Analysis teaches that continuity is a statement about nearness, and that "nearness" can be phrased entirely through open sets. Topology keeps only that open-set structure. This chapter fixes the axioms, shows how to build a topology from a small family of generating sets, and proves that the open-set definition of continuity behaves as it should.
Motivation
In a metric space a function is continuous at when for each there is with . Equivalently, the preimage of each open ball around contains an open ball around . The metric has disappeared from the second formulation except through the notion of "open set". Many spaces that matter (quotients, infinite products, spaces of functions) have no natural metric, or have a natural topology that no metric produces. We therefore take "open set" as the primitive notion and ask which properties of open sets in are essential. Two turn out to suffice: arbitrary unions and finite intersections of open sets are open.
Definitions
Definition 1.1.1
(Topology, open and closed sets)
A topology on a set is a collection of subsets of such that
- and ;
- if is any family, then ;
- if , then .
The pair is a topological space; members of are open sets, and complements of open sets are closed sets. A set can be both open and closed, or neither.
Definition 1.1.2
(Neighborhoods and closure)
A neighborhood of is a set containing an open set that contains .
The closure of is the intersection of all closed sets containing ; it is the smallest closed set containing . A point lies in if and only if every neighborhood of meets : the complement is the union of all open sets disjoint from , and lies in it exactly when some open neighborhood of misses .
Comparison. If are topologies on , then is coarser and is finer.
Definition 1.1.3
(Basis and generated topology)
A collection of subsets of is a basis for a topology if (B1) every lies in some and (B2) whenever with , there is with . The topology generated by is the family of all such that for each there is with .
Definition 1.1.4
(Subspace topology)
For , the subspace topology is .
Definition 1.1.5
(Continuity and homeomorphism)
A function between topological spaces is continuous if is open in for every open . A continuous bijection whose inverse is also continuous is a homeomorphism.
Basic examples.
- Discrete topology on : every subset is open.
- Indiscrete topology: only and are open.
- Cofinite topology: is open iff or is finite.
- Sierpiński space with open sets .
- Standard topology on : generated by the open intervals .
Theorem/proposition
Theorem 1.1.6
(Basis criterion)
Let satisfy (B1) and (B2). Then the family generated by is a topology on , and it equals the collection of all unions of members of .
Theorem 1.1.7
(Continuity via a basis)
Let be a basis for the topology of and let be a function with a topological space. Then is continuous if and only if is open in for every .
Theorem 1.1.8
(Subspaces and restriction)
Let carry the subspace topology. Then is a topology on . The inclusion is continuous. Moreover a function is continuous if and only if is continuous.
Theorem 1.1.9
(Sequences are not enough)
In a metric space, a function is continuous if and only if whenever . In a general topological space, a continuous function preserves convergent sequences, but the converse can fail.
Proof sketch or proof idea
The proofs are given in full.
Proof
(Unions of basis elements form a topology)
Let be the family in the definition. First we show it equals the collection of unions of members of . If , choose for each a with ; then (the empty union gives ). Conversely, if and , then for some , so .
Now the axioms. The empty set is the empty union. By (B1), is a union of basis elements. A union of unions of basis elements is a union of basis elements, which gives axiom 2. For axiom 3 it suffices, by induction, to treat with . Let . Choose with and . By (B2) there is with . So .
Proof
(Testing continuity on basis elements)
If is continuous then each is open because is open (a basis element is a union of one basis element). Conversely let be open. By the basis criterion, with . Preimages commute with unions, so is a union of open sets, hence open.
Proof
(Subspace axioms and the restriction property)
We have and . For unions, ; for finite intersections, . So is a topology. For the inclusion, is open in . If is continuous then is a composite of continuous maps; compositions of continuous maps are continuous because . Conversely, suppose is continuous and let be open in , say with open in . Then , because takes values in . This is open in .
Proof
(Sequences under continuous maps)
Preservation. Let be continuous, (meaning each neighborhood of contains for all large ). If is an open set containing , then is an open set containing , hence contains for large ; thus for large and .
Metric converse. Suppose is not continuous at in a metric space. Then there is such that for every some has and . Then but . (This argument uses a countable family of nested neighborhoods, which a metric supplies.)
Failure in general. See the countable-complement counterexample below.
The explorer below lets you choose a family of subsets of a four-point set, computes the family it generates, and shows which of (B1), (B2) and the topology axioms fail, with the first failing witness.
Worked examples
Worked example 1.1.10
(Cofinite versus standard)
On , the cofinite topology is strictly coarser than the standard one. A nonempty cofinite-open set has the form with finite, and is a finite union of open intervals, so it is standard-open. But is standard-open while its complement is infinite, so it is not cofinite-open. Consequently the identity is continuous, since the preimage of a cofinite-open set is that same set, which is standard-open. The identity is not continuous: the preimage of the open set is , which is not cofinite-open.
Worked example 1.1.11
(Sierpiński space and continuity)
A function is continuous iff is open in (the preimages of and are and ). Hence continuous maps correspond bijectively to open subsets of , via ; given open , the indicator of is continuous. This explains why Sierpiński space is sometimes called the classifying space of open sets. Notice also that is closed but not open, and the closure of is , so closedness and openness are genuinely different here.
