JIA-MING (FRANK) LIOU
Let K be a field. All the vector spaces in this note are over K.
1. Some Linear Algebras
Let V be an n-dimensional vector space over K. A linear functional on V is a linear map ϕ : V → K. The set of all linear functionals, denoted by V∗, forms a vector space. The addition and scalar multiplication are given as follows. Let ϕ, ψ ∈ V∗, and a ∈ k. Define ϕ + ψ and aϕ by
(ϕ + ψ)(v) = ϕ(v) + ψ(v), (aϕ)(v) = aϕ(v).
Theorem 1.1. V∗ has dimension n.
Proof. Let {v1, · · · , vn} be a basis for V. Let ψi : V → K such that ψi(vj) = δij. Claim that the set β = {ψi: 1 ≤ i ≤ n} forms a basis for V∗.
For any ϕ ∈ V∗, one can check that ϕ =
n
X
i=1
ϕ(ei)ψi. Then ϕ is a linear combinaton of elements of β.
Assume that Pn
i=1aiψi = 0 for some a1, · · · , an ∈ K. Evaluate the sum for vk, we see that ak= 0. This shows that β is linearly independent.
Remark. The basis {ψi} for V∗ in Theorem (1.1) is called the dual basis of {vi}.
Let ϕ1, ϕ2 ∈ V∗. We define a map ϕ1∧ ϕ2: V × V → K by (ϕ1∧ ϕ2)(w1, w2) = det [ϕi(wj)]2i,j=1.
We can check that ϕ1∧ ϕ2 : V × V → K is bilinear, and skew-symmetric, i.e. ϕ2∧ ϕ1 =
−ϕ1 ∧ ϕ2. We call ϕ1∧ ϕ2 the wedge product of ϕ1 and ϕ2. Let Λ2V∗ be the set of all skew-symmetric bilinear maps from V × V to K.
Theorem 1.2. Λ2V∗ is a vector space of dimensionn 2
.
Proof. Let {vi} be a basis for V and {ψi} be its dual basis. Claim that {ψi∧ ψj : 1 ≤ i <
j ≤ n} forms a basis for V.
Suppose that
X
i<j
aijψi∧ ψj = 0.
Evaluate the equation for (vk, vl), we obtain akl= 0.
Let B be a skew-bilinear map. Denote bij = B(vi, vj). Then bij = −bji. For any x = Pn
i=1xivi and y =P
iyivi, we see that B(x, y) =X
i,j
bijxiyj 1
=X
i<j
bijxiyj+X
i>j
bijxiyj
=X
i<j
bij(xiyj − xjyi).
Note that (ψi∧ ψj)(x, y) = (xiyj − xjyi). Hence we find B(x, y) = (X
i<j
bijψi∧ ψj)(x, y).
Hence we prove that B =P
i<jbijψi∧ ψj.
Definition 1.1. Let p be a natural number and Sp be the symmetric group on p-letters.
A map f : ×pi=1V → K is alternating if
f (vσ1, · · · , vσp) = (sgn σ)f (v1, · · · , vp), for any v1, · · · , vp ∈ V. f is called p-linear if for each 1 ≤ k ≤ p,
f (v1, · · · , avk+ bwk, · · · , vp) = af (v1, · · · , vk, · · · , vp) + bf (v1, · · · , wk, · · · , vp), for any a, b ∈ K.
Let 1 ≤ p ≤ n. Given ϕ1, · · · , ϕp ∈ V∗, we define ϕ1∧ · · · ∧ ϕp : ×pi=1V → k by (ϕ1∧ · · · ∧ ϕp)(w1, · · · , wp) = det [ϕi(wj)]pi,j=1,
for all w1, · · · , wk ∈ V. Then we can check that ϕ1∧ · · · ∧ ϕp is p-linear and alternating.
Let ΛpV∗ be the set of all p-linear alternating maps from ×pi=1V to k and denote Λ0V∗= V∗.
Theorem 1.3. For 0 ≤ p, ≤ n, the set ΛpV∗ forms a vector space of dimensionn p
. Proof. Let {vi} be a basis for V and {ψi} be a basis for V∗ dual to V. We only need to show that the set {ψi1 ∧ · · · ∧ ψip : 1 ≤ i1< · · · < ip ≤ n} forms a basis for ΛpV∗.
