A.

7 Vector fields

Let M be a smooth manifold. A vector field on M is a map X : M → T M such that X(p) ∈ Tp M

for p ∈ M . Sometimes, we also write Xp instead of X(p). As we have seen, T M carries a canonical
manifold structure, so it makes sense to call X is a smooth vector field if the map X : M → T M
is smooth. Hence, a smooth vector field on M is a smooth assignment of tangent vectors at the
various points of M . From another point of view, recall the natural projection π : T M → M ; the
requirement that X(p) ∈ Tp M for all p is equivalent to having π ◦ X = idM .
More generally, let f : M → N be a smooth mapping. Then a vector field along f is a map
X : M → T N such that X(p) ∈ Tf (p) N for p ∈ M . The most important case is that in which
f is a smooth curve γ : [a, b] → N . A vector field along γ is a map X : [a, b] → T N such that
X(t) ∈ Tγ(t) N for t ∈ [a, b]. A typical example is the tangent vector field γ̇.
Let X be a vector field on M . Given a smooth function f ∈ C ∞ (U ) where U is an open subset
of M , the directional derivative X(f ) : U → R is defined to be the function p ∈ U 7→ Xp (f ).
Further, if (x1 , . . . , xn ) is a coordinate system on U , we have already seen that { ∂x∂ 1 |p , . . . , ∂x∂ n |p }
is a basis of Tp M for p ∈ U . It follows that there are functions ai : U → R such that
n
X ∂
(A.7.1) X|U = ai .
∂xi
i=1

A.7.2 Proposition Let X be a vector field on M . Then the following assertions are equivalent:
a. X is smooth.
b. For every coordinate system (U, (x1 , . . . , xn )) of M , the functions ai defined by (A.7.1) are
smooth.
c. For every open set V of M and f ∈ C ∞ (V ), the function X(f ) ∈ C ∞ (V ).
Proof. Suppose X is smooth and let { ∂x∂ 1 |p , . . . , ∂x∂ n |p } be a coordinate system on U . Then X|U
is smooth and ai = dxi ◦ X|U is also smooth.
Next, assume (b) and let f ∈ C ∞ (V ). Take a coordinate system (U, (x1 , . . . , xn )) with V ⊂ U .
∂f
Then, by using (b) and the fact that ∂x i
is smooth,
n
X ∂f
X(f )|U = ai ∈ C ∞ (U ).
∂xi
i=1

Since V can be covered by such U , this proves (c).

Finally, assume (c). For every coordinate system (U, (x1 , . . . , xn )) of M , we have a corresponding
coordinate system (π −1 (U ), x1 ◦ π, . . . , xn ◦ π, dx1 , . . . , dxn ) of T M . Then

(xi ◦ π) ◦ X|U = xi and dxi ◦ X|U = X(xi )

are smooth. This proves that X is smooth. 

∂
In particular, the proposition shows that the coordinate vector fields associated to a local
∂xi
chart are smooth. The arguments in the proof also show that if X is a vector field on M satisfying
X(f ) = 0 for every f ∈ C ∞ (V ) and every open V ⊂ M , then X = 0. This remark forms the
basis of our next definition, and is explained by noting that in the local expression (A.7.1) for a
coordinate system defined on U ⊂ V , the functions ai = dxi ◦ X|U = X(xi ) = 0.
Next, let X and Y be smoth vector fields on M . Their Lie bracket [X, Y ] is defined to be the
unique vector field on M that satisfies

(A.7.3) [X, Y ](f ) = X(Y (f )) − Y (X(f ))

123
for every f ∈ C ∞ (M ). By the remark in the previous paragraph, such a vector field is unique if
it exists. In order to prove existence, consider a coordinate system (U, (x1 , . . . , xn )). Then we can
write
n n
X ∂ X ∂
X|U = ai and Y |U = bj
∂xi ∂xj
i=1 j=1

for ai , bj ∈ C ∞ (U ). If [X, Y ] exists, we must have

n  
X ∂bj ∂aj ∂
(A.7.4) [X, Y ]|U = ai − bi ,
∂xi ∂xi ∂xj
i=1

because the coefficients of [X, Y ]|U in the local frame { ∂x∂ j }nj=1 must be given by [X, Y ](xj ) =
X(Y (xj )) − Y (X(xj )). We can use formula A.7.4 as the definition of a vector field on U ; note
that such a vector field is smooth and satisfies property (A.7.3) for functions in C ∞ (U ). We finally
define [X, Y ] globally by covering M with domains of local charts: on the overlap of two charts,
the different definitions coming from the two charts must agree by the above uniqueness result; it
follows that [X, Y ] is well defined.

