Solutions To Lee's Smooth Manifolds

SOLUTIONS TO LEE’S INTRODUCTION TO SMOOTH MANIFOLDS
SFEESH
1. Topological Manifolds
Exercise 1.1. Show that equivalent definitions of manifolds are obtained if instead of al-
lowing U to be homeomorphic to any open subset of Rn , we require it to be homeomorphic
to an open ball in Rn , or to Rn itself.
Every open ball is homeomorphic to the the open unit ball centred at the origin (we can
map any two balls onto each other by linear scaling and a translation, both of which are
continuous with continuous inverses). Note that open unit ball is homeomorphic to Rn itself
via the homeomorphism
x
f : int(Bn ) → Rn , x 7→ .
1 − kxk
Hence, it is enough to show that we obtain an equivalent definition of a topological
manifold if we require that U be homeomorphic to an open ball. First, suppose that at
every p ∈ M , there is an open neighbourhood U such that U b ∈ Rn is open and φ : U → U
b
−1
is open. Then φ(p) ∈ B ⊆ U , where B is some open ball. Then φ (B) is an open subset
b
of M containing p that is homeomorphic to an open ball in Rn . Conversely, if every point
p ∈ M is contained in a neighbourhood homeomorphic to an open ball in Rn , then M
already satisfies the given definition.
Exercise 1.6. Show that RPn is Hausdorff and second-countable, and is therefore a topo-
logical n-manifold.
Let [x] and [y] be distinct points of RPn , that is, distinct 1-dimensional subspaces of Rn+1 ,
and let them be spanned by the unit vectors x and y respectively. Since Sn is Hausdorff,
it is not hard to see that we can find pairwise disjoint open sets U, U , V, V ⊆ Sn such that
x ∈ U, −x ∈ U e , y ∈ V, −y ∈ Ve , where A = −a : a ∈ A. Let U b = U ∪U e and Vb = V ∪ Ve .
Define
p
φ : Rn r {0} → Sn , p 7→ .
kpk
We claim that U = π(φ−1 (U b )) and V = π(φ−1 (Vb )) are disjoint and contain [x] and [y]
respectively, and that U and V are open. It is clear that [x] ∈ U and [y] ∈ V . Let
[p] ∈ U ∩ V . Then [p] = π(u) = π(v) for some u ∈ φ−1 (U b ), v ∈ φ−1 (Vb ). Then u = λv
for some λ ∈ R r {0}. Hence, φ(u) = ±φ(v). But this implies that φ(u) ∈ U b ∩ Vb , a
contradiction. We conclude that RPn is Hausdorff.
Let B be a countable basis for Rn . We claim that π(B) = {π(B) : B ∈ B} is a basis
for RPn . Define ft : Rn r {0} → Rn r {0}, p 7→ tp for every t ∈ R r {0}. Note that ft is
continuous and has a continuous inverse f1/t . Hence, for if U is open, then ft (U ) is also
open. We claim that π −1 (π(U )) = t∈Rr{0} ft (U ) for every open set U ∈ Rn r {0}. Let
S
Date: February 23, 2023.

1
2 SOLUTIONS TO LEE’S INTRODUCTION TO SMOOTH MANIFOLDS
p ∈ π −1 (π(U )). Then π(p) ∈ π(U ), implying that there is a u ∈ U such that uSspans the same
vector space as p. Hence, p = λu for some non-zero λ, and therefore p ∈ t∈Rr{0} ft (U ).
S
Conversely, suppose that p ∈ t∈Rr{0} ft (U ). Then p = fλ (u) = λu for some u ∈ U and
some non-zero λ. Hence, π(p) = π(λu) = π(u), so p ∈ π −1 (π(U )), proving the claim. We
are now ready to prove second-countability, equivalently to prove that π(B) is a basis. Let
[p] ∈ π(B1 ) ∩ π(B2 ) for two basis sets B1 , B2 ∈ B. Then p ∈ π −1 (π(B1 )) ∩ π −1 (π(B2 )),
which is open by our previous work. Since this set is nonempty, there is a basis set B3
contained in π −1 (π(B1 )) ∩ π −1 (π(B2 )). Then π(B3 ) ⊆ π(B1 ) ∩ π(B2 ), showing that π(B) is
a basis, and therefore showing that RPn is a manifold.
Exercise 1.7. Show that RPn is compact. [Hint: show that the restriction of π to Sn is
surjective.]
Let [p] = π(p) ∈ RPn . Then π(p/kpk) = π(p), and p/kpk ∈ Sn . Then the restriction of π
to Sn is surjective. Since Sn is compact, and the image of a compact set under a continuous
function is compact, RPn is compact.
Exercise 1.14. Suppose X is a locally finite collection of subsets of a topological space M .
(a) The collection {X : X ∈ X } is also locally finite.
S S
(b) X∈X X = X∈X X.
S
S Let’s prove (b) first. Let x ∈ X∈X X. Then every open set U containing x intersects
X∈X X. By localS finiteness, choose an open set U containing x intersecting X1 , . . . , Xn ∈
X . Suppose x ∈ / X∈X X. Then, for each i ∈ {1, . . . , n}, there is an open set Ui that
contains x butSdoes not intersect Xi . Then U ∩ U1 ∩ · · · ∩ Un contains x, is open,Sbut does
not intersect X∈X X, which is a contradiction. Conversely, suppose that x ∈ X∈X X.
Then x ∈ X S for some X ∈ X ; thus, every S open set containing x intersects X, and therefore
intersects X∈X X as well. Hence, x ∈ X∈X X.
For (a), let x ∈ M , and let U be an open set containing x and intersecting X1 , . . . , Xn ∈ X .
Let Xe = X r {X1 , . . . , Xn }. Since U contains x but does not intersect any of the sets in
S S S
Xe, we have x ∈ / X∈Xe X. Hence, U r X∈Xe X = U r X∈Xe X is an open set containing
x and intersecting only sets in {X1 , . . . , Xn }. Therefore, {X : X ∈ X } is locally finite.
Exercise 1.18. Let M be a topological manifold. Two smooth atlases for M determine the
same smooth structure if and only if their union is a smooth atlas.
(⇒) Let A1 and A2 be two atlases for M determining the same smooth structure. By
Proposition 1.17 (a), these two atlases are contained in a unique maximal smooth atlas A.
Since any two charts in A are smoothly compatible, then any two charts in A1 ∪ A2 are
smoothly compatible. Since each Ai covers M , A1 ∪ A2 covers M and is therefore a smooth
atlas.
(⇐) Let A1 and A2 be smooth atlases and let A1 ∪A2 be a smooth atlas. By part (a), A1
and A2 are each contained in a maximal smooth atlas. Since A1 and A2 are each contained
in A1 ∪ A2 , they are contained in the same maximal atlas, and therefore they determine the
same smooth structure on M .
Exercise 1.20. Every smooth manifold has a countable basis of regular coordinate balls.
We’re not going to worry about our coordinate balls being centred at 0, since a ball in Rn
can always be mapped to a ball centred at 0 in Rn via a translation. Let M be a topological
n-manifold, and let {Ui , φi } be an atlas of charts for M . We may assume that the atlas is
SOLUTIONS TO LEE’S INTRODUCTION TO SMOOTH MANIFOLDS 3
countable, since M is second-countable. Note that φi (Ui ) ⊆ Rn has a countable basis of

regular coordinate balls, namely those of the for Br (x), such that x ∈ Qn , r ∈ Q, and such
that Br (x) ⊆ Br0 (x) ⊆ φi (Ui ) for some r0 > r. We claim that B = {φ−1 i (Br (x))} forms
a basis of regular coordinate balls for the topology on M . Let B1 , B2 ∈ B intersect non-
trivially. Then B1 ∩ B2 ⊆ Ui for some i, and φi (B1 ∩ B2 ) ⊆ φi (Ui ) ⊆ Rn . There is a rational
coordinate ball B whose closure is contained in another coordinate ball with the same centre.
Hence, φ−1 i (B) = B1 ∩ B2 . Hence, B is a basis. It remains to show that the elements of B
are regular. Let φ−1 −1
i (Br (x)) ∈ B. Then φi (φi (Br (x))) = Br (x) ⊆ Br 0 (x) ⊆ φi (Ui ). Hence,
−1 −1 −1
φi (Br (x)) ⊆ φi (Br0 (x), so φi (Br (x)) is a regular coordinate ball.
Exercise 1.39. Let M be a topological n-manifold with boundary.
(a) IntM is an open subset of M and a topological n-manifold without boundary.
(b) ∂M is a closed subset of M and a topological (n − 1)-manifold without boundary.
(c) M is a topological manifold if and only if ∂M = ∅.
(d) If n = 0, then ∂M = ∅ and M is a 0-manifold.
For (a), let x ∈ IntM . Then x ∈ U , where (U, φ) is an interior chart. We claim that
U ⊆ IntM . Let y ∈ U . Then φ(y) ∈ φ(U ) ⊆ Rn . Since φ(U ) is open, there is an open set
Vb such that φ(y) ∈ Vb ⊆ φ(U ). Then y ∈ φ−1 (Vb ) ⊆ U . Therefore, (φ−1 (Vb ), φ|φ−1 (Vb ) ) is
an interior chart containing y, so y ∈ IntM . Hence, U ⊆ IntM , which proves that IntM is
open. Since IntM is a subspace of M , which is Hausdorff and second-countable, we have that
IntM is Hausdorff and second-countable. By definition, every point of IntM is contained in
an interior chart, so IntM is a topological n-manifold.
For (b), by the theorem on topological invariance of the boundary, ∂M = M r IntM .
By part (a), ∂M is closed. ∂M is Hausdorff and second-countable since it is a subspace
of M , which is Hausdorff and second-countable. It remains to show that every x ∈ ∂M is
in some chart (U, φ) with φ(U ) ⊆ Rn−1 . Let x ∈ ∂M , and let (V, ψ) be a boundary chart
containing x. Then ψ(x) ∈ ∂Hn . Let U b = ψ(V ) ∩ ∂Hn . Note that U b contains x, and is
n ∼ n−1 −1 b
open in ∂H = R . Let U = ψ (U ) = V ∩ ∂M . Then U is open in ∂M , and (U, ψ|U ) is
a chart for x ∈ ∂M . This proves that ∂M is an (n − 1)-manifold.
For (c), let M be a topological manifold. By definition, every point is in an interior
chart. By the theorem on topological invariance of the boundary, ∂M = ∅. Conversely, let
∂M = ∅. Again, by the theorem on the topological invariance of the boundary, every point
is then an interior point. M then satisfies the definition of a topological manifold.
For (d), let M be a 0-manifold. Then every point p ∈ M is in a chart ({p}, φ), where
φ : {p} → R0 is a homeomorphism. Hence, every point of M is an interior point, so ∂M = ∅.
By part (c), M is a 0-manifold.
Exercise 1.41. Let M be a topological manifold with boundary.
(a) M has a countable basis of precompact coordinate balls and half-balls.
(b) M is locally compact.
(c) M is paracompact.
(d) M is locally path-connected.
(e) M has countably many components, each of which is an open subset of M and a
connected topological manifold with boundary.
(f ) The fundamental group of M is countable.
For (a), let {(Ui , φi )} be a collection of charts covering M . We may assume that this
collection is countable, since M has a countable basis by definition. Say (U, φ) is an interior
chart. Then φ(U ) ⊆ Rn has a countable basis of balls whose closures lie in φ(U ). Similarly,
if (U, φ) is a boundary chart, then φ(U ) has a countable basis of balls and half-balls whose
closures all lie in φ(U ). Let B be the collection of preimages of all the coordinate balls and
half-balls discussed above, under the respective coordinate maps. Then B is a countable
basis for M . It remains to show that the sets in B are precompact. Let B ∈ B, and let φ be
its coordinate map. Then B ⊆ U , where (U, φ) is a coordinate chart (the φ in this chart is
an extension of the φ in (B, φ)). Then φ(B) = φ(B) ⊆ φ(U ) is compact, so B is compact,
where B is the closure of B in U . Since M is Hausdorff, B is closed in M . It follows that
B is also the closure of B in M . This proves (a).
For (b), part (a) implies that every x ∈ M is contained in a precompact coordinate ball.
Hence, M is locally compact.
For (c), let X be an open cover of M , and let B be an arbitrary basis for the topology on
M . We will prove the stronger result that X has a countable, open refinement consisting
S∞ of
elements of B. Let {Kj }∞j=1 be an exhaustion of M by compact sets; that is X = j=1 Kj
and Kj ⊆ IntKj+1 . Let Aj = Kj+1 r IntKj and let Bj = IntKj+2 r Kj−1 . Then Aj
is compact and Bj is open, and Aj ⊆ Bj . Every x ∈ Aj is contained in some basis set
B ⊆ X ∩ Bj . Since these basis sets cover the compact set Aj , there is a finite cover of
Aj . Taking the union of these basis sets over j ∈ N, we obtain a countable refinement of X
consisting of sets in B. The cover is locally finite, since the basis sets are all contained in
some Bj , Bj and Bj 0 don’t intersect if |j − j 0 | > 2.
For (d), by part (a) M has a basis of coordinate balls and half-balls, which are all path-
connected. Hence, M is locally path-connected.
For (e), let {Ci } be the collection of connected components of M . Then {Ci } is an open
cover of M . Since M is second-countable, {Ci } has a countable subcover. But since the Ci
are disjoint, we must have that {Ci } is countable to begin with. Hence, M has countably
many components. By definition, each component is an open subset. As subspaces of M ,
the Ci are Hausdorff and second-countable. Let x ∈ Ci and let (U, φ) be a chart containing
x. Then (U ∩ Ci , φ|U ∩Ci ) is a chart in Ci containing x. Thus, Ci is a connected topological
manifold with (possibly empty) boundary.
For (f), let B be a countable basis of coordinate balls and half-balls. For any B1 , B2 ∈ B,
the intersection B1 ∩ B2 has countably many components. To see this, note that B1 ∩ B2
is open in M , and therefore its components are open in M . Moreover, every component is
itself a connected manifold with boundary, and therefore is locally path-connected. Since
connected and locally path-connected spaces are path-connected, every component is path-
connected. Since each component contains a distinct basis set, B1 ∩ B2 has countably many
components. Let X be a set containing a single point from each component of B1 ∩ B2 for
every pair B1 , B2 ∈ B. For every pair x, x0 ∈ X contained in some B ∈ B, let hB x,x0 be a
fixed path from x to x0 in B. Select a point p ∈ X as the basepoint of the fundamental
group π1 (M, p). Define a special loop to be a loop starting at p that is the concatenation of
paths of the form hB x,x0 . Since the set of all special loops is countable, it suffices to show that
every loop based at p is homotopic to some special loop. Let f : [0, 1] → M be an arbitrary
loop based at p. Since f ([0, 1]) is compact, there are finitely many basis sets B1 , . . . , Bk ∈ B
that cover it. Hence, there exist 0 = a0 < a1 < · · · < ak = 1 such that f ([ai−1 , ai ]) ⊆ Bi for
each i. Each f (ai ) is contained in the same component of Bi ∩ Bi+1 as some xi ∈ X . Let gi
be a path from xi to f (ai ), where g0 and gk are both equal to the constant path at p. Let
fi be the path obtained by restricting f to [ai−1 , ai ],

f ' f1 · f2 · · · · · fk ' g0 · f1 · ·g1 · g1 · f2 · g2 · · · · · gk−1 · fk · gk .
Since each Bi is simply connected, and the endpoints of gi−1 ·fi ·gi are xi−1 and xi , gi−1 ·fi ·gi
is homotopic to hB xi−1 ,xi . Hence, f is homotopic to a special loop, concluding the proof.
i
Exercise 1.42. Show that every smooth manifold with boundary has a countable basis con-
sisting of regular coordinate balls and half-balls
This is a straightforward adaptation of Exercise 1.20.
Exercise 1.43. Show that the smooth manifold chart lemma (Lemma 1.35) holds with “
Rn ” replaced with “ Rn or Hn ” and “smooth manifold” replaced by “smooth manifold with
boundary.”
Just make the same replacements in the proof of Lemma 1.35. The only external result
that is used is Exercise A.22, which is a result for general topological spaces.
Exercise 1.44. Suppose M is a smooth n-manifold with boundary, and U is an open subset
of M . Prove the following statements:
(a) U is a topological n-manifold with boundary, and the atlas consisting of all smooth
charts (V, φ) for M such that V ⊆ U defines a smooth structure on U . With this
topology and smooth structure, U is called an open submanifold with boundary.
(b) If U ⊆ IntM , then U is actually a smooth manifold (without boundary); in this case,
we call it an open submanifold of M .
(c) IntM is an open submanifold of M (without boundary).
For (a), every point p ∈ U is contained in the domain of some chart (V, φ). Then
(U ∩ V, φ|U ∩V ) is a chart whose domain is contains p and is contained in U . Thus, U is a
topological n-manifold with boundary. These charts clearly cover U , so the form an atlas.
They are smoothly compatible, since this atlas is a subset of the maximal smooth atlas for
M . Hence, this collection of charts defines a smooth structure on U
For (b), we need to show that all of the charts are in fact interior charts. Let (V, φ) be
a smooth chart with V ⊆ U . If φ(V ) ⊆ Hn with some p ∈ V such that φ(p) ∈ ∂Hn . Then
p is a boundary point and an interior point, which is impossible by the theorem on the
topological invariance of the boundary.
For (c), by exercise 1.39 IntM is open in M . Take U = IntM in part (b).
Problem 1.1. Let X be the set of all points (x, y) ∈ R2 such that y = ±1, and let M be the
quotient of X by the equivalence relation generated by (x, −1) ∼ (x, 1) for all x 6= 0. Show
that M is locally Euclidean and second-countable, but not Hausdorff. (This space is called
the line with two origins.)
Let π : X → M, x 7→ [x]. Let [(x, 1)] ∈ M for some x (we are including the possibility
that x = 0, the [(0, −1)] case is similar). Then there is an open set (a, b) × {1} in X
such that x ∈ (a, b). We claim that the map φ : π((a, b) × {1}) → (a, b), [(y, ±1)] 7→ y
is a homeomorphism. If [(y, ±1)] = [(y 0 , ±1)], then y = y 0 , so φ is well-defined. Let
φ([(y, ±1)]) = φ([(y 0 , ±1)]) ⇒ y = y 0 . If y = y 0 = 0, then [(y, ±1)] = [(y 0 , ±1)] = [(0, 1)],
since the points [(y, ±1)] and [(y 0 , ±1)] must be in π((a, b) × {1}). Otherwise, we can ignore
the ±1 since it does not affect the equivalence class, so [(y, ±1)] = [(y 0 , ±1)]. Thus, φ is
injective. It is also clearly surjective, and thus bijective. Let U ⊆ (a, b) be open. Then
φ−1 (U ) = π(U × {1}). Since π −1 (π(U × {1})) = U × {±1} is open in X, we have that
π(U × {1}) is open in M . Let U ⊆ π((a, b) × {1}) be open in M . Then U × {±1} is open
in X, and therefore U is open in (a, b) ⊆ R. Thus, π is a homeomorphism and M is locally
Euclidean.
Let U ∈ X be open. Then π −1 (π(U )) = U ∪ r(U ), where r is the reflection across the
x-axis. Since U ∪ r(U ) is open, π −1 (π(U )) is open. Hence, π is an open map. By the same
argument given in Exercise 1.6, π(B) is a basis for M if B is a basis for X. Since X is a
subspace of a second-countable space, M is second-countable.
Let U and V be open sets in M containing [(0, 1)] and [(0, −1)] respectively. Hence,
π −1 (U ) and π −1 (V ) are open sets in X containing (0, 1) and (0, −1) respectively. Hence, U
and V contain subsets (−, ) × {1} and (−, ) × {−1}, respectively, for some > 0. Then
π((−, ) × {1}) ⊆ U and π((−, ) × {−1}) ⊆ V . However, these two sets have non-trivial
intersection. We conclude that [(0, 1)] and [(0, −1)] cannot be separated by open sets, and
therefore M is not Hausdorff.
Problem 1.2. Show that a disjoint union of uncountably many copies of R is locally Eu-
clidean and Hausdorff, but not second-countable.
Let M be a disjoint union of uncountably many copies of R. Let x ∈ M . Let R be
the component containing x. Then the homeomorphism φ : R → R defines a chart whose
domain contains x. Therefore, M is locally Euclidean. The disjoint union of Hausdorff
spaces is Hausdorff, so M is Hasudorff. Since manifolds have countably many components,
M cannot be a manifold. We conclude that M is not second-countable since it satisfies the
other conditions of the definition.
Problem 1.3. A topological space is said to be σ-compact if it can be expressed as a union
of countably many compact spaces. Show that a locally Euclidean Hausdorff space is a
topological manifold if and only if it is σ-compact.
(⇒) Let M be a topological manifold. Then M is second countable. By Lemma 1.10, M
has a countable basis of precompact coordinate balls. Then M is the union of the closures
of these basis sets, which are compact.
(⇐) Let M be locally Euclidean, Hausdorff, and σ-compact. Let {Ki }i∈N be a countable
cover of M by compact sets. Since M is locally Euclidean, M has an open cover of coordinate
balls. Each Ki is then covered by finitely many such coordinate balls, so we can assume that
M is covered by countably many coordinate balls. Let (B, φB ) be such a coordinate ball.
Then φ(B) ⊆ Rn has a countable basis BB . Let φ−1 B (BB ) be collection of preimages of sets
in BB under φB . Then B φ−1
S
B (BB ) is a basis for M . Therefore, M is second-countable.
Problem 1.4. Let M be a topological manifold, and let U be an open cover of M .
(a) Assuming that each set in U intersects only finitely many others, show that U is
locally finite.
(b) Give an example to show that the converse to (a) may be false.
(c) Now that the sets in U are precompact in M , and prove the converse: if U is locally
finite, then each set in U intersects only finitely many others.
For (a), let p ∈ M be arbitrary. Then p ∈ U for some U ∈ U, and U is a neighbourhood
of p that intersects at most finitely many of the other sets in U.
For (b), let M = R, and let U contain R, and the intervals (i, i + 1) for each i ∈ Z. Then
U is locally finite since every x ∈ R is contained in some open interval of unit length, and
every such open interval intersects at most three of the sets in U.
For (c), let U ∈ U. For every x ∈ U , let Ux be an open set containing x such that Ux
intersects finitely many sets in U. Then {Ux }x∈X is an open cover for U , which is compact.
(1) (k) (i)
Hence, it has a finite subcover {Ux , . . . , Ux }. Since each Ux intersects finitely many sets
in U and they cover U , we conclude that U intersects finitely many sets in U.
Problem 1.5. Suppose M is a locally Euclidean Hausdorff space. Show that M is second-
countable if and only if it is paracompact and has countably many connected components.
(⇒) If M is second-countable, it is a topological manifold. Hence, it is countably many
components and is paracompact by the results from the section.
(⇐) Let M be paracompact and have countably many components. We can assume that
M is connected: if each connected component has a countable basis, and there are countably
many components, then M has a countable basis.
Let U be an open cover of M by coordinate charts and let (U, φ) be such a chart. Note
that φ(U ) has a basis of precompact balls, where the closures of said balls lie in φ(U ).
Pulling back these balls to M yields a covering of M by balls whose closure in U is compact,
and thus closed because M is Hausdorff. Thus, the closure of a coordinate ball in U is the
same as its closure in M , and therefore we have an open cover of precompact coordinate
charts for M . Since M is paracompact, the open cover constructed has a locally finite
open refinement. Since M is Hausdorff, the this locally finite refinement also consists of
precompact coordinate charts. Call this open cover U.
Let U ∈ U be arbitrary, and define V0 = {U }. By part (c) of the preceding problem,
U intersects finitely many sets in U. Denote the collection of these sets W1 , and define
V1 = W1 r V0 . Suppose we have define Vn ⊆ U. Let Wn+1 be the collection of all sets in U
that intersect sets in V0 ∪· · ·∪Vn , and define Vn+1 = Wn+1 S∞r(V0 ∪· · ·∪Vn ). By construction,
each Vi contains finitely many sets. We will show that i=0 Vi covers M . Since M is locally
Euclidean, it is locally path-connected, and because we are assuming that M is connected,
it is path-connected. Let y ∈ M and x ∈ U be arbitrary, and let f : [0, 1] → M be a path
from x to y. Since f ([0, 1]) is compact, it is covered by finitely many of the sets in U, and
therefore there are numbers 0 = a0 < a1 < · · · < ak = 1 such that f ([ai−1 , ai ]) ⊆ Ui for
some Ui ∈ U. Then f ([a0 , a1 ]) = f ([0, a1 ]) ⊆ U1 = U . Suppose that we have shown that
f ([aj−1 , aj ]) is contained in the union of the sets in V0 ∪ · · · ∪ Vj−1 . Then f (aj ) ∈ V for
some V ∈ V0 ∪ · · · ∪ Vj−1 . Hence, f ([aj , aj+1 ]) must be contained in the union of sets in
V0 ∪ · · · ∪ Vj , since it intersects the sets in V0 ∪ · · · ∪ Vj−1 . Therefore, y is contained in a set
in some Vk , which proves the claim. Hence, we have a countable cover of M by precompact
coordinate charts. Each of these charts has a countable basis, and the union of all these
bases is a basis for the topology on M . Hence, M is second countable.
Problem 1.6. Let M be a nonempty topological manifold of dimension n > 1. If M has a
smooth structure, show that it has uncountably many distinct ones.
s−1
Following the hint, for every s > 0 we show that Fs : Bn → Bn , x 7→ |x| x is a homeo-
morphism, and a diffeomorphism if and only if s = 1. Note that Fs is not defined at x = 0
s
for s 6 1. However, |Fs (x)| = |x| , so defining Fs (0) = 0 makes Fs a continuous function.
s
Moreover, |Fs (x)| = |x| 6 1, so the image of Fs does indeed lie in Bn . The inverse of Fs is
1
−1
given by Fs−1 (x) = |x| s x, where we define Fs−1 (0) = 0. Clearly, Fs−1 is continuous, so Fs
is a homeomorphism for each s > 0.
Away from x = 0, Fs is a diffeomorphism since Fs and Fs−1 are products or quotients
of smooth non-vanishing functions. Moreover, F1 = id is a diffeomorphism. Now suppose
that s 6= 1. It suffices to show that Fs is not smooth for s < 1, since the same proof will
show that Fs−1 is not smooth for s > 1. If Fs is smooth, then it should have continuous
derivatives of all orders. However, the partial derivative of the first component of Fs with
respect to the first variable is
∂ s−1 s−3 s−1
|x| x1 = (s − 1)|x| x21 + |x| .
∂x1
To see that this quantity is not continuous at 0, we can take its limit as x → 0 along the
x1 -axis. Setting x2 = · · · = xn = 0, we find
∂ s−1 s−3 s−1 s−1
|x| x1 = (s − 1)|x1 | x21 + |x1 | = s|x1 | → ∞,
∂x1
since s < 1. This concludes our proof of the statement in the hint.
Given that M has a smooth structure, let A be a locally finite atlas for M . Let x ∈ M and
let x be contained in the domains of the charts (U1 , φ1 ), . . . , (Uk , φk ). Define Let A0 = A,
and inductively define
(
Ai−1 r {(Ui , φi )} if Ui−1 is covered by the domains of other charts in Ai−1
Ai =
Ai−1 otherwise
for i = 1, . . . , k. By induction, Ak is an atlas for M . Hence, the charts in Ak cover M .
Since x ∈ M , there must be at least one i ∈ {1, . . . , k} such that (Ui , φi ) ∈ Ak . Hence,
Ui is not covered by the domains in of charts in Ai−1 , and therefore is not covered by
the domains of charts in Ak ⊆ Ai−1 . Thus, there is a p ∈ Ui that is covered only by Ui .
Moreover, φi (Ui ) r {φi (p)} can be covered by open balls Bα = Brα (xα ) entirely contained
in φi (Ui ) r {φi (p)}. There is also a ball B = Br (φi (p)) ⊆ φi (Ui ). Pulling back these balls,
we get that A0 = Ak r {(Ui , φi )}] ∪ {(φ−1 −1
i (Bi ), φi )} ∪ {(φi (B), φi )} is a smooth atlas for
−1
M such that p is only contained the domain of (φi (B), φi ). The coordinate maps are
smoothly compatible, since the new maps are obtained by restricting φ, which is smoothly
compatible with all of the other coordinate maps. There is a diffeomorphism f : B → Bn
obtained by translating and then scaling. Replacing (φ−1 −1
i (B), φi ) with (φi (B), f ◦ φi ) in
0
A , we obtain a new smooth atlas Z.
For every s > 0, define Zs to be the set of charts in Z, except (φ−1 i (B), f ◦ φi ) is
replaced with (φ−1 i (B), Fs ◦ f ◦ φ i ). We claim that each Z s is a smooth atlas, and that
Zs is not smoothly compatible with Zs0 for s 6= s0 . Clearly, every pair of charts not
including (φ−1 i (B), Fs ◦ f ◦ φi ) is smoothly compatible. Now consider the pair of charts
(φ−1
i (B), F ◦ −1
s f ◦ φi ) and (U, φ) and suppose that φi (B) ∩ U 6= ∅. The transition function
−1
is then Fs ◦ f ◦ φi ◦ φ−1 . Since φi (B) ∩ U does not contain p and Fs is a diffeomorphism
away from 0, we have that the transition function is a diffeomorphism. This shows that Zs
is a smooth atlas, and therefore determines a smooth structure for each s. For s 6= s0 , the
charts (φ−1 −1
i (B), Fs ◦ f ◦ φi ) and (φi (B), Fs0 ◦ f ◦ φi ) are not smoothly compatible, since
Fs ◦ f ◦ φi ◦ (Fs0 ◦ f ◦ φi )−1 = Fs ◦ f ◦ φi ◦ φ−1

i ◦f
−1
◦ Fs−1
0 = Fs ◦ Fs−1
0 = Fs/s0
(the last equality is a standard computation) is not a diffeomorphism. We conclude that Zs
and Zs0 are not smoothly compatible and therefore there are uncountably many different
smooth structures on M .
Problem 1.7. Let N denote the north pole (0, . . . , 0, 1) ∈ Sn ⊆ Rn+1 , and let S denote the
south pole (0, . . . , 0, −1). Define the stereographic projection σ : Sn r {N } → Rn by
(x1 , . . . , xn )
σ(x1 , . . . , xn+1 ) = .
1 − xn+1
Let σ̃(x) = −σ(−x) for x ∈ Sn r {S}.

