LECTURE 5

Remark. For simplicity, from now on we will usually suppress the adjective smooth,
by implicitly assuming smoothness of the objects involved: by manifold we mean
smooth manifold, by chart we mean smooth chart, etc.

5.1. The tangent space

Intuitively, the tangent space of a smooth manifold M at a point p, is a vector
space denoted by Tp M of dimension m = dim(M ), which represents all velocities
of curves passing through p.
Example 5.1.1. For example, the tangent space to the 2-sphere S 2 at a point
p ∈ S 2 can be identified with the set of vectors perpendicular to p: Tp S 2 ∼ = {v ∈
R3 : v · p = 0}. Namely, any smooth curve γ : (−, ) → S 2 (with γ(0) = p)
satisfies γ(t) · γ(t) = 1, hence taking the derivative, γ 0 (t) · γ(t) = 0; in particular
v · p = 0, where v = γ 0 (0). On the other hand, given a local parameterization
f : O ⊂ R2 → S 2 (i.e. the inverse of a local chart) with f (q) = p, we can describe
the tangent space as the image of the differential of f :
Tp S 2 ∼
= (dq f )(R2 ).
Example 5.1.2. The tangent space of Rm at p ∈ Rm can be canonically identified
Tp Rm ∼ = Rm : any smooth curve γ : (−, ) → Rm with γ(0) = p induces the
tangent vector γ 0 (0) ∈ Rm . Conversely, any v ∈ Rm is the speed of some curve
through p; for example, γ(t) = p + tv. We think about Tp Rm as being “vectors
based at p”. Let us see what happens when we change coordinates on Rm , i.e.
consider a diffeomorphism χ : U − ∼
→ V , between open sets U, V ⊂ Rm . Then any
smooth curve γ : (−, ) → U , with γ(0) = p, can be transported to a smooth curve
χ ◦ γ : (−, ) → V , and their speeds are related by the chain rule:
(χ ◦ γ)0 (0) = (dp χ)γ 0 (0).
Thus, the differential of the change of coordinate matrix gives a natural identifica-
tion between the tangent spaces.
On a general manifold, a fixed chart induces a model for the tangent space (i.e.
just Rm ), and we use the differential of the change of coordinate maps to identify
these various models.
43
44 IOAN MĂRCUT
, , MANIFOLDS

Definition 5.1.3. Let M be a smooth manifold of dimension m. A tangent vector

at p ∈ M is a map v which associates to each chart (U, ϕ) around p an element
vϕ ∈ Rm satisfying the following rule
vψ = dχ(p) (ψ ◦ χ−1 )(vχ ),
for any two charts (O, χ) and (V, ψ) around p. The set of all tangent vectors at p
is called the tangent space of M at p and is denoted by Tp M .
Since the differential of the change of coordinates map is a linear map, it follows
that Tp M is a vector space with operations
(λv)ϕ = λvϕ , (v + w)ϕ = vϕ + wϕ , ∀v, w ∈ Tp M, λ ∈ R.
We have that:
Lemma 5.1.4. The tangent space of M at p is a vector space of dimension m =
dim(M ). Moreover, for any chart (U, ϕ) around p, the map
Tp M → Rm , v 7→ vϕ
is a linear isomorphism.
Proof. That the map is linear is clear. The kernel of the map is zero: if vϕ = 0
then, by linearity of the differential, vψ = dϕ(p) (ψ ◦ϕ−1 )(vϕ ) = 0 for any other chart
(V, ψ), thus v = 0. Thus the map is injective. To show surjectivity, let u ∈ Rm ,
and define v so that
vψ := dϕ(p) (ψ ◦ ϕ−1 )(u),
for a chart (V, ψ). That v is indeed a tangent vector follows from the chain rule:
for any two charts (V, ψ) and (O, χ), we have
vψ = dϕ(p) (ψ ◦ ϕ−1 )(u) = dϕ(p) (ψ ◦ χ−1 ◦ χ ◦ ϕ−1 )(u) =
= dχ(p) (ψ ◦ χ−1 ) ◦ dϕ(p) (χ ◦ ϕ−1 )(u) = dχ(p) (ψ ◦ χ−1 )(vχ ).
Clearly, vϕ = u; thus the assignment is indeed onto. 
m m
Using the standard chart (R , id) on R , the lemma above allows us to identify
(∗) Tp Rm → Rm , v 7→ vid .

