Closure, Interior, and Boundary

What does it mean for a point to be "on the edge" of a set? Consider the open disk D={(x,y):x2+y2<1}D = \{(x,y) : x^2+y^2 < 1\} in the plane. Every point of the unit circle feels like an edge point of DD. This is true even though none of them belong to DD itself. Making this intuition precise is one of the founding achievements of point-set topology. This must be done without any reference to distance, only to open sets. It produces three operators that pervade the whole subject: the interior, the closure, and the boundary of a set. These operators matter because they translate vague geometric language into precise operations. Phrases like "inside", "on the edge", "arbitrarily close to" become operations defined purely from the axioms. These axioms come from a topological space. Once defined, they satisfy clean algebraic laws: idempotence, monotonicity, duality. These laws let us reason about sets the way we reason about numbers. They also let us detect properties like openness and closedness by simple equalities. A set is open exactly when it equals its own interior. It is closed exactly when it equals its own closure. The big picture is that closure and interior are dual operations. They are linked by complementation. The boundary measures the discrepancy between a set and its closure/interior. This trio gives the first real computational toolkit for a topological space. This is possible well before one has metrics, sequences, or continuity in hand. It also sets up the vocabulary needed for Dense Sets and Separability. A set is dense iff its closure is everything. It also prepares the vocabulary for Limit Points and Sequences, since closure adds exactly the limit points. This article gives the formal definitions. It proves the fundamental structural theorem describing how interior, closure, and boundary interact. It works two detailed non-trivial examples. One example is in R\mathbb{R} with the standard topology. The other is in a non-Hausdorff finite space. It shows exactly why each hypothesis in the theorem is needed. It closes with a map of where these operators recur across mathematics.

Key takeaways
  • The interior int⁡(A)\operatorname{int}(A) is the largest open set contained in AA; the closure A‾\overline{A} is the smallest closed set containing AA.
  • Both exist and are unique in any topological space, by taking a union (resp. intersection) over a suitable family of sets.
  • Closure and interior are dual under complementation: A‾=X∖int⁡(X∖A)\overline{A} = X \setminus \operatorname{int}(X\setminus A).
  • The boundary ∂A=A‾∖int⁡(A)\partial A = \overline{A} \setminus \operatorname{int}(A) measures exactly the points that are limits of both AA and its complement.
  • AA is open iff A=int⁡(A)A = \operatorname{int}(A); AA is closed iff A=A‾A = \overline{A}; both operators are idempotent and monotone.
  • Key idea: closure and interior are the extremal (smallest/largest) closed/open approximations to a set, and every set decomposes as int⁡(A)⊔∂A⊔int⁡(X∖A)\operatorname{int}(A) \sqcup \partial A \sqcup \operatorname{int}(X\setminus A).

Where this lives today

Closure operators recur throughout mathematics as soon as one has any notion of "approximation" or "limiting process." In real and complex analysis, the closure of the set of polynomials in an appropriate function-space topology underlies the Stone–Weierstrass approximation theorem and questions of which functions can be uniformly approximated. In complex analysis, the boundary of a domain (e.g., the unit disk) is exactly the locus where boundary-value problems, Cauchy integral formulas, and conformal mapping theorems (Ahlfors) are stated. In algebraic geometry, the Zariski closure of a set of points is the smallest algebraic variety containing them, and Zariski-closed sets play the role of closed sets in an entirely different, non-metric topology, showing the abstract framework's full generality. In general topology itself, the closure operator is the starting axiomatic object in Kuratowski's approach, from which open sets are recovered as complements of fixed points of the closure operator — an alternate, equivalent foundation to the open-set axioms. In computer science, closure operators appear in formal concept analysis, in the semantics of program analysis (closure under a set of inference rules), and in database theory (closure of a set of functional dependencies). In topology proper, these operators are the direct staircase to Limit Points and Sequences (a point is a limit point of AA iff every deleted neighborhood meets AA, closely tied to A‾\overline A), to the notion of a dense set used in Dense Sets and Separability, and to the separation axioms of Hausdorff and Separation Axioms, where the failure of "nice" boundary behavior in Example 2 above is exactly repaired by assuming points can be separated by disjoint open sets.

Ask the tutor: Can you show me how the closure operator's Kuratowski axioms alone recover the open sets of a topology?

DefinitionInterior, Closure, and Boundary

Let (X,τ)(X,\tau) be a topological space and let A⊆XA \subseteq X.

Interior. The interior of AA, written int⁡(A)\operatorname{int}(A) or A∘A^{\circ}, is the union of all open sets contained in AA:

int⁡(A)=⋃{U∈τ:U⊆A}.\operatorname{int}(A) = \bigcup \{ U \in \tau : U \subseteq A \}.

Closure. The closure of AA, written A‾\overline{A} or cl⁡(A)\operatorname{cl}(A), is the intersection of all closed sets containing AA:

A‾=⋂{C⊆X:C closed, A⊆C}.\overline{A} = \bigcap \{ C \subseteq X : C \text{ closed}, \ A \subseteq C \}.

Boundary. The boundary (or frontier) of AA, written ∂A\partial A, is

∂A=A‾∖int⁡(A)=A‾∩X∖A‾.\partial A = \overline{A} \setminus \operatorname{int}(A) = \overline{A} \cap \overline{X \setminus A}.

