The Gold Standard of Topology | Relative Boundaries and the Ambient Space Behind an Interior

Let A be a subset of a topological space X. A point x is interior to A in X if some open neighbourhood of x in X is contained in A. Write intX(A) for all such points. The neighbourhood is chosen using X’s topology.

In the usual real line, 1/2 is interior to (0,1): a sufficiently small open interval around it stays inside A. Zero is not even a member of A, so it cannot be an interior point of A.

The notation’s subscript is useful work, not decoration. “Open” does not mean “missing endpoints in a drawing” in every space. A neighbourhood in a subspace may end at a point that looks like an endpoint in the larger drawing.

Contents · Next chapter

A point is in clX(A) when every open neighbourhood of it in X meets A. A point is in the boundary ∂X A when every such neighbourhood meets both A and X∖A.

Equivalently, ∂X A = clX(A)∩clX(X∖A) = clX(A)∖intX(A). The second equality follows because a point fails to lie in the closure of the complement exactly when some neighbourhood avoids that complement.

The ambient complement X∖A matters as much as A. Alessandro Carlotto’s ETH Zürich notes on interior, closure and boundary present these equivalent tests. We will now keep the set A fixed while changing which neighbouring points are available.

Contents · Previous chapter · Next chapter

Take X = ℝ with the usual topology and A = (0,1). Every point strictly between zero and one has a small interval inside A, so intℝ(A) = (0,1).

Every neighbourhood of zero or one meets A, while a point strictly outside [0,1] has a neighbourhood missing A. Thus clℝ(A) = [0,1]. The difference between closure and interior is {0,1}, the boundary.

At zero, for example, a small neighbourhood meets A to its right and the complement to its left and at zero itself. Boundary membership does not require membership in A: both boundary points here lie outside A.

The two endpoints were found by the neighbourhood test. Their appearance in a picture suggests the answer, but the test explains why the answer holds.

Contents · Previous chapter · Next chapter

Now put the same A inside Y = (0,1] with the subspace topology inherited from ℝ. Its relative interior is still A. Its closure in Y is all Y, and its boundary in Y is {1}. Zero is no longer an available point of the space.

Next let Y = A itself. Every point of Y has the neighbourhood Y contained in A. Hence intY(A) = Y and clY(A) = Y. The relative complement Y∖A is empty, so the boundary is empty.

With Y = [0,1], both endpoints remain available and the boundary is {0,1}. Changing only the surrounding space can remove one boundary point, remove both, or retain both.

Contents · Previous chapter · Next chapter

For A ⊆ Y ⊆ X with the subspace topology on Y, the general formula is clY(A) = Y∩clX(A). Take y in Y. Its basic relative neighbourhoods have the form Y∩U, where U is open in X and contains y.

Because A is contained in Y, (Y∩U)∩A = U∩A. Every relative neighbourhood meets A exactly when every ambient neighbourhood meets A. Restricting the ambient closure to Y therefore gives the relative closure.

This proof uses no metric, no sequence argument and no assumption that Y is open or closed. Allen Hatcher’s point-set topology notes, the subspaces section provide the wider framework.

The equality describes closure only. It is tempting to copy it for every related operation. The next step shows why interior and boundary require more care.

Contents · Previous chapter · Next chapter

Every point interior to A in X is interior to A in Y, because its ambient open neighbourhood can be intersected with Y. Thus intX(A) ⊆ intY(A). The inclusion can be strict.

Use A = Y = [0,1] inside X = ℝ. The ambient interior is (0,1), while the relative interior is [0,1]. Near zero, a relative neighbourhood can be [0,ε), which is open in Y even though it is not open in ℝ.

In general, intY(A) = Y∖clY(Y∖A) = Y∖clX(Y∖A). This follows by applying the complement test within Y and then using the closure formula. It retains Y∖A, rather than replacing it by X∖A.

If Y is open in X, the interiors agree for every A ⊆ Y. A relative open neighbourhood is then an intersection of two ambient open sets and is itself open in X. Without that condition, equality needs a separate check.

Contents · Previous chapter · Next chapter

Apply the closure-intersection definition within Y: ∂Y A = clY(A)∩clY(Y∖A). Substituting the closure formula gives ∂Y A = Y∩clX(A)∩clX(Y∖A).

Since Y∖A ⊆ X∖A, monotonicity of closure yields ∂Y A ⊆ Y∩∂X A. Every relative boundary point is an ambient boundary point retained in Y. Some retained ambient boundary points can become relative interior points, so equality need not hold.

The counterexample A = Y = [0,1] makes this explicit. Here ∂Y A is empty, while Y∩∂ℝ A = {0,1}. Merely intersecting the old boundary with the new space gives the wrong result.

Use the complement in the space where boundary is being tested. That one substitution prevents the false equality.