A.7.5 Proposition Let X, Y and Z be smooth vector fields on M . Then

a. [Y, X] = −[X, Y ].
b. If f , g ∈ C ∞ (M ), then

[f X, gY ] = f g[X, Y ] + f (Xg)Y − g(Y f )X.

c. [[X, Y ], Z] + [[Y, Z], X] + [[Z, X], Y ] = 0. (Jacobi identity)

We omit the proof of this propostion which is simple and only uses (A.7.3). Note that (A.7.3)
combined with the commutation of mixed second partial derivatives of a smooth function implies
∂
that [ ∂x , ∂ ] = 0 for coordinate vector fields associated to a local chart.
i ∂xj
Let f : M → N be a diffeomorphism. For every smooth vector field X on M , the formula
df ◦ X ◦ f −1 defines a smooth vector field on N which we denote by f∗ X. More generally, if
f : M → N is a smooth map which needs not be a diffeomorphism, smooth vector fields X on M
and Y on N are called f -related if df ◦ X = Y ◦ f . The proof of the next propostion is an easy
application of (A.7.3).

A.7.6 Proposition Let f : M → M ′ be smooth. Let X, Y be smooth vector fields on M , and let
X ′ , Y ′ be smooth vector fields on M ′ . If X and X ′ are f -related and Y and Y ′ are f -related, then
also [X, Y ] and [X ′ , Y ′ ] are f -related.

Flow of a vector field

Next, we discuss how to “integrate” vector fields. Let X be a smooth vector field on M . An integral
curve of X is a smooth curve γ in M such that

γ̇(t) = X(γ(t))

for all t in the domain of γ.

In order to study existence and uniqueness questions for integral curves, we consider local
coordinates. So suppose γ : (a, b) → M is a smooth curve in M , 0 ∈ (a, b), γ(0) = p, (U, ϕ =

124
 P ∂
(x1 , . . . , xn )) is a local chart around p, X is a smooth vector field in M and X|U = i ai ∂xi for
ai ∈ C ∞ (U ). Then γ is an integral curve of X on γ −1 (U ) if and only if

dγi 
(A.7.7)  = (ai ◦ ϕ−1 )(γ1 (t), . . . , γn (t))
dr t
for i = 1, . . . , n and t ∈ γ −1 (U ), where γi = xi ◦ γ. Equation (A.7.7) is a system of first order
ordinary differential equations for which existence and uniqueness theorems are known. These,
translated into manifold terminology yield the following proposition.

A.7.8 Proposition Let X be a smooth vector field on M . For each p ∈ M , there exists a (possibly
infinite) interval (a(p), b(p)) ⊂ R and a smooth curve γp : (a(p), b(p)) → M such that:
a. 0 ∈ (a(p), b(p)) and γp (0) = p.
b. γp is an integral curve of X.
c. γp is maximal in the sense that if µ : (c, d) → M is a smooth curve satisfying (a) and (b),
then (c, d) ⊂ (a(p), b(p)) and µ = γp |(c,d) .
Let X be a smooth vector field on M . Put

Dt = { p ∈ M | t ∈ (a(p), b(p)) }

and define Xt : Dt → M by setting

Xt (p) = γp (t).
Note that we have somehow reversed the rôles of p and t with this definition. The collection of Xt
for all t is called the flow of X.