A point xx is called an interior point of AA if x∈int⁡(A)x \in \operatorname{int}(A), equivalently if some open set UU satisfies x∈U⊆Ax \in U \subseteq A. A point xx is called a closure point (or adherent point) of AA if x∈A‾x \in \overline{A}, equivalently if every open set UU containing xx satisfies U∩A≠∅U \cap A \neq \varnothing. A point xx is a boundary point of AA if every open set containing xx meets both AA and X∖AX \setminus A.

Intuition

Think of τ\tau as the collection of "roomy" neighborhoods available in XX. The interior of AA collects the points that have an entire roomy neighborhood sitting fully inside AA — points that are safely surrounded by AA on all sides. The closure of AA collects points that cannot be separated from AA by any roomy neighborhood — points you can't push away from AA no matter how you try, because every neighborhood you draw still bumps into AA. The boundary is what's left over: points that are adherent to AA but not safely inside it, which forces every neighborhood of such a point to also poke into the complement of AA. A useful mental picture is the "sandwich": int⁡(A)⊆A⊆A‾\operatorname{int}(A) \subseteq A \subseteq \overline{A} always holds, and equality on the left means AA is open, equality on the right means AA is closed. The boundary is the region where this sandwich has genuine bread on both sides — points caught between AA and its complement no matter how finely you zoom in (topologically, no matter which open set you use to zoom).

Figure
The half-open interval A=[0,1)∪{2}A=[0,1)\cup\{2\} in R\mathbb{R}: interior, closure, boundary
0 1 2 A = [0,1) ∪ {2} int(A) = (0,1) closure(A) = [0,1] ∪ {2} ∂A = {0, 1, 2}

Duality between closure and interior

The single most important structural fact, proved below, is that closure and interior determine one another via complementation. This duality is why every fact about interiors has a mirror-image fact about closures, obtained by swapping "open" ↔\leftrightarrow "closed", "union" ↔\leftrightarrow "intersection", and "AA" ↔\leftrightarrow "X∖AX \setminus A".

TheoremFundamental Properties of Interior, Closure, and Boundary

Let (X,τ)(X,\tau) be a topological space and let A,B⊆XA, B \subseteq X. Then: 1. (Extremal characterization) int⁡(A)\operatorname{int}(A) is open, int⁡(A)⊆A\operatorname{int}(A) \subseteq A, and int⁡(A)\operatorname{int}(A) is the largest open subset of AA: if UU is open and U⊆AU \subseteq A, then U⊆int⁡(A)U \subseteq \operatorname{int}(A). 2. (Extremal characterization, dual) A‾\overline{A} is closed, A⊆A‾A \subseteq \overline{A}, and A‾\overline{A} is the smallest closed superset of AA: if CC is closed and A⊆CA \subseteq C, then A‾⊆C\overline{A} \subseteq C. 3. (Point characterizations)

int⁡(A)={x∈X:∃ U∈τ, x∈U⊆A},A‾={x∈X:∀ U∈τ with x∈U, U∩A≠∅}.\operatorname{int}(A) = \{x \in X : \exists \, U \in \tau,\ x \in U \subseteq A\}, \qquad \overline{A} = \{x \in X : \forall\, U \in \tau \text{ with } x \in U,\ U \cap A \neq \varnothing\}.

4. (Duality) X∖A‾=int⁡(X∖A)X \setminus \overline{A} = \operatorname{int}(X \setminus A) and X∖int⁡(A)=X∖A‾X \setminus \operatorname{int}(A) = \overline{X \setminus A}. 5. (Idempotence) int⁡(int⁡(A))=int⁡(A)\operatorname{int}(\operatorname{int}(A)) = \operatorname{int}(A) and A‾‾=A‾\overline{\overline{A}} = \overline{A}. 6. (Monotonicity) If A⊆BA \subseteq B then int⁡(A)⊆int⁡(B)\operatorname{int}(A) \subseteq \operatorname{int}(B) and A‾⊆B‾\overline{A} \subseteq \overline{B}. 7. (Openness / closedness criteria) AA is open iff A=int⁡(A)A = \operatorname{int}(A). AA is closed iff A=A‾A = \overline{A}. 8. (Interaction with union/intersection)

int⁡(A)∪int⁡(B)⊆int⁡(A∪B),A∩B‾⊆A‾∩B‾,\operatorname{int}(A) \cup \operatorname{int}(B) \subseteq \operatorname{int}(A \cup B), \qquad \overline{A \cap B} \subseteq \overline{A} \cap \overline{B},
int⁡(A∩B)=int⁡(A)∩int⁡(B),A∪B‾=A‾∪B‾.\operatorname{int}(A \cap B) = \operatorname{int}(A) \cap \operatorname{int}(B), \qquad \overline{A \cup B} = \overline{A} \cup \overline{B}.

9. (Boundary facts) ∂A\partial A is closed, ∂A=∂(X∖A)\partial A = \partial(X \setminus A), and

X=int⁡(A)⊔∂A⊔int⁡(X∖A)(disjoint union).X = \operatorname{int}(A) \sqcup \partial A \sqcup \operatorname{int}(X \setminus A) \quad \text{(disjoint union)}.

