For each j ∈ {1, . . . , n}, Φ(bj )|V ∈ Γ(E2|V ); hence (by Problem 34) there are
unique gjk ∈ C ∞ (V ) (k = 1, . . . , m) such that

(109) Φ(bj )|V = gjk ck|V .

For each q ∈ V , let h(q) be the linear map E1,q → E2,q which has matrix
(gjk (q)) with respect to the bases b1 (q), . . . , bn (q) and c1 (q), . . . , cm (q); that
is,


(110) h(q) αj bj (q) = αj gjk (q)ck (q), ∀α = (αj )j=1,...,n ∈ Rn .

Note that h(q) ∈ Hom(E1 , E2 )q , and since all gjk ∈ C ∞ (V ), we have

(111) h ∈ Γ(Hom(E1 , E2 )|V ).

Lemma 4. For every s ∈ ΓE1 and every q ∈ V , Φ(s)(q) = h(q)(s(q)).

Proof. Let s ∈ ΓE1 and q ∈ V be given, and take f 1 , . . . , f n ∈ C ∞ (V ) so

that s|V = f j bj|V . By Problem 7(e) there exist an open set U ⊂ V with
j j
q ∈ U and functions fe1 , . . . , fen ∈ C ∞ (M ) such that fe|U ≡ f|U . Hence
 PROBLEMS; “RIEMANNIAN GEOMETRY” 137

s|U = (fej bj )|U , and so by Lemma 3, Φ(s)|U = Φ(fej bj )|U , and in particular


Φ(s)(q) = Φ(fej bj )(q) = fej (q)Φ(bj )(q) = f j (q)gjk (q)ck (q) = h(q) f j (q)bj (q)
= h(q)(s(q)).
(In the second equality we used the assumption that Φ is C ∞ (M )-linear;
in
the third equality we used (109); in the fourth equality we used (110), and
in the last equality we used s|V = f j bj|V .) 
Lemma 5. h(q) depends only on Φ and q, and not on V or b1 , . . . , bn or
c1 , . . . , cm .

Proof. This is clear from Lemma 4 and Problem 35(c). 

It follows from Lemma 5 that h(q) ∈ Hom(E1 , E2 )q can be unambiguously

defined for any point q ∈ M which lies in some V ∈ V. But as we have
pointed out, V covers M ; hence we have in fact defined a map
h : M → Hom(E1 , E2 )
which satisfies π ◦ h = 1M and
(112) Φ(s)(q) = h(q)(s(q))
for all s ∈ ΓE1 and q ∈ M (cf. Lemma 4). But we also have h|V ∈
Γ(Hom(E1 , E2 )|V ) for every V ∈ V; hence h is C ∞ , and so h ∈ Γ(Hom(E1 , E2 )).
Now by (112), Φ = Φh . This completes the proof that h 7→ Φh is surjective.

138 ANDREAS STRÖMBERGSSON

(d). (Cf. [3, Thm. 7.5.5] and stackexchange.)

Incomplete solution: Given any f ∈ Γ(E1 ) and g ∈ Γ(E2 ), let us write
f ⊗ g for the map
(113) f ⊗ g : M → E1 ⊗ E2 ,
(f ⊗ g)(p) := f (p) ⊗ g(p) ∈ E1,p ⊗ E2,p (p ∈ M ).
Then clearly π ◦ (f ⊗ g) = 1M , where π is the projection map E1 × E2 → M .
Let us verify that the map f ⊗g is C ∞ . Assume that (U, ϕj ) is a bundle chart
for Ej , for j = 1, 2. Then by the definition of E1 ⊗E2 – cf. Problem 39 – if we
define τ : π −1 (U ) → U × Rmn by τ (p, v) = (p, τp (v)) where τp = ϕ1,p ⊗ ϕ2,p
(and we have fixed an identification Rm ⊗Rn = Rmn ), then (U, τ ) is a bundle
chart for E1 ⊗ E2 , i.e. τ is a diffeomorphism from π −1 (U ) onto U × Rmn .
Hence it suffices to verify that the map τ ◦ (f ⊗ g) : U → U × Rmn is
C∞ , for any (U, ϕj ) (j = 1, 2) as above. As usual it suffices to verify that
pr1 ◦τ ◦ (f ⊗ g) and pr2 ◦τ ◦ (f ⊗ g) are C ∞ ; the first of these is the identity
map on U which is trivially C ∞ ; and the second map is seen to equal
(114) p 7→ ϕ1,p (f (p)) ⊗ ϕ2,p (g(p)) : U → Rmn .
Here we know that the maps p 7→ ϕ1,p (f (p)) and p 7→ ϕ2,p (g(p)) are C ∞ .
But also the map hv, wi 7→ v ⊗ w : Rm × Rm → Rmn is C ∞ ; hence the map
(114) is C ∞ , as desired. Hence we have proved:
f ⊗ g ∈ Γ(E1 ⊗ E2 ).

We have thus constructed a map

Γ(E1 ) × Γ(E2 ) → Γ(E1 ⊗ E2 ), (f, g) 7→ f ⊗ g.
Note that this map is C ∞ (M )-bilinear; hence there is a unique C ∞ (M )-
linear map
J : Γ(E1 ) ⊗ Γ(E2 ) → Γ(E1 ⊗ E2 )
such that
J(f ⊗ g) = f ⊗ g, ∀f ∈ Γ(E1 ), g ∈ Γ(E2 ).
We claim that J is an isomorphism of C ∞ (M )-modules.
Proof that J is surjective: By Problem 38 28 there exists a finite open cover
U1 , . . . , Ur of M such that E1|Uℓ and E2|Uℓ are trivial for ℓ = 1, . . . , r. Then
for each fixed ℓ ∈ {1, . . . , r} there is a basis of sections b1 , . . . , bm ∈ ΓE1|Uℓ
and a basis of sections b′1 , . . . , b′n ∈ ΓE2|Uℓ (here we are writing m = rank E1
and n = rank E2 ). Now for each p ∈ Uℓ , the vectors bj (p) ⊗ b′k (p) (j ∈

28Apply Problem 38 to both E and E ; then consider the common refinement of the
1 2
two open covers, i.e. the family of pair-wise intersections of the open sets.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 139

{1, . . . , m}, k ∈ {1, . . . , n}}), form a basis of the R-linear space E1,p ⊗ E2,p .
Hence

{bj ⊗ b′k : j ∈ {1, . . . , m}, k ∈ {1, . . . , n}}

is a basis of sections in Γ(E1 ⊗ E2 )|Uℓ . (Here the notation “bj ⊗ b′k ” is the
one introduced above in (113), but applied for the vector bundles E1|Uℓ and
E2|Uℓ .) This means that for every section s ∈ Γ(E1 ⊗ E2 )|Uℓ there exists a
unique choice of functions g jk ∈ C ∞ (Uℓ ) such that
X n
m X m X
X n
jk
s= g · bj ⊗ b′k = (gjk bj ) ⊗ b′k .
j=1 k=1 j=1 k=1

Now let a global section s ∈ Γ(E1 ⊗ E2 ) be given. From the above

discussion (and passing to a slightly different notation) we conclude that
(ℓ) (ℓ)
for each ℓ ∈ {1, . . . , r} there exist sections σj ∈ ΓE1|Uℓ , τj ∈ ΓE2|Uℓ ,
j = 1, . . . , mn, such that
mn
X (ℓ) (ℓ)
s|Uℓ = σj ⊗ τ j .
j=1

By Problem 11(a),(b) (partition of unity), there exist C ∞ functions ϕ1 , . . . , ϕr :

M → [0, 1] satisfying supp ϕℓ ⊂ Uℓ for each ℓ and
r
X
ϕℓ (p)2 = 1, ∀p ∈ M.
ℓ=1

(ℓ) (ℓ) (ℓ)

Define the function σ
ej : M → E1 by σ
ej (p) = ϕℓ (p)σj (p) for p ∈ Uℓ
(ℓ) (ℓ)
and σ
ej = 0 ∈ E1,p for p ∈ M \ Uℓ . Then σ
ej is C ∞ by the argument in
(ℓ)
the solution to Problem 7(a) (indeed the restriction of σ
ej to the two open
∞
sets Uℓ and M \ supp(ϕℓ ) is C , and these two open sets cover M ). Also
(ℓ)
π1 ◦ σ
ej = 1M . Hence
(ℓ)
ej ∈ ΓE1 .
σ

Similarly define
(ℓ)
τej ∈ ΓE2
(ℓ) (ℓ)
by τej (p) = ϕℓ (p)τj (p) for p ∈ Uℓ and zero elsewhere. Now
mn X
X r
(ℓ) (ℓ)
ej ⊗ τej
σ ∈ ΓE1 ⊗ ΓE2 ,
j=1 ℓ=1
140 ANDREAS STRÖMBERGSSON

and for every p ∈ M we have

Xmn Xr  mn X
X r
(ℓ) (ℓ) (ℓ) (ℓ)
J ej ⊗ τej
σ (p) = ej (p) ⊗ τej (p)
σ
j=1 ℓ=1 j=1 ℓ=1
mn X
X r
(ℓ) (ℓ)
= (ϕℓ (p)σj (p)) ⊗ (ϕℓ (p)τj (p))
j=1 ℓ=1
(p∈Uℓ )
r
X mn
X (ℓ) (ℓ)
= ϕℓ (p)2 σj (p) ⊗ τj (p)
ℓ=1 j=1
(p∈Uℓ )
X r
= ϕℓ (p)2 s(p)
ℓ=1
(p∈Uℓ )
= s(p).
Hence
X r
mn X 
(ℓ) (ℓ)
J σ
ej ⊗ τej = s,
j=1 ℓ=1
and we have proved that J is surjective.
Proof that J is injective: This seems somewhat more complicated to
carry out in the direct approach fashion used above, and we skip it for
now... However see [3, Thm. 7.5.5] for an elegant (but slightly less direct)
proof. 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 141

Problem 44:
(a). (Cf. Poor, [16, Prop. 1.60].)
We start by proving that there is a natural bijection between Γf ∗ E and
Γf E. Remeber from Problem 42 that f ∗ E is a differentiable submanifold of
M × E, and we know from Problem 8 that pr2 : M × E → E is a C ∞ map.
Now if s ∈ Γf ∗ E then pr2 ◦s is a C ∞ map from M to E. Also π ◦ pr2 ◦s = f ,
since pr2 (s(p)) ∈ Ef (p) for all p ∈ M . Hence pr2 ◦s ∈ Γf E. Thus we have
constructed a map
(115) Γf ∗ E → Γf E; s 7→ pr2 ◦s.
We next construct the inverse map. Given σ ∈ Γf E, we define the map
b : M → f ∗ E;
σ σ
b(p) := (p, σ(p)).
Note that for any p ∈ M we have σ(p) ∈ Ef (p) and hence
b(p) ∈ {p} × Ef (p) = (f ∗ E)p .
σ
This relation implies in particular σb(p) ∈ f ∗ E, i.e. σ
b is indeed a well-defined
∗
map from M to f E, and it also implies that π e(bσ (p)) = p; thus πe◦σ b = 1M .
In order to prove that σb is C ∞ , it suffices to prove that ϕ e◦σb|f −1 (U ) is C ∞
for every bundle chart (U, ϕ) for (E, π, N ). 29 Now for every p ∈ f −1 (U ),
ϕ(b
e σ (p)) = ϕ(p,
e σ(p)) = (p, pr2 (ϕ(σ(p)))),
and the last expression is clearly a C ∞ function of p ∈ f −1 (U ) (using Prob-
lem 8(d) and the fact that any composition of C ∞ maps is a C ∞ map).
b ∈ Γf ∗ E. Thus we have constructed a map
Hence σ
(116) Γf E → Γf ∗ E; σ 7→ σ
b.

Next we prove that the two maps (115) and (116) are inverses to each
other. For every σ ∈ Γf E we have, for every p ∈ M ,
pr2 ◦b
σ (p) = pr2 (p, σ(p)) = σ(p).
σ = σ. Next, for every s ∈ Γf ∗ E we have, for every p ∈ M ,
Hence pr2 ◦b
p\
r2 ◦s(p) = (p, pr2 (s(p))) = (e
π (s(p)), pr2 (s(p))) = (pr1 (s(p)), pr2 (s(p)))
= s(p).
Hence p\
r2 ◦s = s. Done!
Hence we have proved that the map (115) is a bijection of Γf ∗ E onto Γf E,
with inverse given by (116). Hence there is a unique C ∞ M -module structure
on Γf E which such that (115) is an isomorphism of C ∞ M -modules!

29Here we use the notation from the solution of Problem 42(a); thus ϕ e is the C ∞
diffeomorphism of π e −1 (f −1 (U )) onto f −1 (U ) × Rn given by ϕ(p,
e v) := (p, pr2 (ϕ(v))); then
(f −1 (U ), ϕ)
e is a bundle chart for (f ∗ E, π e , M ).
142 ANDREAS STRÖMBERGSSON

It should here also be pointed out that the C ∞ M -module operations in

Γf E are completely natural, namely they are just pointwise addition and
pointwise multiplication by scalar(s). Indeed, let α ∈ C ∞ M and σ1 , σ2 ∈
Γf E. Then σ1 + σ2 ∈ Γf E is just the pointwise sum, i.e.
(σ1 + σ2 )(p) = σ1 (p) + σ2 (p) ∈ Ef (p) , ∀p ∈ M ;
and ασ1 ∈ Γf E is the pointwise product, i.e.
(ασ1 )(p) = α(p)σ1 (p) ∈ Ef (p) , ∀p ∈ M.
To prove the formula for σ1 + σ2 , note that since we require that (116) is
a C ∞ M -module isomorphism, we should have σ\ 1 + σ2 = σ b1 + σb2 in Γf ∗ E,
and by definition of the C ∞ M -module structure of Γf ∗ E, σ b1 + σb2 is just
pointwise sum, i.e. (b
σ1 + σ
b2 )(p) = σ b2 (p) for all p ∈ M . Hence
b1 (p) + σ
(p, (σ1 + σ2 )(p)) = σ\
1 + σ2 (p) = σ
b1 (p) + σ
b2 (p) = (p, σ1 (p)) + (p, σ2 (p))
= (p, σ1 (p) + σ2 (p)),
and therefore (σ1 + σ2 )(p) = σ1 (p) + σ2 (p). Done! The proof of the formula
for ασ1 is completely similar. 
(b) (Recall that “basis of sections” is defined in Problem 33(c).)
Our task is to prove that for every p ∈ U ,
(117) s1 (f (p)), . . . , sn (f (p)) form a basis for (f ∗ E)p .
However (f ∗ E)p = Ef (p) and f (p) ∈ V , and therefore (117) follows from our
assumption that s1 , . . . , sn is a basis of sections in ΓE|V . 

(c) By Problem 38 there is a finite open cover {U1 , . . . , Ur } of N such

that E|Uj is trivial for each j. For each j ∈ {1, . . . , r}, let sj,1 , . . . , sj,n
be a basis of sections in ΓE|Uj (here n = rank E). Also let ϕ1 , . . . , ϕr
be a subordinate partition of unity as in Problem 11(a), i.e. each ϕj is a
C ∞ function NP→ [0, 1] with supp ϕj ⊂ Uj (but supp ϕj is not necessarily
compact) and rj=1 ϕj (y) = 1 for all y ∈ N . By Problem 11(b), we may
√
furthermore assume that each function ρj := ϕj is C ∞ . Note that these
functions satisfy supp ρj = supp ϕj ⊂ Uj and
r
X
ρj (y)2 = 1, ∀y ∈ N.
j=1

For each j ∈ {1, . . . , r} and k ∈ {1, . . . , n} we define a function sej,k : N →

E by
(
ρj (y)sj,k (y) if y ∈ Uj
sej,k (y) =
0 (in Ey ) if y ∈
/ Uj .
Then supp(esj,k ) ⊂ supp(ρj ) ⊂ Uj and hence by mimicking the solution to
Problem 7(a) one shows that sej,k is C ∞ . Also π ◦ sej,k = 1N ; hence sej,k ∈ ΓE.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 143

Now let s ∈ Γf E be given. We then define αj,k : M → R for j ∈ {1, . . . , r}

and k ∈ {1, . . . , n} by the requirement
( P
s(x) = nk=1 αj,k (x) · sj,k (f (x)) if x ∈ f −1 (Uj );
(118)
αj,1 (x) = · · · = αj,n (x) = 0 / f −1 (Uj ).
if x ∈
By part (b) together with Problem 34, this makes the functions αj,k uniquely
determined, and αj,k|f −1 (Uj ) ∈ C ∞ (f −1 (Uj )) for all j, k. Next define the
ej,k : M → R by
function α
ej,k (x) := ρj (f (x)) · αj,k (x).
α
ej,k|f −1(Uj ) ∈ C ∞ (f −1 (Uj )), and also supp(e
Then α αj,k ) ⊂ supp(ρj ◦ f ) ⊂
−1 −1 30
f (supp(ρj )) ⊂ f (Uj ), and hence by Problem 7(a), α ej,k ∈ C ∞ (M ).
Now for each x ∈ M ,
X n
r X r
X n
X
α
ej,k (x)e
sj,k (f (x)) = ρj (f (x))2 αj,k (x)sj,k (f (x))
j=1 k=1 j=1 k=1
(f (x)∈Uj )
r
X

use (118) = ρj (f (x))2 s(x) = s(x).
j=1
(f (x)∈Uj )

(The sum over j is takenPover all j ∈ {1, . . . , r} for which f (x) ∈ Uj .) The
last equality holds since rj=1 ρj (f (x))2 = 1 and ρj (f (x)) = 0 for all j with
f (x) ∈
/ Uj . Hence we have expressed s as a finite sum of the desired form.


30Here the inclusion supp(ρ ◦f ) ⊂ f −1 (supp(ρ )) holds since f −1 (supp(ρ )) is a closed

j j j
subset of M which contains every x ∈ M satisfying ρj (f (x)) 6= 0.
144 ANDREAS STRÖMBERGSSON

Problem 45:
(a). The solution to Problem 40 is generalized to the present situation
without any new difficulties arising (apart from a little extra amount of
book-keeping):
Let us write nj = rank Ej (j = 1, 2) and let π be the projection map
π : Hom(E1 , f ∗ E2 ) → M .
Let s ∈ Γ(Hom(E1 , f ∗ E2 )). Then for each p ∈ M ,
s(p) ∈ Hom(E1 , f ∗ E2 )p = Hom(E1,p , E2,f (p) ),
and so s gives rise to a map
h : E1 → E2 , h(x) := s(π1 (x))(x) (x ∈ E1 ).
By construction this map h satisfies π2 ◦ h = f ◦ π1 , and furthermore for
each p ∈ M ,
hp := h|E1,p = s(p) ∈ Hom(E1,p , E2,f (p) ),
i.e. hp is a linear map from E1,p to E2,f (p) . Hence if we can only prove that
h is C ∞ then h is a bundle homomorphism E1 → E2 along f .
To prove that h is C ∞ is a local problem, and by passing to appropriate
charts and bundle charts it is seen to follow from the basic fact pointed
out around (99), (100) in the solution of Problem 40. (We leave out the
details...)
Hence, writing H for the set of bundle homomorphisms E1 → E2 along
f , then above we have constructed a map
(119) Γ(Hom(E1 , f ∗ E2 )) → H, “s 7→ h”.
We next construct the inverse map. Thus let h be a bundle homomorphism
E1 → E2 along f . Then by definition, for each p ∈ M , hp = h|E1,p is an
R-linear map from E1,p to E2,f (p) , i.e.
hp ∈ Hom(E1,p , E2,f (p) ) = Hom(E1 , f ∗ E2 )p .
Let us define the map s : M → Hom(E1 , f ∗ E2 ) by s(p) := hp . Clearly
π ◦ s = 1M , and one verifies that s is C ∞ by passing to local coordinates
(we again leave out the details). Hence s ∈ Γ(Hom(E1 , f ∗ E2 )), and so we
have constructed a map
(120) H → Γ(Hom(E1 , f ∗ E2 )), “h 7→ s”.
It is immediate from our definitions (in particular using “s(p) = hp ”) that
the two maps (119) and (120) are inverses to each other. Hence we the two
maps are in fact bijections. 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 145

(b). Consider the special case N = M and f = 1M . Then h : E1 → E2 is

a “bundle homomorphism along f ” iff h is a bundle homomorphism. Also
f ∗ E2 = E2 . Hence in this special case, part (a) of the present problem says
the same as Problem 40 (and one verifies that the bijection is really the
same as there).
Next consider the special case where E1 is the trivial vector bundle of
rank 1 over M , i.e. E1 = M × R (but f : M → N is again a general C ∞ map
between C ∞ manifolds; also (E2 , π2 , N ) is an arbitrary vector bundle). Then
there is an “obvious” bijection between the family of bundle homomorphism
h : E1 → E2 along f and the family Γf E2 of sections s of E2 along f : This
bijection is given by s(p) := h(p, 1) (∀p ∈ M ); inverse: h(p, r) = r · s(p)
(∀(p, r) ∈ E1 = M × R). Furthermore we have an “obvious” identification
Hom(E1 , f ∗ E2 ) = f ∗ E2 , via the identifications
Hom(E1 , f ∗ E2 )p = Hom(E1,p , (f ∗ E2 )p ) = Hom(R, (f ∗ E2 )p ) = (f ∗ E2 )p
(∀p ∈ M ). In the light of these identifications, part (a) of the present
problem now says that there is a natural bijection between Γf ∗ E2 and the
set Γf E2 . This is exactly the statement of Problem 44(a) (and one verifies
that the bijection is really the same as there). 
146 ANDREAS STRÖMBERGSSON

Problem 46:
Let (U, ϕ) be a bundle chart for E with c(t0 ) ∈ U . Take ε > 0 so small
that c(t) ∈ U for all t ∈ (t0 − ε, t0 + ε). Then via (U, ϕ), the section
s|(t0 −ε,t0 +ε) is identified with a C ∞ map from (t0 − ε, t0 + ε) to Rn , where
n := rank E. Let sj be the j:th coordinate of this function; thus sj is a C ∞
map from (t0 − ε, t0 + ε) to R, for j = 1, . . . , n. By Problem 12(b), for each
j there exists some εj ∈ (0, ε) and a C ∞ function gj : U → R such that
gj (c(t)) = sj (t) for all t ∈ (t0 − εj , t0 + εj ). Let g : U → Rn be the function
whose jth coordinate is gj ; this is a C ∞ function from U to Rn , and via our
bundle chart (U, ϕ), g defines a section se ∈ ΓE|U satisfying se(c(t)) = s(t) for
all t ∈ (t0 − ε′ , t0 + ε′ ), where ε′ := min(ε1 , . . . , εn ). Now by Problem 35(a),
there exists some s1 ∈ ΓE and an open set V ⊂ U containing c(t0 ) satisfying
s1|V = se|V . Now take ε′′ ∈ (0, ε′ ] so that c(t) ∈ V for all t ∈ (t0 − ε′′ , t0 + ε′′ ).
Then we have s1 (c(t)) = s(t) for all t ∈ (t0 − ε′′ , t0 + ε′′ ). Done! 

[Some pedantic details: In more precise notation, we have in the previous

discussion:
sj := pr
e j ◦ pr2 ◦ϕ ◦ s : (t0 − ε, t0 + ε) → R,
where pr2 is the projection from U × Rn onto the second factor Rn , and
prj : Rn → R is projection onto the jth coordinate. Also se is defined by
se(p) := ϕ−1 (p, g(p)), ∀p ∈ U,
which by inspection is indeed a C∞
map from U to E with π ◦ se = 1U ;
thus se ∈ ΓE|U . Our choice of g, g 1 , . . . , g n and ε′ implies that for every
t ∈ (t0 − ε′ , t0 + ε′ ) we have
gj (c(t)) = sj (t) = pr
e j ◦ pr2 ◦ϕ ◦ s(t);
hence
g(c(t)) = pr2 ◦ϕ ◦ s(t),
and since also pr1 ◦ϕ ◦ s(t) = π ◦ s(t) = c(t), it follows that:
ϕ(s(t)) = (c(t), g(c(t))),
and thus
s(t) = ϕ−1 (c(t), g(c(t))) = se(c(t)), ∀t ∈ (t0 − ε′ , t0 + ε′ ),
just as we claimed in the above discussion.]
 PROBLEMS; “RIEMANNIAN GEOMETRY” 147

Problem 47:
(a). The map
(121) f 7→ X(Y (f )) − Y (X(f )), C ∞ (M ) → C ∞ (M ),
is clearly R-linear, since X and Y are (or give) R-linear maps on C ∞ (M )
(cf. Problem 15(b)). Furthermore for any f, g ∈ C ∞ (M ), using the fact that
X and Y are derivations we have


X(Y (f g)) = X (Y f ) · g + f · (Y g)
= (X(Y f )) · g + (Y f ) · (Xg) + (Xf ) · (Y g) + f · (X(Y g)).
Similarly,
Y (X(f g)) = (Y (Xf )) · g + (Xf ) · (Y g) + (Y f ) · (Xg) + f · (Y (Xg)),
and subtracting the two we get
   
X(Y (f g)) − Y (X(f g)) = X(Y (f )) − Y (X(f )) · g + f · X(Y (g)) − Y (X(g)) .
Hence the map in (121) is a derivation of C ∞ (M ). Hence by Problem 15(b)
there is a unique vector field Z on M such that
Z(f ) = X(Y (f )) − Y (X(f )), ∀f ∈ C ∞ (M ).
Done. 

(b). ...

(c). For any f ∈ C ∞ (M ) we have, by definition of the Lie product,

[[X, Y ], Z](f ) = [X, Y ](Zf ) − Z([X, Y ]f )
= X(Y (Z(f ))) − Y (X(Z(f ))) − Z(X(Y (f ))) + Z(Y (X(f ))).
Adding this to the corresponding formulas for [[Y, Z], X](f ) and [[Z, X], Y ](f )
we obtain
 
[[X, Y ], Z] + [[Y, Z], X] + [[Z, X], Y ] (f ) = 0.
This is true for all f ∈ C ∞ (M ); hence we obtain (via Problem 15(b)):
[[X, Y ], Z] + [[Y, Z], X] + [[Z, X], Y ] = 0.


(d). ...
148 ANDREAS STRÖMBERGSSON

Problem 48:
(a). The task is to prove that if (V, y) is any other C ∞ chart on M , with
respect to which
X
ω|V = eI dy I
ω
I

eI ∈ C ∞ (U )), then we have

(with ω
X X
(122) dωI ∧ dxI = ωI ∧ dy I
de in U ∩ V.
I I

We start by noticing that the fact that

X X
(123) eI dy I =
ω ωI dxI in U ∩ V
I I

(since both these are = ω|U ∩V in U ∩ V ) means that

X X 
I ∗ I
(124) ω
eI dy = f ωI dx in y(U ∩ V ) ⊂ Rd .
I I

where f is the coordinate transformation

f = x ◦ y −1 : y(U ∩ V ) → x(U ∩ V ).

[Pedantic explanation: In (124) we are stating an equality between two

r-forms living on an open subset of Rd , and so we are no longer viewing dxI
or dy I as r-forms on U ∩ V , but rather as r-forms on x(U ) ⊂ Rd and on
y(V ) ⊂ Rd , respectively. This is the reason why we can both have (123) and
(124) although on first look they seem to contradict each other! Namely, in
(123) we are viewing dxI and dy I as r-forms on (subsets of) M , but in (124)
we are viewing them more concretely as r-forms on (subsets of) Rd . Note
that also ωI and ωeI stand for different things in (124) versus (123): in (123)
ωI and ωeI are functions on U and V respectively, whereas in (124) they are
functions on x(U ) and y(V ), respectively (and a more pedantically correct
notation for these would be ωI ◦ x−1 and ω eI ◦ y −1 ). The situation is exactly
the same regarding the relationship between (122) and the computation
below. Note that we saw a similar example of such31 abuse of notation
k
already when we introduced tangent spaces: We have ∂x∂ j = ∂y ∂
∂xj ∂y k
in Tp M
for any p ∈ U ∩ V , but the corresponding relation certainly does not hold
(in general) when we view ∂x∂ j as vectors in Tx(p) Rd = Rd and ∂y∂ k as vectors
in Ty(p) Rd = Rd .]

31very convenient!
 PROBLEMS; “RIEMANNIAN GEOMETRY” 149

Now in y(U ∩ V ) we have:

X X 
I I
ωI ∧ dy = d
de ω
eI dy
I I
 X 
∗ I
=d f ωI dx
I
 X 
∗ I
=f d ωI dx
I
X 
∗ I
=f dωI ∧ dx .
I
[Details: The first equality holds by the definition of d; the second by (124);
the third by Jost, [12, Lemma 2.1.3]; and finally the fourth equality again
holds by the definition of d.]
The equality proved in the above computation says exactly that (122)
holds! This completes the proof that the map d : Ωr (M ) → Ωr+1 (M ) is
well-defined. It is now immediate to verify that this map is R-linear. 
150 ANDREAS STRÖMBERGSSON

(b). Strictly speaking, we need to prove this formula first in the special
′
case when M and N are open subsets of Rd and Rd , since we make use of this
fact in the proof that the general exterior derivative d : Ωr (M ) → Ωr+1 (M )
is well-defined; cf. part (a) above!
Thus assume that M is an open subset of Rd and N is an open subset
′ ′
of Rd (and f : M → N is a C ∞ map). Let us write x = (x1 , . . . , xd ) for
a variable point in N and y = (y 1 , . . . , y d ) for a variable P
point in M . Let
ω ∈ Ω (N ), and take functions ωI ∈ C (N ) so that ω = I ωI dxI . Then
r ∞

f ∗ (ω) ∈ Ωr (M ) is given by (recall that I = (i1 , . . . , ir ) runs through all

r-tuples with 1 ≤ i1 < · · · < ir ≤ d′ ):
X
f ∗ (ω) = f ∗ (ωI ) f ∗ (dxi1 ∧ · · · ∧ dxir )
I
X
= (ωI ◦ f ) f ∗ (dxi1 ) ∧ · · · ∧ f ∗ (dxir )
I
X
= (ωI ◦ f ) d(xi1 ◦ f ) ∧ · · · ∧ d(xir ◦ f )
I
X
= (ωI ◦ f ) df i1 ∧ · · · ∧ df ir .
I

(In this computation we used some basic properties of f ∗ which we pointed

out in Lecture #8.) Hence (making use of [12, Lemma 2.1.2] for r-forms on
Rd ):
X  
d(f ∗ (ω)) = d (ωI ◦ f ) df i1 ∧ · · · ∧ df ir
I
X
= d(ωI ◦ f ) ∧ df i1 ∧ · · · ∧ df ir
I
+ (ωI ◦ f )(d(df i1 )) ∧ df i2 ∧ · · · ∧ df ir
(125) − (ωI ◦ f )(df i1 ) ∧ (d(df i2 )) ∧ df i3 ∧ · · · ∧ df ir

+ · · · − (−1)r (ωI ◦ f )(df i1 ) ∧ (df i2 ) ∧ · · · ∧ d(df ir ) .