2. Tangent Vectors
Let p be a point in Rn. We denote (v)p = p + v, where + is the addition in Rn. Let TpRn be the set of all (v)p with v ∈ Rn. We define the addition and scalar multiplication on TpRn as follows. Let (v)p, (w)p ∈ TpRn, and a ∈ R. We define
(v)p+ (w)p = (v + w)p, a · (v)p= (av)p.
Note that the addition and scalar multiplication here are different from those of Rn. Proposition 2.1. The set TpRn forms an n-dimensional real vector space.
Proof. The proof is obvious. Let {ei : 1 ≤ i ≤ n} be the standard basis for Rn. Then {(ei)p: 1 ≤ i ≤ n} forms a basis for TpRn.
Remark. We will use the notation vp for (v)p.
We say that TpRn is the tangent space to Rn at p and elements of TpRn are tangent vectors at p. The dual space of TpRn is denoted by T∗Rn and called the cotangent space of Rn at p.
Let U be an open set on Rn. The algebra of (real-valued) smooth functions on U is denoted by C∞(U ). Let p ∈ U and f ∈ C∞(U ), we define the directional derivative of f at p along a vector v ∈ Rn by
vp[f ] = d
dtf (p + tv) t=0
.
Let {ei : 1 ≤ i ≤ n} be the standard basis for Rn. Assume that v = Pn
i=1viei. From calculus, we know
vp[f ] =
n
X
i=1
vi
∂f
∂xi
(p).
Then we know
(1) vp : C∞(U ) → R is a linear functional, i.e. vp(af + bg) = avp(f ) + bvp(g) for all a, b ∈ R and
(2) vp(f g) = vp(f )g(p) + f (p)vp(g).
Note that if the tangent vector is given, the directional derivative only depends on the partial derivatives of the function at the point p. Suppose f, g are two smooth functions defined on some open sets, (not necessarily the same). If f and g agree on an open set containing p, then vp(f ) = vp(g). This leads to another definition of tangent vectors of Rn. Let V, W be open sets containing p in Rn. Let f be a smooth function on V and g smooth function on W. We say that the pair (f, V ) is equivalent to the pair (g, W ) if there exists an open set Z ⊂ V ∩ W contain p such that f = g on Z. The set of all equivalent classes [(f, V )]
is denoted by Cp∞. For simplicity, we will simply denote [(f, V )] by [f ]. An equivalent class [f ] is called a germ at p.
Proposition 2.2. The set Cp∞forms an algebra over R.
Let [f ] ∈ Cp∞. We define [f ](p) = f (p), where f is a representative of [f ]. Then [f ](p) is well-defined. Now, if f1, f2 belong to the same germ [f ] at p, we can check that vp(f1) = vp(f2). Hence we can define vp([f ]) = vp(f ) where f is a representative of [f ].
Definition 2.1. A point derivation δp at p is a linear functional on Cp∞ such that δp([f ][g]) = δp([f ])[g](p) + [f ](p)δp([g]).
The set of all point derivations at p is denoted by Tp0Rn.
Remark. By definition, all tangent vectors are point derivation at p.
Given [f ] ∈ Cp∞, we define
∂
∂xi
(p)[f ] = ∂f
∂xi
(p) for a representative f of [f ]. Then we know that ∂
∂xi(p) is a point derivation.
Theorem 2.1. The set
∂
∂xi(p) : 1 ≤ i ≤ p
forms a basis for Tp0Rn; Then Tp0Rn is an n-dimensional vector space.
Proof. It is easy to verify that the set is linearly independent.
Let δp ∈ TpRn. Denote δp[xi] = vi. Let [f ] ∈ Cp∞. Choose a representative f of [f ].
Consider the Tayler expansion f (x) = f (p) +X
i
∂f
∂xi
(p)(xi− pi) +X
i,j
(xi− pi)(xj− pj) Z 1
0
(1 − t) ∂2f
∂xi∂xj
(p + t(x − p))dt.
