Bases and Subbases for a Topology

How do you describe a topology without listing every open set? For the real line, we do not enumerate every open subset. Instead, we say a set is open if it is a union of open intervals. This single idea is the subject of bases and subbases. The idea is that a small collection of "generating" sets can determine an entire topology. It matters because most topologies encountered in practice are defined this way. Metric topologies, product topologies, order topologies — these are typical examples. They are defined not by their full list of open sets. Instead, they are defined by a compact generating family. Knowing when a family of sets can generate a topology is essential. Knowing how to check openness relative to it is also essential. This is a basic working skill in topology. The big picture is this: a basis is a collection of "elementary" open sets. These sets are large enough to build every open set as a union. They are also small enough to check via a simple local criterion. A subbasis is an even more economical device. It is a collection whose finite intersections form a basis. Both notions let us pass from a small family of sets to a well-defined topology. This family is often geometrically transparent. They also let us verify continuity, openness, and convergence. We can test only on the generating family, rather than on all open sets. This article gives the precise definitions of basis and subbasis. It states and proves the basis criterion. This criterion characterizes which collections of sets qualify as a basis for some topology. It also proves the companion fact. Any collection of subsets of a set generates a topology. This happens via finite intersections and arbitrary unions. We work carefully through the logical order. First, we fix what a basis for a given topology means. Then, we show how an abstract collection can construct a topology from scratch. This collection must satisfy certain axioms. By the end, you will be able to certify whether a given family is a basis. You will be able to construct the topology it generates. You will distinguish basis-generation from the coarser subbasis-generation. You will also see why these tools matter. They are the natural setup for later constructions, such as The Product Topology. They are also useful for testing Continuity of Maps.

Key takeaways
  • A basis for a topology T\mathcal T is a subcollection B⊆T\mathcal B \subseteq \mathcal T such that every open set is a union of basis elements.
  • The basis criterion tells you exactly when an abstract family B\mathcal B of subsets of XX is a basis for some topology: cover XX, and intersections of basis elements are unions of basis elements.
  • A subbasis S\mathcal S need not have the intersection property; the topology it generates is built from finite intersections of S\mathcal S-elements, then arbitrary unions.
  • The topology generated by a basis (or subbasis) is the coarsest topology containing that family — it is unique and well-defined.
  • Openness of a set can be checked purely locally: UU is open iff every point of UU has a basis element around it contained in UU.
  • Key idea: small structured families of sets — bases and subbases — freely generate topologies, and open sets are exactly unions (or unions of finite intersections) drawn from them.

Where this lives today

Bases and subbases are the workhorse for constructing every standard topology efficiently.

  • Product and box topologies: the The Product Topology on a product ∏iXi\prod_i X_i is defined via a subbasis (cylinder sets restricted in one coordinate), then generated exactly by the finite-intersection/union recipe proved here — this is the very next construction that relies directly on this article's corollary.
  • Continuity testing: to prove a map f:X→Yf: X \to Y is continuous, it suffices to check that f−1(B)f^{-1}(B) is open for every BB in a basis (or even subbasis) of YY; this reduces Continuity of Maps checks from "all open sets" to a small generating family.
  • Metric and normed spaces: open balls of rational radius centered at points of a countable dense set form a countable basis for separable metric spaces, underlying second-countability arguments in analysis.
  • Zariski topology in algebraic geometry: the subbasis of complements of zero-sets of single polynomials generates the Zariski topology on affine space, a construction formally identical to the subbasis recipe here.
  • Computer science and domain theory: Scott topologies on partially ordered sets (used in denotational semantics of programming languages) are specified via bases of "way-below" sets, directly invoking the basis criterion.
  • Probability and measure theory: the Borel σ\sigma-algebra is generated from a basis (open intervals, or open sets) via a closure operation analogous in spirit to subbasis generation, though closing under countable rather than finite operations.

Ask the tutor: Can you show why the collection of open balls in a metric space always satisfies the basis intersection property, using the triangle inequality?

DefinitionBasis, Subbasis, and Generated Topology

Let XX be a set.

Basis for a topology. Let (X,T)(X,\mathcal T) be a topological space. A subcollection B⊆T\mathcal B \subseteq \mathcal T is a basis for T\mathcal T if for every open set U∈TU \in \mathcal T and every point x∈Ux \in U, there exists B∈BB \in \mathcal B with

x∈B⊆U.x \in B \subseteq U.

Equivalently, every U∈TU \in \mathcal T can be written as a union of members of B\mathcal B.