(a) For any x ∈ Sn r {N }, show that σ(x) = u, where (u, 0) is the point where the line
through N and x intersects the linear subspace where xn+1 = 0. Similarly, show that
(σ̃(x), 0) is the point where the line through S and x intersects the same subspace.
(For this reason, σ̃ is called stereographic projection from the south pole.)
(b) Show that σ is bijective, and
2
(2u1 , . . . , 2un , |u| − 1)
σ −1 (u1 , . . . , un ) = 2 .
|u| + 1
(c) Compute the transition map σ̃ ◦ σ −1 and verify that the atlas consisting of the two
charts (Sn r {N }, σ) and (Sn r {S}, σ̃) defines a smooth structure on Sn . (The
coordinates defined by σ and σ̃ are called stereographic coordinates.)
(d) Show that this smooth structure is the same as the one defined in Example 1.31.
For (a), let x = (x1 , . . . , xn+1 ) ∈ Sn r {N }. The line through N and x is parametrized
by u1 = x1 t, . . . , un = xn t, un+1 = (xn+1 − 1)t + 1. The intersection of this line with the
un+1 = 0 occurs when t = 1/(1−xn+1 ). Hence, the intersection point is (σ(x), 0), as desired.
Similarly, if x = (x1 , . . . , xn+1 ) ∈ Sn r {S}, then the line through S and x is parametrized
by u1 = x1 t, . . . , un = xn t, un+1 = (xn+1 + 1)t − 1. Setting un+1 = 0, we find that the
intersection point is (σ̃(x), 0).
−1
For (b), it suffices to check that the given definition for σ n is indeed the inverse of σ.
−1
It is straightforward to check that σ (u) = 1 for all u ∈ R , and the last coordinate of

2
σ −1 (u) is |u| −1
|u|2 +1
< 1. Hence, σ −1 (Rn ) ⊆ Sn r {N }. Showing that σ ◦ σ −1 and σ −1 ◦ σ are
the identity maps is a straightforward calculation.
For (c), the transition map is
(u1 , . . . , un )
σ̃ ◦ σ −1 (u1 , . . . , un ) = 2 .
|u|
Note that the domain of σ̃ ◦ σ −1 is σ(Sn r {N, S}). Since σ(S) = 0, this shows that u = 0 is
not in the domain of σ̃ ◦ σ −1 . Hence, σ̃ ◦ σ −1 is smooth. A similar computation shows that
σ ◦ σ̃ −1 is smooth. Since Sn r {N } and Sn r {S} form an open cover of Sn , we conclude
that (Sn r {N }, σ) and (Sn r {S}, σ̃) define a smooth structure on Sn .
For (d), there are three “types” of charts from Example 1.31 to consider. The first is
+ − −
(Un+1 , φ+
n+1 ), which contains N . The second is (Un+1 , φn+1 ), which contains S. Finally,
± ±
the third type is (Ui , φi ) for i = 1, . . . , n, which contain neither N nor S. We will show
that each type of chart is smoothly compatible with σ; the computation showing that they
are smoothly compatible with σ̃ is similar. The first functions for the first and second type
of chart are
(2u1 , . . . , 2un )
φ±
n+1 ◦ σ
−1 1
(u , . . . , un ) = 2 ,
|u| + 1
which are both smooth. The inverses of these transition functions are
(u1 , . . . , un )
σ ◦ (φ±
n+1 )
−1 1
(u , . . . , un ) = q
2
1 ∓ 1 − |u|
Hence, σ ◦ (φ−
n+1 )
−1
is smooth. To see that σ ◦ (φ+
n+1 )
−1
is smooth, note that its domain is
+ +
(φn+1 )(Un+1 r {N }), which does not include 0. For the third type of chart, we have
ci , . . . , 2un , |u|2 − 1)
(2u1 , . . . , 2u
φ±
i ◦σ
−1 1
(u , . . . , un ) = 2 ,
|u| + 1
which is smooth, and
q
2
(u1 , . . . , ui−1 , 1 − |u| , ui , . . . , un−1 )
σ◦ (φ±
i )
−1 1
(u , . . . , un ) = ,
1 − un
which is also smooth, since un 6= 1 in Ui± . Hence, the smooth atlas from Example 1.31
is smoothly compatible with the smooth atlas described in this problem. Hence, they
determine the same smooth structures on Sn .
Problem 1.8. By identifying R2 with C, we can think of the unit circle S1 as a subset of the
complex plane. An angle function on a subset U ⊆ S1 is a continuous function θ : U → R
such that eiθ(z) = z for all z ∈ U . Show that there exists an an angle function θ on an open
subset U ⊆ S1 if and only if U 6= S1 . For any such angle function, show that (U, θ) is a
smooth coordinate chart for S1 with its standard smooth structure.
(⇒) Let θ be an angle function on a subset U ⊆ S1 . For a contradiction, suppose that
U = S1 . Since the image of a connected and compact space under a continuous function is
connected and compact, we have that θ(S1 ) = [a, b] ⊆ R for some a 6 b. We now show that
θ̃ : S1 → [a, b], z 7→ θ(p) is a homeomorphism; this will give us a contradiction, since [a, b] is
simply connected, but S1 is not. It suffices to show that θ̃ is bijective, since a continuous
bijection with compact domain has a continuous inverse. By definition, θ̃ is surjective. If
0
θ̃(z) = θ̃(z 0 ), then z = eiθ(z) = eiθ(z ) = z 0 , so θ̃ is injective, and therefore bijective.
(⇐) Let U ⊂ S1 be a proper open subset. Suppose that p ∈ S1 r U . We will construct an
angle function θ on S1 r {p}. The restriction of θ to U is therefore also an angle function.
Note that θ(z) = −i ln(z) is an angle function, where we must take a branch of the complex
logarithm along the ray through p and the origin.
Let θ : U → R be an angle function. We proved in that θ is injective in the ⇒ direction
above. Hence, by the theorem on invariance of domain, θ is an open map and therefore
a homeomorphism onto its image. This shows that (U, θ) is a chart for S 1 . Suppose that
−1
U = S1 r {N }. Since θ−1 (x) = eiθ(θ (x)) = eix corresponds to (cos x, sin x) ⊆ Rn . Let σ
be the stereographic projection from the north discussed in the previous problem. Then
cos x x π
σ ◦ θ−1 (x) = = tan + ,
1 − sin x 2 4
which is a diffeomorphism from any interval of the form ( π2 + 2πn, 5π 2 + 2πn) to R, where
n ∈ Z (proving the above identity is a fun trigonometry exercise). Now, let U ⊂ S1 be an
arbitrary proper open subset. By rotating S1 , we may assume that N ∈ / U . Then σ ◦ θ−1
is the restriction of the above formula on each connected component of U . Thus, (U, σ)
is smoothly compatible with the stereographic coordinate charts, and is thus a smooth
coordinate chart for S1 with its standard smooth structure.
Problem 1.9. Complex projective n-space, denoted by CPn , is the set of all 1-dimensional
complex-linear subspaces of Cn+1 , with the quotient topology inhereted from the natural
projection π : Cn+1 r {0} → CPn . Show that CPn is a compact 2n-dimensional topological
manifold, and show how to give it a smooth structure analogous to the one we constructed
for RPn .
ei ⊆ Cn+1 r {0} be the open set of points where where z i 6= 0, and let Ui = π(U
Let U ei ).
−1
It is not too hard to show that π (π(U ei )) = Uei . Hence, Ui is open. Moreover, since
the Ui cover C
e n+1
r {0} and π is surjective, the Ui form an open cover of CPn . Define
φi : Ui → Cn ∼
= R2n by
1
z i−1 z i+1 z n+1

1 n+1 z
φi [z , . . . , z ] = ,..., i , i ,..., i .
zi z z z
Since φi [z 1 , . . . , z n+1 ] is unchanged by multiplying (z 1 , . . . , z n+1 ) by a non-zero constant,
φi is well-defined. Moreover, φi is bijective, which is not too hard to see. We claim that
ei → Ui is a quotient map; to show this, we must show that π is an open map, which
π|Uei : U
was done in Exercise 1.6. Hence, by the characteristic property of quotient maps, φi is
continuous if and only if φi ◦ π|Uei is continuous. Since
1
z i−1 z i+1 z n+1

1 n+1 z
φi ◦ π|Uei (z , . . . , z ) = ,..., i , i ,..., i ,
zi z z z
φi is continuous. The inverse of φi is given by
φ−1 1 n 1
i (z , . . . , z ) = [z , . . . , z
i−1
, 1, z i , . . . , z n ].
Let fi : Cn → Cn+1 , (z 1 , . . . , z n ) 7→ (z 1 , . . . , z i−1 , 1, z i , . . . , z n ); note that fi is continuous.
Then φ−1 i = π ◦ fi is continuous. Thus φi is a homeomorphism, and we have shown that
CPn is locally Euclidean.
To prove that CPn is Hausdorff, we will need the following fact from topology: if X is
a compact Hausdorff space and G is a compact topological group acting continuously on
X, then the orbit space X/G is Hausdorff. Let X = S2n+1 ⊆ Cn+1 and G = S1 ⊆ C,
and let S1 × S2n+1 → S2n+1 : (λ, z) 7→ λz be the group action. Then the quotient space
S2n+1 /S1 is Hausdorff. We will now prove that S2n+1 /S1 is homeomorphic to CPn . We will
denote the equivalence classes in S2n+1 /S1 and CPn by [·]S2n+1 /S1 and [·]CPn , respectively.
We claim that the map i : S2n+1 /S1 → CPn , [z 1 , . . . , z n+1 ]S2n+1 /S1 7→ [z 1 , . . . , z n+1 ]CPn is
a homeomorphism. The fact that i is well-defined and bijective is easily verified. For
continuity, observe the following commutative diagram
i
S2n+1 C2n+1 r {0}
π π
i
S2n+1 /S1 CPn
where i is the inclusion map, and the maps denoted π are the corresponding quotient maps.
Commutativity and the continuity of π ◦ i imply that i ◦ π is continuous. The universal
property of quotient maps implies that i is continuous. Since the inverse of a continuous
function with compact domain is continuous, we have that i is a homeomorphism. It follows
that CPn ∼ = S2n+1 /S1 , and is therefore Hausdorff.
The proof that CPn is second-countable is a straightforward adaptation of the proof that
RPn is second-countable given in Exercise 1.6.
We conclude the problem by showing that charts constructed above are smoothly com-
patible, and therefore define a smooth structure on CPn . Let (Ui , φi ) and (Uj , φj ) be two
charts, with i > j. The transition function is
1
z j−1 z j+1 z i−1 1 z i zn

z
φj ◦ φ−1
i (z 1
, . . . , z n
) = , . . . , , , . . . , , , , . . . , .
zj zj zj zj zj zj zj
Since z j 6= 0 for (z1 , . . . , z n ) ∈ φi (Ui ∩ Uj ), the above map is well defined, and therefore a
diffeomorphism from φi (Ui ∩ Uj ) to φj (Ui ∩ Uj ).
Problem 1.10. Let k and n be integers satisfying 0 < k < n, and let P, Q ⊆ Rn be the
linear subspaces spanned by (e1 , . . . , ek ) and (ek+1 , . . . , en ), respectively, where ei is the ith
standard basis vector for Rn . For any k-dimensional subspace S ⊆ Rn that has trivial
intersection with Q, show that the coordinate representation φ(S) constructed in Example
1.36
Ik
is the unique (n − k) × k matrix B such that S is spanned by the columns of the matrix
B , where Ik denotes the k × k identity matrix.
As described in Example 1.36, if πP : Rn → P and πQ : Rn → Q are the projections

onto P and Q, then the map πP |S : S → P is an isomorphism. Hence, there are vectors
b1 , . . . , bk ∈ Q such that ei + qi ∈ S for each i ∈ {1, . . . , k}. Since {e1 + b1 , . . . , ek + bk }
gets mapped to the basis {e1 , . . . , ek } for P , it must be a basis for S. Arranging the bi into
columns, we have a (n − k) × k matrix B such that S is spanned by the columns of IBk .

Now let’s show that this B is unique. Suppose B 0 is a (n − k) × k matrix such that the
Ik
columns of B 0 form a basis for S. Since πP |S (ei + b0i ) = ei and πP |S is an isomorphism,
we must have b0i = bi for each i. Hence, B = B 0 , so B is unique.
Since (πQ |S ) ◦ (πP |S )−1 (ei ) = (πQ |S )(ei + bi ) = bi , the matrix representation of (πQ |S ) ◦
(πP |S )−1 is B. Hence, φ(S) = B.
Problem 1.11. Let M = Bn , the closed unit ball in Rn . Show that M is a topological
manifold with boundary in which each point in Sn−1 is a boundary point and each point in
Bn is an interior point. Show how to give it a smooth structure such that every smooth
interior chart is a smooth chart for the standard smooth structure on Bn .
Bn is Hausdorff and second-countable, since it is a subspace of Rn . Let
Ui+ = {(x1 , . . . , xn ) ∈ Bn : xi > 0}
and Ui− = {(x1 , . . . , xn ) ∈ Bn : xi < 0}. Then the Ui± form an open cover of Bn . Define
n,±
φ± ±
i : Ui → Bi ,
q
(x1 , . . . , xn ) 7→ (x1 , . . . , xi−1 , xi ∓ 1 − (x1 )2 − · · · − (x
[ i )2 − · · · − (xn )2 , xi+1 , . . . , xn ),
where Bn,+ i = {(x1 , . . . , xn ) ∈ Bn : xi > 0} and Bn,− i = {(x1 , . . . , xn ) ∈ Bn : xi 6 0}. The

±
inverse of φi is then
(φ± )−1 : Bn,± → Ui±
qi i
(x1 , . . . , xn ) 7→ (x1 , . . . , xi−1 , xi ± 1 − (x1 )2 − · · · − (x

[ i )2 − · · · − (xn )2 , xi+1 , . . . , xn ).
Since both φ± i and (φi )

± −1
are continuous, we conclude that (Ui± , φ± i ) are boundary charts
covering M , so M is a topological manifold.
Let (x1 , . . . , xn ) ∈ Sn−1 . Say (x1 , . . . , xn ) ∈ Ui± . Then
q
xi = ± 1 − (x1 )2 − · · · − (x [ i )2 − · · · − (xn )2 ,
and φ± 1 n 1
i (x , . . . , x ) = (x , . . . , x
i−1
, 0, xi+1 , . . . , xn ), so every point in Sn−1 is a boundary
point. If (x1 , . . . , xn ) lies in Bn , then it lies in the interior chart (Bn , id), and is thus an
interior point.
Note that φ± i and (φi )
± −1
can be extended to maps from R×· · ·×R×(−1, 1)×R×· · ·×R
to itself, where (−1, 1) is the ith factor, and . Since both extensions of φ± ± −1
i and (φi ) are
both diffeomorphisms, the transition functions are all smooth, so the charts given above
define a smooth structure on M . Moreover, since every chart is smoothly compatible with
(Bn , id), every interior chart is smoothly compatible with the standard smooth structure on
Bn .
Problem 1.12. Prove Proposition 1.45 (a product of smooth manifolds together with one
smooth manifold with boundary is a smooth manifold with boundary).
We will show that if M is a smooth m-manifold and N is a smooth n-manifold with
boundary, then M ×N is a smooth manifold with boundary, and ∂(M ×N ) = M ×∂N . The
general case follows by induction. First, M ×N is Hausdorff and second-countable, since both
M and N are. Given charts (U, φ) and (V, ψ) for M and N , respectively, we let (U ×V, φ×ψ)
be a chart for M × N . The collection of all such charts then gives M × N the structure of a
smooth manifold with boundary, as we now show. Let (U1 ×V1 , φ1 ×ψ1 ) and (U2 ×V2 , φ2 ×ψ2 )
be two intersecting charts. Note that (φ2 × ψ2 ) ◦ (φ1 × ψ1 )−1 = (φ2 ◦ φ−1 −1
1 ) × (ψ2 × ψ1 ) has
domain (φ1 ◦ ψ1 )([U1 × V1 ] ∩ [U2 × V2 ]) = φ1 (U1 ∩ U2 ) × ψ1 (V1 ∩ V2 ). Since ψ1 (V1 ∩ V2 ) can
be extended to an open set such that ψ2 ◦ ψ1−1 is smooth, we have that the product charts
are smoothly compatible. This, M × N is a smooth manifold with boundary.
Let (x, y) ∈ ∂(M ×N ). Then (x, y) is in the domain of some boundary chart (U ×V, φ×ψ).
Since φ(U ) is open in Rm , it follows that (V, ψ) must be a boundary chart. Moreover,
ψ(y) should lie on the boundary of Hn , since otherwise there is an interior chart (V 0 , ψ 0 )
whose domain contains y 0 , and thus (U × V 0 , φ × ψ 0 ) is an interior chart containing (x, y 0 ).
Thus, y ∈ ∂N , so (x, y) ∈ M × ∂N . Conversely, let (x, y) ∈ M × ∂N . Then there is a
boundary chart (V, ψ) such that y ∈ V and ψ(y) is in ∂Hn . Then, if u is in the domain of
a chart (U, φ), then (φ × ψ)(x, y) = (φ(x), ψ(y)) ∈ ∂Hm+n , so (x, y) ∈ ∂(M × N ). Hence,
∂(M × N ) = M × ∂N .
2. Smooth Maps
Exercise 2.1. Let M be a smooth manifold with or without boundary. Show that pointwise
multiplication turns C ∞ (M ) into a commutative ring and a commutative and associative
algebra over R.
Let f, g ∈ C ∞ (M ). We want to show that f + g and f g are also smooth functions.
By Exercise 2.3, f ◦ φ−1 and g ◦ φ−1 are smooth for every coordinate chart (U, φ). Then
(f + g) ◦ φ−1 = (f ◦ φ−1 ) + (g ◦ φ−1 ) and (f g) ◦ φ−1 = (f ◦ φ−1 )(g ◦ φ−1 ) are smooth. Thus
C ∞ (M ) is a commutative ring, since multiplication in R is commutative. Since smooth
functions can be multiplied by real constants without affecting smoothness, C ∞ (M ) is also a
commutative and associative algebra over R, since multiplication of functions is commutative
and associative.
Exercise 2.2. Let U be an open submanifold of Rn with its standard smooth manifold
structure. Show that a function f : U → Rk is smooth in the sense just defined if and only
if it is smooth in the sense of ordinary calculus. Do the same for an open submanifold with
boundary in Hn .
The chart (U, id) defines a smooth structure on U . Let f be smooth in the sense defined in
the section. Then, for each chart p ∈ U , there is a chart (V, ψ) whose domain contains p and
such that f ◦ ψ −1 is smooth. Since ψ ◦ id−1 = ψ is smooth, we have that (f ◦ ψ −1 ) ◦ ψ = f is
smooth on V . Since this holds for every chart, we conclude that f is smooth in the ordinary
calculus sense. Now suppose that f is smooth in the sense of ordinary calculus. Then, since
every p ∈ U and f ◦ id−1 is smooth, f is smooth in the sense of the section.
By Exercise 1.44, if U ⊆ Hn is open, then (U, id) gives U a smooth manifold structure,
since id : U → U can be extended to all of R. Let f : U → R be smooth in the sense of the
section. Then for every p ∈ M , there is an open set V 3 p in U and a chart (V, ψ) such
that f ◦ ψ −1 is smooth on some open extension of V . Since ψ ◦ id−1 = ψ is also smooth on
some open extension of V , we conclude that f is smooth on some open extension of V , in
the sense of ordinary calculus. Conversely, let f : U → R be smooth in the sense of ordinary
calculus. Then f = f ◦ id−1 is smooth on some open extension of U . Then f is smooth in
the sense defined in the section.
Exercise 2.3. Let M be a smooth manifold with or without boundary, and suppose f : M →
Rk is a smooth function. Show that f ◦ φ−1 : φ(U ) → Rk is smooth for every smooth chart
(U, φ).
Let (U, φ) be an arbitrary chart, and let p ∈ U . Since f is smooth, there is a chart
(V, ψ) such that p ∈ V and f ◦ ψ −1 : ψ(V ) → Rk is smooth. Since the transition function
ψ ◦φ−1 : φ(U ∩V ) → ψ(U ∩V ) is smooth, we have that (f ◦ψ −1 )◦(ψ ◦φ−1 ) = f ◦φ−1 : φ(U ∩
V ) → Rk is smooth. Since p ∈ U was chosen arbitrarily, we conclude that f ◦ φ−1 is smooth
on all of φ(U ).
Exercise 2.7. Prove propositions 2.5 and 2.6.
We begin with Proposition 2.5.
(Smoothness ⇒ (b)) By smoothness of F , there exists smooth atlases {(Uα , φα )} and
{(Vβ , ψβ )} for M and N , respectively, such that for every α there exists a β such that
F (Uα ) ⊆ Vβ and ψβ ◦ F ◦ φ−1 α is smooth. Now, let (Uα , φα ) and (Vβ , ψβ ) be arbitrary
charts in the atlases described above. Let (Vβ 0 , ψβ 0 ) be a chart such that F (Uα ) ⊆ Vβ 0 . We
−1
have that ψβ 0 ◦ F ◦ φ−1 α is smooth from φα (Uα ) to ψβ 0 (Vβ 0 ), and ψβ ◦ ψβ 0 is smooth from
−1
ψβ 0 (Vβ ∩ Vβ 0 ) to ψβ (Vβ ∩ Vβ 0 ). Hence, (ψβ ◦ ψβ 0 ) ◦ (ψβ 0 ◦ F ◦ φα ) = ψβ ◦ F ◦ φ−1
−1
α is smooth
−1
from φα (Uα ∩ F (Vβ )) to ψβ (Vβ ) (there is no issue with extending the codomain here).
((b) ⇒ (a)) Since {(Uα , φα )} and {(Vβ , ψβ )}, for every p ∈ M , p ∈ Uα for some α and
F (p) ∈ Vβ for some β. Let U = Uα and V = Vα . By continuity of F , U ∩ F −1 (V ) is open.
By assumption, ψ ◦ F ◦ φ−1 : φ(U ∩ F −1 (V )) → ψ(V ) is smooth.
((a) ⇒ Smoothness) Let p be in M , and let (U b , φ) and (V, ψ) be charts containing p and
−1
b ∩ F (V ), we find that F is smooth.
F (p), respectively. Taking U = U
We turn to Proposition 2.6.
For (a), suppose that every p ∈ M has a neighbourhood U such that F |U is smooth.
Then there is a chart (U e , φ̃) of U containing p and a chart (V, ψ) of N containing F (p) such
that F (U ) ⊆ V and ψ ◦ F ◦ φ−1 is smooth. Since U
e e is open in U and U is open in M , we
have that (U e , φ̃) is a chart of M containing p. Thus F is smooth.
For (b), let U ⊆ M be an open submanifold with its standard smooth structure. Let
p ∈ U and let (U 0 , φ) and (V, ψ) be charts containing p and F (p) such that ψ ◦ F ◦ φ−1 is
smooth. Then the restriction (U ∩ U 0 , φ) is a chart in U having all the required properties
for smoothness. Thus, F |U is smooth.
Exercise 2.9. Suppose F : M → N is a smooth map between smooth manifolds with or
without boundary. Show that the coordinate representation of F with respect to every pair
of smooth charts for M and N is smooth.
Let (U, φ) and (V, ψ) be an arbitrary pair of charts for M and N , respectively. If U ∩
F −1 (V ) = ∅, then there is nothing to check, so suppose that p ∈ U ∩ F −1 (V ), so that
F (p) ∈ V . Since F is smooth, there is a chart (U e , φ̃) containing p and a chart (Ve , ψ̃) such
−1 e ) → ψ̃(Ve ) is smooth. Since φ̃◦φ−1 : φ(U ∩ U
e ) ⊆ Ve and ψ̃◦F ◦ φ̃ : φ̃(U e ) → φ̃(U ∩ U
that F (U e)
−1
and ψ ◦ ψ̃ : ψ̃(V ∩ V ) → ψ(V ∩ V ) are smooth, it follows that
e e
(ψ ◦ ψ̃ −1 ) ◦ (ψ̃ ◦ F ◦ φ̃−1 ) ◦ (φ̃ ◦ φ−1 ) = ψ ◦ F ◦ φ−1 : φ(U ∩ U

e ∩ F −1 (V )) → ψ(V ).
Since p ∈ U ∩ F −1 (V ) was arbitrary, we conclude that ψ ◦ F ◦ φ−1 is smooth on all of
φ(U ∩ F −1 (V )).
Exercise 2.11. Prove parts (a)-(c) of Proposition 2.10.
For (a), let c : M → N, p 7→ c be a constant map, where c is some fixed point in N . Let
p ∈ M be arbitrary, and let (U, φ) be a chart containing p and (V, ψ) a chart containing
F (p) = c. Since U ∩ F −1 (V ) = U ∩ M = U is open and ψ ◦ F ◦ φ−1 : φ(U ) → ψ(V ), x 7→ ψ(c)
is smooth (as a constant map between Euclidean spaces), we have that c : M → N is smooth.
For (b), let id : M → M be the identity map. Let p ∈ M and let (U, φ) be a chart
containing p. Then (U, φ) is a chart containing id(p) = p, and φ ◦ id ◦ φ−1 : φ(U ) → φ(U ) is
smooth since it is the identity map, and the identity map on a subset of Euclidean space is
smooth.
For (c), let U ⊆ M be an open submanifold and let i : U ,→ M be the inclusion map.
Let p ∈ U and let (V, φ) be a chart in U containing p. Then (V, φ) is also a chart in M
containing i(p) = p. Then, φ ◦ i ◦ φ−1 : φ(U ) → φ(U ) is the identity map, which is smooth.
Hence i : U ,→ M is smooth.
Exercise 2.16. Prove Proposition 2.15.
For (a), let F : M → N and G : N → P be diffeomorphisms. We that G ◦ F : M → P
is a bijection and is smooth by Proposition 2.10. Since F −1 and G−1 are also smooth, the
composition F −1 ◦ G−1 = (G ◦ F )−1 is also smooth by Proposition 2.10. We conclude that
G ◦ F is a diffeomorphism.
For (b), let Fi : Mi → Ni be diffeomorphisms between smooth manifolds for i = 1, . . . , n.
We want to show that
F1 × · · · × Fn : M1 × · · · × Mn → N1 × · · · × Nn , (p1 , . . . , pn ) 7→ (F1 (p1 ), . . . , Fn (pn ))
is a diffeomorphism. The inverse of F1 × · · · × Fn is F1−1 × · · · × Fn−1 . By Proposition 2.12,
F = F1 × · · · × Fn is smooth if and only if each πi ◦ F is smooth for each i, where πi is
the projection from N1 × · · · Nn 7→ Ni . Since (πi ◦ F )(p1 , . . . , pn ) = Fi (pi ), we have that
πi ◦ F = Fi ◦ πi , where the second πi is the projection from M1 × · · · × Mn to Mi . Since Fi
and πi are smooth, πi ◦ F is smooth, and therefore F is smooth. A similar proof shows that
F −1 is smooth.
For (c), let F : M → N be a diffeomorphism. Then F and F −1 are continuous by
Proposition 2.4, so F is a homeomorphism. Then F is and open map since homeomorphisms
are open maps.
For (d), let F : M → N be a diffeomorphism, let U ⊆ M be an open submanifold, and let
F |U : U → F (U ) be the restriction. Note that F |U = F ◦i, where i : U ,→ M is the inclusion.
By Exercise 2.11, F |U is smooth. The same argument applies to (F |U )−1 : F (U ) → U , since
F (U ) is an open submanifold by part (c).
For (e), M is always diffeomorphic to M , since id : M → M is a diffeomorphism; being
diffeomorphic is reflexive. If F is a diffeomorphism, then so is F −1 ; being diffeomorphic
is symmetric. Finally, since the composition of diffeomorphisms is a diffeomorphism, being
diffeomorphic is transitive. Hence, being diffeomorphic is an equivalence relation.
Exercise 2.19. Use Theorem 1.46 to prove Theorem 2.18.

Let F (p) ∈ F (∂M ). Then p ∈ ∂M , so there is a boundary chart (U, φ) containing p.
Then (F (U ), φ ◦ F −1 ) is a boundary chart, since F and φ are diffeomorphisms and thus
homeomorphisms. Hence, F (p) ∈ ∂N . Hence, F (∂M ) ⊆ ∂N ; applying F −1 to both sides,
we have ∂M ⊆ F −1 (∂N ). But F −1 is a diffeomorphism, so the same argument given above
shows that F −1 (∂N ) ⊆ ∂M . Thus F (∂M ) = ∂N .
Since IntM is open by Exercise 1.39, F restricts to a diffeomorphism from IntM to
F (IntM ) by Proposition 2.15. Since F (∂M ) = ∂N , and every point is either an interior
point or a boundary point, we have that F (IntM ) = IntN .
Exercise 2.24. Show how the proof of Theorem 2.23 needs to be modified for the case in
which M has nonempty boundary.
We need our bases to be composed of regular balls and half balls. We also need to extend
our definition of bump functions to include coordinate half balls. Otherwise the proof is
essentially the same.
Exercise 2.27. Give a counterexample to show that the conclusion of the extension lemma
can be false if A is not closed.
Let M = R (with its standard smooth structure) and let A = (−∞, 0) ∪ (0, ∞). Let
f : A → R be defined by f (x) = 1 if x > 0 and f (x) = −1 if x < 0. Note that f is smooth
since it is smooth on each component of A. Any extension of f to R is discontinuous, and
therefore not smooth, so the extension lemma does not hold here.
Problem 2.1. Define f : R → R by
(
1, x > 0,
f (x) =
0, x < 0.
Show that for every x ∈ R, there are smooth coordinate charts (U, φ) containing x and (V, ψ)
containing f (x) such that ψ ◦ f ◦ φ−1 is smooth as a map from φ(U ∩ f −1 (V )) to ψ(V ), but
f is not smooth in the sense we have defined in this chapter.
Since f is not continuous, it is not smooth in the sense we have defined in this chapter.
Moreover, f is smooth away from x = 0 since it is constant there. Let > 0, let U = (−, ),
and let V = ( 12 , 32 ). Then U contains x = 0 and V contains f (x) = 1. Moreover, we
let (U, id) and (V, id) be charts for R. Then id ◦ f ◦ id−1 = f is the constant map on
f (U ∩ f −1 (V )) = [0, ), and is therefore smooth.
Problem 2.2. Prove Proposition 2.12.
Let F : N → M1 × · · · × Mk be smooth. By Proposition 2.10, in order to show that
πi ◦ F is smooth, it suffices to show that the projection πi : M1 × · · · × Mk → Mi is smooth.
Let (p1 , . . . , pk ) ∈ M1 × · · · × Mk , and let (U1 × · · · × Uk , φ1 × · · · × φk ) be a product
chart containing (p1 , . . . , pk ). Then (Ui , φi ) is a chart containing pi ∈ Mi . The coordinate
representation of φi ◦ πi ◦ (φ1 × · · · × φk )−1 is just the projection from φ1 (U1 ) × · · · × φk (Uk )
to φi (Ui ), which is smooth. Hence, πi is smooth.
Conversely, let πi ◦ F be smooth for each i = 1, . . . , k. Let p ∈ N , let (V, ψ) be a
chart containing p and let (Ui , φi ) be charts containing πi ◦ F (p) such that πi ◦ F (V ) ⊆
Ui and φi ◦ πi ◦ F ◦ ψ −1 : ψ(V ) → φi (Ui ) is smooth for each i = 1, . . . , k. Now, then
(U1 × · · · × Uk , φ1 × · · · × φk ) is a chart for M1 × · · · × Mk such that F (V ) ⊆ U1 × · · · × Uk .
Then the coordinate representation (φ1 × · · · × φk ) ◦ F ◦ ψ −1 : ψ(V ) → φ1 (U1 ) × · · · × φk (Uk )
is smooth, since the projection onto each φi (Ui ) is equal to φi ◦πi ◦F ◦ψ −1 , which is smooth.
Problem 2.3. For each of the following maps between spheres, compute sufficiently many
coordinate representations to prove that it is smooth.
(a) pn : S1 → S1 is the nth power map for n ∈ Z, given in complex notation by pn (z) =
zn.
(b) α : Sn → Sn is the antipodal map α(x) = −x.
(c) F : S3 → S2 is given by F (w, z) = (z w̄ + wz̄, iwz̄ − iz w̄, z z̄ − ww̄), where we think
2 2
of S3 as the subset {(w, z) : |w| + |z| = 1} of C2 .
For (a), let z ∈ S1 , let (U, θ) be an angle coordinate chart containing z, and let (V, φ) be an
angle coordinate chart containing z n . Then φ◦pn ◦θ−1 (x) = φ◦pn (eix ) = φ(einx ) = nx+2kπ
for some k, which is constant on each component of θ(U ∩ p−1 −1
n (V )). Note that U ∩ pn (V )
is open, since pn is continuous. Hence, pn is smooth.
For (b), let p ∈ Sn , and assume that (Sn r {N }, σ) is the stereographic chart from
the north and it contains p. Then (Sn r S, σ̃) contains α(p), where σ̃ is the stereographic
projection from the south. A computation shows that σ̃ ◦ α ◦ σ −1 (u) = −u, which is smooth.
Hence, α is smooth.
For (c), writing F in real coordinates, we have
F (x1 , x2 , x3 , x4 ) = (2x1 x3 + 2x2 x4 , 2x1 x4 − 2x2 x3 , (x3 )2 + (x4 )2 − (x1 )2 − (x2 )2 ),
so F is continuous, since it is the restriction of a continuous function. A computation shows
that
2 2
(8u1 u3 + 4u2 (|u| − 1), 4u1 (|u| − 1) − 8u2 u3 )
σS2 ◦ F ◦ σS−1
3 (u1 , u2 , u3 ) = 2 ,
1 + (2u1 )2 + (2u2 )2 − (2u3 )2 − (|u| − 1)2
which is smooth on σS3 (S3 r{N }∩F −1 (S2 r{N })). Here σSn is the stereographic projection
from the north of Sn . Similar computations for different pairs of charts show that F is indeed
a smooth function.
Problem 2.4. Show that the inclusion map Bn ,→ Rn is smooth when Bn is regarded as a
smooth manifold with boundary.
We give Bn the smooth structure described in 1.11. Let x ∈ Bn be contained in some
chart (Ui± , φ± n n n
i ). Then i(x) = x is contained in the chart (R , id), where i : B ,→ R is the
± −1 ± −1
inclusion. Then id ◦i ◦ (φi ) = (φi ) , which is smooth, as discussed in Problem 1.11.
Problem 2.5. Let R be the real line with its standard smooth structure, and let R e denote the
same topological manifold with the smooth structure defined in Example 1.23. Let f : R → R
be a function that is smooth in the usual sense.
(a) Show that f is also smooth as a map from R to R. e
(b) Show that f is smooth as a map from R e to R if and only if f (n) (0) = 0 whenever n
is not an integral multiple of 3.
For (a), the coordinate representation ψ ◦ f ◦ id−1 = ψ ◦ f is smooth, since ψ(x) = x3
and f are both smooth in the sense of regular calculus.
For (b), the coordinate representation is id ◦ f ◦ ψ −1 = f ◦ ψ −1 , and f ◦ ψ −1 (x) = f (x1/3 ).
Let f ◦ ψ −1 . We appeal to Faà di Bruno’s formula, which states that for smooth functions
F and G
n (j) mj
dn X n! (m1 +···+mn )
Y G (x)
F (G(x)) = F (G(x)) ,
dxn m1 ! · · · mn ! j=1
j!
where the sum ranges over all n-tuples of nonnegative integers (m1 , . . . , mn ) such that
m1 + 2m2 + · · · + nmn = n. Since we are assuming f ◦ ψ −1 is smooth, we can apply the
above formula to F = f ◦ ψ −1 and G = ψ. Note that ψ (j) (0) = 0 for all j 6= 3. Hence, if
f (n) (0) 6= 0, then there must be an n-tuple (m1 , . . . , mn ) such that mj = 0 for all j 6= 3.
Then n = 3m3 . This proves the first direction.
For the other direction, we first prove the following fact by induction on k: if F ∈ C k (R),
1 1 1 1
then so is xk+ 3 F (x 3 ). For k = 0, the result is clear, since xk+ 3 F (x 3 ) is continuous. For
k
the inductive step, let F ∈ C (R). Then
d h k+ 1 1
i 1 1 1 1 1 1
x 3 F (x 3 ) = x(k−1)+ 3 F (x 3 ) + x(k−1)+ 3 x 3 F 0 (x 3 ).
dx 3
By induction,
1 1
x(k−1)+ 3 F (x 3 ) ∈ C k−1 (R).
1 1
Since xF 0 (x) ∈ C k−1 (R), we also have 13 x(k−1)+ 3 F 0 (x 3 ) ∈ C k−1 (R) by induction. Since it
1 1
derivative is C k−1 (R), we have that xk+ 3 F (x 3 ) ∈ C k (R), concluding the proof of the claim.
Suppose that f (n) (0) = 0 for all non-multiples of 3. Using the Taylor remainder theorem
to write f up to the 3mth term:
f (3) (0) 3 f (3m) (0) 3m
f (x) = f (0) + x + ··· + x + x3m+1 F3m+1 (x)
3! (3m)!
for some smooth function F3m+1 . Then
1 f (3) (0) f (3m) (0) m 1 1
f (x 3 ) = f (0) + x + ··· + x + xm+ 3 F3m+1 (x 3 ).
3! (3m)!
1 1
By the claim, xm+ 3 F3m+1 (x 3 ) ∈ C m (R). Hence, f ◦ ψ −1 ∈ C m (R) for all m > 0, and is
therefore smooth.
Problem 2.6. Let P : Rk+1 r {0} → Rk+1 r {0} be a smooth function, and suppose that
for some d ∈ Z, P (λx) = λd P (x) for all λ ∈ R r {0} and x ∈ Rn+1 r {0}. (Such a function
is said to be homogeneous of degree d.) Show that the map Pe : RPk → RPk defined by
Pe([x]) = [P (x)] is well defined and smooth.
If [x] = [y], then x = λy for some λ ∈ R r {0}. Then Pe([x]) = [P (x)] = [P (λy)] =
d
[λ P (y)] = [P (y)] = Pe([y]). Therefore, Pe is well defined.
Let [x] ∈ Ui and suppose that Pe([x]) ∈ Uj . The coordinate representation φj ◦ Pe ◦ φ−1
i
takes a point (u1 , . . . , uk ) ∈ φi (Ui ) to
(P1 (α), . . . , Pj−1 (α), Pj+1 (α), . . . , Pn+1 (α))
,
Pj (α)
where α = φi (u1 , . . . , uk ) = (u1 , . . . , ui−1 , 1, ui , . . . , uk ) and Pm is the mth component of P .
Since P is smooth, each Pm is smooth. Therefore, the coordinate representation is smooth,
and thus so is Pe.
Problem 2.7. Let M be a nonempty smooth n-manifold with or without boundary, and
suppose n > 1. Show that the vector space C ∞ (M ) is infinite-dimensional.
We begin by proving the suggested claim from the hint, namely that if f1 , . . . , fk ∈
C ∞ (M ) have disjoint support and are nonzero, then they are linearly independent. Consider
a linear combination of these functions set to 0:
α1 f1 + · · · + αk fk = 0.
For each i = 1, . . . , k, choose some xi ∈ M such that fi (xi ) 6= 0. Since fj (xi ) = 0 for each
j 6= i, we have αi f (xi ) = 0 ⇒ αi = 0. Since i was arbitrary, the fi are linearly independent.
Now, let (U, φ) be some chart and let k ∈ N be arbitrary. It is not hard to see that
there are k disjoint open balls Bri (φ(xi )) in φ(U ). By Proposition 2.25, there are smooth
bump functions that are positive on φ−1 (Bri /2 (φ(xi ))) and supported on φ(Bri (φ(xi ))).
Since the support of all these balls are disjoint, the corresponding bump functions are
linearly independent. Since k ∈ N was arbitrarily chosen, we have that C ∞ (M ) is infinite
dimensional.
Problem 2.8. Define F : Rn → RPn by F (x1 , . . . , xn ) = [x1 , . . . , xn , 1]. Show that F is a
diffeomorphism onto a dense open subset of RPn . Do the same for G : Cn → CPn defined
by G(z 1 , . . . , z n ) = [z 1 , . . . , z n , 1].
We claim that F : Rn → Un+1 is a diffeomorphism, where Un+1 is an open submanifold of
RPn . We begin by showing that F is injective; suppose that [x1 , . . . , xn , 1] = [y 1 , . . . , y n , 1].
Then (x1 , . . . , xn , 1) = λ(y 1 , . . . , y n , 1), so 1 = λ · 1 ⇒ λ = 1. To see that F is surjec-
x1 n
tive, let [x1 , . . . , xn+1 ] ∈ Un+1 . Then F ( xn+1 , . . . , xxn+1 ) = [x1 , . . . , xn+1 ]. The coordinate
representation of F is
φn+1 ◦ F : Rn → φn+1 (Un+1 ), (x1 , . . . , xn ) 7→ (x1 , . . . , xn ),
which is of course smooth. The inverse of F is given by
1
xn

