You are on page 1of 2

LECTURE 2

Now that we have defined the notion of manifold, let us consider two simple ways of
generating new manifolds:
Exercise 8. Let X ⊂ Rm and Y ⊂ Rn smooth manifolds, then X × Y ⊂ Rm+n is also a
smooth manifold of dimension dim(X + Y ) = dim X + dim Y .
Definition 9. Let Y ⊂ X ⊂ Rn so that X and Y are both smooth manifolds in Rn . Then
Y is called a submanifold of X. The codimension of Y in X is dim X − dim Y .
Example 10. A manifold X ⊂ Rn is in particular a submanifold of Rn . Given X ⊂ Rn
manifold, any V ⊂ X that is open for the induced topology in X is a submanifold of X.

2.2 Tangent space and derivatives


Definition 11 (Revision from Analysis II). Given m, m ∈ N∗ and U ⊂ Rn open and f : U →
Rm smooth, the differential Dfq at q ∈ U is defined by f (q + h) = f (q) + Dfq (h) + O(|h|2 ) for
h ∼ 0. It is a linear map from Rn to Rm . If we denote f = (f1 , . . . , fm ) and q = (x1 , . . . , xn ),
the linear map Dfq : Rn → Rm identifies in these maps with the matrix
 ∂f1 ∂f1 
∂x1 (q) . . . ∂xn (q)
 ..  .
Dfq =  ... . 
∂fm ∂fm
∂x1 (q) ... ∂xn (q)

In the latter definition the fluctuation h ∈ Rn describes the whole Euclidean space. This is
not necessarily the case anymore when h is restricted to the tangent directions to a manifold.
In other world, the linear approximation is restricted to certain direction.
Definition 12. Given X ⊂ Rn a smooth k-manifold and x ∈ X we define the tangent space
Tx X at x as follows: given U0 ⊂ Rk open in Rk around 0, Vx ⊂ X open in X around x, and
a parametrisation ϕ : U → Vx , the tangent space is (the vector space) Tx X := Dϕ0 (Rk ).
It is not immediately clear that this definition is correct, since it assumes that Tx X is
independent of the choice of parametrisation (remember the parametrisation is not unique).
This is the object of the following proposition.
Proposition 13. Given X ⊂ Rn a smooth k-manifold and x ∈ X, the tangent space Tx X is
independent of the choice of parametrisation around x, and dim Tx X = dim X = k.
Remark 14. Note that the two “dim” applies to slightly different objects in the last equality.

Proof. Assume ϕ : U0 → Vx and ψ : Ũ0 → Ṽx are two parametrisations as in the last
definition. By restricting them to the common range Wx := Vx ∩ Ṽx (which is still an open
neighbourhood of x in X), we obtain the two parametrisations ϕ| : U0 ∩ ϕ−1 (Wx ) → Wx and
ψ| : V0 ∩ψ −1 (Wx ) → Wx . We then define the transition map θ : U0 ∩ϕ−1 (Wx ) → V0 ∩ψ −1 (Wx )
by θ : ψ|−1 ◦ ϕ| . It is a diffeomorphism by composition, and since it is defined between open
sets of Rk , this means the “standard” notion of smoothness and diffeomorphism. By the chain
rule Dϕ0 = Dψ0 ◦ Dθ0 (θ maps 0 to 0), so the range of Dϕ0 is equal to the range of Dψ0 .
The dimension of the tangent space is k since the linear maps involved are invertible.