But note that for any g ∈ C ∞ (M ) we have

 ∂g   ∂g  ∂2g
j j
(126) d(dg) = d dy = d ∧ dy = dy k ∧ dy j = 0.
∂y j ∂y j ∂y k ∂y j
(The last equality holds since we are adding over k, j ∈ {1, . . . , d}, and since
dy k ∧ dy j = −dy j ∧ dy k .) Hence all inner terms in (125) except the first
vanish, i.e. we obtain:
X
(127) d(f ∗ (ω)) = d(ωI ◦ f ) ∧ df i1 ∧ · · · ∧ df ir .
I
 PROBLEMS; “RIEMANNIAN GEOMETRY” 151

(Of course, (126) is a special case of the general relation d ◦ d = 0 [12,

Theorem 2.1.5], and once we know this the proof of (127) is much shorter.)
On the other hand,
X  X
∗ ∗ I
f (dω) = f dωI ∧ dx = f ∗ (dωI ) ∧ f ∗ (dxI )
I I
X
(128) = d(ωI ◦ f ) ∧ df i1 ∧ · · · ∧ df ir .
I
Comparing with (127), we conclude that indeed d(f ∗ (ω)) = f ∗ (dω). 
Later, when we have defined the exterior derivative for general manifolds
M (cf. part (a) of this problem), the above proof carries over, with very
small changes, to the case of a C ∞ map f : M → N between P C ∞ manifolds.
Indeed, one fixes a chart (U, x) on N and assumes ω|U = I ωI dxI with
ωI ∈ C ∞ (U ). Then the computation up to and including (125) is still valid,
in the open subset f −1 (U ) of M . (Note that each f j is a C ∞ function
f −1 (U ) → R, defined by f j := xj ◦ f .) Also for any open set W ⊂ M and
any g ∈ C ∞ (W ), the computation in (126) shows that d(dg) = 0; namely
if we work locally wrt any chart (V, y) on W . Hence we obtain (127), as
an equality of (r + 1)-forms restricted to the set f −1 (U ). Similarly the
computation (128) is valid in f −1 (U ), and so we conclude that
d(f ∗ (ω))|f −1 (U ) = f ∗ (dω)|f −1 (U ) .
But this is true for (U, x) being an arbitrary C ∞ chart on N ; hence we
actually have d(f ∗ (ω)) = f ∗ (dω) on all M . 
152 ANDREAS STRÖMBERGSSON

(c). Let (U, x) be a C ∞ chart on M . We first prove the stated formula

on U , and in the special case when
∂
(129) Xj := ∈ Γ(T U ), (j = 0, . . . , r),
∂xℓj
for some ℓ0 , . . . , ℓr ∈ {1, . . . , d} (d = dim M ). In this case [Xj , Xk ] = 0 for
all j, k ∈ {0, . . . , r}, and hence the formula that we wish to prove states that
for any ω ∈ Ωr (U ),
Xr


(130) [dω](X0 , . . . , Xr ) = (−1)j Xj ω(X0 , . . . , X̂j , . . . , Xr ) .
j=0

Let us first note that both sides of (130) are alternating in X0 , . . . , Xr , i.e. if
we replace X0 , . . . , Xr by Xσ(0) , . . . , Xσ(r) for some σ ∈ Sr+1 (the group of
permutations of {0, 1, . . . , r}) then the effect is that the expressions on both
sides of (130) get multiplied by sgn σ. [Detailed proof: For the left hand side
this holds since dω is alternating by definition. Now consider the right hand
side. It suffices to study what happens when Xi and Xi+1 are switched for
some i ∈ {0, 1, . . . , r − 1}, since Sr+1 is generated by such transpositions.
Then for each j ∈ / {i, i + 1} the corresponding term in the sum is negated,
since ω(X0 , . . . , X̂j , . . . , Xr ) gets negated by the transposition. Furthermore
the contribution from j ∈ {i, i + 1} to the sum after the transposition is



(−1)i Xi+1 ω(X0 , . . . , X̂i+1 , . . . , Xr ) + (−1)i+1 Xi ω(X0 , . . . , X̂i , . . . , Xr ) ,
and this is again equal to −1 times the contribution from j ∈ {i, i+ 1} in the
original sum. Hence we conclude that the whole expression in the right hand
side of (130) gets negated when switching Xi ↔ Xi+1 , and this completes
the proof of the claim.]
It follows that it suffices to prove (130) in the special case
(131) ℓ0 < ℓ1 < · · · < ℓr .
(Note in particular that if any two of the ℓj ’s are equal then the alternating
property proved above implies that both sides of (130) equal zero and so
the equality holds.)
∞ (U ) so that ω =
P I
Now P take ω I ∈ C I ωI dx (notation as before). Then
dω = I dωI ∧ dxI , and recalling I = (i1 , . . . , ir ) and the definition of wedge
product, we get:
X
[dω](X0 , . . . , Xr ) = (dωI ∧ dxI )(X0 , . . . , Xr )
I
X X r
Y
(132) = (sgn σ) · dωI (Xσ(0) ) · dxij (Xσ(j) ).
I σ∈Sr+1 j=1

Recalling (129), we see that the last product equals one if I = (i1 , . . . , ir ) =
(ℓσ(1) , . . . , ℓσ(r) ), otherwise zero. Recalling now (131) and the fact that I
runs through all r-tuples with 1 ≤ i1 < i2 < · · · < ir ≤ d, it follows
 PROBLEMS; “RIEMANNIAN GEOMETRY” 153

that σ ∈ Sr+1 contributes to the last sum iff σ(1) < σ(2) < · · · < σ(r).
There exist exactly r + 1 such permutations σ, namely one for each choice of
a := σ(0) ∈ {0, 1, . . . , r + 1}; explicitly the unique admissible permutation
σ = σa with σa (0) = a is given by σa (j) = j − 1 for 1 ≤ j ≤ a and
σa (j) = j for a < j ≤ r. Note also that this permutation σ can be obtained
as a product of a transpositions i ↔ i + 1; hence sgn(σa ) = (−1)a . Given
σ = σa , we get contribution from I = (ℓσa (1) , . . . , ℓσa (r) ) and no other I, and
P
using ω = I ωI dxI we get for I = (ℓσ(1) , . . . , ℓσ(r) ):

ωI = ω(Xσa (1) , . . . , Xσa (r) ) = ω(X0 , . . . , X̂a , . . . , Xr ) ∈ C ∞ (U )

and thus


dωI (Xσa (0) ) = Xa ω(X0 , . . . , X̂a , . . . , Xr ) .

Using these facts in (132) we get:

r
X

[dω](X0 , . . . , Xr ) = (−1)a Xa ω(X0 , . . . , X̂a , . . . , Xr ) ,
a=0

i.e. (130) is proved! Recall that this was under the assumption that Xj =
∂
ℓj for j = 0, . . . , r; cf. (129).
∂x
Next, let us call the right hand side of the general formula which we wish
to prove “F (X0 , . . . , Xr )”. That is:
r
X

F (X0 , . . . , Xr ) := (−1)j Xj ω(X0 , . . . , X̂j , . . . , Xr )
j=0
X
(133) + (−1)j+k ω([Xj , Xk ], X0 , . . . , X̂j , . . . , X̂k , . . . , Xr ).
0≤j<k≤r

Thus F is a map F : Γ(T M ) × · · · × Γ(T M ) → C ∞ (M ). Let us prove that

F is C ∞ (M )-multilinear! It is clear by inspection that F is R-multilinear.
Let us also note that F is alternating. (Proof: The argument we gave just
below (130) applies to show that the first sum is alternating. For the second
sum a similar argument works; we leave out the details.) Hence to show the
C ∞ (M )-multilinearity, it now suffices to prove that for any f ∈ C ∞ (M ),

F (f X0 , X1 , . . . , Xr ) = f · F (X0 , X1 , . . . , Xr ).

But when we replace X0 by f X0 , the j = 0 term in the first sum of (133)

clearly gets multiplied by f , whereas each j > 0 term becomes (−1)j times



Xj ω(f X0 , . . . , X̂j , . . . , Xr ) = Xj f · ω(X0 , . . . , X̂j , . . . , Xr )


= (Xj f ) · ω(X0 , . . . , X̂j , . . . , Xr ) + f · Xj ω(X0 , . . . , X̂j , . . . , Xr ) .
154 ANDREAS STRÖMBERGSSON

In the second sum, each term with 0 < j < k ≤ r gets multiplied by f ,
whereas each term with 0 = j < k ≤ r becomes (−1)0+k times
ω([f X0 , Xk ], X1 , . . . , X̂k , . . . , Xr )


= ω f [X0 , Xk ] − (Xk f )X0 , X1 , . . . , X̂k , . . . , Xr
= f · ω([X0 , Xk ], X1 , . . . , X̂k , . . . , Xr ) − (Xk f ) · ω(X0 , . . . , X̂k , . . . , Xr ).
(Cf. Problem 47(d).) Hence in total we get:
F (f X0 , X1 , . . . , Xr ) =f · F (X0 , X1 , . . . , Xr )
r
X
+ (−1)j (Xj f ) · ω(X0 , . . . , X̂j , . . . , Xr )
j=1
Xr


+ (−1)k −(Xk f ) · ω(X0 , . . . , X̂k , . . . , Xr )
k=1
= f · F (X0 , X1 , . . . , Xr ).
Hence the C ∞ (M )-multilinearity is proved.
Now the proof of the formula in the general case is easily completed:
Again let (U, x) be an arbitrary C ∞ chart on M ; it suffices to prove that the
desired formula holds in U . Hence from now on we may just as well assume
M = U . We keep the notation “F ” from (133) (but now with M = U ). We
have proved above that F is C ∞ (U )-multilinear, and also that
(134) [dω](X0 , . . . , Xr ) = F (X0 , . . . , Xr )
when each Xj is of the form Xj = ∂ℓj (indeed, cf. (130) and again recall that
∂x
for such X0 , . . . , Xr , all Lie brackets [Xj , Xk ] vanish). But every X ∈ Γ(T U )
∂ ∞ (U ); hence
can be expressed as X = fℓ ∂x ℓ for some (unique) f1 , . . . , fd ∈ C
it follows that (134) holds for arbitrary X1 , . . . , Xr ∈ Γ(T U ). This is the
desired formula! 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 155

Problem 49:
(a). By the definition
Vr of tensor product (of C ∞ (M )-modules), every ele-
ment in Γ(E1 )⊗Γ( M ) can V be written as a finite sum of pure tensors µ1 ⊗ω1
(where µ1 ∈ ΓE1 , ω1 ∈ Γ( r M ) = ΩrV (M )). Hence by Problem 43(d), the
same is true for any section in Γ(E ⊗ r M ) = Ωr (E ). The analogous fact
Vs 1 1
of course also holds for Γ(E2 ⊗ M ) = Ωs (E2 ). Hence the stated formula,
(135) (µ1 ⊗ ω1 ) ∧ (µ2 ⊗ ω2 ) = (µ1 ⊗ µ2 ) ⊗ (ω1 ∧ ω2 ),
∀µ1 ∈ Γ(E1 ), ω1 ∈ Ωr (M ), µ2 ∈ Γ(E2 ), ω2 ∈ Ωs (M ),
together with the requirement that ∧ should be a C ∞ (M )-bilinear map
Ωr (E1 ) × Ωs (E2 ) → Ωr+s (E1 ⊗ E2 ), certainly makes s1 ∧ s2 uniquely deter-
mined for any s1 ∈ Ωr (E1 ), s2 ∈ Ωs (E2 ). Hence it remains to prove that
such a C ∞ (M )-bilinear map exists.
For the existence proof, we start by considering the map
F : Γ(E1 ) × Ωr (M ) × Γ(E2 ) × Ωs (M ) → Ωr+s (E1 ⊗ E2 ),
F (µ1 , ω1 , µ2 , ω2 ) = (µ1 ⊗ µ2 ) ⊗ (ω1 ∧ ω2 ).
This map is immediately verified to be C ∞ (M )-multilinear. Hence by the
definition of tensor product, there exists a unique C ∞ (M )-linear map
Fe : Γ(E1 ) ⊗ Ωr (M ) ⊗ Γ(E2 ) ⊗ Ωs (M ) → Ωr+s (E1 ⊗ E2 )
satisfying
Fe(µ1 ⊗ ω1 ⊗ µ2 ⊗ ω2 ) = (µ1 ⊗ µ2 ) ⊗ (ω1 ∧ ω2 ),
(136) ∀µ1 ∈ Γ(E1 ), ω1 ∈ Ωr (M ), µ2 ∈ Γ(E2 ), ω2 ∈ Ωs (M ).
V
Using Problem 43(d) and Ωr (M ) = Γ( r M ), Fe becomes identified with a
C ∞ (M )-linear map
Fe : Ωr (E1 ) ⊗ Ωs (E2 ) → Ωr+s (E1 ⊗ E2 ).

Composing Fe with the canonical map (s1 , s2 ) 7→ s1 ⊗s2 from Ωr (E1 )×Ωs (E2 )
to Ωr (E1 ) ⊗ Ωs (E2 ) (which is C ∞ (M )-bilinear by the definition of tensor
product), we obtain a C ∞ (M )-bilinear map
Ωr (E1 ) × Ωs (E2 ) → Ωr+s (E1 ⊗ E2 )


which maps µ1 ⊗ ω1 , µ2 ⊗ ω2 to (µ1 ⊗ µ2 ) ⊗ (ω1 ∧ ω2 ) for all µ1 , ω1 , µ2 , ω2 as
in (136). This map satisfies all requirements imposed on “∧”, i.e. we have
proved the (unique) existence of such a map “∧”! 
As an addendum, let us note that the wedge product s1 ∧ s2 ∈ Ωr+s (E1 ⊗
E2 ) of any two sections s1 ∈ Ωr (E1 ) and s2 ∈ Ωs (E2 ) can be computed
“fiber by fiber”:
(s1 ∧ s2 )(p) = s1 (p) ∧ s2 (p), ∀p ∈ M,
156 ANDREAS STRÖMBERGSSON

where in the right hand side, the “∧” 32 denotes the unique R-bilinear map
V V V
(E1,p ⊗ r (Tp∗ (M ))) × (E2,p ⊗ s (Tp∗ (M ))) → (E1,p ⊗ E2,p ) ⊗ r+s (Tp∗ (M ))
satisfying
(v1 ⊗ ϕ1 ) ∧ (v2 ⊗ ϕ2 ) = (v1 ⊗ v2 ) ⊗ (ϕ1 ∧ ϕ2 ),
V V
(137) ∀v1 ∈ E1,p , ϕ1 ∈ r (Tp∗ (M )), v2 ∈ E2,p , ϕ2 ∈ s (Tp∗ (M )).
Indeed, this is clear by parsing through the identifications in the above proof
(in particular see equation (113) in the solution to Problem 43(d)). Note also
that the fact that there indeed exists a unique R-bilinear map “∧” satisfying
(137) is proved by an argument completely similar to what we did above. 

(b). This is quite immediate from the corresponding formulas for the
“standard” wedge product on Ω(M ). Indeed, for the associativity relation,
by C ∞ (M )-multilinearity and the argument at the beginning of our solution
to part (a), it suffices to prove the identity s1 , s2 , s3 of the form sj = µj ⊗ ωj
(j = 1, 2, 3), where µj ∈ Γ(Ej ) and ω1 ∈ Ωr (M ), ω2 ∈ Ωs (M ), ω3 ∈ Ωt (M ).
But in that case we have
 
s1 ∧ (s2 ∧ s3 ) = (µ1 ⊗ ω1 ) ∧ (µ2 ⊗ ω2 ) ∧ (µ3 ⊗ ω3 )
 
= (µ1 ⊗ ω1 ) ∧ (µ2 ⊗ µ3 ) ⊗ (ω2 ∧ ω3 )
= (µ1 ⊗ µ2 ⊗ µ3 ) ⊗ (ω1 ∧ (ω2 ∧ ω3 ))
= (µ1 ⊗ µ2 ⊗ µ3 ) ⊗ ((ω1 ∧ ω2 ) ∧ ω3 )
= ···
= (s1 ∧ s2 ) ∧ s3 .
In the fourth equality we used the fact that the wedge product on Ω(M ) is
associative.
Similarly for any s1 , s2 as above we have
(−1)rs · J(s2 ∧ s1 ) = (−1)rs · J((µ2 ⊗ µ1 ) ⊗ (ω2 ∧ ω1 ))
= (−1)rs · (µ1 ⊗ µ2 ) ⊗ (ω2 ∧ ω1 )
= (µ1 ⊗ µ2 ) ⊗ (ω1 ∧ ω2 )
= s1 ∧ s2 .
In the third equality we used the fact that ω1 ∧ ω2 = (−1)rs ω2 ∧ ω1 . 

(c). This is immediate from the second relation in part (b); indeed note
e
that m′ = m ◦ J : Ωr+s (E2 ⊗ E1 ) → Ωr+s (E).
32Note that we are now using “∧” in quite a few different ways; however it will always
be clear from the type of the two arguments which “∧” is used in each instance.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 157

(d). Similarly, this is fairly immediate from the first relation in part (b);
we leave out a detailed discussion. (One has to change order between certain
operations and this becomes somewhat tedious to spell out.) Instead we give
a direct proof by mimicking the proof of the first relation in part (b): For
any s1 , s2 , s3 of the form sj = µj ⊗ ωj (j = 1, 2, 3), where µj ∈ Γ(Ej ) and
ω1 ∈ Ωr (M ), ω2 ∈ Ωs (M ), ω3 ∈ Ωt (M ), we have (note carefully that now
certain “∧” have a different meaning than in part (b)...):
 
s1 ∧ (s2 ∧ s3 ) = (µ1 ⊗ ω1 ) ∧ (µ2 ⊗ ω2 ) ∧ (µ3 ⊗ ω3 )
 
= (µ1 ⊗ ω1 ) ∧ (µ2 · µ3 ) ⊗ (ω2 ∧ ω3 )
= (µ1 · (µ2 · µ3 )) ⊗ (ω1 ∧ (ω2 ∧ ω3 ))
= ((µ1 · µ2 ) · µ3 ) ⊗ ((ω1 ∧ ω2 ) ∧ ω3 )
= ···
= (s1 ∧ s2 ) ∧ s3 .
Done! 
158 ANDREAS STRÖMBERGSSON

Problem 50: We have

   
j∗ k i∗ ℓ
(µ ◦ η)|U = β ⊗ γk ⊗ µj ◦ α ⊗ βℓ ⊗ ηi .
and by the definitions in Problem 49(a),(c), this is




= β j∗ ⊗ γk ◦ αi∗ ⊗ βℓ ⊗ µkj ∧ ηiℓ .
But using the definitions of composition of homomorphisms and of the iden-
tifications Hom(E2 , E3 ) = E2∗ ⊗ E3 and Hom(E1 , E2 ) = E1∗ ⊗ E2 , we find
that



β j∗ ⊗ γk ◦ αi∗ ⊗ βℓ = δj,ℓ αi∗ ⊗ γk .
(Here δj,ℓ is the standard Kronecker symbol; δj,ℓ = 1 if j = ℓ otherwise = 0.
Note that in terms of matrices the last formula is simply a formula for the
product of a matrix with “1” in position k, j and all other entries zero, and
a matrix with “1” in position ℓ, i and all other entries zero.) Using the last
formula in the previous one, we obtain:



(µ ◦ η)|U = δj,ℓ αi∗ ⊗ γk ⊗ µkj ∧ ηiℓ



= αi∗ ⊗ γk ⊗ µkj ∧ ηij ,
qed. 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 159

Problem 51:
(a). To start, by the definition of tensor product, there is a natural
bijection between the C ∞ (M )-multilinear maps Γ(T M )(r) → ΓE and the
C ∞ (M )-linear maps from
· · ⊗ T M}) = Γ(T0r M )
| M ⊗ ·{z
Γ(T M ) ⊗ · · · ⊗ Γ(T M ) = Γ(T
| {z }
r times r times

to ΓE (here we used Problem 43(d)). But the space of such maps is

Hom(Γ(T0r M ), ΓE) = ΓHom(T0r M, E) = Γ(E ⊗ (T0r M )∗ ) = Γ(E ⊗ Tr0 M ).
(For the first equality see Problem 43(c); for the next equality cf. p. 9 in
Lecture #7.) Working through the identifications used above, we see that
viewing a given s ∈ Γ(E ⊗ Tr0 M ) as a C ∞ (M )-multilinear map Γ(T M )(r) →
ΓE, means that for any vector fields X1 , . . . , Xr ∈ Γ(T M ), the section
s(X1 , . . . , Xr ) ∈ ΓE
is explicitly given by
(138)



s(X1 , . . . , Xr ) (p) = Cp s(p) ⊗ (X1 (p) ⊗ · · · ⊗ Xr (p)) (∀p ∈ M ),
where Cp is the unique R-linear map (“contraction at p”)


Cp : Ep ⊗ Tr0 (M )p ⊗ T0r (M )p → Ep
which maps
(139) Cp (w ⊗ η ⊗ α) = η(α) · w, ∀w ∈ Ep , η ∈ Tr0 (M )p , α ∈ T0r (M )p .
V
Next, note that r M by definition is a subset of Tr0 (M ), and oneV
verifies
immediately that it is in fact a subbundle of Tr0 (M ). 33 Hence E ⊗ r M is
a subbundle of E ⊗ Tr0 (M ). 34
In order to complete the solution,
V we have to prove that for any s ∈
Γ(E ⊗ Tr0 (M )), we have s ∈ Γ(E ⊗ r M ) iff s as a V
C ∞ (M )-multilinear map
Γ(T M )(r) →VΓE is alternating. Now s ∈ Γ(E ⊗ r M ) is equivalent with
s(p) ∈ Ep ⊗ r (Tp∗ M ), ∀p ∈ M , and on the other hand s : Γ(T M )(r) → ΓE
is alternating iff
s(Xσ(1) , . . . , Xσ(r) )(p) = (sgn σ) · s(X1 , . . . , Xr )(p),
∀σ ∈ Sr , X1 , . . . , Xr ∈ Γ(T M ), p ∈ M.

33Indeed, using standard bundle charts, this boils down to the local fact that U ×
Vr
((Rn )∗ ) is a subbundle of U × (Rn ⊗ · · · ⊗ Rn )∗ (for U open ⊂ M ).
34General fact (complement to Problems 39(a) and 41): If E , E , F are vector bundles
1 2
over M and E1 is a subbundle of E2 , then E1 ⊗ F is a subbundle of E2 ⊗ F .
160 ANDREAS STRÖMBERGSSON

Hence, by also using (138) and Problem 35(c) (applied to the vector bundle
T M ) we see that it suffices to prove the following:
V For every p ∈ M and
every z ∈ Ep ⊗ Tr0 (M )p , we have z ∈ Ep ⊗ r (Tp∗ M ) iff
Cp (z ⊗ (vσ(1) ⊗ · · · ⊗ vσ(r) )) = (sgn σ) · Cp (z ⊗ (v1 ⊗ · · · ⊗ vr )),
(140) ∀σ ∈ Sr , v1 , . . . , vr ∈ Tp M.
Proof of the last claim: Let b1 , . . . , bP n be a basis of Ep . Then every z ∈
Ep ⊗Tr0 (M )p can be expressed as z = nj=1 bj ⊗ηj with uniquely determined
V V
η1 , . . . , ηn ∈ Tr0 (M )p , and we have z ∈ Ep ⊗ r (Tp∗ M ) iff ηj ∈ r (Tp∗ M ) for
all j. It now follows from the definition of Cp , (139), that
n
X
Cp (z ⊗ α) = ηj (α) · bj , ∀α ∈ T0r (M )p .
j=1

Hence since b1 , . . . , bn is a basis of Ep , we see that (140) holds iff

ηj (vσ(1) ⊗ · · · ⊗ vσ(r) ) = ηj (v1 ⊗ · · · ⊗ vr ),
∀j ∈ {1, . . . , n}, σ ∈ Sr , v1 , . . . , vr ∈ Tp M.
V
In other words, (140) is equivalent with ηj ∈ r (Tp∗ M ) forVall j, and as we
have already pointed out this is equivalent with z ∈ Ep ⊗ r (Tp∗ M ). Done!


Addendum: Note that for any s ∈ Ωr (E) of the form s = µ ⊗ ω with

µ ∈ Γ(E) and ω ∈ Ωr (M ) (and more generally for any s ∈ Γ(E ⊗ Tr0 M ) of
the form s = µ ⊗ ω with µ ∈ Γ(E) and ω ∈ Γ(Tr0 M )), the corresponding
multilinear map satisfies
(141) (µ ⊗ ω)(X1 , . . . , Xr ) = ω(X1 , . . . , Xr ) · µ (in ΓE),
∀X1 , . . . , Xr ∈ Γ(T M ).
Indeed, this is clear from (138) and (139). 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 161

(b). The expressions on both sides in the given formula clearly depends
C ∞ (M )-linearly on s1 and C ∞ (M )-linearly on s2 . Hence, by the argument
at the beginning of the solution to Problem 49(a), it suffices to prove the
stated formula when sj = µj ⊗ ωj (j = 1, 2), with µj ∈ Γ(Ej ) and ω1 ∈
Ωr (M ), ω2 ∈ Ωs (M ). In this case we get by (141):
1 X
sgn(σ) s1 (Xσ(1) , . . . , Xσ(r) ) ⊗ s2 (Xσ(r+1) , . . . , Xσ(r+s) )
r!s!
σ∈Sr+s
1 X    
= sgn(σ) ω1 (Xσ(1) , . . . , Xσ(r) ) · µ1 ⊗ ω2 (Xσ(r+1) , . . . , Xσ(r+s) ) · µ2 ,
r!s!
σ∈Sr+s
 X 
1
= sgn(σ) ω1 (Xσ(1) , . . . , Xσ(r) ) · ω2 (Xσ(r+1) , . . . , Xσ(r+s) ) · µ1 ⊗ µ2 .
r!s!
σ∈Sr+s

By the definition of exterior product on Ω(M ), this is

= (ω1 ∧ ω2 )(X1 , . . . , Xr+s ) · (µ1 ⊗ µ2 ).
But we have s1 ∧ s2 = (µ1 ⊗ µ2 ) ⊗ (ω1 ∧ ω2 ), and so by one more application
of (141), the above is
= (s1 ∧ s2 )(X1 , . . . , Xr+s ).
Done! 

Finally, the analogous formula for s1 · s2 is:

(s1 · s2 )(X1 , . . . , Xr+s )
1 X
= sgn(σ) s1 (Xσ(1) , . . . , Xσ(r) ) · s2 (Xσ(r+1) , . . . , Xσ(r+s) ),
r!s!
σ∈Sr+s
∀X1 , . . . , Xr+s ∈ Γ(T M ),
162 ANDREAS STRÖMBERGSSON

Problem 52:
(a). This was proved in the lecture.
(b). We first prove that the connection D|U , if it exists, is unique. Let
s ∈ Γ(E|U ) be given. We claim that if V is any open subset of U and
s′ ∈ Γ(E) is such that s′|V = s|V , then

(142) (D|U s)|V = (Ds′ )|V .

By Problem 35(a), U can be covered by open sets V for which such a section
s′ ∈ Γ(E) exists; hence (142) implies that the whole section D|U s is uniquely
determined.
[Proof of (142): It follows from the requirement on D|U that
(143) D|U (s′|U ) = (Ds′ )|U
Furthermore, applying part (a) to D|U , and using (s′|U )|V = s′|V = s|V , we
have
(144) (D|U (s′|U ))|V = (D|U s)|V .
Combining (143) and (144) we get (142).]
Next let us verify that there indeed exists a well-defined section “D|U s” in
Γ(E|U ⊗ T ∗ U ) such that (142) holds whenever V is open in U and s′ ∈ Γ(E)
is such that s′|V = s|V . For this, it suffices to verify that for any two open
subset V1 , V2 ⊂ U and any s′1 , s′2 ∈ Γ(E), if s′j|Vj = s|Vj for j = 1, 2 then
(Ds′1 )|V1 ∩V2 = (Ds′2 )|V1 ∩V2 . However this is clear from (s′1 )|V1 ∩V2 = s|V1 ∩V2 =
(s′2 )|V1 ∩V2 , and part (a) of the lemma (i.e. the fact that D is local).
Hence we have shown that our requirements on D|U imply that D|U is
a uniquely defined map from Γ(E|U ) to Γ(E|U ⊗ T ∗ U ). Note that it is
immediate from our construction that this map has the desired property,
i.e. that (Ds)|U = D|U (s|U ) for all s ∈ Γ(E). (Indeed, let s ∈ Γ(E) be given.
Then for the section s|U ∈ Γ(E|U ), (142) applies with s′ = s and V = U ,
and then says that (D|U (s|U ))|U = (Ds)|U , as desired.)
It remains to prove that D|U is a connection on E|U . Clearly D|U is
R-linear, and so it remains to prove that
(145) D|U (f s) = s ⊗ df + f D|U s, ∀f ∈ C ∞ (U ), s ∈ Γ(E|U ).
For this let f ∈ C ∞ (U ) and s ∈ Γ(E|U ) be given. It suffices to prove that
for any given point p ∈ U , there is an open subset V ⊂ U with p ∈ V such
that
(D|U (f s))|V = (s ⊗ df + f D|U s)|V .
However, given p ∈ U , we know by Problem 35(a) (applied to the two vector
bundles E and M × R) that there exist f ′ ∈ C ∞ (M ) and s′ ∈ Γ(E) such
 PROBLEMS; “RIEMANNIAN GEOMETRY” 163

′ = f ′
that f|V |V and s|V = s|V for some open set V ⊂ U containing p. Then
also (f ′ s′ )|V = (f s)|V , and now
(D|U (f s))|V = (D(f ′ s′ ))|V = (s′ ⊗ df ′ + f ′ Ds′ )|V = s′|V ⊗ d(f|V
′ ′
) + f|V (Ds′ )|V
= s|V ⊗ d(f|V ) + f|V (D|U s)|V = (s ⊗ df + f D|U s)|V ,
as desired! Hence we have proved (145), and so D|U is a connection on E|U .

(c) The requirement on D is that
(146) (Ds)|Uα = Dα (s|Uα ), ∀s ∈ Γ(E), α ∈ A.
Let s ∈ Γ(E) be given. Since ∪α∈A Uα = M , the condition (146) determines
Ds uniquely (if it exists at all). To prove that there really exists a section
Ds ∈ Γ(E ⊗ T ∗ M ) which satisfies (146), it suffices to verify that for any
two α, β ∈ A with V := Uα ∩ Uβ 6= ∅, we have (Dα (s|Uα ))V = (Dβ (s|Uβ ))V .
However this is immediate from our assumption that (Dα )|V = (Dβ )|V .
Hence we have shown that our requirement on D implies that D is a
uniquely defined map from Γ(E) to Γ(E ⊗ T ∗ M ). Clearly this map is R-
linear. In order to prove that D is a connection, it remains to verify that
D(f s) = s ⊗ df + f Ds for all f ∈ C ∞ (M ), s ∈ Γ(E). Thus let f ∈ C ∞ (M )
and s ∈ Γ(E) be given. Since ∪α∈A Uα = M , it suffices to prove that
(D(f s))|Uα = (s ⊗ df + f Ds)|Uα for every α ∈ A. Thus let α ∈ A be given.
Then
(D(f s))|Uα = Dα ((f s)|Uα ) = s|Uα ⊗ df|Uα + f|Uα Dα (s|Uα )
= (s ⊗ df + f Ds)|Uα ,
as desired! 
Remark 1. The following (apriori) stronger version of part (b) is actually technically
slightly more direct to prove:
Let (Uα )α∈A be an open covering of M , and for each α ∈ A let Dα be a connection
on E|Uα . Assume that (Dα (s|Uα ))|Uα ∩Uβ = (Dβ (s|Uβ ))|Uα ∩Uβ for all s ∈ ΓE and all
α, β ∈ A. Then there exists a unique connection D on E such that (Ds)|Uα = Dα (s|Uα )
for all s ∈ ΓE and α ∈ A.

Proof. Given any s ∈ ΓE, the section Ds ∈ Γ(T ∗ M ⊗ E) is clearly uniquely determined
(if it exists) by the given requirement, since M = ∪α∈A Uα . On the other hand the given
compatibility assumption easily implies that Ds is indeed as well-defined (C ∞ !) section
of T ∗ M ⊗ E! Hence we obtain a well-defined map from Γ(E) to Γ(T ∗ M ⊗ E). The rest
is as above! 
164 ANDREAS STRÖMBERGSSON

Problem 53: Let p ∈ M be the base point of v, so that v ∈ Tp M .