5.1.5. The differential of a smooth map

Next, we extend the notion of the differential of a smooth map. We will use the
same notation as for maps between open sets in Euclidean space; the standard
notion is obtained by using the identification (∗) above.
Definition 5.1.6. Let f : M → N be a smooth map. The differential of f at
p ∈ M is the linear map
dp f : Tp M → Tf (p) N
dp f (v) ψ = dϕ(p) (ψ ◦ f ◦ ϕ−1 )(vϕ ),

where (U, ϕ) is a chart around p and (V, ψ) is a chart around f (p).

To show that the above assignment ψ 7→ (dp f (v))ψ yields indeed a tangent
vector, one needs to show that it transforms accordingly to the rule from Definition
5.1.3; however, this is a straightforward application of the chain rule, which we omit
(as in the proof of Lemma 5.1.4).
 LECTURE 5. 45

With the canonical identification of the tangent space of Rm , note that the
differential of a chart (U, ϕ) around p ∈ M becomes
dp ϕ : Tp M → Tϕ(p) Rm , dp ϕ(v) = vϕ .
The chain rule also extends to the setting of smooth manifolds, and its proof
is again a straightforward application of the ‘classical’ chain rule:
Lemma 5.1.7. Let f : M → N and g : N → P be smooth maps between smooth
manifolds. The differentials satisfy the chain rule:
dp (g ◦ f ) = df (p) g ◦ dp f, ∀ p ∈ M.

5.2. Notations for tangent vectors

Let M be a smooth manifold and consider a chart (U, ϕ) around p ∈ M inducing
local coordinates
xiϕ : U → R, ϕ = (x1ϕ , . . . , xm
ϕ ).

Let e1 , . . . , em be the standard basis of Rm . The corresponding basis of Tp M will

be denoted by
∂ ∂ ∂
, , ... , ∈ Tp M,
∂x1ϕ p ∂x2ϕ p ∂xm
ϕ p
 ∂   ∂ 
i
= dp ϕ i
= ei , 1 ≤ i ≤ m.
∂xϕ p ϕ ∂xϕ p
To explain the notation, let f ∈ C ∞ (M ) be a smooth function on M . Then its
differential is given in this standard basis by:
 ∂ 
−1 i ∂f ◦ ϕ−1
dp f = dϕ(p) (f ◦ ϕ )(e ) = (ϕ(p));
∂xiϕ p ∂xi
in other words the differential of f on the standard basis is given by the partial
derivatives of the local expression f ◦ ϕ−1 of f .
More generally, let f : M → N be a smooth map, and consider a smooth chart
(V, ψ) around q = f (p), with local coordinates ψ = (yψ1 , . . . , yψn ).The matrix of dp f
in the bases
  ( )
∂ ∂
: 1≤i≤m and : 1≤j≤n
∂xiϕ p ∂yψj q
of Tp M and Tq N , respectively, is the Jacobian matrix of the local representation
ψ ◦ f ◦ ϕ−1 (x) = (f 1 (x), . . . , f n (x))
of f in these charts:
∂f 1 ∂f 1 ∂f 1
 
∂x1 ∂x2 ... ∂xm
  ∂f 2 ∂f 2 ∂f 2

dp f = 
 ∂x1 ∂x2 ... ∂xm  (ϕ(p));