−1 n 1 n+1 x
F : Un+1 → R , [x , . . . , x ] 7→ , . . . , n+1 .
xn+1 x
The coordinate representation is F −1 ◦ φ−1 n+1 = (φn+1 ◦ F )
−1
= id which is smooth. Hence,
F is a diffeomorphism. It remains to show that Un+1 is open and dense in RPn . We know
that Un+1 is open since it is the domain of a chart of RPn .
For density, let [x1 , . . . , xn+1 ] ∈ RPn r Un+1 and let U ⊆ RPn be an open set containing
[x , . . . , xn+1 ]. Since π is surjective, we have that U = π(π −1 (U )). Since π is continuous,
1
we have that π −1 (U ) is open, and therefore it contains a point (y 1 , . . . , y n+1 ) such that
y n+1 6= 0. Hence, [y 1 , . . . , y n+1 ] ∈ U ∩ Un+1 . Thus, every open set containing [x1 , . . . , xn+1 ]
intersects Un+1 , and therefore Un+1 is dense in RPn .
Doing the same for G : C → CPn is completely analogous.
Problem 2.9. Given a polynomial p in one variable with complex coefficients, not identically
zero, show that there is a unique smooth map pe : CP1 → CP1 that makes the following
diagram commute, where CP1 is 1-dimensional complex projective space and G : C → CP1
is the map from Problem 2.8:
G
C CP1
p p̃
G
C CP1
Let p(z) = an z n + · · · + a1 z + a0 with an 6= 0. Define P : C2 r {0} → C2 r {0} by
P (z 1 , z 2 ) = (an (z 1 )n + an−1 (z 1 )n−1 z 2 + · · · + a1 z 1 (z 2 )n−1 + a0 (z 2 )n , (z 2 )n ).
To see that the image of this map does not include 0, not that if z 2 = 0, then z1 6= 0 and
P (z 1 , 0) = (an (z 1 )n , 0) 6= 0. Since P is homogeneous of degree n, p̃ : CP1 → CP1 , [z 1 , z 2 ] 7→
[P (z 1 , z 2 )] is well defined. It follows directly from the definition that p̃ ◦ G = G ◦ p. Since P
is smooth, we have that p̃ is smooth by Problem 2.6. If there is another smooth p̃0 making
the diagram commute, then p̃0 must agree with p̃ on the image of G, which is dense in CP1 .
Since p̃ and p̃0 are smooth, they are equal.
Problem 2.10. For any topological space M , let C(M ) denote the algebra of continuous
functions f : M → R. Given a continuous map F : M → N , define F ∗ : C(N ) → C(M ) by
F ∗ (f ) = f ◦ F .
(a) Show that F ∗ is a linear map.
(b) Suppose M and N are smooth manifolds. Show that F : M → N is smooth if and
only if F ∗ (C ∞ (N )) ⊆ C ∞ (M ).
(c) Suppose F : M → N is a homeomorphism between smooth manifolds. Show that it
is a diffeomorphism if and only if F ∗ restricts to an isomorphism from C ∞ (N ) to
C ∞ (M ).
For (a), let f, g ∈ C ∞ (M ) and let α, β ∈ R. Then F ∗ (αf + βg) = (αf + βg) ◦ F =
αf ◦ F + βg ◦ F . Hence, F ∗ is linear.
For (b), let F : M → N be smooth, and let f ∈ C ∞ (N ). Since the composition of smooth
functions is smooth, F ∗ (f ) = f ◦ F ∈ C ∞ (M ). Hence, F ∗ (C ∞ (N )) ⊆ C ∞ (M ). Conversely,
suppose that F ∗ (C ∞ (N )) ⊆ C ∞ (M ). Let B = {(V, ψ)} be an atlas of regular coordinate
balls in N and let A = {(U, φ)} any atlas for M . By the extension lemma for smooth
functions (Lemma 2.26), each coordinate map ψ : V → Rn can be extended to a smooth
map ψ : N → Rn , note that each extended component function ψ i : N → R is in C ∞ (N ).
Let p ∈ M be arbitrary and be contained in a smooth chart (U, φ), and let F (p) be in the
chart (V, ψ). We want to show that ψ ◦ F ◦ φ−1 : φ(U ∩ F −1 (V )) → ψ(V ) is smooth. Note
that U ∩F −1 (V ) is open since F is assumed to be continuous. We have that ψ ◦F : M → Rn
is smooth, since each of the component functions are F ∗ (ψ i ) ∈ C ∞ (M ). The restriction to
U ∩ F −1 (V ) is also smooth, and since φ−1 is smooth, we have that ψ ◦ F ◦ φ−1 is smooth.
Hence, F is smooth.
For (c), let F be a diffeomorphism. Then F ∗ (C ∞ (N )) ⊆ C ∞ (M ) and (F −1 )∗ (C ∞ (M )) ⊆
C (N ). Since F is invertible, F ∗ must be injective. We conclude that F ∗ : C ∞ (N ) →
∞
C ∞ (M ) is a bijection. It is also linear by part (a). Finally, if f, g ∈ C ∞ (N ), then F ∗ (f g) =

(f g) ◦ F = (f ◦ F )(g ◦ G). Thus, F ∗ is an algebra isomorphism when restricted to C ∞ (N ).
Conversely, suppose that F ∗ restricts to an isomorphism from C ∞ (N ) to C ∞ (M ). Then
F ∗ (C ∞ (N )) ⊆ C ∞ (M ) and (F −1 )∗ (C ∞ (M )) ⊆ C ∞ (N ), so part (b) implies that F and
F −1 are smooth, and therefore that F is a diffeomorphism.
Problem 2.11. Suppose V is a real vector space of dimension n > 1. Define the pro-
jectivization of V , denoted by P(V ), to be the set of 1-dimensional linear subspaces of V ,
with the quotient topology induced by the map π : V r {0} → P(V ) that sends x to its
span. (Thus P(Rn ) = RPn .) Show that P(V ) is a topological (n − 1)-manifold, and has a
unique smooth structure with the property that for each basis (E1 , . . . , En ) for V , the map
E : RPn−1 → P(V ) defined by E[v 1 , . . . , v n ] = [v i Ei ] (where the brackets denote equivalence
classes) is a diffeomorphism.
Let V and W be n-dimensional vector spaces, and let T : V → W be a linear homeomor-
phism (V and W have topologies inherited from any of their norms and the topologies are
independent of the chosen norms). Define Te : P(V ) → P(W ), [v] 7→ [T (v)]. Linearity of T
guarantees that Te is well defined. If [T (v1 )] = [T (v2 )], then T (v1 ) = λT (v2 ) = T (λv2 ) ⇒
v1 = λv2 ⇒ [v1 ] = [v2 ], so Te is injective; Te is clearly surjective, so it is bijective.
We want to show that Te is a homeomorphism. Let πV : V → P(V ) and πW : W →
P(W ) be the natural projections, and let U ⊆ P(W ) be open. Te−1 (U ) is open if and
only πV−1 (Te−1 (U )) = (Te ◦ πV )−1 (U ) is. But Te ◦ πV = πW ◦ T , which is continuous, hence
πV−1 (Te−1 (U )) is open, so Te−1 (U ) is open. This show that Te is continuous. The same idea
shows that Te−1 (which is given by Te−1 (w) = [T −1 (w)]) is continuous, since T −1 is also a
linear homeomorphism. We conclude that P(V ) and P(W ) are homeomorphic. Since V is
homeomorphic to Rn via a linear homeomorphism taking any basis of V to the standard basis
of Rn , we have that P(V ) ∼ = P(Rn ) = RPn−1 . Since RPn−1 is a topological (n − 1)-manifold,
P(V ) is too.
Fix a basis (B1 , . . . , Bn ) for V , and let T : Rn → V be the linear homeomorphism that
sends the standard basis vector ei to Bi for each i = 1, . . . , n. Let Te : RPn−1 → P(V ) be
the induced homeomorphism. Consider the standard smooth atlas {(Ui , φi )}ni=1 for RPn−1 .
I claim that {(Te(Ui ), φi ◦ Te−1 } is a smooth atlas for P(V ). Indeed, the transition functions
are (φi ◦ Te−1 ) ◦ (φj ◦ Te−1 )−1 = φi ◦ φ−1 j which are smooth. Now let (E1 , . . . , En ) be a basis
for V , and let E : RPn−1 → P(V ) be defined as above. Then the coordinate representation
of E is
(φi ◦ Te−1 ) ◦ E ◦ φ−1 (x1 , . . . , xn−1 ) = φi [T −1 F (x1 , . . . , xj−1 , 1, xj , . . . , xn−1 )],
j
where F : R → V is defined by F (v 1 , . . . , v n ) = v i Ei . Since T −1 F is an invertible linear

n
map from Rn to itself, it is a diffeomorphism. Since φi and π : Rn → RPn−1 are smooth, the
coordinate representation is smooth. Therefore, E is smooth. The coordinate representation
of E −1 is
φi ◦ E −1 ◦ (φj ◦ Te−1 )−1 (x1 , . . . , xn−1 ) = φi [F T −1 (x1 , . . . , xj−1 , 1, xj , . . . , xn−1 )],
so E −1 is smooth as well. Therefore, E is a diffeomorphism.
Let A be a smooth atlas for P(V ) such that Te is a diffeomorphism, and let (U, φ) ∈ A
be a smooth chart. Then φi ◦ Te−1 ◦ φ−1 is smooth for each i. But this just means that
A is smoothly compatible with the smooth structure described in this section. This proves
uniqueness of the smooth structure.
Problem 2.12. State and prove the analogue of Problem 2.11 for complex vector spaces.
If V is a complex vector space of dimension > 1, define the projectivization of V , denoted
by P(V ) to be the set of all 1-dimensional complex linear subspaces of V , with the quotient
topology induced by the map π : V r {0} → P(V ) that sends x to its span. Then P(V ) is a
topological (2n − 2)-manifold, and has a unique smooth structure with the property that for
each basis (E1 , . . . , En ) fpr V , the map E : CPn−1 → P(V ) defined by E[z 1 , . . . , z n ] = [z i Ei ]
is a diffeomorphism.
The proof of the previous problem only used the fact that all n-dimensional real vector
spaces are isomorphic (and homeomorphic) via a linear map that sends bases to bases, as well
as the structure of the coordinate charts on RPn−1 . Since the same fact is true of complex
vector spaces, and the coordinate charts on CPn−1 are defined completely analogously to
the coordinate charts on RPn−1 , the proof of the above statement is given in the previous
problem, with the exception that we must replace R with C whenever it appears.
Problem 2.13. Suppose that M is a topological space with the property that for every
indexed open cover X of M , there exists a partition of unity subordinate to X . Show that
M is paracompact.
Let X = {Xα }α be an open cover, and let (ψα )α be a partition of unity subordinate
to it. Let Uα = P{p ∈ M : ψα (p) > 0} = ψα−1 ((0, ∞)). Then Uα is open, since ψα is
continuous. Since α ψα (p) = 1 for all p ∈ M , we have that the Uα cover M . Moreover,
since Uα ⊆ supp(ψα ) and {sup(ψα )} is locally finite, {Uα }α is locally finite. Moreover,
Uα ⊆ Xα , so {Uα }α is a refinement of X . Hence, M is paracompact.
Problem 2.14. Suppose A and B are disjoint closed subsets of a smooth manifold M .
Show that there exists f ∈ C ∞ (M ) such that 0 6 f (x) 6 1 for all x ∈ M , f −1 (0) = A, and
f −1 (1) = B.
By Theorem 2.29, there are functions g, h : M → [0, ∞) such that g −1 (0) = A and
g
h−1 (0) = B. It is immediate to check that g+h has the desired property.
3. Tangent Vectors
Exercise 3.5. Prove Lemma 3.4.
For (a), let f1 : M → R be the constant function at 1. Then
vf1 = v(f12 ) = f1 (p)vf1 + f1 (p)vf1 = 2vf1 ,
so vf1 = 0. If f is the constant function at c ∈ R, then vf = v(cf1 ) = c · vf1 = 0.
For (b), we have
v(f g) = f (p)vg + g(p)vf = 0.
For (a), let u, v ∈ Tp M be derivations, let a, b ∈ R, and let f ∈ C ∞ (N ). Then
dFp (au + bv)(f ) = (au + bv)(f ◦ F ) = au(f ◦ F ) + bv(f ◦ F ) = adFp (u)(f ) + bdFp (v)(f ).
Since f was arbitrary, we have that dFp (au + bv) = adFp (u) + bdFp (v).
For (b), let v ∈ Tp M be a derivation and let f ∈ C ∞ (P ). Then
d(G ◦ F )p (v)(f ) = v(f ◦ G ◦ F ) = dFp (v)(f ◦ G) = dGF (p) ◦ dFp (v)(f ).
For (c), we have
d(IdM )p (v)(f ) = v(f ◦ IdM ) = vf.
For (d), let F : M → N be a diffeomorphism. Parts (b) and (c) imply IdTp M = d(IdM )p =
d(F −1 ◦ F )p = d(F −1 )F (p) ◦ dFp and IdTp N = d(IdN )F (p) = d(F ◦ F −1 )F (p) = dFp ◦
d(F −1 )F (p) . This shows that dFp is an isomorphism and that (dFp )−1 = d(F −1 )F (p) .
Exercise 3.17. Let (x, y) denote the standard coordinates on R2 . Verify that (x̃, ỹ) are
global smooth coordinates on R2 , where
x̃ = x, ỹ = y + x3 .
Let p be the point (1, 0) ∈ R2 (in standard coordinates), and show that

∂ ∂
6= ,
∂x p ∂ x̃ p
even though the coordinate functions x and x̃ are identically equal.
The map φ : R2 → R2 , (x, y) 7→ (x, y + x3 ) is smooth with smooth inverse φ−1 : R2 →
R , (x, y) 7→ (x, y − x3 ). Therefore, (x̃, ỹ) are global smooth coordinates on R2 . Now,
2

∂ ∂ x̃ ∂ ∂ ỹ ∂ ∂ ∂ ∂
= + = +3 6= .
∂x p ∂x p ∂ x̃ ∂x p ∂ ỹ ∂ x̃ ∂ ỹ ∂ x̃ p
Exercise 3.19. Suppose M is a smooth manifold with boundary. Show that T M has a
natural topology and smooth structure making it into a smooth manifold with boundary,
such that if (U, (xi )) is any smooth boundary chart for M , then rearranging the coordinates
in the natural chart (π −1 (U ), (xi , v i )) for T M yields a boundary chart (π −1 (U ), (v i , xi )).
We begin by defining the topology on T M , which I believe was not done explicitly in
Lee. Let A = {(Uα , φα )} be a smooth atlas for M . For every (U, φ) ∈ A, we have that
φ̃ : π −1 (U ) → φ(U ) × Rn , (p, v i ∂/∂xi p ) 7→ (φ(p), v i ei )

is a bijection. Hence, we obtain a topology on π −1 (U ) by pulling back the topology on

φ(U ) × Rn . We then define a subset of T M to be open if its intersection with each π −1 (U )
is open; it is not hard to check that this defines a topology on T M . We define the coordinate
charts as in the proof of Proposition 3.18, and the proof that they define a smooth structure
on T M is the same.
It remains to show that if (U, φ) is a boundary chart, then (π −1 (U ), φ̃) is a boundary
chart (after some rearranging of coordinates). Simply note that φ(U ) ⊆ Hn , so φ̃(φ−1 (U )) =
φ(U )×Rn ⊆ Hn ×Rn ∼ = H2n . Moreover, φ̃(p, v) ∈ ∂H2n for any p ∈ U such that φ(p) ∈ ∂Hn .
−1
Hence, (π (U ), φ̃) is a boundary chart. The statement of the exercise asks us to rearrange
coordinates so that the last coordinate of the chart intersects the boundary, but this is not
so important.
Exercise 3.27. Show that any (covariant or contravariant) functor from C to D takes
isomorphisms in C to isomorphisms in D.
Let F be a covariant functor from C to D, and let f ∈ homC (X, Y ) be an isomorphism.
Then IdF (X) = F(f −1 ◦ f ) = F(f −1 ) ◦ F(f ) and IdF (Y ) = F(f ◦ f −1 ) = F(f ) ◦ F(f −1 ).
Hence, F(f ) is an isomorphism, with inverse F(f −1 ). The proof is similar in the case where
F is a contravariant functor.
Problem 3.1. Suppose M and N are smooth manifolds with or without boundary, and
F : M → N is a smooth map. Show that dFp : Tp M → TF (p) N is the zero map for each
p ∈ M if and only if F is constant on each component of M .
(⇒) Let dFp be the zero map for each p ∈ M . Let q ∈ N be in the image of F . We
will show that F −1 (q) is both open and closed. Since {q} ⊆ N is closed and F is smooth,
F −1 (q) is closed. On the other hand, let p0 ∈ M be such that F (p0 ) = q, and let (B, φ) be
a coordinate ball centred at p0 . If p1 is any other point in B, then there is a straight-line
smooth path γ : [0, 1] → φ(B) such that γ(0) = φ(p0 ) and γ(1) = φ(p1 ). Then φ−1 ◦ γ is
a smooth path from p0 to p1 and F ◦ φ−1 ◦ γ is a smooth path from F (p0 ) to F (p1 ). By
Proposition 3.24, (F ◦ φ−1 ◦ γ)0 (t) = dF ((φ−1 ◦ γ)0 (t)) = 0 for every t ∈ [0, 1].
We will now show that a smooth curve with velocity identically equal to 0 is a constant
path. Let P be a smooth manifold, and let γ : J → N be a path with velocity identically
equal to zero (where J ⊆ R is some interval). Let (V, ψ) be a chart containing γ(t) for some
t. Then ψ ◦ γ is a curve in ψ(V ) ⊆ Rn , and (F ◦ γ)0 (t) = dF (γ 0 (t)) = dF (0) = 0. A smooth
curve of velocity 0 in Rn is constant, and therefore γ is constant.
Hence, the curve F ◦ φ−1 ◦ γ is constant, so F (p0 ) = F (p1 ) = q. We conclude that
B ⊆ F −1 (q), so F −1 (q) is open. Therefore, F −1 (q) is either a component or a union of
components. We conclude that F is constant on each component of M .
(⇐) If F is constant, then since dFp only depends on F in a neighbourhood of p, and F
is constant on the component of p, we have that dFp is the zero map for each p.

For each i = 1, . . . , k, let βi : Mi → M1 × · · · × Mk , x 7→ (p1 , . . . , pi−1 , x, pi+1 , . . . , pk ), and
let
β : Tp1 M1 ⊕· · ·⊕Tpk Mk → Tp (M1 ×· · ·×Mk ), v1 ⊕· · ·⊕vk 7→ d(β1 )p1 (v1 )+· · ·+d(βk )pk (vk ).
It is straightforward to check that β is a linear map. Let α be the map defined in Proposition
3.14. Note that πi ◦βi is the identity on Mi if i = j, and is the constant map at pj otherwise.
It follows from this observation and part (b) of Proposition 3.6 that α ◦ β is the identity on
Tp1 M1 ⊕· · ·⊕Tpk Mk . Then α is surjective.
P Since Tp1 M1 ⊕· · ·⊕Tpk Mk and Tp (M1 ×· · ·×Mk )
have the same dimension (namely i dimMi ), and α is a surjection, α is an isomorphism.
If one of the Mi is a smooth manifold with boundary, then M1 × · · · × Mk is a smooth
manifold with boundary. Hence it has a well defined tangent space at each point, and
differentials of maps are defined. The same proof then applies in this case.
Problem 3.3. Prove that if M and N are smooth manifolds, then T (M × N ) is diffeomor-
phic to T M × T N .
By the preceding problem, the tangent space T(p,q) (M × N ) can be identified with Tp M ⊕
Tq N for every pair (p, q) ∈ M × N . Thus, we can define a map
F : T (M × N ) → T M × T N, ((p, q), u ⊕ v) 7→ ((p, u), (q, v)).
We claim that F is a diffeomorphism. The inverse of F is given by F −1 ((p, u), (q, v)) =
((p, q), u ⊕ v). Let (U, φ) and (V, ψ) be charts for M and N , respectively. Let πM , πN , πM ×N
be projections from T M, T N, T (M × N ) to M, N, M × N , respectively. We have a chart
−1
(πM ×N (U × V ), α) for T (M × N ), where
−1 m n
α : πM ×N (U × V ) → φ(U ) × ψ(V ) × R × R
!
∂ ∂
(p, q), ui i ⊕ v j j

7→ (φ(p), ψ(q), u, v).
∂x (p,q) ∂y (p,q)
−1 −1
We also have a corresponding chart (πM (U ) × πN (V ), β) for T M × T N , where
−1 −1
β : πM (U ) × πN (V ) → φ(U ) × Rm × ψ(V ) × Rn
! !!
∂ ∂
p, ui i , q, v j j

7→ (φ(p), u, ψ(q), v).
∂x p ∂y q
In both maps, u = (u1 , . . . , um ) and v = (v 1 , . . . , v n ). The coordinate representation of F
is β ◦ F ◦ α−1 (x, y, u, v) = (x, u, y, v) and the coordinate representation of F −1 is α ◦ F ◦
β −1 (x, u, y, v) = (x, y, u, v). Both of these maps are smooth, so F is a diffeomorphism.
Problem 3.4. Show that T S1 is diffeomorphic to S1 × R.
View S1 ⊆ C. We have two angle charts (U, θ) and (V, φ) covering S1 , where U = S1 r{1},
θ : U → (0, 2π), and V = S1 r {−1}, φ : V → (−π, π). Define F : T S1 → S1 × R as follows:
d d
if z ∈ U , let F (z, v dθ |z ) = (z, v) and if z ∈ V , then let F (z, v dφ |z ) = (z, v). It is not hard
d d
to show that dθ |z = dφ |z for z ∈ U ∩ V ; hence, F is well defined. Note that F restricted to
π −1 (U ) or π −1 (V ) is a diffeomorphism from π −1 (U ) to U × R and from π −1 (V ) to V × R,
and it agrees on the overlap π −1 (U ) ∩ π −1 (V ). By the gluing lemma for smooth maps
(Corollary 2.8), F is a diffeomorphism.
Problem 3.5. Let S1 ⊆ R2 be the unit circle, and let K ⊆ R2 be the boundary of the
square of side 2 centred at the origin: K = {(x, y) : max(|x|, |y|) = 1}. Show that there is
a homeomorphism F : R2 → R2 such that F (S1 ) = K, but there is no diffeomorphism with
the same property.
Let T1 ⊆ R2 be the set {(x, y) : x > 0, |x| > |y|}. Let r : R2 → R2 be the π/2 angle
clockwise rotation about the origin, and let T2 = r(T1 ), T3 = r2 (T1 ), and T4 = r3 (T1 ). Note
S4
that R2 = i=1 Ti . We begin by defining a homeomorphism F1 : T1 → T1 . For (x, y) ∈ T1 ,
let (q 2
1 + xy 2 (x, y) if (x, y) 6= (0, 0)
F1 (x, y) =
0 if (x, y) = (0, 0)
q
2 √
For (x, y) 6= (0, 0), we have |F1 (x, y)| = 1 + xy 2 |(x, y)| 6 2|(x, y)|, so F1 (x, y) tends to 0
as (x, y) tends to (0, 0) in T1 . Thus, F1 is continuous. The inverse of F1 is given by
( x
√ 2 2 (x, y) if (x, y) 6= (0, 0)
−1 x +y
F1 (x, y) =
0 if (x, y) = (0, 0)
as one can check. In T1 , we have F1 (x, y) 6 |(x, y)|, so F1−1 is continuous on T1 . Thus,
−1
F1 is a homeomorphism. We similarly obtain a homeomorphism of Ti via the function

Fi = ri−1 ◦ F1 ◦ r−(i−1) . Finally, define
F : R2 → R2 , (x, y) 7→ Fi (x, y) if (x, y) ∈ Ti .
This function is well defined, since the Fi agree on the overlaps of their domains. By the
pasting lemma, F : R2 → R2 is a homeomorphism. We check that F (S1 ) = K. Assume that
(x, y) ∈ S1 ∩ T1 . Then
r
y2 1p 2 1 y
F (x, y) = F1 (x, y) = 1 + 2 (x, y) = x + y 2 (x, y) = (x, y) = 1, ∈ K∩T1 .
x x x x
Then F (S1 ) ⊆ K, since S1 and K are both invariant under rotations by π/2. Now let
(x, y) ∈ K ∩ T1 . Then
1
F1−1 (x, y) = F1−1 (1, y) = p (1, y),
1 + y2
so F1−1 (x, y) = 1, meaning F1−1 (x, y) ∈ S1 ∩ T1 . Thus, F (S1 ) = K.

Suppose that there is a diffeomorphism F : R2 → R2 such that F (S1 ) = K. Let γ(t) : R →

R , t 7→ (cos t, sin t). Note that γ(t) ∈ S1 and γ 0 (t) 6= 0 for all t ∈ R. Let tc ∈ R such that
2
F ◦γ(tc ) = (1, 1). Then, for some > 0, there are intervals I− = (tc −, tc ) and I+ = (tc , tc +)
such that F ◦ γ(I− ) ⊆ {1} × (−1, 1) and F ◦ γ(I+ ) ⊆ (−1, 1) × {1} (assuming that F sends
γ counter-clockwise around K). Then
(
(1, y ◦ F ◦ γ(t)) if t ∈ I−
F ◦ γ(t) =
(x ◦ F ◦ γ(t), 1) if t ∈ I+
and
0 d(x ◦ F ◦ γ) ∂ d(y ◦ F ◦ γ) ∂
(F ◦ γ) (tc ) = (tc ) + (tc ) .
dt ∂x (1,1) dt ∂y (1,1)
◦γ) ◦γ)
By continuity of d(x◦F
dt and d(y◦F
dt , they must both vanish at tc . Hence, (F ◦ γ)0 (tc ) = 0.
But Proposition 3.24 implies (F ◦ γ) (tc ) = dF (γ 0 (tc )). Since F is a homeomorphism, dFp
0
is an isomorphism for every p ∈ R2 . In particular, dF (γ 0 (tc )) = dFγ(tc ) (γ 0 (tc )) 6= 0, since

γ 0 (tc ) 6= 0. Thus, F cannot be a diffeomorphism.
Problem 3.6. Consider
S3 as the unit sphere in C2 under the usual identification 2 4
C ↔R .
1 2 3 3 it 1 it 2
For each z = z , z ∈ S , define a curve γz : R → S by γz (t) = e z , e z . Show that
γz is a smooth curve whose velocity is never zero.
Letting z k = xk + iy k for k = 1, 2, we have that
x1 cos t − y 1 sin t, x1 sin t + y 1 cos t, x2 cos t − y 2 sin t, x2 sin t + y 2 cos t .