Contents · Previous chapter · Next chapter

If Y is open in X, then ∂Y A = Y∩∂X A for every A ⊆ Y. We already proved one inclusion. For the other, take y in Y∩∂X A and an ambient neighbourhood U of y.

The intersection U∩Y is an open neighbourhood in X because Y is open. Ambient boundary membership says it meets A and X∖A. Its second intersection actually lies in Y∖A. Therefore every relative neighbourhood meets both sides, proving relative boundary membership.

For an example, let Y = (−1,2) and A = [0,1]. Both old boundary points remain, and ∂Y A = {0,1}.

Openness is a sufficient condition for equality across all subsets. Equality can also happen for particular A when Y is not open, as our earlier A = (0,1), Y = [0,1] example shows. Do not turn a sufficient hypothesis into a necessary one without proof.

Contents · Previous chapter · Next chapter

Let A be the horizontal segment {(x,0): 0 ≤ x ≤ 1} inside X = ℝ². Every open disk about a point of A contains points off the segment. No disk lies inside A, so its ambient interior is empty.

The segment is closed in the plane. Consequently its boundary as a subset of ℝ² is the entire segment. Every disk at each segment point meets A and the plane outside A.

Now take Y to be the whole horizontal axis with its subspace topology. The relative interior is {(x,0): 0 < x < 1}, and the relative boundary consists of (0,0) and (1,0).

Finally regard A as the whole space. Its boundary as a subset of itself is empty. This last statement concerns subset boundary; it must not be confused with the different definition of the boundary of a manifold with boundary.

Contents · Previous chapter · Next chapter

First, a boundary need not be a thin outline. In ℝ, let A = ℚ. Every real interval contains rational and irrational points. Thus clℝ(A) = ℝ, intℝ(A) is empty and ∂ℝ A = ℝ. This boundary has nonempty interior.

Inside Y = ℚ, the same set A is the whole space and its relative boundary is empty. The rational numbers did not change; the irrational complement ceased to be part of the space.

Second, being closed does not mean being only boundary. The closed interval [0,1] in ℝ contains its interior as well as its endpoints. For a closed A, ∂X A = A∖intX(A), which can be smaller than A.

Michael Müger’s Topology for the Working Mathematician, Sections 2.6–2.7 develops the subspace and density questions behind these tests. State the topology before trusting the outline suggested by a sketch.

Contents · Previous chapter · Next chapter

Let A = [0,1) and X = ℝ. Find its interior, closure and boundary, then repeat inside Y = [0,2). Work from neighbourhoods before comparing with the answer.

In ℝ, the interior is (0,1), closure [0,1] and boundary {0,1}. Inside Y, the point zero becomes interior: [0,ε) is a relative open neighbourhood contained in A. Hence intY(A) = [0,1), clY(A) = [0,1] and ∂Y A = {1}.

The closure formula still works: Y∩[0,1] = [0,1]. The false boundary shortcut gives {0,1}, keeping a point that has become interior. Explain the failure by identifying the actual relative complement Y∖A = [1,2).

For a second check, set Y = [0,1). A is now the whole space, so its closure and interior are both A and its relative boundary is empty.

Contents · Previous chapter · Next chapter

Is a point always either interior or boundary?

Relative to A in X it can also be exterior: interior to X∖A. Interior, boundary and exterior partition X. A point outside the space being discussed has no membership in that partition.

Do these proofs need a distance?

No. The subspace arguments use open neighbourhoods and set operations. Metric balls are convenient examples, not a hidden requirement of the formulas.

Is a boundary always empty in its own space?

Every space has empty boundary as a subset of itself. That does not say every subset has empty boundary, or redefine the separate notion of manifold boundary.

Sources checked 22 October 2026: the linked ETH Zürich notes, Hatcher’s subspace discussion and Müger’s topology text support the definitions and general framework. The interval ledger, planar comparison and transfer solutions are explicit worked teaching examples.

Choose the next question

Closure, neighbourhoods and dense sets

• Closure of Intersections and Dense Sets
• Derived Sets, Isolated Points and the Limits Left After Removal
• When Sequences Miss Closure and Continuity

Maps and altered spaces

• Quotient Seams and Why Saturated Sets Matter
• A Continuous Bijection with a Discontinuous Inverse
• Dense Equivalence Classes and Separation in a Quotient Space

Compactness and subspace reasoning

• Compact Sets, Closedness and the Missing Hausdorff Hypothesis
• Closed Projections and the Compact Factor That Keeps Limits from Escaping
• Sequential Compactness and Compactness in Metric Spaces
• Topology

Return to the Topology subject root · The Gold Standard of Education: learning transfers into life · Explore How X Works.

Contents · Previous chapter · Return to Topology