Moreover AA is open iff ∂A∩A=∅\partial A \cap A = \varnothing, and AA is closed iff ∂A⊆A\partial A \subseteq A.

Proof

We prove each part in turn, using only the axioms of a topology (arbitrary unions of open sets are open, finite intersections of open sets are open, ∅\varnothing and XX are open) and the definition of closed set as complement of open.

(1) Extremal characterization of interior. By definition int⁡(A)=⋃{U∈τ:U⊆A}\operatorname{int}(A) = \bigcup \{U \in \tau : U \subseteq A\}, a union of members of τ\tau, hence open by the topology axiom that arbitrary unions of open sets are open. Each UU in the union satisfies U⊆AU \subseteq A, so the union itself satisfies int⁡(A)⊆A\operatorname{int}(A) \subseteq A. Finally if UU is open with U⊆AU \subseteq A, then UU is one of the sets being unioned over, so U⊆int⁡(A)U \subseteq \operatorname{int}(A). Thus int⁡(A)\operatorname{int}(A) is open, contained in AA, and contains every open subset of AA — it is the largest such set (largest in the sense of set inclusion: any open U⊆AU \subseteq A satisfies U⊆int⁡(A)U \subseteq \operatorname{int}(A)).

(2) Extremal characterization of closure. By definition A‾=⋂{C:C closed,A⊆C}\overline{A} = \bigcap \{C : C \text{ closed}, A \subseteq C\}. An arbitrary intersection of closed sets is closed: if {Ci}\{C_i\} are closed then X∖⋂iCi=⋃i(X∖Ci)X \setminus \bigcap_i C_i = \bigcup_i (X\setminus C_i) is a union of open sets, hence open, so ⋂iCi\bigcap_i C_i is closed. Thus A‾\overline{A} is closed. Each CC in the intersection contains AA, so A⊆A‾A \subseteq \overline{A}. If CC is closed with A⊆CA \subseteq C, then CC is one of the sets intersected, so A‾⊆C\overline{A} \subseteq C. (Note: the family of closed supersets of AA is nonempty since XX itself is closed and contains AA, so the intersection is well-defined and not vacuously all of XX by convention — it genuinely equals the stated intersection.)

(3) Point characterizations. Interior: If x∈int⁡(A)x \in \operatorname{int}(A), then since int⁡(A)\operatorname{int}(A) is open (by part 1) and contained in AA, taking U=int⁡(A)U = \operatorname{int}(A) gives an open set with x∈U⊆Ax \in U \subseteq A. Conversely, if some open UU satisfies x∈U⊆Ax \in U \subseteq A, then by part (1) U⊆int⁡(A)U \subseteq \operatorname{int}(A), so x∈int⁡(A)x \in \operatorname{int}(A). This proves the two sides are equal as sets of points. Closure: We show x∉A‾x \notin \overline{A} iff some open UU contains xx and misses AA entirely, which is the negation of the claimed characterization, hence equivalent to it. If x∉A‾x \notin \overline{A}, since A‾\overline{A} is closed (part 2), U:=X∖A‾U := X \setminus \overline{A} is open and contains xx. Also U∩A=∅U \cap A = \varnothing because A⊆A‾A \subseteq \overline{A}. Conversely, suppose open U∋xU \ni x has U∩A=∅U \cap A = \varnothing; then A⊆X∖UA \subseteq X \setminus U, and X∖UX\setminus U is closed, so by the extremal property (part 2) A‾⊆X∖U\overline{A} \subseteq X \setminus U, giving x∉A‾x \notin \overline{A} since x∈Ux \in U. This establishes the contrapositive, hence the stated characterization.

(4) Duality. We show X∖A‾=int⁡(X∖A)X \setminus \overline{A} = \operatorname{int}(X \setminus A); the second identity follows by replacing AA with X∖AX\setminus A (using X∖(X∖A)=AX\setminus(X\setminus A) = A). X∖A‾X \setminus \overline{A} is open, since A‾\overline{A} is closed (part 2). It is contained in X∖AX \setminus A: because A⊆A‾A \subseteq \overline{A} (part 2), complementing (which reverses inclusion) gives X∖A‾⊆X∖AX \setminus \overline{A} \subseteq X \setminus A. So X∖A‾X\setminus \overline{A} is an open subset of X∖AX \setminus A; by the extremal property of interior (part 1), X∖A‾⊆int⁡(X∖A)X \setminus \overline{A} \subseteq \operatorname{int}(X \setminus A). For the reverse inclusion, int⁡(X∖A)\operatorname{int}(X\setminus A) is open and contained in X∖AX \setminus A (part 1), so complementing, X∖int⁡(X∖A)X \setminus \operatorname{int}(X\setminus A) is closed and contains AA. By the extremal property of closure (part 2), A‾⊆X∖int⁡(X∖A)\overline{A} \subseteq X \setminus \operatorname{int}(X \setminus A), and complementing again gives int⁡(X∖A)⊆X∖A‾\operatorname{int}(X \setminus A) \subseteq X \setminus \overline{A}. Both inclusions give X∖A‾=int⁡(X∖A)X \setminus \overline{A} = \operatorname{int}(X\setminus A).