γz (t) =
Let (U1+ , φ+ 3
1 ) be the coordinate chart for S described in Example 1.4. The coordinate
representation of γz with respect to this chart is
−1
φ+ + 3 1 1 2 2 2 2

1 ◦ γz : γz (U1 ) → R , t 7→ x sin t + y cos t, x cos t − y sin t, x sin t + y cos t .
Since γz is continuous, γz−1 (U1+ ) is open, and therefore φ+

1 ◦ γz is smooth. A similar compu-
tation shows that the coordinate representation of γz is smooth with respect to every chart
(Ui± , φ± +
i ). The velocity of γz at some t such that γz (t) ∈ U1 is
γz0 (t) = (x1 cos t − y 1 sin t, −x2 sin t − y 2 cos t, x2 cos t − y 2 sin t),
in the coordinates of U1+ . Since the first coordinate of γz (t) is x1 cos t−y 1 sin t and is positive
for γz (t) in U1+ , γ 0 (t) has positive first in U1+ and is therefore nonzero. A similar calculation
shows that γ 0 (t) is nonzero in the other charts (Ui± , φ± i ).
Problem 3.7. Let M be a smooth manifold with or without boundary and p be a point of
M . let Cp∞ (M ) denote the algebra of germs of smooth real-valued functions at p, and let
Dp M denote the vector space of derivations of Cp∞ (M ). Define a map Φ : Dp M → Tp M by
(Φv)f = v([f ]p ). Show that Φ is an isomorphism.
Let’s begin by checking that Φv is indeed a derivation at p. It is clearly linear. Let
f, g ∈ C ∞ (M ). Then
(Φv)(f g) = v[f g]p = f (p)v[g]p + g(p)v[f ]g = f (p)(Φv)g + g(p)(Φv)f,
so Φv is a derivation at p.
To check that Φ is an isomorphism, we first note that it is clearly linear. For injectivity,
let Φv = 0. Let [f ]p be the germ of some pair (f, U ), where U is open an contains p. Let B 0
be a coordinate ball contained in U and centred at p, and let B be a coordinate ball centred
at p such that B ⊆ B. Moreover, by the extension lemma for smooth functions, there is
a smooth function ψ supported in U such that ψ ≡ 1 on U . Then f˜ = ψf is a smooth
function such that [f˜]p = [f ]p . Then v[f ]p = v[f˜]p = (Φv)(f˜) = 0. Since [f ]p was arbitrary,
we conclude that v = 0, and therefore that Φ is injective.
Finally, let w ∈ Tp M be an arbitrary derivation. Define v ∈ Dp M by v[f ]p = wf . Since
wf = wg if f = g on some neighbourhood of p (Proposition 3.8), v is well defined. Then
(Φv)f = v[f ]p = wf for any f ∈ C ∞ M , so Φv = w. This proves surjectivity, and therefore
proves that Ψ is an isomorphism.
Problem 3.8. Let M be a smooth manifold with or without boundary and p ∈ M . Let
Vp M denote the set of equivalence classes of smooth curves starting at p under the relation
γ1 ∼ γ2 if (f ◦ γ1 )0 (0) = (f ◦ γ2 )0 (0) for every smooth real-valued function defined on a
neighbourhood of p. Show that the map Ψ : Vp M → Tp M defined by Ψ[γ] = γ 0 (0) is well
defined and bijective.
Let [γ1 ] = [γ2 ] for two smooth curves starting at p. Let (U, φ) be a chart containing p.
In the coordinates of this chart, we have

0 0 ∂ 0 ∂
i
γ1 (0) = (x ◦ γ1 ) (0) i
= (x ◦ γ2 ) (0) = γ20 (0),
∂xi p ∂xi p
since (f ◦ γ1 )0 (0) = (f ◦ γ2 )0 (0) for every smooth real-valued function f defined in a neigh-
bourhood of p. Hence, Ψ is well defined.
For injectivity, let γ10 (0) = γ20 (0), and let f be any smooth real-valued function defined
on a neighbourhood of p. Then

0 d d
(f ◦ γ1 ) (0) = d (f ◦ γ)0 = dfγ(0) ◦ dγ0
dt 0
dt 0
= dfγ(0) ◦ γ10 (0) = dfγ(0) ◦ γ20 (0) = (f ◦ γ2 )0 (0),
so [γ1 ] = [γ2 ], proving injectivity.
For surjectivity, let v be an arbitrary derivation at p, and let (U, φ) be some coordinate
chart containing p such that φ(p) = 0. Then we can write

∂
v = vi
∂xi p
in the coordinates of (U, φ). For some > 0, we can define a curve
γ̃ : [0, ) → U, t 7→ (tv 1 , . . . , tv n ),
and then let γ = φ−1 ◦ γ̃. Then γ is a smooth curve starting at p, since it is the composition
of smooth functions. Then
dγ i d(φi ◦ γ) d(tv 1 )

0 ∂ ∂ ∂ i ∂

γ (0) = (0) i
= (0) i
= (0) i
= v i
= v.
dt ∂x
γ(0) dt ∂x dt
γ(0) ∂x ∂x
γ(0) p
4. Submersions, Immersions, and Embeddings

Exercise 4.3. Verify the claims made in the preceding example.
For (a), let (U1 × · · · × Uk , φ1 × · · · × φk ) be a smooth chart for M1 × · · · × Mk . Then πi
has the coordinate representation
φi ◦ πi ◦ (φ1 × · · · × φk )−1 (v1 , . . . , vk ) = vi .
Let dim Mi = ni and let vi = (x1i , . . . , xni 1 ). Thus, in coordinates, d(πi )p is the (n1 + · · · +
nk ) × ni matrix (∂xji /∂xlm ), where j ranges from 1 to ni , m ranges from 1 to k, and for each
m, l ranges from 1 to nm . This matrix is of full rank, since the only nonzero terms occur
when j = l and i = m simultaneously. Hence, d(πi )p is surjective and is therefore a smooth
submersion.
For (b), suppose that the smooth curve γ is a smooth immersion. Then dγt0 is injective
for every t0 ∈ J. Then γ 0 (t0 ) = dγt0 ( d/dt|t0 ) 6= 0. Conversely, suppose γ 0 (t0 ) 6= 0 for every
t0 ∈ J. Suppose that dγt0 (v) = 0 for some v ∈ Tt0 J. Since Tt0 J is spanned by d/dt|t0 ,
we have that v = α d/dt|t0 for some α ∈ R. Then 0 = dγt0 (α d/dt|t0 ) = αdγt0 ( d/dt|t0 ) =
αγ 0 (t0 ), implying α = 0. Hence, dγt0 is injective, and therefore γ is a smooth immersion.
For (c), let (U, φ) be an open chart on M , and let (π −1 (U ), φ̃) be the corresponding
chart on on T M . Then the coordinate representation of π with respect to these charts is
φ ◦ π ◦ φ̃−1 (x, v) = x. A similar computation as the one done in (a) shows that dπ|p is a
surjection for every p ∈ T M , and therefore π is a submersion.
For (d), dX(u,v) has the matrix representation

 
−2π sin 2πu cos 2πv −2π(2 + cos 2πu) sin 2πv
dX(u,v) =  −2π sin 2πu sin 2πv 2π(2 + cos 2πu) cos 2πv  .
2π cos 2πu 0
Note that at least one of the (1, 2) or (2, 2) entries is nonzero. If 2π cos 2πu = 0, then the
bottom row is a row of zeros, and the determinant of the 2 × 2 matrix formed by the top
two rows is ±8π. Hence, the columns are linearly independent, and thus dX(u,v) has full
rank. Therefore, X is a smooth immersion.
Exercise 4.4. Show that a composition of smooth submersions is a smooth submersion,
and a composition of smooth immersion is a smooth immersion. Give a counterexample to
show that a composition of maps of constant rank need not have constant rank.
Let F : M → N and G : N → P be smooth [submersions/immersions]. Then for every
p ∈ M , dFp and dGF (p) are [surjective/injective]. Hence, dGF (p) ◦ dFp = d(G ◦ F )p is
[surjective/injective]. Therefore, G ◦ F is a smooth [submersion/immersion].
For the counterexample, let f : R → R2 , x 7→ (x, x2 ) and g : R2 → R, (x, y) 7→ y. Note
that g is a a submersion and therefore has constant rank 1. Moreover, dfx = (1 2x)> , so f
has constant rank 1. However, g ◦ f : R → R, x 7→ x2 has derivative d(g ◦ f )x = 2x, so g ◦ f
has rank 1 for each x 6= 0 and rank 0 at x = 0.
For (a), let F : M → N and G : N → P be local diffeomorphisms and let p ∈ M . Let
U ⊆ M be an open set containing p such that F |U : U → F (U ) is a diffeomorphism. Let
V ⊆ N be an open set containing F (p) such that G|V : V → G(V ) is a a diffeomorphism.
Then G ◦ F |U ∩F −1 (V ) : U ∩ F −1 (V ) → (G ◦ F )(U ) ∩ G(V ) is a diffeomorphism, where
U ∩ F −1 (V ) ⊆ M is an open set containing p. Hence, G ◦ F is a local diffeomorphism.
For (b), let Fi : Mi → Ni be a local homeomorphism for i = 1, . . . , k, and consider the
product
F1 × · · · × Fk : M1 × · · · × Mk → N1 × · · · × Nk .
Let p = (p1 , . . . , pk ) ∈ M1 × · · · × Mk and let Ui be open an open set containing pi such that
Fi |Ui : Ui → Fi (Ui ) is a diffeomorphism. By Exercise 2.16,
F1 |U1 × · · · × Fk |Uk : U1 × · · · × Uk → F1 (U1 ) × · · · × Fk (Uk )
is a diffeomorphism. Since this is the same map as
(F1 × · · · × Fk )|U1 ×···×Uk : U1 × · · · × Uk → (F1 × · · · × Fk )(U1 × · · · × Uk ),
we have that F1 × · · · × Fk is a local diffeomorphism.
For (c), let F : M → N be a local diffeomorphism. Since every point p ∈ M has an open
neighbourhood U such that F |U : U → F (U ) is a diffeomorphism. Since diffeomorphisms
are homeomorphisms, F is a local homeomorphism. Let U ∈ M be an arbitrary open set.
Let p0 ∈ F (U ) and let p ∈ U be such that F (p) = p0 . Since U is open, we can choose an open
set V such that p ∈ V ⊆ U and F |V : V → F (V ) is a diffeomorphism. Then F (V ) ⊆ F (U )
is open and contains p0 . Thus, F (U ) is open and F is an open map.
For (d), let F : M → N be a local diffeomorphism and let U ⊆ M be an open submanifold.
Let p ∈ U be arbitrary, and let V ⊆ M be an open set such that F |V : V → F (V ) is
a diffeomorphism. By Exercise 2.16, (F |V )|U ∩V = F |(U ∩V ) : U ∩ V → F (U ∩ V ) is a
diffeomorphism. Hence, F |U is a local diffeomorphism.
For (e), let F : M → N be a diffeomorphism and let p ∈ M . Note that M itself is an open
set containing p and F |M = F is a diffeomorphism. Therefore, F is a local diffeomorphism.
For (f), let F : M → N be a bijective local diffeomorphism. To see that F is smooth, note
that every point of M is contained in a chart (U, φ) such that F |U is a diffeomorphism. Then
the coordinate representation of F in the chart (U, φ) is smooth, so F is smooth. To see that
F −1 is also smooth, let F (p) ∈ N and let U ⊆ M be an open set containing p such that F |U
is a diffeomorphism. Then F −1 |F (U ) : F (U ) → U is also a diffeomorphism. Let (V, ψ) be a
smooth chart containing F (p) such that V ⊆ F (U ). Then F −1 |V is a diffeomorphism onto
its image, and therefore F −1 has a smooth coordinate representation in (V, ψ). Therefore,
F −1 is smooth and F is a diffeomorphism.
For (g), let F : M → N be a local diffeomorphism. Let p ∈ M and let U be an open
set containing p such that F |U : U → F (U ) is a diffeomorphism. We can then find a
chart (V, φ) such that V ⊆ U . Then the coordinate representation of F in (V, φ) is a
diffeomorphism, and therefore a local diffeomorphism by part (e). Conversely, suppose that
every point p ∈ M is contained in a chart (U, φ) such that F (p) is in some chart (V, ψ) and
ψ ◦ F ◦ φ−1 : φ(U ∩ F −1 (V )) → ψ(F (U ∩ F −1 (V ))) is a local diffeomorphism. Since F is
smooth (coordinate representations are only defined for smooth maps), φ(U ∩ F −1 (V )) is
open in Rn (or Hn ). Since local diffeomorphisms are open maps, ψ(F (U ∩F −1 (V ))) is open in
Rn (or Hn ). Since the restriction of a local diffeomorphism to an open submanifold is a local
diffeomorphism, (ψ ◦ F ◦ φ−1 ) ◦ φ|U ∩F −1 (V ) = ψ ◦ F |U ∩F −1 (V ) . Similarly, ψ −1 |ψ(F (U ∩F −1 (V )))
is a local diffeomorphism, and therefore F |U ∩F −1 (V ) : U ∩ F −1 (V ) → F (U ∩ F −1 (V )) is a
local diffeomorphism. Then there is an open set W ⊆ U ∩ F −1 (V ) containing p such that
F |W : W → F (W ) is a diffeomorphism. Hence, F is a local diffeomorphism.
Exercise 4.9. Show that the conclusions of Proposition 4.8 still hold if N is allowed to be
a smooth manifold with boundary, but not if M is.
To show that the result holds when N has nonempty boundary, we just need to show that
we can apply the inverse function theorem for manifolds when F is a smooth immersion and
a smooth submersion. To do this, it suffices to show that F (M ) ⊆ IntN . This is the result
of 4.2.
The inclusion map Hn ,→ Rn is both a smooth immersion and a smooth submersion (it’s
differential is given by the identity). On the other hand, it is not a local diffeomorphism: if
U ⊆ Hn is an open neighbourhood of 0, then F (U ) is not open in Rn . In particular, F (U )
contains no open ball centred at 0.
Exercise 4.10. Suppose M, N, P are smooth manifolds with or without boundary, and
F : M → N is a local diffeomorphism. Prove the following:
(a) If G : P → M is continuous, then G is smooth if and only if F ◦ G is smooth.
(b) If in addition F is surjective and G : N → P is any map, then G is smooth if and
only if G ◦ F is smooth.
For (a), let G be smooth. Since local diffeomorphisms are smooth, F ◦ G is smooth.
Conversely, suppose that F ◦ G is smooth. Let p ∈ P , and let U ⊆ M be an open set
containing g(p) such that F |U : U → F (U ) is a diffeomorphism. Since G is continuous,
G−1 (U ) is open, and therefore (F ◦ G)|G−1 (U ) : G−1 (U ) → F (U ) is smooth. Then (F |U )−1 ◦
(F ◦ G)|G−1 (U ) = G|G−1 (U ) : G−1 (U ) → U is smooth. Thus, G is smooth.
For (b), let G be smooth. Since local diffeomorphisms are smooth, G ◦ F is smooth.
Conversely, suppose that G ◦ F is smooth. Let q ∈ N be arbitrary, and let F (p) = q
(using the fact that F is surjective). Let U ⊆ M be an open neighbourhood op p such that
F |U : U → F (U ) is a diffeomorphism. Then G◦F |U is smooth, and so is (G◦F |U )◦(F |U )−1 =

G|F (U ) : F (U ) → N . Thus, G is smooth.
Exercise 4.17. Show that every composition of smooth embeddings is a smooth embedding.
The composition of smooth immersions is a smooth immersion (Exercise 4.4). It remains
to show that a composition of topological embeddings is a topological embedding. Let
F : M → N and G : N → P be topological embeddings. Then F : M → F (M ) and G : N →
G(N ) are homeomorphisms. Then G|F (M ) : F (M ) → G ◦ F (M ) is a homeomorphism, so
G ◦ F : M → G ◦ F (M ) is a homeomorphism. Therefore G ◦ F is a topological embedding.
Exercise 4.24. Give an example of a smooth embedding that is neither an open map nor
a closed map.
Let F : R → R2 be given by F (t) = (et , 0). Then F is a continuous bijection onto
(0, ∞) × {0} and F −1 (x, 0) = log x for x > 0. The inverse is continuous, so we have that
F is a topological embedding. Next, dFt = (et 0) is of full rank for all t, so F is a smooth
immersion and therefore is a smooth embedding. Note that F (R) = (0, ∞) × {0} is neither
open nor closed in R2 despite the fact that R is both open and closed in R. Hence, F is
neither an open nor a closed map.
Exercise 4.27. Give an example of a smooth map that is a topological submersion but not
a smooth submersion.
Let F : R → R be given by F (x) = x3 . Then F is smooth and bijective with continuous
inverse F −1 (x) = x1/3 . Therefore F is a topological submersion. However, dF0 = 0, so F is
not a smooth submersion.
Exercise 4.32. Prove Theorem 4.31.
By Theorem 4.30, there are smooth maps F : N1 → N2 and G : N2 → N1 such that
F ◦ π2 = π1 and G ◦ π1 = π2 . Then G ◦ F ◦ π2 = π2 . Since π2 is surjective, G ◦ F = Id. A
similar argument shows that F ◦ G = Id. Hence, G = F −1 and F is a diffeomorphism.
M
π2 π1
N1 N2
G

For (a), let π : E → M be a smooth covering map. Let p ∈ M and let q ∈ E such
that π(q) = p. There is a neighbourhood U of p such that V ⊆ E is the component
of π −1 (U ) containing q. Then π|V : V → U is a diffeomorphism, meaning π is a local
diffeomorphism. By Proposition 4.8, π is also a smooth submersion (the fact that E and M
can have nonempty boundary is not a problem for this direction of the proof of Proposition
4.8). Since local diffeomorphisms are open maps, π is an open map. Surjective open maps
are quotient maps, so π is a quotient map.
For (b), if π is injective, then it is bijective. Since bijective local diffeomorphisms are
diffeomorphisms, π is a diffeomorphism.
For (c), let π : E → M be a topological covering map that is also a smooth covering map.
By part (a), π is also a local diffeomorphism. Conversely, let π : E → M be a topological
covering map that is also a local diffeomorphism. Let p ∈ M and let U be an evenly covered
neighbourhood of p, and let π −1 (U ) = α∈A Uα , where π|Uα is a homeomorphism onto its
S
image. Then π|Uα is a bijective local diffeomorphism, and therefore a diffeomorphism.
Exercise 4.37. Suppose π : E → M is a smooth covering map. Show that every local section
of π is smooth.
Let U ⊆ M be open, and let σ : U → E be a local section. Note that every p ∈ U is
contained in an evenly covered open neighbourhood Up ⊆ U . By Proposition 4.36, there is
a unique smooth local section τ : Up → E of π such that τ (p) = σ(p). Then π ◦ σ|Up = π ◦ τ ,
and since π is bijective on π −1 (Up ), we have that σ|Up = τ . Therefore, σ is smooth.
Exercise 4.38. Suppose E1 , . . . , Ek and M1 , . . . , Mk are smooth manifolds (without bound-
ary), and πi : Ei → Mi is a smooth covering map for each i = 1, . . . , k. Show that
π1 × · · · × πk : E1 × · · · × Ek → M1 × · · · × Mk is a smooth covering map.
We already have that π1 × · · · × πk is smooth and surjective. Let pi ∈ Mi be an arbitrary
point, and let Ui be an evenly covered neighbourhood of pi for each i = 1, . . . , k. Let Ui,αi
be a component of π −1 (Ui ) for each i = 1, . . . , k. Since the product of diffeomorphisms is a
diffeomorphism, we have that
(π1 )|U1,α1 × · · · × (πk )|Uk,αk : U1,α1 × · · · × Uk,αk → U1 × · · · × Uk
is a diffeomorphism. Since the connected components of (π1 × · · · × πk )−1 (U1 × · · · × Uk )
are all of the form U1,α1 × · · · × Uk,αk , this shows that π1 × · · · × πk is a smooth covering
map.
Exercise 4.39. Suppose that π : E → M is a smooth covering map. Since π is also a
topological covering map, there is a potential ambiguity about what it means for a subset U ⊆
M to be evenly covered: does π map the components of π −1 (U ) diffeomorphically onto U , or
merely homeomorphically? Show that the two concepts are equivalent: if U ⊆ M is evenly
covered in the topological sense, then π maps each component of π −1 (U ) diffeomorphically
onto U .
Let V be a component of π −1 (U ). Note that every p ∈ U has an open neighbourhood Up ⊆
U such that π|V ∩π−1 (Up ) : V ∩ π −1 (Up ) → Up is a diffeomorphism. Hence, π|V is a bijective
local diffeomorphism, and therefore a diffeomorphism. Hence, π maps the components of
π −1 (U ) diffeomorphically onto U .
The facts that E is second-countable and Hausdorff are proven in exactly the same
way as in the proof of Proposition 4.40. To show that E is a topological n-manifold with
boundary, let p ∈ E and first suppose that π(p) ∈ ∂M . Then there is an evenly covered
open neighbourhood U of π(p) such that (U, φ) is also a boundary chart for M . Letting
V be the component of π −1 (U ) containing p, we find that (V, φ ◦ π|V ) is a boundary chart
containing p. We can similarly construct an interior boundary chart of any point p ∈ E
mapping to IntM . Therefore, E is a topological n-manifold with boundary.
Let p ∈ ∂E and let U ⊆ M be an evenly covered open neighbourhood of π(p), such that
(U, φ) is a chart with domain U . As described above, we obtain a chart (V, φ◦π|V ) containing
p. Since p is a boundary point, φ(π(p)) ∈ ∂Hn , which implies that π(p) is a boundary point
of M . Then p ∈ π −1 (π(p)) ⊆ π −1 (∂M ). Conversely, suppose that p ∈ π −1 (∂M ). Let U
be an evenly covered open neighbourhood of π(p) that is also the domain of a chart (U, φ).
Then φ(π(p)) ∈ ∂Hn , which implies that p is a boundary point of the chart (V, φ ◦ π|V ), and
therefore of E. Thus, we have proven ∂E = π −1 (∂M ).
To prove the existence of a unique smooth structure on E making π into a smooth
covering map, we need to show that the charts (V, φ ◦ π|V ) constructed above are smoothly
compatible. This is done in the same way as in the proof of Proposition 4.40. The proof of
uniqueness will be given in Problem 4.9.
Exercise 4.44. Prove Corollary 4.43.
We know that any connected and locally simply connected space admits a unique universal
cover. Since topological manifolds admit bases of coordinate balls, they are locally simply
connected. Thus, if M is a connected (smooth) manifold, it admits a topological covering
map π : Mf → M where M̃ is simply connected. By Proposition 4.40, M f has a unique
smooth structure making π into a smooth covering map. This proves the existence of a
smooth covering manifold of M .
For uniqueness, suppose that π c → M is another smooth covering of M by a simply
b: M
connected space M . By uniqueness of topological universal coverings, there is a homeomor-
f
phism φ making the following diagram commute
φ
M
f M
c
.
π π
b
M
Every point of M has a neighbourhood that is evenly covered by π and π b (by intersecting
two possibly different evenly covered sets). Let V ⊆ M be a component of π −1 (U ). Then
φ(V ) is a component of πb−1 (U ), and we have the following commutative diagram
U
(π|V )−1 π |φ(V ) )−1
(b
.
φ|V
V φ(V )
−1
By Theorem 4.29, since (π|V ) is a surjective smooth submersion (it is a diffeomorphism)
φ|V ◦ (π|V )−1 is smooth if and only if φ|V is smooth. Since φ|V ◦ (π|V )−1 = (bπ |φ(V ) )−1
is smooth, we have that φ|V is smooth. Since M is covered by sets of the form V , φ is
smooth. A similar argument shows that φ−1 is smooth, so φ is a diffeomorphism, which
proves uniqueness of the universal covering manifold of M .
Exercise 4.45. Generalize the preceding corollary to smooth manifolds with boundary.
By Proposition 4.41, every smooth manifold with boundary M admits a universal covering
π: Mf → M where M f is a smooth manifold with boundary.
We will prove uniqueness in a simpler way than we did in Exercise 4.44 that does not
require the use of Theorem 4.29 (which is not stated for manifolds with boundary). Again,
if there is another universal covering π c → M , then there is a homeomorphism φ making
b: M
φ
M
f M
c
.
π π
b
M
commute. Let U ⊆ M be an evenly covered neighbourhood, and let V be a component of
π −1 (U ). Then φ(V ) is a component of π
b−1 (U ), and we have that π
b|φ(V ) ◦ φ|V = π|V . Since
b|φ(V ) and π|V are diffeomorphisms, we have that φ|V is a diffeomorphism. Then φ is a
π
bijective local diffeomorphism, and therefore is a diffeomorphism. This concludes the proof
of uniqueness.
Problem 4.1. Use the inclusion map Hn ,→ Rn to show that Theorem 4.5 does not extend
to the case in which M is a manifold with boundary.
Denote the inclusion by i : Hn ,→ Rn . Then i is smooth and di(0,0) = Id is invertible.
Suppose there are connected open neighbourhoods U0 ⊆ Hn of 0 and V0 ⊆ Rn of 0 such
that i|U0 : U0 → V0 is a diffeomorphism. But U0 is a manifold with nonempty boundary,
while V0 has empty boundary. This shows that Theorem 4.5 (the inverse function theorem
for manifolds) does not hold when the domain is a manifold with boundary.
Problem 4.2. Suppose M is a smooth manifold (without boundary), N is a smooth manifold
with boundary, and F : M → N is smooth. Show that if p ∈ M is a point such that dFp is
nonsingular, then F (p) ∈ IntN .
Suppose there is a p ∈ M such that F (p) ∈ ∂N . Let (U, φ) be a chart containing p and
let (V, ψ) be a boundary chart containing F (p). Suppose that the coordinates of U are xi
for i = 1, . . . , m and those of V are y i (x1 , . . . , xm ) for i = 1, . . . , n. Then y n (x1 , . . . , xn ) has
a local minimum at (x1 , . . . , xm ) = φ(p). Therefore, ∂y n /∂xi |φ(p) = 0 for each i = 1, . . . , m,
which implies that the bottom row of the matrix representation of dFp is a row of zeros.
Hence, dFp is singular, a contradiction.
Problem 4.3. Formulate and prove a version of the rank theorem for a map of constant
rank whose domain is a smooth manifold with boundary.
Formulation of the “domain manifold with boundary” version of the rank theorem: let
M be a smooth manifold with boundary and let N be a smooth manifold of dimensions m
and n, respectively, and let F : M → N be a smooth map with constant rank r. For each
p ∈ M there exist smooth charts (U, φ) for M centred at p and (V, ψ) for N centred at F (p)
such that F (U ) ⊆ V , in which F has a coordinate representation of the form
Fb(x1 , . . . , xr , xr+1 , . . . , xm ) = (x1 , . . . , xr , 0, . . . , 0).
We now begin the proof of this version of the rank theorem. After choosing smooth
coordinates, we can replace M and N by open subsets U ⊆ Hm and V ⊆ Rn . Moreover, we
may assume that U intersects ∂Hm , otherwise the proof is the same as the one given in the
text. By translating, we may assume that p = 0 and F (p) = 0. Since F is smooth, it has
an extension Fe : Ue → V , where U ⊆ U e is open in Rm . We will now show that there
e and U
is a projection π : Rn → Rr such that π ◦ Fe is a submersion. Since F has constant rank r,
the matrix (∂ Fei /∂xj ) has an r × r submatrix with nonzero determinant at (x1 , . . . , xm ) = 0
(assume it is the upper-left submatrix by reordering coordinates). Therefore, by shrinking U e
if necessary, we may assume that this submatrix has non-vanishing determinant on all of U e.
Let π : Rn → Rr be the projection onto the first r coordinates; then π ◦ Fe is a submersion.
By the usual rank theorem, for each q ∈ U e , there are charts (U0 , φ0 ) for U
e centred at q
and (W, α) for π(V ) centred at π ◦ F (q) such that π ◦ F (U0 ) ⊆ W and
e e
α ◦ π ◦ Fe ◦ φ−1 1 r
0 (x , . . . , x , x
r+1
, . . . , xm ) = (x1 , . . . , xr ).
Note that α ◦ π ◦ Fe ◦ φ−1 −1
0 = π ◦ (α × idRn−r ) ◦ F ◦ φ0 , so
e
(α × idRn−r ) ◦ Fe ◦ φ−1 1 r
0 (x , . . . , x , x
r+1
, . . . , xm ) = (x1 , . . . , xr , R(x1 , . . . , xr , xr+1 , . . . , xm ))
for some smooth function R defined on φ0 (U0 ). By shrinking U0 if necessary, we may assume
that φ0 (U0 ) is an open cube. By the same argument given in the proof of the usual rank
theorem, R|φ0 (U0 ∩U ) is independent of xr+1 , . . . , xm (this uses that F has constant rank r).
Define the open set
V0 = {(y 1 , . . . , y r , y r+1 , . . . , y n ) ⊆ (W × Rn−r ) ∩ V : (y 1 , . . . , y r , 0, . . . , 0) ⊆ φ0 (U0 )}.
Moreover, define ψ0 : V0 → Rn by ψ0 (v, w) = (v, w − R(v, 0)), where v = (y 1 , . . . , y r ) and
w = (y r+1 , . . . , y n ). Note that ψ0 is a diffeomorphism onto its image since it has inverse
given by ψ0−1 (v, w) = (v, w + R(v, 0)). Therefore, (V0 , ψ0 ) is a chart for V centred at 0.
Note that on φ0 (U0 ∩ U ) we have
ψ0 ◦ (α × idRn−r ) ◦ F ◦ φ−1 1 r
0 (x , . . . , x , x
r+1
, . . . , xm ) = (x1 , . . . , xr , 0, . . . , 0)
since R is independent of xr+1 , . . . , xm there.
What remains is to show how (U0 , φ0 ) can be made into a boundary chart for U . Simple
restriction will not work here and an idea similar to the one used in the proof of Theorem
4.15 will not work, since we do not have that F is an immersion. Unfortunately, I don’t
know how to do this step.
Problem 4.4. Let γ : R → T2 be the curve of Example 4.20. Show that the image set γ(R)
is dense in T2 .
We claim that for every t ∈ R, the set {e2πiα(t+n) : n ∈ Z} is dense in S1 . Since multipli-
cation by e2πiαt defines a homeomorphism of S1 , it is enough to show that {e2πiαn : n ∈ Z}
is dense in S1 . It is then enough to show that for every N ∈ N and every k = 0, . . . , N − 1,
k k+1
there is an n ∈ Z such that bαnc ∈ ( N , N ), or equivalently, that there exists an m ∈ Z
such that αn − m ∈ ( N , N ). By Dirichlet’s approximation theorem, there are m0 , n0 ∈ Z
k k+1
such that |αn0 − m0 | < N1 . If αn0 − m0 ∈ (− N1 , 0), replace m0 and n0 by −m0 and −n0 ,
respectively, so that αn0 − m0 ∈ (0, N1 ). Note that we are using the fact that α is ir-
rational here to justify our use of open intervals. Then, for some l ∈ N, we have that
l(αn0 − m0 ) = α(ln0 ) − (lm0 ) ∈ ( N , N ). Choose m = lm0 and n = ln0 .
k k+1
1 1
Now let U ⊆ S × S be open. We may then assume that U is of the form U1 × U2 , where
Ui ⊆ S1 is open. Choose t ∈ R such that e2πit ∈ U . By our previous work, there is an n ∈ Z
such that e2πiα(t+n) ∈ U2 . Hence,
γ(t + n) = (e2πi(t+n) , e2πiα(t+n) ) = (e2πit , e2πiα(t+n) ) ∈ U1 × U2 .
Hence, γ(R) intersects every open set in T2 , implying that γ(R) is dense in T2 .
Problem 4.5. Let CPn denote the n-dimensional complex projective space, as define in 1.9.
(a) Show that the quotient map π : Cn+1 r{0} → CPn is a surjective smooth submersion.
(b) Show that CP1 is diffeomorphic to S2 .
For (a), let U ek = {(z 1 , . . . , z n+1 ) ∈ Cn+1 r {0} : z k 6= 0} and let Uk = π(U
ek ). Let
1 n+1
(z , . . . , z ) ∈ Uk . Then, with respect to the charts (Uk , id) and (Uk , φk ) (see Problem
e e
1.9), the coordinate representation of π is
1
z k−1 z k+1 z n+1

1 n+1 z
φk ◦ π(z , . . . , z ) = ,..., k , k ,..., k ,
zk z z z
which is smooth since z k 6= 0. Hence, π is smooth and surjective. It remains to show that
π is a submersion. Letting z j = xj + iy j , we have that
zj xj xk + y j y k xk y j − xj y k
k
= k 2 k 2
+i k 2 .
z (x ) + (y ) (x ) + (y k )2
In the coordinates of Ui , one can then check that dπ has a 2n × 2n submatrix of the form
 