A.7.9 Example Take M = R2 and X = ∂r∂1 . Then Dt = R2 for all t ∈ R and Xt (a1 , a2 ) =
(a1 + t, a2 ) for (a1 , a2 ) ∈ R2 . Note that if we replace R2 by the punctured plane R2 \ {(0, 0)}, the
sets Dt become proper subsets of M . ⋆

A.7.10 Theorem a. For each p ∈ M , there exists an open neighborhood V of p and ǫ > 0 such
that the map
(−ǫ, ǫ) × V → M, (t, p) 7→ Xt (p)
is well defined and smooth.
b. The domain dom(Xs ◦ Xt ) ⊂ Ds+t and Xs+t |dom(Xs ◦Xt ) = Xs ◦ Xt . Further, dom(Xs ◦ Xt ) =
Xs+t if st > 0.
c. Dt is open for all t, ∪t>0 Dt = M and Xt : Dt → D−t is a diffeomorphism with inverse X−t .
Proof. Part (a) is a local result and is simply the smooth dependence of solutions of ordinary
differential equations on the intial conditions. We prove part (b). First, we remark the obvious
fact that, if p ∈ Dt , then s 7→ γp (s + t) is an integral curve of X with initial condition γp (t) and
maximal domain (a(p) − t, b(p) − t); therefore (a(p) − t, b(p) − t) = (a(γp (t)), b(γp (t))). Next, let
p ∈ dom(Xs ◦ Xt ). This means that p ∈ dom(Xt ) = Dt and γp (t) = Xt (p) ∈ dom(Xs ) = Ds . Then
s ∈ (a(γp (t)), b(γp (t))), so s + t ∈ (a(γp (t)) + t, b(γp (t)) + t) = (a(p), b(p)), that is p ∈ Ds+t . Further,
Xs+t (p) = γp (s + t) = γγp (t) (s) = Xs (Xt (p)) and we have already proved the first two assertions.
Next, assume that s, t > 0 (the case s, t ≤ 0 is similar); we need to show that Ds+t ⊂ dom(Xs ◦Xt ).
But this follows from reversing the argument above as p ∈ Ds+t implies that s + t ∈ (a(p), b(p)),
and this implies that t ∈ (a(p), b(p)) and s = (s + t) − t ∈ (a(p) − t, b(p) − t) = (a(γp (t)), b(γp (t))).
Finally, we prove part (c). The statement about the union follows from part (a). Note that

125
D0 = M . Fix t > 0 and p ∈ Dt ; we prove that p is an interior point of Dt and Xt is smooth on a
neighborhood of p (the case t < 0 is analogous). Indeed, since γp ([0, t]) is compact, part (a) yields
an open neighborhood W0 of this set and ǫ > 0 such that (s, q) ∈ (−ǫ, ǫ) × W0 7→ Xs (q) ∈ M is
well defined and smooth. Take an integer n > 0 such that t/n < ǫ and put α1 = X t |W0 . Then,
n
define inductively Wi = α−1
i (Wi−1 ) ⊂ Wi−1 and αi = X t |Wi −1 for i = 2, . . . , n. It is clear that
n
αi is smooth and Wi is an open neighborhood of γp ( n−i
n t) for all i. In particular, Wn is an open
neighborhood of p in W . Moreover,

α1 ◦ α2 ◦ · · · ◦ αn |Wn = (X t )n |Wn = Xt |Wn

n

by the last assertion of part (b), so Xt is smooth on Wn . Now Dt is open and Xt is smooth on Dt .
It is obvious that the image of Xt is D−t . Since X−t is also smooth on D−t , it follows from part
(b) that Xt and X−t are inverses one of the other and this completes the proof of the theorem. 