(5) Idempotence. int⁡(A)\operatorname{int}(A) is open (part 1). For any open set UU, U⊆UU \subseteq U trivially, and UU is the largest open subset of itself, so int⁡(U)=U\operatorname{int}(U) = U: indeed int⁡(U)⊆U\operatorname{int}(U) \subseteq U always, and UU itself is an open subset of UU so U⊆int⁡(U)U \subseteq \operatorname{int}(U) by the extremal property. Applying this with U=int⁡(A)U = \operatorname{int}(A) gives int⁡(int⁡(A))=int⁡(A)\operatorname{int}(\operatorname{int}(A)) = \operatorname{int}(A). Symmetrically A‾\overline{A} is closed (part 2), and any closed set CC satisfies C‾=C\overline{C} = C by the same argument dualized, so A‾‾=A‾\overline{\overline{A}} = \overline{A}.

(6) Monotonicity. Suppose A⊆BA \subseteq B. If UU is open with U⊆AU \subseteq A, then U⊆BU \subseteq B, so UU is one of the open sets whose union defines int⁡(B)\operatorname{int}(B); hence U⊆int⁡(B)U \subseteq \operatorname{int}(B). Taking the union over all such U⊆AU \subseteq A gives int⁡(A)⊆int⁡(B)\operatorname{int}(A) \subseteq \operatorname{int}(B). Dually, if CC is closed with B⊆CB \subseteq C, then A⊆CA \subseteq C too, so CC is one of the closed sets intersected to form A‾\overline{A}, giving A‾⊆C\overline{A} \subseteq C; intersecting over all such C⊇BC \supseteq B gives A‾⊆B‾\overline{A} \subseteq \overline{B}.

(7) Openness/closedness criteria. If AA is open, then AA is an open subset of itself, so by the extremal property A⊆int⁡(A)A \subseteq \operatorname{int}(A); combined with int⁡(A)⊆A\operatorname{int}(A) \subseteq A (always true), A=int⁡(A)A = \operatorname{int}(A). Conversely if A=int⁡(A)A = \operatorname{int}(A), then since int⁡(A)\operatorname{int}(A) is open (part 1), AA is open. The closed case is dual: if AA is closed, AA is a closed superset of itself so A‾⊆A\overline{A} \subseteq A by the extremal property; with A⊆A‾A \subseteq \overline{A} always true, A=A‾A = \overline{A}. Conversely if A=A‾A = \overline{A}, since A‾\overline{A} is closed, AA is closed.

(8) Interaction with union/intersection. Interior and intersection (equality): A∩B⊆AA \cap B \subseteq A and A∩B⊆BA \cap B \subseteq B, so by monotonicity (part 6), int⁡(A∩B)⊆int⁡(A)\operatorname{int}(A \cap B) \subseteq \operatorname{int}(A) and int⁡(A∩B)⊆int⁡(B)\operatorname{int}(A\cap B) \subseteq \operatorname{int}(B), giving int⁡(A∩B)⊆int⁡(A)∩int⁡(B)\operatorname{int}(A \cap B) \subseteq \operatorname{int}(A) \cap \operatorname{int}(B). For the reverse, int⁡(A)∩int⁡(B)\operatorname{int}(A) \cap \operatorname{int}(B) is an intersection of two open sets, hence open (finite intersection axiom), and it is contained in A∩BA \cap B since int⁡(A)⊆A\operatorname{int}(A) \subseteq A and int⁡(B)⊆B\operatorname{int}(B) \subseteq B. By the extremal property (part 1), int⁡(A)∩int⁡(B)⊆int⁡(A∩B)\operatorname{int}(A) \cap \operatorname{int}(B) \subseteq \operatorname{int}(A \cap B). Both inclusions give equality. Interior and union (inclusion only): A⊆A∪BA \subseteq A \cup B and B⊆A∪BB \subseteq A \cup B, so by monotonicity int⁡(A)⊆int⁡(A∪B)\operatorname{int}(A) \subseteq \operatorname{int}(A\cup B) and int⁡(B)⊆int⁡(A∪B)\operatorname{int}(B) \subseteq \operatorname{int}(A \cup B); taking the union of these two inclusions gives int⁡(A)∪int⁡(B)⊆int⁡(A∪B)\operatorname{int}(A) \cup \operatorname{int}(B) \subseteq \operatorname{int}(A \cup B). Closure and union (equality): By duality (part 4), A∪B‾=X∖int⁡(X∖(A∪B))=X∖int⁡((X∖A)∩(X∖B))\overline{A \cup B} = X \setminus \operatorname{int}(X \setminus (A \cup B)) = X \setminus \operatorname{int}((X\setminus A) \cap (X \setminus B)). By the just-proved interior/intersection equality, this equals X∖(int⁡(X∖A)∩int⁡(X∖B))X \setminus (\operatorname{int}(X\setminus A) \cap \operatorname{int}(X \setminus B)), which by De Morgan equals (X∖int⁡(X∖A))∪(X∖int⁡(X∖B))(X \setminus \operatorname{int}(X\setminus A)) \cup (X \setminus \operatorname{int}(X \setminus B)), which by duality again equals A‾∪B‾\overline{A} \cup \overline{B}. Closure and intersection (inclusion only): dualize the interior/union inclusion the same way: apply duality to int⁡(A)∪int⁡(B)⊆int⁡(A∪B)\operatorname{int}(A) \cup \operatorname{int}(B) \subseteq \operatorname{int}(A \cup B) with A,BA, B replaced by X∖A,X∖BX\setminus A, X \setminus B, then complement (which reverses ⊆\subseteq), yielding A∩B‾⊆A‾∩B‾\overline{A \cap B} \subseteq \overline{A} \cap \overline{B}.