Using the properties of tangent vectors, we find δp[f ] =X
i
vi ∂
∂xi
(p)[f ].
In other words, we find δp =X
i
vi
∂
∂xi
(p).
We know that TpRnis a vector subspace of Tp0Rn. Since they both have the same dimen- sion, they must be equal. In fact, by elementary calculus,
(ei)p[f ] = ∂f
∂xi(p).
We find that (ei)p = ∂
∂xi
(p) by definition. Hence we conclude that the notion of tangent vectors are equivalent to the notion of point derivations.
Remark. Since we always choose a representative of a germ at p, from now on, we will simply use the notation f for [f ].
Let f be a germ at p. We define a linear functional dfp: TpRn→ R by dfp(vp) = vp(f )
Then dfp ∈ Tp∗Rn. Notice that for any vp = (v1, · · · , v)p ∈ TpRn, we have (dxi)p(vp) = vi. This implies that (dxi)p((ei)p) = δij for all i, j. Hence {(dxi)p : 1 ≤ i ≤ p} is the dual basis to {(ei)p: 1 ≤ i ≤ n}. We conclude that:
Theorem 2.2. The set {(dxi)p: 1 ≤ i ≤ n} forms a basis for Tp∗Rn. 3. Tangent Maps
Let U be an open subset of Rn. Suppose F : U → Rm is a smooth map, where F = (F1, · · · , Fm). For each p ∈ U, we can define a linear map, called the tangent map,
dFp : TpRn→ TF (p)Rm as follows.
Let f be a germ at F (p). Then f ◦ F is a germ at p. Given any vp∈ TpRn, we set (dFp(vp)) (f ) = vp(f ◦ F ).
Suppose F = F (y1, · · · , ym). By chain rule, ∂
∂xi(p)(f ◦ F ) =
m
X
j=1
∂f
∂yj(F (p))∂Fj
∂xi(p). By definition, dFp
∂
∂xi
(p)
(f ) =
m
X
j=1
∂Fj
∂xi
(p)∂f
∂yj
(F (p)). Hence we see that
dFp
∂
∂xi(p)
=
m
X
j=1
∂Fj
∂xi(p) ∂
∂yj(F (p)).
If we choose the standard basis for TpRnand TF (p)Rm, the tangent map dFp is represented by the Jacobi matrix of F at p.
Let U be an open set in Rn and F : U → Rm be a smooth map. Assume that V is an open set in Rm containing F (U ) and G : V → Rkis a smooth map. For any x ∈ U, we have (chain rule)
(3.1) d(G ◦ F )x: TxRn −−−−→ TdFx F (x)Rm
dGF (x)
−−−−→ TG(F (x)). 4. Differential Forms on Rn
Let U be an open set in Rn. We denote T U = [
p∈U
TpRn, T∗U = [
p∈U
Tp∗Rn.
Then we find that T U can be identify with U ×Rnand T∗U can be identified with U ×(Rn)∗. We have a natural projection π : T∗U → U.
Let f be a smooth function on an open set U ⊂ Rn. The total derivative of f has the formal expression:
df =X
i
∂f
∂xidxi.
Since f is smooth, all the partial derivatives of f are all smooth. We shall use this idea to define the notion of one-form. A smooth one form ω on U is a formal expression
ω =X
i
ωidxi
with ωi ∈ C∞(U ). Now we can think of ω as a map from U to T∗U. In fact, for each p ∈ U, ω(p) =P
iωi(p)(dxi)p ∈ Tp∗Rn. Moreover, (π ◦ ω)(p) = p for all p ∈ U. Therefore a smooth one-form ω is a smooth map ω : U → T∗U such that π ◦ ω = 1U.
Let ΛkT∗U =S
p∈U ΛkTp∗Rn. We can also consider the natural projection1 π : ΛkT∗U → U. We can check that ΛkT∗U can be identified with U × Λk(Rn)∗.
Definition 4.1. A smooth k-form on U is a smooth map η : U → ΛkT∗U such that π ◦ η = 1U. The set of all smooth k forms on U is denoted by Ωk(U ). We denote Ω∗(U ) = L
k≥0Ωk(U ).