The Frobenius theorem

Let M be a smooth manifold of dimension n. A distribution D of rank (or dimension) k is a choice
of k-dimensional subspace Dp ⊂ Tp M for each p ∈ M . A distribution D of rank k is called smooth
if for every p ∈ M there exists an open neighborhood U of p and k smooth vector fields X1 , . . . , Xk
on U such that Dq equals the span of X1 (q), . . . , Xk (q) for every q ∈ U . A vector field X is said
to belong to (or lie in) the distribution D (and we write X ∈ D) if X(p) ∈ Dp for p ∈ M . A
distribution D is called involutive if X, Y ∈ D implies that [X, Y ] ∈ D. A submanifold N of M is
called an integral manifold of a distribution D if Tp N = Dp for p ∈ N .
If X is a nowhere zero smooth vector field on M , then of course the line spanned by Xp in Tp M
for p ∈ M defines a smooth distribution on M . In this special case, Proposition A.7.8 guarantees
the existence and uniqueness of maximal integral submanifolds. Our next intent is to generalize this
result to arbitrary smooth distributions. A necessary condition is given in the following proposition.
The contents of the Frobenius theorem is that the condition is also sufficient.

A.7.11 Proposition A smooth distribution D on M admitting integral manifolds through any

point of M must be involutive.
Proof. Given smooth vector fields X, Y ∈ D and p ∈ M , we need to show that [X, Y ]p ∈ Dp .
By assumption, there exists an integral manifold N passing thorugh p. By shrinking N , we may
further assume that N is embedded. Denote by ι the inclusion of N into M . Then dιι−1 (p) :
Tι−1 (p) N → Tp M is an isomorphism onto Dp . Therefore there exist vector fields X̃ and Ỹ on N
which are ι-related to resp. X and Y . Due Theorem A.4.6, X̃ and Ỹ are smooth, so by using
Proposition A.7.6 we finally get that [X, Y ]p = dι([X̃, Ỹ ]ι−1 (p) ) ∈ Dp . 
It is convenient to use the following terminology in the statement of the Frobenius theorem.
A coordinate system (U, ϕ = (x1 , . . . , xn )) of a smooth manifold M of dimension m will be called
cubic if ϕ(U ) is an open cube centered at the origin of Rm , and it will be called centered at a point
p ∈ U if ϕ(p) = 0.

A.7.12 Theorem (Frobenius, local version) Let D be a smooth distribution of rank k on a

smooth manifold of dimension n. Suppose that D is involutive. Then, given p ∈ M , there exists
an integral manifold of D containing p. More precisely, there exists a cubic coordinate system
(U, ϕ = (x1 , . . . , xn )) centered at p such that the “slices”

xi = constant for i = k + 1, . . . , n

126
are integral manifolds of D. Further, if N ⊂ U is a connected integral manifold of D, then N is
an open submanifold of one of these slices.

Proof. We proceed by induction on k. Suppose first that k = 1. Choose a smooth vector field
X ∈ D defined on a neighborhood of p such that Xp 6= 0. It suffices to construct a coordinate
system (U, ϕ = (x1 , . . . , xn )) around p such that X|U = ∂x∂ 1 |U . Indeed, it is easy to get a coordinate
system (V, ψ = (y1 , . . . , yn )) centered at p such that ∂y∂ 1 = Yp . The map

σ(t, a2 , . . . , an ) = Xt (ψ −1 (0, a2 , . . . , an ))

is well defined and smooth on (−ǫ, ǫ) × W for some ǫ > 0 and some neighborhood W of the origin
in Rn−1 . We immediately see that
   
∂  ∂  ∂  ∂ 
dσ  = Xp =  and dσ  =  for i = 2, . . . , n.
∂r1 0 ∂y1 p ∂ri 0 ∂yi p

By the inverse function theorem, σ is a local diffeomorphism at 0, so its local inverse yields the
desired local chart ϕ.
We next assume the theorem is true for distributions of rank k − 1 and prove it for a given
distribution D of rank k. Choose smooth vector fields X1 , . . . , Xn spanning D on a neighborhood
Ṽ of p. The result in case k = 1 yields a coordinate system (V, y1 , . . . , yn ) centered at p such that
V ⊂ Ṽ and X1 |V = ∂y∂ 1 |V . Define the following smooth vector fields on V :

Y1 = X1
Yi = Xi − Xi (y1 )X1 for i = 2, . . . , k

Plainly, Y1 , . . . , Yk span D on V . Let S ⊂ V be the slice y1 = 0 and put

Zi = Yi |S for i = 2, . . . , k.

