- Prove that the tangent space
T
p
M
T_pM
TpM is an n-dimensional vector space by showing that
the tangent vectors { ∂ ∂ x i ∣ p } 1 ≤ i ≤ n \{\frac{\partial}{\partial x^i}\big|_p\}_{1\leq i \leq n} {∂xi∂∣∣p}1≤i≤n forms a basis.
Proof: \textbf{Proof:} Proof:
We know that T p M T_pM TpM is the functional space: { X p ∣ X p is linear funcitonal on C ∞ ( M ) satisfies Leibniz rule } \{X_p|X_p \text{ is linear funcitonal on }C^{\infty}(M) \text{ satisfies Leibniz rule}\} {Xp∣Xp is linear funcitonal on C∞(M) satisfies Leibniz rule} only need to show it has a finite basis.
∀
f
∈
C
∞
(
M
)
\forall f\in C^{\infty}(M)
∀f∈C∞(M), we have a general Taylor’s formula with remainder:
f
(
x
)
=
f
(
p
)
+
(
x
i
−
p
i
)
∂
f
∂
x
i
+
g
(
x
)
,
x
∈
M
f(x)=f(p)+(x^i-p^i)\frac{\partial f}{\partial x^i}+g(x), x\in M
f(x)=f(p)+(xi−pi)∂xi∂f+g(x),x∈M
X
p
(
g
)
=
0
,
X
p
(
c
)
=
0
,
∀
c
∈
R
X_p(g)=0,X_p(c)=0,\forall c\in \mathbb{R}
Xp(g)=0,Xp(c)=0,∀c∈R, since
p
p
p is second order root of
g
(
x
)
g(x)
g(x) and
X
p
X_p
Xp satisfies Leibniz rule.
Thus,
X
p
(
f
)
=
X
p
(
(
x
i
−
p
i
)
∂
f
∂
x
i
)
=
X
p
(
x
i
)
∂
∂
x
i
∣
p
⋅
f
X_p(f)=X_p((x^i-p^i)\frac{\partial f}{\partial x^i})=X_p(x^i)\frac{\partial}{\partial x^i}|_p\cdot f
Xp(f)=Xp((xi−pi)∂xi∂f)=Xp(xi)∂xi∂∣p⋅f i.e.
X
p
=
X
p
(
x
i
)
∂
∂
x
i
∣
p
X_p=X_p(x^i)\frac{\partial}{\partial x^i}|_p
Xp=Xp(xi)∂xi∂∣p, which means that
X
p
X_p
Xp can be represented in a linear combination of
{
∂
∂
x
i
∣
p
}
1
≤
i
≤
n
\{\frac{\partial}{\partial x^i}\big|_p\}_{1\leq i \leq n}
{∂xi∂∣∣p}1≤i≤n, and the coefficient is
X
p
(
x
i
)
X_p(x^i)
Xp(xi).
Next, we need to show { ∂ ∂ x i ∣ p } 1 ≤ i ≤ n \{\frac{\partial}{\partial x^i}\big|_p\}_{1\leq i \leq n} {∂xi∂∣∣p}1≤i≤n is linear independent:
Suppose a i ∂ ∂ x i ∣ p = 0 a_i \frac{\partial}{\partial x^i}\big|_p=\textbf{0} ai∂xi∂∣∣p=0, then a i ∂ ∂ x i ∣ p ⋅ x j = a i δ i j = a j = 0 , ∀ 1 ≤ j ≤ n a_i \frac{\partial}{\partial x^i}\big|_p \cdot x^j=a_i \delta_{ij}=a_j=0, \forall 1\leq j\leq n ai∂xi∂∣∣p⋅xj=aiδij=aj=0,∀1≤j≤n.
Thus
T
p
M
T_pM
TpM is an n-dimensional vector space, with basis
{
∂
∂
x
i
∣
p
}
1
≤
i
≤
n
\{\frac{\partial}{\partial x^i}\big|_p\}_{1\leq i \leq n}
{∂xi∂∣∣p}1≤i≤n.\
- Let
{
v
i
}
i
=
1
n
⊂
T
p
M
\{v_i\}^n_{i=1}\subset T_pM
{vi}i=1n⊂TpM be a basis of tangent vectors at
p
p
p. Show that there is a local chart
(
U
,
φ
;
x
i
)
(U,\varphi;x^i)
(U,φ;xi) around
p
p
p such that
v i = ∂ ∂ x i ∣ p , i = 1 , … , n . v_i=\frac{\partial}{\partial x^i}\big|_p, i=1,\dots, n. vi=∂xi∂∣∣p,i=1,…,n.
