Ma · Mathematics

Topological Boundary vs. Manifold Boundary

Published By the 9to5Science editorial team

At a glance

  1. 1. Fix the ambient space

    A topological boundary depends on the space in which a set is viewed.

  2. 2. Use closure and interior

    Compute the topological boundary as closure minus interior.

  3. 3. Use local charts

    A manifold boundary consists of points modeled on a Euclidean half-space.

  4. 4. Compare the definitions

    They agree in some full-dimensional cases, but differ for lower-dimensional manifolds.

Advertisement

The words topological boundary and manifold boundary describe related but different ideas. The topological boundary is defined using closure and interior in an ambient topological space. The manifold boundary is defined by the local coordinate model of a manifold.

They agree for familiar full-dimensional regions such as a closed disk in R2\mathbb{R}^2. They do not generally agree when the manifold has lower dimension than the space containing it.

Topological boundary

Let AA be a subset of a topological space XX. Its topological boundary, written ∂XA\partial_X A, is

∂XA=A‾ X∖int⁡X(A).\partial_X A=\overline{A}^{\,X}\setminus \operatorname{int}_X(A).

Equivalently, a point x∈Xx\in X belongs to the boundary of AA precisely when every open neighborhood of xx intersects both AA and X∖AX\setminus A.

The subscript XX matters. The same set can have different boundaries when considered inside different ambient spaces.

For example, let

A=[0,1].A=[0,1].

Viewed as a subset of R\mathbb{R}, its closure is [0,1][0,1] and its interior is (0,1)(0,1), so

∂R[0,1]={0,1}.\partial_{\mathbb{R}}[0,1]=\{0,1\}.

Viewed as a subset of itself, however, AA is both open and closed in the subspace topology. Therefore

∂[0,1][0,1]=∅.\partial_{[0,1]}[0,1]=\varnothing.

Thus “the boundary of AA” is incomplete unless the relevant ambient space is understood.

Manifold boundary

An nn-dimensional manifold with boundary is a space in which every point has a chart onto a relatively open subset of the closed half-space

Hn={(x1,…,xn)∈Rn:xn≥0}.\mathbb{H}^n=\{(x_1,\ldots,x_n)\in\mathbb{R}^n:x_n\ge 0\}.

A point is an interior point if its chart image lies in the interior of Hn\mathbb{H}^n; equivalently, it has a chart modeled on an open subset of Rn\mathbb{R}^n. It is a manifold-boundary point if its chart image lies in the boundary hyperplane xn=0x_n=0.

The manifold boundary is therefore

∂M={p∈M:a boundary chart sends p to xn=0}.\partial M=\{p\in M: \text{a boundary chart sends }p\text{ to }x_n=0\}.

This definition is intrinsic: it uses only the topology and local manifold structure of MM. It does not require MM to be sitting inside another space.

A standard theorem says that the classification is independent of the chosen chart. A point cannot be an interior point in one valid manifold chart and a boundary point in another.

A case where the boundaries agree

Consider the closed unit disk

D={(x,y)∈R2:x2+y2≤1}.D=\{(x,y)\in\mathbb{R}^2:x^2+y^2\le 1\}.

As a subset of R2\mathbb{R}^2,

int⁡R2(D)={x2+y2<1},\operatorname{int}_{\mathbb{R}^2}(D)=\{x^2+y^2<1\},

and DD is closed. Hence its topological boundary is the unit circle:

∂R2D={(x,y):x2+y2=1}.\partial_{\mathbb{R}^2}D=\{(x,y):x^2+y^2=1\}.

The disk is also a 2-dimensional manifold with boundary. Points satisfying x2+y2<1x^2+y^2<1 have neighborhoods modeled on R2\mathbb{R}^2, while points on the circle have neighborhoods modeled on a half-plane. Therefore

∂manifoldD={(x,y):x2+y2=1}.\partial_{\text{manifold}}D=\{(x,y):x^2+y^2=1\}.

In this full-dimensional example, the two notions produce the same set.

A case where they differ: an interval in the plane

Now consider the same interval embedded in R2\mathbb{R}^2:

M=[0,1]×{0}.M=[0,1]\times\{0\}.

As a 1-dimensional manifold with boundary, MM has manifold boundary

∂manifoldM={(0,0),(1,0)}.\partial_{\text{manifold}}M=\{(0,0),(1,0)\}.