Basis for generating a topology (abstract basis). A collection B\mathcal B of subsets of XX is called a basis (in the generative sense) if: 1. (Covering) For every x∈Xx \in X, there exists B∈BB \in \mathcal B with x∈Bx \in B. 2. (Intersection property) For all B1,B2∈BB_1, B_2 \in \mathcal B and every x∈B1∩B2x \in B_1 \cap B_2, there exists B3∈BB_3 \in \mathcal B with

x∈B3⊆B1∩B2.x \in B_3 \subseteq B_1 \cap B_2.

Given such B\mathcal B, define

TB={ U⊆X:for every x∈U, ∃ B∈B with x∈B⊆U }.\mathcal T_{\mathcal B} = \{\, U \subseteq X : \text{for every } x \in U,\ \exists\, B \in \mathcal B \text{ with } x \in B \subseteq U \,\}.

This TB\mathcal T_{\mathcal B} is called the topology generated by B\mathcal B.

Subbasis. A collection S\mathcal S of subsets of XX with ⋃S∈SS=X\bigcup_{S \in \mathcal S} S = X is called a subbasis. The topology generated by S\mathcal S, denoted TS\mathcal T_{\mathcal S}, is defined as the smallest topology on XX containing S\mathcal S; equivalently, it is TB\mathcal T_{\mathcal B} where

B={⋂i=1nSi:n≥0, S1,…,Sn∈S}\mathcal B = \left\{ \bigcap_{i=1}^{n} S_i : n \ge 0,\ S_1,\dots,S_n \in \mathcal S \right\}

is the collection of all finite intersections of members of S\mathcal S (with the empty intersection, n=0n=0, taken to be XX).

Intuition

Think of B\mathcal B as a stock of "elementary tiles" you're allowed to lay down. A set UU counts as open exactly when it can be assembled — perhaps using infinitely many tiles — as a union of these elementary pieces. On R\mathbb{R}, the tiles are open intervals (a,b)(a,b); every open set, however wild its shape, decomposes as a countable union of disjoint open intervals. The interval basis is small (parametrized by two real numbers) yet it generates the entire, vastly larger, standard topology. The two basis axioms encode exactly what you need for "union of tiles" to behave like a topology. Covering ensures XX itself, which must be open, is assembled from tiles. The intersection property ensures that when two tiles overlap, the overlap region is itself coverable by tiles — this is precisely what guarantees that finite intersections of open sets (unions of tiles) remain open (unions of tiles). Without it, intersecting two "open" unions of tiles might produce a set that cannot be redecomposed into tiles, breaking the topology axioms. A subbasis relaxes this further: its elements need not overlap nicely at all. To repair this, we first close the subbasis under finite intersections — manufacturing the missing overlap regions by brute force — and then take all unions. This two-stage process (finite intersect, then arbitrarily union) is the generic recipe for building a topology from any starting family whatsoever, however badly behaved.

Figure
Basis tiles generating an open set on R\mathbb{R}: three basic intervals unioning to an open set UU
ℝ -3 0 1 3 4 5 B₁=(-3,1) B₂=(0,3) B₃=(3,5) U = B₁ ∪ B₂ ∪ B₃ = (-3,5)
TheoremThe Basis Criterion

Let XX be a set and let B\mathcal B be a collection of subsets of XX satisfying: 1. (Covering) ⋃B∈BB=X\bigcup_{B \in \mathcal B} B = X. 2. (Intersection property) For all B1,B2∈BB_1, B_2 \in \mathcal B and every x∈B1∩B2x \in B_1 \cap B_2, there is B3∈BB_3 \in \mathcal B with x∈B3⊆B1∩B2x \in B_3 \subseteq B_1 \cap B_2. Define

TB={ U⊆X:∀x∈U, ∃B∈B, x∈B⊆U }.\mathcal T_{\mathcal B} = \{\, U \subseteq X : \forall x \in U,\ \exists B \in \mathcal B,\ x \in B \subseteq U \,\}.

Then TB\mathcal T_{\mathcal B} is a topology on XX, and B⊆TB\mathcal B \subseteq \mathcal T_{\mathcal B} with B\mathcal B a basis for TB\mathcal T_{\mathcal B} in the sense of the definition above. Moreover TB\mathcal T_{\mathcal B} is the coarsest topology on XX containing B\mathcal B: if T′\mathcal T' is any topology on XX with B⊆T′\mathcal B \subseteq \mathcal T', then TB⊆T′\mathcal T_{\mathcal B} \subseteq \mathcal T'. Conversely, if B\mathcal B is a basis (in the first, relative sense) for some topology T\mathcal T on XX, then B\mathcal B satisfies conditions 1 and 2 above, and T=TB\mathcal T = \mathcal T_{\mathcal B}.

