LECTURE 6

6.1. The inverse function theorem

The inverse function theorem gives a very simple condition to check that a map is
locally a diffeomorphism, namely, it suffices that its differential be invertible. For
a proof of this classical result, see for example [11]. This can be directly extended
to smooth manifolds:
Theorem 6.1.1 (The inverse function theorem). Let f : M → N be a smooth map
and let p ∈ M . If dp f : Tp M → Tf (p) N is a linear isomorphism, then there exists
an open neighborhood O ⊂ M of p, such that f (O) is open and f restricts to a
diffeomorphism:
f |O : O −∼→ f (O).
Proof. Let (U, ϕ) be a chart around p and (V, ψ) be a chart around f (p). By
shrinking U , we may assume that f (U ) ⊂ V . By the chain rule, we have that
−1
dϕ(p) (ψ ◦ f ◦ ϕ−1 ) = df (p) ψ ◦ dp f ◦ (dp ϕ) .
Since the charts are diffeomorphisms, their differentials are linear isomorphisms;
therefore, the right hand side is the composition of linear isomorphisms. We obtain
that also dϕ(p) (ψ ◦f ◦ϕ−1 ) is a linear isomorphism. The map ψ ◦f ◦ϕ−1 is a smooth
map between open sets in Rm , and its Jacobian matrix at ϕ(p) coincides with its
differential, thus it is invertible. By the standard inverse function theorem (see e.g.
[11]), there exists an open neighborhood W of ϕ(p) such that ψ ◦ f ◦ ϕ−1 (W ) is
open and ψ ◦ f ◦ ϕ−1 |W : W − ∼→ ψ ◦ f ◦ ϕ−1 (W ) is a diffeomorphism. Since the
charts are diffeomorphisms, this implies that conclusion with O := ϕ−1 (W ). 

6.2. Immersions and submersions

The following two classes of maps play an important role:
Definition 6.2.1. Let f : M → N be a smooth map, and let p ∈ M .
(1) f is called an immersion at p if dp f : Tp M → Tf (p) N is injective. If this
holds at all points in M , then f is called an immersion.
(2) f is called a submersion at p if dp f : Tp M → Tf (p) N is surjective. If this
holds at all points in M , then f is called a submersion.
53
54 IOAN MĂRCUT
, , MANIFOLDS

The most basic example of an immersion is the linear inclusion:

ikl : Rk −→ Rl , ikl (x1 , . . . , xk ) := (x1 , . . . , xk , 0, . . . , 0), for k ≤ l,

and the most basic example of a submersion is the linear projection

prlk : Rl −→ Rk , prlk (x1 , . . . , xl ) := (x1 , . . . , xk ), for k ≤ l.

The following two theorems say that locally, up to a change of coordinates,

these are the only examples:

Theorem 6.2.2 (The local immersion theorem). Let f : M → N be a smooth map.

If f is an immersion at p ∈ M , then there are charts (U, ϕ) and (V, ψ) around p
and f (p), respectively, such that f (U ) ⊂ V and, in these coordinates, f becomes
the map imn ; i.e.

ψ ◦ f ◦ ϕ−1 = im n
n |ϕ(U ) : ϕ(U ) −→ R ,

ψ ◦ f ◦ ϕ−1 (x1 , . . . , xm ) = (x1 , . . . , xm , 0, . . . , 0).

Theorem 6.2.3 (The local submersion theorem). Let f : M → N be a smooth

map. If f is a submersion at p ∈ M , then there are charts (U, ϕ) and (V, ψ) around
p and f (p), respectively, such that f (U ) ⊂ V and, in these coordinates, f becomes
the map prmn ; i.e.

ψ ◦ f ◦ ϕ−1 = prm n
n |ϕ(U ) : ϕ(U ) −→ R ,

ψ ◦ f ◦ ϕ−1 (x1 , . . . , xm ) = (x1 , . . . , xn ).

The proofs of these two results are completely analogous, therefore, we only
give the details for the second result; the proof of the local immersion theorem is
left as an exercise.