4
Example 15. • Given U ⊂ Rn open and x ∈ U , the tangent space Tx U = Rn (use the trivial
identity parametrisation). In particular Rn is its own tangent space at every point (which is
why the notion of tangent space is “obscured” in Euclidean geometry).
• Given p ∈ S1 , the tangent space Tp S1 at p is the line p + p⊥ ⊂ R2 . More generally given
p ∈ Sn , the tangent space Tp Sn at p is the affine hyperplan p + p⊥ ⊂ Rn+1 . Check it with the
parametrisations of the previous example.
Now that we have a notion of tangent space, the next step is obviously to define a notion
of differential for smooth maps between manifolds.
Definition 16. Given X ⊂ Rn and Y ⊂ Rm smooth manifolds of dimensions k and l
and f : X → Y smooth and x ∈ X, we define Dfx the differential of f at x as follows:
given Ux ⊂ X open around x in X, Vf (x) open around f (x) in Y , with f (Ux ) ⊂ Vf (x) , and
parametrisations ϕ : U0 ⊂ Rk → Ux ⊂ X and ψ : V0 ⊂ Rl → Vf (x) ⊂ Y , and the transition
map θ = ψ −1 ◦ f ◦ ϕ : Rk → Rl , then Dfx := Dψ0 ◦ Dθ0 ◦ (Dϕ0 )−1 : Tx X → Tf (x) Y .
When X is open in Rn it coincides with the usual differential by using the identity
parametrisations in X. However, we first need to prove that the definition is correct:
Proposition 17. Dfx thus defined is independent of the choice of parametrisations.
Proof. Consider two pairs of parametrisations (ϕ, ψ) and (ϕ̃, ψ̃) and, without any effect on the
definition, we reduce if necessary the size of the open sets so that ϕ : Ux → Ux , ϕ̃ : Ũx → Ux ,
ψ : V0 → Vx , ψ̃ : Ṽ0 → Vx (the arrival neighbourhoods are the same). Then the two
corresponding transition maps θ and θ̃ are related by θ = ψ −1 ◦ ψ̃ ◦ θ̃ ◦ ϕ̃−1 ◦ ϕ. We apply the
(standard) chain rule in Euclidean spaces:
   
Dθ0 = D ψ −1 ◦ ψ̃ ◦ Dθ̃0 ◦ D ϕ̃−1 ◦ ϕ .
0 0
Therefore the definition of the differential with the first parametrisation is
   
Dψ0 ◦ Dθ0 ◦ (Dϕ0 )−1 = Dψ0 ◦ D ψ −1 ◦ ψ̃ ◦ Dθ̃0 ◦ D ϕ̃−1 ◦ ϕ ◦ (Dϕ0 )−1
0 0

and since ψ̃ = ψ ◦ (ψ −1 ◦ ψ̃) locally and ϕ = ϕ̃ ◦ (ϕ̃−1 ◦ ϕ) locally we deduce


   
Dψ0 ◦ D ψ −1 ◦ ψ̃ = Dψ̃0 and D ϕ̃−1 ◦ ϕ ◦ (Dϕ0 )−1 = (Dϕ̃0 )−1
0 0
which finally proves
Dψ0 ◦ Dθ0 ◦ (Dϕ0 )−1 = Dψ̃0 ◦ Dθ̃0 ◦ (Dϕ̃0 )−1
and concludes the proof.
Remark 18. Note that this notion of differential also satisfies the familiar chain rule: given
X, Y , Z three manifolds, f : X → Y and g : Y → Z smooth, ϕ parametrisation around
x ∈ X, ψ parametrisation around f (x) ∈ Y , ζ parametrisation around g ◦ f (x) ∈ Z, theta
and λ the two corresponding transition maps, we have
D(g ◦ f )x = Dζ0 ◦ D(λ ◦ θ)0 ◦ (Dϕ0 )−1 = Dζ0 ◦ Dλ0 ◦ Dθ0 ◦ (Dϕ0 )−1
by using the standard chain rule on the transition maps, and using (Dψ0 )−1 ◦ Dψ0 = Id:
D(g ◦ f )x = Dζ0 ◦ Dλ0 ◦ (Dψ0 )−1 ◦ Dψ0 ◦ Dθ0 ◦ (Dϕ0 )−1 = Dgf (x) ◦ Dfx .

You might also like