Take an open neighborhood U ⊂ M of p such that both T M|U = T U
and E|U are trivial, and choose bases of sections X1 , . . . , Xd ∈ Γ(T U ) and
σ1 , . . . , σn ∈ Γ(E|U ). Let Γkij ∈ C ∞ (U ) be the corresponding Christoffel
symbols, so that DXi σj = Γkij σk for all i ∈ {1, . . . , d} and j ∈ {1, . . . , n}.
Take a1 , . . . , an ∈ C ∞ (U ) and b1 , . . . , bn ∈ C ∞ (U ) so that
s1|U = aj σj and s2|U = bj σj
(cf. Problem 34). Also take γ 1 , . . . , γ n ∈ R so that
v = γ i · Xi (p) ∈ Tp M.
Then as shown in Lecture #9, we have
(147) Dv (s1 ) = v(ak ) · σk (p) + γ j · ak (p) · Γℓjk (p) · σℓ (p).
and
(148) Dv (s2 ) = v(bk ) · σk (p) + γ j · bk (p) · Γℓjk (p) · σℓ (p).
[Details: In Lecture #9 we noted a formula saying that if X = γ i Xi ∈ Γ(T U )
then DX (s1 ) = X(ak ) · σk + γ j ak Γℓjk σℓ in Γ(E|U ), and evaluating this section
at p we get (147). But of course we don’t need to refer to Lecture #9; the
proof of (147) is immediate using Leibniz’ rule: We have
Dv (s1 ) = Dv (ak σk ) = v(ak ) · σk (p) + ak (p) · Dv (σk ),


and here Dv (σk ) = γ i DXi (p) (σk ) = γ i DXi (σk ) (p) = γ i Γℓik (p)σℓ (p), and
combining these two we get (147). The proof of (148) is the same.]
Now we are assuming v = ċ(0); hence (by “fact” on p. 9 in Lecture #2;
cf. Problem 13(f)):
v(aj ) = (aj ◦ c)′ (0) and v(bj ) = (bj ◦ c)′ (0).
We are also assuming that for every t ∈ (−ε, ε) we have s1 (c(t)) = s2 (c(t)),
i.e.
aj (c(t)) · σj (c(t)) = bj (c(t)) · σj (c(t)) in Ec(t) ,
and thus
aj (c(t)) = bj (c(t)), ∀j ∈ {1, . . . , n}, t ∈ (−ε, ε).
Combining the above facts, we conclude that
v(aj ) = v(bj ), ∀j ∈ {1, . . . , n}.
Also of course aj (p) = aj (c(0)) = bj (c(0)) = bj (p). Hence we see by inspec-
tion in (147) and (148) that Dv (s1 ) = Dv (s2 ). 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 165

Problem 54: The map d is clearly R-linear. Furthermore, for any f ∈

C ∞U ,
d(f ak sk ) = sk ⊗ d(f ak ) = sk ⊗ (ak · df + f · dak )
= (ak sk ) ⊗ df + f · sk ⊗ dak
= (ak sk ) ⊗ df + f · d(ak sk ).
Hence d is indeed a connection. 

Problem 55: The first statement is an immediate verification; indeed

the restricted map is, by assumption, a map D : ΓE ′ → Γ(E ′ ⊗ T ∗ M ), and
it satisfies D(f s) = f · Ds + s ⊗ df for all f ∈ C ∞ (M ), s ∈ ΓE ′ , since this
holds more generally when s ∈ ΓE.
Here’s a simple example showing that the condition is not always satisfied:
Let E be the trivial vector bundle E = M × R2 over M = R, equipped
with the corresponding ’trivial’ connection D (i.e. the connection which Jost
would call “d” with respect to the bundle chart ϕ = 1E : E → M × R2 ). Set
E ′ = {(x, v) ∈ E : v ∈ R(cos x, sin x)}.
This is easily verified to be a vector subbundle of E. Now consider the
section s ∈ ΓE ′ , s(x) := (cos x, sin x). Then Ds(x) = (− sin x, cos x) in
Γ(E ⊗ T ∗ M ) = ΓE. (Note: For our M = R we have T ∗ M = M × R and
thus E ⊗ T ∗ M = E under obvious identifications.) Hence Ds is not in
Γ(E ′ ⊗ T ∗ M ) = Γ(E ′ ); indeed Ds(x) ∈
/ E ′ for every x ∈ M . 

Problem 57: The requirement on f ∗ D is:

(149) (f ∗ D)(s ◦ f ) = Ddf (·) (s) ∈ Γ(Hom(T M, f ∗ E)), ∀s ∈ ΓE
Let us first verify that this formula makes sense! For any s ∈ ΓE we have
s ◦ f ∈ Γf E = Γf ∗ E (cf. Problem 44); hence if f ∗ D is a connection on f ∗ E
then we should indeed have
(f ∗ D)(s ◦ f ) ∈ Γ(f ∗ E ⊗ T ∗ M ) = Γ(Hom(T M, f ∗ E)).
Next let us prove that Ddf (·) (s), i.e. the map v 7→ Ddf (v) (s), is indeed a
bundle homomorphism T M → E along f , so that it can be viewed as an el-
ement of Γ(Hom(T M, f ∗ E)) by Problem 45. First of all, df is a C ∞ function
T M → T N (cf. Problem 17(a)); also D(s) ∈ Γ(Hom(T N, E)) and so D(s)
can be viewed as a bundle homomorphism T N → E (cf. Problem 40). Now
Ddf (·) (s) is the composition of these two C ∞ maps, hence itself C ∞ . Also
Ddf (·) (s) clearly restricts to an R-linear map Tp M → Ef (p) for all p ∈ M .
Hence Ddf (·) (s) is indeed a bundle homomorphism T M → E along f . Done!
By Problem 44(c), every section in Γf ∗ E be written as a finite C ∞ (M )-
linear combination of sections of the form s ◦ f with s ∈ ΓE. Hence the
166 ANDREAS STRÖMBERGSSON

connection f ∗ D, if it exists at all, is uniquely determined by the relation

(149).
It remains to prove that there exists such a connection f ∗ D. Let us first
prove the existence of f ∗ D in the special case when (E, π, N ) is a trivial
vector bundle. Then fix a basis of sections s1 , . . . , sn ∈ ΓE (cf. Problem 33).
Then for any σ ∈ Γf ∗ E there exist unique ’coefficient functions’ α1 , . . . , αn ∈
C ∞ (M ) such that
n
X
σ= αj · (sj ◦ f )
j=1

(cf. Problem 44(b) and Problem 34), and we now define the map
e : Γf ∗ E → Γ(f ∗ E ⊗ T ∗ M )
D
by setting
n 
X 
(150) e
D(σ) := αj · Ddf (·) (sj ) + (sj ◦ f ) ⊗ dαj .
j=1

(Here Ddf (·) (sj ) ∈ Γ(Hom(T M, f ∗ E)) = Γ(f ∗ E ⊗ T ∗ M ) as in (149).) This

map D e is clearly well-defined and R-linear. Furthermore, for any g ∈
P
C (M ) and σ = nj=1 αj · (sj ◦ f ) as above, we have
∞

n
X
g·σ = (g · αj ) · (sj ◦ f ),
j=1

and hence by definition:

n 
X 
e · σ) =
D(g gαj · Ddf (·) (sj ) + (sj ◦ f ) ⊗ d(gαj )
j=1
n o
Use d(gαj ) = g · dαj + αj · dg.
n 
X  n
X
=g· αj · Ddf (·) (sj ) + (sj ◦ f ) ⊗ dαj + αj (sj ◦ f ) ⊗ dg
j=1 j=1
e
= g · D(σ) + σ ⊗ dg,
e is a connection! Finally we prove that D
This proves that D e satisfies (149).
. . . , βn ∈ C ∞ (N ) such
∈ ΓE be given. Then there exist unique β1 ,P
Thus let sP
that s = j=1 βj sj (cf. Problem 34); hence s ◦ f = nj=1 (βj ◦ f ) · (sj ◦ f );
n

e
and so by our definition of D,
n 
X 
(151) e ◦ f) =
D(s (βj ◦ f ) · Ddf (·) (sj ) + (sj ◦ f ) ⊗ d(βj ◦ f ) .
j=1
 PROBLEMS; “RIEMANNIAN GEOMETRY” 167

On the other hand for any p ∈ M and v ∈ Tp M we have

X
n 
Ddf (v) (s) = Ddf (v) βj sj
j=1
n 
X 
= βj (f (p)) · Ddf (v) (sj ) + [dfp (v)](βj ) · sj (f (p)) .
j=1

Here

[dfp (v)](βj ) = (dβj )f (p) (dfp (v)) = [d(βj ◦ f )](v) ∈ R

(by the chain rule), and hence by comparing the last two formulas we see
e ◦ f ) = Ddf (·) (s). Hence we have proved that the connection D
that D(s e
∗ ∗
satisfies the relation required for f D, (149). This proves that f D exists,
namely f ∗ D = D! e
We have thus proved that the pullback of any connection on a trivial
vector bundle exists; and it is unique since we have noted that the pullback
of any connection is unique if it exists.
Finally we will prove that the pullback connection f ∗ D exists for an arbi-
trary vector bundle (E, π, N ). Fix an open covering (Vα )α∈A of N such that
E|Vα is trivial for each α ∈ A. Let Uα = f −1 (Vα ); then (Uα )α∈A is an open
covering of M . Now f|Uα is a C ∞ map of manifolds Uα → Vα , and D|Vα is a
connection on the trivial vector bundle E|Vα (cf. Problem 52); hence by what
we have proved above, there exists a uniquely defined pullback connection
De α := f ∗ (D|V ) on f ∗ (E|V ) = (f ∗ E)|U . 35 Let us prove that these
|Uα α |Uα α α
e
connections Dα (α ∈ A) are compatible in the appropriate sense. Thus let
α, β ∈ A and set U ′ := Uα ∩ Uβ ; assume U ′ 6= ∅. Note that U ′ = f −1 (V ′ )
where V ′ := Vα ∩ Vβ . We claim that

(152) e α|U ′ (s′ ◦ f|U ′ ) = Ddf (·) (s′ ),

D ∀s′ ∈ ΓE|V ′ .
|U ′

(Here in the right hand side, “D” really stands for “D|U ′ ”; we will use this
type of mild abuse of notation several times in the discussion below.) To
prove (152), let s′ ∈ ΓE|V ′ be given, and take p ∈ U ′ . By Problem 35(a),
there is a section s ∈ ΓE|Vα such that s|V ′′ = s′|V ′′ for some open set V ′′ ⊂ V ′
containing f (p). Then s ◦ f|Uα and s′ ◦ f|U ′ have the same restrictions to
U ′′ := f −1 (V ′′ ), i.e. (s ◦ f|Uα )|U ′′ = (s′ ◦ f|U ′ )|U ′′ . Note also that p ∈ U ′′ .

35Prove this identification, “f ∗ (E ∗

|Uα |Vα ) = (f E)|Uα ”, as a complement to Problem 42.
168 ANDREAS STRÖMBERGSSON

Now we get:
e α|U ′ (s′ ◦ f|U ′ )|U ′′
D
 
e α (s ◦ f|U ))|U ′′ since (s ◦ f|Uα )|U ′′ = (s′ ◦ f|U ′ )|U ′′ ; cf. the
=(D α
solution to Problem 52(b).

=(Ddf|Uα (·) (s))|U ′′ eα.
by the defining condition for D
 

 indeed for every v ∈ T (U ′′ ) we have df (v) ∈ 

 
T (V ′′ ) and thus Ddf (v) (s) = Ddf (v) (s′ ), since
=(Ddf|U ′ (·) (s′ ))|U ′′ .

 D(s)|V ′′ = D(s′ )|V ′′ (cf. the solution to Prob- 

 
lem 52(b)).
Since every p ∈ U ′ has such a neighborhood U ′′ , it follows that (152) holds!
But (152) says exactly that D e α|U ′ is the f|U ′ -pullback of D|V ′ (which we
know is unique if it exists). Changing the roles of α and β, it also follows
that D e β|U ′ is the f|U ′ -pullback of D|V ′ . Hence by the uniqueness of “the
f|U ′ -pullback of D|V ′ ”, we conclude:

(153) e α|U ′ = D
D e β|U ′ .

The fact that (153) holds for all α, β ∈ A with Uα ∩ Uβ 6= ∅ implies by

e on f ∗ E satisfying
Problem 52(c) that there exists a unique connection D
e |U = D
D e α for all α ∈ A. One easily proves that
α

(154) e ◦ f ) = Ddf (·) (s),

D(s ∀s ∈ ΓE.

(Indeed, let s ∈ ΓE be given. Then for every α ∈ A we have (D(s e ◦ f ))|U =

α

De α ((s ◦ f )|U ) = D
e α (s|V ◦ f|U ) = Ddf (·) (s|V ) = Ddf (·) (s)|U , where
α α α |Uα α α
each step is justified by arguments similar to those in the proof of (152).
The fact that (D(s e ◦ f ))|U = Ddf (·) (s)|U for each α ∈ A implies that
α α
e ◦ f ) = Ddf (·) (s). Done!)
D(s
e satisfies the re-
The relation (154) says exactly that the connection D
quirements on the pullback bundle f D; hence we have proved that f ∗ D
∗

exists! 

(b). Let us first prove that the connection f ∗ D defined in part (a) indeed
satisfies the stated condition. Thus let s ∈ Γf ∗ E = Γf E and let c : (−ε, ε) →
M be a C ∞ curve; also let s1 ∈ ΓE, and assume that s1 (f (c(t))) = s(c(t))
for all t ∈ (−ε, ε). The assumption means that the two sections s1 ◦ f and
s in Γf ∗ E are equal along the curve c. Hence by Problem 53,
(f ∗ D)ċ(0) (s) = (f ∗ D)ċ(0) (s1 ◦ f ).
 PROBLEMS; “RIEMANNIAN GEOMETRY” 169

But also, by the defining property of f ∗ D,

(f ∗ D)ċ(0) (s1 ◦ f ) = Ddf (ċ(0)) (s1 ).
Hence (f ∗ D)ċ(0) (s) = Ddf (ċ(0)) (s1 ), and so we have proved that f ∗ D satisfies
the desired condition.
Next let us prove that f ∗ D is the only connection on f ∗ E which satisfies
the stated condition. Thus let ∇ be any connection on f ∗ E such that for
any s ∈ Γf ∗ E = Γf E, any s1 ∈ ΓE, and any curve c : (−ε, ε) → M , if
s1 (f (c(t))) = s(c(t)) (∀t ∈ (−ε, ε)), then ∇ċ(0) (s) = Ddf (ċ(0)) (s1 ).
Consider an arbitrary section s1 ∈ ΓE and an arbitrary v ∈ T M . Then
there is a C ∞ curve c : (−ε, ε) → M such that v = ċ(0). Note that s1 ◦ f ∈
Γf ∗ E and obviously s1 (f (c(t))) = (s1 ◦ f )(c(t)) for all t ∈ (−ε, ε). Hence by
our assumption, ∇ċ(0) (s1 ◦ f ) = Ddf (ċ(0)) (s1 ), i.e.
∇v (s1 ◦ f ) = Ddf (v) (s1 ).
The fact that this holds for all s1 ∈ ΓE and all v ∈ T M means exactly that
∇ satisfies the defining condition for “f ∗ D”; hence by the uniqueness proved
in part (a) we must have ∇ = f ∗ D. 
170 ANDREAS STRÖMBERGSSON

(c). Let V ⊂ N be an open set containing q such that there exists a basis
of sections s1 , . . . , sn ∈ ΓE|V . Set U = f −1 (V ) ⊂ M . Then, given s ∈ Γf ∗ E,
there exist unique α1 , . . . , αn ∈ C ∞ (U ) such that
X n
s|U = αj · (sj ◦ f|U )
j=1

(cf. Problem 44(b) and Problem 34). Now by the definition of f ∗ D (cf. part
a),
n 
X 
(f ∗ D)(s)|U = (sj ◦ f|U ) ⊗ dαj + αj · Ddf (·) (sj ) in Γ((f ∗ E ⊗ T ∗ M )|U ).
j=1
In particular,
n 
X 
(f ∗ D)ċ(0) (s) = (αj ◦ c)′ (0) · sj (q) + αj (c(0)) · Ddf (ċ(0)) (sj ) .
j=1

Here df (ċ(0)) = 0 since f ◦ c is constant; thus we are left with:

n
X
(155) (f ∗ D)ċ(0) (s) = (αj ◦ c)′ (0) · sj (q).
j=1

On the other hand we have s(c(t)) = αj (c(t)) · sj (q) for all t ∈ (−ε, ε), and
d
hence the right hand side of (155) equals the tangent vector ( dt (s ◦ c)(t))|t=0
in Ts(c(0)) (Eq ) = Eq . Done! 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 171

Problem 58: The general structure of the following proof is very similar
to the solution to Problem 57(a).
The requirement on D is
(156) D(µ ⊗ ν) = (D1 µ) ⊗ ν + µ ⊗ (D2 ν), ∀µ ∈ ΓE1 , ν ∈ ΓE2 .
By the definition of tensor product (of C ∞ (M )-modules), every element in
Γ(E1 ) ⊗ Γ(E2 ) can be written as a finite sum of pure tensors µ1 ⊗ µ2 (µ1 ∈
ΓE1 , µ2 ∈ ΓE2 ); hence by Problem 43(d), the same is true for any section in
Γ(E1 ⊗ E2 ). (Note that a main portion of the solution to Problem 43(d) was
spent on proving exactly this fact.) Hence the formula (156), together with
the requirement that D should be R-linear (or merely additive) certainly
makes the connection D uniquely defined, if it exists at all.
Thus it remains to prove that there exists such a connection D. Let us
first prove the existence of D in the special case when both E1 and E2 are
trivial vector bundles. Then fix a basis of sections µ1 , . . . , µn ∈ ΓE1 and a
basis of sections ν1 , . . . , νm ∈ ΓE2 (cf. Problem 33; we set n = rank E1 and
m = rank E2 ). Then {µi ⊗ νj } (with i ∈ {1, . . . , n}, j ∈ {1, . . . , m}) is a
basis of sections in Γ(E1 ⊗ E2 ), and so for any s ∈ Γ(E1 ⊗ E2 ) there exist
unique ’coefficient functions’ αi,j ∈ C ∞ (M ) such that
n X
X m
(157) s= αi,j µi ⊗ νj ,
i=1 j=1

and we now define the map

D : Γ(E1 ⊗ E2 ) → Γ(E1 ⊗ E2 ⊗ T ∗ M )

by setting36
m 
n X
X 
D(s) := µi ⊗ νj ⊗ dαi,j + αi,j (D1 µi ) ⊗ νj + αi,j µi ⊗ (D2 νj ) .
i=1 j=1

(Here we use the natural isomorphism between Γ(E1 ⊗ T ∗ M ⊗ E2 ) and

Γ(E1 ⊗ E2 ⊗ T ∗ M ) to identify (D1 µi ) ⊗ νj ∈ Γ(E1 ⊗ T ∗ M ⊗ E2 ) with an
element in Γ(E1 ⊗ E2 ⊗ T ∗ M ).) This map D is clearly well-defined and
R-linear. Furthermore, for any f ∈ C ∞ (M ) and s as in (157), we have
n X
X m
fs = f αi,j µi ⊗ νj ,
i=1 j=1

36Note that this definition of D apriori depends on the choice of the bases of sections
µ1 , . . . , µn and ν1 , . . . , νn ; however we will soon prove that D is a connection satisfying
(156), and then it follows that our D is in fact independent of the choices of µ1 , . . . , µn and
ν1 , . . . , νn , since we noted from start that there exists at most one connection satisfying
(156)!
172 ANDREAS STRÖMBERGSSON

and hence by our definition,

Xn Xm  
D(f s) = µi ⊗ νj ⊗ d(f αi,j ) + f αi,j (D1 µi ) ⊗ νj + f αi,j µi ⊗ (D2 νj )
i=1 j=1

use d(f αi,j ) = αi,j df + f dαi,j and our formula for D(s)
n X
X m
= f · D(s) + αi,j µi ⊗ νj ⊗ df
i=1 j=1

= f · D(s) + s ⊗ df.
This proves that D is a connection! We next prove that D satisfies (156).
Thus let µ ∈ ΓE1 and ν ∈ P
ΓE2 be given. Then there exist unique α1 , . . . , αn ∈
n
C ∞ (M ) such that µ = Pm i=1 αi µi , and there exist unique β1 , . . . , βm ∈
∞
C (M ) such that ν = j=1 βj νj . Then
n X
X m
µ⊗ν = αi βj µi ⊗ νj ,
i=1 j=1

and hence by our definition,

Xn Xm  
D(µ ⊗ ν) = µi ⊗ νj ⊗ d(αi βj ) + αi βj (D1 µi ) ⊗ νj + αi βj µi ⊗ (D2 νj )
i=1 j=1
Xn X m 
= µi ⊗ (βj νj ) ⊗ dαi + (αi D1 µi ) ⊗ (βj νj )
i=1 j=1

+ (αi µi ) ⊗ νj ⊗ dβj + αi µi ⊗ (βj D2 νj )
X m
n X n X
X m
= (D1 (αi µi )) ⊗ (βj νj ) + (αi µi ) ⊗ (D2 (βj νj ))
i=1 j=1 i=1 j=1

= (D1 µ) ⊗ ν + µ ⊗ (D2 ν).

Hence we have proved that D satisfies (156). This completes the proof in
the special case when both E1 and E2 are trivial vector bundles.
Finally we will prove that the connection D on Γ(E1 ⊗ E2 ) exists when
E1 , E2 are arbitrary vector bundles over M . Fix an open covering (Uα )α∈A
of M such that both E1|Uα and E2|Uα are trivial for all α ∈ A. Then by
what we have proved above, for each α ∈ A there exists a unique connection
Dα on (E1 ⊗ E2 )|Uα = E1|Uα ⊗ E2|Uα satisfying
(158) Dα (µ ⊗ ν) = (D1 µ) ⊗ ν + µ ⊗ (D2 ν), ∀µ ∈ ΓE1|Uα , ν ∈ ΓE2|Uα .
(Here “D1 ” really stands for D1|Uα and similarly for D2 ; cf. Problem 52.)
Now one proves that these connections Dα are compatible in the sense that
(Dα )|Uα ∩Uβ = (Dβ )|Uα ∩Uβ for all α, β ∈ A. (We leave out the details for
 PROBLEMS; “RIEMANNIAN GEOMETRY” 173

this; but cf. the solution to Problem 57 where we give a detailed proof of the
same kind of compatibility in a different situation.) Hence by Problem 52(c),
there exists a unique connection D on Γ(E1 ⊗ E2 ) satisfying D|Uα = Dα for
all α ∈ A. One easily proves that this connection D satisfies (156). This
completes the proof. 
174 ANDREAS STRÖMBERGSSON

Problem 59: (a). By the now familiar argument (cf., e.g., the beginning
of the solution to Problem 58), any section in Γ(E1 ⊗ E2 ) can be written as
a finite sum of pure tensor sections s1 ⊗ s2 with s1 ∈ ΓE1 and s2 ∈ ΓE2 .
The analogous fact holds for Γ(E1∗ ⊗ E3 ). Hence, by R-linearity, it suffices
to prove the desired formula when α = s1 ⊗ s2 and β = u ⊗ s3 for some
s1 ∈ ΓE1 , s2 ∈ ΓE2 , u ∈ ΓE1∗ , s3 ∈ ΓE3 . In this case,
(α, β) = (s1 ⊗ s2 , u ⊗ s3 ) = (s1 , u) · s2 ⊗ s3 ,
and hence
D(α, β) = s2 ⊗ s3 ⊗ d(s1 , u) + (s1 , u) · D(s2 ⊗ s3 )
   
= s2 ⊗ s3 ⊗ (Ds1 , u) + (s1 , Du) + (s1 , u) · (Ds2 ) ⊗ s3 + s2 ⊗ (Ds3 )
 
= (Ds1 ) ⊗ s2 + s1 ⊗ (Ds2 ) , u ⊗ s3
 
+ s1 ⊗ s2 , (Du) ⊗ s3 + u ⊗ (Ds3 )
= (Dα, β) + (α, Dβ).
Done! 

(b). We have standard identifications Γ(Hom(E2 , E3 )) = Γ(E2∗ ⊗ E3 )

and Γ(Hom(E1 , E2 )) = Γ(E1∗ ⊗ E2 ), and under these identifications, the
composition α ◦ β is the section in Γ(E1∗ ⊗ E3 ) obtained by contracting the
E2∗ -part of α against the E2 -part of β. (Cf. the solution to Problem 50 where
this is discussed in matrix notation.) Hence the stated formula is equivalent
to the formula proved in part a. (Of course the formula in part a remains
true regardless of the exact ordering of the factors in the tensor products.)


(c). We have the standard identification Γ(Hom(E1 , E2 )) = Γ(E1∗ ⊗ E2 ),

and under this identification the composition α ◦ β is the section in Γ(E2 )
obtained by contracting the E1∗ -part of α against β. Hence the stated for-
mula is a special case of the formula in part a (namely: take E3 = M × R
and replace E1 by E1∗ in the formula in part a). 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 175

Problem 60:
(a). The general structure of the following proof is again similar to the
solutions of Problems 57(a) and 58.
The requirement on D : Ωp (E) → Ωp+1 (E) is R-linearity and
(159) D(µ ⊗ ω) = (Dµ) ∧ ω + µ ⊗ dω, ∀µ ∈ ΓE, ω ∈ Ωp (M ).
We call such a map an exterior covariant derivative
V with respect to the
connection D. It follows from Ωp (E) = Γ(E ⊗ p M ) = Γ(E) ⊗C ∞ M Ωp (M )
(cf. Problem 43(d)) that every s ∈ Ωp (E) can be written as a finite sum
of pure tensors s ⊗ ω (s ∈ Γ(E), ω ∈ Ωp (M )). Hence, since D is required
to be R-linear (in particular additive), the condition (159) certainly makes
D : Ωp (E) → Ωp+1 (E) uniquely determined, if it exists at all.
Thus it remains to prove that there exists such an exterior covariant
derivative. We start by proving three lemmata:
Lemma 6. Let (Uα )α∈A be an open covering of M , and for each α ∈ A let
Dα : Ωp (E|Uα ) → Ωp+1 (E|Uα ) be an exterior covariant derivative wrt the con-
nection D|Uα on E|Uα . Assume that (Dα (s|Uα ))|Uα ∩Uβ = (Dβ (s|Uβ ))|Uα ∩Uβ
for all s ∈ Ωp (E) and all α, β ∈ A. Then there exists a unique R-linear map
D : Ωp (E) → Ωp+1 (E) satisfying (Ds)|Uα = Dα (s|Uα ) for all s ∈ Ωp (E) and
α ∈ A, and this map is an exterior covariant derivative wrt D.

This lemma is proved by the same type of arguments as in the solution

to Problem 52 (cf. in particular Remark 1).
Lemma 7. Let U2 ⊂ U1 be open subsets of M , and for j = 1, 2 let
Dj : Ωp (E|Uj ) → Ωp+1 (E|Uj ) be an exterior covariant derivative wrt the
connection D|Uj . Then
(160) (D1 s)|U2 = D2 (s|U2 ), ∀s ∈ Ωp (E|U1 ).

Proof. Since every s ∈ Ωp (E|U1 ) can be written as a finite sum of pure

tensors µ ⊗ ω, where µ ∈ Γ(E|U1 ) and ω ∈ Ωp (U1 ), it suffices to prove (160)
when s is such a pure tensor; s = µ ⊗ ω. But then
(D1 s)|U2 = (D1 (µ ⊗ ω))|U2 = ((D|U1 µ) ∧ ω + µ ⊗ dω)|U2
= (D|U1 µ)|U2 ∧ ω|U2 + µ|U2 ⊗ d(ω|U2 )
= (D|U2 µ|U2 ) ∧ ω|U2 + µ|U2 ⊗ d(ω|U2 )
= D2 (µ|U2 ⊗ ω|U2 ) = D2 (s|U2 ).
(In the above computation we used the fact that (D|U1 )|U2 = D|U2 – this
fact is immediate from the solution of Problem 52. 
Lemma 8. Let U be any open subset of M such that both the vector bundles
T U and E|U are trivializable. Then there exists a unique exterior covariant
e : Ωp (E|U ) → Ωp+1 (E|U ) wrt D|U .
derivative D
176 ANDREAS STRÖMBERGSSON

Proof. The assumptions

imply that there exists a basis of sections ω1 , . . . , ωr
for Ωp (U ) (r = dp ) and a basis of sections µ1 , . . . , µn for Γ(E|U ). Then
(µj ⊗ ωk ) form a basis of sections of Ωp (E|U ) and hence every s ∈ Ωp (E|U )
can be uniquely expressed as s = ajk µj ⊗ ωk with ajk ∈ C ∞ U . We now
define the map De : Ωp (E|U ) → Ωp+1 (E|U ) by

(161) e jk µj ⊗ ωk ) := (D(ajk µj )) ∧ ωk + ajk µj ⊗ dωk .

D(a
(In the right hand side, “D” of course stands for D|U .) This map D e is
clearly R-linear. Let us verify that D e is an exterior covariant derivative wrt
D|U . Thus let µ ∈ Γ(E|U ) and ω ∈ Ωp (U ) be given. Then there exist unique
b1 , . . . , bn , c1 , . . . , cr ∈ C ∞ (U ) such that µ = bj µj and ω = ck ωk , and thus
µ ⊗ ω = bj ck µj ⊗ ωk . Hence by our definition,
e ⊗ ω) = (D(bj ck µj )) ∧ ωk + bj ck µj ⊗ dωk ,
D(µ
and this can be manipulated as follows:


= ck D(bj µj ) + bj µj ⊗ dck ∧ ωk + (bj µj ) ⊗ (ck · dωk )
= (D(bj µj )) ∧ (ck ωk ) + (bj µj ) ⊗ (dck ∧ ωk + ck · dωk )
= (Dµ) ∧ ω + µ ⊗ d(ck ωk )
= (Dµ) ∧ ω + µ ⊗ dω.
e is indeed an exterior covariant derivative wrt D|U .
Hence D
The uniqueness follows by the argument immediately below (159). (In
particular this shows that D e is independent of the choice of bases of sections
ω1 , . . . , ωr and µ1 , . . . , µn .) 

We now complete the proof of existence: Let {Uα } be a family of open

subsets satisfying the assumption of Lemma 8, covering M . Let Dα :
Ωp (E|Uα ) → Ωp+1 (E|Uα ) be the exterior covariant derivative provided by
Lemma 8. For any α, β ∈ A the set V := Uα ∩ Uβ also satisfies the as-
sumption of Lemma 8 (assume V 6= ∅ for nontriviality), and so there exists
e : Ωp (E|V ) → Ωp+1 (E|V ) wrt D|V .
a unique exterior covariant derivative D
Then Lemma 7 (applied twice) implies that for every s ∈ Ωp (E) we have
e |V ) = (Dα (s|U ))|V .
(Dα (s|U ))|V = D(s
α β

Hence all assumptions of Lemma 6 are fulfilled and now that lemma proves
the existence of an exterior covariant derivative Ωp (E) → Ωp+1 (E) wrt D.
Done! 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 177

(b). In order to simplify the notation let us replace M by U ; thus from

now on we can write “E” in place of “E|U ”.
Note that both sides of the stated formula are R-linear in µ; hence by
the argument below (159) it suffices to prove the formula for µ of the form
µ = s ⊗ ω, with s ∈ Γ(E) and ω ∈ Ωp (M ). For such µ we have by the
definition in part (a):
Dµ = D(s ⊗ ω) = (Ds) ∧ ω + µ ⊗ dω = (ds + As) ∧ ω + s ⊗ dω


= ds ∧ ω + s ⊗ dω + As ∧ ω
= d(s ⊗ ω) + As ∧ ω.
(In the last equality we again used the definition in part (a), this time for
the naive connection d.) Note also that the “∧” in “As∧ω” can be viewed as
the combined vector-wedge-product (cf. Problem 49(c)) Ω1 (E) × Ωp (M ) →
Ωp+1 (E) coming from the standard “scalar product map” Γ(E)×C ∞ (M ) →
Γ(E). As in the problem formulation, let us also write “∧” for the vector-
wedge-product
(162) Ω1 (End E) × Ωr (E) → Ωr+1 (E)
coming from the standard contraction (“evaluation”) Γ(End E)×ΓE → ΓE.
(Thus “As” appearing above is the same as A∧s, namely the image of A and
s under the map in (162) with r = 0.) Noticing that the given multiplication
rules Γ(End E) × ΓE → ΓE and ΓE × C ∞ (M ) → ΓE satisfy the obvious
associativity relation37, it follows by Problem 49(d) that
As ∧ ω = (A ∧ s) ∧ ω = A ∧ (s ∧ ω).
Using this in the previous computation gives
Dµ = d(s ⊗ ω) + A ∧ (s ∧ ω) = dµ + A ∧ µ.
Done! 