Proof

We use direct verification of the topology axioms — the standard proof technique for basis-generation theorems — together with a double-inclusion argument for coarseness, and finish with the converse direction.

Step 1: ∅,X∈TB\emptyset, X \in \mathcal T_{\mathcal B}. The empty set vacuously satisfies the defining condition of TB\mathcal T_{\mathcal B} (there is no x∈∅x \in \emptyset to check), so ∅∈TB\emptyset \in \mathcal T_{\mathcal B}. For XX: by the covering axiom, every x∈Xx \in X lies in some B∈BB \in \mathcal B, and trivially B⊆XB \subseteq X; hence X∈TBX \in \mathcal T_{\mathcal B}.

Step 2: TB\mathcal T_{\mathcal B} is closed under arbitrary unions. Let {Ui}i∈I⊆TB\{U_i\}_{i \in I} \subseteq \mathcal T_{\mathcal B} and let U=⋃i∈IUiU = \bigcup_{i \in I} U_i. Take any x∈Ux \in U. Then x∈Ui0x \in U_{i_0} for some i0∈Ii_0 \in I. Since Ui0∈TBU_{i_0} \in \mathcal T_{\mathcal B}, there exists B∈BB \in \mathcal B with x∈B⊆Ui0⊆Ux \in B \subseteq U_{i_0} \subseteq U. Since x∈Ux \in U was arbitrary, U∈TBU \in \mathcal T_{\mathcal B}.

Step 3: TB\mathcal T_{\mathcal B} is closed under finite intersections. It suffices to show closure under intersection of two sets, then induct. Let U1,U2∈TBU_1, U_2 \in \mathcal T_{\mathcal B} and x∈U1∩U2x \in U_1 \cap U_2. Since U1∈TBU_1 \in \mathcal T_{\mathcal B}, there is B1∈BB_1 \in \mathcal B with x∈B1⊆U1x \in B_1 \subseteq U_1. Since U2∈TBU_2 \in \mathcal T_{\mathcal B}, there is B2∈BB_2 \in \mathcal B with x∈B2⊆U2x \in B_2 \subseteq U_2. Now x∈B1∩B2x \in B_1 \cap B_2, so by the intersection property (axiom 2) there is B3∈BB_3 \in \mathcal B with

x∈B3⊆B1∩B2⊆U1∩U2.x \in B_3 \subseteq B_1 \cap B_2 \subseteq U_1 \cap U_2.

Since x∈U1∩U2x \in U_1 \cap U_2 was arbitrary, U1∩U2∈TBU_1 \cap U_2 \in \mathcal T_{\mathcal B}. By induction on nn, any finite intersection U1∩⋯∩UnU_1 \cap \cdots \cap U_n of members of TB\mathcal T_{\mathcal B} lies in TB\mathcal T_{\mathcal B}: the base case n=1n=1 is trivial and the inductive step applies the two-set case to (U1∩⋯∩Un−1)(U_1 \cap \cdots \cap U_{n-1}) and UnU_n. Steps 1–3 show TB\mathcal T_{\mathcal B} satisfies all three topology axioms, so TB\mathcal T_{\mathcal B} is a topology on XX.

Step 4: B⊆TB\mathcal B \subseteq \mathcal T_{\mathcal B}, and B\mathcal B is a basis for TB\mathcal T_{\mathcal B}. Let B0∈BB_0 \in \mathcal B. For any x∈B0x \in B_0, take B=B0∈BB = B_0 \in \mathcal B; then x∈B0⊆B0x \in B_0 \subseteq B_0, so the defining condition of TB\mathcal T_{\mathcal B} is met with U=B0U = B_0. Hence B0∈TBB_0 \in \mathcal T_{\mathcal B}, proving B⊆TB\mathcal B \subseteq \mathcal T_{\mathcal B}. To see B\mathcal B is a basis for TB\mathcal T_{\mathcal B} in the relative sense: let U∈TBU \in \mathcal T_{\mathcal B} and x∈Ux \in U. By definition of TB\mathcal T_{\mathcal B}, there is B∈BB \in \mathcal B with x∈B⊆Ux \in B \subseteq U. This is exactly the basis condition.