Proof: \textbf{Proof:} Proof:
Here we define { v i } i = 1 n \{v_i\}^n_{i=1} {vi}i=1n as the equivalent relation of curves on M M M, marked v i : = [ v i ] v_i:=[v_i] vi:=[vi]. Suppose these is define under ( U , ϕ ) (U,\phi) (U,ϕ).
Then since
{
v
i
}
i
=
1
n
⊂
T
p
M
\{v_i\}^n_{i=1}\subset T_pM
{vi}i=1n⊂TpM is a basis, we know
(
[
v
i
]
⋅
ϕ
1
,
…
,
[
v
i
]
⋅
ϕ
n
)
([v_i]\cdot \phi^1,\dots,[v_i]\cdot \phi^n)
([vi]⋅ϕ1,…,[vi]⋅ϕn) is linear independent
∀
1
≤
i
≤
n
\forall 1\leq i\leq n
∀1≤i≤n, also we have addition and production on
T
p
M
T_pM
TpM, i.e. we can represent the operator in matrix form, and thus
A
A
A is invertible:
(
[
v
1
]
[
v
2
]
⋮
[
v
n
]
)
⋅
(
ϕ
1
ϕ
2
…
ϕ
n
)
=
A
\begin{pmatrix} [v_1]\\ [v_2]\\ \vdots \\ [v_n]\\ \end{pmatrix} \cdot \begin{pmatrix} \phi^1 & \phi^2 & \dots & \phi^n \end{pmatrix} =A
⎝⎜⎜⎜⎛[v1][v2]⋮[vn]⎠⎟⎟⎟⎞⋅(ϕ1ϕ2…ϕn)=A
Then we can take
A
−
1
A^{-1}
A−1 on both sides:
(
[
v
1
]
[
v
2
]
⋮
[
v
n
]
)
⋅
(
(
ϕ
1
ϕ
2
…
ϕ
n
)
⋅
A
−
1
)
=
I
\begin{pmatrix} [v_1]\\ [v_2]\\ \vdots \\ [v_n]\\ \end{pmatrix} \cdot \Big(\begin{pmatrix} \phi^1 & \phi^2 & \dots & \phi^n \end{pmatrix} \cdot A^{-1} \Big) =I
⎝⎜⎜⎜⎛[v1][v2]⋮[vn]⎠⎟⎟⎟⎞⋅((ϕ1ϕ2…ϕn)⋅A−1)=I
Since
ϕ
\phi
ϕ is a local chart, we know the linear combination of
ϕ
i
\phi^i
ϕi is still a local chart:
(
U
,
φ
)
(U,\varphi)
(U,φ), where
φ
=
(
ϕ
1
ϕ
2
…
ϕ
n
)
⋅
A
−
1
\varphi=\begin{pmatrix} \phi^1 & \phi^2 & \dots & \phi^n \end{pmatrix} \cdot A^{-1}
φ=(ϕ1ϕ2…ϕn)⋅A−1. And
φ
\varphi
φ satisfies that
φ
j
∘
v
i
=
δ
i
j
t
\varphi^j\circ v_i=\delta_{ij}t
φj∘vi=δijt.
Thus, by definition on
(
U
,
φ
)
(U,\varphi)
(U,φ), we have
v
i
=
∂
∂
x
i
∣
p
,
i
=
1
,
…
,
n
v_i=\frac{\partial}{\partial x^i}\big|_p, i=1,\dots, n
vi=∂xi∂∣∣p,i=1,…,n.\
- Show that the cotangent space T p ∗ M T_p^*M Tp∗M is an n-dimensional vector space by showing that { d x i ∣ p } ⊂ T p ∗ M \{dx^i|_p\}\subset T_p^*M {dxi∣p}⊂Tp∗M forms a basis. Moreover, T p ∗ M T_p^*M Tp∗M is the dual space of the tangent space T p M T_pM TpM.
Proof: \textbf{Proof:} Proof:
First, claim that on a local chart ( U , ϕ ) (U, \phi) (U,ϕ), ∀ f ∈ C ∞ ( M ) \forall f\in C^{\infty}(M) ∀f∈C∞(M), d f ∣ p = ∂ ∂ x i ( f ∘ ϕ − 1 ) d x i ∣ p df|_p=\frac{\partial}{\partial x^i}(f\circ\phi^{-1})dx^i|_p df∣p=∂xi∂(f∘ϕ−1)dxi∣p.