37This merely captures the fact that the map Γ(End E) × ΓE → ΓE is C ∞ (M )-linear
in its second argument (in fact it is also C ∞ (M )-linear in its first argument).
178 ANDREAS STRÖMBERGSSON

(c). Again by R-(bi-)linearity it suffices to prove the stated formula for

µ1 , µ2 of the form µ1 = s1 ⊗ω1 and µ2 = s2 ⊗ω2 , with µj ∈ ΓEj , ω1 ∈ Ωr (M )
and ω2 ∈ Ωs (M ). In this case we have
D(µ1 ∧ µ2 ) = D((s1 · s2 ) ⊗ (ω1 ∧ ω2 ))
= (D(s1 · s2 )) ∧ (ω1 ∧ ω2 ) + (s1 · s2 ) ⊗ d(ω1 ∧ ω2 ),
where we used the definition of ∧ (Problem 49(c)) and then the definition
of D (part (a) of this problem). Using now the assumption that the given
connections respect our “·”, we get
= ((Ds1 ) ∧ s2 ) ∧ (ω1 ∧ ω2 ) + (s1 ∧ Ds2 ) ∧ (ω1 ∧ ω2 )
(163) +(s1 · s2 ) ⊗ (dω1 ∧ ω2 ) + (−1)r (s1 · s2 ) ⊗ (ω1 ∧ dω2 ).
Now note that the multiplication rule from E1 , E2 to E e satisfy the asso-
ciativity relation (s1 · s2 ) · f = s1 · (s2 · f ) for all s1 ∈ ΓE1 , s2 ∈ ΓE2 ,
f ∈ C ∞ (M ) = Γ(M × R) (namely since the multiplication rule is C ∞ (M )-
linear in s2 ). By Problem 49(d), this implies that
(ϕ1 ∧ ϕ2 ) ∧ ϕ3 = ϕ1 ∧ (ϕ2 ∧ ϕ3 ),
∀ϕ1 ∈ Ωr1 (E1 ), ϕ2 ∈ Ωr2 (E2 ), ϕ3 ∈ Ωr3 (M ).
Similarly we also have
(ϕ1 ∧ ϕ2 ) ∧ ϕ3 = ϕ1 ∧ (ϕ2 ∧ ϕ3 ),
∀ϕ1 ∈ Ωr1 (E2 ), ϕ2 ∈ Ωr2 (M ), ϕ3 ∈ Ωr3 (M ),
and other similar associativity relations. Furthermore by Problem 49(c),
ϕ1 ∧ ϕ2 = (−1)r1 r2 ϕ2 ∧ ϕ1 , ∀ϕ1 ∈ Ωr1 (E2 ), ϕ2 ∈ Ωr2 (M ).
Using these facts (and Ds2 ∈ Ω1 (E2 ), and, again, the definition of ∧), the
expression in (163) is seen to be
= (Ds1 ) ∧ ω1 ∧ s2 ∧ ω2 + (−1)r s1 ∧ ω1 ∧ (Ds2 ) ∧ ω2
+ (s1 ⊗ dω1 ) ∧ (s2 ⊗ ω2 ) + (−1)r (s1 ⊗ ω1 ) ∧ (s2 ⊗ dω2 )
   
= (Ds1 ) ∧ ω1 + s1 ⊗ dω1 ∧ µ2 + (−1)r µ1 ∧ (Ds2 ) ∧ ω2 + s2 ⊗ dω2
= (Dµ1 ) ∧ µ2 + (−1)r µ1 ∧ (Dµ2 ).
Done! 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 179

(d). The statement that the given connections respect the multiplication
rule can be expressed as:
(164)


D m(s1 ⊗ s2 ) = m(Ds1 ⊗ s2 ) + m(s1 ⊗ Ds2 ) (∀s1 ∈ ΓE1 , s2 ∈ ΓE2 ),
where in the right hand side, “m(α)” for α ∈ Ω1 (E1 ⊗ E2 ) is the output of
the vector-wedge-product
e × Ω1 (E1 ⊗ E2 ) → Ω1 (E)
ΓHom(E1 ⊗ E2 , E) e
which extends the standard evaluation map
e × Γ(E1 ⊗ E2 ) → Γ(E).
ΓHom(E1 ⊗ E2 , E) e
However by Problem 59(c) and Problem 58 we have:


D m(s1 ⊗ s2 ) = (Dm)(s1 ⊗ s2 ) + m(Ds1 ⊗ s2 ) + m(s1 ⊗ Ds2 ),
for any s1 ∈ ΓE1 , s2 ∈ ΓE2 . Hence (164) is equivalent with:
(165) (Dm)(s1 ⊗ s2 ) = 0 (∀s1 ∈ ΓE1 , s2 ∈ ΓE2 ).
But every section in Γ(E1 ⊗ E2 ) can be written as a finite sum of sections
of the form s1 ⊗ s2 ; hence (165) is equivalent with (Dm)(s) = 0 for all
s ∈ Γ(E1 ⊗ E2 ). This is equivalent with Dm = 0 in Ω1 (Hom(E1 ⊗ E2 , E))e
(via Problem 35(c)). 
180 ANDREAS STRÖMBERGSSON

Problem 61:
It suffices to prove the stated formula when s = µ ⊗ ω (µ ∈ Γ(E), ω ∈
Ωr (M )), since an arbitrary section in Ωr (E) can be expressed as a finite sum
of such “pure tensor” sections. Now when s = µ ⊗ ω, we find that the right
hand side of the stated formula equals
Xr


(−1)j DXj ω(X0 , . . . , X̂j , . . . , Xr ) · µ
j=0
X
+ (−1)j+k ω([Xj , Xk ], X0 , . . . , X̂j , . . . , X̂k , . . . , Xr ) · µ.
0≤j<k≤r
Xr


= (−1)j Xj ω(X0 , . . . , X̂j , . . . , Xr ) · µ
j=0
r
X
+ (−1)j ω(X0 , . . . , X̂j , . . . , Xr ) · DXj µ
j=0
X
+ (−1)j+k ω([Xj , Xk ], X0 , . . . , X̂j , . . . , X̂k , . . . , Xr ) · µ.
0≤j<k≤r

On the other hand we have by definition Ds = (Dµ) ∧ ω + µ ⊗ dω (cf.

Problem 60(a)), and thus


[Ds](X0 , . . . , Xr ) = (Dµ) ∧ ω (X0 , . . . , Xr ) + [dω](X0 , . . . , Xr ) · µ.
For the first term we now use the definition of wedge product38, and for the
second term we apply Problem 48(c); this gives:
r
X
[Ds](X0 , . . . , Xr ) = (−1)j · ω(X0 , . . . , X̂j , . . . , Xr ) · (Dµ)(Xj )
j=0
r
X

+ (−1)j Xj ω(X0 , . . . , X̂j , . . . , Xr ) · µ
j=0
X
+ (−1)j+k ω([Xj , Xk ], X0 , . . . , X̂j , . . . , X̂k , . . . , Xr ) · µ.
0≤j<k≤r

Here “(Dµ)(Xj )” stands for the contraction of the form part of Dµ ∈ Ω1 (E)
against Xj ; thus (Dµ)(Xj ) = DXj µ. Hence the last expression equals our
previous expression for the right hand side of the stated formula. Hence the
stated formula is proved! 

38together with a computation reducing the “A-sum” over S

r+1 to a sum over only
r + 1 distinct permutations; we leave this step to the reader.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 181

Problem 62:
By Problem 59, for each j ∈ {0, . . . , r} we have
 
∇]
[D se (X0 , . . . , X̂j , . . . , Xr )
Xj
  Xj−1  
= DXj s(X0 , . . . , X̂j , . . . , Xr ) − s X0 , . . . , ∇Xj Xk , . . . , X̂j , . . . , Xr
k=0
Xr  
− s X0 , . . . , X̂j , . . . , ∇Xj Xk , . . . , Xr
k=j+1
  Xj−1  
= DXj s(X0 , . . . , X̂j , . . . , Xr ) − (−1)k s ∇Xj Xk , X0 , . . . , X̂k , . . . , X̂j , . . . , Xr
k=0
Xr  
− (−1)k−1 s ∇Xj Xk , X0 , . . . , X̂j , . . . , X̂k , . . . , Xr ,
k=j+1

where in the last step we used the fact that the form part of s is alternating.
Using the above it follows that
r
X  
(−1)j [ D
∇] se (X0 , . . . , X̂j , . . . , Xr )
Xj
j=0
r
X  
= (−1)j DXj s(X0 , . . . , X̂j , . . . , Xr )
j=0
X  
− (−1)j+k s ∇Xj Xk , X0 , . . . , X̂k , . . . , X̂j , . . . , Xr
0≤k<j≤r
X  
+ (−1)j+k s ∇Xj Xk , X0 , . . . , X̂j , . . . , X̂k , . . . , Xr .
0≤j<k≤r

In the middle sum we change names between j and k; this gives:

r
X  
= (−1)j DXj s(X0 , . . . , X̂j , . . . , Xr )
j=0
X  
+ (−1)j+k s ∇Xj Xk − ∇Xk Xj , X0 , . . . , X̂j , . . . , X̂k , . . . , Xr .
0≤j<k≤r
Xr  
= (−1)j DXj s(X0 , . . . , X̂j , . . . , Xr )
j=0
X  
+ (−1)j+k s [Xj , Xk ], X0 , . . . , X̂j , . . . , X̂k , . . . , Xr ,
0≤j<k≤r
182 ANDREAS STRÖMBERGSSON

where in the last step we used the assumption that ∇ is torsion free. By
Problem 61, the above equals [Ds](X0 , . . . , Xr ). Hence we have proved the
desired formula. 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 183

Problem 63:
(a). Let (U, ϕ) be any bundle chart for E, and let s1 , . . . , sn ∈ ΓE|U be
the corresponding basis of sections. Then we get a corresponding bundle
e for End E by mapping any B ∈ End Ep (p ∈ U ) to the matrix
chart (U, ϕ)
for B with respect to the basis s1 (p), . . . , sn (p) of Ep .
(Then ϕ e is a C ∞ diffeomorphism from End E|U onto U × Mn (R); pedan-
tically for this to be a bundle chart we also need to fix an identification
2
of Mn (R) with Rn . Furthermore: The bundle chart described here is the
same as the one which we give in the solution to Problem 39, after identifying
Hom(Rn , Rn ) with Mn (R) in the obvious way.)
Now if (U, ϕ) is a metric bundle chart then the image of AdE|U under ϕe is
exactly U ×o(n) where o(n) is the set of skew-symmetric matrices in Mn (R).
Hence since o(n) is a linear subspace of Mn (R) 39, and since (E, π, M ) can
be covered with metric bundle charts [12, Thm. 2.1.3], it follows that AdE
is a vector subbundle of End E. 
Remark: Recall that we write gl(E) for the vector bundle End E equipped
with its standard Lie algebra bundle structure. Similarly we write gln (R)
for Mn (R) equipped with its standard Lie algebra structure; and o(n) is in
fact a Lie subalgebra of gln (R). Also for each p ∈ U , ϕ
ep is in fact a Lie
∼
algebra isomorphism gl(Ep ) − → gln (R) which maps AdEp onto o(n). Hence
AdE is a Lie algebra subbundle of gl(E).

39and so we can fix a linear isomorphism M (R) = Rn2 under which o(n) becomes
n
2
identified with Rk = {(∗, · · · , ∗, 0, · · · , 0)} ⊂ Rn for some k. (In fact k = n(n − 1)/2.)
184 ANDREAS STRÖMBERGSSON

(b). Assume s ∈ Γ(AdE). Let (U, ϕ) be any metric bundle chart for
E, and let µ1 , . . . , µn be the corresponding basis of sections in ΓE|U . With
respect to (U, ϕ) we write D|U = d + A with A = (Akj ) ∈ Ω1 (End E); thus
each Akj is in Ω1 (U ) and D(µj ) = A(µj ) = µk ⊗ Akj for all j ∈ {1, . . . , n}; cf.
#9, p. 6. Then Akj = −Ajk for all j, k ∈ {1, . . . , n}, by Lemma 2 in #11.
Let µ1∗ , . . . , µn∗ be the basis of sections in ΓE|U
∗ which is dual to µ , . . . , µ ;
1 n
j∗
then {µ ⊗ µk : j, k ∈ {1, . . . , n}} is a basis of sections in Γ End E. Take
akj ∈ C ∞ (U ) for j, k ∈ {1, . . . , n} so that s|U = akj µj∗ ⊗ µk . This means that
for any p ∈ U , (akj (p)) is the matrix for s(p) ∈ End Ep with respect to the
basis for Ep which comes from (U, ϕ); hence by the definition of AdE (Cf.
#11, Def. 2) we have akj = −ajk throughout U , for all j, k ∈ {1, . . . , n}.40
Now we have
(Ds)|U = ds + [A, s].
(This was seen in the proof of the second Bianchi identity; cf. #11, p. 6.)
Here since A and s have the matrices (Akj ) and (akj ), respectively, we find
that [A, s] has the matrix (Akj aji − akj Aji )i,k . 41 Hence:


(Ds)|U = ds + [A, s] = µi∗ ⊗ µk ⊗ daki + aji Akj − akj Aji .
Using now the fact that akj = −ajk and Akj = −Ajk throughout U (∀j, k) it
follows that daki = −daik and aji Akj − akj Aji = −(ajk Aij − aij Ajk ) throughout
U (∀i, k). Hence (Ds)|U is represented by a skew-symmetric matrix wrt
the basis coming from (U, ϕ), and therefore (Ds)|U ∈ Γ(AdE|U ). Since this
holds for any metric bundle chart (U, ϕ) for E, it follows that s ∈ Γ(AdE).
Done! 

See also alternative solution on the next page.

40Here’s a more explicit version of exactly the same argument: By the definition of
AdE we have hs(µi ), µℓ i = −hµi , s(µℓ )i throughout U , for all i, ℓ ∈ {1, . . . , n}. But
hs(µi ), µℓ i = haki µk , µℓ i = aℓi and similarly hµi , s(µℓ )i = aiℓ . Hence aℓi = −aiℓ throughout
U.
41Details: We have [A, s] = A ◦ s − s ◦ A since A ∈ Ω1 (End E ) and s ∈ Ω0 (End E);
|U
cf. #11, p. 7. Now
[A, s] = A ◦ s − s ◦ A
= (µj∗ ⊗ µk ⊗ Akj ) ◦ (aℓi µi∗ ⊗ µℓ ) − (akj µj∗ ⊗ µk ) ◦ (µi∗ ⊗ µℓ ⊗ Aℓi )
= µi∗ ⊗ µk ⊗ (aji Akj − akj Aji ).
 PROBLEMS; “RIEMANNIAN GEOMETRY” 185

Alternative (not using local coordinates): For any s ∈ Γ(End E)

and X, Y ∈ ΓE we have
(Ds)(X) = D(s(X)) − s(DX)
by Problem 59(c), and hence
(166) h(Ds)(X), Y i = hD(s(X)), Y i − hs(DX), Y i.
(Here of course h·, ·i stands for the vector-wedge-product Ω1 (E) × Γ(E) →
Ω1 (M ) which comes from the given bundle metric Γ(E) × Γ(E) → Γ(M ).)
Using also the fact that D is metric, we can write the above relation as:
(167) h(Ds)(X), Y i = dhs(X), Y i − hs(X), DY i − hs(DX), Y i.
Switching X and Y we also have:
(168) hX, (Ds)(Y )i = dhX, s(Y )i − hDX, s(Y )i − hX, s(DY )i.

Now assume s ∈ Γ(AdE). Then hs(Z1 ), Z2 i = −hZ1 , s(Z2 )i for any two
sections Z1 , Z2 ∈ ΓE. This implies that more generally
(169) hs(µ1 ), µ2 i = −hµ1 , s(µ2 )i, ∀µ1 ∈ Ωp (E), µ2 ∈ Ωq (E).
[Indeed, if µ1 = Z1 ⊗ ω1 and µ2 = Z2 ⊗ ω2 with Z1 , Z2 ∈ ΓE, ω1 ∈ Ωp (M ),
ω2 ∈ Ωq (M ), then
hs(µ1 ), µ2 i = hs(Z1 ) ⊗ ω1 , Z2 ⊗ ω2 i = hs(Z1 ), Z2 i · ω1 ∧ ω2
= −hZ1 , s(Z2 )i · ω1 ∧ ω2 = −hµ1 , s(µ2 )i,
i.e. (169) holds. The general case follows by R-(bi)-linearity.] Applying (169)
it follows that the right hand side of (167) equals the negative of the right
hand side of (168). Hence:
(170) h(Ds)(X), Y i = −hX, (Ds)(Y )i in Ω1 (M ).
The fact that this holds for all X, Y ∈ ΓE implies that
(171) Ds ∈ Ω1 (AdE).
Done! 
[Detailed proof that (170) implies (171): Let (U, x) be any C ∞ chart
for M ; then dx1 , . . . , dxd is a basis of sections in ΓT ∗ U . Hence there exist
unique β1 , . . . , βd ∈ Γ End E|U such that Ds|U = βj ⊗ dxj , and now the
above relation says that
hβj (X), Y i · dxj = −hX, βj (Y )i · dxj in Ω1 (U ),
and therefore
hβj (X), Y i = −hX, βj (Y )i in C ∞ (U ), ∀j.
Using Problem 35(c) and the definition of AdE, this implies that βj (p) ∈
AdEp , ∀p ∈ U , i.e. βj ∈ Γ(AdE|U ), for all j. Therefore Ds|U ∈ Ω1 (AdE|U ).
Since M can be covered by C ∞ charts, it follows that Ds ∈ Ω1 (AdE).]
186 ANDREAS STRÖMBERGSSON

Problem 64:
V
(a). First assume that the statement in r (V ) holds. Note that v1 ∧ · · · ∧
vr 6= 0 implies that v1 , . . . , vr are linearly independent. (Proof: exercise!)
Hence r ≤ n := dim V and we can choose vr+1 , . . . , vn ∈ V so that v1 , . . . , vn
is a basis for V . Then we know (cf., Vr e.g., “Prop. 3” in Sec. 7.2 in the lecture
notes) that (vI )I∈I is a basis for (V ), where I is the family of all r-tuples
I = (i1 , . . . , ir ) ∈ {1, . . . , n}r with i1 < · · · < ir , and
vI := vi1 ∧ · · · ∧ vir .
Vr

(Thus dim (V ) = #I = nr .) Also since v1 , . . . , vn is a basis for V , there
exist unique constants ckj ∈ R (j ∈ {1, . . . , r}, k ∈ {1, . . . , n}) such that
wj = ckj vk for j = 1, . . . , r. Then
w1 ∧ · · · ∧ wr = (ck11 vk1 ) ∧ (ck22 vk2 ) ∧ · · · ∧ (ckr r vkr )
= ck11 ck22 · · · ckr r · vk1 ∧ vk2 ∧ · · · ∧ vkr .
Note that the last expression is a sum over all (k1 , . . . , kr ) ∈ {1, . . . , n}r ,
and for each such (k1 , . . . , kr ) there exist a unique I ∈ I and a unique
permutation σ ∈ Sr such that
(k1 , . . . , kr ) = (iσ(1) , . . . , iσ(r) ).
Hence:
X X i i i
w1 ∧ · · · ∧ wr = c1σ(1) c2σ(2) · · · crσ(r) · viσ(1) ∧ viσ(2) ∧ · · · ∧ viσ(r)
I∈I σ∈Sr
X X i i i

= (sgn σ)c1σ(1) c2σ(2) · · · crσ(r) · vI
I∈I σ∈Sr
X i
= det(cℓj ) · vI .
I∈I
(In the second equality we made repeated use of the rule u1 ∧ u2 = −u2 ∧ u1 ,
i
∀u1 , u2 ∈ V . In the last line (cℓj ) is an r × r-matrix; ℓ, j ∈ {1, . . . , r}.) Now
from our assumption v1 ∧ · · · ∧ vr = c · w1 ∧ · · · ∧ wr and the fact that (vI )I∈I
V i
is a basis for r (V ), it follows that c 6= 0, det(cℓj ) = c−1 for I = (1, . . . , r)
i
(in other words: det(cjℓ ) = c−1 ), while det(cℓj ) = 0 for all I ∈ I \{(1, . . . , r)}.
In other words, in the r × n matrix
 1 2 
c1 c1 · · · cn1
 .. .. ..  ,
. . .
cr cr · · · cnr
1 2

the r × r minor which is furthest to the left equals c−1 6= 0, while all
other r × r minors vanish! This implies that the first r columns of the
above matrix (viewed as vectors in Rr ) form a basis for Rr . Furthermore,
it follows that every other column vanishes, i.e. ciℓ = 0 for all i > r and
ℓ ∈ {1, . . . , r}. (Proof: Suppose that there is some i > r such that the
 PROBLEMS; “RIEMANNIAN GEOMETRY” 187

ith column is not 0. Then there exists a subset of r − 1 among the first
r columns which together with the ith column form a basis for Rr . This
implies that the corresponding r × r minor is non-zero, a contradiction.)
Therefore wℓ = ckℓ vk ∈ Span{v1 , . . . , vr } for each ℓ ∈ {1, . . . , r}; furthermore
since the matrix (ckℓ )ℓ,k∈{1,...,r} is invertible we get vℓ ∈ Span{w1 , . . . , wr }
for each ℓ ∈ {1, . . . , r}; hence Span{v1 , . . . , vr } = Span{w1 , . . . , wr }, as we
wanted to prove!
Conversely, now assume that v1 , . . . , vr are linearly independent and v1 , . . . , vr
and w1 , . . . , wr span the same r-dimensional linear subspace of V . This
means that wℓ = ckℓ vk for some constants ckℓ ∈ R (ℓ, k ∈ {1, . . . , r}) such
Vr
that the r × r matrix (ckℓ ) is non-singular. ThenVr v 1 ∧ · · · ∧ vr 6
= 0 in (V ),
since v1 ∧ · · · ∧ vr can be part of a basis for (V ) by “Prop. 3” in Sec. 7.2
in the lecture notes. Let (γij ) := (ckℓ )−1 ∈ Mr (R); then vk = γkℓ wℓ and so
v1 ∧ · · · ∧ vr = (γ1ℓ1 wℓ1 ) ∧ · · · ∧ (γrℓr wℓr ) = det(γkℓ ) · w1 ∧ · · · ∧ wr ,
and det(γkℓ ) 6= 0. Done! 

(b).

(c). If v1 , . . . , vr are not linearly independent then the parallelotope in

question is contained in some r − 1 dimensional subspace and so has r-
dimensional volume 0; also v1 ∧ · · · ∧ vr = 0 and so kv1 ∧ · · · ∧ vr k = 0, i.e.
the formula holds.
Now assume that v1 , . . . , vr are linearly independent. Pick an ON-basis
e1 , . . . , er for the r-dimensional subspace spanned by v1 , . . . , vr , and choose
er+1 , . . . , en such that e1 , . . . , en is an ON-basis for V . Take ckj ∈ R so that
vj = ckj ek . Then the volume of the r-dimensional parallelotope spanned by
v1 , . . . , vr equals | det(ckj )| (basic fact about volumes). On the other hand
v1 ∧ · · · ∧ vr = det(ckj ) · e1 ∧ · · · ∧ er (by a computation similar to a step in
the solution to part a) and hence
hv1 ∧ · · · ∧ vr , v1 ∧ · · · ∧ vr i = (det(ckj ))2 · he1 ∧ · · · ∧ er , e1 ∧ · · · ∧ er i
= (det(ckj ))2 ,
where the last equality holds by property (i) in part b. Hence
q
kv1 ∧ · · · ∧ vr k = (det(ckj ))2 = | det(ckj )|.
Done! 
188 ANDREAS STRÖMBERGSSON

Problem 65:
(a). (i) ⇒ (ii): Let A be an oriented atlas for M . By a simple modifi-
cation of the standard C ∞ charts for T M (cf. Lecture #2, pp. 10–11, and
Problem 16), one proves that for any C ∞ chart (U, x) for M , (U, ηx ) is a
bundle chart for (T M, π, M ), where ηx is the map
ηx : T U → U × Rd ;
ηx (w) = (π(w), dxπ(w) (w)).
In particular the following is an atlas of bundle charts for (T M, π, M ):
A′ := {(U, ηx ) : (U, x) ∈ A}.
We claim that A′ makes (T M, π, M ) an oriented vector bundle. To prove
this consider any two charts (U, x), (V, y) ∈ A, and any point p ∈ U ∩ V .
Our task is to prove that the linear map
−1
ηy,p ◦ ηx,p : Rd → Rd
is in GL+d (R). However ηx,p = dxp and ηy,p = dyp (both are linear maps
−1 = dy ◦ dx−1 = d(y ◦ x−1 )
from Tp (M ) to Rd ); hence ηy,p ◦ ηx,p p p x(p) . However
−1
y ◦ x : x(U ∩ V ) → y(U ∩ V ) is the chart transition map between (U, x)
and (V, y); hence by our assumption on A, det d(y ◦ x−1 )x(p) > 0, i.e.
−1
ηy,p ◦ ηx,p = d(y ◦ x−1 )x(p) ∈ GL+
d (R).
Hence A′ indeed makes (T M, π, M ) an oriented vector bundle.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 189

(ii) ⇒ (i): Let A′ be an atlas of bundle charts for (T M, π, M ) with respect

to which (T M, π, M ) is an oriented vector bundle. We will prove that M
possesses an oriented atlas. Let A1 be an arbitrary C ∞ atlas for M . Set
n
A2 := (W, x|W ) : (U, x) ∈ A1 , (V, ϕ) ∈ A′ , and W is a path-connected
o
component of U ∩ V .
Then A2 is also a C ∞ atlas for M , and it has the convenient property that
whenever (U, x) is a chart in A2 , U is path-connected and there is a bundle
chart (V, ϕ) in A′ such that U ⊂ V .
Now consider any (U, x) ∈ A2 and (V, ϕ) ∈ A′ subject to U ⊂ V . (More
generally, the following argument applies to any C ∞ chart (U, x) on M such
that U is path-connected and U ⊂ V for some (V, ϕ) ∈ A′ .) Then for any
p ∈ U both dxp and ϕp are linear isomorphisms from Tp U onto Rd ; hence
dxp ◦ ϕ−1 d
p is a linear isomorphism of R onto itself, and so the determinant
det(dxp ◦ ϕ−1
p ) is well-defined and non-zero. Hence by continuity (crucially
using the fact that U is path-connected),42 we either have det(dxp ◦ϕ−1 p )>0
for all p ∈ U or det(dxp ◦ ϕ−1p ) < 0 for all p ∈ U . Let us define the “sign
of (U, x) wrt (V, ϕ)” to be s = +1 in the first case and s = −1 in the
second case. In this situation, we note that: for every (W, η) ∈ A′ and every
p ∈ U ∩ W , we have det(dxp ◦ ηp−1 ) = s. (Proof: For each p ∈ U we have
dxp ◦ ηp−1 = dxp ◦ ϕ−1 −1 −1
p ◦ (ϕp ◦ ηp ) and det(ϕp ◦ ηp ) > 0 since A makes
′
−1 −1
(T M, π, M ) oriented; hence det(dxp ◦ ηp ) and det(dxp ◦ ϕp ) have the same
sign.)
From the previous discussion we conclude: Every (U, x) ∈ A2 (and more
generally every C ∞ chart (U, x) on M such that U is path-connected and
U ⊂ V for some (V, ϕ) ∈ A′ ) has a well-defined sign s ∈ {−1, 1} wrt A′ ,
with the property that
(172) ∀(W, η) ∈ A′ , ∀p ∈ U ∩ W : det(dxp ◦ ηp−1 ) = s.

Now let us fix a non-singular linear map R ∈ GLd (R) with det R < 0 (e.g.
a reflection). For any (U, x) ∈ A2 we define
(
x if (U, x) has sign +1 wrt A′
x
b :=
R ◦ x if (U, x) has sign −1 wrt A′ .

42some more details: we have to prove that det(dx ◦ ϕ−1 ) is a continuous function
p p
of p ∈ U . But we know that α := dx ◦ ϕ−1 |T U is a C
∞
diffeomorphism from U × Rd onto
d
T (x(U )) = x(U ) × R , and for any m, n ∈ {1, . . . , d}, the (m, n)-entry of the matrix
representing the linear map dxp ◦ ϕ−1 p equals em · pr2 (α(p, en )), where em is the mth
standard unit vector in Rd , · is the standard scalar product on Rd , and pr2 : x(U ) × Rd →
Rd is the projection onto the second factor. From this we see that each matrix entry of
(the matrix representing) dxp ◦ϕ−1 p depends continuously on p; hence also the determinant
of dxp ◦ ϕ−1
p depends continuously on p.
190 ANDREAS STRÖMBERGSSON

b) is a C ∞ chart on M , and (U, x

Note that then also (U, x b) has sign +1 wrt
′
A . We set:

A3 := (U, x b) : (U, x) ∈ A2 .
Then A3 is also a C ∞ atlas for M , and it has the property that for any chart
(U, x) ∈ A3 , U is path-connected, there is some (V, ϕ) ∈ A′ with U ⊂ V , and
(U, x) has sign +1 wrt A′ . We claim that A3 is an oriented atlas. To prove
this, consider any two charts (U, x), (V, y) ∈ A3 , and any point p ∈ U ∩ W .
We have to prove that det(dxp ◦ dyp−1 ) > 0. Take a bundle chart (W, η) ∈ A′
with p ∈ W . Since both (U, x) and (V, y) have sign +1 wrt A′ , we have both
det(dxp ◦ ηp−1 ) = +1 and det(dyp ◦ ηp−1 ) = +1 (cf. (172)). Hence


det(dxp ◦ dyp−1 ) = det (dxp ◦ ηp−1 ) ◦ (dyp ◦ ηp−1 )−1


−1
= det dxp ◦ ηp−1 · det dyp ◦ ηp−1 = 1.
Done! 
(i) ⇔ (iii): Cf., e.g., [1, Def. V.7.5, Thm. V.7.6].

(b). Let A be any C ∞ atlas for M ; then we know from Lecture #2, pp.
10–11 (cf. also Problem 16) that the family
A′ := {(T U, dx) : (U, x) ∈ A}
is a C ∞ atlas for T M . Let us prove that A′ is an oriented atlas! Thus
fix any two charts (U, x), (V, y) ∈ A. Set W := U ∩ V and η := y ◦ x−1 ;
then η is a C ∞ diffeomorphism from x(W ) onto y(W ). We have to prove
that the diffeomorphism dy ◦ (dx)−1 = dη from T (x(W )) = x(W ) × Rd onto
T (y(W )) = y(W ) × Rd has everywhere positive Jacobian determinant. For
any (p, v) ∈ x(W ) × Rd we have
dη(p, v) = (η(p), dηp (v)).
Hence the Jacobian matrix of dη at (p, v) has a block decomposition
 
dηp 0
,
∗ dηp
where “dηp ”, “0” and “∗” are d × d matrices (here ∗ stands for a matrix
which we don’t care exactly what it is; note also that the bottom right d × d
matrix equals dηp since the differential of a linear map at any point equals
the map itself). The determinant of the above 2d × 2d-matrix is (det(dηp ))2 ,
which is everywhere positive. Done! 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 191

Problem 66:
(a). Let us use the short-hand notation
∂
∂i := ∈ Γ(T U ).
∂xi
By definition we have
Aji;k = (∇∂k A)(∂i ⊗ dxj ),
where the right hand side stands for the contraction of ∇∂k A ∈ Γ(T11 U )
against ∂i ⊗ dxj ∈ T11 U . This gives, via Problem 59:
     
Aji;k = ∂k A(∂i ⊗ dxj ) − A (∇∂k ∂i ) ⊗ dxj − A ∂i ⊗ (∇∂k (dxj ))
   
= ∂k Aji − A (Γℓki ∂ℓ ) ⊗ dxj − A ∂i ⊗ (−Γjkℓ dxℓ )

= ∂k Aji − Γℓki · A(∂ℓ ⊗ dxj ) + Γjkℓ · A(∂i ⊗ dxℓ )

= ∂k Aji − Γℓki · Ajℓ + Γjkℓ · Aℓi .
(In the second equality we used [12, (4.1.22)] for the last term.) Done! 

