See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/267222794

An Introduction to Distributions and Foliations

Conference Paper · December 2007

CITATIONS READS

2 2,092

1 author:

Samuel Otten
University of Missouri
69 PUBLICATIONS 587 CITATIONS

SEE PROFILE

All content following this page was uploaded by Samuel Otten on 09 December 2014.

The user has requested enhancement of the downloaded file.

 An Introduction to Distributions and Foliations

Samuel Otten
Michigan State University
MTH 864 · Spring 2008

In smooth manifold theory, the notion of a tangent space makes it possi-

ble for differentiation to take place on an abstract manifold. In this paper,
the notion of a distribution will be presented which makes it possible for in-
tegration to take place on an abstract manifold. The first section introduces
terminology and builds intuition via an analogy to the concept of integral
curves. The second section presents the Frobenius theorem–one of the foun-
dational results in smooth manifold theory. The concluding section leaves the
reader with foliations and a brief look at their connection to the Frobenius
theorem.

1 Preliminaries
Let M be an m-dimensional manifold, Tp M the tangent space to p ∈ M , and
T M the tangent bundle of M . Throughout this work, things are implicitly
smooth. Recall that a vector field on M is a section of T M and can be thought
of as a choice of tangent vector at every point in the manifold. Given a vector
field V , one can consider an associated integral curve.

Definition. An integral curve of a vector field V is a curve γ : (a, b) → M

(a and b real numbers) such that

γ ′ (t) = Vγ(t) for all t ∈ (a, b).

1
 That is, the derivative at any point of the curve with respect to time is
precisely the value of the vector field at that point (see figure 1). In this way,

Figure 1: An integral curve of a vector field.

an integral curve can be intuitively conceived of as the path a particle would

trace if it were to flow along V . Note also that if V is non-vanishing, then its
integral curves are connected, immersed 1-dimensional submanifold of M .
Broadly speaking, the work that follows is aimed toward generalizing this
idea to connected, immersed k-dimensional submanifolds. So let us begin by
taking the concept of vector field and increasing its dimension.

Definition. A k-dimensional (tangent) distribution on M is a choice of k-

dimensional linear subspace Dp ⊂ Tp M at each point p ∈ M . We will denote
this by D, where a
D= Dp ⊂ T M.
p∈M

If D is a k-dimensional distribution, then we can find a vector field V 1

such that Vp1 ∈ Dp for all p in some neighborhood U ⊂ M . We can continue
(possibly shrinking the neighborhood) until we have vector fields V 1 , . . . , V k
such that Vp1 , . . . , Vpk form a basis for Dp at each p ∈ U .
Now, let D be a k-dimensional distribution and consider a k-dimensional
immersed submanifold S ⊂ M . Then for s ∈ S, Ts S is a k-dimensional linear
subspace of Ts M and natural questions arise: Does Ts S correspond to Ds ?

2
Does there exist an immersed submanifold such that Ts S = Ds for all s ∈ S?
Such an S would be analogous to the integral curves above, and therefore we
have the following definitions.

Definition. An immersed submanifold S is an integral manifold of the dis-

tribution D if Ts S = Ds for all s ∈ S, and D is integrable if each point of
M is contained in an integral manifold of D.

To answer the second question above, there are distributions for which no
integral manifolds exist. The reason for this is that the basis vector fields of
a distribution may not interact in a nice way. To be more precise about the
word “nice” let us look at the most common way in which two vector fields
interact to form a new vector field.

Definition. The Lie bracket of the vector fields V and W is [V, W ] = V W −

WV .

It is well known that [V, W ] itself defines a (natural) vector field. What
we will find below is that “nice” means given any vector fields V , W with
Vp , Wp ∈ Dp for all p in some neighborhood U , it follows that [Vp , Wp ] ∈ Dp .
If this condition is satisfied (it is only necessary to check it on the basis vector
fields), then the distribution D is called involutive.

Lemma. If D is an integrable distribution, then D is necessarily involutive.