Step 5: Coarseness — TB\mathcal T_{\mathcal B} is the smallest topology containing B\mathcal B. Let T′\mathcal T' be any topology on XX with B⊆T′\mathcal B \subseteq \mathcal T'. Let U∈TBU \in \mathcal T_{\mathcal B}. For each x∈Ux \in U choose Bx∈B⊆T′B_x \in \mathcal B \subseteq \mathcal T' with x∈Bx⊆Ux \in B_x \subseteq U (possible by definition of TB\mathcal T_{\mathcal B}). Then

U=⋃x∈UBx,U = \bigcup_{x \in U} B_x,

since each Bx⊆UB_x \subseteq U gives one inclusion, and each x∈Ux \in U lies in its own BxB_x giving the reverse inclusion. Each Bx∈T′B_x \in \mathcal T', and T′\mathcal T' is closed under arbitrary unions (it is a topology), so U∈T′U \in \mathcal T'. Hence TB⊆T′\mathcal T_{\mathcal B} \subseteq \mathcal T', as claimed.

Step 6: Converse. Suppose B\mathcal B is a basis for a topology T\mathcal T on XX, in the relative sense: B⊆T\mathcal B \subseteq \mathcal T and every U∈TU \in \mathcal T is a union of members of B\mathcal B. Covering: Since X∈TX \in \mathcal T, XX is a union of members of B\mathcal B, so ⋃B∈BB=X\bigcup_{B \in \mathcal B} B = X; axiom 1 holds. Intersection property: Let B1,B2∈B⊆TB_1, B_2 \in \mathcal B \subseteq \mathcal T. Since T\mathcal T is a topology, B1∩B2∈TB_1 \cap B_2 \in \mathcal T. Since B\mathcal B is a basis for T\mathcal T, B1∩B2B_1 \cap B_2 is a union of members of B\mathcal B; in particular for any x∈B1∩B2x \in B_1 \cap B_2 there is some B3∈BB_3 \in \mathcal B in that union with x∈B3⊆B1∩B2x \in B_3 \subseteq B_1 \cap B_2. This is axiom 2. So B\mathcal B satisfies conditions 1 and 2, and by Steps 1–5 above, TB\mathcal T_{\mathcal B} is a well-defined topology with B\mathcal B as a basis for it. It remains to see T=TB\mathcal T = \mathcal T_{\mathcal B}. T⊆TB\mathcal T \subseteq \mathcal T_{\mathcal B}: since B⊆T\mathcal B \subseteq \mathcal T, Step 5 (applied with T′=T\mathcal T' = \mathcal T) gives TB⊆T\mathcal T_{\mathcal B} \subseteq \mathcal T — wait, we need the reverse; let us verify directly. Let U∈TU \in \mathcal T. Since B\mathcal B is a basis for T\mathcal T, for each x∈Ux \in U there is B∈BB \in \mathcal B with x∈B⊆Ux \in B \subseteq U; this is exactly the condition for U∈TBU \in \mathcal T_{\mathcal B}. So T⊆TB\mathcal T \subseteq \mathcal T_{\mathcal B}. TB⊆T\mathcal T_{\mathcal B} \subseteq \mathcal T: this is Step 5 applied with T′=T\mathcal T' = \mathcal T, since B⊆T\mathcal B \subseteq \mathcal T. Both inclusions hold, so T=TB\mathcal T = \mathcal T_{\mathcal B}. ■\blacksquare

∎
CorollarySubbasis generation

Let XX be a set and S\mathcal S a collection of subsets of XX with ⋃S∈SS=X\bigcup_{S \in \mathcal S} S = X. Let

B={S1∩⋯∩Sn:n≥0, Si∈S}\mathcal B = \left\{ S_1 \cap \cdots \cap S_n : n \ge 0,\ S_i \in \mathcal S \right\}

be the collection of all finite intersections of members of S\mathcal S (with n=0n=0 giving XX by convention). Then B\mathcal B satisfies the basis axioms of the Basis Criterion, and TB\mathcal T_{\mathcal B} is a topology on XX — the topology generated by S\mathcal S — which is the smallest topology on XX containing S\mathcal S.

Proof

Covering: X∈BX \in \mathcal B (the empty intersection), so trivially ⋃B∈BB=X\bigcup_{B \in \mathcal B} B = X. Intersection property: Let B1=S1∩⋯∩SmB_1 = S_1 \cap \cdots \cap S_m and B2=S1′∩⋯∩Sn′B_2 = S_1' \cap \cdots \cap S_n' be two finite intersections of subbasis elements. Then