(b). (Cf. [14, Lemma 4.8].) Suppose that

∂ ∂
A|U = Aji11···i
···jr
· dxi1 ⊗ · · · ⊗ dxis ⊗ ⊗ · · · ⊗ jr ,
s
∂xj1 ∂x
and write
∂ ∂
∇ ∂ A = Aji11···i
···jr
s ;k
· dxi1 ⊗ · · · ⊗ dxis ⊗
j
⊗ · · · ⊗ jr
∂xk ∂x 1 ∂x
in U . Then by the same type of computation as in part a we find:
s r
j1 ···jr ∂ j1 ···jr X ℓ j1 ···jr
X j
Ai1 ···is ;k = A − Γk ip · Ai1 ···ℓ···is + Γkpℓ · Aij11···i
···ℓ···jr
.
∂xk i1 ···is s
p=1 p=1

192 ANDREAS STRÖMBERGSSON

Problem 67:
We first make some computations useful for all parts of the problem: As
in [12, p. 3, Ex. 1]), we also introduce the chart (V, z) on S d , with
 x xd 
1
V = S d \ {(0, . . . , 0, 1)}; z(x) = ,...,
1 − xd+1 1 − xd+1
Note that both y and z are diffeomorphism onto all of Rd . We compute that
the inverse map of y is given by
 2y 2yd 1 − kyk2 
1
x= , · · · , , , ∀y ∈ Rd .
1 + kyk2 1 + kyk2 1 + kyk2
Here kyk is the standard Euclidean norm; thus kyk2 = y12 + · · · yd2 . Similarly,
the inverse map of z is given by
 2z 2zd kzk2 − 1 
1
x= , · · · , , , ∀z ∈ Rd .
1 + kzk2 1 + kzk2 kzk2 + 1
Note also that U ∩ V = S d \ {(0, . . . , 0, ±1)} and
y(U1 ∩ U2 ) = z(U1 ∩ U2 ) = Rd \ {0},
and so y ◦ z −1 is a diffeomorphism from Rd \ {0} onto itself. We compute
that y ◦ z −1 is explicitly given by
2zj

kzk2 +1 zj
yj = kzk2 −1
= 2
(z ∈ Rd \ {0}, j = 1, . . . , d).
1+ 2 kzk
kzk +1
Note also that
kzk2 1
kyk2 = 4
= ,
kzk kzk2
and hence
yj
zj = kzk2 yj = (y ∈ Rd \ {0}, j = 1, . . . , d).
kyk2
(Note that y ◦ z −1 and z ◦ y −1 are in fact the same map from Rd \ {0} onto
Rd \ {0}. Geometrically this map is inversion in the sphere S d−1 ⊂ Rd .)
Next we compute, for all y ∈ Rd \ {0}:
∂zk ∂  yk  δjk kyk2 − yk · 2yj
= = = δjk kzk2 − 2zk zj (j, k ∈ {1, . . . , d}).
∂yj ∂yj kyk2 kyk4
Hence
∂ ∂zk ∂ ∂ ∂
(173) = = kzk2 − 2zj zk (j ∈ {1, . . . , d}).
∂yj ∂yj ∂zk ∂zj ∂zk
(The last relation is an equality between vector fields on U ∩ V ⊂ S d .)
 PROBLEMS; “RIEMANNIAN GEOMETRY” 193

(a). Assume the opposite, i.e. that there exists a vector field X ∈ Γ(T S 2 )
∂
such that X|U = y1 (wrt the chart (U, y)). Then there exist unique
∂y1
∞
functions α1 , α2 ∈ C (V ) such that
∂ ∂
X|V = α1 + α2
∂z1 ∂z2
(wrt the chart (V, z)). However by (173) we have on U ∩ V :
∂ ∂ ∂ ∂
∂ ∂
= kzk2 − 2z12 − 2z1 z2 = −z12 + z22 − 2z1 z2 ,
∂y1 ∂z1 ∂z1 ∂z2 ∂z1 ∂z2
and y1 = z1 /kzk2 . Hence we must have


z1 −z12 + z22 2z12 z2
α1 (z) = , α2 (z) = − ,
kzk2 kzk2
for all z ∈ R2 \ {0}. We will now prove that the above formula implies
that α2 cannot be extended to a smooth function on all of R2 ; this gives
a contradiction against α2 ∈ C ∞ (V ) and so the solution to part a will be
complete.
The above formula implies
∂α2 4z1 z23 4z1 z23
=− = − , ∀z ∈ R2 \ {0},
∂z1 kzk4 z12 + z22
and the limit of this function as z → 0 in R2 does not exist! (Indeed, for
4ab3
z = t(a, b) 6= 0 the above expression equals − 2 , and for any fixed
(a + b2 )2
4ab3
(a, b) ∈ R2 \ {0} this tends to − 2 as t → 0. Now one immediately
(a + b2 )2
4ab3
verifies that − 2 can take different values for different choices of
(a + b2 )2
∂α2
(a, b) ∈ R2 \ {0}. This shows that has different limits as z approaches
∂z1
2
the origin along different lines in R , and therefore the ’full 2-dim limit’
∂α2
of as z → (0, 0) does not exist.) This proves that we cannot have
∂z1
α2 ∈ C 1 (R2 ), and hence, afortiori, we cannot have α2 ∈ C ∞ (R2 ). (In fact
∂αj
the same argument applies to any of the partial derivatives , j, k ∈ {1, 2},
∂zk
and in particular we cannot have α1 ∈ C ∞ (R2 ) either.) 
194 ANDREAS STRÖMBERGSSON

(b). In U ∩ V we have
(174)
∂ ∂ 1 − y12 − y22 + y32 ∂
(y1 y3 − y2 ) + (y2 y3 + y1 ) +
∂y1 ∂y2 2 ∂y3
z1 z3 − z2 kzk 2   z2 z3 + z1 kzk2  ∂ 
2 ∂ ∂ 2 ∂
= · kzk − 2z z
1 k + · kzk − 2z z
2 k
kzk4 ∂z1 ∂zk kzk4 ∂z2 ∂zk
4 2 2
kzk − z1 − z2 + z3 2  ∂ ∂ 
+ 4
· kzk2 − 2z3 zk
2kzk ∂z3 ∂zk
 
1 2 ∂ 2 ∂ kzk4 − z12 − z22 + z32 ∂
= (z1 z3 − z2 kzk ) + (z2 z3 + z1 kzk ) +
kzk2 ∂z1 ∂z2 2 ∂z3
 
1 2 2 4 2 2 2

∂
+ 4
−2z1 (z1 z3 − z2 kzk ) − 2z2 (z2 z3 + z1 kzk ) − z3 kzk − z1 − z2 + z3 zk
kzk | {z } ∂zk
(∗)

Here the expression called “(*)” equals:

 
−2z12 z3 − 2z22 z3 − z3 kzk4 − z12 − z22 + z32 = −z3 z12 + z22 + z32 + kzk4
= −z3 kzk2 (1 + kzk2 ).
Hence we can continue; the vector field in (174) equals

1 ∂ ∂
2
(z1 z3 − z2 kzk2 − z3 z1 (1 + kzk2 )) + (z2 z3 + z1 kzk2 − z3 z2 (1 + kzk2 ))
kzk ∂z1 ∂z2
 kzk4 − z 2 − z 2 + z 2  ∂ 
1 2 3
+ − z32 (1 + kzk2 )
2 ∂z3
Here
1  kzk4 − z12 − z22 + z32 2 2
 1  kzk4 − kzk2 2 2

− z3 (1 + kzk ) = − z3 kzk
kzk2 2 kzk2 2
kzk2 − 1 − 2z32 z 2 + z22 − z32 − 1
= = 1 ,
2 2
and hence we finally conclude that the above vector field equals
∂ ∂ z 2 + z22 − z32 − 1 ∂
(175) −(z2 + z1 z3 )+ (z1 − z2 z3 ) + 1 .
∂z1 ∂z2 2 ∂z3
Recall that this computation was performed in the set U ∩ V ; however the
expression in (175) clearly defines a (C ∞ ) vector field on all of V . Hence we
can define the (C ∞ ) vector field X ∈ Γ(T S 3 ) to be given by the expression
in (the first line of) (174) in U , and by the expression in (175) in V ; the
above computation shows that this vector field is well-defined, i.e. the two
formulas really give the same vector field in the region of overlap, U ∩ V . 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 195

(c). In U ∩ V we have
∂yj δjk kzk2 − 2zj zk
dyj = dzk = dzk ,
∂zk kzk4
i.e.
z22 − z12 2z1 z2
dy1 = dz1 − dz2
kzk4 kzk4
and
2z1 z2 z12 − z22
dy2 = − dz1 + dz2 .
kzk4 kzk4
Hence:
1
 

dy1 ⊗ dy1 = 4 z22 − z12 dz1 − 2z1 z2 dz2 ⊗ z22 − z12 dz1 − 2z1 z2 dz2
η

1
2


= 4 z22 − z12 dz1 ⊗ dz1 − 2z1 z2 z22 − z12 dz1 ⊗ dz2 + dz2 ⊗ dz1
η

2
+ 4(z1 z2 ) dz2 ⊗ dz2 .

The formula for dy2 ⊗ dy2 is exactly the same except that all z1 ’s and z2 ’s
are swapped. Hence, noticing also

2
2
z22 − z12 + 4(z1 z2 )2 = z22 + z12 = kzk4
and
1 1 1 1 1
· = = = ,
(1 + y12
+ y22 )4 kzk8 2 4
(1 + kyk ) kzk8 −2 4
(1 + kzk ) kzk8 (1 + kzk2 )4
we obtain:
1

dy1 ⊗ dy1 + dy2 ⊗ dy2
(1 + y12 + y22 )4
kzk4

(176) 2
=4
dz1 ⊗ dz1 + dz2 ⊗ dz2 .
(1 + kzk )
Recall that this computation was performed in the set U ∩ V ; however the
last expression clearly defines a (C ∞ ) section of all of T20 (V ). Hence we can
define the (C ∞ ) section
m ∈ Γ(T20 (S 2 ))
to be given by the expression in the left hand side of (176) in U , and by the
expression in the right hand side of (176) in V ; the above equality (which
is valid in U ∩ V ) shows that this section in m is well-defined.
For each p ∈ S 2 , m(p) is a vector in T20 (S 2 )p = Tp∗ (S 2 ) ⊗ Tp∗ (S 2 ) and
can thus be viewed as a bilinear form on Tp (S 2 ). We see by inspection in
(176) that this bilinear form is symmetric at every p ∈ S 2 . Furthermore it
is positive definite at every p ∈ U , since its matrix with respect to the basis
∂ ∂ 2 −4 times the 2 × 2 identity
∂y1 , ∂y2 equals the positive number (1 + kyk )
196 ANDREAS STRÖMBERGSSON

matrix. Hence m|U indeed defines a Riemannian metric on U . However at

the point (0, 0, −1) ∈ S 2 , which corresponds to z = (0, 0) ∈ V , m(p) is the
zero form, thus not positive definite. Hence m does not define a Riemannian
metric on S 2 . 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 197

(d). Such a vector bundle can in fact be constructed for any given C ∞
function µ : U ∩ V → GL2 (R); and similarly a rank n vector bundle over S 2
can be constructed having any given C ∞ function µ : U ∩ V → GLn (R) as
transition function; there is simply no obstruction present!
The easiest solution is to simply refer to Jost’s [12, Thm. 2.1.1]. How-
ever that theorem is not very clearly formulated; let us attempt to give an
alternative, more precise statement here:
Theorem. Let M be a C ∞ manifold, let (Uα )α∈A be a covering of M by
open sets, and for any α, β ∈ A let ϕβα be a C ∞ function from Uα ∩ Uβ to
GL(n, R). Assume that for all α, β, γ ∈ A, the following holds:
ϕαα (x) = idRn , ∀x ∈ Uα ;
ϕαβ (x)ϕβα (x) = idRn , ∀x ∈ Uα ∩ Uβ ;
ϕαγ (x)ϕγβ (x)ϕβα (x) = idRn , ∀x ∈ Uα ∩ Uβ ∩ Uγ .
Then there exists a vector bundle E over M (unique up to isomorphism of
vector bundles over M ) which has a bundle atlas {(Uα , ϕα )}α∈A for which
the transition functions are given by the above ϕβα ’s.
Using the above theorem, the existence of the desired vector bundle E
over S 2 is immediate; simply take A = {1, 2}; U1 = U , U2 = V , ϕ12 ≡ µ
and ϕ21 ≡ µ−1 in U ∩ V , ϕ11 ≡ idR2 in U and ϕ22 ≡ idR2 in V . One verifies
that these ϕαβ ’s satisfy all conditions in the above theorem; and the vector
bundle which the theorem gives has the desired property! 
Exercise: Prove the above theorem, e.g. using Problem 36! (Cf. also [15,
Exc. 10-6].)

Alternative: We will construct the desired vector bundle E over S 2

using Problem 36. As a set we define
E := S 2 × R2 .
We also set π := pr1 : E → S 2 , i.e. projection onto the first coordinate. Note,
though, that (unless m = 0) E will not become equipped with the standard
product C ∞ manifold structure of S 2 × R2 ! Note that π −1 (U ) = U × R2 and
π −1 (V ) = V × R2 . Also Ep = π −1 (p) = {p} × R2 for every p ∈ S 2 , and we
equip each such fiber with the standard vector space structure of R2 . Let φ
be the identity map on U × R2 , and let ψ be the map
ψ : V × R2 → V × R2 ,
(
(p, µ(p) · v) if p ∈ U ∩ V
ψ(p, v) :=
(p, v) if p = (0, 0, −1).
(This map is well-defined since V is the disjoint union of U ∩V and {(0, 0, −1)}.
Note that “µ(p) · v” denotes the product of the matrix µ(p) ∈ GL2 (R) and
the vector v ∈ R2 viewed as a 2 × 1 column matrix.)
198 ANDREAS STRÖMBERGSSON

Now π : E → M toghether with the family {(U, φ), (V, ψ)} is easily seen
to satisfy all the assumptions of Problem 36. (In particular ψ ◦ φ−1 (p, v) =
(p, µ(p) · v) and φ ◦ ψ −1 (p, v) = (p, µ(p)−1 · v) for all (p, v) ∈ (U ∩ V ) × R2 ,
from which we see43 that both the maps ψ ◦ φ−1 and φ ◦ ψ −1 are C ∞ maps
from (U ∩ V ) × R2 onto itself.)
Hence by Problem 36, E possesses a unique C ∞ manifold structure such
that (E, π, M ) is a vector bundle of rank 2 and (U, φ) and (V, ψ) are bundle
charts. Note that the transition function from (U, φ) to (V, ψ) equals µ by
construction!

Remark: From a conceptual point of view the “discontinuity” of the

above map ψ at p = (0, 0, −1) is confusing and ugly! The way to think
about this is that our initial definition of E as a set, “S 2 × R2 ” is only a
technical device used to fit the construction into the result from Problem 36,
where we need from start E to be a given (well-defined) set! Note that this
“S 2 ×R2 ” carries no topology from start, so it is actually meaningless to speak
about continuity/discontinuity of the maps ψ and φ! The only topology and
differential structure which we endow “S 2 × R2 ” with, is the one imposed
by requiring that the two bundle charts should be diffeomorphisms! Hence
conceptually it is much better to think of E as the result of gluing the two
vector bundles V ×R2 and U ×R2 together in line with the above description
— and forget about the set “S 2 × R2 ” used in the construction.

43Of course here it is crucial to note that µ is a C ∞ map from U ∩ V to GL (R). This
2
is clear from the formula defining µ, if we view α(y) = arg(y 1 + iy 2 ) as a C ∞ function
from R2 \ {0} to the circle R/2πZ and then use the fact that both cos(mα) and sin(mα)
are well-defined C ∞ functions on R/2πZ.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 199

Problem 68:
e k we have
(a). By the definition of Γ ij
e kij · (sk ◦ f ) = (f ∗ D)
Γ ∂ (sj ◦ f ) in Γ(f ∗ E)|V ,
∂y i

for all i ∈ {1, . . . , d′ } and j ∈ {1, . . . , n}. Let us evaluate the above at an
arbitrary point p ∈ V , using the identity from Problem 57(a); this gives:
(177) e kij (p) · sk (f (p)) = Dv (sj ),
Γ
where v := dfp ( ∂y∂ i ) ∈ Tf (p) N . We have
∂f ℓ ∂
v= (p) · ℓ (f (p)).
∂y i ∂x
(Here we write f ℓ := xℓ ◦ f , as usual.) Hence the right hand side of (177)
can be evaluated as
∂f ℓ
Dv (sj ) = (p) · Γkℓj (f (p)) · sk (f (p)).
∂y i
Comparing with (177), and using the fact that s1 (f (p)), . . . , sn (f (p)) is a
basis of Ef (p) , we conclude:
ℓ
e k (p) = ∂f (p) · Γk (f (p)).
Γ ij ℓj
∂y i
This can also be expressed as:
ℓ
ekij = ∂f · (Γk ◦ f ).
Γ ℓj
∂y i

200 ANDREAS STRÖMBERGSSON

(b). Proof of existence of f ∗ : Ωr (E) → Ωr (f ∗ E), first alternative:

We wish to prove that there exists a unique R-linear map f ∗ : Ωr (E) →
Ωr (f ∗ E) satisfying
(178) f ∗ (µ ⊗ ω) = (µ ◦ f ) ⊗ f ∗ (ω) for all µ ∈ ΓE and ω ∈ Ωr (N ).
V V
It follows from Ωr (E) = Γ(E⊗ r (T ∗ N )) = Γ(E)⊗Γ( r (T ∗ N )) (cf. Problem
43(d)) that every section in Ωr (E) can be expressed as a finite sum of sections
of the form µ ⊗ ω with µ ∈ ΓE and ω ∈ Ωr (N ). Hence the formula (178)
together with the R-linearity certainly makes the map f ∗ uniquely defined,
if it exists at all. The problem is that different decompositions of a given
section s ∈ Ωr (E) as a sum of “pure” sections µ ⊗ ω might apriori lead to
different answers for what “f ∗ (s)” should be. To resolve this we will give
an alternative, “pointwise”, definition for f ∗ (s).
V V
For each p ∈ M , let Ap : r (Tf∗(p) N ) → r (Tp∗ M ) be the map given by
V
Ap (α)(v1 , . . . , vr ) := α(dfp (v1 ), . . . , dfp (vr )), ∀α ∈ r (Tf∗(p) N ), v1 , . . . , vr ∈ Tp M.
This map Ap is clearly R-linear. (Note that this map Ap in principle appears
in Definition 5 in Lecture #8; namely we have f ∗ (ω)p = Ap (ωf (p) ) for any
ω ∈ Ωr (N ) and any p ∈ M .) Hence for each p ∈ M there is a unique
R-linear map
V V
Bp := 1Ef (p) ⊗ Ap : Ef (p) ⊗ r (Tf∗(p) N ) → Ef (p) ⊗ r (Tp∗ M )
V
satisfying Bp (v ⊗ α) = v ⊗ Ap (α) for all v ∈ Ef (p) and α ∈ r (Tf∗(p) N ). Note
that under standard identifications, Bp can equivalently be viewed as a map
V V
Bp : (E ⊗ r T ∗ N )f (p) → (f ∗ E ⊗ r T ∗ M )p .

Now let us define f ∗ (s), for any s ∈ Ωr (E), by44

(179) (f ∗ (s))(p) := Bp (s(f (p))), ∀p ∈ M.
V
Then for every s ∈ Ωr (E), f ∗ (s)
is a function
V from M to f ∗ E ⊗ r (T ∗ M ),

mapping each p ∈ M into the fiber (f ∗ E ⊗ r (T ∗ M ))p . It is also clear from

(179) that f ∗ (s) is C ∞ 45 and hence f ∗ (s) ∈ Ωr (f ∗ E). Therefore f ∗ is a
map from Ωr (E) to Ωr (f ∗ E), and it is immediate from (179) that this map
is R-linear. Finally, for any s = µ ⊗ ω with µ ∈ ΓE and ω ∈ Ωr (N ), we
have, for all p ∈ M :
 
(f ∗ (s))(p) = Bp (s(f (p))) = Bp µ(f (p)) ⊗ ω(f (p)) = µ(f (p)) ⊗ Ap (ω(f (p)))
(180) = µ(f (p)) ⊗ f ∗ (ω)(p) = f ∗ (µ ⊗ ω)(p),
and hence our map f ∗ satisfies (178). Done! 

44The way one initially finds out that this formula (179) should/must hold, is by a
computation similar to (180), but “in the other direction”.
45The fact that f ∗ (s) is C ∞ can also be seen as follows: Decompose s in some way as
a finite sum s = µ1 ⊗ ω1 + · · · + µm ⊗ ωm with µ1 , . . . , µm ∈ ΓE
Pmand ω1 , . . . , ωm ∈ Ωr (N ).
Then similarly as in the computation (180) we have f (s) = j=1 (µj ◦ f ) ⊗ f ∗ (ωj ), and
∗

the right hand side is C ∞ by inspection.

 PROBLEMS; “RIEMANNIAN GEOMETRY” 201

Proof of existence of f ∗ : Ωr (E) → Ωr (f ∗ E), second alternative:

Using Ωr (E) = ΓE ⊗ Ωr (N ) and Ωr (f ∗ E) = Γf ∗ E ⊗ Ωr (M ), it is tempting
to simply say that, by the machinery from Problem 43 etc., “it suffices to
show that the corresponding map from ΓE × Ωr (N ) to Γf ∗ E ⊗ Ωr (M ) is
bilinear” (cf. (182)). However there are complications due to the fact that
we here have a mix of C ∞ (N )-modules and C ∞ (M )-modules, and one has
to be careful about what “bilinear” really means... One can fill in the details
as follows.
On C ∞ (N ) = Ω0 (N ) the map f ∗ is
(181) f ∗ : C ∞ (N ) → C ∞ (M ); f ∗ (g) = g ◦ f (∀g ∈ C ∞ (N )),
and one verifies that f ∗ is a ring homomorphism from C ∞ (N ) to C ∞ (M ).
Using this ring homomorphism, any C ∞ (M )-module gets an induced struc-
ture of a C ∞ (N )-module.
Now consider the map
(182) J : ΓE × Ωr (N ) → Γf ∗ E ⊗ Ωr (M ); J(µ, ω) = (µ ◦ f ) ⊗ f ∗ (ω).
Note that J is a map from a Cartesian product of two C ∞ (N )-modules to
the C ∞ (M )-module Γf ∗ E ⊗Ωr (M ); 46 however by what we have said above,
Γf ∗ E ⊗ Ωr (M ) also has an induced structure of a C ∞ (N )-module, via the
homomorphism f ∗ in (181). Now one verifies that the map J is C ∞ (N )-
bilinear. [Details: One immediately verifies that J(µ1 + µ2 , ω) = J(µ1 , ω) +
J(µ2 , ω) and J(µ, ω1 + ω2 ) = J(µ, ω1 ) + J(µ, ω2 ) for all µ1 , µ2 , µ ∈ ΓE and
ω1 , ω2 , ω ∈ Ωr (M ). Next for arbitrary µ ∈ ΓE, ω ∈ Ωr (M ) and g ∈ C ∞ (N )
we have (g · µ) ◦ f = (g ◦ f ) · (µ ◦ f ), and therefore
J(g · µ, ω) = (g ◦ f ) · J(µ, ω) = f ∗ (g) · J(µ, ω).
Also f ∗ (g · ω) = f ∗ (g) · f ∗ (ω) and therefore
J(µ, g · ω) = f ∗ (g) · J(µ, ω).
Hence J is C ∞ (N )-bilinear.]
The fact that the map (182) is C ∞ (N )-bilinear now implies, via the
defining property of tensor product (of C ∞ (N )-modules) that there exists a
unique C ∞ (N )-linear map
f ∗ : ΓE ⊗ Ωr (N ) → Γf ∗ E ⊗ Ωr (M )
such that
f ∗ (µ ⊗ ω) = J(µ, ω) = (µ ◦ f ) ⊗ f ∗ (ω).
This is the desired map! (Indeed recall ΓE ⊗ Ωr (N ) = Ωr (E) and Γf ∗ E ⊗
Ωr (M ) = Ωr (f ∗ E). The fact that f ∗ is C ∞ (N )-linear implies in particular
that f ∗ is R-linear, as desired.) 
46Recall that the tensor product in “Γf ∗ E ⊗ Ωr (M )” always stands for tensor product
of C ∞ (M )-modules. A more precise notation is “Γf ∗ E ⊗C ∞ (M ) Ωr (M )”.
202 ANDREAS STRÖMBERGSSON

∗
Proof of the formula involving df D . We now turn to the second task
of the problem, i.e. to prove that for any s ∈ Ωr (E) we have
∗D
(183) (df )(f ∗ (s)) = f ∗ (dD s).
We first prove the auxiliary result that the map f ∗ : Ωr (E) → Ωr (f ∗ E)
respects wedge product, i.e.
(184)
f ∗ (σ ∧ η) = f ∗ (σ) ∧ f ∗ (η) in Ωr+s (E), ∀σ ∈ Ωr (E), η ∈ Ωs (N ).
By R-linearity in σ it suffices to prove (184) for σ = σ1 ⊗ η1 with σ1 ∈ ΓE
and η1 ∈ Ωr (N ). In this case,
f ∗ (σ ∧ η) = f ∗ (σ1 ⊗ (η1 ∧ η)) = (σ1 ◦ f ) ⊗ f ∗ (η1 ∧ η)
= (σ1 ◦ f ) ⊗ f ∗ (η1 ) ∧ f ∗ (η) = f ∗ (σ) ∧ f ∗ (η),
where in the third equality we used the fact that f ∗ : Ω(N ) → Ω(M ) respects
wedge product (cf. #8, p. 9). Hence (184) is proved.
Now we prove (183). In the case r = 0 (i.e., s ∈ ΓE and dD = D and
∗
df D = f ∗ D) we see that (183) is equivalent with the identity in Prob-
lem 57(a) if we can only prove that
(185) Ddf (·) (s) = f ∗ (Ds),
P
and assuming Ds = m j=1 µj ⊗ ωj with µ1 , . . . , µm ∈ ΓE and ω1 , . . . , ωm ∈
1
Ω (N ) we have, for every p ∈ M and X ∈ Tp M :
m
X m
X



f ∗ (Ds) (X) = (µj ◦ f ) ⊗ f ∗ (ωj ) (X) = ωj (df (X)) · µj (f (p)
j=1 j=1
= (Ds)(df (X)) = Ddf (X) (s).
Hence (185) holds, and so we have proved that (183) holds when r = 0.
Finally we prove (183) for r ≥ 1. By R-linearity, it is enough to check
that (183) holds when s = µ ⊗ ω for some µ ∈ ΓE, ω ∈ Ωr (N ). Then:
f ∗ (dD s) = f ∗ (dD (µ ⊗ ω))
= f ∗ (Dµ ∧ ω + µ ⊗ dω)
[1]
= f ∗ (Dµ) ∧ f ∗ (ω) + (µ ◦ f ) ⊗ f ∗ (dω)
[2]


= (f ∗ D)(µ ◦ f ) ∧ f ∗ (ω) + (µ ◦ f ) ⊗ d f ∗ (ω)
∗

= (df D ) (µ ◦ f ) ⊗ f ∗ ω
∗

= (df D ) f ∗ (µ ⊗ ω)
∗D
= (df )(f ∗ (s)).
(Here equality [1] holds by R-linearity and (184), and equality [2] holds by
[(183) for r = 0].) Hence we have proved that (183) holds for any r ≥ 0. 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 203

Problem 70: Let ∇ be the Levi-Civita connection for (M, h·, ·i). Then
∇ is metric also wrt [·, ·], since for any vector fields X, Y ∈ Γ(T M ) we have
d[X, Y ] = d(chX, Y i) = c · dhX, Y i = c · h∇X, Y i + c · hX, ∇Y i
= [∇X, Y ] + [X, ∇Y ]
in Ω1 (M ). Also ∇ is torsion free (recall that this notion is independent of
the Riemannian metric). Hence, by the uniqueness in Theorem 1 in #13, ∇
is the Levi-Civita connection also for (M, [·, ·]).
Hence (M, h·, ·i) and (M, [·, ·]) also have the same curvature tensor, R =
∇ ◦ ∇ ∈ Ω2 (End T M ). (However the tensor field “Rm” – cf. p. 1 in Lecture
#14 – is not the same for (M, h·, ·i) and (M, [·, ·]), since its definition involves
the inner product.) Now from Definition 1 in #15 it follows that, if K and
Ke denote sectional curvature on (M, h·, ·i) and on (M, [·, ·]), respectively,
then for any p ∈ M and any linearly independent X, Y ∈ Tp M ,
e [R(X, Y )Y, X] c hR(X, Y )Y, Xi
K(X ∧Y) = = 2 = c−1 K(X ∧ Y ).
[X ∧ Y, X ∧ Y ] c hX ∧ Y, X ∧ Y i)

Example: Let (M, h·, ·i) be the standard unit sphere S d in Rd+1 , and
let [·, ·] be the Riemannian metric obtained by instead using the embedding
x 7→ Rx of S d into Rd+1 , for some fixed R > 0 (still using the standard
Riemannian metric on Rd+1 ). In other words (M, [·, ·]) is the sphere of
radius R in Rd+1 . Then [·, ·] = R2 h·, ·i and thus
e
K(X ∧ Y ) = R−2 K(X ∧ Y )
for any X, Y as above. This agrees with the fact that the sphere of radius
R has constant sectional curvature R−2 .
204 ANDREAS STRÖMBERGSSON

Problem 71: Let p ∈ M and X ∈ Tp M with kXk = 1. Choose an ON

basis X1 , . . . , Xd of Tp M with Xd = X. Then the uniform average of the
sectional curvatures of all planes in Tp M containing X equals:
Z
1

A= d−2
K Xd ∧ (α1 X1 + · · · + αd−1 Xd−1 ) dω(α),
ω(S ) S d−2
where S d−2 = {α ∈ Rd−1 : α21 + · · · + α2d−1 = 1} is the standard d − 2
dimensional unit sphere and ω is the is its standard volume measure (cf.,
e.g., [6, Thm. 2.49]). Using the fact that for any α ∈ S d−2 , the two vectors
Xd and α1 X1 + · · · + αd−1 Xd−1 in Tp M are orthogonal and have unit length,
we get
Z   d−1
X X
d−1 
1
A= R Xd , αj Xj αj Xj , Xd dω(α)
ω(S d−2 ) S d−2
j=1 j=1
d−1 Z
d−1 X
X    
1
= R Xd , αj Xj αk Xk , Xd dω(α).
ω(S d−2 ) S d−2
j=1 k=1
d−1 X
X d−1 Z
1
= hR(Xd , Xj )Xk , Xd i αj αk dω(α).
ω(S d−2 ) S d−2
j=1 k=1
R
Here we note that for any j 6= k ∈ {1, . . . , d−1} we have S d−2 αj αk dω(α) =
0, since the measure ω is preserved by the reflection in the hyperplane
αj = 0. ROn the other hand for each j ∈ {1, . . . , d − 1}, the integral
ω(S d−2 )−1 S d−2 α2j dω(α) equals a constant which is independent of j, since
ω is invariant under any permutation of the coordinates α1 , . . . , αd−1 . Let
us define Cd by
Z
1
Cd−1 := α2 dω(α)
ω(S d−2 ) S d−2 j
(any j ∈ {1, . . . , d − 1}); this is a positive number which only depends on
the dimension d.
(For d = 2 we immediately compute C2 = 1; indeed note that in this case
S d−2 = {1, −1} ⊂ R and ω({1}) = ω({−1}) = 21 . Furthermore for d = 3 we
R 2π
have C3−1 = (2π)−1 0 (cos ϕ)2 dϕ = 12 , i.e. C3 = 2.)
We get:
d−1
X d
X
A = Cd−1 hR(Xd , Xj )Xj , Xd i = Cd−1 hR(Xd , Xj )Xj , Xd i
j=1 j=1

= Cd−1 · Ric(X, X),

where in the second equality we used the fact that hR(Xd , Xd )Xd , Xd i = 0,
and in the last equality we used the definition of the Ricci tensor, Def.
2 in Lecture #15. (Details for the last step: Fix any chart (U, x) on
 PROBLEMS; “RIEMANNIAN GEOMETRY” 205

M with p ∈ U , such that Xj = ∂x∂ j at p, for j = 1, . . . , d. Then (gij )

equals the identity matrix at p, and hence also the inverse matrix, (gij )
equals the identity matrix. Hence Ric(X, X) = g jℓ hR(X, ∂x∂ j ) ∂x∂ ℓ , Xi =
Pd ∂ ∂ Pd
j=1 hR(X, ∂xj ) ∂xj , Xi = j=1 hR(Xd , Xj )Xj , Xd i, as claimed.)