D 0 ··· 0
0 D ··· 0
A = . ..  ,
 
.. . .
 .. . . .
0 0 ··· D
where
xk −y k
!
(xk )2 +(y k )2 (xk )2 +(y k )2
D = yk xk
.
(xk )2 +(y k )2 (xk )2 +(y k )2
Then det A = 1, so dπ is of full rank, and therefore π is a submersion.
For (b), we define a diffeomorphism F : S2 → CP1 as follows:
(
x y
[1, 1−z + i 1−z ] if (x, y, z) ∈ S2 r {N }
F (x, y, z) = x y
[ 1+z − i 1+z , 1] if (x, y, z) ∈ S2 r {S}.
We check that F is well defined: if (x, y, z) ∈ S2 r {N, S}, then

x y x y x y x y
1, +i = −i 1, +i = −i ,1 .
1−z 1−z 1+z 1+z 1−z 1−z 1+z 1+z
Next, note that F |S2 r{N } = φ−1 2 ◦ i ◦ σ, where φ2 is one of the standard coordinate maps
1
defined on CP , i is the identification of R2 with C, and σ is the stereographic projection
from the north. Since each of these are diffeomorphisms (i is by definition essentially), we
have that F |S2 r{N } is a diffeomorphism onto its image. Similarly, F |S2 r{S} = φ−1
1 ◦ τ ◦ i ◦ σ̃,
where τ is complex conjugation and σ̃ is the stereographic projection from the south. Thus,
F |S2 r{S} is also a diffeomorphism onto its image. Since both restrictions are defined on
open sets and they agree on the overlap, we have that F is a diffeomorphism onto its image.
It remains to show that F is surjective. This is straightforward, since the image of F |S2 r{N }
is U1 , the image of F |S2 r{S} is U2 , and CP1 = U1 ∪ U2 . Hence, CP1 is diffeomorphic to S1 .
Problem 4.6. Let M be a nonempty smooth compact manifold. Show that there is no
smooth submersion F : M → Rk for any k > 0.
By Proposition 4.28, if F is a smooth submersion then it is an open map. Since M is
open in itself, F (M ) is open. Since M is compact and F is continuous, F (M ) is compact.
Since open sets in Rk are not compact, this is a contradiction. Hence, there is no smooth
submersion from M to Rk . Note that this result can easily be generalized: if M is a compact
topological space, then there is no open map F : M → Rk for k > 0.
Problem 4.7. Suppose M and N are smooth manifolds, and π : M → N is a surjective
submersion. Show that there is no other smooth manifold structure on N that satisfies the
conclusion of Theorem 4.29; in other words, assuming that N e represents the same set as N
with a possibly different topology and smooth structure, and that for every smooth manifold
P with or without boundary, a map F : N e → P is smooth if and only if F ◦ π is smooth,
show that IdN is a diffeomorphism between N and N e.
Let π̃ : M → N e be the same smooth surjective submersion as π, but viewed as a map

e . Note that π : M → N is equal to Id−1 ◦ π
into N e and is smooth. Therefore Id−1
N N is smooth.
Note that IdNe : N → N is smooth, and therefore that IdNe ◦ π̃ = π̃ is smooth. Then IdN is
e e
smooth, since IdN ◦ π = π̃ is smooth. This shows that IdN is a diffeomorphism.
Problem 4.8. This problem shows that the converse of Theorem 4.29 is false. Let π : R2 →
R be defined by π(x, y) = xy. Show that π is surjective and smooth, and for each smooth
manifold P , a map F : R → P is smooth if and only if F ◦ π is smooth; but π is not a smooth
submersion.
Since π(x, 1) = x for all x ∈ R, we have that π is surjective. Moreover, π is smooth since
it is a polynomial. Suppose that F : R → P is smooth; then F ◦ π is smooth by composition.
Conversely, suppose that F ◦ π is smooth. Since f : R → R2 , x 7→ (x, 1) is smooth, we have
that F ◦ π ◦ f = F is smooth by composition. To see that π is not a submersion, note that

y
dπ(x,y) = ,
x
which has rank 0 at (0, 0).
Problem 4.9. Let M be a connected smooth manifold, and π : E → M be a topological
covering map. Complete the proof of Proposition 4.40 by showing that there is only one
smooth structure on E such that π is a smooth covering map.
We will prove this for manifolds with boundary as well. We already know that E has a
smooth structure making π : E → M a smooth covering map. Suppose that E e is the same
topological manifold as E, but with a possibly different smooth structure making π̃ : Ee→M
into a smooth covering map, where π̃ is the same map as π, just viewed as a map whose
domain is E. e We will prove that the identity map Id : E → E
e is a diffeomorphism; it suffices
to show that Id is a local diffeomorphism since it is obviously bijective.
Every point p ∈ E is in the preimage of some evenly covered V ⊆ M . Let U be the
component of π −1 (U ) containing p. Since there is no distinction between being smoothly
evenly covered and topologically evenly covered (Exercise 4.39), Id(U ) is also a component
of π̃ −1 (V ), and there is a smooth section σ : V → Id(U ) of π̃. Thus, Id|U = σ ◦ π|U is a
diffeomorphism, so Id is a local diffeomorphism, proving our claim.
Problem 4.10. Show that the map q : Sn → RPn defined in Example 2.13(f ) is a smooth
covering map.
As discussed in Example 2.13(f), we have that q is smooth. By our solution to Exercise
1.7, we have that q is surjective as well. Consider the usual smooth charts (Ui , φi ) for RPn
and the smooth charts (Vi± , ψi± ) for Sn (see Example 1.4) for i = 1, . . . , n + 1. It is not
too hard to see that q −1 (Ui ) = Vi+ t Vi− . Hence, if we can show that q|V ± : Vi± → Ui is a
i
diffeomorphism, we will be done. In coordinates, we have
(x1 , . . . , xn )
φi ◦ q|V ± ◦ (ψi± )−1 (x1 , . . . , xn ) = ± q .
i 2
1 − |x|
Then φi ◦ q|V ± ◦ (ψi± )−1 : Bn → Rn is a smooth map with smooth inverse f : Rn → Bn given
i
by
(x1 , . . . , xn )
f (x1 , . . . , xn ) = q .
2
|x| + 1
Hence, φi ◦ q|V ± ◦ (ψi± )−1 is a diffeomorphism, which means that q|V ± is a diffeomorphism.
i i
Problem 4.11. Show that a topological covering map is proper if and only if its fibres are
finite, and therefore the converse of Proposition 4.46 is false.
Let π : E → M be a topological covering map. Suppose that π is proper. Then {p} is

compact for every p ∈ M , which implies that π −1 (p) is compact in E. Let U be an evenly
covered neighbourhood of p. Then each component of π −1 (U ) contains exactly one point of
π −1 (p) since each component maps homeomorphically onto U . This collection of components
then forms an open cover for π −1 (p), and must therefore be finite by compactness. Thus,
π −1 (p) is finite.
Conversely, suppose that π −1 (p) is finite for each p ∈ M . Let K ⊆ M be compact and
let {Uα }α∈A be an open covering of π −1 (K). Let p ∈ K be arbitrary, and let π −1 (p) =
{q1p , . . . , qkpp }. By the fact that π is an open map (topological covering maps are open)
and π −1 (p) is finite, we can choose open neighbourhoods Vip of qip such that Vip ⊆ Uα for
each i = 1, . . . , kp and for some α ∈ A, and such that π|Vip is a homeomorphism onto its
image W p = π(Vip ). By compactness of K, let {W p1 , . . . , W pn } be an open covering of K.
We claim that collection of components of all the π −1 (W pi ) forms a finite open covering
of π −1 (K). To see this, let q ∈ π −1 (K), so that π(q) ∈ K. Then π(q) ∈ W pj for some
j ∈ {1, . . . , n}. Then q is in one of the components of π −1 (W pj ), as claimed. The fact that
p
the covering is finite follows from the fact that the fibres are finite. For each component Vi j
of W pj , choose a Uα that contains it. This finite subcollection of {Uα }α∈A is then a finite
subcover of π −1 (K). Hence, π −1 (K) is compact, which proves that π is proper.
Since : R → S1 defined in Example 2.13(b) is a smooth covering map with (countably)
infinite fibres, it is not proper. Hence, the converse of Proposition 4.46 is false.
Problem 4.12. Using the covering map 2 : R2 → T2 (see Example 4.35), show that the
immersion X : R2 → R3 defined in Example 4.2(d) descends to a smooth embedding of T2
into R3 . Specifically, show that X passes to the quotient to define a smooth map X
e : T2 →
3
R , and then show that X is a smooth embedding whose image is the given surface of
e
revolution.
It is straightforward to check that X is constant on the fibres of 2 . Since 2 is a smooth
submersion, Theorem 4.30 guarantees that there is a smooth map X e making
R2
X
2
T2 R3
X
e
commute. Because 2 is actually a local diffeomorphism (Example 4.11) and X is a smooth

immersion, it follows that X e must be a smooth immersion. More specifically, at each p ∈ R2 ,
we have dX2 (p) ◦ dp = dXp . Since d2p is invertible and dXp is injective, it follows that
e 2
dXe2 (p) is injective for every p; surjectivity of 2 then implies that X e is an immersion. We
2πis 2πis e 2πit1 , e2πit2 ).
will now show that X e is an injection. To this end, let X(e
e 1
,e 2
) = X(e
Then, by the definition of X,
(2 + cos 2πs1 ) cos 2πs2 = (2 + cos 2πt1 ) cos 2πt2
(2 + cos 2πs1 ) sin 2πs2 = (2 + cos 2πt1 ) sin 2πt2
sin 2πs1 = sin 2πt1 .
Since 2 + cos x is never zero, the first two equations imply that cos 2πs2 = cos 2πt2 and
sin 2πs2 = sin 2πt2 . Since cos x and sin x are never zero for the same x, at least one of
the first two equations yields 2 + cos 2πs1 = 2 + cos 2πt1 , which gives cos 2πs1 = cos 2πt1 .
Putting this together with the third equation, we have that (e2πis1 , e2πis2 ) = (e2πit1 , e2πit2 ),
which proves injectivity.
Since T2 is compact and X e is an injective smooth immersion, Proposition 4.22 implies
that Xe : T2 → R3 is an embedding.
Problem 4.13. Define a map F : S2 → R4 by F (x, y, z) = (x2 − y 2 , xy, xz, yz). Using
the smooth covering map of Example 2.13(f ) and Problem 4.10, show that F descends to a
smooth embedding of RP2 into R4 .
Since F = G ◦ ι, where G : R3 → R4 is the map given by the same formula as F but

extended to all of R3 and ι : S2 ,→ R3 is the inclusion, we have that F is smooth. Note that
q : S2 → RP2 is a smooth submersion and that F is constant on the fibres q −1 (x, y, x) =
{±(x, y, z)}. By Theorem 4.30, there is then a map Fe such that
S2
q F
RP2 R4
F
e
commutes. The commutativity of the diagram forces

1
Fe[x, y, z] = (x2 − y 2 , xy, xz, yz).
x2 + y 2 + z 2
In the coordinates of the usual chart (U1 , φ1 ), we have

1
Fe ◦ φ−1
1 (u, v) = F [1, u, v] =
e (1 − u2 , u, v, uv),
1 + u2 + v 2
and the matrix representation of dFep for p ∈ U1 is then
−4u − 2uv 2 1 − u2 + v 2 v − uv +3

1 −2uv
.
(1 + u2 + v 2 )2 2u2 v − 2v −2uv 1 + u2 − v 2 u + u3 − uv 2
The first minor is (up to a nonzero factor)

−4u − 2uv 2 1 − u2 + v 2
2 = 2v(1 + 2u2 + v 2 + u4 + u2 v 2 ),
2u v − 2v −2uv
which can only be zero if v = 0. But plugging v = 0 into the matrix representation of dFe
shows that the rows are linearly independent. Thus, Fe is an immersion on U1 . Similar
computations show that Fe is an immersion on the other charts.
We now show that Fe is injective. To this end, let Fe[x, y, x] = Fe[u, v, w]. We may assume
that (x, y, z) and (u, v, w) are normalized. Then
x2 − y 2 = u2 − v 2 , xy = uv, xz = uw, yz = vw.
Using these equations along with x2 + y 2 + z 2 = u2 + v 2 + w2 = 1, it can be shown that

(x, y, z) = ±(u, v, w), and therefore that [x, y, z] = [u, v, w]. Then Fb is an injective smooth
submersion with compact domain, and is therefore an embedding. Hence, Fe : RP2 → R4 is
an embedding.
5. Submanifolds
Exercise 5.10. Show that spherical coordinates (Example C.38) form a slice chart for S2
in R3 on any open subset where they are defined.
Let ψ : U → R3 be spherical coordinates defined on some open set of R3 . Then ψ(U ∩S2 ) =
{(x, y, z) ∈ U : z = 1}. Thus, spherical coordinates define form a slice chart for S2 in R3 .
Exercise 5.20. Suppose M is a smooth manifold and S ⊆ M is an immersed submanifold.
Show that every subset of S that is open in the subspace topology is also open in its given
manifold topology; and the converse is true if and only if S is embedded.
Let U ⊆ S be open in the subspace topology. Then there is an open set V ⊆ M such that
U = S ∩ V . If ι : S ,→ M is the inclusion map, then U = ι−1 (U ) = ι−1 (S ∩ V ) = ι−1 (V ) is
open, since the inclusion is a smooth immersion and therefore continuous.
Suppose that every subset open in the manifold topology of S is also open in the subspace
topology of M . Then ι(U ) is open in S (as a subspace of M ) for every U open in the
manifold topology of S. Hence, ι is a topological embedding and a smooth submersion, and
is therefore a smooth embedding. Hence, S is embedded. The proof of the converse follows
the same steps in reverse order.
Assume that X : U → M is a smooth local parametrization of S. Let V = ι−1 (X(U )),
which is open in S, and let φ = X −1 ◦ ι : V → U ⊆ Rk . Then φ is a diffeomorphism since
X is a diffeomorphism onto its image (with respect to S’s smooth manifold structure), and
therefore (V, φ) is a smooth chart for S. Moreover, X = ι ◦ φ−1 by definition.
Conversely, suppose that (V, φ) is a smooth chart for S, and let X = ι ◦ φ−1 . Then X is
a diffeomorphism onto its image with respect to the smooth manifold topology on S, since
φ−1 is a diffeomorphism. So X is a smooth local parametrization of S.
Since every point of S is in the image of φ−1 for some smooth chart (V, φ), the last claim
of the proposition follows directly from the work above.
Let dim M = n and dim S = k. Let v ∈ Tp M and suppose that v ∈ Tp S. By Proposition
5.22, there is an open neighbourhood V of p such that ι|V : V ,→ M is an embedding, and
by Theorem 5.8, there is a smooth chart (U, φ) for M centred at p such that V ∩ U is a
k-slice in U . Letting (x1 , . . . , xn ) be the coordinates, we have

∂ n ∂
v = v1

1
+ · · · + v n
∂x p
∂x
p
for some v i ∈ R. On one hand, we have v(xi ) = v(xi |S ) = 0 if i = k + 1, . . . , n (we are

viewing the xi as the coordinate functions of φ here). On the other,
!
∂ ∂
v1 + · · · + vn (xi ) = v i ,

∂x1 p ∂xn p
so we have
∂
1 k ∂

v = v + ··· + v .
∂x1 p ∂xk p
Let γ̃ : J → φ(U ) be the curve defined by γ̃(t) = (v 1 t, . . . , v k t, 0, . . . , 0) for some interval
J ⊆ R containing 0. Letting γ = φ−1 ◦ γ̃, we have that the image of γ is in S, γ(0) = p, and
γ 0 (0) = v. In fact, the image of γ is contained in V , which is embedded in M . Corollary

5.30 implies that γ : J → V is also smooth. Since V is an open submanifold of S, we have
that γ : J → S is smooth.
Conversely, suppose there is a smooth curve γ : J → M such that γ(J) ⊆ S, γ is smooth
as a map into S, 0 ∈ J, γ(0) = p, and γ 0 (0) = v. Let γ̃ : J → S be the curve as γ, but
viewed as mapping into S. Then, for a smooth function f , we have

d
dιp (γ̃ 0 (0))f = γ̃(0)(f ◦ ι) = dγ̃ (f ◦ ι)
dt 0

d d
= (f ◦ ι ◦ γ̃) = (f ◦ γ) = γ 0 (0)(f ) = vf.
dt 0 dt 0
Hence, v = dιp (γ̃ 0 (0)) ∈ Tp S.
Exercise 5.40. Suppose S ⊆ M is a level set of a smooth map Φ : M → N with constant
rank. Show that Tp S = Ker dΦp for each p ∈ S.
By Proposition 5.12, S is a properly embedded submanifold of M of codimension r, where
r is the rank of Φ. The proof that Tp S ⊆ Ker dΦp is the same as the one given in Proposition
5.38. By the rank-nullity,
dim Ker dΦp = dim Tp M − dim Im dΦp = dim Tp M − r = dim Tp S.
Hence, Tp S = Ker dΦp .
This uses Theorem 5.11, which we prove in Problem 5.2. Let (U, φ) be a smooth boundary
chart containing p. Let v ∈ Tp ∂M be a vector tangent to ∂M . What this means is that
there is a a vector w ∈ Tp ∂M such that v = dιp (w), where ι : ∂M ,→ M is the inclusion.
Note that (φ−1 (∂Hn ), φ) is a smooth chart ∂M containing p. We will denote this chart by
(V, ψ). Note that, for every function f , if
n−1 !
X ∂
v i d(ψ −1 )ψ(p)

w = ,
i=1
∂xi ψ(p)
then
!
−1 ∂ i ∂
i
(f ◦ ι ◦ ψ −1 )

dιp (w)(f ) = v d(ι ◦ ψ )|ψ(p) (f ) = v
∂xi ψ(p) ∂xi ψ(p)
!
i ∂ −1 i −1 ∂

= v (f ◦ φ ) = v d(φ )φ(p) (f ).
∂xi φ(p) ∂xi ψ(p)
Therefore, the xn -component of v is zero. Conversely, suppose that v has xn -component
equal to zero. Specifically, suppose that v = (v 1 , . . . , v n−1 , 0) in coordinates. The same
computation above shows that v = dιp (w), where w = (v 1 , . . . , v n−1 ) in the coordinates of
(V, ψ). Thus, v ∈ Tp ∂M .
Now suppose that v ∈ Tp M is an inward pointing vector; let γ : [0, ) → M be a smooth
curve such that γ(0) = p and γ 0 (0) = v. Note that φn ◦ γ(t) is nnon-decreasing at t = 0
d
(where φn is the nth component function of φ). Therefore dt 0
(φ ◦ γ) > 0. Recall that
n

d(φ ◦ γ) ∂
γ 0 (0) = (0) ,
dt ∂xi p
so the xn -component of v = γ 0 (0) is greater than or equal to zero. But we assumed that v
is inward pointing and therefore not tangent to ∂M . Since we proved that it is exactly the
vectors tangent to the boundary that have xn -component equal to zero, we conclude that
the xn -component of v is greater than zero. Conversely, suppose that v = (v 1 , . . . , v n ) ∈
Tp M and v n > 0. Then v ∈ Tp M r Tp ∂M since xn -component of v is nonzero. Define
γ(t) = (v 1 t, . . . , v n t) (in coordinates), where t ∈ [0, ) for some > 0 small enough. If
γ(0) 6= p, subtract φ(p) from the coordinate representation of γ to obtain a new smooth
curve. This new curve has derivative v at 0, and therefore v is an inward-pointing vector.
The proof that outward-pointing vectors are exactly those with negative xn -component
is similar.
Exercise 5.44. Suppose M is a smooth manifold with boundary, f is a boundary defining
function, and p ∈ ∂M . Show that a vector v ∈ Tp M is inward pointing if and only if vf > 0,
outward pointing if and only if vf < 0, and tangent to ∂M if and only if vf = 0.
Let (U, φ) be a boundary chart centred at p and let

1 ∂ n ∂

v = v 1
+ ··· + v n
∂x p ∂x p
∂f
in coordinates. Since f is constant on ∂M , we have that ∂x i (p) = 0 for i = 1, . . . , n − 1.
n ∂f
Hence, vf = v ∂xn (p). Since dfp 6= 0, there is a vector w ∈ Tp M such that dfp (w) 6= 0.
Then there is a smooth function g : [0, ∞) → R such that
!
1 ∂ n ∂

dfp (w)(g) = w(g ◦ f ) = w + ··· + w (g ◦ f )
∂x1 p ∂xn p
∂(g ◦ f ) dg ∂f
= wn (p) = wn (f (p)) n (p) 6= 0.
∂xn dt ∂x
∂f
Then ∂x n (p) 6= 0. Moreover, since the coordinate representation of f has a minimum at
n ∂f
x = 0 (where the other coordinates are fixed), we have that ∂x n (p) > 0. This discussion,
together with the result of Exercise 5.42 proves the desired result.
For (a), let U ⊆ M be an open submanifold of M . By Exercise 1.44, U is a smooth
manifold with boundary of the same dimension as M . Moreover, ι : U ,→ M is an immersion
since its differential at any point is the identity, and it is a topological embedding since U
has the subspace topology.
For (b), see the proof of Proposition 5.2.
For (c), see the proof of Proposition 5.5.
For (d), see the proof of Proposition 5.22.
This is a straightforward generalization of Theorem 5.8. The same technique works since
we have Theorem 4.15 (the local immersion theorem for manifolds with boundary).
For (a), see the proof of Theorem 5.27.
For (b), see the proof of Theorem 5.29. Note that since we assume that S is embedded
in M , we have the continuity assumption. Moreover, instead of invoking Proposition 5.22
and Theorem 5.8, invoke Proposition 5.49(d) and Theorem 5.51 instead.
Problem 5.1. Consider the map Φ : R4 → R2 defined by

Φ(x, y, s, t) = (x2 + y, x2 + y 2 + s2 + t2 + y).
Show that (0, 1) is a regular value of Φ, and that the level set Φ−1 (0, 1) is diffeomorphic to
S2 .
Note that Φ−1 (0, 1) = {(x, y, s, t) : y = −x2 , y 2 + s2 + t2 = 1}. We have

2x 1 0 0
dΦ(x,y,s,t) = .
2x 2y + 1 2s 2t
Assume that (x, y, s, t) ∈ Φ−1 . First assume that y 6= 0. Then x 6= 0, and the two leftmost
columns of dΦ are linearly independent. If y = 0, then the condition y 2 + s2 + t2 = 1
implies that either s or t is nonzero. In this case, one of the two leftmost columns of dΦ is
linearly independent, and is linearly independent of the second column. In both cases, dΦ
has full rank and is therefore surjective. This proves that (0, 1) is a regular value. Therefore,
Φ−1 (0, 1) is a 2-dimensional, properly embedded submanifold of of R4 .
Consider the map F : Φ−1 (0, 1) → R3 , (x, y, s, t) → (x, s, t). Since Φ−1 (0, 1) is embedded,
F is smooth. Note that F (Φ−1 (0, 1)) ⊆ {(x, s, t) : x4 + s2 + t2 = 1} = S, and S is a properly
embedded submanifold of R3 , since 1 is a regular value of f (x, s, t) = x4 + s2 + t2 . Hence, we
can restrict the codomain to obtain a smooth map F : Φ−1 (0, 1) → S. We now prove that
this F is a diffeomorphism. This is easy because it has inverse F −1 (x, s, t) = (x, −x2 , s, t),
which is smooth from R4 to R3 , and is therefore smooth when restricted to the embedded
domain S and codomain Φ−1 (0, 1).
It remains to prove that S2 is diffeomorphic to S. We will use coordinates (x, y, z) for
R . Define G : S → S2 by
3
(x, y, z)
G(x, y, z) = p .
x2 + y 2 + z 2
Then G is smooth, since S2 and S are actually embedded in R3 r{0} and G is the restriction
of a smooth map from R3 r {0} to itself. We compute the inverse of G to be
s √
−1 x2 − 1 + 5x4 − 2x2 + 1
G (x, y, z) := (x, y, z)
2x4
whenever x 6= 0, and when x = 0 we have G−1 (x, y, z) = (x, y, z). It then follows that G
is bijective. To check that G is actually a diffeomorphism, it is enough to show that it has
constant rank (by the Global Rank Theorem). Note that G has rank 2 everywhere when
viewed as a map from R3 r {0} to S2 and that the vectors in the kernel of dF are precisely
the ones that point directly towards/away from the origin (where we are identifying the
tangent spaces Tp (R3 r {0}) ∼
= Tp R3 with R3 in the standard way). It suffices then to show
that S has no tangent vectors that point towards/away from the origin.
For a contradiction, suppose that S had some tangent vector pointing towards/away from
the origin. Then there would be some cross-section of S that has a tangent vector pointing
towards/away from the origin in R2 . The cross sections of S are all of the form
x4 + y 2 = c, x4 + z 2 = c, y2 + z2 = c
for some constant c. Clearly, a circle y 2 + z 2 = c has no tangent vectors pointing to-
wards/away from the origin. By symmetry, it then suffices to consider the cross-sections of
the form x4 + y 2 = c. By considering the derivatives dx/dy and dy/dx, it is not hard to see
that this curve doesn’t have any tangent vectors that point towards/away from the origin.
Problem 5.2. Prove Theorem 5.11 (the boundary of a manifold with boundary is an em-
bedded submanifold).
Exercise 1.39(c) shows that ∂M is a topological (n − 1)-manifold and that is is given the
subspace topology with respect to M . Therefore, ι : ∂M ,→ M is a topological embedding.
We now show that the charts defined in Exercise 1.39(c) are smoothly compatible. Let
(U, φ) and (V, ψ) be two charts for ∂M . Then there is are boundary charts (U 0 , φ0 ) and
(V 0 , ψ 0 ) for M such that U = U 0 ∩ ∂M and V = V 0 ∩ ∂M , and φ and ψ are the restrictions
of φ0 and ψ 0 respectively. Let j : ∂Hn ,→ Hn be the inclusion and let π : Hn → ∂Hn be the
projection; note that j and π are both smooth. Hence,
ψ ◦ φ−1 = π ◦ ψ 0 ◦ (φ0 )−1 ◦ j,
so ψ ◦ φ−1 is smooth as it is the composition of smooth maps. Therefore, ∂M is a smooth
manifold (without boundary).
We now want to show that ι is a smooth embedding. Let (U, φ) be a smooth chart for ∂M
and let (U 0 , φ0 ) be a boundary chart for M such that U = U 0 ∩ ∂M and φ is the restriction
of φ0 . The coordinate representation of ι with respect to these charts is simply the inclusion
of j restricted to some open set of ∂Hn . Since this map is a smooth immersion, we conclude
that ∂M is an embedded submanifold of M . To prove that ∂M is properly embedded we
will prove that it is closed (Proposition 5.5). This is true by Proposition 1.38.
Problem 5.3. Prove proposition 5.21 (sufficient conditions for immersed submanifolds to
be embedded).
For (a), apply Proposition 4.22(d).
For (b), apply Proposition 4.22(b).
For (c), apply Proposition 4.22(c).
Problem 5.4. Show that the image of the curve β : (−π, π) → R2 of Example 4.19 is not
an embedded submanifold of R2 .
Let X = β(−π, π) and assume that X is an embedded submanifold of R2 . Then X is
either a 0-, 1- or 2-dimensional manifold. Since the codimension-0 embedded submanifolds
are exactly the open ones and X is not open in R2 , X is either 0- or 1-dimensional. Since X
doesn’t have the discrete topology, it is not 0-dimensional (to see this, note that β −1 (0, 0) = 0
is not open in (−π, π) despite β being continuous). Hence, X is 1-dimensional.
By the classification of 1-dimensional topological manifolds, X is homeomorphic to either
R or S1 . Since X is compact but R is not, this implies that X is homeomorphic to S1 . But
removing any single point from S1 results in a connected 1-manifold, which is not true of
X. Hence, X cannot be an embedded submanifold R2 , since it is not homeomorphic to any
topological manifold.
Problem 5.5. Let γ : R → T2 be the curve of Example 4.20. Show that γ(R) is not an
embedded submanifold of the torus.
First we prove that γ(R) is not an open submanifold of γ(T2 ). Assume that it is; let
x ∈ γ(T2 ) and let U, V ⊆ S1 be open such that x ∈ U × V ⊆ γ(T2 ). By the density of the
Q in R, there exist p, q ∈ Q r Z such that (e2πip , e2πiq ) ∈ U × V . Then there is a t ∈ R
such that γ(t) = (e2πit , e2πiαt ) = (e2πip , e2πiq ). The assumption that (e2πip , e2πiq ) 6= (1, 1)
is essential here: we have (p + m)α = (q + n) for integers n, m, which implies that α is
rational, a contradiction. Therefore γ(T2 ) is not open; hence, γ(T2 ) must be 1-dimensional
(it clearly does not have the discrete topology and so is not 0-dimensional).
By the classification of 1-dimensional manifolds, γ(R) is homeomorphic to either R or S1 .