(9) Boundary facts. ∂A=A‾∩X∖A‾\partial A = \overline{A} \cap \overline{X \setminus A} (this equals A‾∖int⁡(A)\overline{A}\setminus \operatorname{int}(A) by duality, part 4: A‾∖int⁡(A)=A‾∩(X∖int⁡(A))=A‾∩X∖A‾\overline{A} \setminus \operatorname{int}(A) = \overline{A} \cap (X\setminus \operatorname{int}(A)) = \overline{A} \cap \overline{X\setminus A}). This is an intersection of two closed sets (part 2 applied to AA and to X∖AX\setminus A), hence closed. Symmetry ∂A=∂(X∖A)\partial A = \partial(X\setminus A) is immediate from the symmetric formula A‾∩X∖A‾\overline{A}\cap\overline{X\setminus A}. For the trichotomy: int⁡(A)\operatorname{int}(A), ∂A\partial A, int⁡(X∖A)\operatorname{int}(X\setminus A) are pairwise disjoint and cover XX. Disjointness: int⁡(A)∩int⁡(X∖A)⊆A∩(X∖A)=∅\operatorname{int}(A) \cap \operatorname{int}(X \setminus A) \subseteq A \cap (X\setminus A) = \varnothing; and ∂A∩int⁡(A)=(A‾∖int⁡(A))∩int⁡(A)=∅\partial A \cap \operatorname{int}(A) = (\overline{A} \setminus \operatorname{int}(A)) \cap \operatorname{int}(A) = \varnothing by definition of set difference; similarly ∂A∩int⁡(X∖A)=∅\partial A \cap \operatorname{int}(X\setminus A) = \varnothing since ∂A=∂(X∖A)=X∖A‾∖int⁡(X∖A)\partial A = \partial(X\setminus A) = \overline{X\setminus A} \setminus \operatorname{int}(X\setminus A). Covering: for any x∈Xx \in X, either x∈A‾x \in \overline{A} or not. If x∉A‾x \notin \overline{A}, then by duality (part 4) x∈int⁡(X∖A)x \in \operatorname{int}(X \setminus A). If x∈A‾x \in \overline{A}, then either x∈int⁡(A)x \in \operatorname{int}(A), or x∉int⁡(A)x \notin \operatorname{int}(A) in which case x∈A‾∖int⁡(A)=∂Ax \in \overline{A}\setminus \operatorname{int}(A) = \partial A. In all cases xx lies in one of the three pieces, proving X=int⁡(A)⊔∂A⊔int⁡(X∖A)X = \operatorname{int}(A) \sqcup \partial A \sqcup \operatorname{int}(X\setminus A). Finally, AA open iff A=int⁡(A)A = \operatorname{int}(A) (part 7) iff ∂A∩A=∅\partial A \cap A = \varnothing: if A=int⁡(A)A = \operatorname{int}(A) then A∩∂A=int⁡(A)∩∂A=∅A \cap \partial A = \operatorname{int}(A) \cap \partial A = \varnothing by disjointness just shown. Conversely if ∂A∩A=∅\partial A \cap A = \varnothing, take x∈Ax \in A; since x∉∂Ax \notin \partial A and A⊆A‾A \subseteq \overline A so x∈A‾x\in\overline A, the trichotomy forces x∈int⁡(A)x \in \operatorname{int}(A) (it cannot be in int⁡(X∖A)⊆X∖A\operatorname{int}(X\setminus A) \subseteq X\setminus A since x∈Ax \in A). Hence A⊆int⁡(A)A \subseteq \operatorname{int}(A), and with the always-true reverse inclusion, A=int⁡(A)A = \operatorname{int}(A), so AA is open. The closed case: AA closed iff A=A‾A = \overline{A} (part 7) iff ∂A⊆A\partial A \subseteq A, by an entirely dual argument applied to X∖AX \setminus A using ∂(X∖A)=∂A\partial(X\setminus A) = \partial A and the open case just proved. This completes the proof of all nine parts. ■\blacksquare

∎

Worked examples

Example 1 (asymmetric subset of R\mathbb{R}, standard topology). Let X=RX = \mathbb{R} with the standard topology (open sets are unions of open intervals), and let

A=(0,1)∪{2}∪([3,4]∩Q).A = (0,1) \cup \{2\} \cup ([3,4] \cap \mathbb{Q}).