Summing up, we have proved

Ric(X, X) = Cd · A,
and this is the desired formula. It only remains to compute Cd . Since
Cd only depends on the dimension d, it can be conveniently computed by
considering any manifold of constant sectional curvature (6= 0). For example
let M be the unit sphere S d with its standard Riemannian metric. Then the
sectional curvature is everywhere equal to 1, and so A = 1 for any p ∈ M
and any unit vector X ∈ Tp M . On the other hand by again choosing an ON-
basis X1 , . . . , Xd ∈ Tp M with Xd = X then as above we have Ric(X, X) =
Pd Pd−1
j=1 hR(Xd , Xj )Xj , Xd i = j=1 1 = d − 1. Hence:
Cd = d − 1.


Remark: It is also somewhat satisfactory to compute Cd directly from its

2π (d−1)/2
definition. Using basic properties of ω, in particular ω(S d−2 ) = Γ((d−1)/2) ,
we get:
Z 1
Γ( d−1
2 ) 2π (d−2)/2 d−4
Cd−1 = (d−1)/2 α21 · d−2
(1 − α21 ) 2 dα1
2π −1 Γ( 2 )
Z
Γ( d−1
2 )
1 d−4 dx
=√ d−2
· 2 x(1 − x) 2 √
π Γ( 2 ) 0 2 x
d−1 Z 1
Γ( 2 ) 1 d−4
=√ · x 2 (1 − x) 2 dx
π Γ( d−2
2 ) 0

Γ( d−1
2 ) Γ( 23 )Γ( d−2
2 )
=√ d−2
· d+1
π Γ( 2 ) Γ( 2 )
1
= .
d−1

206 ANDREAS STRÖMBERGSSON

Problem 72: First note that, for any X, Y, Z ∈ V :

2R(X, Z, Z, Y ) = K(X + Y, Z) − K(X, Z) − K(Y, Z).

Next, for any X, Y, Z, W ∈ V :

R(X, Z, W, Y ) + R(X, W, Z, Y )
= R(X, Z + W, Z + W, Y ) − R(X, Z, Z, Y ) − R(X, W, W, Y )


= 21 K(X + Y, Z + W ) − K(X, Z + W ) − K(Y, Z + W )
1


− 2 K(X + Y, Z) − K(X, Z) − K(Y, Z)
1


− 2 K(X + Y, W ) − K(X, W ) − K(Y, W ) .

Using also R(X, W, Z, Y ) = −R(W, X, Z, Y ), the above identity can be

rewritten as

2R(X, Z, W, Y ) =2R(W, X, Z, Y )


(186) + K(X + Y, Z + W ) − K(X, Z + W ) − K(Y, Z + W )


− K(X + Y, Z) − K(X, Z) − K(Y, Z)


− K(X + Y, W ) − K(X, W ) − K(Y, W ) .

Here is the same identity with X, Z, W cyclically permuted:

2R(W, X, Z, Y ) =2R(Z, W, X, Y )


(187) + K(W + Y, X + Z) − K(W, X + Z) − K(Y, X + Z)


− K(W + Y, X) − K(W, X) − K(Y, X)


− K(W + Y, Z) − K(W, Z) − K(Y, Z) .

We rewrite (186) by solving for R(W, X, Z, Y ):

2R(W, X, Z, Y ) =2R(X, Z, W, Y )


(188) − K(X + Y, Z + W ) − K(X, Z + W ) − K(Y, Z + W )


+ K(X + Y, Z) − K(X, Z) − K(Y, Z)


+ K(X + Y, W ) − K(X, W ) − K(Y, W ) .
 PROBLEMS; “RIEMANNIAN GEOMETRY” 207

Now add (187) and (188) and add an extra 2R(W, X, Z, Y ) on both sides,
and then use the first Bianchi identity in the right hand side. This gives:


6R(W, X, Z, Y ) = K(W + Y, X + Z) − K(W, X + Z) − K(Y, X + Z)


− K(W + Y, X) − K(W, X) − K(Y, X)


− K(W + Y, Z) − K(W, Z) − K(Y, Z)


− K(X + Y, Z + W ) − K(X, Z + W ) − K(Y, Z + W )


+ K(X + Y, Z) − K(X, Z) − K(Y, Z)


+ K(X + Y, W ) − K(X, W ) − K(Y, W )
(189) = K(W + Y, X + Z) − K(W, X + Z) − K(Y, X + Z)
− K(W + Y, X) + K(Y, X)
− K(W + Y, Z) + K(W, Z)
− K(X + Y, Z + W ) + K(X, Z + W ) + K(Y, Z + W )
+ K(X + Y, Z) − K(X, Z)
+ K(X + Y, W ) − K(Y, W ).
This is an explicit formula for R in terms of K! Changing letters (W → X,
X → Y , Y → W ) the formula reads:
6 · R(X, Y, Z, W ) = K(X + W, Y + Z) − K(X, Y + Z) − K(W, Y + Z)
− K(X + W, Y ) + K(W, Y )
− K(X + W, Z) + K(X, Z)
− K(Y + W, Z + X) + K(Y, Z + X) + K(W, Z + X)
+ K(Y + W, Z) − K(Y, Z)
+ K(Y + W, X) − K(W, X),
which is exactly the formula which Jost states in his [12, Lemma 4.3.3],
except for the factor “6” in the left hand side. 
208 ANDREAS STRÖMBERGSSON

Problem 73: Cf., e.g., [13, p. 292, Thm. 1] or [14, Prop. 7.8].
Assume that there is a function c : M → R such that (wrt any C ∞ chart
(U, x) on M ):
(190) Rik = c · gik .
(Cf. the note on p. 5 in Lecture #15.)
Fix a point p ∈ M and assume that (U, x) are normal coordinates around
p. Then by the second Bianchi identity,
∂h Rijkℓ + ∂k Rijℓh + ∂ℓ Rijhk = 0 at p.
(Here ∂h := ∂x∂ h .) Multiply the above relation with gik gjℓ and add over all
i, k, j, ℓ; this gives:
g ik gjℓ · ∂h Rijkℓ + gik g jℓ · ∂k Rijℓh + gik gjℓ · ∂ℓ Rijhk = 0 at p.
However we have ∂h gik = 0 at p, for all h, i, k; hence the above relation is
equivalent with:




(191) ∂h gik g jℓ Rijkℓ + ∂k gik gjℓ Rijℓh + ∂ℓ gik gjℓ Rijhk = 0 at p.
Recall now that gjℓ Rijkℓ = Rik , by definition. Hence using also (190) we get
gik gjℓ Rijkℓ = gik Rik = gik · c · gik = c · δii = d · c,
where d := dim M . Also
gik gjℓ Rijℓh = −gik gjℓ Rijhℓ = −g ik Rih = −gik · c · gih = −δhk · c.
and
gik gjℓ Rijhk = −g ik gjℓ Rjihk = −g jℓ Rjh = −g jℓ · c · gjh = −δhℓ · c
Substituting these relations in (191) we obtain, at p:
0 = d · ∂h c − ∂h c − ∂h c = (d − 2)∂h c.
Hence since we are assuming d ≥ 3, we conclude that ∂h c = 0 at p. This is
true for all h; hence dcp = 0. This is true for all p ∈ M ; hence c is constant,
qed. 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 209

Problem 76:
(a) By Problem 35(c), every vector in Ef (p) can be obtained as s(f (p))
for some s ∈ ΓE; hence it suffices to prove that for any s ∈ ΓE:
e ◦ f )(X1 , X2 ) = (Rs)(df (X1 ), df (X2 ))
R(s in (f ∗ E)p = Ef (p) ,
∀p ∈ M, X, Y ∈ Tp (M ).
Using (for r = 2) the map f∗
: Ωr (E)
→ Ωr (f ∗ E) defined in Problem 68(b),
the above relation can be expressed:
e ◦ f ) = f ∗ (Rs)
R(s in Ω2 (f ∗ E).
By the definition of f ∗ : Ω0 (E) → Ω0 (f ∗ E), this can also be expressed
(slightly more nicely?) as
(192) e ∗ (s)) = f ∗ (Rs)
R(f in Ω2 (f ∗ E).

Recall that, by definition,

e = df ∗ D ◦ f ∗ D
R and R = dD ◦ D.
∗
(Here, as in Problem 68(b), we write df D and dD for the exterior covari-
ant derivatives, and not just “f ∗ D” and “D” as we usually do.) Now we
compute, for any s ∈ ΓE:


e ∗ (s)) = df ∗ D (f ∗ D)(f ∗ s)
R(f
 
Apply Problem 57(a); cf. also Problem 68(b).
∗

= df D f ∗ (Ds)
 
Apply Problem 68(b).
 
= f ∗ dD (Ds)
= f ∗ (R(s)).
Hence (192) is proved! 
210 ANDREAS STRÖMBERGSSON

(b). Here we are considering a C ∞ map c ≡ F : H → M , where

H = [a, b] × (−ε, ε),
and where M is a Riemannian manifold with ∇ being the Levi-Civita con-
nection on T M . (Actually, in order not to have to consider a manifold with
boundary, we should instead take H = (a − ε′ , b + ε′ ) × (−ε, ε); cf. section
3.1 in the lecture notes.) Also (t, s) are the standard coordinates on H; thus
∂ ∂
∂t and ∂s are well-defined vector fields in Γ(T H). Finally recall that in
“∇ ∂ ” and “∇ ∂ ”, the “∇” is really a short-hand notation for the pullback
∂t ∂s
connection F ∗ ∇ on Γ(F ∗ (T M )). Hence our task is to prove:
∂ ∂  ∂ 
R F, F F
∂t ∂s ∂s
∂  ∂ 
(193) = (F ∗ ∇) ∂ (F ∗ ∇) ∂ F − (F ∗ ∇) ∂ (F ∗ ∇) ∂ F .
∂s ∂t ∂s ∂t ∂s ∂s
∂ ∂ ∂ ∂
(Here ∂s F = dF ◦ ∂s and ∂t F = dF ◦ ∂t are sections in ΓF (T M ) =
Γ(F ∗ (T M )), so that the expression in the right-hand side makes sense.
Note also that the expressions on both sides of the equality are sections
in ΓF (T M ) = Γ(F ∗ (T M )).)
∂ ∂
From now on let us use the short-hand notation ∂t := ∂t , ∂s := ∂s .
Note that for every p ∈ H we have (∂t F )(p) = dF (∂t (p)) and (∂s F )(p) =
dF (∂s (p)) in TF (p) (M ). Hence by part a,


e ∂t , ∂s (∂s F )
R(∂t F, ∂s F )(∂s F ) = R in ΓF (T M ),
e is the curvature of the connection F ∗ ∇ on F ∗ (T M ). However by
where R
Theorem 1 in Lecture #11, using [∂t , ∂s ] ≡ 0, we have




e ∂t , ∂s (∂s F ) = (F ∗ ∇)∂ (F ∗ ∇)∂ (∂s F ) − (F ∗ ∇)∂ (F ∗ ∇)∂ (∂s F ) .
R t s s t

This proves (193)! 

 PROBLEMS; “RIEMANNIAN GEOMETRY” 211

Problem 78: (Cf. [14, Prop. 10.9].) Recall that the fact that the chart
(U, x) gives normal coordinates means that there is some r > 0 such that
expp|Br (0) is a diffeomorphism onto U , where Br (0) is the open ball of radius
r about the origin in Tp (M ) (wrt the Riemannian metric h·, ·i); also we fix
an identification Tp (M ) = Rd which carries h·, ·i on Tp (M ) to the standard
scalar product on Rd ; thus Br (0) is now the open ball of radius r about
the origin in Rd ; and finally x := (expp|Br (0) )−1 : U → Rd , with image
x(U ) = Br (0).
Now fix a point x ∈ Br (0) \ {0}. Set T := kxk ∈ (0, r) and consider the
geodesic


c : [0, T ] → M, c(t) = expp tkxk−1 x .
Note that c is parametrized by arc length, i.e. kċ(t)k = 1 for all t ∈ [0, T ].
Using the chart (U, x) to identify U and Br (0), the map expp : Br (0) → U
becomes simply the identity map on Br (0); the geodesic c becomes c(t) = tx,
and also for any w ∈ Br (0) the differential of expp : Tp (M ) → M at w,
(194) (d expp )w : Tw (Tp (M )) = Rd → Texpp (w) (M ) = Rd ,
gets47 identified with the identity map on Rd . Hence by Cor. 1 in Lec-
ture #17, for any v ∈ Rd the formula
(195) X(t) := t · v ∈ Tc(t) (M ) = Rd (t ∈ [0, T ])
defines a Jacobi field along c.
The Riemannian metric h·, ·i on U carries over to a Riemannian metric
h·, ·i on Br (0), which is given by
hv, wi = gij (x)v i wj
for any x ∈ Br (0) and v, w ∈ Rd .
Let us first assume that the vector v in (195) satisfies v · x = 0. Since
gij (0) = δij (by Lemma 1 in #4), this implies that v and kxk−1 x are orthog-
onal when viewed as tangent vectors in Tp (M ). Note also ċ(0) = kxk−1 x;
hence by “Gauss’ Lemma” (Cor. 2 in Lecture #17), hX(t), ċ(t)i = 0 for all
t ∈ [0, T ], i.e. X is a normal Jacobi field along c. Hence by the discussion
on pp. 5–6 in Lecture #17 we have
(196) X(t) = sρ (t) · X1 (t), ∀t ∈ [0, T ],
where X1 (t) is a parallel vector field along c, and

 −1/2 sin(ρ1/2 t)
ρ if ρ > 0
sρ (t) = t if ρ = 0

 −1/2 1/2
|ρ| sinh(|ρ| t) if ρ < 0.
47In (194), the last identification “T d
expp (w) (M ) = R ” of course comes from using the
∂
basis of sections ∂x1
, . . . , ∂x∂ d ∈ Γ(T U ), at the point expp (w) ∈ U .
212 ANDREAS STRÖMBERGSSON

Using X(t) = t · v and gij (0) = δij we see that ht−1 X(t), t−1 X(t)i → kvk2
as t → 0+ . Combining this with (196) and t−1 sρ (t) → 1 as t → 0+ we
conclude that hX1 (t), X1 (t)i → kvk2 as t → 0+ . However hX1 (t), X1 (t)i is
independent of t since X1 is parallel along c; hence

hX1 (t), X1 (t)i = kvk2 , ∀t ∈ [0, T ].

In particular, since T ·v = X(T ) = sρ (T )·X1 (T ), it follows that for v viewed

as a vector in Tx (M ) = Tc(T ) (M ) (recall T = kxk):

 

 sin2 (ρ1/2 kxk) 

 if ρ > 0 


 ρ kxk2 


 

hv, vi = T −2 sρ (T )2 kvk2 = 1 2
if ρ = 0  · kvk .

 

 



 sinh2 (|ρ|1/2 kxk) 

 if ρ < 0. 

|ρ| kxk2

On the other hand if v is proportional to x then we know (by Gauss’

Lemma or by Problem 23 = [12]) that for v viewed as a vector in Tc(t) (M )
for any t ∈ [0, T ],

hv, vi = kvk2 .

In particular this holds at x = c(T ).

Finally let v be an arbitrary vector in Rd . We can then write v = ax + w
where a = kxk−2 (v · x) and w = v − ax; then ax is proportional to x while
v · w = 0. It follows from Gauss’ Lemma (or Problem 23 = [12]) that
hax, wi = 0 when ax and w are viewed as vectors in Tx (M ). Hence
 

 sin2 (ρ1/2 kxk) 

 if ρ > 0 


 ρ kxk2 

2

 

(v · x) 2
hv, vi = hax, axi + hw, wi = + 1 if ρ = 0  · kwk .
kxk2 
 

 

 sinh2 (|ρ|1/2 kxk)
 


 if ρ < 0. 

|ρ| kxk2

Note here that

(v · x)2
kwk2 = kvk2 − .
kxk2
 PROBLEMS; “RIEMANNIAN GEOMETRY” 213

By polarization (i.e. using hv, v ′ i = 12 (hv + v ′ , v + v ′ i − hv, vi − hv ′ , v ′ i)), the

above formula leads to
 

 sin2 (ρ1/2 kxk) 

 if ρ > 0  


 ρ kxk 2 


 
 
′ (v · x)(v ′ · x) ′ (v · x)(v ′ · x) 
hv, v i = + 1 if ρ = 0  · v · v −
kxk2 
  kxk2

 2


 1/2 
 sinh (|ρ| kxk) if ρ < 0. 

 

|ρ| kxk 2

for any v, v ′ ∈ Rd viewed as vectors in Tx (M ). Inserting here v = ei , v ′ = ej

we conclude:


 xi xj sin2 (ρ1/2 kxk)  xi xj 

 + δij − if ρ > 0

 kxk2 ρ kxk2 kxk2


gij (x) = hei , ej i = δij if ρ = 0




 xi xj

 sinh2 (|ρ|1/2 kxk)  xi xj 
 + δij − if ρ < 0,
kxk2 |ρ| kxk2 kxk2
which is the desired formula.
(Using the fact that the Taylor series for ( sinr r )2 and for ( sinh r 2
r ) has the
form 1 + c1 r 2 + c2 r 4 + · · · , which converges for all r ∈ R, one immediately
verifies that the last expression is C ∞ also at x = 0, if extended by continuity
to this point.) 
214 ANDREAS STRÖMBERGSSON

Problem 79:
For symmetry reasons we may assume i = 1, j = 2. Since (U, x) gives
normal coordinates, we know that x(U ) is an open ball in Rd centered at
the origin; take R > 0 so that x(U ) = BR (0). As usual we identify U and
BR (0) via x. Let us furthermore introduce the short-hand notation “(x, y)”
for (x, y, 0, . . . , 0) ∈ Rd .
Given a point (x, y) with 0 < k(x, y)k < R, we consider the two tangent
∂ ∂ ∂ ∂
vectors x +y and −y +x in T(x,y) (M ). We note that for an
∂x ∂y ∂x ∂y
arbitrary choice of polar coordinates (r, θ1 , . . . , θd−1 ) on Rd 48, these two
vectors are given by
X d−1
∂ ∂ ∂ ∂ ∂ ∂
(197) x +y =r , and −y +x = αj
∂x ∂y ∂r ∂x ∂y ∂θj
j=1

for some α1 , . . . , αd−1 ∈ R (which depend on (x, y) and on the choice of polar
∂ ∂
coordinates). The first formula in (197) follows from the fact that x ∂x + y ∂y
is the tangent vector of the curve c(t) := (tx, ty, 0, . . . , 0) at t = 1, and in
polar coordinates this curve is given by c(t) = (tr, θ1 , . . . , θd−1 ) for some fixed
θ1 , . . . , θd−1 . Similarly the second formula in (197) follows from the fact that
∂ ∂
−y ∂x + x ∂y is the tangent vector of the curve γ(t) = (r cos t, r sin t) at a
p
certain t = t0 , where r := k(x, y)k = x2 + y 2 , and in polar coordinates
this curve takes the form γ(t) = (r, θ1 (t), . . . , θd−1 (t)) where r = k(x, y)k is
fixed and θ1 (t), . . . , θd−1 (t) are some smooth real-valued functions of t.
It follows from (197) and Problem 23 that, for any point (x, y) in the
punctured disc 0 < k(x, y)k < R,
D ∂ ∂ ∂ ∂ E
(198) x + y ,x +y = r 2 = x2 + y 2
∂x ∂y ∂x ∂y
and
D ∂ ∂ ∂ ∂ E
(199) x + y , −y +x = 0,
∂x ∂y ∂x ∂y
as stated in the hint to the problem. Note that (198) and (199) are also
valid at (x, y) = (0, 0), by inspection (or by continuity). [Remark on no-
tation: In (198) and (199), “h·, ·i” denotes the Riemannian scalar product
on T(x,y) (M ), whereas we use “k(x, y)k” to denote the standard Euclidean
p
norm, x2 + y 2 .]

48by this we mean: We have fixed a chart (V, ϕ) on S d−1 and we then consider the
corresponding chart (R+ V, (r, θ1 , . . . , θd−1 )) on Rd , where r(z) = kzk (standard Euclidean
length of the vector z) and (θ1 (z), . . . , θd−1 (z)) = ϕ(kzk−1 z) for all z in the open cone
R+ V . This is just as in Problem 23, but with different variable names. Of course, we
assume that the given point (x, y) = (x, y, 0, . . . , 0) lies in the cone R+ V .
 PROBLEMS; “RIEMANNIAN GEOMETRY” 215

Expanding the left hand sides of (198) and (199) we get

(200) x2 g11 + 2xyg12 + y 2 g22 = x2 + y 2
and
(201) −xyg11 + (x2 − y 2 )g12 + xyg22 = 0,
where gij of course stands for gij (x, y) = gij (x, y, 0, . . . , 0). The relations
(200) and (201) are valid for all (x, y) with k(x, y)k < R.
Taking y = 0 in (200) we obtain g11 (x, 0) = 1 for all x ∈ (−R, R) \ {0};
and by continuity this is also valid for x = 0. This implies that all iterated
derivatives g11,1 (x, 0), g11,11 (x, 0), g11,111 (x, 0), . . ., vanish identically for all
x ∈ (−R, R). In particular
g11,11 (0) = 0,
as desired. Note that a symmetric argument (exchanging the roles of x and
y) also gives g22 (0, y) = 1 and g22,2 (0, y) = g22,22 (0, y) = . . . = 0 for all
y ∈ (−R, R).
Differentiating (201) with respect to x we get
−yg11 − xyg11,1 + 2xg12 + (x2 − y 2 )g12,1 + yg22 + xyg22,1 = 0,
and differentiating this three times with respect to y gives49
y · * + x · * − 3g11,22 − 6g12,12 + 3g22,22 = 0,
where each “ * ” stands for a sum where each term is a polynomial in x
and y times a partial derivative of g11 or g12 or g22 . The above is valid for
all (x, y) in the disc k(x, y)k < R. Setting now (x, y) = (0, 0), and using
g22,22 (0, 0) = 0, we conclude that
g11,22 (0, 0) = −2g12,12 (0).
By the symmetric argumment (x ↔ y) we also have
g22,11 (0, 0) = −2g12,12 (0).


49 using the general Leibniz rule; specifically

 ∂ 3

a(y)b(y) = a′′′ (y)b(y) + 3a′′ (y)b′ (y) + 3a′ (y)b′′ (y) + a(y)b′′′ (y).
∂y
216 ANDREAS STRÖMBERGSSON

Problem 80:
(See wikipedia; Bertrand-Diquet-Puiseux Theorem.)
Let the chart (U, z) on M be normal coordinates centered at p such that
X := ∂z∂ 1 (p) and Y := ∂z∂ 2 (p) form an ON-basis of Π. (Such a chart is easily
obtained by choosing the identification between Tp M and Rd appropriately
in the construction of normal coordinates.) As usual, we will identify U with
the open ball z(U ) ⊂ Rd (via z). In particular the submanifold expp (Dr )
gets identified with

Dr = {(x, y, 0, . . . , 0) : x2 + y 2 < r 2 }.

From now on we will write “(x, y)” as a short-hand for “(x, y, 0, . . . , 0)”
(denoting either a point in Rd = Tp M or a point in M ).
Let the Riemannian metric be represented by (gij (z)) with respect to
(U, z). Then the induced Riemannian metric on the submanifold expp (Dr )
is represented by the 2 × 2 matrix function
 
g11 (x,y) g12 (x,y)
g21 (x,y) g22 (x,y)
, ∀(x, y) ∈ Dr .

(this is immediate from the definition of the induced Riemannian metric; cf.
Problem 18). Hence by the definition of the volume measure on a Riemann-
ian manifold (cf. #12, p. 1), we have
Z p
(202) Ar = g11 (x, y)g22 (x, y) − g12 (x, y)2 dx dy.
Dr

Recall that gij (0) = δij and gij,k (0) = 0 for all i, j, k ∈ {1, . . . , d} (cf. Lemma
1 in Lecture #4). Hence we have the Taylor expansion
gij,11 (0) 2 gij,22 (0) 2
gij (x, y) = δij + x + gij,12 (0)xy + y + O((x2 + y 2 )3/2 )
2 2
for all (x, y) near (0, 0). This gives

g11 (x, y)g22 (x, y) − g12 (x, y)2

 g11,11 2 g11,22 2  g22,11 2 g22,22 2 
= 1+ x + g11,12 xy + y 1+ x + g22,12 xy + y
2 2 2  2 
+ O (x2 + y 2 )3/2 ,

where all the gij,kl ’s in the right hand side are evaluated at 0. Multiplying
out and using g11,11 (0) = g22,22 (0) = 0 (cf. Problem 79), we get

g11 (x, y)g22 (x, y) − g12 (x, y)2

g22,11 2
g11,22 2  
=1+ x + g11,12 + g22,12 xy + y + O (x2 + y 2 )3/2 .
2 2
 PROBLEMS; “RIEMANNIAN GEOMETRY” 217
√
Hence, using the fact that 1 + α = 1+ 21 α+O(α2 ) for α near 0, we conclude
that for all (x, y) sufficiently near 0:
p
g11 (x, y)g22 (x, y) − g12 (x, y)2
g22,11 2 g11,12 + g22,12 g11,22 2  
=1+ x + xy + y + O (x2 + y 2 )3/2 .
4 2 4
Inserting this in (202), we note that the xy-term gives a 0-contribution, since
the function xy is odd wrt x. Passing to polar coordinates we now get, for
r > 0 sufficiently small
Z r Z 2π    
2 g22,11 2 g11,22 2 3
Ar = 1 + r1 cos ϕ + sin ϕ + O(r1 ) r1 dϕ dr1
0 0 4 4
πr 4

= πr 2 + g22,11 + g11,22 + O(r 5 ).
16
Hence
πr 2 − Ar 3
3
lim 12 4
= − g22,11 + g11,22 = − g11,22 ,
r→0 + πr 4 2
where the last equality holds by Problem 79. On the other hand we have
D ∂ ∂ E
k
K(Π) = K(X ∧ Y ) = K(X, Y ) = hR(X, Y )Y, Xi = R212 ,
∂xk ∂x1
1 1
3
= R212 = g12,12 + g12,12 − g22,11 − g11,22 = − g11,22 ,
2 2
where at the end we used Lemma 3 of Lecture #14 and then the relations
from Problem 79. Hence
πr 2 − Ar
lim 12 = K(Π),
r→0+ πr 4
as desired! 
218 ANDREAS STRÖMBERGSSON

Problem 81: Let I : S → R3 be the inclusion map, i.e. I(x, α) =

(x, f (x) cos α, f (x) sin α). Then
∂ 
dI(x,α) = (1, f ′ (x) cos α, f ′ (x) sin α)
∂x
and
 ∂ 
dI(x,α) = (0, −f (x) sin α, f (x) cos α).
∂α
Hence on S we have (cf. Problem 18):
D∂ ∂ E
, = 12 + (f ′ (x) cos α)2 + (f ′ (x) sin α)2 = 1 + f ′ (x)2 ;
∂x ∂x
D∂ ∂ E

, = f ′ (x)f (x) − cos α sin α + cos α sin α = 0;
∂x ∂α
D ∂ ∂ E
, = f (x)2 .
∂α ∂α
In other words, the matrix representing the Riemannian metric on S wrt
the (x, α) coordinates is:
 
1 + f ′ (x)2 0
g(x, α) = .
0 f (x)2

We also note that the inverse matrix is:

 
−1 (1 + f ′ (x)2 )−1 0
g(x, α) = .
0 f (x)−2

From this, using the formula for the Christoffel symbols of the Levi-Civita
connection,

Γijk = 12 g il (gjl,k + gkl,j − gjk,l ),

we compute:

f ′ (x)f ′′ (x) f ′ (x)

Γ111 (x, α) = ; Γ212 (x, α) = Γ221 (x, α) = ;
1 + f ′ (x)2 f (x)
f (x)f ′ (x)
Γ122 (x, α) = − ,
1 + f ′ (x)2

while all other functions Γijk are identically zero. Suppose now that c(t) =
(x(t), α(t)) is a C ∞ curve on S and s(t) is a vector field along c; we write

∂ ∂
(203) s(t) = a1 (t) + a2 (t) .
∂x ∂α
 PROBLEMS; “RIEMANNIAN GEOMETRY” 219

Then by the formula for ṡ(t) in local coordinates (cf. Lecture #9, p. 8):
∂ ∂  ∂ ∂ 
ṡ(t) = ȧ1 (t) + ȧ2 (t) + ċj (t)ak (t) Γ1jk (c(t)) + Γ2jk (c(t))
∂x ∂α ∂x ∂α
∂ ∂ f ′ f ′′ ∂ f ′  ∂ ff′ 2 ∂
= ȧ1 + ȧ2 + 2 ẋa 1
+ ẋa 2
+ α̇a1
− 2 α̇a
∂x ∂α 1 + f ′ ∂x f ∂α 1 + f ′ ∂x
(204)
 f ′ f ′′ ff′   f′ 2  ∂
2 ∂
= ȧ1 + ẋa 1
− α̇a + ȧ2
+ ẋa + α̇a 1
.
1 + f ′2 1 + f ′2 ∂x f ∂α

(a). The equation for c being a geodesic is ∇ċ ċ = 0. But the vector field
s(t) = ċ(t) is given by (203) with a1 = ẋ and a2 = α̇. Hence the equation
becomes (cf. (204)):

 f ′ (x)f ′′ (x) 2 f (x)f ′ (x) 2

 ẍ + ẋ − α̇ = 0
 1 + f ′ (x)2 1 + f ′ (x)2
(205)

 f ′ (x)

α̈ + 2 ẋα̇ = 0.
f (x)

To prove that f (x(t))2 α̇(t) remains constant along any geodesic, we simply
note that
d 
f (x)2 α̇ = 2f (x)f ′ (x)ẋα̇ + f (x)2 α̈ = 0,
dt
where the last equality holds by the second equation in (205).
Remark: The fact that f (x)2 α̇ remains constant around any geodesic is
called Clairaut’s relation. Note that f (x(t))α̇(t) = kċ(t)k sin ψ(t), where
ψ(t) is the angle between ċ(t) and the meridians of S. Hence (since kċ(t)k
is constant along any geodesic) an equivalent formulation of the relation is
to say that f (x) sin ψ remains constant along any geodesic.
From (205) we see that the geodesics with x ≡ constant are exactly the
curves c(t) = (k1 , k2 + k3 t) with k1 , k2 , k3 ∈ R, and [k3 = 0 or f ′ (k1 ) = 0].
Any curve c(t) = (k1 , k2 + k3 t) with k3 6= 0 is called a parallel of S, and
what we have just shown is that a parallel of S is a geodesic iff its x-value
satisfies f ′ (x) = 0.
Finally, let us consider a curve with α ≡ constant, i.e. c(t) = (x(t), α).
By (205), this is a geodesic iff
f ′ (x)f ′′ (x) 2
(206) ẍ + ẋ ≡ 0.
1 + f ′ (x)2
We may note that the function x(t) satisfies (206) iff
C
(207) ẋ(t) ≡ p for some constant C ∈ R.
1 + f ′ (x(t))2
220 ANDREAS STRÖMBERGSSON

[Proof: Note that for every real-valued C ∞ function x(t) we have

d p  p f ′ (x)
ẋ(t) 1 + f ′ (x(t))2 = ẍ 1 + f ′ (x)2 + ẋ · p · f ′′ (x) · ẋ
dt 1 + f ′ (x)2
p  f ′ (x)f ′′ (x) 2 
= 1 + f ′ (x)2 ẍ + · ẋ .
1 + f ′ (x)2
p
This implies that the function ẋ(t) 1 + f ′ (x(t))2 is constant iff (206) holds.
This is the desired equivalence.]
The equation (207) is seen to be equivalent with the statement that c(t) =
(x(t), α) viewed as a curve in R3 (that is, c(t) = (x(t), f (x(t)) cos α, f (x(t)) sin α)),
is parametrized proportionally to arc length. In particular, up to rescaling
the parametrization and changing direction, there exists a unique geodesic
α ≡ constant for every choice of the constant α! Such a curve is called a
meridian of S. 