 ... ... ... ... 
∂f n ∂f n ∂f n
∂x1 ∂x2 ... ∂xm
n
 ∂  X ∂f j ∂
dp f i
= i
(ϕ(p)) j .
∂xϕ p j=1
∂x ∂yψ q
46 IOAN MĂRCUT
, , MANIFOLDS

5.3. The commutative algebra C ∞ (M )

The “cleanest” way to introduce the tangent space is by using the commutative
R-algebra of smooth functions:
(C ∞ (M ), +, ·), (f + g)(x) := f (x) + g(x), (f · g)(x) := f (x) · g(x).
Note that smooth maps induce algebra homomorphisms:
Proposition 5.3.1. Any smooth map ϕ : M → N induces a homomorphism of
algebras, called the pullback along ϕ:
ϕ∗ : C ∞ (N ) → C ∞ (M ), ϕ∗ (f ) = f ◦ ϕ.
The more surprising result is the converse of this Proposition, which implies
that the commutative algebra C ∞ (M ) encodes algebraically the entire information
about the manifold M :
Theorem ♣ 5.3.2. Any algebra homomorphism s : C ∞ (N ) → C ∞ (M ) is of the
form s = ϕ∗ for a unique smooth map ϕ : M → N .
In particular, this result implies that two manifolds are diffeomorphic if and
only if their commutative algebras of smooth functions are isomorphic. A proof of
Theorem 5.3.2 is outlined in the exercises at the end of this lecture.

5.4. Tangent vectors as derivations

Recall that on Rm the differential of a function f : U ⊂ Rm → Rn is defined using
directional derivatives:
d
dp f (v) = f (p + tv) t=0 .
dt
We introduce the analog terminology on manifolds:
Definition 5.4.1. Let M be a smooth manifold. The derivative along v ∈ Tp M
is the map
Dv : C ∞ (U ) → R, Dv (f ) := dp f (v).
The following holds:
Lemma 5.4.2. For all f, g ∈ C ∞ (M ), and v ∈ Tp M we have that:
Dv (f · g) = f (p)Dv (g) + Dv (f )g(p).
Proof. Using a chart, it suffices to prove this on open sets in Rm . Here it follows
from the usual Leibniz rule (α · β)0 = α · β 0 + α0 · β applied to the functions
α(t) = f (p + tv) and β(t) = g(p + tv) at t = 0. 

Definition 5.4.3. A derivation of C ∞ (M ) at p ∈ M is a linear map

D : C ∞ (M ) −→ R
satisfying the derivation rule:
D(f · g) = f (p)D(g) + D(f )g(p), ∀ f, g ∈ C ∞ (M )
The space of derivations of C ∞ (M ) at p will be denoted by Derp (C ∞ (M )).
 LECTURE 5. 47

Note that the space of derivations is a vector space: for all λ, µ ∈ R and
D1 , D2 ∈ Derp (C ∞ (M )),
λD1 + µD2 : C ∞ (M ) → R, (λD1 + µD2 )(f ) := λD1 (f ) + µD2 (f )
is also derivation of C ∞ (M ) at p.
The following result shows that all derivations come from derivatives along
tangent vectors.
Theorem 5.4.4. Let M be a smooth manifold, and let p ∈ M . The map
Tp M → Derp (C ∞ (M )), v 7→ Dv
is a linear isomorphism.
To prove the theorem, we first show that derivations of C ∞ (M ) at p are local,
in the sense that they depend only on the behavior of the functions around p. The
following terminology will be therefore rather useful:
Definition 5.4.5. Let M be a manifold and p ∈ M . Two functions f, g ∈ C ∞ (M )
are said to have the same germ at p, if there is an open neighborhood W of p such
that f |W = g|W .
Lemma 5.4.6. Let f, g ∈ C ∞ (M ) have the same germ at p ∈ M , then for any
D ∈ Derp (C ∞ (M )), we have that D(f ) = D(g).
Proof. Let h := f − g; then h|W = 0, for some neighborhood W of p. We show
that D(h) = 0. Consider a bump function χ ∈ C ∞ (M ) such that χ(p) = 1 and
supp(χ) ⊂ W . Then, χh = 0. Therefore:
0 = D(0) = D(χh) = h(p)D(χ) + χ(p)D(h) = D(h). 