Since

(A.7.13) Yi (y1 ) = Xi (y1 ) − Xi (y1 ) X1 (y1 ) = 0 for i = 2, . . . , k,

| {z }
=1

we have Zi (q) ∈ Tq S for q ∈ S, so Z2 , . . . , Zk span a smooth distribution D ′ of rank k − 1 on S;

we next check that D ′ is involutive. Since Zi and Yi are related under the inclusion S ⊂ V , also
[Zi , Zj ] and [Yi , Yj ] are so related. Eqn. (A.7.13) gives that [Yi , Yj ](y1 ) = 0, so, on V

k
X
[Yi , Yj ] = cijk Yℓ
ℓ=1

for cijk ∈ C ∞ (V ). Hence

k
X
[Zi , Zj ] = cijk |S Zℓ ,
ℓ=1

as we wished. By the inductive hypothesis, there exists a coordinate system (w2 , . . . , wn ) on some
neighborhood of p in S such that the slices wi = constant for i = k + 1, . . . , n define integral
manifolds of D ′ .

127
 Let π : V → S be the linear projection relative to (y1 , . . . , yn ). Set

x1 = y1 ,
xi = wi ◦ π for i = 2, . . . , n.

It is clear that there exists an open neighborhood U of p in V such that (U, ϕ = (x1 , . . . , xn )) is a
cubic coordinate system of M centered at p. Now the first assertion in the statement of the theorem
follows if we prove that Yi (xj ) = 0 on U for i = 1, . . . , k and j = k + 1, . . . , n, for this will imply
that ∂x∂ 1 |q , . . . , ∂x∂ k |q is a basis of Dq for every q ∈ U . In order to do that, note that

∂xj 1, if j = 1,
=
∂y1 0, if j = 2, . . . , n

on U . Hence
n
∂ X ∂xj ∂ ∂
Y1 = X1 = = = ,
∂y1 ∂y1 ∂xj ∂x1
j=1

so Y1 (xj ) = 0 or j > k. Next, take i ≤ k and j > k. Owing to the involutivity of D,

k
X
[Y1 , Yi ] = ciℓ Yℓ
ℓ=1

for some cik ∈ C ∞ (U ). Therefore

k
∂ X
(Yi (xj )) = Y1 (Yi (xj )) − Yi (Y1 (xj )) = Y1 (Yi (xj )) = ciℓ Yℓ (xj ),
∂x1
ℓ=2

which, for fixed x2 , . . . , xn , is a system of k − 1 homogeneous linear ordinary differential equations

in the functions Yℓ (xj ) of the variable x1 . Of course, the initial condition x1 = 0 corresponds to
S ∩ U along which we have
Yi (xj ) = Zi (xj ) = Zi (wj ) = 0,
where the latter equatility follows from the fact that Zi lies in D ′ and wj = constant for j > k
define integral manifolds of D ′ . By the uniqueness theorem of solutions of ordinary differential
equations, Yi (xj ) = 0 on U .
Finally, suppose that N ⊂ U is a connected integral manifold of D. Let ι denote the inclusion of
N into U let and π : Rn → Rk ×Rn−k be the projection. Then d(π◦ϕ◦ι)q (Tq N ) = d(π◦π)q (Dq ) = 0
for q ∈ U . By connectedness of N , π ◦ ϕ ◦ ι is a constant function on N . Thus N is contained in
one of the slices xi = constant for i = k + 1, . . . , n, say S. The inclusion of N into M is continuous
(since N is a submanifold of M ) and has image contained in S; since S is embedded in M , the
inclusion of N into S is continuous and thus smooth by Theorem A.4.6. Since N is a submanifold
of M , this inclusion is also an immersion. Of course, dim N = dim S, so this inclusion is indeed a
local diffeomorphism. Hence N is an open submanifold of S. 

A.7.14 Theorem Integral manifolds of involutive distributions are quasi-embedded submanifolds.

More precisely, suppose that f : M → N is smooth, P is an integral manifold of an involutive
distribution D on M , and f (M ) ⊂ P . Consider the induced map f0 : M → P that satisfies
ι ◦ f0 = f , where ι : P → N is the inclusion. Then f0 is continuous (and hence smooth by
Theorem A.4.6).