(b). The equation for parallel transport along a given curve c(t) =
(x(t), α(t)) is ṡ(t) = 0, i.e., by (204):

 1 f ′ (x)f ′′ (x) 1 f (x)f ′ (x) 2

 ȧ + ẋa − α̇a = 0
 1 + f ′ (x)2 1 + f ′ (x)2
 2 f ′ (x)  2



ȧ + ẋa + α̇a1 = 0.
f (x)
We should consider this for the curve c(t) = (x, t), t ∈ [0, 2π]. Then the
above equation becomes:

 1 f (x)f ′ (x) 2

 ȧ − a =0
 1 + f ′ (x)2
(208)

 ′
ȧ2 + f (x) a1 = 0.

f (x)

If f ′ (x) = 0 then the equation implies that both a1 and a2 are constant.
Now assume f ′ (x) 6= 0. Then differentiating the first equation and sub-
stituting the second into the result gives:
f ′ (x)2
ä1 (t) = − a1 (t).
1 + f ′ (x)2
f ′ (x)2
Here − < 0; hence the general solution is
1 + f ′ (x)2
 
1 f ′ (x)
a (t) = C1 sin C2 + p t
1 + f ′ (x)2
 PROBLEMS; “RIEMANNIAN GEOMETRY” 221

with C1 , C2 ∈ R. Then from the first equation in (208) we get

 
2 1 + f ′ (x)2 f ′ (x) f ′ (x)
a (t) = C1 p cos C2 + p t
f (x)f ′ (x) 1 + f ′ (x)2 1 + f ′ (x)2
p  
1 + f ′ (x)2 f ′ (x)
= C1 cos C2 + p t .
f (x) 1 + f ′ (x)2
It is nicer to express s(t) in terms of the following ON basis for T(x,α) S:
1 ∂ 1 ∂
(209) b1 := p , b2 := .
1+ f ′ (x)2 ∂x f (x) ∂α
We get:



(210) s(t) = Ce1 sin C2 + σx t b1 + cos C2 + σx t b2
′ p
where σx := √ f (x)
′ 2
e1 = C1 1 + f ′ (x)2 . Thus in these ON co-
and C
1+f (x)
ordinates the vector is simply rotating at constant speed along the curve.


Alternative; a more geometrical solution (outline) (cf., e.g., stackexchange):

Let S ′ be the cone (or cylinder, if f ′ (x) = 0) in R3 which is tangent to S
along c. It follows from [12, Thm. 4.7.1] that the Levi-Civita connections
for S and S ′ are equal at every point along c; hence also parallel trans-
port along c is the same in S and S ′ . But we can imagine the cone S ′
being constructed by rolling a paper; unrolling the paper then gives a (lo-
cal) isometry between S ′ and R2 with its standard Riemannian metric; and
in R2 parallel transport of any vector along any curve means simply keeping
the vector constant in the standard coordinates on Tc(t) R2 = R2 . Our curve
′
c is mapped by the isometry to an arc of angle √ |f′ (x)|2 2π along a circle
f (x) +1
√ ′ 2
f (x) +1
with radius f (x) · |f ′ (x)| . Parallel transport of any vector v ∈ R2 along
this circle means that the angle between v and ċ increases/decreases with
′
a constant rate, with the total change being √ |f′ (x)|2 2π as t goes from 0
f (x) +1
to 2π. Hence in the basis b1 , b2 (209) The general form of such a motion
′
is indeed given by (210), with σx := ± √ |f (x)|
′ 2
. Further inspection of the
1+f (x)
cone unfolding argument shows that in the (x, α) coordinates, v rotates in
positive direction if f ′ (x) < 0, and negative direction if f ′ (x) > 0; thus in
′
fact in (210) we have σx := √ f (x) ′ 2
. 
1+f (x)
222 ANDREAS STRÖMBERGSSON

(c). Since dim Tp S = dim S = 2 we can indeed speak of the sectional

curvature of S at a point p ∈ S, and this sectional curvature equals K(X ∧Y )
where X, Y is any basis for Tp S; cf. Problem 69. We compute this with
∂ ∂
X = ∂x , Y = ∂α at the point p = (x, α). First note:
∂ ∂  D ∂ ∂  ∂ ∂ E D ∂ ∂ ∂ E
1 2
K , = R , , = R212 + R212 ,
∂x ∂α ∂x ∂α ∂α ∂x ∂x ∂α ∂x
= (1 + f ′ (x)2 ) · R212
1
(x, α).

Now use the formula from Lecture #11, p. 3:

∂Γ122 ∂Γ112
1
R212 = − + Γ11j Γj22 − Γ12j Γj12
∂x ∂α
∂  f (x)f ′ (x)  f ′ (x)f ′′ (x) f (x)f ′ (x)
=− − ·
∂x 1 + f ′ (x)2 1 + f ′ (x)2 1 + f ′ (x)2
f (x)f ′ (x) f ′ (x)
+ ·
1 + f ′ (x)2 f (x)
(f ′ 2 + f f ′′ )(1 + f ′ 2 ) − 2f f ′ 2 f ′′ f f ′ 2 f ′′ f ′2
=− − +
(1 + f ′ 2 )2 (1 + f ′ 2 )2 1 + f ′ 2
f (x)f ′′ (x)
=− .
(1 + f ′ (x)2 )2
Hence
∂ ∂  f (x)f ′′ (x)
K , =− .
∂x ∂α 1 + f ′ (x)2
∂ ∂
Also, since h ∂x , ∂α i = 0,
2 2 2
∂ ∂
∧ = ∂ · ∂ = (1 + f ′ (x)2 )f (x)2 .
∂x ∂α ∂x ∂α
Hence
 
∂ ∂ f (x)f ′′ (x) 1 f ′′ (x)
K ∧ =− · = − .
∂x ∂α 1 + f ′ (x)2 (1 + f ′ (x)2 )f (x)2 f (x)(1 + f ′ (x)2 )2

f ′′ (x)
Answer: The sectional curvature at (x, α) is − .
f (x)(1 + f ′ (x)2 )2

In particular we note that the sectional curvature at (x, α) is positive iff

f ′′ (x) < 0 and negative iff f ′′ (x) > 0.
√
Finally for f (x) = r 2 − x2 we compute that
f ′′ (x) 1
− = 2 for all x ∈ (−r, r).
f (x)(1 + f ′ (x)2 )2 r
This is indeed the scalar curvature at any point of a sphere of radius r. 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 223

Problem 82:
(Cf. [14, p. 128, Problem 7-3].)
We have
(∇2 η)(X, Y, Z) = (∇Z (∇η))(X, Y )


= Z (∇η)(X, Y ) − (∇η)(∇Z X, Y ) − (∇η)(X, ∇Z Y )


= Z (∇Y η)(X) − (∇Y η)(∇Z X) − (∇∇Z Y (η))(X)
 

= Z Y (η(X)) − η(∇Y X) − Y η(∇Z X) + η(∇Y ∇Z X)
− (∇Z Y )(η(X)) + η(∇∇Z Y X).
(In the above computation, in the first and the third equalities we used the
r (M ) given in the problem formulation,
definition of ∇ : ΓTsr (M ) → ΓTs+1
while in the second and the fourth equalities we used the formula from Prob-
lem 59(a).) Subtracting the corresponding expression for (∇2 η)(X, Z, Y )
from the above expression, and using
Z(Y (η(X))) − Y (Z(η(X))) − (∇Z Y )(η(X)) + (∇Y Z)(η(X))
 
= [Z, Y ] − ∇Z Y + ∇Y Z (η(X)) = 0,
we obtain:
(∇2 η)(X, Y, Z) − (∇2 η)(X, Z, Y )
 
= η ∇Y ∇Z X − ∇Z ∇Y X + ∇∇Z Y X − ∇∇Y Z X
 
= η ∇Y ∇Z X − ∇Z ∇Y X − ∇[Y,Z]X
 
= η R(Y, Z)X .
Done! 
224 ANDREAS STRÖMBERGSSON

Alternative: It suffices to prove that the two functions (∇2 η)(X, Y, Z)−
(∇2 η)(X, Z, Y ) and η(R(Y, Z)X) have the identical restrictions to the set
U for any chart (U, x). Given such a chart, by expanding each of X, Y, Z
in the basis of sections ∂x∂ 1 , . . . , ∂x∂ d ∈ Γ(T U ), and using the fact that both
(∇2 η)(X, Y, Z)−(∇2 η)(X, Z, Y ) and η(R(Y, Z)X) are C ∞ (M )-linear in each
of X, Y, Z, we see that it suffices to prove that, for any i, j, k,
(211)
 ∂ ∂ ∂   ∂ ∂ ∂    ∂ ∂  ∂ 
(∇2 η) , , − (∇ 2
η) , , = η R , .
∂xi ∂xj ∂xk ∂xi ∂xk ∂xj ∂xj ∂xk ∂xi
∂
From now on let us use the short-hand notation ∂i := ∂x i . Take η1 , . . . , ηd ∈
∞
C (U ) so that
η|U = ηl dxl .
Then in U we have


∇η = ∇(ηl dxl ) = dxl ⊗ dηl + ηl ∇(dxl ) = dxl ⊗ (∂m ηl )dxm − ηl Γlab dxb ⊗ dxa
= (∂a ηb − ηl Γlab ) dxb ⊗ dxa ,
where we used [12, (4.1.22)]. It follows that, for any k,
 
l b a
∇∂k (∇η) = ∇∂k (∂a ηb − ηl Γab ) dx ⊗ dx
 
= ∂k (∂a ηb − ηl Γlab ) dxb ⊗ dxa + (∂a ηb − ηl Γlab ) ∇∂k (dxb ) ⊗ dxa
+ (∂a ηb − ηl Γlab ) dxb ⊗ ∇∂k (dxa )
 
= ∂k ∂a ηb − (∂k ηl )Γlab − (∂k Γlab )ηl dxb ⊗ dxa
− (∂a ηb − ηl Γlab ) Γbkc dxc ⊗ dxa − (∂a ηb − ηl Γlab ) Γakc dxb ⊗ dxc

= ∂k ∂a ηb − (∂k ηl )Γlab − (∂k Γlab )ηl − (∂a ηc ) Γckb + ηl Γlac Γckb

− (∂c ηb ) Γcka + ηl Γlcb Γcka dxb ⊗ dxa .
This means that
    
(∇2 η) ∂i , ∂j , ∂k = ∇∂k (∂η) ∂i , ∂j
= ∂k ∂j ηi − (∂k ηl )Γlji − (∂k Γlji )ηl − (∂j ηc ) Γcki + ηl Γljc Γcki − (∂c ηi ) Γckj + ηl Γlci Γckj .
Subtracting the corresponding expression with j and k swapped (and using
Γckj = Γcjk , which holds since the Levi-Civita connection is torsion free; cf.
Lemma 2 in #13), we obtain
     
(∇2 η) ∂i , ∂j , ∂k − (∇2 η) ∂i , ∂k , ∂j = −∂k Γlji + Γljc Γcki + ∂j Γlki − Γlkc Γcji ηl .
Comparing with the formula for R on p. 3 in Lecture #11, we get
l
= Rijk ηl = η(R(∂j , ∂k )∂i ).
l ∂ and η = η dxl .) Hence
(The last equality holds since R(∂j , ∂k )∂i = Rijk l l
we have proved (211)! 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 225

Problem 83:
(a). One simple construction is as follows: Equip M with an arbitrary
Riemannian metric. This is possible by [12, Thm. 1.4.1]. Let exp : D → M
be the corresponding exponential map with its maximal domain D (an open
subset of T M ). By a simple compactness argument (using the fact that
0p ∈ D for all p ∈ M ), there exists some ε > 0 such that s · Y (t) ∈ D for all
t ∈ [0, 1] and all s ∈ (−ε, ε). Now define
(212) c(t, s) := exp(s · Y (t)), (t ∈ [0, 1], s ∈ (−ε, ε)).
This is easily verified to be a smooth variation of the given curve c with
c′ = Y , which is furthermore proper if Y (0) = 0 = Y (1). 

(b). (Outline.) First assume c(0) 6= c(1). Then it is possible to choose

the Riemannian metric on M so that for each j ∈ {0, 1}, either Y (j) = 0
or else γj is a geodesic. [Indeed, inspecting the construction in [12, Thm.
1.4.1] we see that it suffices to prove that if γ̇j (0) = Y (j) 6= 0 then there
exist ε > 0 and an open neighborhood U of γj (0) which can be equipped
with a Riemannian metric such that γj|[−ε,ε] is a geodesic in U . And this
can be constructed by letting U be the domain of a chart (U, x) such that
x(γj (t)) = (t, 0, . . . , 0) for all t near 0 (as is possible by Problem 12), and
then equipping U with the Riemannian metric inherited from the standard
Riemannian metric on Rd via x : U → Rd .] With this choice, the variation
in (212) again has all the desired properties.
If c(0) = c(1) then the above construction can be modified e.g. as follows:
For j = 0, 1, choose a Riemannian metric mj on M such that if Y (j) 6= 0
then γj is a geodesic wrt mj . Then also um0 + (1 − u)m1 is a Riemannian
metric on M for each u ∈ [0, 1], and we denote by exp( · ; u) : Du → M the
corresponding exponential map. Now one can prove that there is some ε > 0
such that s · Y (t) ∈ Dt for all t ∈ [0, 1] and s ∈ (−ε, ε). Then define
c(t, s) := exp(s · Y (t); t).
This is a variation having the required properties. 
Remark: Note that when we apply the result of this Problem 83 in prac-
tice, M often comes already equipped with a Riemannian metric; however
the Riemannian metric which is chosen in the above construction may well
be another (“completely unrelated”) Riemannian metric!
226 ANDREAS STRÖMBERGSSON

Problem 84: Lemma 1 in Lecture #16 implies that, in our situation,

E ′ (s) = L′ (s) = 0 for all s. Hence E(s) and L(s) are indeed constant
functions. 

Problem 85:
(a). Identify Tp M with Rd by some fixed linear map respecting the inner
product. Let γ1 and γ2 be the following C ∞ curves in Tp M = Rd :
γ1 (t) = te1 , t ∈ [0, π],
where e1 = (1, 0, . . . , 0), and
γ2 (t) = (π cos t, π sin t, 0, 0, . . . , 0), t ∈ [0, a],
where a is any fixed positive constant. Then set:
γ = γ1 · γ2 · γ2 .
In order to fit into the problem formulation, this product path should be
understood to be reparametrized in some way so that the domain of γ is
[0, 1].
Note then that
γ(1) = γ1 (π) = πe1 .
Also kγ2 (t)k = π and thus expp (γ2 (t)) = −p for all t ∈ [0, a]! Hence
L(expp ◦γ) = L(expp ◦γ1 ) = kγ1 (π)k = kγ(1)k,
where the second equality holds since t 7→ expp ◦γ1 (t) = expp (te1 ) is a
geodesic. However γ is certainly not a reparametrization of the curve t 7→
t · γ(1) = t · πe1 (t ∈ [0, 1]), since the image of γ in Rd contains points outside
the line Re1 . 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 227

(b). Writing e1 , . . . , ed for the standard basis vectors in Rd , our task is to

prove that


(213) d(expp ◦y −1 )ỹ (e1 ), d(expp ◦y −1 )ỹ (e1 ) = 1
and


(214) d(expp ◦y −1 )ỹ (e1 ), d(expp ◦y −1 )ỹ (ej ) = 0 for j = 2, . . . , d.
Now for any v ∈ Rd we have
d(expp ◦y −1 )ỹ (v) = (d expp )y−1 (ỹ) ◦ (dy −1 )ỹ (v).
Furthermore, the “polar coordinates” assumption implies that, for any ỹ ∈
y(W ), if w = y −1 (ỹ) ∈ W then
(215) kwk = ỹ 1 > 0; (dy −1 )ỹ (e1 ) = kwk−1 w;
and


(216) (dy −1 )ỹ (ej ), w = 0, for j = 2, . . . , d.
[Proof: We have y 1 (w′ ) = kw′ k ≥ 0 for all w′ ∈ W , and y(W ) is open; hence
y 1 (w′ ) > 0 for all w′ ∈ W , and in particular the first relation in (215) holds.
Next consider the curve c(t) = (1+t)w (−ε < t < ε) in Tp M ; for ε sufficiently
small this curve is contained in W , and the polar coordinates assumption
implies that y(c(t)) = ((1 + t)ỹ 1 , ỹ 2 , . . . , ỹ d ) for all t ∈ (−ε, ε). Considering
the tangent vector of this curve at t = 0 we find dyw (w) = ye1 e1 , and this
implies the second relation in (215). Finally for given j ∈ {2, . . . , d} consider
the curve γ(t) = ỹ + tej (−ε < t < ε) in Rd ; for ε sufficiently small this
curve is contained in y(W ), and the polar coordinates assumption implies
that the curve y −1 ◦γ is contained in the sphere {w′ ∈ W : kw′ k = kwk > 0}.
Note also y −1 (γ(0)) = w. Hence the tangent vector of y −1 (γ(t)) at t = 0 is
orthogonal to w, and since γ̇(0) = ej this implies (216).]
In view of the above, (213) is equivalent with50


(217) (d expp )w (kwk−1 w), (d expp )w (kwk−1 w) = 1, ∀w ∈ W
and in order to prove (214) it suffices to prove that


(218) (d expp )w (kwk−1 w), (d expp )w (v) = 0,
whenever w ∈ W , v ∈ Tp M , hw, vi = 0.
However these two statements are clearly implied by “Gauss Lemma”! 

50Recall that we saw in (215) that kwk > 0 for all w ∈ W ; thus the statements (217)
and (218) make sense.
228 ANDREAS STRÖMBERGSSON

(c). Outline: We follow the argument on pp. 8–9 in Lecture #4 (mutatis

mutandis) using the fact proved in part b (where of course the bottom right
(d − 1) × (d − 1) submatrix must be everywhere positive semidefinite). Note
that in that argument, in
Z 1 Z 1
L(γ) = ·
k(expp ◦γ) (t)k dt ≥ |ṙ(t)| dt ≥ |r(1) − r(0)| = kvk,
0 0
we now cannot claim that equality in the first inequality holds iff ϕ̇(t) = 0
∀t ∈ (0, 1), namely since (d expp )γ(t) may be singular. However equality
in the second equality holds iff ṙ(t) ≥ 0 ∀t ∈ (0, 1). Thus, since we are
assuming L(γ) = kvk, the function r(t) = kγ(t)k must be increasing, and so
r(t) ≤ kγ(1)k = kvk for all t ∈ [0, 1].
Now assume that there is some t ∈ (0, 1) with γ(t) ∈ / [0, 1]·kvk, and let t0 ∈
(0, 1] be the supremum of the set of such t. By continuity, γ(t0 ) ∈ [0, 1] · kvk,
say γ(t0 ) = t1 kvk (t1 ∈ [0, 1]). Now since the point c(t1 ) = exp(t1 v) is
not conjugate to c(0) along c, we have that (d expp )γ(t0 ) is non-singular, and
hence (d expp )w is non-singular for all w in some open neighborhood Ω ⊂ Dp
of γ(t0 ). On the other hand it follows from the definition of t0 that there
exist t-values < t0 arbitrarily near t0 where γ(t) ∈ / [0, 1] · kvk. Hence there
also exist t-values < t0 arbitrarily near t0 where ϕ̇(t) 6= 0. Since such a t can
be found arbitrarily near t0 , we can ensure that γ(t) ∈ Ω. Both ϕ̇(t) 6= 0
and γ(t) ∈ Ω are “open” conditions; hence there must in fact exist a whole
open interval (t2 , t2 + η) ⊂ (0, t0 ) (η > 0) such that ϕ̇(t) 6= 0 and γ(t) ∈ Ω
for all t ∈ (t2 , t2 + η). Now we can conclude
Z t2 +η Z t2 +η
·
k(expp ◦γ) (t)k dt > kṙ(t)| dt,
t2 t2
since k(expp ◦γ)· (t)k > kṙ(t)| for all t ∈ (t2 , t2 + η), and so in total we must
have a strict inequality L(γ) > kvk, contradicting our assumption.
Hence we must have γ(t) ∈ [0, 1] · kvk for all t; and since also r(t) is
increasing, it follows that γ is a reparametrization of the curve t 7→ tv. 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 229

Problem 86: As stated in the lecture notes, this is clear from Cor. 1
in Lecture #17. Indeed, after possibly shrinking and reparametrizing the
geodesic, and possibly changing its direction, we may assume t0 = a = 0,
t1 = b = T > 0. Set also p = c(0) and q = c(T ). Then our task is to prove
that c(0) and c(T ) are conjugate along c iff
(d expp )T ·ċ(0) : Tp (M ) → Tq (M )
is singular. By Cor. 1 in #17, for every v ∈ Tp (M ) we have that (d expp )T ·ċ(0) (v)
equals X(T ) when X is the unique Jacobi field along c with X(0) = 0,
Ẋ(0) = v. 51 Hence, since also Tp (M ) and Tq (M ) have the same dimension,
(d expp )T ·ċ(0) is singular iff there is some v 6= 0 in Tp (M ) with such that the
unique Jacobi field along c with X(0) = 0 and Ẋ(0) = v satisfies X(T ) = 0.
In other words, (d expp )T ·ċ(0) is singular iff there is some Jacobi field X 6≡ 0
along c with X(0) = 0 and X(T ) = 0, i.e. iff c(0) and c(T ) are conjugate
along c. 

51The fact that there indeed exists a unique such Jacobi field is provided by Lemma 1
in #17.
230 ANDREAS STRÖMBERGSSON

Problem 88: As in the proof of Lemma 1 in #17, let X1 , . . . , Xd ∈ Vc be

parallel vector fields along c such that X1 (t), . . . , Xd (t) forms an ON-basis
Tc(t) (M ) for each t ∈ [a, b], and define ρki ∈ C ∞ ([a, b]) by R(Xi , ċ)ċ = ρki Xk .
Then any pw C ∞ vector field Y along c can be (uniquely) expressed as
Y = ξ i Xi ,
where ξ1 , . . . , ξd are pw C ∞ functions [a, b] → R. We then also have
Z b  

(219) I(Y, Y ) = ∇ ∂ Y k2 − hR(Y, ċ)ċ, Y i dt
∂t
a
Z b X
d d X
X d 
= ξ̇i (t)2 − ρki (t)ξi (t)ξk (t) dt.
a i=1 i=1 k=1
Hence the problem is reduced to a problem in real analysis: We see that it
now suffices to prove that if f is any pw C ∞ function f : [a, b] → R, then
there exists a sequence f1 , f2 , . . . of functions in C ∞ ([a, b]) such that
(220) kfk − f kL∞ → 0 and kfk′ − f ′ kL1 → 0,
and kfk′ kL∞ stays bounded as k → ∞,
and that we may furthermore choose this sequence so that fk (t) = f (t) for
all k and all t ∈ [a, b] \ ∪m−1
j=1 (tj − ε, tj + ε), where ε > 0 is any fixed constant
and t1 < t2 < · · · < tm−1 are any given numbers in (a, b) including all
’break-points’ of f (viz., f|[tj−1 ,tj ] ∈ C ∞ ([tj−1 , tj ]) for j = 1, . . . , m, where
t0 = a and tm = b).
(Indeed, if we can prove the statement of the last sentence, then we apply
it to each of the functions ξ1 , . . . , ξd describing the given pw C ∞ vector field
Y , with fixed ε > 0 and with t1 < · · · < tm−1 being all the ’break-points’
of Y in (a, b). This gives a sequence of C ∞ vector fields Z1 , Z2 , . . . along c
each satisfying Zk (t) = Y (t) for all t ∈ [a, b] \ ∪m−1 j=1 (tj − ε, tj + ε), and, as
one verifies using (219): I(Zk , Zk ) → I(Y, Y ) as k → ∞.)
Outline of real analysis argument: Thus assume that f : [a, b] → R is
pw C ∞ and let a = t0 < t1 < · · · < tm = b be such that f|[tj−1 ,tj ] ∈
C ∞ ([tj−1 , tj ]) for j = 1, . . . , m. In fact we may extend f to a pw C ∞
function (a − ε, b + ε) → R so that f|(a−ε,t1 ] and f|[tm−1 ,b+ε) are C ∞ . (Cf.
∞
the lecture notes, Sec. 3.1.)R Fix a C function φ : R → R≥0 with support
−1 φ(η −1 t);
contained in (−1, 1) and R φ = 1. For every η > R 0 set φη (t) := η
then φη has support contained in (−η, η) and R φη = 1. Now choose any
sequence ε > η1 > η2 > · · · > 0 with ηk → 0, and set fk := φηk ∗ f
(convolution), i.e.
Z
fk (t) = f (t − x)φηk (x) dx.
R
It follows from supp(φηk ) ⊂ (−ε, ε) that fk is well-defined and C ∞ for t ∈
[a, b]. One now verifies that (220) holds (in fact |fk′ |L∞ ≤ kf ′ kL∞ for all k).
 PROBLEMS; “RIEMANNIAN GEOMETRY” 231

Finally one can use a partition of unity argument, creating fknew weighting
appropriately between fk and f , to also ensure fk (t) = f (t) for all k and all
t ∈ [a, b] \ ∪m−1
j=1 (tj − ε, tj + ε) (while (220) remains true). 

Problem 89: We are assuming that expp is defined and injective on the
open ball Br (0) in Tp (M ). We claim that (d expp )v is non-singular for every
v ∈ Br (0). Assume the opposite, i.e. assume that v ∈ Br (0) is such that
(d expp )v is singular. Take b > 1 so that bv ∈ Br (0), and let c be the geodesic
c : [0, b] → M ; c(t) := expp (tv).
Our assumption that (d expp )v is singular implies that the point c(1) is
conjugate to c(0) along c (by Problem 86). Hence by Theorem 1 in Lecture
#18, c is not a local minimum for L among pw C ∞ curves from p to q :=
c(b). This implies in particular that d(p, q) < L(c). Now by Problem 28
there is a geodesic from p to q realizing the distance d(p, q), i.e. there is
some w ∈ Tp (M ) such that q = expp (w) and kwk = d(p, q). Of course
w ∈ Br (0) and w 6= bv, since kwk = d(p, q) < L(c) = kbvk < r. Now
we have expp (bv) = q = expp (w), contradicting the fact that expp|Br (0) is
injective. This completes the proof that (d expp )v is non-singular for every
v ∈ Br (0).
From the non-singularity just proved it follows, via the Inverse Func-
tion Theorem, that expp|Br (0) is a local diffeomorphism and that U :=
expp (Br (0)) is an open subset of M . Since expp|Br (0) is injective, this map
is a bijection of Br (0) onto U . Let f : U → Br (0) be the inverse map. The
fact that expp|Br (0) is a local diffeomorphism implies that f is C ∞ in all U .
Hence expp|Br (0) is a diffeomorphism onto the open set U . 
232 ANDREAS STRÖMBERGSSON

Problem 90: Let (U, x) and (V, y) be two charts with p ∈ U , p ∈ V , and
assume that
∂ ∂
(221) · · · jr f = 0 at p,
∂xj1 ∂x
for any 1 ≤ r ≤ k and any j1 , . . . , jr ∈ {1, . . . , d}. We know that
∂ ∂
= ϕji (∀i ∈ {1, . . . , d})
∂y i ∂xj
(equality of vector fields in ΓT (U ∩ V )), where
∂xj
ϕji := ∈ C ∞ (U ∩ V ).
∂y i
Now take any r ∈ {1, . . . , k} and i1 , . . . , ir ∈ {1, . . . , d}. Then in U ∩ V we
have
   
∂ ∂ j1 ∂ jr ∂
· · · ir f = ϕi1 · · · ϕir f,
∂y i1 ∂y ∂xj1 ∂xjr
and this can be expanded as a sum where each term is of the form “A · B”
where each “A” is a product of partial derivatives of some of the functions
∂ ∂
ϕjill , and each “B” equals jl(1) · · · f for some 1 ≤ l(1) < l(2) < · · · <
∂x ∂xjl(s)
l(s) ≤ r (with 1 ≤ s ≤ r). Evaluating this sum at p, each B-factor vanishes,
because of (221) (and since s ≤ r ≤ k). Hence
∂ ∂
i
· · · ir f = 0 at p.
∂y 1 ∂y

 PROBLEMS; “RIEMANNIAN GEOMETRY” 233

Problem 91:
(a). Let (U, x) be any chart on S d and let (gij (x)) give the standard
Riemannian metric h·, ·i with respect to (U, x). Then the Riemannian metric
[·, ·] is given by (hij (x)) where52
hij (x) = f (x) · gij (x), ∀i, j ∈ {1, . . . , d}, x ∈ x(U ).

Now for any j, l, k, and for all x ∈ x(U ):

∂ ∂    ∂  ∂
hjl (x) = f (x)gjl (x) = f (x) · gjl (x) + f (x) · gjl (x).
∂xk ∂xk ∂xk ∂xk
If x lies on the curve c then f (x) = 1 and ∂x∂ k f (x) = 0 (since f ∈ F1 ), and
hence
∂ ∂
(222) hjl (x) = gjl (x).
∂xk ∂xk
Also of course hjl (x) = f (x)gjl (x) = gjl (x) for all x along c. Let ∇ and ∇ e
be the Levi-Civita connections on T (S d ) corresponding to the Riemannian
metrics h·, ·i and [·, ·], respectively, and let Γijk and Γ e i be the Christoffel
jk
e
symbols for ∇ and ∇, respectively, with respect to the basis of sections
∂
∂x1
, . . . , ∂x∂ d ∈ Γ(T U ). Then
 
i 1 il ∂ ∂ ∂
(223) Γjk (x) = g (x) gjl (x) + j gkl (x) − l gjk (x)
2 ∂xk ∂x ∂x
and
 
e i 1 il ∂ ∂ ∂
Γjk (x) = h (x) hjl (x) + j hkl (x) − l hjk (x)
2 ∂xk ∂x ∂x
(for all x ∈ x(U )). It follows from the observations made above that
e i (x) = Γi (x)
Γ for all x along c.
jk jk

Since c is a geodesic on S d we have

c̈i (t) + Γijk (c(t)) ċj (t)ċk (t) = 0,
for all i ∈ {1, . . . , d} and all t ∈ [0, π] with c(t) ∈ x(U ). (Here ci := xi ◦ c.)
It follows that also
e i (c(t)) ċj (t)ċk (t) = 0
c̈i (t) + Γjk
for all t ∈ [0, π] with c(t) ∈ x(U ). The fact that this holds for every chart
(U, x) on S d implies that c is a geodesic in Sfd . 