Proof of the local submersion theorem. Let (U e , ϕ)

e and (V, ψ) be two charts
around p and f (p), respectively, such that f (U ) ⊂ V . Denote the local expression
e
of f by h := ψ ◦ f ◦ ϕ e−1 . We are going to “complete h to a diffeomorphism”.
By applying Exercise 6.1 to the linear map A = dϕ(p) e h (which, by the chain rule,
is surjective), we obtain a linear map B : Rm → Rm−n such that (dϕ(p) e h, B) :
Rm → Rm is a linear isomorphism. This map is precisely the differential at ϕ(p) e
of the map (h, B) : ϕ( eUe ) → Rm . Therefore, by the inverse function theorem,
there is a neighborhood O ⊂ ϕ( eU e ) of ϕ(p)
e such that W := (h, B)(O) is open and
∼
(h, B) : O −→ W is a diffeomorphism; denote its inverse by g := (h, B)−1 : W − ∼
→ O.
Then we have that (h(g(x)), B(g(x))) = x for all x ∈ W , hence h ◦ g = prm n |W .
Equivalently, ψ ◦ f ◦ ϕ−1 = prmn |W , where

ϕ := g −1 ◦ ϕ e−1 (O);
e : U −→ Rm , U := ϕ
e = (h, B) ◦ ϕ

hence, the charts (U, ϕ) and (V, ψ) satisfy the conditions from the statement. 

Remark 6.2.4. In the local submersion theorem, the chart on the codomain (V, ψ)
can be chosen arbitrary; this follows from the proof. Similarly, in the local immer-
sion theorem the chart (U, ϕ) on the domain can be chosen arbitrary.
 LECTURE 6. 55

6.3. Embedded submanifolds

The following generalizes the notion of embedded manifolds in Rn from Section 2.4:
Definition 6.3.1. A subset S ⊂ M of a smooth m-dimensional manifold M is
called a k-dimensional embedded submanifold if for every point p ∈ S there
exists a chart (U, ϕ) around p, such that
ϕ(S ∩ U ) = Rk × {0} ∩ ϕ(U ) ⊂ Rm .


(*)
A chart (U, ϕ) with these property is called a chart adapted to S. The codimen-
sion of S is the number m − k = dim(M ) − dim(S).
Next, we show that embedded submanifolds are indeed smooth manifolds:
Theorem 6.3.2. Let S ⊂ M be a k-dimensional embedded submanifold. Then S,
endowed with the subset topology, has a k-dimensional differentiable structure, with
atlas obtained by restricting adapted charts:
AS = {(U ∩ S, prm
k ◦ ϕ|U ∩S ) : (U, ϕ) is a chart adapted to S}.
Moreover, the inclusion map i : S ,→ M , i(x) = x is a smooth immersion.
Proof. The proof that S, endowed with the subset topology, has a k-dimensional
differentiable structure for which AS is a smooth atlas can be proven exactly as in
the case of embedded manifolds in Rn (see Proposition 2.4.3.
Next, since S has the induced topology, it follows that i is continuous. Let
(U, ϕ) be an adapted chart around p ∈ S. Then, note that in the pair of charts
(U ∩ S, prmk ◦ ϕ|U ∩S ) and (U, ϕ) the local expression of i is the canonical immersion
ikm : Rk → Rm . This proves that i is a smooth immersion. 
In Definition 6.3.1 above, denote that the coordinates by ϕ = (x1ϕ , . . . , xm
ϕ ).
Then condition (*) means that S is locally described by the l = m − k equations:
S ∩ U = {q ∈ U : xk+1 k+2 m
ϕ (q) = 0, xϕ (q) = 0, . . . , xϕ (q) = 0}.
The converse also holds, and turns out to be a very useful criterion: any subset
which is locally described as the zeroes of by l independent equations is an embedded
submanifold of codimension l (here, independent means that the equations form a
submersion):
Proposition 6.3.3. A subset S ⊂ M is an embedded submanifold of codimension
l if and only if for every p ∈ S there exists an open set p ∈ U ⊂ M and smooth
map h : U → Rl which is a submersion at p, such that S ∩ U = h−1 (0).
Proof. Assume first that S is an embedded submanifold of codimension l. Let
p ∈ S and let (U, ϕ) be a chart adapted to S with p ∈ U . Define
h : U −→ Rl , h := prm
l ◦ ϕ,
where prm l : R
m
→ Rl denotes the projection onto the last l coordinates. Then h
is a submersion and S ∩ U = h−1 (0).
Conversely, assume that S ⊂ M is a subset such that, around any p ∈ S, we can
find an open set U and a smooth map h : U → Rl which is a submersion at p and
satisfies U ∩ S = h−1 (0). We apply the local submersion theorem to h. By Remark
6.2.4, we do not need to change the coordinates on Rl . Thus, after shrinking U , we
may assume that there is a chart (U, ϕ) on M such that
h ◦ ϕ−1 (x1 , . . . , xm ) = (xk+1 , . . . , xm ), where k = m − l;
56 IOAN MĂRCUT
, , MANIFOLDS