Proof. Let V and W be local sections of D ⊂ T M defined on some U ⊂ M

and let p ∈ U . Since D is integrable, we have some S an integral manifold
of D containing p. Since V and W are sections of D, we know that V and
W are tangent to S on U . By properties of the naturality of the Lie bracket,
[V, W ] is also tangent to S and so [V, W ]p ∈ Dp . This holds for all points in
U and so D is involutive.

As an example, consider the 2-dimensional distribution D on R3 with

basis vector fields
∂ ∂ ∂
V = +y , W = .
∂x ∂z ∂y

3
We have [V, W ] = V (1) ∂/∂y − W (1) ∂/∂x − W (y) ∂/∂z = −∂/∂z which is
clearly not a linear combination of V and W and so is not a section of D.
Thus, by the preceding lemma, D is not integrable. Indeed, consider the
origin in R3 and suppose to the contrary that an integral manifold S exists
containing the origin. Then an integral curve of V or W must stay in S near
the origin for at least a short period of time because V and W are tangent to
S by definition. Since the x-axis is an integral curve of V , S must contain an
open subset of it. Since lines parallel to the y-axis are integral curves of W , S
must also contain (for some small a) an open subset {(a, y, 0) | −ε < y < ε}.
Taking these together, we see that S must contain an open subset of the
(x, y)-plane near the origin. However, Dp is not equal to the (x, y)-plane for
any point off the x-axis, a contradiction.
So we see that involutivity is a useful concept as it is a necessary condition
for distributions to be integrable.

2 The Frobenius Theorem

Let U ⊂ M be an open neighborhood with parametrization φ : U → Rm . If
D is a k-dimensional distribution, then what is the best possible situation
we can hope for from the integral manifolds? First of all, we can hope for
the existence of integrable manifolds; i.e., that D is integrable. More than
that, though, we can hope for a simple product structure on the integral
manifolds. This brings us to our next definitions.

Definition. A parametrization φ is flat for D if φ(U ) ⊂ Rm is a product

of connected open sets in Rk × Rm−k and for each p ∈ U , DP is spanned
by precisely the first k basis vector fields. A distribution D is completely
integrable if there exists a flat parametrization for D in a neighborhood of
every point of M .

So at the intuitive level, we have a k-dimensional subspace for every point

of M . These vary smoothly, but need not be entirely neat and tidy. A flat

4
parametrization takes each Dp for p ∈ U and pushes it forward in such a way
that its image is the span of ∂/∂x1 , . . . , ∂/∂xk (see figure 2).

xk+1 ,...,xm

x2 ,...,xk

x1

Figure 2: A flat parametrization of a distribution.

Since being completely integrable requires an added condition on the inte-

gral manifolds of a distribution, it is clear that complete integrability implies
integrability. Furthermore, we saw in the lemma above that integrability
implies involutivity. The result below implies that these implications are
bidirectional.

Theorem. [Frobenius] If a distribution D is involutive, then D is com-

pletely integrable.

The hypothesis gives us a basis V 1 , . . . , V k of vector fields for D such that

[V i , V j ]p ∈ Dp for all i, j and all p ∈ U . Intuitively, the “commutator” of V i
and V j locally remains within the distribution. The idea behind the proof
of this theorem is to flow (briefly) along each vector field one at a time and
get from each a new coordinate for a parametrization. The involutivity just
discussed guarantees that nothing will go terribly wrong during this process,
and at the end of it all you will have a flat parametrization.
The Frobenius theorem comes in other forms. The original version as
presented by Frobenius regarded Pfaffian systems (whatever those may be).

5
It can also be stated in terms of differential forms rather than with vector
fields as we’ve done here. Another version goes like this: A distribution is
involutive (and thus completely integrable) if and only if it corresponds to
the tangent spaces of the leaves of a foliation of M . This leads us to our next
section.