However, both of these manifolds are locally path connected, but γ(R) is not. Hence, γ(R)
is not an embedded submanifold of T2 .
Problem 5.6. Suppose M ⊆ Rn is an embedded m-dimensional submanifold, and let U M ⊆
T Rn be the set of all unit tangent vectors to M :
U M = {(x, v) ∈ T Rn : x ∈ M, v ∈ Tx M, |v| = 1}.
It is called the unit tangent bundle of M . Prove that U M is an embedded (2m − 1)-
dimensional submanifold of T Rn ∼
= Rn × R n .
We begin by showing that T M is an embedded submanifold of T Rn . Let x ∈ M be
arbitrary and let (U, φ) be a slice chart for M in Rn containing x. We have
φ(U ∩ M ) = {(x1 , . . . , xn ) ∈ φ(U ) : xm+1 = · · · = xn = 0}.
Let (U 0 , φ0 ) be the chart for T Rn associated to (U, φ). Then
φ0 (U 0 ∩ T M ) = {(x, v) ∈ φ(U ) × Rn : xm+1 = · · · = xn = 0 = v m+1 = · · · = v n },
so (U 0 , φ0 ) is a slice chart for T M in T Rn containing (x, v) for every v ∈ Tx M . Thus, T M
satisfies the local 2m-slice condition in T Rn and is therefore an embedded submanifold of
T Rn .
2
Identifying T Rn with Rn ×Rn , consider the smooth map F : T Rn → R, (x, v) 7→ |v| . Note
that F is smooth, and since T M is embedded, F restricts to a smooth map on T M . More-
over, U M = F −1 (1). To show that U M is a (2m − 1)-dimensional embedded submanifold of
T M , and therefore of T Rn , it is enough to show that 1 is a regular value of F when viewed
as a map on T M . Let (x, v) ∈ F −1 (1) and let γ : (0, ∞) → T M, t 7→ (x, (1 + t)v). Note that
because x is held constant, the image of γ is contained in T M (the scalar multiple of a vector
2
in Tx M is always in Tx M ). Moreover, dF(x,v) γ 0 (0) = dtd
0
(F ◦γ) = d
dt 0 (1+t)2
|v| = 2 6= 0.
−1
Therefore, dF(x,v) is surjective at each point of F (0) so U M is an embedded (2m − 1)-
dimensional submanifold of T Rn .
Problem 5.7. Let F : R2 → R be defined by F (x, y) = x3 + xy + y 3 . Which level sets of F
are embedded submanifolds of R2 ? For each level set, prove either that it is or that it is not
an embedded submanifold.
We begin by computing the regular values of F since the regular level sets are embedded.
We will deal with the remaining level sets afterwards. The differential of F is given by
dF(x,y) = 3x2 + y x + 3y 2 ,

which is not surjective only when 3x2 + y = x + 3y 2 = 0. This system of equations has
solutions (x, y) = (0, 0) and (x, y) = (− 31 , − 13 ). Since F (0, 0) = 0 and F (− 13 , − 13 ) = 27
1
,
−1 2
we have determined that F (c) is an embedded 1-dimensional submanifold of R for every
1
c 6= 0, 27 .
We will now show that F −1 ( 27
1
) is an embedded submanifold by first determining F −1 ( 27 1
)
2
as a subset of R . Note that

3 3 1 1 2 1 2 1 1
x + xy + y − = x+y− y − (x − )y + (x + x + ) .
27 3 3 x 9
The first factor is zero on the line given by y = 13 −x. To figure out when the second factor is
zero, we can solve for y in terms of x using the quadratic equation. Doing this, we find that
the discriminant is nonnegative only when x = − 13 ; when this happens, y = − 13 . Hence,
F −1 ( 27
1
) consists of the line y = 13 − x and the point (− 13 , − 13 ). It this subset corresponds to
an embedding of the disjoint union of a point (which is a 0-dimensional submanifold) and
R. Note that if we require manifolds to have the same dimension everywhere then F −1 ( 27 1
)
is not an embedded submanifold.
We finish by showing that F −1 (0) is not an embedded submanifold. Note that F −1 (0)
is actually the folium of Descartes. Assume that it is embedded. Then since F −1 (0) is not
open (it is closed, nonempty, and not equal to all of R2 ) it cannot be 2-dimensional, and
is therefore an embedded 1-dimensional submanifold of R2 . Since F −1 (0) is not compact,
it must then be homeomorphic to R. However, deleting (0, 0) from F −1 (0) splits it into
three components, but deleting any point from R breaks it up into two components. We
conclude that F −1 (0) is not an embedded submanifold of R2 . [Another (perhaps more
rigorous) approach would be to take two curves whose image is contained in F −1 (0) whose
derivatives span a 2-dimensional subspace, contradicting that F −1 (0) is 1-dimensional.]
Problem 5.8. Suppose M is a smooth n-manifold and B ⊆ M is a regular coordinate ball.
Show that M r B is a smooth manifold with boundary, whose boundary is diffeomorphic to
Sn−1 .
We begin by constructing a smooth atlas for M r B. Let p be a point in the topological
interior of M r B. Then there is a smooth chart for M containing p whose domain does not
intersect B. This chart is then also a chart for M r B containing p.
Now suppose that p is a point in the topological boundary of M r B. Since B is a regular
coordinate ball, there is a coordinate ball B 0 such that B ⊆ B 0 . Note that B is embedded
in B 0 , which is in turn embedded in M . Hence, the inclusion B ,→ M is an embedding.
Therefore, B satisfies the local n-slice condition for submanifolds with boundary. This means
that there is a chart (U, φ) for M containing p such that (U ∩ B, φ|U ∩B ) is a boundary chart
for B. But then (U r B, φ|U rB ) is a boundary chart for M r B containing p.
All of the charts described above are smoothly compatible, since they are restrictions of
smoothly compatible charts for M . Moreover, the construction of our charts show that the
manifold boundary is equal to ∂B, which is diffeomorphic to Sn−1 .
Problem 5.9. Let S ⊆ R2 be the boundary of the square of side 2 centred at the origin.
Show that S does not have a topology and smooth structure in which is is an immersed
submanifold of R2 .
A single side (without endpoints) of the square is an embedded 1-dimensional submani-
fold, so if S had a topology and smooth structure making it immersed, it would have to be
1-dimensional. Assume S has such a topology and smooth structure. Then there is a curve
γ : (−, ) → S such that γ(0) = (1, 1) and γ 0 (0) 6= 0. Write γ(t) = (x(t), y(t)). Then x(t)
and y(t) either are constant on one side of t = 0, or they have a local maximum there. In
any case, x0 (0) = y 0 (0) = 0, giving γ 0 (t) = 0, a contradiction. We conclude that S is not an
immersed submanifold of R2 .
Problem 5.10. For each a ∈ R, let Ma be the subset of R2 defined by
Ma = {(x, y) : y 2 = x(x − 1)(x − a)}.
For which values of a is Ma an embedded submanifold of R2 ? For which values can Ma be
given a topology and smooth structure making it into an immersed submanifold.
Letting F (x, y) = y 2 − x(x − 1)(x − a), we have
−3x2 + 2(a + 1)x − a

dF |(x,y) = 2y .
Since dF is surjective unless both entries are zero, we find that y = 0 at all singular points
of F . The points of F −1 (0) with y = 0 are (0, 0), (1, 0), and (a, 0). Plugging these into dF ,
we find that dF is surjective unless a = 0 or a = 1. Thus, Ma is an embedded submanifold
for a 6= 0, 1.
We now consider the case where a = 0. We are then considering the set of points (x, y)
satisfying y 2 = x2 (x − 1). Ignoring the case where x = 0, we have that x = (y/x)2 + 1,
so x > 1. We then see that M0 is the disjoint union of the point (0, 0) and the curve
defined by y 2 = x2 (x − 1) for x > 1. It is easy to show that this curve is embedded in
{(x, y) : x > 32 }, and is therefore embedded in R2 . If we allow components of our manifolds
to have different dimensions, then M0 is an embedded submanifold. Otherwise, M0 cannot
be given a topology making it an immersed submanifold (since the components have different
dimensions).
Now, let a = 1; we are considering the set of points satisfying y 2 = x(x − 1)2 . It is not
hard to show that removing the point (1, 0) breaks M1 into three components. Thus, M1
cannot be an embedded submanifold, since neither R nor S1 have more than two components
after removing an arbitrary point. Intuitively, M1 is immersed and inherits the topology
of two disjoint copies of R, which “crash into each other” at (1, 0). Since x > 0 in M1 , it
makes sense to let x = t2 , which gives y = ±t(t2 − 1) (the ± determines the direction of the
parametrization. Then
f : R r {1}, t 7→ (t2 , ±t(t2 − 1))
is an injective immersion.
Problem 5.11. Let Φ : R2 → R be defined by Φ(x, y) = x2 − y 2 .
(a) Show that Φ−1 (0) is not an embedded submanifold of R2 .
(b) Can Φ−1 (0) be given a topology and smooth structure making it into an immersed
submanifold of R2 ?
(c) Answer the same two questions for Ψ : R2 → R defined by Ψ(x, y) = x2 − y 3 .
For (a), note that Φ−1 (0) is 1-dimensional since it is not an open submanifold of R2 .
Removing the point (0, 0) breaks Φ−1 (0) into four connected components. Therefore Φ−1 (0)
cannot be homeomorphic to R or S1 , and is therefore not an embedded submanifold of R2 .
For (b), the answer is yes. Let M be the disjoint union of three copies of R, which we
label R1 , R2 , R3 . On R1 , we define φ(t) = (t, −t). On R2 , we define φ(t) = (et , et ). Finally,
on R3 , we define φ(t) = (−et , −et ). This gives a well defined bijection φ : M → Φ−1 (0).
We give Φ−1 (0) a topology by declaring φ to be a homeomorphism. It is a straightforward
computation to show that φ is an immersion.
For (c), we will show that Φ−1 (0) is not even immersed (and therefore not embedded).
If Φ−1 (0) was immersed, then there would be an injective curve γ : (−, ) → Φ−1 (0) such
that γ(0) = (0, 0) and γ 0 (0) 6= 0. Let γ(t) = (x(t), y(t)). Since y has a minimum at t = 0, we
have that y 0 (0) = 0. Wep will assume that x(t) > 0 for t ∈ [0, ). Then x(t) = (y(t))3/2 there,
which implies x (t) = 2 y(t)y 0 (t) ⇒ x0 (0) = 0. Therefore, γ 0 (0) = 0, and we conclude that
0 3
Φ−1 (0) is not immersed.

Problem 5.12. Suppose E and M are smooth manifolds with boundary, and π : E → M is
a smooth covering map. Show that the restriction of π to each connected component of ∂E
is a smooth covering map onto a component of ∂M .
We begin by showing that the restriction π : ∂E → ∂M is smooth. First, note that
π(∂E) ⊆ ∂M because π is a local diffeomorphism. By Theorem 5.11, the boundary of a
manifold is a properly embedded submanifold; combined with the result of Theorem 5.30,
this gives that π : ∂E → ∂M is smooth.
Let Fp ∈ ∂M , and let B be some coordinate half ball centred at p that lifts to a disjoint
union α Vα ⊆ E, where π maps each Vα diffeomorphicallyF onto B. Then ∂B is a neigh-
bourhood of p in ∂M and it lifts to the disjoint union α ∂Vα . The restriction of π to each
∂Vα is again a diffeomorphism, since boundaries are embedded submanifolds. We further
restrict π to a connected component of ∂E in order to satisfy the definition of a covering
map.
Problem 5.13. Prove that the image of the dense curve on the torus described in Example
4.20 is a weakly embedded submanifold of T2 .
Let S ⊆ T2 be the image of the dense curve in the torus. Let N be a smooth manifold
and let f : N → T2 be a smooth map whose image is contained S. Let (U, φ) and (V, ψ)
be smooth coordinate balls for N and T2 , respectively, centred at p and f (p), respectively.
By shrinking U and V if necessary, we can choose V so that it is a product neighbourhood,
and S ∩ V is the disjoint union of sets diffeomorphic to open intervals (in the topology of S,
not of T2 ). Moreover, we also choose U and V so that f (U ) ⊆ V . Since f (U ) is connected,
it must be contained in one of the intervals of S ∩ V , which we will denote by J. Let
τ : ψ(J) → I ⊆ R be a diffeomorphism from ψ(J) to an open interval I. Then (J, τ ◦ ψ|J )
is a smooth coordinate chart for S, since J is an embedded submanifold of V . We conclude
that τ ◦ ψ|J ◦ f ◦ φ−1 is a smooth coordinate representation for f at p as a map into S, and
therefore f is smooth as a map into S. Therefore, S is weakly embedded.
Problem 5.14. Prove Theorem 5.32.
Let S and Se be immersed submanifolds of M , where
Problem 5.15. Show by example that an immersed submanifold S ⊆ M might have more
than one topology and smooth structure with respect to which it is an immersed submanifold.
Let S be the figure-eight subset of R2 described in Example 4.19. We can parametrize
S by β0 : (−π, π) → R2 , t 7→ (sin 2t, sin t) or β1 : (−π, π) → (− sin 2t, sin t). Both makes S
homeomorphic to an interval, but the open sets are different. The resulting topologies are
different, and thus the smooth structures are as well. It is not hard to come up with similar
examples where the two topologies do not yield homeomorphic spaces.
Let ι̃ : Se → M, p 7→ p be a weak embedding, where S and Se have the same underlying
sets. Because ι̃ is smooth and ι̃(S)e = S, we have that the restricted map ι̃0 : Se → S is
smooth. Since ι̃ = ι ◦ ι̃0 (where ι : S → M is the inclusion) is an immersion, we have we
have that the differential dp ι̃ = dp ι ◦ dp ι̃0 is injective at every point p ∈ S.
e But ι is also
an immersion, so we must have that ι̃0 is also an immersion. But ι̃0 is also bijective, and
therefore a diffeomorphism by the global rank theorem. Hence, S and Se are diffeomorphic,
proving that the smooth structure on S is unique.
Problem 5.17. Prove Lemma 5.34.
(a) Let S be a k-dimensional submanifold of the n-dimensional manifold M . Cover S by
slice charts (Ui , ϕi ) such that all points in each ϕi (S ∩ Ui ) have first k coordinates equal to
zero. We can extend f ◦ ϕ−1 i |ϕi (S∩Ui ) : ϕi (S ∩ Ui ) → R to all of ϕi (Ui ) by defining
Fi : ϕi (Ui ) → R, (x1 , . . . , xk , xk+1 , . . . , xn ) 7→ f ◦ ϕ−1

i |ϕi (S∩Ui ) (0, . . . , 0, xk+1 , . . . , xn ).
S
Let U = i Ui . Then the Ui form an open cover for the submanifold U , so we can choose a
partition of unity ψi : U → R subordinate to it. Hence,
X
f˜: U 7→ ψi · Fi ◦ ϕ−1

i
is an extension of f to U .
(b) By Proposition 5.5, S is closed in M . Let U be a neighbourhood of S such that
f˜: U → R extends f . Let {ϕ1 , ϕ2 } be a partition of unity subordinate to the open covering
{U, M − S}. Then ϕ1 · f˜ is an extension of f to all of M (where we take this function to be
zero outside of U ).
Problem 5.18. Suppose M is a smooth manifold and S ⊆ M is a smooth submanifold.
(a) Show that S is embedded if and only if every f ∈ C ∞ (S) has a smooth extension to
a neighbourhood of S in M .
(b) Show that S is properly embedded if and only if every f ∈ C ∞ (S) has a smooth
extension to all of M .
The forward directions of each part are given by Lemma 5.34.
(a) Suppose that S is not embedded, so there is a point p ∈ S that is not in the domain of
any slice chart for S. Since S is immersed, it is locally embedded by Proposition 5.22. Let
U ⊆ S be an open neighbourhood of p such that U is an embedded submanifold of M , and
let (V, ψ) be a slice chart for U containing p. If V ∩ S ⊆ U , then V is also a slice chart for
S containing p, a contradiction. Hence, V contains some point of S − U . Since slice charts
containing p can be made arbitrarily small, this shows that every open neighbourhood (in
M ) of p containing a point of S − U .
Now, let f ∈ C ∞ (S) be a function supported in U and such that f (p) = 1. Suppose,
for a contradiction, that f has a smooth (or even just continuous) extension f˜ to an open
neighbourhood A of S. By the above paragraph, there is a sequence of points in (S − U ) ∩ A
converging to p. By continuity of f˜, we have that 0 = f˜(p) = f (p) = 1, a contradiction. We
conclude that S is embedded.
(b) From part (a) we immediately obtain that S is embedded. By Proposition 5.5, it
suffices to show that S is closed in M . For a contradiction, suppose that it is not. Then
there is an accumulation point p of S that is not contained in S. Let (xn )n∈N be a sequence
converging to p. By Proposition 2.28, S admits a smooth exhaustion function f : S → R
(so f −1 (]−∞, c[) is compact for every c ∈ R). For every j ∈ N, M − f −1 (]−∞, j[) is
then an open neighbourhood of p, and therefore contains some point xnj . This implies that
f (xnj ) > nj for every j ∈ N, which means that f cannot be extended continuously at p. We
conclude that S is properly embedded.
Problem 5.19. Suppose S ⊆ M is an embedded submanifold and γ : J → M is a smooth
curve whose image happens to lie in S. Show that γ 0 (t) is in the subspace Tγ(t) S of Tγ(t) M
for all t ∈ J. Give a counterexample if S is not embedded.
Let (U, ϕ) be a slice chart for S containing γ(t). We also assume ϕ(U ) = Rn and
ϕ(S ∩ U ) = Rk ⊆ Rn , where Rk is viewed as the subspace of Rn whose first k coordinates
are zero. It then follows that (ϕ ◦ γ)0 (t) ∈ Tϕ(γ(t)) Rk ⊆ Tϕ(γ(t)) Rn . Since dϕ(γ(t)) ϕ−1 maps
Tϕ(γ(t)) Rk into Tγ(t) S, we have that γ 0 (t) ∈ Tγ(t) S.
For a counterexample, consider the folium of Descartes, which is the locus of points
defined by x3 +y 3 = 3xy. We realise the folium of Descartes as an immersion of S = R−{−1}
into R2 by
3t2

3t
F : S → R2 , t 7→ , .
1 + t3 1 + t3
Checking that F is indeed an immersion is straightforward. Moreover, T(0,0) S is canonically
identified with the x-axis of T(0,0) R2 = R2 . A painful but straightforward check shows that
the image of the curve
√ s √ s √
 
2 3 2 − 2t 3 2 − 2t 
γ : ]−1, 1[ →, t 7→ t−t √ ,t + t √ .
2 6t + 3 2 6t + 3 2
lies in S, but γ 0 (0) lies in the y-axis, so γ 0 (t) ∈

/ Tγ(0) S = T(0,0) S.
Remark. Even though the solution is ugly, the idea is simple. Descartes’s folium was the
first example of an immersed submanifold of R2 that “crashes into itself” that I could think
of. Plotting the curves I’ve given in a graphing calculator such as Desmos will give an idea
of how to come up with other more simple examples.
Problem 5.20. Show by giving a counterexample that the conclusion of Proposition 5.37
may be false if S is merely immersed.
As in the previous problem, we will make use of the folium of Descartes. Let F : S → R2
and γ : ]−1, 1[ be as in the previous problem, and let f be a function vanishing on S ⊆ R2 .
Then f ◦ γ ≡ 0. Hence,

∂f
0 = d0 (f ◦ γ) = d(0,0) f ◦ d0 γ = c
∂y (0,0)
∂ 2
for some c ∈ R. But ∂y ∈ T(0,0) R − T(0,0) S. Hence, Proposition 5.37 does not apply to
general submanifolds.
(a) We will show that f −1 (]−∞, b]) satisfies the local slice condition at every point. First,
−1
f (]−∞, b[) is an open submanifold of M , and so vacuously satisfies the local slice condition.
It therefore suffices to show that the points of f −1 (b) satisfy the local slice condition. Let
p ∈ f −1 (b). Since b is a regular value, dfp is surjective. Thus, there is a neighbourhood of p
on which f is a submersion. By the rank theorem (Theorem 4.12), there are charts (U, ϕ) in
M and (V, ψ) in R on which the coordinate representation is fb(x1 , . . . , xn ) = x1 . It is then
not hard to see that (U ∩ f −1 (]−∞, b])) is a boundary slice chart. Thus, f −1 (]−∞, b]) is
an embedded submanifold with boundary. Since f is continuous and ]−∞, b] ⊆ R is closed,
f −1 (]−∞, b]) is properly embedded by Proposition 5.5. (b) follows similarly.
By the boundary slice condition, at every point of ∂D, we can find a chart (U, ϕ) such that
the nth coordinate of ϕ(U ∩∂D) (resp. ϕ(U ∩intD), resp. ϕ(U ∩(M −D))) is zero (resp. neg-
ative, resp. positive). We can also take our charts to be such that ϕ(c1 , . . . , cn−1 , xn ) is an
increasing function of xn for each point (c1 , . . . , cn−1 ). Moreover, we can assume that
∂
∂xn ϕ > 0 at every point, since if it is not we can replace ϕ with ϕ + xn . Now, cover intM
with charts whose nth coordinate is negative and M − D with charts whose nth coordinate
is positive. Let {(Uα , ϕα )}α∈I be the collection of all charts mentioned above, and let {ψα }α
be a partition of unity subordinate to the cover by charts. Define
X
f := (xn ◦ ϕα ) · ψα .
α
Then 0 is a regular value of f by the derivative condition and f −1 (]−∞, 0]) = D. Thus, f
is a defining function for D.
Now assume that D is compact. We take the same charts as above, except we moreover
assume that they are all precompact and that the cover is countable, and that there are only
finitely many charts intersecting D. Let {(U−t , ϕ−t ), . . . , (U0 , ϕ0 )} be the charts intersecting
D, and let {(Uj , ϕj )}j∈N be the charts in M − D. Let {ψ−t , . . . , ψ0 } ∪ {ψj }j∈N be a partition
of unity subordinate to the cover. Define
0
X ∞
X
g := (xn ◦ ϕj ) · ψj + jψj .
j=−t j=1
Then g is a defining function for D, and g −1 (]−∞, c]) is compact for any c ∈ R by precom-
pactness of the charts Uj and compactness of D.
Problem 5.23. Suppose M is a smooth manifold with boundary, N is a smooth manifold
and F : M → N is a smooth map. Let S = F −1 (c), where c ∈ N is a regular value for both
F and F |∂M . Prove that S is a smooth submanifold with boundary in M , with ∂S = S ∩∂M .
We will show that S satisfies the local k-slice condition for submanifolds with boundary
(see Theorem 5.51), where k = dim M − dim N . Fix m = dim M and n = dim N , and note
that dim M > dim N since c is a critical value of F |∂M .
Restricting F to the interior of M , we see that S ∩ int M is a properly embedded smooth
submanifold of int M and therefore satisfies the local k-slice condition at these points. It
remains to check the condition at points of S ∩ ∂M . By choosing suitable coordinate charts,
we assume that M = Hm = {(x1 , . . . , xm ) : x1 > 0} and N = Rn , and that p = 0 in these
coordinates. Let F be a smooth extension of F to an open subset of Rm . Note that dF p
has full rank, and in fact it has full rank when restricted to the subset x1 = 0. Hence, we
can shuffle the coordinates x2 , . . . , xm so that dFp has the form

dF p = A B
where B is an n × n invertible matrix. By the implicit function theorem, there is an open
neighbourhood U of 0 ∈ Rk such that
S = {(x1 , . . . , xk , f (x1 , . . . , xk )) : x1 > 0, (x1 , . . . , xk ) ∈ U }
for some smooth function f : Rk → Rn . Since S is locally the graph of some function, it is
locally an embedded k-dimensional smooth manifold with boundary, and therefore satisfies
the local k-slice condition. It is clear from the proof that ∂S = S ∩ ∂M .
6. Sard’s Theorem
Exercise 6.1. Show that open rectangles can be replaced by open balls or open cubes in the
definition of subsets of measure zero.
We will say a subset A ⊆ Rn has rectangle-measure zero if, for any δ > 0, A can be
covered by countably many rectangles whose volumes sum is less than δ. The definitions of
cube-measure zero and ball-measure zero are similar.
Suppose A ⊆ Rn has rectangle-measure zero and let R be a rectangle with sides of
lengths 1 = l1 6 · · · 6 ln . Then R can be covered by a rectangle R0 with sides of length
dl1 e 6 · · · 6 dln e, which can can be tessellated by cubes of sidelength 1. Moreover
vol R0 < (l1 + 1)(l2 + 1) · · · (ln + 1) 6 (2l1 )(2l2 ) · · · (2ln ) = 2n vol R.
We chose l1 = 1 to keep things simple, but, by rescaling, the above argument shows that
every rectangle R can be covered by cubes whose volumes sum to 2n vol R. It is then clear
that A also has cube-measure zero. Since cubes are rectangles, we conclude that cube-
measure zero is equivalent to rectangle-measure zero.
It remains to show that A has cube-measure zero if and only A has ball-measure zero. To
see this, note that a ball of radius r can be inscribed in a cube
√
of sidelength 2r. Conversely,
a cube of sidelength l can be inscribed in a ball of radius 2n l. Since the volumes of balls
and cube scale like the nth power of the radius and sidelength, respectively, the conclusion
follows.
Exercise 6.7. Let M be a smooth manifold with or without boundary. Show that a countable
union of sets of measure zero in M has measure zero.
It is enough to consider the case M = Rn and verify the claim in the identity chart. Let
{Ai }i∈N be a countable collection of sets of measure zero in M and let δ > 0 be arbitrary.
For each i ∈ N, let
S Ci be a collection of Srectangles covering Ai whose volume sum is less
than 2δi . Let C = i∈N Ci . Then C covers i∈N Ai and its volume sum is less than δ.
Problem 6.1. Use Proposition 6.5 to give a simpler proof of Corollary 6.11 that does not
use Sard’s theorem.
Let F : M → N be a smooth map of manifolds (with or without boundary) and suppose
that dim M < dim N . Then goal is to show that F (M ) has measure zero in N . Since M
is second countable and the countable union of measure zero sets has measure zero, it is
enough to show that the image of each chart has measure zero. Hence, we assume that
M = Rm . Cover N with countably many charts, each diffeomorphic to Rn ; by Lemma 6.6
it is enough to check that the image of F (M ) in each chart has measure zero. It is therefore
enough to assume that F is a map of the form Rm → Rn with m < n, and the goal is now
to show that F (Rm ) has measure 0 in Rn .
View Rm as a proper affine subspace of Rn . By Lemma 2.26, there is a smooth extension
F : Rn → Rn of F . By Corollary 6.4, Rm has measure zero in Rn and finally, by Proposition
e
6.5, Fe(Rm ) = F (Rm ) has measure zero in Rn .
Problem 6.2. Prove Theorem 6.18 (the Whitney immersion theorem) in the special case
∂M = ∅.
We follow the hint (see the textbook). Let n = dim M ; by Theorem 6.15 (the Whitney
Embedding Theorem), we may view M as a smoothly and properly embedded submani-
fold of R2n+1 . Letting U M ⊆ T R2n+1 be the unit tangent bundle of M , let G : U M →
RP2n , (x, v) 7→ [v]. Recall that dim U M = 2n − 1, so G(U M ) has measure zero in RP2n
by Sard’s theorem. The image of R2n+1 r R2n in RP2n under the standard projection is
open, and therefore intersects the complement of G(U M ); let v ∈ R2n+1 r R2n be such that
[v] ∈ RP2n is not in the image of G.
We claim that the linear projection R2n+1 → R2n with kernel Rv restricts to a smooth
immersion on M . This is equivalent to showing that, for all p ∈ M , no vectors of Tp M are
parallel to v. But this is guaranteed by the fact that v is not in the image of G, so we are
done.
Problem 6.3. Let M be a smooth manifold, let B ⊆ M be a closed subset, and let δ : M → R
be a positive continuous function. Show that there is a smooth function δe: M → R that is
zero on B positive on M r B, and satisfies δ(x)
e < δ(x) everywhere.
By Theorem 2.29, there exists a smooth nonnegative function f : M → R such that B =

f −1 (0). By Corollary 6.22, there is a smooth function e : M → R such that 0 < e(x) < δ(x)
everywhere. The desired function is

f (x)
δ(x) :=
e e(x).
1 + f (x)
Problem 6.4. Let M be a smooth manifold, let B be a closed subset of M , and let δ : M → R
be a positive continuous function.
(a) Given any continuous function f : M → Rk , show that there is a continuous function
fe: M → Rk that is smooth on M r B, agrees with f on B, and is δ-close to f .
(b) Given a smooth manifold N and a continuous map F : M → N , show that F is
homotopic relative to B to a map that is smooth on M r B.
(a) Let δe: M → R be a nonnegative smooth function such that δe−1 (0) = B and δ(x) e 6
δ(x) for all x; such a function exists by the result of the previous problem. By Theorem 6.21
(the Whitney Approximation Theorem for Functions), there is a smooth function g : M r
B → Rk that is δ-close
e to f |M rB . Let
(
g(x) if x ∈ M r B
fe(x) := .
f (x) if x ∈ B
It is clear that fe is smooth on M r B and that it agrees with f on B, so it remains to check

that it is continuous. To this end, let (xn )n∈N be a sequence of points converging to a point
x ∈ B. We will assume that xn ∈ B for each n ∈ N. Then

f (xn ) − fe(x) = |g(xn ) − f (x)| 6 δ(x
e n)
e
and δ(x
e n ) converges to 0, since δe is continuous. We conclude that fe(xn ) converges to fe(x),
and therefore that fe is continuous.
(b) In the proof of Theorem 6.26, we can take a map Fe : M → N that agrees with F
on B, is smooth on M r B, and such that F is homotopic relative to B to r(Fe). Here r is
a retraction of a tubular neighbourhood, so again r(Fe) agrees with F on B. The notation
is confused between the problem stated here and the notation of the proof, but the idea
should be clear; it is a straightforward application of part (a) to the proof of Theorem 6.26.
Problem 6.5. Let M ⊆ Rn be an embedded submanifold. Show that M has a tubular
neighbourhood U with the following property: for each y ∈ U , r(y) is the unique point in M
closest to y, where r : U → M is the retraction defined in Proposition 6.25.
We begin by showing that if y is a closest point x ∈ M , then y − x is orthogonal to Tx M .
Define
f : Rn → R, p 7→ |y − p|.
The directional derivative of f in the direction v at x is zero for any v ∈ Tx M , so But the
directional derivative is
y−x
∇f · v|x = · v = 0,
|y − x|
which shows that y − x is orthogonal to Tx M .
Let U be the tubular neighbourhood of M defined by a positive function ρ : M → R
as in the proof of Theorem 6.24. Let ρe = 12 ρ; we claim that the tubular neighbourhood
Ue defined by ρe satisfies the desired property. Suppose there is a point y ∈ M such that
|x − y| < |x − r(x)|, where x ∈ U e . We will assume that y is a closest point to x. Then
1 1
|y − r(x)| 6 |y − x| + |x − r(x)| < δ + δ = δ
2 2
giving that y ∈ Vδ (x). But then r(x) = y, since r(x) is the unique point p of M ∩ Vδ (x)
such that x − p is orthogonal to Tp M .
Problem 6.6. Suppose M ⊆ Rn is a compact embedded submanifold. For any ε > 0,
let Mε be the set of points in Rn whose distance from M is less than ε. Show that for
sufficiently small ε, ∂Mε is a compact embedded hypersurface in Rn , and Mε is a compact
regular domain in Rn whose interior contains M .
Let U be a tubular neighbourhood of M defined by a function δ : M → R. Because
M is compact, δ attains a minimum ε > 0 on M , so we can take δ to be the constant
function ε, and therefore U = Mε . It is also clear that we can choose ε small enough so
that U is diffeomorphic to the subset of N M whose vectors have length 6 ε. This is an
n-manifold with boundary (locally it looks like Rk × Bn−k , where dim M = k), and ∂Mε is
its diffeomorphic image in Rn . Moreover, ∂Mε is a boundary and therefore closed, and it
is bounded since Mε is bounded. We conclude that ∂Mε is a compact hypersurface of Rn .
The closure Mε is a proper submanifold since it is of codimension 0 and it is defined as a
subset of Rn . Moreover, its interior is Mε , which clearly contains M .
Problem 6.7. By considering the map F : R → H2 given by F (t) = (t, |t|) and the subset
A = [0, ∞[ ⊆ R, show that the conclusions of Theorem 6.26 and Corollary 6.27 can be false
when M has nonempty boundary.
Note that F |A : A → H2 is smooth. We will prove that any extension of F |A to R is
not smooth at t = 0, which will show that the conclusions of Theorem 6.26 and Corollary
6.27 are false. For a contradiction, suppose that f is a smooth extension of F |A to R. The
derivative of f at t = 0 is
f (h) − f (0) f (h)
d0 f = lim = lim = (1, 1),
h→0 h−0 h→0 h
where the last equality is obtained by taking the limit as h → 0+ . On the other hand,
f (h)
lim−
h→0 h
has a nonpositive y-component, since h is approaches 0 from the negatives here, and f (h)
always has a positive y-component. We conclude that f is not even C 1 , let alone smooth.
Problem 6.8. Prove that every proper continuous map between smooth manifolds is homo-
topic to a proper smooth map.
Following the hint, we prove that the map Fe constructed in the proof of Theorem 6.26 is
proper if F : N → M is. We use the notation of the proof, and assume that M is a properly
embedded submanifold of Rn . Let K be a compact subset of M and let L be a compact
subset of M that contains all the points that are of distance at most 1 from K (we can
take the intersection of M with some ball in Rn of sufficiently large radius for this). Since
Fe is δ-closed
e to F and δe is uniformly bounded by 1, we obtain Fe−1 (K) ⊆ F −1 (L). Since
F is proper, Fe−1 (K) is then a closed subset of a compact set and is therefore compact,
concluding the proof. A slightly more careful analysis of the proof given here and that of
Theorem 6.26 shows that F and Fe can be taken to be homotopic through proper maps.
Problem 6.9. Let F : R2 → R3 be the map F (x, y) = (ey cos x, ey sin x, e−y ). For which
positive numbers r is F transverse to the sphere Sr (0) ⊆ R3 ? For which positive numbers r
is F −1 (Sr (0)) an embedded submanifold of R2 ?
We have
∂x F (x, y) = (−ey sin x, ey cos x, 0) and ∂y F (x, y) = (ey cos x, ey sin x, −e−y ).
2
The map F will not√ be transverse to Sr (0) if and only if there is a point (x, y) ∈ R such
that |F (x, y)| = e2y + e−2y = r and the vectors ∂x F (x, y) and ∂y F (x, y) are parallel to
Sr (0). This last condition is in turn equivalent to the following orthogonality relations:
∂x F (x, y) · F (x, y) = 0 and ∂y F (x, y) · F (x, y) = 0.
2y −2y
The first of these holds everywhere, and √ the second is equivalent to e − e =
√ 0, which
has solution y = 0 and therefore r = 2. So F is transverse to Sr (0) unless r = √ 2.
By Theorem
√ 6.30(a), F −1 (Sr (0)) is an embedded submanifold of R2 for r 6= 2. In the
case r = 2, we have
F −1 (Sr (0)) = {(x, y) : e2y + e−2y = 2},
which is just the x-axis and is clearly an embedded submanifold. Hence, F −1 (Sr (0)) is an
embedded submanifold for all positive values of r.
Problem 6.10. Suppose F : N → M is a smooth map that is transverse to an embed-
ded submanifold X ⊆ M , and let W = F −1 (X). For each p ∈ W , show that Tp W =
(dFp )−1 (TF (p) X). Conclude that if two embedded submanifolds X, X 0 ⊆ M intersect trans-
versely, then Tp (X ∩ X 0 ) = Tp X ∩ Tp X 0 for every p ∈ X ∩ X 0 .
The diagrams
F
N M
ιW ιX
W X
F |W
commutes. Let v ∈ Tp W . By functoriality,