B1∩B2=S1∩⋯∩Sm∩S1′∩⋯∩Sn′B_1 \cap B_2 = S_1 \cap \cdots \cap S_m \cap S_1' \cap \cdots \cap S_n'

is itself a finite intersection of subbasis elements, hence B1∩B2∈BB_1 \cap B_2 \in \mathcal B. So for any x∈B1∩B2x \in B_1 \cap B_2, taking B3=B1∩B2∈BB_3 = B_1 \cap B_2 \in \mathcal B gives x∈B3⊆B1∩B2x \in B_3 \subseteq B_1 \cap B_2 trivially (equality, in fact). This is even stronger than required by axiom 2. By the Basis Criterion, TB\mathcal T_{\mathcal B} is a topology on XX containing B⊇S\mathcal B \supseteq \mathcal S. To see minimality: let T′\mathcal T' be any topology with S⊆T′\mathcal S \subseteq \mathcal T'. Since T′\mathcal T' is closed under finite intersections, every finite intersection of subbasis elements lies in T′\mathcal T', so B⊆T′\mathcal B \subseteq \mathcal T'. By the coarseness clause of the Basis Criterion (Step 5 of its proof), TB⊆T′\mathcal T_{\mathcal B} \subseteq \mathcal T'. Hence TB\mathcal T_{\mathcal B} is the smallest topology containing S\mathcal S. ■\blacksquare

∎

Worked examples

Example 1: A basis on R2\mathbb{R}^2 that is not the standard one, built from half-open "tick" rectangles. Let X=R2X = \mathbb{R}^2 and let

B={ [a,b)×[c,d):a<b, c<d, a,b,c,d∈R }\mathcal B = \{\, [a,b) \times [c,d) : a < b,\ c < d,\ a,b,c,d \in \mathbb{R} \,\}

be the collection of half-open rectangles, closed on the left/bottom edges, open on the right/top edges. We check the basis axioms directly. Covering: Given any point (x,y)∈R2(x,y) \in \mathbb{R}^2, the rectangle [x,x+1)×[y,y+1)[x, x+1) \times [y, y+1) contains (x,y)(x,y) and belongs to B\mathcal B. So ⋃B=R2\bigcup \mathcal B = \mathbb{R}^2. Intersection property: Take B1=[a1,b1)×[c1,d1)B_1 = [a_1,b_1) \times [c_1,d_1) and B2=[a2,b2)×[c2,d2)B_2 = [a_2,b_2) \times [c_2,d_2). Their intersection is

B1∩B2=[max⁡(a1,a2),min⁡(b1,b2))×[max⁡(c1,c2),min⁡(d1,d2)),B_1 \cap B_2 = [\max(a_1,a_2), \min(b_1,b_2)) \times [\max(c_1,c_2), \min(d_1,d_2)),

which is again of the required rectangular form — provided it is nonempty, i.e. max⁡(a1,a2)<min⁡(b1,b2)\max(a_1,a_2) < \min(b_1,b_2) and similarly in the second coordinate. If it is nonempty it is itself a member of B\mathcal B, so for any point xx in the intersection we may take B3=B1∩B2B_3 = B_1 \cap B_2. If the intersection is empty, axiom 2 is vacuous. Either way the axiom holds. Concretely, let B1=[0,3)×[1,4)B_1 = [0,3) \times [1,4) and B2=[2,5)×[−1,2)B_2 = [2,5) \times [-1,2). Then

B1∩B2=[max⁡(0,2),min⁡(3,5))×[max⁡(1,−1),min⁡(4,2))=[2,3)×[1,2),B_1 \cap B_2 = [\max(0,2), \min(3,5)) \times [\max(1,-1), \min(4,2)) = [2,3) \times [1,2),

which is a nonempty member of B\mathcal B, confirming the mechanics of the calculation. By the Basis Criterion, TB\mathcal T_{\mathcal B} is a genuine topology on R2\mathbb{R}^2: it is called the lower-limit (or Sorgenfrey-type) product topology. Notice that a set like [0,1)×[0,1)[0,1) \times [0,1) is open here but (0,1)×(0,1)(0,1) \times (0,1) is also open, since it is a union of basis rectangles [a,1)×[b,1)[a,1) \times [b,1) with a,ba,b decreasing to 00 — but the singleton edge behavior differs from the standard topology: the set {0}×[0,1)\{0\} \times [0,1) is not open in the standard topology on R2\mathbb{R}^2 but is not open here either (no rectangle [a,b)×[c,d)[a,b)\times[c,d) with a<ba<b fits inside a set of zero width in the first coordinate). This shows the generated topology, while sharing many open sets with the Euclidean topology, is a genuinely different, strictly finer topology than the standard one on R2\mathbb{R}^2 (every standard-open rectangle is a union of these half-open ones, but not conversely).