52As usual, “x” denotes two things, namely a map from U to Rd and also a general
point in x(U ); also “f (x)” really stands for “f (x−1 (x))”.
234 ANDREAS STRÖMBERGSSON

(b). Let f ∈ F2 . Then c is a geodesic in Sfd , by part (a). We also know

that in S d , c(0) and c(π) are conjugate along c, and there is no point before
c(π) conjugate to c(0) along c. (This can for example be easily verified from
the explicit formula for a general Jacobi field along a geodesic in constant
curvature; cf. pp. 5–6 in #17. Alternatively, the statement can be proved
using Theorem 1 in #18 combined with the known facts that any arc of
length < π of a great circle on S d is a strict local minimum for L, while any
arc of length > π of a great circle on S d is not a local minimum for L.)
Hence it now suffices to prove that an arbitrary vector field along c is a
Jacobi field in Sfd iff it is a Jacobi field in Sf .
Thus consider an arbitrary vector field X along c. Note that apriori
“Ẋ(t)” stands for different things in S d and Sfd , since it is defined in terms of
the Levi-Civita connection. However in local coordinates, the expression for
Ẋ(t) only involves the Christoffel symbols evaluated at points along c; 53 and
we know from part (a) that these agree for S d and Sfd (since f ∈ F2 ⊂ F1 ).
Hence “Ẋ(t)” means the same thing in S d and Sfd , for any vector field X
along c. Repeated use of this fact implies that also Ẍ(t) means the same
thing in S d and Sfd , for any vector field X along c.

Recall that by definition, X is a Jacobi field iff “Ẍ + R(X, ċ)ċ ≡ 0”;
hence it now only remains to prove that “R(X, ċ)ċ ” stands for the same
thing in S d and Sfd . However, if (U, x) is an arbitrary chart on M , and
J := {t ∈ [0, π] : c(t) ∈ U }, and if X is represented by the functions
a1 , . . . , ad ∈ C ∞ (J) (viz., X(t) = aj (t) · ( ∂x∂ j )c(t) , ∀t ∈ J), then
 ∂ 
k
R(X(t), ċ(t))ċ(t) = Rjim (c(t)) · ai (t) ċm (t) ċj (t) , ∀t ∈ J.
∂xk c(t)
Hence it suffices to prove that “Rjim k (c(t))” is the same for S d and S d , for
f
all t ∈ J. In view of the formula
k
∂Γkmj ∂Γkij
Rjim = − + Γkil Γlmj − Γkml Γlij (valid in all U ),
∂xi ∂xm
∂
it suffices to prove that “Γkmj ” and “ ∂x k
i Γmj ” are the same for S
d and S d ,
f
at every point c(t), t ∈ J (and for all m, j, k, i ∈ {1, . . . , d}). For Γkmj this
53Indeed, let (U, x) be an arbitrary chart on M and set J := {t ∈ [0, π] : c(t) ∈ U };
then there exist unique functions a1 , . . . , ad ∈ C ∞ (J) such that
 ∂ 
X(t) = aj (t) · , ∀t ∈ J.
∂xj c(t)
Then if Γljk ∈ C ∞ (x(U )) are the Christoffel symbols for the Levi-Civita connection, we
have
  ∂ 
Ẋ(t) = ∇ċ(t) X(t) = ȧl (t) + ċj (t) ak (t) Γljk (c(t)) , ∀t ∈ J.
∂xl c(t)
Cf. Lecture # 9, p. 8.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 235

was proved in the solution to part (a). In order to prove it also for the
∂ k
derivative “ ∂x i Γmj ”, we see from the formula (223) that it suffices to prove

that gjl (x) and all its first and second derivatives “are the same” for S d and
Sfd , at every point c(t) (t ∈ J), and also that the corresponding thing holds
for gil (x) and all its first derivatives. We already know this fact for gjl (x)
and gil (x) themselves, and also for all the first derivatives of gjl (x); cf. part
(a). Thus, in the notation from the solution to part (a), what remains to
prove is that, for any k, i, l, j ∈ {1, . . . , d},
∂ il ∂ il
(224) k
g (x) = h (x)
∂x ∂xk
and also
∂2 ∂2
(225) gjl (x) = hjl (x)
∂xi ∂xk ∂xi ∂xk
for all x along c. Note that hil (x) = f (x)−1 · gil (x) throughout x(U ), and
also that for every x along c we have f (x)−1 = 1 and ∂x∂ k (f (x)−1 ) =
−f (x)−2 ∂x∂ k f (x) = 0; with these observations, (224) follows in the same
way as (222). On the other hand, (225) follows from
∂2  ∂2   ∂  ∂
gjl (x) = f (x) · gjl (x) + f (x) · gjl (x)
∂xi ∂xk ∂xi ∂xk ∂xi ∂xk
 ∂  ∂ ∂2
+ f (x) · gjl (x) + f (x) · i k gjl (x),
∂xk ∂xi ∂x ∂x
∂2 ∂
and the fact that if x lies on c then f (x) = 1 and ∂xi ∂xk
f (x) = ∂xi f (x) =
∂
∂xk f (x) = 0 (since f ∈ F2 ). Done! 

(c). Take t1 ∈ [0, π] so that q := c(t1 ) lies in U . Then also c(t) ∈ U for
all t in some neighborhood of t1 in [0, π]. Hence we may assume from start
that t1 ∈ (0, π). Now take r > 0 so small that Bq (r) ⊂ U , where Bq (r) is
the open ball in Sfd of radius r about the point q. Let c1 : [0, π] → Sfd be any
pw C ∞ curve with c1 (0) = c(0) and c1 (π) = c(π) and df (c1 (t), c(t)) < r,
∀t ∈ [0, π]. We then claim that Lf (c1 ) ≥ Lf (c) with equality only if c1 is
a reparametrization of c. Here and in the following we write df and Lf for
the metric and length of curves in Sfd , and we will write d and L for the
corresponding things in S d .
Note that [v, w] ≥ hv, wi for all p ∈ S d , v, w ∈ Tp S d , since f ≥ 1 every-
where. Hence
Z πp Z πp
(226) Lf (c1 ) = [ċ1 (t), ċ1 (t)] dt ≥ hċ1 (t), ċ1 (t)i dt = L(c1 )
0 0
≥ L(c) = Lf (c),
236 ANDREAS STRÖMBERGSSON

where L(c1 ) ≥ L(c) holds since we know that c is a distance minimizing

geodesic in S d , and L(c) = Lf (c) holds since f = 1 along c. Hence we have
proved the desired inequality, Lf (c1 ) ≥ Lf (c), and it only remains to prove
the statement about when equality holds.
Thus assume Lf (c1 ) = Lf (c). Then equality must hold in both “≥” in
(226). The fact that equality holds in the second “≥” in (226) implies that
c1 is a geodesic in S d , up to reparametrization (cf. Problem 24). However we
know that the geodesics in S d are exactly the (pieces of) great circles in S d ;
hence we conclude that c1 is an arc of a great circle between the antipodal
points c(0) and c(π). If this great circle is not equal to (the image of) c
itself then c1 (t) ∈/ c([0, π]) for all t ∈ (0, π). In particular c1 (t1 ) ∈
/ c([0, π]).
But we have df (c1 (t1 ), c(t1 )) < r by assumption, i.e. c1 (t1 ) ∈ Bq (r). Hence
c1 (t1 ) ∈ U \ c([0, π]), and therefore f (c1 (t1 )) < 1. This implies that
[ċ1 (t), ċ1 (t)] > hċ1 (t), ċ1 (t)i
for all t in some neighborhood of t1 , and therefore the first “≥” in (226)
must be a strict inequality, contradicting Lf (c1 ) = Lf (c)! Hence the great
circle c1 ([0, π]) must be equal to c([0, π]), i.e. c1 is a reparametrization of c,
qed. 

(d). Take t1 ∈ (0, π) and r as in part (c). Let c1 : [0, π] → S d be any great
circle from c(0) to c(π), parametrized by arc length, not equal to c itself and
satisfying df (c1 (t1 ), c(t1 )) < r. We will then prove that Lf (c1 ) < Lf (c).
Since such curves c1 can be chosen with supt∈[0,π] df (c1 (t), c(t)) arbitrarily
small54 this will complete the proof that c is not a local minimum for L in
Sfd among pw C ∞ curves with fixed endpoints (in fact it even follows that
there exists a proper variation c(t, s) of c such that L(cs ) < L(c) for all s 6= 0
near 0).
We have:
Z π p Z π p
Lf (c1 ) = [ċ1 (t), ċ1 (t)] dt < hċ1 (t), ċ1 (t)i dt = L(c1 ) = π
0 0
= L(c) = Lf (c).
The “<” in the above computation holds since f ≤ 1 throughout S d and
since by an argument as in part (c) we have
[ċ1 (t), ċ1 (t)] < hċ1 (t), ċ1 (t)i
for all t in some neighborhood of t1 . Hence we have proved Lf (c1 ) < Lf (c),
as desired! 

54Indeed, note that d (p, q) ≤ (sup d √f ) · d(p, q), ∀p, q ∈ S d , and we can make
f S
supt∈[0,π] d(c1 (t), c(t)) arbitrarily small.
 PROBLEMS; “RIEMANNIAN GEOMETRY” 237

Problem 92: (Cf. Cheeger & Ebin [2, Cor. 1.30].)

In view of the definition of the length of a curve in a Riemannian manifold,
d
and the fact that dt (expp (c(t))) = d(expp )c(t) (ċ(t)) (and similarly for expp0 ),
it suffices to prove that
kd(expp )c(t) (ċ(t))k ≥ kd(expp0 )c(t) (ċ(t))k, ∀t ∈ [a, b].
(Here in the left hand side k · k is the norm on Texpp (c(t)) (M ) coming from
the Riemannian metric on M , while in the right hand side k · k is the norm
on Texpp (c(t)) (M0 ) coming from the Riemannian metric on M0 .) We will
0
prove the stronger statement that
(227) kd(expp )x (v)k ≥ kd(expp0 )x (v)k, ∀x ∈ Br (0), v ∈ Tx Rd = Rd .
If v = 0 then (227) is trivial. If x = 0 then d(expp )x is the identity
map on Tp (M ) = Rd = T0 (Rd ), and similarly for d(expp0 )x , and there-
fore (227) holds with equality for all v ∈ Rd . From now on we assume both
v 6= 0 and x 6= 0. Let us decompose v as v = u + w where u ∈ Rx and
v·x
w · x = 0. (Thus u = · x.) Then by Gauss’ Lemma (= Cor. 2 in Lec-
x·x
ture #17), d(expp )x (u) and d(expp )x (w) are orthogonal in Texpp (x) (M ), and
kd(expp )x (u)k = kuk (the standard length of u as a vector in Rd ); hence
q
(228) kd(expp )x (v)k = kuk2 + kd(expp )x (w)k2 .

The analogous formula holds for kd(expp0 )x (v)k, and hence, in order to prove
(227), it suffices to prove the corresponding inequality with w in place of v.
In other words, it suffices to prove (227) under the extra assumption that
v · x = 0. We impose this assumption from now on.
Set
x̂ := kxk−1 x and v̂ := kxk−1 v.
Let γ : [0, kxk] → M be the geodesic γ(t) = expp (tx̂); note that γ is
parametrized by arc length, i.e. kγ̇(t)k = 1 for all t, since kx̂k = 1. Set
J(t) := (d expp )tx̂ (tv̂) for t ∈ [0, kxk].
Then by Corollary 1 (and Lemma 1) in Lecture #17, J is a Jacobi field
along γ with J(0) = 0, J(0) ˙ = v̂. Note that J tan (0) = 0, and J˙tan (0) = 0
since v̂ · x̂ = 0, i.e. hv̂, x̂i = 0 in Tp M . Also J tan is a Jacobi field along γ by
Lemma 3 in Lecture #17; hence J tan ≡ 0 by Lemma 1 in Lecture #17.
Similarly if we also set γ0 (t) = expp0 (tx̂) and
J0 (t) := (d expp0 )tx̂ (tv̂) for t ∈ [0, kxk]
then γ0 is a geodesic in M0 and J0 is a Jacobi field along γ0 with J0 (0) = 0,
J˙0 (0) = v̂ and J0tan ≡ 0.
238 ANDREAS STRÖMBERGSSON

Now let X1 , . . . , Xd be parallel vector fields along γ0 which form an ON-

basis in Tγ(0) (M0 ) (and hence in each tangent space Tγ(t) (M0 )), and which
are chosen so that X1 (0) = γ̇0 (0) (thus X1 (t) = γ̇0 (t) for all t ∈ [0, 1]) and
v̂ = k · X2 (0) for some k > 0
(recall that we assume v 6= 0; thus v̂ 6= 0). Then by pp. 5–6 in Lecture #17,
J0 (t) = k · sµ (t) · X2 (t), ∀t ∈ [0, kxk],
where

 −1/2 sin(µ1/2 t)
µ (µ > 0)
sµ (t) = t (µ = 0)

 −1/2 1/2
|µ| sinh(|µ| t) (µ > 0).
Recall that we are assuming that expp0 restricted to Br (0) is a diffeomor-
phism. This implies that (d expp0 )tx̂ is a linear bijection for each t ∈ [0, kxk],
and in particular J0 (t) 6= 0 for all t ∈ [0, kxk]. This implies that sµ (t) > 0
for all t ∈ (0, kxk]. Now the Rauch Comparison Theorem (Theorem 1 in
Lecture #19) applies to our situation, with
fµ = kJk˙(0) · sµ = k · sµ
˙
(indeed we have kJk˙(0) = kJ(0)k = kv̂k since J(0) = 0), and that theorem
implies
kJ(t)k ≥ fµ (t) = kJ0 (t)k, ∀t ∈ [0, kxk].
Taking t = kxk in the last inequality, we conclude that (227) holds! 
 PROBLEMS; “RIEMANNIAN GEOMETRY” 239

Problem 93:
Assume that τ ∈ (a, b) and that c(τ ) is a focal point of γ along c. Let X
be a nontrivial Jacobi field along c satisfying X(τ ) = 0 and
(229) X(a) ∈ Span(γ̇(0)) and hẊ(a), γ̇(0)i = 0 in Tc(a) (M ).
It follows that X tan (a) = 0 (since hċ(a), γ̇(0)i = 0) and X tan (τ ) = 0; hence
by Lemmata 2,3 in Lecture #17 we must have X tan ≡ 0. Define the pw C ∞
vector field Y along c by
(
X(t) if t ∈ [a, τ ]
Y (t) =
0 if t ∈ [τ, b].
(Note in particular that Y is a well-defined and continuous function, since
X(τ ) = 0.) We have:
Z τ 
1 1
I(Y, Y ) = I(X , X ) = hẊ, Ẋi − hR(ċ, X)X, ċi dt
a
Z τ  
d
= hẊ, Xi − hẌ + R(X, ċ)ċ, Xi dt
a dt
= hẊ(τ ), X(τ )i − hẊ(a), X(a)i = 0,
where the last equality holds since X(τ ) = 0 and because of (229).
◦
Consider any Z ∈ V c (viz., Z ∈ Γc (T M ) with Z(a) = 0 = Z(b)). Write
Z 1 for the restriction of Z to [a, τ ]. Then
I(Y + Z, Y + Z) = I(Y, Y ) + 2 · I(X 1 , Z 1 ) + I(Z, Z)
= 2 · I(X 1 , Z 1 ) + I(Z, Z),
and here
Z τ 
1 1
I(X , Z ) = hẊ, Żi − hR(ċ, X)Z, ċi dt
a
Z τ 
d
= hẊ, Zi − hẌ + R(X, ċ)ċ, Zi dt
a dt
= hẊ(τ ), Z(τ )i − hẊ(a), Z(a)i
= hẊ(τ ), Z(τ )i.
◦
As in the proof of Theorem 1 in #18 we now fix a vector field Z ∈ V c which
is normal and which satisfies Z(τ ) = −Ẋ(τ ).55 Applying the above formulas
with ηZ (η ∈ R) in place of Z gives
I(Y + ηZ, Y + ηZ) = −2 · kẊ(τ )k2 · η + I(Z, Z) · η 2 .

55To see that this can be done, note that −Ẋ(τ ) is normal against ċ(τ ); hence we can
simply set Z equal to the parallel transport of −Ẋ(τ ) along c, multiplied by some smooth
function f : [a, b] → R with f (τ ) = 1 and f (a) = f (b) = 0.
240 ANDREAS STRÖMBERGSSON

Here kẊ(τ )k > 0 since X is nontrivial and X(τ ) = 0 (cf. Lemma 1 in #17).
Hence for η > 0 small enough we have
I(Y + ηZ, Y + ηZ) < 0.
Fix such an η. Note that Y + ηZ is a normal pw C ∞ vector field along c,
and by Problem 88 there is a C ∞ vector field U along c which also satisfies
I(U, U ) < 0, as well as U (t) = Y (t) + ηZ(t) for all t near a or b; in particular
U (a) = X(a) ∈ Span(γ̇(0)) and U (b) = 0. Using the fact that Y + ηZ is
normal along c and inspecting the solution to Problem 88, we see that we
can also take U to be normal along c.
Take k ∈ R so that U (a) = k · γ̇(0). Now by Problem 83(b) there exists
a C ∞ variation c : [a, b] × (−ε, ε) → M of c satisfying c′ ≡ U as well as
c(a, s) = γ(ks) and c(b, s) = c(b) for all s ∈ (−ε, ε). Set L(s) := L(cs ). By
Lemma 1 in #16 (and a remark on p. 3 of #16, and using hU (a), ċ(0)i = 0
and U (b) = 0) we have L′ (0) = 0. Next we apply Theorem 1 in #16. Note
that c′ ⊥ ≡ c′ since U is normal along c. Note also that ∇ ∂ c′ = 0 at t = a
∂s
and at t = b, since c(a, s) = γ(ks) (a geodesic) and c(b, s) = c(b) for all
s ∈ (−ε, ε). Hence Theorem 1 in #16 says:
1
L′′ (0) = I(U, U ).
kċk
Hence by what we proved above, L′′ (0) < 0. This means that after shrinking
ε if necessary, we have L(s) < L(0) for all s ∈ (−ε, ε) \ {0}. Done! 
(Remark: An alternative solution, which is perhaps a bit simpler and
even closer to the proof of Theorem 1 in #18, is to construct the vector field
U without insisting that U is normal – but with all the other properties
required above. Then we use Theorem 1 in #16 to deduce E ′′ (0) < 0
in place of L′′ (0) = 0. We also have E ′ (0) = 0; and then the argument
in the notes for #18 applies and lets us conclude what we want, i.e. that
L(s) < L(0) for all s 6= 0 sufficiently near 0.)
 PROBLEMS; “RIEMANNIAN GEOMETRY” 241

Problem 94: See Lee, [14, Lemma 11.6].

Problem 95: (We follow the proof of [14, Theorem 11.12].)
Let the constant sectional curvature of M be ρ ∈ R.
Let us first assume ρ ≤ 0. If ρ = 0 then set H := Rn ; if ρ < 0 then
set H = H n (|ρ|), i.e. hyperbolic n-space scaled to have constant curvature
ρ.56 Fix any point p ∈ M . By the Cartan-Hadamard Theorem (=Theorem
2 in Lecture #19), expp : Tp M → M is a surjective diffeomorphism. Hence
if we fix any identification of Tp M with Rn respecting the inner product,
then the C ∞ chart (M, exp−1p ) give normal coordinates on all M with center
p. Similarly, for any fixed point q ∈ H, (H, exp−1q ) give normal coordinates
on all H with center q. (Here expq is the exponential map from Rn =
Tq H to H.) Now by Problem 78, since M and H have the same constant
curvature ρ everywhere, the two C ∞ functions Rn → Mn (R) which give the
Riemannian metric of M wrt. (M, exp−1 p ) and the Riemannian metric of H
wrt. (H, expq ), respectively, are equal; indeed this function Rn → Mn (R) is
−1

given explicitly in Problem 78. Hence the map expp ◦ exp−1 q is an isometry
of H onto M , and we are done!

Next assume ρ > 0. Let S ⊂ Rn+1 be the n-dimensional sphere with

radius r := ρ−1/2 about the origin, with its standard Riemannian metric.
S has constant sectional curvature ρ. For any q ∈ S we know that expq :
Tq S → S restricted to Bπr (0) ⊂ Tq S is a diffeomorphism onto S \ {−q};
hence the chart (S \ {−q}, exp−1
q ) give normal coordinates on S with center
q. Also for any p ∈ M it follows from Cor. 1 in #19 that expp restricted
to Bπr (0) ⊂ Tp M has everywhere non-singular differential, and hence is a
local diffeomorphism (by the Inverse Function Theorem). Hence the ball
Bπr (0) ⊂ Tp M can be endowed with a unique Riemannian metric M such
that expp : Bπr (0) → M is a local isometry (cf. Problem 18). Note that
Bπr (0) with the Riemannian metric M has constant sectional curvature = ρ,
since M has so. We fix identifications Tq S = Rn and Tp M = Rn respecting
the inner products. Note that then (Bπr (0), I) (where I is the identity
map) form normal coordinates on Bπr (0) wrt the metric M, since for any
unit vector v ∈ Rn the curve c : (−πr, πr) → Bπr (0), c(t) = tv, is a geodesic.
Hence the metric M is explicitly given in by the formula in Problem 78! On
the other hand the expression for the metric on S wrt (S \ {−q}, exp−1 q )
must be given by the same formula. Hence the map
Φ := expp ◦ exp−1
q

from S \ {−q} to M is a local isometry.

56We obtain H n (|ρ|) by replacing the Riemannian metric h·, ·i on the standard hyper-
bolic n-space, H n , by the Riemannian metric [·, ·] := |ρ|−1 h·, ·i. Cf. Problem 70. Note
that there’s a misprint in Jost, [12, p. 228 (line -1)]; his “ρ” should be “ρ−1 ”.
242 ANDREAS STRÖMBERGSSON

Next fix any q ′ ∈ S \ {q, −q} and set p′ := Φ(q ′ ) ∈ M . Then by repeating
the above discussion we obtain a local isometry
(230) e := expp′ ◦ exp−1
Φ ′ q

from S \ {−q ′ } to M with Φ(q e ′ ) = p′ . Note that this Φ

e depends on which
n n
identifications Tq′ S = R and Tp′ M = R we choose; more to the point
what matters is how Tq′ S gets identified with Tp′ M , since this is what is
needed to make unique sense of (230). Any identification respecting the
respective Riemannian scalar products on Tq′ S and Tp′ M is ok. Let us
choose the identification of Tq′ S and Tp′ M to be given by dΦq′ : Tq′ S →
Tp′ M ; this is indeed a linear bijection respecting the respective Riemannian
scalar products, since Φ is a local isometry. Having made this choice, our
map Φ e satisfies
e q′ = dΦq′ ,
dΦ
since (d expp′ )0 : Tp′ M = Tp′ (Tp′ M ) → Tp′ M is the identity map, and
similarly for (d expq′ )0 . Hence by the following lemma, we actually have
(231) e |S\{−q,−q′ } ≡ Φ|S\{−q,−q′ } .
Φ

Lemma 9. Let N and N e be Riemannian manifolds and let ϕ, ψ : N → N e

be local isometries. Suppose that for some point p ∈ N we have ϕ(p) = ψ(p)
and dϕp = dψp . Then ϕ ≡ ψ.

Proof. First assume that q ∈ N is such that there exists a geodesic from p to
q, say c(t) = expp (tv), t ∈ [0, T ], for some v ∈ Tp (M ). (Thus expp (T v) = q.)
Then since ϕ and ψ are local isometries, both ϕ ◦ c and ψ ◦ c are geodesics
in Ne . These two geodesics have ϕ ◦ c(0) = ϕ(p) = ψ(p) = ψ ◦ c(0) and
(ϕ ◦ c)˙(0) = (dϕp )(ċ(0)) = (dψp )(ċ(0)) = (ψ ◦ c)˙(0);
hence by uniqueness of geodesics, ϕ ◦ c(t) = ψ ◦ c(t) for all t ∈ [0, T ], and in
particular
ϕ(q) = ϕ ◦ c(T ) = ψ ◦ c(T ) = ψ(q).

It follows from the above that if U is any open subset of N such that
every point q ∈ U can be reached by a geodesic from p, then ϕ|U = ψ|U .
Now let q be an arbitrary point in N . By Problem 1 there is a curve
c : [0, 1] → M with c(0) = p, c(1) = q. Consider the set
F := {t ∈ [0, 1] : ϕ(c(t)) = ψ(c(t)) and dϕc(t) = dψc(t) }.
This is a closed subset of [0, 1], since ϕ, ψ, dϕ, dψ are all continuous. Also
0 ∈ F , since ϕ(p) = ψ(p) and dϕp = dψp by assumption. Let us prove that F
is also an open subset of [0, 1]. Thus take any t ∈ F . Then ϕ(c(t)) = ψ(c(t))
and dϕc(t) = dψc(t) . By Theorem 3 in #4 there exists an open neighborhood
U of c(t) in N such that every point in U can be reached by a geodesic from
 PROBLEMS; “RIEMANNIAN GEOMETRY” 243

c(t). Now by the above argument applied to the point c(t) in place of p,
ϕ|U = ψ|U , and hence ϕ(q) = ψ(q) and dϕq = dψq for all q ∈ U . Hence F
contains the set {t′ ∈ [0, 1] : c(t′ ) ∈ U }, and this is an open neighborhood
of t in [0, 1]. Since every point t ∈ F has such an open neighborhood, it
follows that F is open in [0, 1]. Hence F is connected, being both open and
closed. This together with 0 ∈ F implies F = [0, 1]. In particular 1 ∈ F ,
and thus ϕ(q) = ψ(q). Since q was arbitrary, this completes the proof that
ϕ ≡ ψ. 

Continuing with our proof of the Killing-Hopf Theorem, we have now

proved (231) and this means that Φ and Φ e together define a C ∞ map Ψ
from the whole of S to M , and this map Ψ is a local isometry since Φ and
e are.
Φ
Lemma 10. Suppose that f is a local diffeomorphism from a C ∞ manifold
N1 to a C ∞ manifold N2 . Assume that N1 is compact. Then f is a covering
map.

Proof. Since N1 is compact and f is continuous, f (N1 ) is a compact subset

of N2 . In particular f (N1 ) is a closed subset of N2 . But f (N1 ) is also an
open subset of N2 since f , being a local diffeomorphism, is an open map.
Hence f (N1 ) (which is of course non-empty) is a connected component of
N2 , i.e. f (N1 ) = N2 . Hence we have proved that f is surjective (and also
that N2 is compact).
Now let p be an arbitrary point in N2 . Then f −1 (p) is a closed subset
of N1 , and hence compact, since N1 is compact. Furthermore f −1 (p) is a
discrete subset of N1 , since f is a local diffeomorphism. (Indeed, for any
q ∈ f −1 (p) there is an open set U ⊂ N1 with q ∈ U such that f|U is a
−1
diffeomorphism; then U ∩ f −1 (p) = f|U (p) = {q}, and this says that {q} is
−1 −1
an open subset of f (p) when f (p) is endowed with the relative topology
as a subset of N1 . Hence f −1 (p) with this topology is discrete.) Hence
f −1 (p), being both compact and discrete, is finite. Note that f −1 (p) 6= ∅
since f is surjective. Let us write f −1 (p) = {q1 , . . . , qm } (with q1 , . . . , qm
pairwise distinct).
Now since f is a local diffeomorphism, there exist open subsets U1 , . . . , Um ⊂
N1 such that qj ∈ Uj and f|Uj is a diffeomorphism onto an open subset
of N2 , for each j. Also, since N1 is Hausdorff, there exist open subsets
′ ⊂ N such that q ∈ U ′ for all j and U ′ ∩ U ′ = ∅ for all i 6= j.
U1′ , . . . , Um 1 j j i j
Set Uj := Uj ∩ Uj′ for j = 1, . . . , m. Then for each j, Uj′′ is open, qj ∈ Uj′′
′′

and f|Uj′′ is a diffeomorphism onto an open subset of N2 , and furthermore

′′ are pairwise disjoint. Set V := ∩m f (U ′′ ). This is an
the sets U1′′ , . . . , Um j=1 j
open subset of N2 containing p. Let V = V0 ⊃ V1 ⊃ V2 ⊃ · · · be a sequence
of open subsets of V which form a neighborhood basis for the point p. (Viz.,
p ∈ Vk for all k, and for every open set V ′ containing p there is some k
244 ANDREAS STRÖMBERGSSON

such that Vk ⊂ V ′ . For example we can take V1 , V2 , . . . to be a decreasing

sequence of open balls around p with radius tending to zero, with respect
to any fixed chart containing p.) Assume that f −1 (Vk ) 6⊂ ∪m ′′
j=1 Uj for every
k ≥ 1. This means that for every k ≥ 1 there exists some point p′k ∈ N1 ,
/ ∪m
p′k ∈ ′′ ′
j=1 Uj , such that f (pk ) ∈ Vk . Since N1 is compact, after passing to a
subsequence we may assume that p′k tends to a limit point p′ ∈ N1 as k → ∞.
Then f (p′k ) → f (p′ ) in N2 , and f (p′k ) ∈ Vk , and by our choice of V1 , V2 , . . .
this implies that f (p′ ) = p. On the other hand p′ ∈ / {q1 , . . . , qm }, since for
each j we have qj ∈ Uj′′ with Uj′′ open, and p′k ∈ / Uj′′ . This is a contradiction
against f −1 (p) = {q1 , . . . , qm }. Hence we must have f −1 (Vk ) ⊂ ∪m ′′
j=1 Uj
for some k ≥ 1. For this k, f −1 (Vk ) equals the disjoint union of the sets
−1
Wj := Uj′′ ∩ f −1 (Vk ) = f|U ′′ (Vk ), j = 1, . . . , m, and f|Wj is a diffeomorphism
j
of Wj onto Vk (since f|Uj′′ is a diffeomorphism and Vk ⊂ f (Uj′′ )). The fact
that every point p ∈ N2 has such an open neighborhood Vk proves that f is
a covering map. 

The lemma applies to our situation, and gives that Ψ : S → M is a

covering map. Hence since we are assuming that M is simply connected, Ψ
must be a homeomorphism, hence an isometry of S onto M . 

References
1. W. M. Boothby, An introduction to differentiable manifolds and riemannian geometry,
Academic Press, 1986.
2. Jeff Cheeger and David G. Ebin, Comparison theorems in Riemannian geometry,
AMS Chelsea Publishing, Providence, RI, 2008, Revised reprint of the 1975 original.
MR 2394158
3. Lawrence Conlon, Differentiable manifolds, second ed., Birkhäuser Advanced Texts:
Basler Lehrbücher. [Birkhäuser Advanced Texts: Basel Textbooks], Birkhäuser
Boston, Inc., Boston, MA, 2001. MR 1821549
4. Anton Deitmar and Siegfried Echterhoff, Principles of harmonic analysis, second ed.,
Universitext, Springer, Cham, 2014. MR 3289059
5. Karl-Heinz Fieseler, Riemannian geometry, lecture notes; from his www, 2015.
6. Gerald B. Folland, Real analysis, second ed., Pure and Applied Mathematics (New
York), John Wiley & Sons Inc., New York, 1999, Modern techniques and their appli-
cations, A Wiley-Interscience Publication. MR 1681462 (2000c:00001)
7. Allen Hatcher, Algebraic topology, Cambridge University Press, Cambridge, 2002.
8. Sigurdur Helgason, Differential geometry, lie groups, and symmetric spaces, Academic
Press, 1978.
9. Witold Hurewicz and Henry Wallman, Dimension theory, Princeton Univ. Press,
Princeton, NJ, 1948.
10. L. Hörmander, The analysis of linear partial differential operators i, Springer-Verlag,
1990.
11. Jürgen Jost, Riemannian geometry and geometric analysis, fourth ed., Universitext,
Springer-Verlag, Berlin, 2005.
12. Jürgen Jost, Riemannian geometry and geometric analysis, sixth ed., Universitext,
Springer, Heidelberg, 2011. MR 2829653
 PROBLEMS; “RIEMANNIAN GEOMETRY” 245

13. Shoshichi Kobayashi and Katsumi Nomizu, Foundations of differential geometry. Vol.
I, Wiley Classics Library, John Wiley & Sons, Inc., New York, 1996, Reprint of the
1963 original, A Wiley-Interscience Publication. MR 1393940
14. John M. Lee, Riemannian manifolds, Graduate Texts in Mathematics, vol. 176,
Springer-Verlag, New York, 1997, An introduction to curvature. MR 1468735
15. , Introduction to smooth manifolds, second ed., Graduate Texts in Mathemat-
ics, vol. 218, Springer, New York, 2013. MR 2954043
16. Walter A. Poor, Differential geometric structures, McGraw-Hill Book Co., New York,
1981. MR 647949