This set is deliberately asymmetric: an open piece, an isolated point, and a "porous" piece (rationals in [3,4][3,4], which contain no interval). Interior. We claim int⁡(A)=(0,1)\operatorname{int}(A) = (0,1). First, (0,1)(0,1) is open and (0,1)⊆A(0,1) \subseteq A, so (0,1)⊆int⁡(A)(0,1) \subseteq \operatorname{int}(A) by the extremal property. For the reverse, suppose x∈int⁡(A)x \in \operatorname{int}(A); then some open interval (x−ε,x+ε)⊆A(x-\varepsilon,x+\varepsilon) \subseteq A for some ε>0\varepsilon>0. Such an interval cannot be centered at 22 (since any interval around 22 contains points outside {2}∪(0,1)∪([3,4]∩Q)\{2\}\cup(0,1)\cup([3,4]\cap\mathbb Q), e.g. points in (2−ε,2)(2-\varepsilon,2) not equal to 22 and not in (0,1)∪([3,4]∩Q)(0,1)\cup([3,4]\cap\mathbb Q) once ε<1\varepsilon<1). It also cannot be centered at any point of [3,4]∩Q[3,4]\cap\mathbb{Q}, or any point x∈[3,4]x \in [3,4] at all, because every open interval around a point of [3,4][3,4] contains irrational numbers (by density of irrationals), and irrationals in (3,4)(3,4) lie outside AA. Hence any interior point of AA must come from the (0,1)(0,1) piece, forcing x∈(0,1)x\in (0,1) (endpoints 0,10,1 fail since any interval around 00 contains negative numbers not in AA, similarly at 11). So int⁡(A)=(0,1)\operatorname{int}(A) = (0,1). Closure. We claim A‾=[0,1]∪{2}∪[3,4]\overline{A} = [0,1] \cup \{2\} \cup [3,4]. The set [0,1]∪{2}∪[3,4][0,1]\cup\{2\}\cup[3,4] is closed (finite union of closed sets — a closed interval, a point, a closed interval — and finite unions of closed sets are closed, dually to finite intersections of opens), and it contains AA, so by the extremal property of closure, A‾⊆[0,1]∪{2}∪[3,4]\overline{A} \subseteq [0,1]\cup\{2\}\cup[3,4]. For the reverse inclusion we check each piece is adherent: every point of [0,1][0,1] is a limit of points of (0,1)⊆A(0,1) \subseteq A (standard density of open intervals in their closed hulls); the point 22 lies in AA, hence in A‾\overline{A}; every point of [3,4][3,4] is a limit of rationals in [3,4]∩Q⊆A[3,4] \cap \mathbb{Q} \subseteq A, since rationals are dense in every real interval. Hence [0,1]∪{2}∪[3,4]⊆A‾[0,1]\cup\{2\}\cup[3,4] \subseteq \overline{A}, giving equality. Boundary. ∂A=A‾∖int⁡(A)=([0,1]∪{2}∪[3,4])∖(0,1)={0,1}∪{2}∪[3,4]\partial A = \overline{A} \setminus \operatorname{int}(A) = \big([0,1]\cup\{2\}\cup[3,4]\big) \setminus (0,1) = \{0,1\} \cup \{2\} \cup [3,4]. Note the entire interval [3,4][3,4] is boundary — every point there is a limit of both AA (rationals) and X∖AX \setminus A (irrationals, and reals outside [3,4][3,4]), matching the theorem's characterization that boundary points see both AA and its complement in every neighborhood.

Example 2 (a finite non-Hausdorff space, showing closure need not be "nice"). Let X={a,b,c,d}X = \{a,b,c,d\} with topology

τ={∅, {a}, {a,b}, {a,b,c}, X}.\tau = \{\varnothing,\ \{a\},\ \{a,b\},\ \{a,b,c\},\ X\}.

One checks this is a topology: it is closed under arbitrary unions and finite intersections directly by inspection (it is a nested chain of sets, so unions/intersections of any subfamily equal the largest/smallest member of that subfamily, which is again in τ\tau). Let A={b,d}A = \{b, d\}. Interior. The open subsets of A={b,d}A = \{b,d\} are exactly the members of τ\tau contained in {b,d}\{b,d\}. Checking: ∅⊆A\varnothing \subseteq A (yes), {a}⊈A\{a\} \not\subseteq A, {a,b}⊈A\{a,b\}\not\subseteq A (contains a∉Aa\notin A), {a,b,c}⊈A\{a,b,c\}\not\subseteq A, X⊈AX \not\subseteq A. So the only open subset of AA is ∅\varnothing, and int⁡(A)=∅\operatorname{int}(A) = \varnothing. Closure. First find the closed sets: complements of τ\tau are X,{b,c,d},{c,d},{d},∅X, \{b,c,d\}, \{c,d\}, \{d\}, \varnothing. We need the smallest closed set containing A={b,d}A=\{b,d\}. Checking each closed set against AA: ∅\varnothing fails (A⊈∅A\not\subseteq\varnothing); {d}\{d\} fails (b∉{d}b\notin\{d\}); {c,d}\{c,d\} fails (b∉{c,d}b \notin \{c,d\}); {b,c,d}\{b,c,d\} works ({b,d}⊆{b,c,d}\{b,d\}\subseteq\{b,c,d\}); XX works but is bigger. The smallest closed superset is {b,c,d}\{b,c,d\}, so A‾={b,c,d}\overline{A} = \{b,c,d\}. Boundary. ∂A=A‾∖int⁡(A)={b,c,d}∖∅={b,c,d}\partial A = \overline{A}\setminus\operatorname{int}(A) = \{b,c,d\} \setminus \varnothing = \{b,c,d\}. This example is instructive: the point cc, which is not even in AA, is nonetheless in the boundary of AA — every open set containing cc (namely {a,b,c}\{a,b,c\} or XX) meets both A={b,d}A=\{b,d\} (via bb or dd) and X∖A={a,c}X\setminus A = \{a,c\} (via aa or cc itself). This shows boundary points can be "far" from AA in a naive sense yet topologically inseparable from it — a phenomenon impossible in the more familiar Hausdorff/metric setting where points are always separable from disjoint closed sets not containing them, foreshadowing Hausdorff and Separation Axioms.