Lemma 5.4.7. Let U ⊂ M be an open neighborhood of p. Then the map

I : Derp (C ∞ (U )) → Derp (C ∞ (M )), I(D)(g) := D(g|U )
is a linear isomorphism.
Proof. Let χ ∈ C ∞ (M ) be a bump function such that χ|W = 1, on some neigh-
borhood W ⊂ U of p and such that supp(χ) ⊂ U . We define the map
E : C ∞ (U ) → C ∞ (M ), E(f )|U := χf, E(f )|M \supp(χ) := 0.
Note that f and E(f )|U = χf have the same germ at p. Consider the map
R : Derp (C ∞ (M )) → Derp (C ∞ (U )), R(D0 )(f ) := D0 (E(f )).
Since E(f g) and E(f )E(g) have the same germ at p, R(D0 ) is indeed a derivation:
R(D0 )(f g) =D0 (E(f g)) = D0 (E(f )E(g)) = E(f )(p)D0 (E(g))
+ E(g)(p)D0 (E(f )) = f (p)R(D0 )(g) + g(p)R(D0 )(f ).
Note that I ◦ R(D0 )(f ) = D0 (E(f )|U ) = D0 (f ), hence I ◦ R(D0 ) = D0 ; and similarly,
that R ◦ I(D)(g) = D(E(g|U )) = D(g), hence R ◦ I(D) = D. We obtain that I is
invertible with I −1 = R. 

For the proof, we will need the following version of Taylor’s theorem:
48 IOAN MĂRCUT
, , MANIFOLDS

Lemma 5.4.8. Let B ⊂ Rm be a convex open neighborhood of 0. For any f ∈

C ∞ (B) there exist smooth functions gi ∈ C ∞ (B), 1 ≤ i ≤ m, such that
m
X
f (x) = f (0) + gi (x)xi ,
i=1
∂f
and moreover, gi (0) = ∂xi (0).

Proof. Fix x ∈ B, and consider the function h : [0, 1] → R, h(t) = f (tx). Clearly,
h is smooth, therefore the fundamental theorem of calculus applies:
Z 1
h(1) − h(0) = h0 (t)dt.
0
This equation gives:
Z 1 Z 1Xm m
d ∂f X
f (x) − f (0) = f (tx1 , . . . , txm )dt = i
(tx)xi
dt = gi (x)xi ,
0 dt 0 i=1 ∂x i=1
R 1 ∂f ∞ ∂f
where gi (x) = 0 ∂xi (tx)dt. Clearly, gi ∈ C (B) and gi (0) = ∂xi (0). 
Proof of Theorem 5.4.4. Consider a chart (U, ϕ) on M around p, with coordi-
nates ϕ = (x1ϕ , . . . , xm
ϕ ) such that ϕ(p) = 0 and ϕ(U ) = B is convex.
By Lemma 5.4.7, it suffices to show that the map
Tp M → Derp (C ∞ (U )), v 7→ Dv
is an isomorphism.
Let v ∈ Tp M and denote vϕ = (v 1 , . . . , v m ). By the formulas in Section 5.2,
m
X ∂xi
Dv (xiϕ ) = dp xiϕ (v) = vj (0) = v i ;
j=1
∂xj
thus, the map is injective.
Let D ∈ Derp (C ∞ (U )). We show that D vanishes on constant functions. Since
D(1) = D(1 · 1) = 1 · D(1) + D(1) · 1 = 2 · D(1),
we obtain that D(1) = 0. By linearity, D(c) = D(c · 1) = c · D(1) = 0, for all c ∈ R.
Let v ∈ Tp M be the vector with vϕ = (v 1 , . . . , v m ) ∈ Rm , where v i := D(xiϕ ).
We show that D = Dv . Let f ∈ C ∞ (U ). Applying Lemma 5.4.8, we find functions
gi ∈ C ∞ (U ), for 1 ≤ i ≤ m, such that
Xm  ∂ 
f = f (p) + xiϕ gi , gi (p) = dp f .
i=1
∂xiϕ p
Applying D, we obtain:
m
X
D(f ) = D(f (p)) + D(xiϕ gi ) =
i=1
m
X
0 · D(gi ) + D(xiϕ ) · gi (p) =