128
 Proof. Let U be an open subset of P , q ∈ U and p ∈ f0−1 (q). It suffices to prove that p is an
interior point of f0−1 (U ). By the local version of the Frobenius theorem (A.7.12), there exists a
cubic coordinate system (V, ψ = (x1 , . . . , xn )) of N centered at q such that the slices

(A.7.15) xi = constant for i = k + 1, . . . , n

are the integral manifolds of D in V . Also, we can shrink V so that V ∩ U is exactly the slice

(A.7.16) xk+1 = · · · = xn = 0.

We have that f −1 (V ) an open neighborhood of q in M ; let W be its connected component containing

p. Of course, W is open. It is enough to show that f0 (W ) ⊂ V ∩ U , or which is the same, f (W )
is contained in (A.7.16). Since f (W ) is connected, it is contained in a component of V ∩ P with
respect to the relative topology. Since f (W ) meets (A.7.16) at least at the point q, it suffices
to show that the components of V ∩ P in the relative topology are contained in the slices of the
form (A.7.15). Let C be a component of V ∩ P with respect to the relative topology; note that C
need not be connected in the topology of P , but, by second-countability of P , C is a countable union
of connected integral manifolds of D in V , each of which is contained in a slice of the form (A.7.15).
Let π : V → Rn−k be given by π(r) = (xk+1 (r), . . . , xn (r)). It follows that π(C) is a countable
connected subset of Rn−k ; hence, it is a single point. 
A maximal integral manifold of a distribution D on a manifold M is a connected integral
manifold N of D such that every connected integral manifold of D which intersects N is an open
submanifold of N .

A.7.17 Theorem (Frobenius, global version) Let D be a smooth distribution on M . Suppose

that D is involutive. Then through any given point of M there passes a unique integral manifold
of D.

Proof. Let dim M = n and dim D = k. Given p ∈ M , define N to be the set of all points
of M reachable from p by following piecewise smooth curves whose smooth arcs are everywhere
tangent to D. By the local version A.7.12 and the σ-compactness of its topology, M can be covered
by countably many cubic coordinate systems (Ui , xi1 , . . . , xin ) such that the integral manifolds of D
in Ui are exactly the slices

(A.7.18) xij = constant for j = k + 1, . . . , n.

Note that a slice of the form (A.7.18) that meets N must be contained in N , and that N is covered
by such slices. We equip N with the finest topology with respect to which the inclusions of all
such slices are continuous. We can also put a differentiable structure on N by declaring that such
slices are open submanifolds of N . We claim that this turns N into a connected smooth manifold of
dimension k. N is clearly connected since it is path-connected by construction. N is also Hausdorff,
because M is Hausdorff and the inclusion of N into M is continuous. It only remains to prove that
N is second-countable. It suffices to prove that only countably many slices of Ui meet N . For this,
we need to show that a single slice S of Ui can only meet countably many slices of Uj . For this
purpose, note that S ∩ Uj is an open submanifold of S and therefore consists of at most countably
many components, each of which being a connected integral manifold of D and hence lying entirely
in a slice of Uj . Now it is clear that N is an integral manifold of D through p.
Next, let N ′ be another connected integral manifold of D meeting N at a point q. Given q ′ ∈ N ′ ,
there exists a piecewise smooth curve integral curve of D joining q to q ′ since connected manifolds

129
are path-connected. Due to q ∈ N , this curve can be juxtaposed to a piecewise smooth curve
integral curve of D joining p to q. We get q ′ ∈ N and this proves that N ′ ⊂ N . Since N ′ is a
submanifold of M , the inclusion of N ′ into M is continuous. By Theorem A.7.14, the inclusion of
N ′ into N is smooth. Hence N ′ is an open submanifold of N .
Finally, suppose that N ′ is in addition to the above a maximal integral manifold. The above
argument shows that N ⊂ N ′ and N is an open submanifold of N ′ . It follows that the identity
map N → N ′ is a diffeomorphism and this proves the uniqueness of N . 

130