Example 2: A subbasis on Z\mathbb{Z} generating a non-obvious topology. Let X=ZX = \mathbb{Z} and let S={Sn:n≥2}\mathcal S = \{ S_n : n \ge 2\} where Sn={k∈Z:n∣k}S_n = \{ k \in \mathbb{Z} : n \mid k \} is the set of multiples of nn. First check the covering requirement: ⋃n≥2Sn\bigcup_{n \ge 2} S_n omits only ±1\pm 1 and 00? Let's check: 0∈Sn0 \in S_n for every nn (since n∣0n \mid 0), so 00 is covered. But an integer like 11 or −1-1 is divisible by no n≥2n \ge 2, so 1∉⋃Sn1 \notin \bigcup S_n. Hence S\mathcal S as stated fails the covering axiom for a subbasis on all of Z\mathbb{Z}. To fix this, enlarge the subbasis: let S′=S∪{Z}\mathcal S' = \mathcal S \cup \{\mathbb{Z}\}, or equivalently work on X=Z∖{−1,1}X = \mathbb{Z} \setminus \{-1,1\} together with 00 excluded from consideration if desired — but the cleanest fix is to adjoin the whole space. Set S′={Sn:n≥2}∪{Z∖{0}}\mathcal S' = \{S_n : n \ge 2\} \cup \{\mathbb{Z}\setminus\{0\}\}; now 1∈Z∖{0}1 \in \mathbb{Z}\setminus\{0\}, and every integer lies in at least one member, so S′\mathcal S' covers Z\mathbb{Z}. Now compute a piece of the generated basis B\mathcal B from the corollary: finite intersections of SnS_n's. Take S2={…,−4,−2,0,2,4,…}S_2 = \{\ldots,-4,-2,0,2,4,\ldots\} (even integers) and S3={…,−6,−3,0,3,6,…}S_3 = \{\ldots,-6,-3,0,3,6,\ldots\} (multiples of 3). Their intersection is

S2∩S3={k:2∣k and 3∣k}={k:6∣k}=S6,S_2 \cap S_3 = \{ k : 2 \mid k \text{ and } 3 \mid k \} = \{ k : 6 \mid k \} = S_6,

since 22 and 33 are coprime, lcm(2,3)=6\mathrm{lcm}(2,3) = 6. More generally Sm∩Sn=Slcm(m,n)S_m \cap S_n = S_{\mathrm{lcm}(m,n)} for any m,n≥2m,n \ge 2, by the standard number-theoretic fact that an integer is divisible by both mm and nn exactly when it is divisible by lcm(m,n)\mathrm{lcm}(m,n). So the basis generated by {Sn:n≥2}\{S_n : n \ge 2\} alone is {Sn:n≥2}∪{∅}\{S_n : n \ge 2\} \cup \{\emptyset\} up to relabeling by lcm's — no genuinely new sets appear beyond further SnS_n's, because the family {Sn}\{S_n\} is already closed under the "intersection produces another SnS_n" operation. This is a case where the subbasis, remarkably, is already a basis (once we also throw in the covering fix above), since finite intersections of subbasis elements stay inside the same family. The topology TS′\mathcal T_{\mathcal S'} on Z\mathbb{Z} (after the covering fix) consists precisely of arbitrary unions of sets of the form SnS_n (multiples of nn) together with Z∖{0}\mathbb{Z}\setminus\{0\}-type pieces — a set is open iff it is a union of "multiples of nn" congruence classes-unions and the co-zero patch, an arithmetic topology quite different from the discrete or cofinite topologies on Z\mathbb{Z}.

CounterexampleDropping the intersection property

The intersection property (axiom 2 in the Basis Criterion) cannot be dropped: without it, TB\mathcal T_{\mathcal B} as defined may fail to be a topology, because finite intersections of "basis-unions" can fail to be basis-unions. Let X={1,2,3}X = \{1,2,3\} and let

B={ {1,2}, {2,3} }.\mathcal B = \big\{\, \{1,2\},\ \{2,3\}\,\big\}.

Covering holds: {1,2}∪{2,3}={1,2,3}=X\{1,2\} \cup \{2,3\} = \{1,2,3\} = X. Intersection property fails: consider B1={1,2}B_1 = \{1,2\} and B2={2,3}B_2 = \{2,3\}; their intersection is B1∩B2={2}B_1 \cap B_2 = \{2\}. For x=2∈B1∩B2x = 2 \in B_1 \cap B_2 we would need some B3∈BB_3 \in \mathcal B with 2∈B3⊆{2}2 \in B_3 \subseteq \{2\}. But every element of B\mathcal B has two points, so no B3∈BB_3 \in \mathcal B satisfies B3⊆{2}B_3 \subseteq \{2\}. Axiom 2 fails. Now define TB\mathcal T_{\mathcal B} exactly as before: U∈TBU \in \mathcal T_{\mathcal B} iff every x∈Ux \in U has some B∈BB \in \mathcal B with x∈B⊆Ux \in B \subseteq U. Let us list the members of TB\mathcal T_{\mathcal B}.