or equivalently, in the usual notation for coordinates ϕ = (x1ϕ , . . . , xm

ϕ ), we have:

h(q) = (xk+1 m
ϕ (q), . . . , xϕ (q)), for q ∈ U.
Thus, for q ∈ U : h(q) = 0 iff ϕ(q) ∈ Rk × {0}, and since U ∩ S = h−1 (0), we obtain:
ϕ(S ∩ U ) = (Rk × {0}) ∩ ϕ(U ). This concludes the proof. 

Remark 6.3.4. Consider an embedded submanifold S ⊂ M . By Theorem 6.3.2

the inclusion map i : S ,→ M is a smooth immersion. Therefore if p ∈ S the map
dp i : Tp S → Tp M is injective, and so it sends Tp S isomorphically onto a linear
subspace of Tp M . We will identify these spaces:
Tp S = dp i(Tp S) ⊂ Tp M.
Assume now that h : U → Rl is a local submersion around p such that S = h−1 (0).
We claim that
(*) Tp S = ker(dp h) = {v ∈ Tp M : dp h(v) = 0}.
Since h ◦ i(U ∩ S) = 0, it follows that dp h ◦ dp i(Tp S) = 0. Hence Tp S ⊂ ker(dp h).
The other inclusion follows because dp h is surjective, and so these vector spaces
have the same dimension:
dim ker(dp h) = dim Tp M − dim im(dp h) = dim M − codim S = dim S = dim Tp S.
Let us introduce some standard terminology:
Definition 6.3.5. A regular value of a smooth map f : M → N is a point q ∈ N
such that f is a submersion at every point in f −1 (q). A point q ∈ N which is not
a regular value is called a critical value for f .
Corollary 6.3.6 (Preimage Theorem). Let f : M → N be a smooth map. If q ∈ N
is a regular value f , then f −1 (q) is an embedded submanifold of M with
codim(f −1 (q)) = dim(N ), Tp (f −1 (q)) = ker(dp f ), for p ∈ f −1 (q).
Proof. Take a chart (U, ϕ) around p with ϕ(q) = 0. Then h := ϕ ◦ f : V → Rn is
a submersion, and f −1 (q) = h−1 (0). Proposition 6.3.3 implies the result. 

Let us also mention a deep result of mathematical analysis, which states that
most points are regular values (note that points that are not in the image are always
regular values). Without introducing all notions involved, we state this result below
(for details, the reader is encouraged to consult [8] or [4]).
Theorem ♣ 6.3.7 (Sard’s Theorem). The set of critical values of a smooth map
f : M → N has Lebesgue measure zero. In particular, the set of regular values of f
is dense in N .
Remark 6.3.8. If dimM < dimN , then every point in M is a critical point of
f . In this situation, Sard’s Theorem says that f (M ) has measure zero in N ; and
therefore N \f (M ) is dense in N . In particular, one cannot have a surjective map
smooth f : M → N if dimM < dimN .
For this property, smoothness is essential. In the continuous setting, there are
the co-called space-filling curves (e.g. Peano’s curve [21]). For example, one can
prove that any connected, nonempty manifold is the image of a continuous curve!!
 LECTURE 6. 57

6.4. Preimages of submanifolds

The following extends the notion of a regular value to submanifolds.
Definition 6.4.1. A smooth map f : M → N is said to be transverse to an
embedded submanifold Q ⊂ N , if
Tf (p) N = Tf (p) Q + dp f (Tp M ), ∀ p ∈ f −1 (Q).
Remark 6.4.2. Note that the sum in the decomposition is not assumed to be
a direct sum, i.e. the vector spaces Tf (p) Q and dp f (Tp M ) may have a nontrivial
intersection. For example, a submersion is transverse to any embedded submanifold
of the codomain.
Theorem 6.4.3. Let f : M → N be a smooth map that is transverse to the
embedded submanifold Q ⊂ N . Then P := f −1 (Q) is an embedded submanifold of
M with codim P = codim Q and
−1
Tp P = (dp f ) (Tf (p) f −1 (Q)), ∀p ∈ P.
Before giving the proof, let us consider an example.
Example 6.4.4. Consider the orthogonal projection from the 2-sphere to R2 ,
f : S 2 −→ R2 , f (x, y, z) := (x, y).
Note that f is a submersion outside of the equatorial circle, which we denote by Z.
Consider two embedded submanifold Q1 and Q2 in R2 which are indicated on the
picture below:

f −1 (Q1 )
2
S

f −1 (Q2 )
p2
p

p0 Z p1

f (p2 )
f (p) Q1 Q2

R2
f (p0 ) f (Z) f (p1 )

Note that f is not transverse to Q1 , because at the point of intersection f (p0 ) of

f (Z) and Q1 , we have that Tf (p0 ) Q1 = dp0 f (Tp0 S 2 ). Note also that f −1 (Q1 ) is not
58 IOAN MĂRCUT
, , MANIFOLDS

an embedded submanifold of S 2 , because it has a singularity which looks like the

intersection of two lines.
On the other hand, f is transverse to Q2 ; outside of Z this is clear because f
is a submersion, and at the two points p1 , p2 ∈ Z ∩ f −1 (Q2 ) note that
dpi f (Tpi S 2 ) + Tf (pi ) (Q2 ) = Tf (pi ) R2 , i = 1, 2.
As predicted by the Theorem 6.4.3, f −1 (Q2 ) is an embedded submanifold of S 2 ; in
fact it is diffeomorphic to a circle.
Proof of Theorem 6.4.3. We check that P satisfies the condition in Proposition
6.3.3. Consider p ∈ P , and denote q := f (p) ∈ Q. Since Q is an embedded
submanifold there exists an open neighborhood U of q and a smooth map h : U →
Rl , where l = codimQ, such that h is a submersion at q and h−1 (0) = U ∩ Q. On
the open neighborhood f −1 (U ) of p, consider the smooth map h ◦ f : f −1 (U ) → Rl .
We show that this map satisfies the conditions from Proposition 6.3.3. Note that
(h ◦ f )−1 (0) = f −1 (h−1 (0)) = f −1 (U ∩ Q) = f −1 (U ) ∩ f −1 (Q) = f −1 (U ) ∩ P.
By Remark 6.3.4, Tq Q = ker dq h; hence dq h(Tq Q) = 0. Using this, that h is a
submersion at q, that f is transverse to Q, and the chain rule, we obtain:
T0 Rl = dq h (Tq N ) = dq h (Tq Q + dp f (Tp M )) = dq h ◦ dp f (Tp M ) = dp (h ◦ f )(Tp M ).
Thus h◦f is a submersion at p. So Proposition 6.3.3 implies that P is an embedded
manifold of codim P = l = codim Q. Finally, again by (*), we have that
Tp P = ker dp (h ◦ f ) = ker (dq h ◦ dp f ) = (dp f )−1 (ker dq h) = (dp f −1 )(Tq Q).

Remark 6.4.5. Note that the empty set ∅ is a smooth manifold of any dimension
d ≥ 0: it has a (unique) topology; it is locally homeomorphic to the empty set in Rd
(which is open); and the only transition map is the id∅ , which is smooth, when we
regard ∅ ⊂ Rd (prove this “by contradiction”!). In the setting of Theorem 6.4.3, it
could happen that f (M ) ∩ Q = ∅; in this case f is transverse to Q and f −1 (Q) = ∅.
Theorem 6.4.3 can be used to intersect submanifolds. First, we need the fol-
lowing definition:
Definition 6.4.6. Two embedded submanifold R, S ⊂ M are said to intersect
transversely, if for every p ∈ R ∩ S we have that
Tp M = Tp R + Tp S.
This condition is equivalent to the inclusion i : R ,→ M being transverse to S.
Note also that
i−1 (S) = R ∩ S, (dp i)−1 (Tp R) = Tp R ∩ Tp S, p ∈ R ∩ S.
Hence, Theorem 6.4.3 implies the following:
Corollary 6.4.7. The intersection R ∩ S of two embedded submanifolds R, S ⊂ M
which intersect transversally is an embedded submanifold with
codim R ∩ S = codim R + codim S;
moreover, for each p ∈ R ∩ S we have that
Tp (R ∩ S) = Tp R ∩ Tp S.
 LECTURE 6. 59