Set
v
i
=
∂
∂
x
i
(
f
∘
ϕ
−
1
)
v_i=\frac{\partial}{\partial x^i}(f\circ\phi^{-1})
vi=∂xi∂(f∘ϕ−1), then we need to show
d
f
∣
p
=
d
v
i
x
i
df|_p=dv_ix^i
df∣p=dvixi.
i.e.
f
∼
v
i
x
i
f\sim v_ix^i
f∼vixi
⟺
\iff
⟺
∂
∂
x
j
(
f
∘
ϕ
−
1
)
=
∂
∂
x
j
⋅
v
i
x
i
\frac{\partial}{\partial x^j}(f\circ\phi^{-1})=\frac{\partial}{\partial x^j}\cdot v_ix^i
∂xj∂(f∘ϕ−1)=∂xj∂⋅vixi, by the definition of
∂
∂
x
i
∣
p
⋅
\frac{\partial}{\partial x^i}|_p\cdot
∂xi∂∣p⋅ we know that the latter equals
v
j
=
L
H
S
v_j=LHS
vj=LHS.
Thus, the claim is proved. Before we show { d x i ∣ p } , i = 1 , … , n \{dx^i|_p\}, i=1,\dots,n {dxi∣p},i=1,…,n is linear independent, I prefer to show T p ∗ M T_p^*M Tp∗M is the dual space of the tangent space T p M T_pM TpM:
which is equivalent to show T p ∗ M = ( T p M ) ∗ T_p^*M=(T_pM)^* Tp∗M=(TpM)∗, i.e. ∀ [ f ] p ∈ T p ∗ M \forall [f]_p\in T_p^*M ∀[f]p∈Tp∗M, it can be realized a linear function on T p M T_pM TpM. Define d f ∣ p : T p M → R 1 , [ γ ] p ↦ ( d f ∣ p , [ γ ] p ) : = d d t ∣ t = 0 ( f ∘ γ ) df|_p:T_pM\rightarrow \mathbb{R}^1, [\gamma]_p\mapsto (df|_p, [\gamma]_p):=\frac{d}{dt}\big|_{t=0} (f\circ \gamma) df∣p:TpM→R1,[γ]p↦(df∣p,[γ]p):=dtd∣∣t=0(f∘γ).
Then we need to show linearity:
∀
d
f
∣
p
,
d
g
∣
p
∈
T
p
∗
M
,
a
,
b
∈
R
1
\forall df|_p,dg|_p\in T_p^*M,a,b\in\mathbb{R}^1
∀df∣p,dg∣p∈Tp∗M,a,b∈R1, for
a
⋅
d
f
∣
p
+
b
⋅
d
g
∣
p
a\cdot df|_p+b\cdot dg|_p
a⋅df∣p+b⋅dg∣p
a
⋅
d
(
f
∘
ϕ
−
1
)
p
+
b
⋅
d
(
g
∘
ϕ
−
1
)
p
=
d
(
(
a
f
+
b
g
)
∘
ϕ
−
1
)
∣
p
a\cdot d(f\circ \phi^{-1})_p+b\cdot d(g\circ \phi^{-1})_p=d\big((af+bg)\circ \phi^{-1})|_p
a⋅d(f∘ϕ−1)p+b⋅d(g∘ϕ−1)p=d((af+bg)∘ϕ−1)∣p, follows the linearity of
C
∞
(
M
)
C^{\infty}(M)
C∞(M). Thus we can define
a
⋅
d
f
∣
p
+
b
⋅
d
g
∣
p
=
d
(
(
a
f
+
b
g
)
∘
ϕ
−
1
)
∣
p
a\cdot df|_p+b\cdot dg|_p=d\big((af+bg)\circ \phi^{-1})|_p
a⋅df∣p+b⋅dg∣p=d((af+bg)∘ϕ−1)∣p
Thus, T p ∗ M = ( T p M ) ∗ T_p^*M=(T_pM)^* Tp∗M=(TpM)∗.
Finally, it’s time to show the linear independent of
{
d
x
i
∣
p
}
\{dx^i|_p\}
{dxi∣p}: suppose
a
i
d
x
i
∣
p
=
0
a_idx^i|_p=\textbf{0}
aidxi∣p=0, act
[
γ
j
]
[\gamma^j]
[γj] both sides,
a
i
⋅
(
d
x
i
∣
p
,
[
γ
j
]
)
=
a
i
δ
i
j
=
a
j
=
0.
a_i\cdot (dx^i|_p, [\gamma^j])=a_i\delta_{ij}=a_j=0.
ai⋅(dxi∣p,[γj])=aiδij=aj=0.