  • ∅\emptyset: vacuously in TB\mathcal T_{\mathcal B}.
  • {1,2}\{1,2\}: for x=1x=1, take B={1,2}⊆{1,2}B=\{1,2\}\subseteq\{1,2\}; for x=2x=2, take B={1,2}B=\{1,2\}. So {1,2}∈TB\{1,2\} \in \mathcal T_{\mathcal B}.
  • {2,3}\{2,3\}: similarly in TB\mathcal T_{\mathcal B}.
  • X={1,2,3}X = \{1,2,3\}: for each point, the corresponding B∈BB \in \mathcal B works (both basis sets sit inside XX). So X∈TBX \in \mathcal T_{\mathcal B}.
  • Is {2}\{2\} in TB\mathcal T_{\mathcal B}? We would need B∈BB \in \mathcal B with 2∈B⊆{2}2 \in B \subseteq \{2\}; neither {1,2}\{1,2\} nor {2,3}\{2,3\} works. So {2}∉TB\{2\} \notin \mathcal T_{\mathcal B}.
  • Is {1}\{1\} in TB\mathcal T_{\mathcal B}? Need BB with 1∈B⊆{1}1 \in B \subseteq \{1\}; no such BB. So {1}∉TB\{1\} \notin \mathcal T_{\mathcal B}. Similarly {3}∉TB\{3\} \notin \mathcal T_{\mathcal B}.

So TB={∅,{1,2},{2,3},{1,2,3}}\mathcal T_{\mathcal B} = \{\emptyset, \{1,2\}, \{2,3\}, \{1,2,3\}\}. But this collection is not closed under intersection: {1,2}∩{2,3}={2}\{1,2\} \cap \{2,3\} = \{2\}, and we just showed {2}∉TB\{2\} \notin \mathcal T_{\mathcal B}. Hence TB\mathcal T_{\mathcal B} fails the finite-intersection topology axiom and is not a topology on XX. This shows precisely why the intersection property is essential in the Basis Criterion: without it, the "unions of basis elements" construction can produce a family that is not closed under finite intersections, so it fails to be a topology at all.

Scope, edge cases & common traps

SituationWhat holdsWhat to watch for
B\mathcal B already known to be a basis for a given topology T\mathcal TAxioms 1, 2 automatically hold (proved in the converse direction)Don't re-derive T\mathcal T; just use T=TB\mathcal T = \mathcal T_{\mathcal B}
B\mathcal B is an abstract family with no topology yet specifiedMust verify axioms 1 and 2 before concluding TB\mathcal T_{\mathcal B} is a topologySkipping axiom 2 is the single most common error (see counterexample)
S\mathcal S is a subbasisTake finite intersections first, then arbitrary unionsTaking unions of S\mathcal S directly, without first intersecting, generally gives too small a family, not a topology
Empty intersection conventionThe empty finite intersection of subbasis elements equals XXForgetting this makes XX fail to appear as generated open, breaking the topology axiom that XX is open
Two bases B1,B2\mathcal B_1, \mathcal B_2 generate the same topologyPossible even if B1≠B2\mathcal B_1 \ne \mathcal B_2 (e.g. open balls vs. open balls of rational radius in Rn\mathbb{R}^n)"Basis" is not unique to a topology; never say "the" basis without context

The mathematician behind it

Origins · 1 minFelix Hausdorff and how this was discovered
FH
Felix Hausdorff
1868–1942 · Founder of point-set topology, Bonn/Greifswald/Bonn

Hausdorff's 1914 treatise Grundzüge der Mengenlehre laid down the axiomatic apparatus of general topological spaces, including the systematic use of neighborhood bases to define topologies economically. Long before "basis" became standard vocabulary, Hausdorff showed that specifying a rich-enough local structure — what we would now call a basis of neighborhoods at each point — suffices to reconstruct an entire consistent notion of open set, continuity, and convergence. His name is permanently attached to the separation axiom ("Hausdorff space") that guarantees points can be separated by disjoint open sets drawn from any basis.