By definition, a smooth k-form has the formal expression
η = X
1≤i1<···<ik≤n
ηi1···ikdxi1 ∧ · · · ∧ dxik with ηi1···ik ∈ C∞(U ).
Let I = (i1, · · · , ik) be a k-tuple. We write
dxI = dxi1 ∧ · · · ∧ dxik. A k-form η is also denoted by η =P
IηIdxI. The sum of two k-forms on U ω =P
IωIdxI and η =P
IηIdxI is defined to be
ω + η =X
I
(ω + η)dxI.
1Let us use the notation π for the projection but we have to know that this projection is different from the above.
For any smooth function f on U, we define f ω =X
I
(f ωI)dxI. One can check that
Proposition 4.1. The set Ωk(U ) is a free C∞(U )-module of rankn k
.
On Ω∗(U ), we define the exterior (wedge) product as follows. Given a s-form ω = P
IωIdxI and a k-form η =P
JηJdxJ, we define a s + k form by ω ∧ η =X
I,J
ωIηJdxI∧ dxJ.
The exterior product of forms on U has the following properties.
Theorem 4.1. Let ω, η, and θ be k, s, r-forms respectively. Then (1) (ω ∧ η) ∧ θ = ω ∧ (η ∧ θ),
(2) ω ∧ η = (−1)ksη ∧ ω,
(3) If r = s, then ω ∧ (η + θ) = ω ∧ η + ω ∧ θ.
Proof. Exercise.
Let ω =P
IωIdxI. We define the exterior derivative dω of ω by dω =X
I
dωI∧ dxI.
Proposition 4.2. d : Ω∗(U ) → Ω∗(U ) is a linear map such that (1) d : Ωk(U ) → Ωk+1(U )
(2) d(ω ∧ η) = dω ∧ η + (−1)kω ∧ dη (3) d2 = 0.
Proof. Exercise.
Let dk= d|Ωk(U ). Then d2 = 0 implies that dk+1dk = 0. In other words, Im dk−1⊂ ker dk. Definition 4.2. The k-th de Rham cohomology of U is the quotient space defined by
Hk(U ) = ker dk/ Im dk−1.
Later, we will study more about the de Rham cohomology of a smooth manifold.
4.1. Pull back on Differential Forms. Let F : U ⊂ Rn→ Rm be a smooth map. Then F induces a linear map F∗ from Ωk(V ) to Ωk(U ), where V is an open set containing F (U ).
Given a k-form ω on V, with k ≥ 1, we define a k-form F∗ω on U by
(F∗ω)(p)(v1, · · · , vk) = ω(F (p))(dFp(v1), · · · , dFp(vk)), ∀p ∈ U, where v1, · · · , vk ∈ TpRn. For k = 0, set F∗g = g ◦ F.
Proposition 4.3. Let F : U ⊂ Rn→ Rm and g ∈ C∞(U ). Suppose ω, η are k-forms. Then (1) F∗(ω + η) = F∗ω + F∗η,
(2) F∗(gω) = (F∗g)(F∗ω),
(3) If ϕ1, · · · , ϕk are one-forms, F∗(ϕ1∧ · · · ∧ ϕk) = F∗ϕ1∧ · · · ∧ F∗ϕk.
Let us assume F = (F1, · · · , Fm), i.e. Fi = yi ◦ F. Suppose that ω = P
IωIdyI. Then F∗ω =P
IF∗ωIF∗dyI. Using the properties, we have F∗dyi1∧· · ·∧F∗dyik = dFi1∧· · ·∧dFik. Then
F∗ω =X
I
ωI(F1, · · · , Fm)dFi1∧ · · · ∧ dFik. Using this identity, it is easy for us to prove the following corollary.
Corollary 4.1. Let ω, η be forms. Then F∗(ω ∧ η) = F∗ω ∧ F∗η.
Proposition 4.4. Let F : U ⊂ Rn→ Rm and G : V ⊂ Rm → Rk, where V is an open set containing F (U ). Then
(G ◦ F )∗ω = F∗(G∗ω) for all k-forms ω.
Proposition 4.5. Let F : U → Rm be a smooth map and ω be a k-form. Then d(F∗ω) = F∗dω.