Thus,
{
d
x
i
∣
p
}
,
i
=
1
,
…
,
n
\{dx^i|_p\}, i=1,\dots,n
{dxi∣p},i=1,…,n is linear independent.
Moreover,
(
d
x
i
,
∂
∂
x
j
∣
p
)
=
(
d
x
i
,
[
γ
j
]
)
=
d
d
t
∣
t
=
0
(
x
i
∘
γ
j
)
=
δ
i
j
(dx^i,\frac{\partial}{\partial x^j}\big|_p)=(dx^i, [\gamma^j])=\frac{d}{dt}\big|_{t=0} (x^i\circ \gamma^j)=\delta_{ij}
(dxi,∂xj∂∣∣p)=(dxi,[γj])=dtd∣∣t=0(xi∘γj)=δij. This suggests that
{
d
x
i
∣
p
}
\{dx^i|_p\}
{dxi∣p} is the dual basis of
(
T
p
M
)
∗
(T_pM)^*
(TpM)∗.\
- Show that under a change of local charts, the natural basis of the cotangent space
T
p
∗
M
T_p^*M
Tp∗M transforms by
d x i ∣ p = ∂ x i ∂ y j ( y 0 ) d y j ∣ p . dx^i|_p=\frac{\partial x^i}{\partial y^j}(y_0)dy^j|_p. dxi∣p=∂yj∂xi(y0)dyj∣p.
Proof:
\textbf{Proof:}
Proof:
Suppose we have two local charts:
(
U
,
ϕ
;
x
i
)
,
(
V
,
φ
;
y
i
)
(U,\phi;x^i),(V,\varphi;y^i)
(U,ϕ;xi),(V,φ;yi), thus we have two basis:
{
d
x
i
∣
p
}
,
{
d
y
i
∣
p
}
⊂
T
p
∗
M
\{dx^i|_p\}, \{dy^i|_p\}\subset T^*_pM
{dxi∣p},{dyi∣p}⊂Tp∗M,
∀
d
x
i
\forall dx^i
∀dxi, we can represent it with
{
d
y
i
∣
p
}
\{dy^i|_p\}
{dyi∣p} :
d
x
i
∣
p
=
a
k
d
y
k
∣
p
dx^i|_p=a_kdy^k|_p
dxi∣p=akdyk∣p.
By the result of partial derivative on different local charts, we know
∂
∂
y
l
∣
p
=
∂
x
j
∂
y
l
(
y
0
)
∂
∂
x
j
∣
p
\frac{\partial}{\partial y^l}\big|_p=\frac{\partial x^j}{\partial y^l}(y_0) \frac{\partial}{\partial x^j}\big|_p
∂yl∂∣∣p=∂yl∂xj(y0)∂xj∂∣∣p
act on
d
x
i
∣
p
dx^i|_p
dxi∣p, also note that
(
d
x
i
,
∂
∂
x
j
∣
p
)
=
δ
i
j
(dx^i,\frac{\partial}{\partial x^j}\big|_p)=\delta_{ij}
(dxi,∂xj∂∣∣p)=δij:
d
x
i
(
∂
∂
y
l
∣
p
)
=
d
x
i
(
∂
x
j
∂
y
l
(
y
0
)
∂
∂
x
j
∣
p
)
=
∂
x
j
∂
y
l
δ
i
j
=
∂
x
i
∂
y
l
=
a
k
d
y
k
(
∂
∂
y
l
∣
p
)
=
a
l
dx^i(\frac{\partial}{\partial y^l}\big|_p)=dx^i(\frac{\partial x^j}{\partial y^l}(y_0) \frac{\partial}{\partial x^j}\big|_p)=\frac{\partial x^j}{\partial y^l}\delta_{ij}=\frac{\partial x^i}{\partial y^l}=a_kdy^k(\frac{\partial}{\partial y^l}\big|_p)=a_l
dxi(∂yl∂∣∣p)=dxi(∂yl∂xj(y0)∂xj∂∣∣p)=∂yl∂xjδij=∂yl∂xi=akdyk(∂yl∂∣∣p)=al
thus, we know
a
k
=
∂
x
i
∂
y
k
a_k=\frac{\partial x^i}{\partial y^k}
ak=∂yk∂xi
therefore,
d
x
i
∣
p
=
∂
x
i
∂
y
k
d
y
k
∣
p
dx^i|_p=\frac{\partial x^i}{\partial y^k}dy^k|_p
dxi∣p=∂yk∂xidyk∣p