The interior points (t,0)(t,0) with 0<t<10<t<1 have neighborhoods in MM homeomorphic to open intervals. The endpoints have neighborhoods homeomorphic to [0,ε)[0,\varepsilon).

But the topological boundary of MM as a subset of R2\mathbb{R}^2 is all of MM:

∂R2M=M.\partial_{\mathbb{R}^2}M=M.

Indeed, MM is closed in R2\mathbb{R}^2, so its closure is MM, and it has empty interior in R2\mathbb{R}^2 because it is only one-dimensional. Thus

∂R2M=M∖∅=M.\partial_{\mathbb{R}^2}M=M\setminus\varnothing=M.

The points in the open segment are not manifold-boundary points, even though they belong to the ambient topological boundary.

This example captures the main distinction: a lower-dimensional subset usually has empty ambient interior, so its ambient topological boundary includes every point of the subset.

Why the ambient space matters

There are three different boundaries one might accidentally confuse:

ObjectBoundary of [0,1][0,1] or MM
[0,1][0,1] in R\mathbb{R}{0,1}\{0,1\}
[0,1]×{0}[0,1]\times\{0\} in R2\mathbb{R}^2the entire segment
[0,1][0,1] in its own topologyempty set
[0,1][0,1] as a 1-manifold with boundaryits two endpoints

The first and fourth rows happen to give the same set. The second and fourth do not.

The manifold boundary is not simply “the topological boundary of the manifold in whatever larger space contains it.” An abstract manifold may not have been given any embedding at all, so such an ambient boundary would not even be defined.

When do they coincide?

Suppose MM is an nn-dimensional manifold with boundary embedded in an nn-dimensional ambient manifold XX without boundary. Identify MM with its image in XX. If the image is closed in XX and the embedding is locally flat, or smoothly embedded in a way that gives the expected local half-space model, then the ambient topological boundary is exactly the manifold boundary:

∂XM=∂manifoldM.\partial_X M=\partial_{\text{manifold}}M.

Indeed, near an interior point the image is open in XX, while near a manifold-boundary point it looks like a closed half-space. Closedness ensures that limit points outside the image do not add further points to the ambient boundary. The closed disk is the standard example.

If the ambient dimension is larger than the manifold dimension, this equality generally fails. For a standard embedded nn-manifold in an NN-manifold with N>nN>n, the embedded manifold has empty ambient interior. If its image is closed, its ambient topological boundary is the whole manifold, regardless of whether it has a manifold boundary. If its image is not closed, its ambient boundary is its closure, which may contain additional limit points outside the manifold.

There are also generalizations involving manifolds with corners, whose local model is

[0,∞)k×Rn−k.[0,\infty)^k\times\mathbb{R}^{n-k}.

In the standard convention, points with exactly one zero coordinate are ordinary codimension-one boundary points, while points with two or more zero coordinates are corners. All points with at least one zero coordinate belong to the boundary of the manifold with corners, but corner points carry a richer higher-depth stratum structure. Topologically, corners are still boundary points of the underlying manifold-with-boundary.

Common mistakes

Treating “boundary” as intrinsic in topology

Topological boundary is relative to an ambient space. Always write ∂XA\partial_XA when the choice of XX could be unclear.

Assuming every ambient boundary point is a manifold-boundary point

The open segment of an interval in R2\mathbb{R}^2 disproves this. It is in the ambient boundary but is an interior point of the 1-manifold.

Taking the boundary inside the manifold itself

If MM is viewed as a topological space, then MM is open in itself. Therefore the topological boundary of the whole space MM relative to MM is empty. This does not compute the manifold boundary.

Confusing an open set with a manifold having boundary

The open half-plane {(x,y):y>0}\{(x,y):y>0\} is a 2-manifold without boundary, even though its topological boundary in R2\mathbb{R}^2 is the line y=0y=0. The closed half-plane {(x,y):y≥0}\{(x,y):y\ge 0\} is a 2-manifold with boundary, and its manifold boundary is that line.

References

  1. John M. Lee, Introduction to Topological Manifolds
  2. John M. Lee, Introduction to Smooth Manifolds
  3. James R. Munkres, Topology

Spotted an error? Send a correction. Read how we write and check articles in our editorial policy.