LECTURE 5

2.6 Manifolds with boundary

We want to define rigorously a notion of “border”. Two simple model cases to keep in mind
are (1) the closed unit ball in Rn which is n-dimensional but whose border is Sn−1 which is
an (n − 1)-manifold; (2) the closed half-space of Rn which is n-dimensional but whose border
is Rn−1 . We start with the second example because it gives the local parametrisation to the
first example as well as to the general case.

Definition 40. The n-half space is Hn := {(x1 , . . . , xn ) ∈ Rn : xn ≥ 0} ⊂ Rn . We denote

its boundary ∂Hn = {(x1 , . . . , xn ) ∈ Rn : xn = 0} ⊂ Hn ⊂ Rn .

We now use this building block to define the general case by parametrisation:

Definition 41. Given k, n ∈ N∗ , a subset X ⊂ Rn is a k-manifold with boundary if each

x ∈ X has a neighbourhood diffeomorphic to an open set of Hk : there is U ⊂ X open around
x in X and U open of Hk and ϕ : U → U diffeomorphism (parametrisation). The boundary
∂X of X is the set of points that are in the image of ∂Hk under parametrisation. The interior
of X is defined as X \ ∂X.

Remark 42. Do not confuse the boundary and interior of X as defined above with the to-
pological notions of interior and boundary of X as a subset of Rn (for example think about
X = (0, 1] × R ⊂ R2 but a good exercise is to think about more subtle examples). Moreover,
be careful with the construction of manifolds by product: the product of two manifolds with
boundary is not necessarily smooth (think about the corners of [0, 1] × [0, 1]). In the example
sheet 1, it is proved that if X is a k-manifold with boundary then ∂X is a (k − 1)-manifold
without boundary, and X \ ∂X is a k-manifold without boundary. Another exercise is to
prove that when x ∈ ∂X, then all parametrisations ϕ around x are such ϕ−1 (x) ∈ ∂Hk .
To define the tangent space we need first to discuss extensions of continuous functions
from Hn to Rn :

Proposition 43. Consider ϕ on B(0, 1) ∩ Hn that is C k for k ∈ N ∪ {+∞}, then there is ϕ̃

on B(0, 1) ⊂ Rn that extends ϕ and is C k .

Proof (non-examinable). Denote x = (x̂, xn ) ∈ Rn . When k = 0, on can take ϕ̃(x̂, xn ) =

ϕ(x̂, −xn ) for xn < 0. For k = 1, one can take ϕ̃(x̂, xn ) = 4ϕ(x̂, −xn /2) − 3ϕ(x̂, −xn ) for
xn < 0. General k ∈ N can be obtained similarly by linear combinations. The case k =P+∞ is
obtained by constructing two sequences (aj ) and (bj ) so that bj < 0 and bj → −∞ and aj bm j
P
summable and equal to 1 for all m ∈ N. Then take ϕ̃(x̂, xn ) = j≥1 aj ζ(bj xn )ϕ(x̂, bj xn ) for
xn < 0, where ζ is a smooth function on R that is 1 on (−∞, 1/2] and 0 on [2, +∞).

The tangent space is then defined as follows:

Definition 44. Given X k-manifold with boundary and x ∈ X, the tangent space Tx X
is the image of the Euclidean space by the differential of a parametrisation ϕ around x, if
necessary by extending it to a ball if 0 = ϕ−1 (x) ∈ ∂Hk is on the boundary.

11
Remark 45. This means in particular that dim Tx X = k even for x ∈ ∂X in the boundary.
We now turn to the main “lego” procedure to construct manifolds with boundary.
Proposition 46. Given X a manifold without boundary and f : X → R smooth so that
0 ∈ R is a regular value and f (X) includes 0 and some positive values, then the subset
Z := {x ∈ X | f (x) ≥ 0} ⊂ X is a smooth submanifold with boundary. Moreover it has
dimension dim X and the boundary is given by f −1 (0).
Remark 47. This is consistent with the pre-image theorem, as we know f −1 (0) is a manifold
of dimension dim X − 1 since 0 is a regular value, and we know the boundary of a manifold
X with boundary is a manifold itself with dimension dim X − 1.
Proof. The set Z = {x ∈ X | f (x) > 0} is open in X by continuity, therefore it is a submani-
fold of X of dimension dim X, and every point has a local parametrisation by an open set of
Rk . The set {x ∈ X | f (x) = 0} is a submanifold of dimension k − 1 by the pre-image the-
orem since 0 is regular value of f . However the proof of the pre-image theorem gives us more:
given x0 ∈ X so that f (x0 ) = 0, the inverse of the local diffeomorphism F (y) = (f (y), T (y))
provides a local parametrisation of {x ∈ X | f (x) ≥ 0} around x0 by an open set of Hk , which
proves that Z is a submanifold with boundary of dimension k, with boundary f −1 (0).
Example 48. The closed unit ball Bn ⊂ Rn can be written {x ∈ Rn | f (x) ≥ 0} with
f : Rn → |x|2 − 1. Since Rn is a manifold without boundary and f is smooth we deduce that
Bn is a manifold with boundary of dimension n, and the boundary is given by f −1 (0) = Sn−1 .
We can also use the pre-image theorem on manifolds with boundary:
Theorem 49 (Pre-image theorem for manifolds with boundary). Given m > n in N∗ , X an
m-manifold with boundary and Y manifold an n-manifold without boundary, and f : X → Y
a smooth map, and y ∈ Y a regular value of f and of f|∂X . Then f −1 (y) is a smooth
(m − n)-manifold with boundary equal to f −1 (y) ∩ ∂X.
Proof. To check that the set f −1 (y) ⊂ X is a manifold, it is enough to check the definition
locally around each point x ∈ f −1 (y). Then by composition F = f ◦ ϕ of f with the inverse
of a parametrisation of X around this point, it is enough to check it when replacing f by F ,
X by Hm , and x ∈ X by x′ = ϕ−1 (x) ∈ Hm . When x′ ∈ Hm \ ∂Hm we can restrict F to
an open set of Rm and use the pre-image theorem to obtain a local parametrisation by an
open set of Rm−n . When x′ ∈ ∂Hm , we first extend F into F̃ smooth on U ∈ Rm around
x′ . Then since DF̃x′ = DFx′ is surjective (we assume that y is a regular value), by the pre-
image theorem F̃ −1 (y) is locally around x′ a manifold of dimension dim X − dim Y = m − n
in Rm . Then consider π : Rm → R defind by π(x1 , . . . , xm ) = xm . It is smooth and
F −1 (y) = π −1 ([0, +∞)) ∩ F̃ −1 (y) locally. If we prove that 0 is a regular value of π we can
apply the previous proposition to conclude that F −1 (y) is locally a submanifold of F̃ −1 (y)
with boundary π −1 (0) = F −1 (y) ∩ ∂Hn (and dim F −1 (y) = dim F̃ −1 (y) = m − n). Observe
that the tangent space of F −1 (y) at a point x′ ∈ F −1 (y) ∩ ∂Hn is equal to the kernel of
DFx′ : Rm → Rn . Since Dπ = π (linear function), to prove that Dπ : Tx′ F −1 (y) → R is
surjective, i.e. π : Ker(DFx′ ) → R is not identically zero, it is enough to prove that Ker(DFx′ )
is not included in ∂Hn : indeed otherwise by the rank theorem this image of (DFx′ )|π−1 (0)
would have dimension n − 1, and the point y would be a critical value for f|∂X . Hence 0 is a
regular value for π and the previous proposition allows to conclude.

12