CounterexampleWhy "closed sets" (not just any sets) must appear in the definition of closure

The theorem crucially defines A‾\overline{A} as the intersection of closed supersets of AA, and relies on closed sets being closed under arbitrary intersection. If we instead tried to define a "closure-like" operator using an intersection over an arbitrary collection of supersets — not required to be closed, or drawn from some non-topology family F\mathcal{F} that is not closed under intersections — the extremal/idempotence properties can fail. Concretely, work in X=RX = \mathbb{R} with the standard topology, and consider the family F\mathcal{F} of all supersets of A=(0,1)A = (0,1) that are countable unions of closed intervals (this family is not closed under arbitrary intersection: it fails to contain all the intersections one might need, and in particular is not the collection of closed sets in any topology on R\mathbb R finer than making arbitrary infinite intersections work). Define a naive operator

cl⁡~(A)=⋂{F∈F:A⊆F}.\widetilde{\operatorname{cl}}(A) = \bigcap \{ F \in \mathcal{F} : A \subseteq F\}.

Take Fn=[1/n,1−1/n]F_n = [1/n, 1-1/n]-style exhaustions is not quite it — instead here is the cleanest failure: let F\mathcal F consist of all sets of the form A∪SA \cup S where SS is any finite subset of R∖A\mathbb R\setminus A (so F\mathcal F is not required to contain complements of open sets at all). Each such A∪SA\cup S is a "superset of AA", and

⋂S finite(A∪S)=A,\bigcap_{S \text{ finite}} (A \cup S) = A,

since for any point y∉Ay \notin A, the single-element set S={y}S=\{y\} excludes yy from being forced into every member's necessity — more precisely A∪{y}A \cup \{y\} is one member of the family, but AA itself (take S=∅S = \varnothing) is also a member, and AA is in the intersection, so the naive intersection collapses to exactly AA. This "closure" is not even a superset-respecting operator in the topological sense: it produces cl⁡~(A)=A=(0,1)\widetilde{\operatorname{cl}}(A) = A = (0,1), missing the honestly adherent points 00 and 11 that every genuine neighborhood of AA must approach. The reason the real closure does not collapse this way is precisely that we restrict to the family of closed sets, a family engineered (via the topology axioms) to be stable under arbitrary intersections and to correctly detect adherent points via the point-characterization (part 3 of the theorem). Dropping the requirement "CC is closed" breaks both the well-definedness of the extremal characterization and the correct detection of limit points, showing this hypothesis in the definition is not decorative — it is load-bearing.

Scope, edge cases & common traps

SituationWhat happensWhy
AA already openint⁡(A)=A\operatorname{int}(A) = ATheorem part 7
AA already closedA‾=A\overline{A} = ATheorem part 7
A=∅A = \varnothing or A=XA = Xint⁡(A)=A‾=A\operatorname{int}(A)=\overline{A}=A, ∂A=∅\partial A = \varnothingboth ∅,X\varnothing, X are open and closed
Infinite union of interiors⋃int⁡(Ai)⊆int⁡(⋃Ai)\bigcup \operatorname{int}(A_i) \subseteq \operatorname{int}(\bigcup A_i), not always equalitye.g. Aq={q}A_q = \{q\} for each rational qq; each interior is empty but ⋃qAq=Q\bigcup_q A_q = \mathbb Q still has empty interior in R\mathbb R here so equality can hold or fail depending on example — care needed, only finite unions/intersections have the clean laws in general
Infinite intersection of closures⋂Ai‾⊆⋂Ai‾\overline{\bigcap A_i} \subseteq \bigcap \overline{A_i}, not always equalitye.g. An=(0,1/n)A_n = (0, 1/n) in R\mathbb R: ⋂nAn=∅\bigcap_n A_n = \varnothing so LHS is ∅\varnothing, but ⋂nAn‾=⋂n[0,1/n]={0}\bigcap_n \overline{A_n} = \bigcap_n [0,1/n] = \{0\}, which is nonempty
Non-Hausdorff spacesBoundary points can lie outside both AA and any "visible" separation of AA from its complementsee Example 2
Confusing interior/closure of complement with complement of interior/closureThese agree exactly (duality), but sign/direction errors are commonTheorem part 4

The mathematician behind it

Origins · 1 minFelix Hausdorff and how this was discovered
FH
Felix Hausdorff
1868–1942 · founder of modern point-set topology

Felix Hausdorff, working in Germany in the early twentieth century, gave the first systematic axiomatic treatment of topological spaces via neighborhood systems in his 1914 book Grundzüge der Mengenlehre. He isolated the separation property now bearing his name and developed closure, interior, and related operators as basic algebraic tools for reasoning about abstract spaces, independent of any underlying metric. His framework unified earlier, more ad hoc treatments of limit points and convergence due to Cantor and Fréchet into a single coherent axiomatic system, and it remains essentially the framework used today.