3 Foliations
One can think of a foliation as an equivalence relation on an m-manifold M
in which the equivalence classes are connected, immersed submanifolds of a
common dimension k. Locally, the equivalence classes should be analogous
to the “leaves” of Rk which make up Rm . We shall be more precise before
turning to some examples.
Definition. A k-dimensional foliation on an m-manifold M is a collection of
disjoint, connected, immersed k-dimensional submanifolds of M (the leaves
of the foliation) such that (i) the union of the leaves is all of M , and (ii)
there is a parametrization φ around each p ∈ U ⊂ M such that φ(U ) is
a product of connected open sets in Rk × Rm−k and each leaf intersects U
in the empty set or a countable union of k-dimensional slices of the form
xk+1 = ck+1 , . . . , xm = cm .
Let us consider some examples of foliations.
1. As alluded to above, Rn can be foliated by copies of Rk in the obvious
way.

2. The collection of all spheres in Rn centered at the origin forms an

(n − 1)-dimensional foliation.

3. The solid torus (D2 × S 1 ) can be foliated as in figure 3, which is known

as the Reed foliation.

4. Let f : M m → N n be a submersion. For each x ∈ M , the submersion

theorem gives a set of coordinates (x1 , . . . , xm ) on a neighborhood U ⊂

6
 Figure 3: A foliation of a solid torus.

M as well as coordinates (y1 , . . . , yn ) around f (x) = y ∈ V ⊂ N

such that f (x1 , . . . , xm ) = (x1 , . . . , xn ) locally. Thus the level sets of
f are embedded submanifolds of M of dimension m − n, and so the
connected components of the nonempty level sets are the leaves of an
(m − n)-dimensional foliation of M . The point is that submersions give
us foliations.

These examples are pleasant, but what we would really like to see is
the foliation-version of the Frobenius theorem. We approach each direction
separately.

Lemma. If F is a k-dimensional foliation of M , then the collection of tan-

gent spaces to the leaves of F form an involutive distribution.

Proof. The tangent spaces to the leaves clearly give a k-dimensional distribu-
tion on M because for each point we have identified a k-dimensional subspace

7
of the tangent space at that point. We must verify that this distribution is in-
volutive. This follows easily, however, because we can see through the leaves
that the distribution is integrable and we proved earlier that integrability
implies involutivity.

Theorem. [Alternate Frobenius] If D is an involutive distribution on M ,

then the collection of all maximal connected integral manifolds of D forms a
foliation of M .

A complete proof of this is beyond the scope of this paper (and its author).
However, if we enlist the help of the following lemma (without proof) then
we can prove the theorem to a reasonable degree.

Lemma. If D is an involutive distribution and {Nα } is a collection of integral

S
manifolds of D with a point in common, then N = Nα is a connected
integral manifold of D in which each Nα is an open submanifold.

Let p ∈ M and let Lp be the union of all connected integral manifolds of

D containing p. By the lemma, Lp itself is a connected integral manifold con-
taining p and is maximal by construction. Let Lq be another such connected
integral manifold. Either Lp and Lq are disjoint or they intersect, and by
maximality, their intersection implies Lp = Lq . So these integral manifolds
as constructed satisfy the intuitive notion of equivalence classes.
Since D is involutive we have a flat parametrization φ on a neighborhood
U around p. By the definition of flat parametrization, an integral manifold
intersected with U comprises a countable union of subsets of slices because
the basis vector fields are represented precisely in the first k coordinates. In
particular, Lp ∩ U is a collection of such slices. Let S be a slice and suppose
that Lp ∩ U is some proper subset of S. Then Lp ∪ S contains Lp and is
a connected integral manifold of D. This contradicts the maximality of Lp ,
and so Lp ∩ U is either empty or all of S and so is a countable union of slices.
Therefore {Lp | p ∈ M } is a foliation.

8
 References
[1] A. Candel, and L. Conlon, Foliations I, American Mathematical Society,
Providence, RI, 2000.

[2] V. Guillemin, and A. Pollack, Differential Topology, Prentice-Hall, Inc.,

Englewood Cliffs, NJ, 1974.

[3] J.M. Lee, Introduction to Smooth Manifolds, Springer, New York, 2003.

View publication stats