History
- 1906 — Maurice Fréchet introduces abstract metric spaces, giving the first general framework in which "basis of open balls" makes sense.
- 1914 — Felix Hausdorff publishes Grundzüge der Mengenlehre, formalizing topological spaces via neighborhood systems, an ancestor of the modern basis concept.
- 1922 — Kazimierz Kuratowski axiomatizes topology via the closure operator, giving an equivalent route to the same open-set structures.
- 1930s — Bases and subbases become standard tools in the Bourbaki-influenced formalization of general topology, crystallizing the modern basis/subbasis vocabulary.
- 1940s–1950s — Textbook treatments (Kelley, and later Munkres) standardize the basis criterion as the entry point for constructing new topologies, including products and order topologies.
What exactly is the difference between a basis "for a topology" and a basis that "generates a topology"?

These are two views of the same object at different logical stages. If a topology T\mathcal T is already given, a basis for T\mathcal T is simply a subcollection B⊆T\mathcal B \subseteq \mathcal T such that every open set is a union of members of B\mathcal B — this is a relative, descriptive notion. Separately, if you start with only an abstract family B\mathcal B of subsets (no topology yet), and it satisfies the covering and intersection axioms, you can generate a topology TB\mathcal T_{\mathcal B} from it. The Basis Criterion proves these views coincide: the topology generated by B\mathcal B has B\mathcal B as a basis for it in the relative sense, and conversely, any basis for a topology satisfies the generative axioms.

Do finite intersections of basis elements always need to equal another single basis element, or just contain one?

Just contain one — this is a frequent point of confusion. The intersection axiom only requires that for each point xx in B1∩B2B_1 \cap B_2, there exists some B3∈BB_3 \in \mathcal B with x∈B3⊆B1∩B2x \in B_3 \subseteq B_1 \cap B_2; it does not require B3=B1∩B2B_3 = B_1 \cap B_2. In many familiar examples (open balls in a metric space) the intersection itself is not a ball, but every point of the intersection has a smaller ball around it contained in the intersection, which is exactly what the axiom demands.

Why do we need the intersection property at all — isn't covering enough?

No: covering alone only guarantees XX is open in the generated family, not that the family is closed under finite intersections. The counterexample in this article, with X={1,2,3}X = \{1,2,3\} and B={{1,2},{2,3}}\mathcal B = \{\{1,2\},\{2,3\}\}, covers XX perfectly but fails the intersection axiom, and the resulting "unions of basis elements" family is not closed under intersection — hence not a topology. The intersection property is precisely the condition that repairs this, guaranteeing overlaps of basis elements are themselves coverable by basis elements.

Is every subbasis-generated basis (finite intersections of S\mathcal S) automatically a basis in the generative sense?

Yes, and this is exactly what the corollary on subbasis generation proves. The key computation is that a finite intersection of finite intersections of S\mathcal S-elements is again a finite intersection of S\mathcal S-elements — this makes the intersection axiom hold with equality (B3=B1∩B2B_3 = B_1 \cap B_2 itself lies in B\mathcal B), which is even stronger than the basis criterion requires. Covering is arranged by hypothesis on S\mathcal S (or by convention that the empty intersection, n=0n=0, equals XX).

Can a topology fail to have a basis at all?

No — trivially, every topology T\mathcal T is its own basis, since every open set is (trivially) a union consisting of itself. The interesting question is not existence but size: whether a topology admits a small basis, e.g. a countable one (this is the property of second countability). Many important spaces, like Rn\mathbb{R}^n with the Euclidean topology, have a countable basis (rational-radius balls at rational-coordinate centers), which is a much stronger and more useful fact than mere existence of a basis.

How does the basis criterion interact with continuity of functions?

If f:X→Yf: X \to Y is a function and B\mathcal B is a basis for the topology on YY, then ff is continuous if and only if f−1(B)f^{-1}(B) is open in XX for every B∈BB \in \mathcal B. This follows because every open V⊆YV \subseteq Y is a union V=⋃iBiV = \bigcup_i B_i of basis elements, so f−1(V)=⋃if−1(Bi)f^{-1}(V) = \bigcup_i f^{-1}(B_i), a union of open sets (hence open) in XX provided each preimage of a basis element is open. This reduces the Continuity of Maps check to a much smaller test family, exactly analogous to the reduction bases give for describing open sets.

Does the choice of basis or subbasis affect properties like compactness or connectedness of the resulting space?

No — once the topology TB\mathcal T_{\mathcal B} or TS\mathcal T_{\mathcal S} is generated, all topological properties (compactness, connectedness, Hausdorffness, etc.) depend only on T\mathcal T itself, not on which generating family produced it. Different bases or subbases can generate the identical topology, and topological properties are properties of the topology, not of the choice of generators. The generators are a computational convenience for verifying membership in T\mathcal T, not additional structure on the space.

Before this

Where this leads