=0+
i=1
m
X  ∂ 
= vi dp f = dp f (v) = Dv (f ).
i=1
∂xiϕ p
Thus, D = Dv . 
 LECTURE 5. 49

5.5. Exercises
Exercise 5.1. Let M ⊂ Rn be a manifold embedded in Rn , as in Definition 2.4.1,
and let p ∈ M . Let Vp ⊂ Rn consist of all vectors γ 0 (0) ∈ Rn , where γ : (−, ) → M
is a smooth curve such that γ(0) = p. Show that Vp is a linear subspace of Rn , and
that there is a linear isomorphism Vp − ∼→ Tp M , such that γ 0 (0) ∈ Vp corresponds
d
to d0 γ( dt |t=0 ) ∈ Tp M .
The following exercise gives an alternative definition of the tangent space:
Exercise 5.2. Let M be a manifold and let p ∈ M . Denote by Ip ⊂ C ∞ (M ) the
ideal of functions vanishing at p:
Ip = {f ∈ C ∞ (M ) : f (p) = 0}.
Denote by Ip2 the square of this ideal, i.e. Ip2 consists of finite sums of the form:
k
X
f= gj hj , gj , hj ∈ Ip , 1 ≤ j ≤ k.
j=1

(a) Prove that Ip /Ip2 has dimension m = dim(M ).

Hint: use Lemma 5.4.8 and Lemma 4.1.2.
(b) Show that dp f (v) = 0 for all f ∈ Ip2 and v ∈ Tp M .
(c) Prove that the following map is a linear isomorphism:
E : Tp M → (Ip /Ip2 )∗ , E(v)(f + Ip2 ) := dp f (v), f ∈ Ip .
We introduce some terminology for the following exercises.
Definition 5.5.1. An algebra over R is a real vector space A endowed with an
associative, bilinear multiplication · : A × A → A, and with a unit 1A ∈ A; i.e.
(f · g) · h = f · (g · h), 1A · f = f = f · 1A ,
(λf + µg) · h = λ(f · h) + µ(g · h), f · (λg + µh) = λ(f · g) + µ(f · h),
for all f, g, h ∈ A and all λ, µ ∈ R. The algebra A is called commutative if
f · g = g · f for all f, g ∈ A.
A homomorphism between the R-algebras A and B is a linear map s : A → B,
which preserves multiplication and units:
s(f · g) = s(f ) · s(g) and s(1A ) = 1B .
Definition 5.5.2. Let A and R be commutative algebras over R, and let s : A → R
be an algebra homomorphism. A derivation at s of A is a linear map
D : A −→ R s.t. D(u · v) = s(u)D(v) + D(u)s(v), for all u, v ∈ A.
We denote the space of all derivations at s by:
Ders (A, R).
Definition 5.5.3. Let A be a commutative algebra over R. A character of A is
an algebra homomorphism χ : A → R.
Definition 5.5.4. For a commutative algebra R, let R[X] denote the polynomial
algebra with coefficients in R, i.e. each element in p ∈ R[X] can be written as a
finite sum:
p = u0 + u1 X + . . . + uk X k ,
50 IOAN MĂRCUT
, , MANIFOLDS