History
- 1872: Georg Cantor introduces limit points of sets of real numbers, an early precursor of closure.
- 1906: Maurice Fréchet introduces abstract metric spaces and studies closure via limits of sequences.
- 1914: Felix Hausdorff publishes Grundzüge der Mengenlehre, axiomatizing topological spaces via neighborhoods and formalizing interior/closure operators abstractly.
- 1922: Kazimierz Kuratowski gives an axiomatic characterization of the closure operator via four axioms (the "Kuratowski closure axioms"), showing closure alone can generate the whole topology.
- 1930s: The modern open-set axiomatization (as used in this article) becomes standard, cementing interior/closure/boundary as the first working tools taught in any topology course.
Is int⁡(A)\operatorname{int}(A) always nonempty if AA is nonempty?

No. Interior can be empty even for large or "thick-looking" sets. The classic example is A=QA = \mathbb{Q} inside R\mathbb{R} with the standard topology: every open interval contains irrational numbers, so no open interval is contained in Q\mathbb{Q}, forcing int⁡(Q)=∅\operatorname{int}(\mathbb{Q}) = \varnothing. More generally, any set whose complement is dense will have empty interior, since no open set can avoid a dense complement. This is why "interior" genuinely measures topological thickness rather than cardinality: Q\mathbb Q is countably infinite yet topologically "thin," while a single open interval, though it may contain fewer points in some naive sense is impossible to compare that way, is topologically "thick" with nonempty interior.

Does closure commute with taking complements the same way it commutes with unions?

Not directly — closure and complement don't commute with each other in a simple sense, but closure and interior are dual under complement, which is a different (and precise) relationship. The exact statement is X∖A‾=int⁡(X∖A)X \setminus \overline{A} = \operatorname{int}(X \setminus A), proven above. This says the complement of the closure of AA equals the interior of the complement of AA — a genuine identity — but it is not the same as saying X∖A‾=X∖A‾\overline{X \setminus A} = X \setminus \overline{A}, which is false in general (e.g., take AA a nonempty proper open subset of R\mathbb R that is not closed, such as A=(0,1)A=(0,1); then check the two sides differ at the boundary points 0,10,1).

Can the boundary of a set be all of XX?

Yes, when both AA and X∖AX\setminus A are dense but neither is open nor closed in a way that creates interior. The rationals again: ∂Q=Q‾∩R∖Q‾=R∩R=R\partial \mathbb{Q} = \overline{\mathbb{Q}} \cap \overline{\mathbb{R}\setminus \mathbb{Q}} = \mathbb{R} \cap \mathbb{R} = \mathbb{R}, since both rationals and irrationals are dense in R\mathbb R. So the entire real line is the "boundary" of the rationals — every real number, rational or not, has every neighborhood meeting both Q\mathbb Q and its complement.

How do closure and interior relate to continuity?

A function f:X→Yf: X \to Y between topological spaces is continuous if and only if f−1(B‾)⊇f−1(B)‾f^{-1}(\overline{B}) \supseteq \overline{f^{-1}(B)} for every B⊆YB \subseteq Y, or equivalently, if and only if f(A‾)⊆f(A)‾f(\overline{A}) \subseteq \overline{f(A)} for every A⊆XA \subseteq X. These closure-based characterizations are frequently used as alternate definitions of continuity, especially in contexts (like Kuratowski's closure-axiom approach) where closure is taken as the primitive notion instead of open sets. They generalize the familiar calculus fact that continuous functions send convergent sequences (and their limits) to convergent sequences.

Is the boundary operator idempotent, like interior and closure?

No, and this is a common point of confusion. While int⁡(int⁡(A))=int⁡(A)\operatorname{int}(\operatorname{int}(A)) = \operatorname{int}(A) and A‾‾=A‾\overline{\overline A} = \overline A always, it is not true in general that ∂(∂A)=∂A\partial(\partial A) = \partial A. In fact ∂A\partial A is always closed (theorem part 9), and for closed sets CC one can show ∂(∂C)⊆∂C\partial(\partial C) \subseteq \partial C but the containment can be strict. For example, in R\mathbb R, let A=Q∩[0,1]A = \mathbb{Q} \cap [0,1]; then ∂A=[0,1]\partial A = [0,1] (a similar density argument as for Q\mathbb Q restricted to [0,1][0,1]), and since [0,1][0,1] is closed, ∂([0,1])={0,1}\partial([0,1]) = \{0,1\}, which is strictly smaller than [0,1]=∂A[0,1] = \partial A. So ∂(∂A)≠∂A\partial(\partial A) \neq \partial A here.

What's the fastest way to check whether a given point is in the closure of AA?

Use the point-characterization directly (theorem part 3): x∈A‾x \in \overline A if and only if every open set containing xx intersects AA. In a metric space this simplifies further to: x∈A‾x \in \overline A iff for every ε>0\varepsilon > 0 there exists a∈Aa \in A with d(x,a)<εd(x,a) < \varepsilon, i.e., xx is a limit of a sequence in AA (or x∈Ax \in A itself). This sequential test, valid in metric and more generally first-countable spaces, is usually the quickest practical check, though it can fail in more exotic (non-first-countable) topological spaces, where the neighborhood-based definition is the one that always works.

Before this

Where this leads