(dFp ◦ dιp )(v) = (dιF (p) ◦ d(F |W )p )(v),
which implies dFp (v) ∈ TF (p) X (one sees this by removing the inclusion maps labelled ι
from the notation). This shows that Tp W ⊆ (dFp )−1 (TF (p) X).
Now focus on the reverse inclusion. Choose a basis of the form {u1 , . . . , uk , v1 , . . . , vl } for
TF (p) M , where the vectors ui form a basis for TF (p) X and the vectors vi are in dFp (Tp W ).
This is possible because F is transverse to X. By Theorem 6.30, W is an embedded sub-
manifold of N whose codimension is equal to that of X in M . Hence, we can choose a basis
{x1 , . . . , xj , y1 , . . . , yl } for Tp N such that the vectors xi are in Tp W and dFp (yi ) = vi for
each i = 1, . . . , l. It is then clear that (dFp )−1 (TF (p) X) ⊆ Tp W . We have thus shown that
Tp W = (dFp )−1 (TF (p) X).
For the last claim, consider the inclusion ι : X ,→ M , where X intersects X 0 transversely.
Then X ∩ X 0 = ι−1 (X 0 ), so it remains to show that (dιp )−1 (Tp X 0 ) = Tp X ∩ Tp X 0 . This is
clear however, since dιp is injective.
Problem 6.11. Suppose F : M → N and G : N → P are smooth maps, and G is transverse
to an embedded submanifold X ⊆ P . Show that F is transverse to the submanifold G−1 (X)
if and only if G ◦ F is transverse to X.
The condition that G be transverse to X ⊆ P means that for every p ∈ G−1 (X), we have
(†) dGp (Tp N ) + TG(p) X = TG(p) P.
Assume that F is transverse to G−1 (X). Then for every q ∈ F −1 G−1 (X), we have
(∗) dFq (Tq M ) + TF (q) G−1 (X) = TF (q) N.
Note that for q ∈ F −1 G−1 (X), we have
dGF (q) (TF (q) N ) + TGF (q) X = TGF (q) P
from (†). From (∗), we have
dGF (q) (dFq (Tq M ) + TF (q) G−1 (X)) ⊆ d(GF )q (Tq (M )) + TGF (q) X,
and therefore
d(GF )q (Tq M ) + TGF (q) X = TGF (q) P,
which shows that GF is transverse to X.
For the converse, now assume that GF is transverse to X; the points p and q are as
above. We claim that
(dGF (q) )−1 (d(GF )q (Tq M ) + TGF (q) X) =

(dGF (q) )−1 (d(GF )q (Tq M )) + (dGF (q) )−1 (TGF (q) X).
To see this, first note that we can choose a basis for TGF (q) P of the form
{u1 , . . . , uk , v1 , . . . , vl },
where the vectors ui form a basis for TGF (q) X and the vectors v1 , . . . , vl are in d(GF )q (Tq M ).
Moreover, since G−1 (X) is a submanifold of N of codimension l, there is a basis
{x1 , . . . , xj , y1 , . . . , yl }
for TF (q) N such that
dGF (q) (xi ) = vi and dGF (q) (yi ) ⊆ span{u1 , . . . , uk },
where i runs over all the indices involved. The claim then follows, and from the assumption
that GF is transverse to X, we can conclude that
TF (q) N = (dGF (q) )−1 (d(GF )q (Tq M )) + (dGF (q) )−1 (TGF (q) X).
By the previous problem,
(dGF (q) )−1 (TGF (q) X) = TF (q) G−1 (X)
and it is true in general that
(dGF (q) )−1 (d(GF )q (Tq M )) ⊆ dFq (Tq M ).
Hence,
TF (q) N = dFq (Tq M ) + TF (q) G−1 (X)
holds and we conclude that F is transverse to G−1 (X).
Problem 6.12. Let M be a compact smooth n-manifold. Prove that if N > 2n, every
smooth map from M to RN can be uniformly approximated by smooth immersions.
Let F : M → RN be a smooth map, and consider the composition ι ◦ F : M → RN +1 ,

where ι includes RN into the first N coordinates of RN +1 . By Corollary 6.17, there is a
smooth embedding G : M ,→ RN +1 that uniformly approximates ι ◦ F .
Examining our solution of Problem 6.2, we see that there is a dense set of vectors in
RN +1 r RN such that the projection along these vectors restricts to an immersion of G(M ).
Hence, choose a vector v arbitrarily close to (0, . . . , 0, 1) so that πv ◦G is a smooth immersion
approximating F .
Problem 6.13. Let M be a smooth manifold. In this chapter, we defined what it means for
two embedded submanifolds of M to intersect transversely, and for a smooth map into M to
be transverse to an embedded submanifold. More generally, if F : N → M and F 0 : N 0 → M
are smooth maps into M , we say that F and F 0 are transverse to each other if for every
x ∈ N and x0 ∈ N 0 such that F (x) = F 0 (x0 ), the spaces dFx (Tx N ) and dFx0 0 (Tx0 N 0 ) together
span TF (x) M . Prove the following statements.
(a) With N , N 0 , F , F 0 as above, F and F 0 are transverse to each other if and only if the
map F ×F 0 : N ×N 0 → M ×M is transverse to the diagonal ∆M = {(x, x) : x ∈ M }.
(b) If S is an embedded submanifold of M , a smooth map F : N → M is transverse to
S if and only if it is transverse to the inclusion ι : S ,→ M .
Remark. There is a part (c) that appears in the text; however, it is incorrect as is pointed
out in the errata as of April 14, 2022.
(a) Let F and F 0 be transverse and let (x, x0 ) ∈ (F × F 0 )−1 (∆M ). Then F (x) = F 0 (x0 )
and so dFx (Tx N ) and dFx0 0 (Tx0 N 0 ) span TF (x) M . Then
d(F × F 0 )(x,x0 ) (T(x,x0 ) (N × N 0 )) + T(F (x),F (x)) (∆M )
= dFx (Tx N ) ⊕ dFx0 0 (Tx0 N 0 ) + ∆TF (x) M
where ∆TF (x) M = {(v, v) : v ∈ TF (x) M } 6 TF (x) M ⊕ TF (x) M = T(F (x),F (x)) (M × M ). Also
note that
dFx (Tx N ) ⊕ dFx0 0 (Tx0 N 0 ) + ∆TF (x) M = TF (x) M ⊕ TF (x) M
by transversality of F and F 0 . We conclude that F × F 0 is transverse to ∆M .
Conversely, assume that F ×F 0 is transverse to ∆M . Then the previous centred inequality
holds, and from here it is easy to deduce that dFx (Tx N ) and dFx0 0 (Tx0 N 0 ) span TF (x) M .
(b) F is transverse to S if and only if for every x ∈ F −1 (S),
dFx (Tx N ) + TF (x) S = TF (x) M
holds. But the left-hand side can be rewritten dFx (Tx N ) + dιF (x) (TF (x) S), and then the
condition is equivalent to F and ι being transverse.
Problem 6.14. This problem illustrates how badly Theorem 6.30 can fail if the transver-
sality hypothesis is removed. Let S = Rn × {0} ⊆ Rn+1 , and suppose A is an arbitrary
closed subset of S. Prove that there is a properly embedded hypersurface S 0 ⊆ Rn+1 such
that S ∩ S 0 = A.
By Theorem 2.29, there is a smooth nonnegative function f : S → R such that f −1 (0) =
A. Consider the map
F : S → Rn+1 , x 7→ (x, f (x)),
and note that F is a proper embedding by Proposition 5.7. Then S 0 = F (S) satisfies the
conditions in the problem statement.
Problem 6.15. Prove Theorem 6.32 (global characterization of graphs).
If (a) holds, then πM |S is a diffeomorphism onto M , which proves (b).

Suppose (b) holds. Then for each p ∈ M , the submanifolds S and {p} × N intersect only
at the point (p, (πM |S )−1 (p)), which we label x. To see that the intersection is transverse,
we note that d((πM |S )−1 )p (Tp (M )) = Tx S and that d(πM |S )p maps Tx ({p} × N ) to zero.
Thus, Tx (M × N ) = Tx S ⊕ Tx ({p} × N ), proving that S and {p} × N intersect transversely.
Thus, (c) holds.
Suppose (c) holds. Then πM |S : S → M is a smooth bijection, and the transversality
hypothesis implies that it is of constant full rank. Theorem 4.14 (the global rank theorem)
then gives that πM |S is a diffeomorphism. Thus, (b) holds.
Finally, suppose that (b) holds. Then S is the graph of the smooth map f : M → N, p 7→
(πM |S )−1 (p).
Problem 6.16. Suppose M and N are smooth manifolds. A class F of smooth maps from
N to M is said to be stable if it has the following property: whenever {Fs : s ∈ S} is
a smooth family of maps from N to M , and Fs0 ∈ F for some s0 ∈ S, then there is a
neighbourhood U of s0 in S such that Fs ∈ F for all s ∈ U . (Roughly speaking, a property of
smooth maps is stable if it persists under small deformations.) Prove that if N is compact,
then the following classes of smooth maps from N to M are stable:
(a) immersions,
(b) submersions,
(c) embeddings,
(d) diffeomorphisms,
(e) local diffeomorphisms,
(f ) maps that are transverse to a given properly embedded submanifold.
Let the smooth family F be parametrized by F : N × S → M . Fix charts for N , S,
and M and write F in coordinates: we see that the Jacobian of Fs depends smoothly on s.
Thus, if Fs0 has full rank at p, then Fs has full rank at x for all (x, s) ∈ O × U , where O and
U are open neighbourhoods of p and s0 , respectively. By compactness, we can cover N by
finitely many sets Oi , whereT Fs has full rank at x for all (x, s) ∈ Oi × Ui . Then Fs has full
rank on all of N for s ∈ i Ui . This proves that the classes of (a), (b), and (e) are stable.
Next, we prove (c), i.e. that embeddings are stable. We will show that there is a neigh-
bourhood U of s0 such that Fs is injective for all s ∈ U . This will suffice, since we know
immersions are stable and injective immersions on compact domains are embeddings. Sup-
pose there is no such neighbourhood U . Then there is a sequence (ti ) converging to s0 in
S such that Fti is not injective. Thus there are pairs (xi , yi ) ∈ N × N r ∆N such that
Fti (xi ) = Fti (yi ). By compactness of N , we can assume that the sequences (xi ) and (yi )
converge. Since Fs0 is injective, the sequences must converge to a common point x. Consider
the map
G : N × S → M × S, (p, s) 7→ (Fs (p), s).
Its differential at (x, s0 ) is of the form

d(Fs0 )x ∗
dG(x,s0 ) = ,
0 id
so dG(x,s0 ) is injective, since d(Fs0 )x is injective. But this implies that G is injective in a
neighbourhood of (x, s0 ), contradicting the work above. We conclude that the class of (c)
is stable.
For (d), let Fs0 be a diffeomorphism. By the previous paragraph, Fs is an embedding
for all s in some open neighbourhood of s0 . Then Fs (N ) is a compact, and hence closed,
submanifold of M . Since Fs0 is a diffeomorphism, N and M have equal dimensions. Since

Fs is an embedding, Fs (N ) is of codimension 0 in M , and therefore is open. The only
nonempty open and closed subset of M is M itself, so we conclude that the maps Fs are
bijective embeddings, and therefore diffeomorphisms.
Finally we prove (f). Let X be the given properly embedded submanifold, and suppose
that Fs0 is transverse to it. Let p ∈ Fs−1
0
(X) and let U be a chart centred at Fs0 (p) that is a
slice chart for X. More specifically, give U coordinates (x1 , . . . , xm ) and let X ∩U correspond
to the subset of points of the form (x1 , . . . , xk , 0, . . . , 0). The transversality assumption
guarantees that d(Fs0 )p (Tp N ) projects onto the n − k last coordinates of TFs0 (p) M . This
is an open condition, since it can be phrased equivalently as the nonvanishing of minors
in a matrix representation for d(Fs0 )p . Since d(Fs0 )p depends smoothly on s0 and p, we
conclude that there are open neighbourhoods U and O of s0 and p, respectively, such that
this holds. This implies that Fs |O is transverse to X. By compactness of N , we can cover
it with finitely many such neighbourhoods O, and thus conclude that Fs is transverse to X
for all s in some open neighbourhood of s0 .
Problem 6.17. Let ϕ : R → R be a compactly supported smooth function such that ϕ(0) = 1.
Use the family {Fs : s ∈ R} of maps from R to R given by Fs (x) = xϕ(sx) to show that the
classes of maps described in Problem 6.16 need not be stable when N is not compact.
We take s0 = 0 here, since in this case Fs0 = idR is a diffeomorphism. We will show that
for s 6= 0, Fs does not belong to any of the classes of (a), (b), (c), (d), (e) of Problem 6.16.
Note that
d
Fs = ϕ(sx) + sxϕ0 (sx),
dx
d
so for x large enough Fs (x) = 0, given that s 6= 0. This shows that Fs is neither an
dx
immersion nor a submersion, and proves the claim above.
Finally, note that Fs0 is transverse to 0 ∈ R. If s 6= 0, however, there are sufficiently large
x ∈ R such that Fs = 0 in a neighbourhood of x. We conclude that Fs is not transverse to
0 for s 6= 0.
7. Lie Groups
Exercise 7.2. If G is a smooth manifold with a group structure such that the map G×G → G
given by (g, h) 7→ gh−1 is smooth, then G is a Lie group.
The inversion map factors as
G → G × G → G, g 7→ (e, g) 7→ eg −1 = g
and is therefore smooth. Hence, the map G × G, (g, h) 7→ (g, h−1 ) is smooth. Then, the
multiplication map factors as
G × G → G × G → G, (g, h) 7→ (g, h−1 ) 7→ g(h−1 )−1 = gh
and is therefore smooth. We conclude that G is a Lie group.
Exercise 7.8. Complete the proof of Theorem 7.7 by showing that x−1 x = xx−1 = ee.
Consider the map f : G e x 7→ x−1 x and note that π ◦ f = ce , the constant map at
e → G,
e ∈ G, so f is a lift of ce . Moreover, cee is also a lift of ce and it agrees with f at ee. By the
unique lifting property, we conclude that f = cee, which shows that x−1 x = ee for all x ∈ G. e
−1
A similar argument shows that xx = ee.
Exercise 7.13. Given a group G and a subset S ⊆ G, show that the subgroup generated by
S is equal to the set of all elements of G that can be expressed as finite products of elements
of S and their inverses.
Let X denote the set of all elements of G that can be expressed as finite products of
elements of S and their inverses. Then X is clearly a subgroup and it contains S. On the
other hand, if H is a subgroup of G containing S, then it contains X, since H is closed
under inversion and multiplication. We conclude that X is the subgroup generated by S.
Exercise 7.20. Let S ⊆ T3 be the image of the subgroup H of Example 7.19 under the
obvious embedding T2 ,→ T3 . Show that S is a Lie subgroup of T3 that is not closed,
embedded, or dense; but its closure is a properly embedded Lie subgroup of T3 .
S is a Lie subgroup of T2 , which in turn is an embedded Lie subgroup of T3 ; hence, S is
a Lie subgroup of T3 . Since S is not closed in T2 , it is not closed in T3 . It is not embedded,
because it does not inherit a manifold topology from T2 , which inherits its topology from
T3 . Finally, S is not dense in T3 , since S ⊆ T2 , and T2 ⊆ T3 is closed. On the other hand,
the closure of S is T2 , which is a properly embedded Lie subgroup of T3 .
Exercise 7.31. Verify that (7.10) does indeed define a Lie group structure on the manifold
N × H, with (e, e) as identity and (n, h)−1 = (θh−1 (n−1 ), h−1 ).
The verification that the semidirect product N o H is indeed a group is tedious and we
will omit it. This is usually done in a course on abstract algebra. That the multiplication
and inversion are smooth follows from the fact that θ is a smooth left action. Thus, N o H
is indeed a Lie group.
(a) The subsets N e and He are closed since slices in product manifolds are closed. The
previous exercise shows that they are closed under multiplication. The obvious maps N → Ne
and H → H are Lie group isomorphisms.
e
(b) That Ne is normal in N o H follows from the fact that the elements of H “slide” past
those of N when multiplying pairs. More precisely, we have
(n0 , h0 ) · (n, eH ) · (n0 , h0 )−1 = (n0 θh0 (n), h0 ) · (θh−1 (n−1 −1 −1
0 ), h0 ) = (n0 θh0 (n)n0 , eH ).
0
Note that if N is Abelian then N

e is central in N o H.
(c) It is obvious that N ∩ H = {(e, e)}, and
e e
(n, h) = (n, e) · (e, h)
shows that N
eHe = N o H.
Problem 7.1. Show that for any Lie group G, the multiplication map m : G × G → G is a
smooth submersion.
Recall that Theorem 4.26 (the local section theorem) states that a smooth map is a
submersion if and only if every point of the domain is in the image of a local section. In the
caseof Lie group multiplication, we can actually construct global sections as follows. Let
(g, h) ∈ G × G be arbitrary. Then the map G → G × G, x 7→ (xh−1 , h) is a smooth section
of m and (g, h) is the image of gh. We conclude that m is a smooth submersion.
Problem 7.2. Let G be a Lie group.
(a) Let m : G×G → G denote the multiplication map. Using Proposition 3.14 to identify
T(e,e) (G × G) with Te G ⊕ Te G, show that the differential dm(e,e) : Te G ⊕ Te G → Te G
is given by
dm(e,e) (X, Y ) = X + Y.
(b) Let i : G → G denote the inversion map. Show that die : Te G → Te G is given by
die (X) = −X.
(a) The isomorphism Te G ⊕ Te G ∼ = T(e,e) (G × G) is induced by the inclusions ι1 , ι2 : G →
G × G, where ι1 (g) = (g, e) and ι2 (g) = (e, g). By functoriality, the diagrams
ιj d(ιj )e
G G×G Te G Te G ⊕ Te G
m and dm(e,e)
id id
G Te G
commute for j = 1, 2. It follows that dm(e,e) (X, 0) = X and dm(e,e) (0, Y ) = Y . By linearity,
we conclude that dm(e,e) (X, Y ) = X + Y .
(b) The constant map ce at e factors as
m
G → G × G −→ G, g 7→ (g, g −1 ) 7→ e.
This induces the zero map on tangent spaces, which factors as
dm(e,e)
Te G → Te G ⊕ Te G −−−−−→ Te G, X 7→ (X, die (X)) 7→ X + die (X) = 0
by functoriality and by part (a). We conclude that die (X) = −X.
Problem 7.3. Our definition of Lie groups includes the requirement that both the multipli-
cation map and the inversion map are smooth. Show that smoothness of the inversion map
is redundant: if G is a smooth manifold with a group structure such that the multiplication
map m : G × G → G is smooth, then G is a Lie group.
It is not hard to see that F : G × G → G × G, (g, h) 7→ (g, gh) is bijective; F is smooth
since m is smooth. Fixing coordinates, we see that

id 0
dF(g,h) = ,
∗ d(mg )h
where mg : G → G is defined by mg (h) = gh. Note that mg is a diffeomorphism with inverse
mg−1 (this does not depend on the inversion being smooth). Thus, dF(g,h) has full rank for
all points (f, g). By Theorem 4.14 (the global rank theorem), F is a diffeomorphism, as it
is a bijection with constant rank. Thus, the inverse
F −1 : G × G → G × G, (g, h) 7→ (g, g −1 h)
is smooth. Hence, the inversion is smooth, since it can be written as the composition
F −1 π
2
G → G × G −−−→ G × G −→ G, g 7→ (g, e) 7→ (g, g −1 ) 7→ g −1 .
Problem 7.4. Let det : GL(n, R) → R denote the determinant function. Use Corollary
3.25 to compute the differential of det, as follows.
(a) For any A ∈ M(n, R), show that

d
det(In + tA) = tr A,
dt t=0

where tr Aij = i Aii is the trace of A.
P
(b) For X ∈ GL(n, R) and B ∈ TX GL(n, R) ∼= M(n, R), show that

d(det)X (B) = (det X) tr X −1 B .

(a) Recall that

n
X Y
det A = sgn(σ)Aiσ(i) .
σ∈Sn i=1
Hence, det(In + tA) is some linear polynomial
Qn in t, so its derivative at 0 is just the coefficient
of the linear term. In every product i=1 sgn(σ)Aiσ(i) , if there is one off-diagonal term
Aiσ(i) , then there must be a second off-diagonal term Ajσ(j) . Thus, the only linear term in
Qn
det(In + tA) comes from i=1 (1 + tAii ), and the linear term of this product is tr(A).
By Corollary 3.25, the differential of a smooth map at a point p can be computed by
calculating its derivative along a smooth curve at p. For any matrix B ∈ TX GL(n, R), for
small enough values of t we have that X + tB ∈ GLn (R). Moreover, the derivative of the
curve X + tB is B. Hence, by Corollary 3.25, we have

d
d(det)X (B) = det(X + tB)
dt t=0

d
det(X) det In + tX −1 B

=
dt t=0
= det(X) tr X −1 B ,

where the last line follows from part (a).

Problem 7.5. Prove Theorem 7.9 (uniqueness of universal covering group).
We use the notation in the statement of Theorem 7.9. By the lifting criterion, there are
continuous maps f and f 0 making the diagrams
e0
G G
e
f and f0
π0 π
G
e G e0
G G
π π0
commute, and moreover f and f 0 preserve identity elements. Note that f and f 0 are
smooth, since π and π 0 are smooth covering maps and thus local diffeomorphisms (see
Exercise 4.10(a)). Note that f 0 ◦ f is a lift of π under π such that (f 0 ◦ f )(e e) = ee. By the
uniqueness of lifts, f 0 ◦ f = idGe . Similarly, f ◦ f 0 = idGe0 , so f is a diffeomorphism with
inverse f 0 .
It remains to show that f is a group homomorphism. The diagram
f ◦m
e×G
G e e
e0
G
π×π π0
G×G m G
commutes, where m and m e denote multiplication maps. Moreover, the square still commutes
after replacing f ◦ m
e with the map F : G e×G e →G e 0 , (g, h) 7→ f (g)f (h). Since f ◦ m
e and
F agree at the point (e e, ee), they must be equal by uniqueness of lifts. Thus, f (gh) =
(f ◦ m)(g,
e h) = F (g, h) = f (g)f (h) for all g, h ∈ G.
e
Problem 7.6. Suppose G is a Lie group and U is any neighbourhood of the identity. Show
that there exists a neighbourhood V of the identity such that V ⊆ U and gh−1 ∈ U whenever
g, h ∈ V .
For each n ∈ N, let Vn be a neighbourhood of e and assume that if gn ∈ Vn for each n ∈ N,
then (gn ) converges to e. For example, we could take the sets Vn to be shrinking open balls
centred at the identity. Now, assume that for each n ∈ N, there is a pair (gn , hn ) ∈ Vn × Vn
such that gn h−1
n ∈/ U . Then ((gn , hn )) converges to (e, e), but gn h−1
n does not converge to
−1
e = ee , violating the continuity of the inverse and multiplication functions.
Remark. Note that we could have done this problem with more general words over the
Lie group G: a word over G is a function G × · · · × G → G which is the composition of
some product of identity and inversion maps with the multiplication map. For instance
G × G → G, (g, h) 7→ gh−1 is a word, as is G × G × G → G, (g1 , g2 , g3 ) 7→ g1−1 g2−1 g3 . The
number of factors of G in the domain is called the length of the word. Note that this is an
unusual definition of words. Our solution to Problem 7.6 applies to all words over G, not
just (g, h) 7→ gh−1 .
Problem 7.7. Prove Proposition 7.15 (properties of the identity component of a Lie group).
Let G be a Lie group and let G0 be its identity component.
Claim 1. G0 is a normal subgroup of G. We first show that G0 is a subgroup. Let
g, h ∈ G0 . Since the components and path-components of manifolds (of CW complexes more
generally) coincide, there are paths γg and γh from e to g and from e to h, respectively.
Then
m
I → G × G −→ G, t 7→ (γg (t), γh (t)) 7→ m(γg (t), γh (t))
is a path from e to gh, showing that G0 is closed under multiplication. To see that G is
closed under inversion, note that i ◦ γg is a path from e to g −1 . To see that G0 is normal,
let s ∈ G. Then sγg s−1 is a path from e to sgs−1 .
Claim 2. G0 is the only connected open subgroup of G. Let H be a connected open sub-
group of G. Then H 6 G0 , since H contains the identity and is connected. By Proposition
7.14(c), H generates G0 . But then H = G0 since H is a subgroup.
Claim 3. Every connected component of G is diffeomorphic to G0 . Let C be a connected
component of G, and let g ∈ C. If h ∈ G0 , then there is a path γ from e to h. Thus, gγ is a
path from g to gh. Hence, left multiplication by g induces a smooth map ϕ : G0 → C, which
we claim is a diffeomorphism. Being the restriction of a bijective map, ϕ is injective. Now
let s ∈ C be arbitrary and let τ be a path from g to s. Then g −1 τ is a path from e to g −1 s.
Since ϕ(g ∈ s) = s, this shows that ϕ is surjective. Its inverse is given by left multiplication
by g −1 , which is also smooth. We thus conclude that G0 and C are diffeomorphic. Note
that we have shown that the connected components of G are in fact the cosets of G0 in G.
Problem 7.8. Suppose a connected topological group G acts continuously on a discrete
space. Show that the action is trivial.
The map ρ : G × K → K defining the G-action is continuous by assumption. Images of
connected sets are connected. Since G is connected and the only connected subsets of K
are singletons, it follows that ϕ is constant on subsets of the form G × {x}, where x ∈ K is
arbitrary. Since e · x = x, it follows that ϕ(G × {x}) = {x}; in other words, the action is
trivial.
Problem 7.9. Show that the formula
A · [x] = [Ax]
defines a smooth, transitive left action of GL(n + 1, R) on RPn .

Suppose [x] = [y], so that there is some α 6= 0 such that x = αy. Then Ax = A(αy) =
αAy, showing that [Ax] = [Ay]. Hence, the action is well-defined. Let π : Sn → RPn be the
double cover. Define an action of GL(n + 1, R) on Sn by A · x = Ax/kAxk; the action is
smooth since matrix multiplication and taking norms are smooth operations. The diagram
GL(n + 1, R) × Sn Sn
id ×π π
n
GL(n + 1, R) × RP RPn
commutes, where the horizontal maps are the maps defining the GL(n + 1, R) actions. Since
each of the vertical maps are local diffeomorphisms and the the top map is smooth, we
conclude that action on RPn is smooth (see Exercise 4.10(b)). Finally, to see that the action
is transitive, let x ∈ Rn be any nonzero vector. Then there is a matrix A ∈ GL(n + 1, R)
whose first column is x, and therefore Ae1 = x, where e1 = (1, 0, . . . , 0). This proves the
stronger statement that GL(n + 1, R) acts transitively on the nonzero vectors of Rn+1 .
Problem 7.10. Repeat Problem 7.9 for GL(n + 1, C) and CPn .
That the action is well-defined and transitive is proved just as in Problem 7.9. For
smoothness, the diagram
f
GL(n + 1, C) × (Cn+1 r {0}) Cn+1 r {0}
id ×π π
n
GL(n + 1, C) × CP CPn
commutes, where π is the standard projection map and the vertical maps are the action
defining maps. By Problem 4.5(b), π is a smooth submersion. By Theorem 4.26 (the local
section theorem), at each point p ∈ GL(n + 1, C) × CPn has an open neighbourhood such
that there is a smooth section σ : U → GL(n + 1, C) × (Cn+1 r {0}) of id ×π. Then the
restriction of the bottom map to U is π ◦ f ◦ σ, which is smooth. Hence, the action is locally
smooth and therefore smooth (since smoothness is a local condition).
Problem 7.11. Considering S2n+1 as the unit sphere in Cn+1 , define an action of S1 on
S2n+1 , called the Hopf action, by
z · (w1 , . . . , wn+1 ) = (zw1 , . . . , zwn+1 ).
Show that this action is smooth and its orbits are disjoint unit circles in Cn+1 .
The action map S1 × S2n → S2n is the restriction of another action map C × C2n+1 →
2n+1
C , again given by complex scalar multiplication. This is clearly smooth, and by the
restriction results of Chapter 5 (Theorem 5.27 and Corollary 5.30), we have that S1 × S2n →
S2n is smooth.
The orbits of the action are of the form {eit w : w ∈ S2n+1 }, which are unit circles in
the plane spanned by the R-linearly independent vectors w and eiπ/2 w. The orbits of any
G-action always partition the G-set, so we know that these circles are disjoint for this very
general reason.
Problem 7.12. Use the equivariant rank theorem to give another proof of Theorem 7.5
by showing that every Lie group homomorphism F : G → H is equivariant with respect to
suitable smooth G-actions on G and H.
The actions are G × G → G, (s, t) 7→ st and G × H → H, (g, h) 7→ F (g)h. These are

smooth by smoothness of F and of Lie group multiplication, and it is easy to check that the
actions are transitive with respect to F . We conclude that any Lie group homomorphism
has constant rank.
Problem 7.13. For each n > 1, prove that U(n) is a properly embedded n2 -dimensional
Lie subgroup of GL(n, C).
Note that U(n) is the level set Φ−1 (In ), where Φ is the map
Φ : GL(n, C) → M(n, C), A 7→ A∗ A.
It is clear that Φ is smooth; we will show that it has constant rank, which will imply that
U(n) is a smooth submanifold by Theorem 5.12 (the constant-rank level set theorem). To
see that Φ has constant rank, note that it is GL(n, C)-equivariant with respect to the smooth
(right) actions
GL(n, C) × GL(n, C) → GL(n, C), (A, B) 7→ AB
and
M(n, C) × GL(n, C) → M(n, C), (A, B) 7→ B ∗ AB.
This shows that U(n) is a properly embedded submanifold of GL(n, C).
To find the dimension of U(n), it suffices to determine the rank of Φ; since Φ has constant
rank, it suffices to determine its rank at In . Let B ∈ M(n, C) and consider the curve defined
by γ(t) = In + tB; for small values of t, the image of γ lies in GL(n, C). Then

d d
dΦIn (B) = Φ(γ(t)) = (In + tB)∗ (In + tB) = B ∗ + B,
dt t=0 dt t=0
so the image of dΦIn (B) lies in the subgroup of Hermitian matrices. On the other hand, if B
is a given Hermitian matrix, then dΦIn ( 21 B) = B, so the image of dΦIn (B) is the subgroup
of Hermitian matrices, which has dimension n2 . Thus, the codimension of U(n) in GL(n, C)
is n2 . Since GL(n, C) has dimension 2n2 , we conclude that U(n) has dimension n2 .
Problem 7.14. For each n > 1, prove that SU(n) is a properly embedded (n2 − 1)-
dimensional Lie subgroup of U(n).
The determinant map det : U(n) → S1 , A 7→ det(A) is a homomorphism of Lie groups,
and therefore has constant rank. Here we are viewing S1 as embedded in C and with group
structure given by complex multiplication. Moreover, SU(n) = det−1 ({1}), so SU(n) is a Lie
subgroup of U(n) by Theorem 5.12 (the constant-rank level set theorem). The determinant
map is not locally constant, and since S1 is 1-dimensional we conclude that det has full rank
equal to 1. Thus, SU(n) is an (n2 − 1)-dimensional Lie subgroup of U(n) (see the previous
problem for the calculation of the dimension of U(n)).
Problem 7.15. Show that SO(2), U(1), and S1 are all isomorphic as Lie groups.
Regard S1 as the unit-norm complex numbers in C. By definition, U(1) consists of all
1 × 1 complex matrices. There is an obvious diffeomorphism between S1 and U(1). The
group SO(2) is the group of orientation and inner product preserving linear transformations
of R2 . These are exactly the rotations of R2 about (0, 0) and thus, SO(2) consists of all
matrices of the form
cos θ sin θ
− sin θ cos θ
where θ ∈ R. Thus, there is a map