for unique u0 , . . . , uk ∈ R; the algebraic operations are defined as usually for poly-
nomials. Consider the following algebra:
R[] := R[X]/(X 2 ),
where (X 2 ) ⊂ R[X] is the ideal generated by X 2 , i.e. each element p ∈ R[] can be
uniquely written as
p = u0 + u1 ,
where  := X + (X 2 ) ∈ R[X]/(X 2 ); and the multiplication is such that 2 = 0:
(u0 + u1 ) · (v0 + v1 ) = u0 v0 + (u0 v1 + u1 v0 ).
Exercise 5.3. (a) Let s : A → R be an algebra homomorphism, and let D ∈
Ders (A, R). Prove that the map
s + D : A −→ R[], (s + D)(u) := s(u) + D(u)
is an algebra homomorphism.
(b) Prove that any algebra homomorphism σ : A → R[] is of the form described
at (a).
Exercise 5.4. Let s : A → R be a homomorphism of algebras.
(a) Prove that a homomorphism of algebras t : B → A induces a linear map
between the derivation spaces, defined as follows:
t∗s : Ders (A, R) −→ Ders◦t (B, R), t∗s (D) := D ◦ t.
(b) If t is surjective, prove that t∗s is injective.
(c) Prove that the following “chain rule” holds: if t : B → A and r : C → B are
algebra homomorphisms, then
(t ◦ r)∗s = rs◦t
∗
◦ t∗s : Ders (A, R) −→ Ders◦t◦r (C, R).
Exercise 5.5. Let X be a topological space, and let C(X) be the algebra of
continuous maps X → R. Consider a subalgebra A ⊂ C(X), with unit 1A the
constant map 1A (p) = 1, and which satisfies the condition:
(*) (f ∈ A, f (x) 6= 0, ∀ x ∈ X) =⇒ 1/f ∈ A.
We denote by A∨ the set of characters. For χ ∈ A∨ , let Iχ denote the kernel of χ:
Iχ := {f ∈ A : χ(f ) = 0}.
(a) Let p ∈ X. Prove that the following map gives a character of A:
χp : A −→ R, χp (f ) := f (p).
We denote Ip := Iχp .
(b) For χ ∈ A∨ , and p ∈ X, prove that Ip ⊂ Iχ , implies χ = χp .
(c) Prove that if f ∈ Iχ , then there exists p ∈ X such that f (p) = 0.
(d) Let χ ∈ A∨ be a general character, and assume that χ 6= χp for every p ∈ X.
Prove that for any point p ∈ X there exists f ∈ Iχ such that f (p) > 0.
(e) Prove that if X is compact, then every χ ∈ A∨ is of the form χ = χp for some
p ∈ X.
(f) Assume that there exists f ∈ A such that f −1 (λ) ⊂ X is compact for any
λ ∈ R. Show that every χ ∈ A∨ is of the form χ = χp for some p ∈ X.
 LECTURE 5. 51

Exercise 5.6. Prove that for any algebra homomorphism χ : C ∞ (M ) → R (i.e.

any character of C ∞ (M )) there exists a unique point p ∈ M such that:
χ(f ) = f (p), for all f ∈ C ∞ (M ).
Hint: use Exercise 5.5 and Corollary 4.4.4.
Exercise 5.7. Let M be a manifold. Show that there is a one-to-one correspon-
dence between tangent vectors on M and algebra homomorphisms C ∞ (M ) → R[].
Hint: use Exercises 5.3 and 5.6.
The following exercise proves Theorem 5.3.2.
Exercise 5.8. Let s : C ∞ (N ) → C ∞ (M ) be an algebra homomorphism.
(a) Prove that there exists a unique function ϕ : M → N such that s(f ) = f ◦ ϕ
for all f ∈ C ∞ (N ). Hint: use Exercise 5.6.
(b) Prove that for any smooth function f ∈ C ∞ (N ), the set
Uf := {p ∈ N : f (p) 6= 0}
is open, and that these sets forms a basis for the topology of N .
Hint: use Lemma 4.1.2.
(c) Prove that ϕ is continuous.
(d) Prove that ϕ is smooth.