6.5. Exercises
Exercise 6.1. Let k ≤ n, and let A : Rn → Rk be a surjective linear map, and let
C : Rk → Rn be an injective linear map.
(a) Prove that there exists a linear map B : Rn → Rn−k such that the map
(A, B) : Rn → Rk × Rn−k = ∼ Rn , (A, B)(x) := (A(x), B(x)),
is a linear isomorphism.
(b) Prove that there exists a linear isomorphism G : Rn → Rn such that
A ◦ G = prnk .
(c) Prove that there exists a linear map D : Rn−k → Rn such that the map
C + D : Rk × Rn−k ∼ = Rn → Rn , (C + D)(y, z) := C(y) + D(z),
is a linear isomorphism.
(d) Prove that there exists a linear isomorphism H : Rn → Rn such that
H ◦ C = ikn .
Exercise 6.2. Let a ∈ R. Show that the following map is an immersion:
fa : R −→ S 1 × S 1 , fa (t) := (eit , eita ).
√
Make a sketch of fa (R) for a = 0, 1/2, 1, 2, 3. For which a ∈ R is fa injective?
Exercise 6.3. Let f : P2 (R) → R3 be the map defined by
1
f ([x, y, z]) = 2 (yz, xz, xy).
x + y2 + z2
Show that f is smooth and show that it only fails to be an immersion at 6 points.
Make a sketch of the image of f .
Being an immersion/submersion is an open condition:
Exercise 6.4. If a smooth map f : M → N is an immersion (resp. a submersion)
at p ∈ M prove that there is an open neighborhood V ⊂ M of p such that f |V is
an immersion (resp. a submersion).
Do not use the local immersion/submersion theorem!
Exercise 6.5. Prove the local immersion theorem.
Exercise 6.6 (Alternative definition of immersions and submersions). Let f : M →
N be a smooth map, and let p ∈ M . Prove the following:
(a) f is an immersion at p iff it has a local left inverse, i.e. there exist open neigh-
borhoods U ⊂ M of p and V ⊂ N of f (p) and a smooth map σ : V → U such
that f (U ) ⊂ V and
σ(f (x)) = x, for all x ∈ U.
(b) f is a submersion at p iff it has a local right inverse, i.e. there exist open
neighborhoods U ⊂ M of p and V ⊂ N of f (p) and a smooth map j : V → U
such that j(f (p)) = p and
f (j(y)) = y, for all y ∈ V.
Exercise 6.7. Let f : M → N be a submersion. Prove that f is an open map (i.e.
for every open set U ⊂ M , f (U ) is open in N ).
60 IOAN MĂRCUT
, , MANIFOLDS

Exercise 6.8. Let f : M → N be a smooth map, and let K ⊂ M be a compact

subset. If f |K : K → N is injective, and dp f : Tp M → Tf (p) N is a linear isomor-
phism for every p ∈ K, prove that there is an open set U such that K ⊂ U , f (U )
is open in N and f : U → f (U ) is a diffeomorphism.
Exercise 6.9. Use the Preimage Theorem (Corollary 6.3.6) to prove that S n is an
embedded submanifold of Rn+1 .
Exercise 6.10. Let Sb1 ⊂ R3 be the circle in the plane z = 0 of radius b > 0 and
with center at the origin. The 2-dimensional torus of radii 0 < a < b, denoted by
2
Ta,b , is the set of points in R3 at distance a from Sb1 . Use the Preimage Theorem
2
(Corollary 6.3.6) to prove that Ta,b is a 2-dimensional manifold (Find an explicit
equation describing it! )
Exercise 6.11. Let M (n) denote the space of all real n × n matrices. Let S(n)
denote the space of symmetric matrices:
S(n) := {B ∈ M (n) : B = B t }.
Prove that I is a regular value of the map
f : M (n) −→ S(n), f (A) = AAt .
Using the Preimage Theorem (Corollary 6.3.6), prove that the set of orthogonal
matrices
O(n) := {A ∈ M (n) : AAt = I},
is a compact embedded submanifold of M (n) of dimension n(n − 1)/2.
2
Hint: For compactness, show that O(n) ⊂ S n −1 .