1 iθ cos θ sin θ
f : S → SO(2), e 7→
− sin θ cos θ
which is well-defined and easily shown to be smooth (since the trigonometric functions
are known to be smooth). Moreover, f is bijective and has constant rank, since it is a
homomorphism of Lie groups. Theorem 4.14 (the global rank theorem) implies that f is a
diffeomorphism.
Problem 7.16. Show that SU(2) is diffeomorphic to S3 .
2 2
We will use the definition S3 = {(z1 , z2 ) ∈ C2 : |z1 | + |z2 | = 1}. The map

z −z2
f : S3 → SU(2), (z1 , z2 ) 7→ 1
z2 z1
is a smooth injection. We claim it is also surjective. Let

a b
A= ∈ SU(2).
c d
Then we have the equations

aa + bb ac + bd 1 0
A∗ A = =
ca + db cc + dd 0 1
and det A = ad − bc = 1. Multiplying the determinant equation by b and using the self-
adjointness relations, we obtain
b = adb − bbc = −aac − bbc = −c.
If c 6= 0, then ca + db = 0 implies a = d. If c = 0, then aa = dd = ad = 1, which implies
that a = d. We have thus shown that A = f (a, −d). The inverse map is then given by

a b
7→ (a, −b),
c d
which is also smooth by the restriction results of Chapter 5. We conclude that S3 is diffeo-
morphic to SU(2). This endows S3 with a Lie group structure.
Problem 7.17. Determine which of the following Lie groups are compact:
GL(n, R), SL(n, R), GL(n, C), SL(n, C), U(n), SU(n).
We begin with the case n = 1. First, GL(1, R) ∼ = R∗ and GL(n, C) ∼= C∗ are noncompact.
On the other hand, SL(1, R) ∼ = SL(n, C) ∼= {1} is compact. As shown in Problem 7.15,
U(1) ∼ = S1 and is therefore compact. Since SU(1) is a closed Lie subgroup of U(1), it is also
compact.
Now assume n > 2. First note that SL(n, R) contains matrices with coefficients that are
1
arbitrarily large in absolute value. Hence, it is not bounded as a subset of Rn and hence
noncompact. Since SL(n, R) is a closed Lie subgroup of GL(n, R), GL(n, C), and SL(n, C),
all of these Lie groups are noncompact. To see that U(n) is compact, note that it is a closed
and bounded subset of Cn ∼
2 2
= R2n . It is closed because it can be expressed as a level set
and it is bounded because the (complex) norm of every column is equal to 1.
Problem 7.18. Prove Theorem 7.35 (characterization of semidirect products).
We use the notation in the theorem statement given in the text. The map (n, h) 7→ nh is
smooth, since multiplication in G is smooth. It is surjective since N H = G and it is injective
since N ∩ H = {e}. Checking that the map is a homomorphism is a straightforward exercise
in the definition of the semidirect product. Since Lie group homomorphisms have constant
rank (Theorem 7.5) and constant rank smooth bijections are diffeomorphisms (Theorem
4.14), the (n, h) 7→ nh is a diffeomorphism.
Problem 7.19. Suppose G, N , and H are Lie groups. Prove that G is isomorphic to a
semidirect product N o H if and only if there are Lie group homomorphisms ϕ : G → H and
ψ : H → G such that ϕ ◦ ψ = idH and ker ϕ = N .
There is an obvious split short exact sequence
1→N →N oH →H →1
of Lie groups, where the splitting is given by h 7→ (e, h). Thus, if G ∼
= N o H, then the
required maps ϕ and ψ exist.
Conversely, suppose there is a split short exact sequence 1 → N → G → H → 1 of Lie
groups; we will identify N as a subgroup of G. Then there is a map
f : N o H → G, (n, h) 7→ nψ(h),
which is a group homomorphism when we make H act on N via h · n := ψ(h)nψ(h)−1 . This
is a left action and it is well-defined because N is normal in G. For any g ∈ G, the element
ψ(ϕ(g))g −1 ∈ ker φ = N . Thus, the image of ψ contains a representative of each coset of
N , which shows that f is onto. To see that f is injective, suppose that nψ(h) = e. Then
ψ(h) ∈ N , so e = ϕ(ψ(h)) = h, which in turn implies n = e. Thus, f is an isomorphism.
Problem 7.20. Prove that the following Lie groups are isomorphic to semidirect products
as shown.
(a) O(n) ∼
= SO(n) o O(1).
(b) U(n) ∼
= SU(n) o U(1).
(c) GL(n, R) ∼
= SL(n, R) o R∗ .
∼
(d) GL(n, C) = SL(n, C) o C∗ .
det
In all cases, the short exact sequence of interest is of the form 1 → SA → A −−→ B → 1,
where the first map is the inclusion and the second map is the determinant map. The
splittings can all be constructed similarly: map b ∈ B to the diagonal matrix in A whose
diagonal entries are all 1, except the (1, 1) entry which is b. Here, we identify O(1) with
{±1} and U(1) with S1 ⊆ C.
Problem 7.21. Prove that when n > 1, none of the groups in Problem 7.20 are isomorphic
to direct products of the indicated groups.
(a) When n is odd, we do have an isomorphism O(n) ∼ = SO(n) × O(1), so there is a
mistake in the statement of the problem. On the other hand, when n is even, Z(O(n)) =
{±In } = Z(SO(n)) = {±In }. It follows that Z(O(n)) and Z(SO(n) × O(n)) have different
orders, and therefore that O(n) is not isomorphic to the direct product SO(n) when n is
even.
(b) Z(U(n)) = {λIn : λ ∈ S1 } ∼ = S1 and Z(SU(n)) = {λIn : λn = 1} ∼ = Z/n. It
is therefore enough to show that S is not isomorphic to Z/n × S1 . Indeed, S1 has the
1
property that for any x ∈ S1 and any m ∈ Z, there is a y ∈ S1 such that y m = x. The same
property is not true of Z/n × S1 .
(c) When n is odd, we have that GL(n, R) ∼ = SL(n, R) × R∗ , so again there is an error
here in the statement of the problem. When n is even, Z(GL(n, R)) = {λIn : λ 6= 0} ∼ = R∗ ,
∗
while Z(SL(n, R)) = {±In }. Therefore, Z(SL(n, R) × R ) has multiple elements of order 2,
while Z(GL(n, R)) only has one.
(d) Z(GL(n, C)) ∼= C∗ , while Z(SL(n, C) × C∗ ) ∼= Z/n × C∗ , and these groups are not
isomorphic. For instance we can take kth roots of all elements in C∗ for all k, while this is
not true in Z/n and therefore neither in Z/n × C∗ .
Problem 7.22. (a) Show that quaternionic multiplication is associative but not com-
mutative.
(b) Show that (pq)∗ = q ∗ p∗ for all p, q ∈ H.
(c) Show that hp, qi = 21 (p∗ q + q ∗ p) is an inner product on H, whose associated norm
satisfies |pq| = |p| · |q|.
(d) Show that every nonzero quaternion has a two-sided multiplicative inverse given by
p−1 = |p|−2 p∗ .
(e) Show that the set H∗ of nonzero quaternions is a Lie group under quaternionic
multiplication.
(a) Associativity follows from the associativity of the multiplication on {1, i, j, k}. To see
that the multiplication is noncommutative, simply note that, for example, ij = −ji.
(b) Let p = (a, b) and q = (c, d). Then
q ∗ p∗ = (c, d)∗ (a, b)∗
= (c, −d)(a, −b)
= (ac − bd, −bc − ad)
= (ac − db, −(ad + cb))
= (pq)∗ .
(c) It is obvious that h·, ·i is linear, since the operation (·)∗ is linear. Next,
1 ∗ 1
(q p + p∗ q) = (p∗ q + q ∗ p) = hp, qi,
hq, pi =
2 2
so the inner product is symmetric. Next, if p = (a, b), then
p∗ p = (a, −b)(a, b) = (aa + bb, 0) = (|a|2 + |b|2 , 0),
whence it is easy to see that the inner product is positive definite. Finally,
2 2 2 2
|pq| = hpq, pqi = (pq)∗ pq = q ∗ p∗ pq = q ∗ |p| q = |p| |q| .
−2 −2 −2 2 −2 2 −2
(d) We have p · |p| p∗ = |p| · pp∗ = |p| |p∗ | = |p| |p| = 1. Showing that |p| p∗
is a right inverse is similar.
(e) The set H∗ is naturally identified with R4 r {0}, on which the operations of multipli-
cation and inversion are clearly smooth. Therefore, H∗ has the structure of a Lie group.
Problem 7.23. Let H∗ be the Lie group of nonzero quaternions (Problem 7.22), and let
S ⊆ H∗ be the set of unit quaternions. Show that S is a properly embedded Lie subgroup of
H∗ , isomorphic to SU(2).
It is clear that S is indeed a subgroup of H, since the norm is multiplicative by part (c)
of Problem 7.22 and the inverse of a unit norm element is also of unit norm by part (d) of
Problem 7.22. By Proposition 7.11, we conclude that S is a Lie subgroup of H∗ . It is also

clear that S is properly embedded since it is a compact subset of H∗ .
The isomorphism is given by

a −b
S → SU(2), (a, b) 7→ .
b a
We omit the proof that this is an isomorphism. The map is clearly injective. To see that it
is surjective, let
a b
A= ∈ SU(2).
c d
It is then not hard to see that the condition A∗ A = I2 implies that d = a and b = −c.
Problem 7.24. Prove that each of the maps τdn : GL(n, R) → GL(Pdn ) described in Example
7.36(g) is a faithful representation of GL(n, R).
Denote the variables of the polynomials in Pdn by x1 , . . . , xn and let A ∈ GL(n, R) be
a matrix such that τdn (A) = In . Let p(x1 , . . . , xn ) = xi . Then xi = τdn (A)p = (p ◦ A−1 ),
which implies that A(xi ) = xi . We conclude that A is the identity and therefore that τdn is
a faithful representation.
8. Vector Fields
(a) If (x, X(x)) and (x, Y (x)) are coordinate representations of X in Y in some chart
of M , then f X + gY is represented by (x, f (x)X(x) + g(x)Y (x)), which is smooth. Thus,
f X + gY is smooth by Proposition 8.1.
(b) This is obvious given (a).
Exercise 8.18. Prove the claim in Example 8.17 in two ways: directly from the definition,
and using Proposition 8.16.
With the standard coordinates (t) and (x, y) on R and R2 , respectively, we obtain

d ∂ ∂
dFt0 = − sin(t0 ) + cos(t0 ) = YF (t0 ) ,
dt ∂x ∂y
which shows that X and Y are F -related.
Now let’s use Proposition 8.16. Let f : V → R be a smooth function, where V is an open
subset of N . Then

d
X(f ◦ F )(t0 ) = f (cos t, sin t)
dt t=t0

∂f ∂f
=− sin(t0 ) + cos(t0 )
∂x F (t0 ) ∂y F (t0 )
= (Y f )(F (t0 )).
Exercise 8.29. Prove part (d) of Proposition 8.28.
Let h ∈ C ∞ (M ). Then
[f X, gY ]h = f X(gY h) − gY (f Xh)
= (f Xg)Y h + f gXY h − (gY f )Xh − f gY Xh
= f g[X, Y ]h + (f Xg)Y h − (gY f )Xh,
and the claim follows.

Exercise 8.34. Verify that the kernel and image of a Lie algebra homomorphism are Lie
subalgebras.
Let A : g → h be a Lie algebra homomorphism. Let X, Y ∈ ker A. Then A[X, Y ] =
[AX, AY ] = [0, 0] = 0, which shows that ker A is closed under the Lie bracket. Moreover,
since [AX, AY ] = A[X, y], the image of A is also closed under the bracket. Thus, the kernel
and image of A are Lie subalgebras of g and h, respectively.
Exercise 8.35. Suppose that g and h are finite-dimensional Lie algebras and A : g → h
is a linear map. Show that A is a Lie algebra homomorphism if and only if A[Ei , Ej ] =
[AEi , AEj ] for some basis (E1 , . . . , En ) of g.
If A is a Lie algebra homomorphism, then Pn clearly A[Ei , Ej ] P= [AEi , AEj ] for any basis
n
(E1 , . . . , En ). For the converse, let X = i=1 αi Ei and Y = j=1 βj Ej be vectors in g.
Then X X
[AX, AY ] = αi βj [AEi , AEj ] = A [αi Ei , βj Ej ] = A[X, Y ],
i,j i,j
so A is a Lie algebra homomorphism.

Exercise 8.43. Prove Corollary 8.42 by choosing a basis for V and applying Proposition
8.41.
Choosing a basis for V determines an isomorphism V → Rn , which in turn induces the
vertical Lie algebra isomorphism in the following diagram
Lie(GL(V )) Tid GL(V ) gl(V )
Lie(GL(n, R)) TIn GL(n, R) gl(n, R) .
which commutes by naturality of the horizontal maps. The bottom maps are all isomor-
phisms by Proposition 8.41, and therefore the top maps are all isomorphisms.
Problem 8.1. Prove Lemma 8.6 (the extension lemma for vector fields).
For each p ∈ A, let Up ⊆ M be a neighbourhood of p such that there is a smooth vector
field X
ep on Up satisfying Xep |U ∩A = X|U ∩A . By shrinking the sets Up , we can assume that
p p
Up ⊆ U for all p ∈ A. Let
{ϕp : Up →: p ∈ A} ∪ {ψ : M r A → R}
be a partition of unity subordinate to the open cover {Up : p ∈ A} ∪ {M r A}. Then the
vector field X
Xe= ϕp X
ep
p∈A
is a smooth extension of X whose support is contained in U .

Problem 8.2 (Euler’s Homogeneous Function Theorem). Let c be a real number, and let
f : Rn r {0} → R be a smooth function that is positively homogeneous of degree c, meaning
that f (λx) = λc f (x) for all λ > 0 and x ∈ Rn r {0}. Prove that V f = cf , where V is the
Euler vector field defined in Example 8.3.
On one hand, we have

d ∂ ∂ 1
f (λx) = x1 · f + · · · + x n · f = Vλx f
dλ ∂x1 λx ∂xn λx λ
for all λ > 0 and all x ∈ Rn r {0}. On the other hand,
d d c
f (λx) = λ f (x) = cλc−1 f (x)
dλ dλ
and therefore
Vλx f = cλc f (x) = cf (λx),
proving the claim.
Problem 8.3. Let M be a nonempty positive-dimensional smooth manifold with or without
boundary. Show that X(M ) is infinite-dimensional.
Since M is positive-dimensional, there is
• a sequence A1 ⊂ U1 ⊂ A2 ⊂ U2 ⊂ · · · , where each Ai (resp. Ui ) is an open
(resp. closed) ball, and
• a sequence of vector fields X 1 , X 2 , . . . such that X i is supported in Ui and is non-
vanishing on Ai for every i.
Now suppose that there is some n and there are constants αi ∈ R such that
α1 X 1 + · · · + αn X n = 0.
We will prove that αi = 0 for each i by induction on n. If n = 1, then this is clear. Let
p ∈ An r Un−1 ; evaluating the previous line at p yields αn Xpn = 0, and since Xpn 6= 0, we
conclude that αn = 0. But then
α1 X 1 + · · · + αn−1 X n−1 = 0,
so the claim follows by induction. We conclude that {Xi : i ∈ N} is a linearly independent
set in X(M ), and therefore that X(M ) is infinite-dimensional.
Problem 8.4. Let M be a smooth manifold with boundary. Show that there exists a global
smooth vector field on M whose restriction to ∂M is everywhere inward-pointing, and one
whose restriction to ∂M is everywhere outward-pointing.
We will only construct the inward-pointing vector field, as the construction of the outward-
pointing vector field is completely analogous. For each p ∈ ∂M let Up be a neighbourhood
of p that is diffeomorphic to Hn . Let X p be an inward-pointing vector field on Up and
let {ϕp : p ∈ ∂M } be a partition of unity subordinate to {Up : p ∈ ∂M }. Let X be the
restriction of X
ϕp X p
p∈∂M
to ∂M . Then X is an inward-pointing smooth vector field along ∂M . Since ∂M is closed
in M , we can extend it to an inward pointing smooth vector field on M by Proposition 8.6.
Problem 8.5. Prove Proposition 8.11 (completion of local frames).
(a) Let p ∈ U and let Xk+1 , . . . , Xn be smooth vector fields on M such that
{X1 |p , . . . , Xn |p }
spans Tp M (we can arrange this using Proposition 8.6, for example). We claim that
{X1 |q , . . . , Xn |q } spans Tq M for all q in some neighbourhood of p. Fix coordinates (xi )
on some neighbourhood V of p and let A be the matrix with ith column Xi (x). Since
{X1 (x), . . . , Xn (x)} is linearly independent if and only if det A(x) 6= 0. Since the deter-
minant map is continuous (it is in fact smooth), we conclude that the Xi are linearly
independent (and therefore spanning) in some neighbourhood of p.
(b) Let X1 , . . . , Xk ∈ X(M ) be such that Xi |p = vi for all i = 1, . . . , k. Using a similar
argument as in (a) one shows that the vector fields Xi are linearly independent on a neigh-
bourhood of p. To give a few more details, put coordinates on an open neighbourhood of
p, that the linear independence of the Xi is detected by the nonvanishing of minors in a
matrix, and conclude that linear independence is an open condition. By part (a), we can
extend (X1 , . . . , Xk ) to a local frame at p.
(c) By Lemma 8.6, each Xi has an extension X ei whose support is contained in some open
T e1 |p , . . . , X
en |p ) is
neighbourhood Ui of A. Let U = i Ui . Since the set of points p where (X
linearly independent is open (again using similar arguments as above) we conclude that the
vector fields X ei form a local frame on some neighbourhood of A.
Problem 8.6. Let H be the algebra of quaternions and let S ⊆ H be the group of unit
quaternions (see Problems 7.22 and 7.23).
(a) Show that if p ∈ H is imaginary, then qp is tangent to S at each q ∈ S.
(b) Define vector fields X1 , X2 , X3 on H by
X1 |q = qi, X2 |q = qj, X3 |q = qk.
Show that these vector fields restrict to a smooth left-invariant global frame on S.
(c) Under the isomorphism (x1 , x2 , x3 , x4 ) ↔ x1 1 + x2 i + x3 j + x4 k between R4 and H,
show that these vector fields have the following coordinate representations:
∂ ∂ ∂ ∂
X1 = −x2 + x1 2 + x4 3 − x3 4
∂x1 ∂x ∂x ∂x
∂ ∂ ∂ ∂
X2 = −x3 1 − x4 2 + x1 3 + x2 4
∂x ∂x ∂x ∂x
∂ ∂ ∂ ∂
X3 = −x4 1 + x3 2 − x2 3 + x1 4 .
∂x ∂x ∂x ∂x
(a) It is enough to prove the claim for p = i, j, or k, since the sum of three vectors tangent
to S at p will also be a tangent vector. We only do the case p = i, since the other two cases
are similar. Let q = a + bi + cj + dk. Then qi = −b + ai + dj − cj, and
(a, b, c, d) · (−b, a, d, −c) = 0
(the product here is the standard dot product), which show that q and qi are perpendicular.
Thus qi is tangent to S at q.
(b) Since S is a 3-dimensional manifold, it suffices that show that (X1 |p , X2 |p , X3 |p ) is
linearly independent at each point p ∈ S. Suppose that α1 qi + α2 qj + α3 qk = 0 for some
α1 , α2 , α3 ∈ R. Then q(α1 i + α2 j + α3 k) = 0, and multiplying on the left by q −1 gives
α1 i + α2 j + α3 j = 0, whence we conclude that α1 = α2 = α3 = 0.
(c) IF q = x1 + x2 i + x3 j + x4 k, then qi = −x2 + x1 i + x3 j − x3 j, which corresponds to
the vector
∂ ∂ ∂ ∂
−x2 1 + x1 2 + x4 3 − x3 4 .
∂x ∂x ∂x ∂x
This shows that X1 has the formed claimed in the problem statement. Using the same ideas
one shows that the other two vector fields have the required form.
Problem 8.7. The algebra of octonions (also called Cayley numbers) is the 8-dimensional
real vector space O = H × H (where H is the space of quaternions defined in Problem 7.22)
with the following bilinear product:
(p, q)(r, s) = (pr − sq ∗ , p∗ s + rq), for p, q, r, s ∈ H.
Show that O is a noncommutative, nonassociative algebra over R, and prove there exists a
smooth global frame on S7 by imitating Problem 8.6 as you can.
Since H is noncommutative and H ⊆ O as a subalgebra, we conclude that O is non-
commutative. We will work with the following basis for O. First, we have the elements
1, i, j, k ∈ H ⊆ O, where H is embedded in O = H × H as the first factor. We complete the
basis with the elements
w = (0, 1), x = (0, −i), y = (0, −j), z = (0, −k).
The calculations
(ij)x = kx = (k, 0)(0, −i) = (0, ki) = (0, j) = −y
and
i(jx) = i((j, 0)(0, −i)) = (i, 0)(0, ji) = (i, 0)(0, −k) = (0, ik) = (0, −j) = y
show that O is nonassociative.
We will now show that the vectors Xa |q = qa, where a ∈ {i, j, k, w, x, y, z}, are tangent
to the unit sphere S ⊆ O. Let q = a1 + bi + cj + dk + ew + f x + gy + hz. Then
qi = −b1 + ai + dj − ck + f w − ex − hy + gz
and therefore the dot product q · qi vanishes, which shows that qi is tangent to S at q.
Similar computations prove the other tangencies.
We now show that the vector fields Xa form a global frame for S7 . It suffices to show
that that O has no zerodivisors. First, note that qq ∗ , q ∗ q ∈ R1 for ever q ∈ O, and that
qq ∗ = 0 if and only if q ∗ q = 0 if and only if q = 0. Now suppose that pq = 0 for some
p, q ∈ O. Then p∗ (pq) = (p∗ p)q = 0 (this follows from weak associativity of the octonions,
see the hint), which implies that either p∗ p = 0 or q = 0.
Problem 8.8. The algebra of sedenions is the 16-dimensional real vector space SO × O
with the product defined as in Problem 8.7 (i.e. apply the Cayley–Dickson construction to
the octonions to obtain the sedenions). Why does sedenionic multiplication not yield a global
frame for S15 ?
The fact that complex, quaternionic, and octonionic multiplication yield global frames
for S1 , S3 , and S7 , respectively, is due to the fact that C, H, and O have no zerodivisors.
It is well-known that the sedenions have zerodivisors (see the Wikipedia page on sedenions
for example), and therefore the vector fields obtained from sedenionic multiplication are not
linearly independent at every point of S15 .
Problem 8.9. Show by finding a counterexample that Proposition 8.19 is false if we replace
the assumption that F is a diffeomorphism by the weaker assumption that it is smooth and
bijective.
Consider the map
F : [0, 1[ → S1 , t 7→ e2πti ,
which is clearly smooth and bijective. Note that [0, 1[ has a smooth vector field Xt =
d
cos(πt) dt which is not F -related to any vector field on S1 .
Problem 8.10. Let M be the open submanifold of R2 where both x and y are positive, and
let F : M → M be the map F (x, y) = (xy, y/x). Show that F is a diffeomorphism, and
compute F∗ X and F∗ Y , where
∂ ∂ ∂
X=x +y , Y =y .
∂x ∂y ∂y
A direct computation of the inverse map of F yields
p √
F −1 (x, y) = ( x/y, xy)
which is smooth, so F is a diffeomorphism. Moreover, the differential dF in the standard
coordinates on R2 gives
y −y/x2

dF = .
x 1/x
From here we obtain
p √
q −q/p2 pq(1 − p/q 2 )

p/q
(F∗ X)(p,q) = √ = p .
p 1/p pq p/q(1 + p)
The computation of F∗ Y is similar.
Problem 8.11. For each of the following vector fields on the plane, compute its coordinate
representation in polar coordinates on the right half-plane {(x, y) : x > 0}.
∂ ∂
(a) X = x ∂x + y ∂y .
∂ ∂
(b) Y = x ∂y − y ∂x .
∂
(c) Z = (x2 + y 2 ) ∂x
Let (r, θ) be the polar coordinates defined by x = r cos θ and y = r sin θ. These derivatives
will be useful:
∂r x ∂r y
=p = cos θ =p = sin θ
∂x 2
x +y 2 ∂y x + y2
2
∂θ y sin θ ∂θ x cos θ
=− 2 =− = 2 = .
∂x x + y2 r ∂y x + y2 r
Some routine calculations then yield
∂ ∂ ∂ ∂
X=r , Y = , Z = r2 cos θ − r sin θ .
∂r ∂θ ∂r ∂θ
Problem 8.12. Let F : R2 → RP2 be the smooth map F (x, y) = [x, y, 1], and let X ∈ X
be defined by X = x∂/∂y − y∂/∂x. Prove that there is a vector field Y ∈ X(RP2 ) that is
F -related to X, and compute its coordinate representation in terms of each of the charts
defined in Example 1.5.
The image of F is the domain of one of the charts of RP2 defined in Example 1.5 (it is
U3 , to be exact). We first compute the coordinate representation of X in the charts U1 and
U2 of Example 1.5. We will then see that the vector field defined on U3 can be smoothly
extended to all of RP2 .
Give U1 , U2 , and U3 coordinates as in Example 1.5:
ϕ1 : U1 → R2 , [1, a, b] 7→ (a, b)
2
ϕ2 : U2 → R , [c, 1, d] 7→ (c, d)
2
ϕ3 : U3 → R , [x, y, 1] 7→ (x, y).
Note that F = ϕ−1 3 . The transition functions are easy to compute (they are also computed
in Example 1.33). It is then easy to show that the coordinate representations are as follows:
∂ ∂
X = (1 + a2 ) + ab in U1 ∩ U3
∂a ∂b
∂ ∂
X = −(1 + c2 ) − cd in U2 ∩ U3
∂c ∂d
∂ ∂
X=x −y in U3 .
∂y ∂x
It is then clear that the vector fields in U1 ∩ U3 and U2 ∩ U3 can be smoothly extended to
all of U1 and U2 , respectively. We conclude that there is a global vector field Y ∈ X(RP2 )
such that X is F -related to Y .
Problem 8.13. Show that there is a smooth vector field on S2 that vanishes at exactly one
point.
Consider the charts on S2 coming from the stereographic projection:
(x, y)
ϕn : Un → R2 , (x, y, z) 7→ = (a, b)
1−z
(x, y)
ϕs : Us → R2 , (x, y, z) 7→ = (c, d).
1+z
Consider the vector field
∂
X=
∂a
on R2 , which defines a nonvanishing vector field on Un via ϕn . In the coordinates of Us , we
have
∂ ∂
X = (d2 − c2 ) − 2cd on Un ∩ Us .
∂c ∂d
We can thus extend the definition of X to all of Us by the same formula as above to obtain
a vector field on all of S2 that only vanishes at one point (namely, the south pole).
Problem 8.14. Let M be a smooth manifold with or without boundary, let N be a smooth
manifold, and let f : M → N be a smooth map. Define F : M → M ×N by F (x) = (x, f (x)).
Show that for every X ∈ X(M ), there is a smooth vector field on M × N that is F -related
to X.
Note that F properly embeds M in M × N as the graph of f (see Proposition 5.7). We
then obtain a vector field F∗ X on F (M ) ⊆ M × N . By the the next problem, there is an
extension Y ∈ X(M × N ) of F∗ X and Y is F -related to X.
Problem 8.15 (Extension Lemma for Vector Fields on Submanifolds). Suppose M is a
smooth manifold and S ⊆ M is an embedded submanifold with or without boundary. Given
X ∈ X(S), show that there is a smooth vector field Y on a neighbourhood of S in M such
that X = Y |S . Show that every such vector field extends to all of M if and only if S is
properly embedded.
For each p ∈ S, let Up be a slice chart. Using the coordinate representation given by the
slice map, it is clear we can extend X|S∩Up to a vector field Yp on Up . Let S{ϕp : p ∈ S} be
a partition of unity subordinate to {Up }. Then we define a vector field on p Up by
X
Xq = ϕp (q)(Yp )|q .
p
The sum makes sense, because if q ∈ / Up , then ϕp (q) = 0. Moreover, if q ∈ S, then

(Yp )|q = Xq (for all p where this makes sense) and therefore
X X
Xq = ϕp (q)(Yp )|q = ϕp (q)Xq = Xq ,
p p
so Y is an extension of X to some open neighbourhood of S. (A similar and probably easier

argument using the fact that every embedded submanifold has a tubular neighbourhood.)
If S is properly embedded in M , then it is closed (see Proposition 5.5). By the work
above, we can extend any vector field X on S to some open neighbourhood of S, which
implies that X is in fact a smooth vector field along S. By Lemma 8.6, we can extend X
to all of M .
Conversely, suppose that S is not properly embedded. Then it is not closed, so there is
some p ∈ M r S that is a limit point of S. Let (xn ) and (yn ) be disjoint sequences in S
which both converge to p, and which are each closed in S (we can construct such sequences
by taking concentric balls around p, we omit the details). Let v ∈ Tp M be any nonzero
vector and let Xxn ∈ Txn M be a sequence of vectors converging (in T M ) to v. Similarly,
let Xyn ∈ Tyn M be a sequence converging to −v. Since {xn } ∪ {yn } is closed in S, there is
an extension of X to all of S by Lemma 8.6. However, there can be no extension of X to
all of M , because we cannot consistently define the extension at p.

Solutions To Lee's Smooth Manifolds

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Solutions To Lee's Smooth Manifolds

Uploaded by

Copyright:

Available Formats

SOLUTIONS TO LEE’S INTRODUCTION TO SMOOTH MANIFOLDS

Date: February 23, 2023.

countable, since M is second-countable. Note that φi (Ui ) ⊆ Rn has a countable basis of

fi be the path obtained by restricting f to [ai−1 , ai ],

Fs ◦ f ◦ φi ◦ (Fs0 ◦ f ◦ φi )−1 = Fs ◦ f ◦ φi ◦ φ−1

Let σ̃(x) = −σ(−x) for x ∈ Sn r {S}.

As described in Example 1.36, if πP : Rn → P and πQ : Rn → Q are the projections

where Bn,+ i = {(x1 , . . . , xn ) ∈ Bn : xi > 0} and Bn,− i = {(x1 , . . . , xn ) ∈ Bn : xi 6 0}. The

(x1 , . . . , xn ) 7→ (x1 , . . . , xi−1 , xi ± 1 − (x1 )2 − · · · − (x

Since both φ± i and (φi )

(ψ ◦ ψ̃ −1 ) ◦ (ψ̃ ◦ F ◦ φ̃−1 ) ◦ (φ̃ ◦ φ−1 ) = ψ ◦ F ◦ φ−1 : φ(U ∩ U

Exercise 2.19. Use Theorem 1.46 to prove Theorem 2.18.

C ∞ (M ) is a bijection. It is also linear by part (a). Finally, if f, g ∈ C ∞ (N ), then F ∗ (f g) =

where F : R → V is defined by F (v 1 , . . . , v n ) = v i Ei . Since T −1 F is an invertible linear

is a bijection. Hence, we obtain a topology on π −1 (U ) by pulling back the topology on

Problem 3.2. Prove Proposition 3.14.

F1 is a homeomorphism. We similarly obtain a homeomorphism of Ti via the function

Suppose that there is a diffeomorphism F : R2 → R2 such that F (S1 ) = K. Let γ(t) : R →

is an isomorphism for every p ∈ R2 . In particular, dF (γ 0 (tc )) = dFγ(tc ) (γ 0 (tc )) 6= 0, since

Since γz is continuous, γz−1 (U1+ ) is open, and therefore φ+

4. Submersions, Immersions, and Embeddings

For (d), dX(u,v) has the matrix representation

F |U : U → F (U ) is a diffeomorphism. Then G◦F |U is smooth, and so is (G◦F |U )◦(F |U )−1 =

Exercise 4.34. Prove Proposition 4.33.

Let π̃ : M → N e be the same smooth surjective submersion as π, but viewed as a map

Let π : E → M be a topological covering map. Suppose that π is proper. Then {p} is

commute. Because 2 is actually a local diffeomorphism (Example 4.11) and X is a smooth

Since F = G ◦ ι, where G : R3 → R4 is the map given by the same formula as F but

commutes. The commutativity of the diagram forces

In the coordinates of the usual chart (U1 , φ1 ), we have

and the matrix representation of dFep for p ∈ U1 is then

The first minor is (up to a nonzero factor)

x2 − y 2 = u2 − v 2 , xy = uv, xz = uw, yz = vw.

Using these equations along with x2 + y 2 + z 2 = u2 + v 2 + w2 = 1, it can be shown that

for some v i ∈ R. On one hand, we have v(xi ) = v(xi |S ) = 0 if i = k + 1, . . . , n (we are

γ 0 (0) = v. In fact, the image of γ is contained in V , which is embedded in M . Corollary

Problem 5.1. Consider the map Φ : R4 → R2 defined by

By the classification of 1-dimensional manifolds, γ(R) is homeomorphic to either R or S1 .

Φ−1 (0) is not immersed.

Fi : ϕi (Ui ) → R, (x1 , . . . , xk , xk+1 , . . . , xn ) 7→ f ◦ ϕ−1

lies in S, but γ 0 (0) lies in the y-axis, so γ 0 (t) ∈

By Theorem 2.29, there exists a smooth nonnegative function f : M → R such that B =

It is clear that fe is smooth on M r B and that it agrees with f on B, so it remains to check

commutes. Let v ∈ Tp W . By functoriality,

(dGF (q) )−1 (d(GF )q (Tq M ) + TGF (q) X) =

Let F : M → RN be a smooth map, and consider the composition ι ◦ F : M → RN +1 ,

If (a) holds, then πM |S is a diffeomorphism onto M , which proves (b).

submanifold of M . Since Fs0 is a diffeomorphism, N and M have equal dimensions. Since

Note that if N is Abelian then N

(b) For X ∈ GL(n, R) and B ∈ TX GL(n, R) ∼= M(n, R), show that

(a) Recall that

where the last line follows from part (a).

defines a smooth, transitive left action of GL(n + 1, R) on RPn .

The actions are G × G → G, (s, t) 7→ st and G × H → H, (g, h) 7→ F (g)h. These are

where θ ∈ R. Thus, there is a map

Problem 7.22. By Proposition 7.11, we conclude that S is a Lie subgroup of H∗ . It is also

and the claim follows.

so A is a Lie algebra homomorphism.

Lie(GL(V )) Tid GL(V ) gl(V )

Lie(GL(n, R)) TIn GL(n, R) gl(n, R) .

is a smooth extension of X whose support is contained in U .

On one hand, we have

The sum makes sense, because if q ∈ / Up , then ϕp (q) = 0. Moreover, if q ∈ S, then

commute. Because 2 is actually a local diffeomorphism (Example 4.11) and X is a smooth