Counterexample mode
A sequence criterion that fails
Claim: a function that carries every convergent sequence to a sequence converging to the corresponding image is continuous. Let with the countable-complement topology ( open iff or is countable) and let with the discrete topology. Let . A sequence converges to in iff for all large . Indeed, if infinitely many terms differ from , the set has countable complement, so it is open, and it contains ; but it contains none of the terms different from , so the sequence is not eventually in . Therefore carries convergent sequences to convergent sequences, since eventually-constant sequences converge in the discrete topology. But is not continuous: is open in , while is not open in . This shows that sequences cannot detect continuity in this space; one needs nets or filters, or first-countability.
Worked example 1.1.12
(The subspace topology of )
For each , is open in the subspace topology, so the subspace topology on is discrete. The subspace shows a subtlety: is open in although it is not open in . Openness depends on the ambient space.
Concept lensStructural
An open set as a two-valued observation
The Sierpiński example gives a second view of an open set: it is exactly the region where a continuous two-valued observation returns . The asymmetry of the topology on matters. Only the inverse image of is required to be open; the inverse image of need not be. If the target were discrete, both inverse images would have to be open, and the observation would instead describe a clopen subset.
Thus the target's open sets specify which observations must be stable under small changes of input. This viewpoint explains why continuity uses inverse images: a permitted observable region in the target pulls back to a permitted region in the domain. It also makes the difference between a topology and a metric concrete. We retain which observations are locally valid, without recording a numerical distance between inputs. Testing the same function against a different target changes the continuity condition; one cannot judge continuity from the formula for the function alone while ignoring the topology of the target.
Continuity, closure and images
Continuity is defined using inverse images, but arguments often start with a set in the domain and its closure. The following equivalent test makes that change of viewpoint precise. The bars in the statement are taken in their respective spaces.
Theorem 1.1.13
(The closure criterion for continuity)
For topological spaces , a function is continuous if and only if, for every ,
Equality is not required and can fail even when is continuous.
Proof
(Closures pulled back through neighborhoods)
Suppose is continuous and . Every open neighborhood of has inverse image open and containing . It therefore meets , so meets . The neighborhood criterion for closure gives .
Conversely assume the inclusion for every . If is closed in , put . Then , and consequently . Thus , making closed. Preimages of closed sets being closed is equivalent, by complements, to continuity.
Proof X-ray
Follow a neighborhood backwards
The argument does not say that images of open neighborhoods are open. It pulls an arbitrary target neighborhood back to a domain neighborhood, where closure forces a point of . This one point supplies the required image in the original target neighborhood.
A map is open if the image of every open set is open in the target, and closed if the image of every closed set is closed in the target. These are separate properties from continuity. They do not require a bijection or even continuity. For a bijection, however, an open map has continuous inverse: the preimage of an open under is precisely . The same holds for a closed map using closed sets. Thus a continuous bijection is a homeomorphism if it is also open or closed.
Worked example 1.1.14
(A continuous map need be neither open nor closed)
Give the subspace topology from , and let be inclusion. It is continuous. The whole domain is both open and closed in itself, but its image is neither open nor closed in : zero has no real open neighborhood contained in it, while is in its closure but outside it. Hence is neither an open nor a closed map. The ambient space is essential: viewed as a map onto the subspace is a homeomorphism.
Worked example 1.1.15
(Why the closure inclusion can be strict)
Let be and let . Then , but
The endpoint is approached by images of points tending to infinity, although it is not the image of any point of . Continuity does not manufacture a domain limit for such an escaping sequence.
For the converse distinction, let be indiscrete and discrete. The identity is both open and closed: the only open or closed subsets of are and , whose images have both properties in . It is not continuous, since the inverse image of the open singleton is not open in .
Quick check
(Show that a continuous map is closed exactly when for every .)
Solution · Guided solution
If is closed, is a closed set containing , hence contains . The closure criterion gives the reverse inclusion. Conversely, if the equality holds and is closed, then , so is closed. Continuity supplies the first implication's reverse inclusion; it is part of the hypothesis.
First countability and the sequence test
In the metric proof above, the key property needed at the domain point is a countable local base. To allow an arbitrary target here, we replace the target metric ball by an open neighborhood. A space is first countable if every point has a countable local base: open neighborhoods such that every neighborhood of contains some . We may make the base decreasing by replacing with . For a metric space, the balls give such a base. First countability is weaker than having one countable base for the whole topology.
Theorem 1.1.16
(Sequential continuity on a first-countable domain)
Let be first countable and be any topological space. A function is continuous if and only if, for every sequence in , one has in . No separation or countability condition on is needed.
Proof
(Witnesses from a decreasing local base)
Continuity preserves convergent sequences, as shown above. For the converse, suppose is not continuous. Then some open has non-open preimage. Choose that is not an interior point of that preimage. Let be a decreasing countable local base at . Each meets , so choose in that intersection. Given any neighborhood of , some ; for , . Thus . Yet every lies outside the open neighborhood of , so cannot converge to . This contradicts the sequence hypothesis.
Proof X-ray
Where countability is used
The proof selects one witness from each neighborhood in a decreasing local base. Decreasingness guarantees that all sufficiently late witnesses lie in any prescribed neighborhood. Without a countable local base there may be no sequence that visits enough neighborhoods to expose the failure of continuity, as the countable-complement example above shows.
Worked example 1.1.17
(A sequence finds a discontinuity)
Define by and for . The sequence converges to , but does not converge to . The theorem therefore detects discontinuity at zero. Away from zero the function is locally constant, so this is its only discontinuity.
Worked example 1.1.18
(A non-Hausdorff target still works)
Let have open sets , and let be the indicator of . The sequence has image constantly , which does not converge to because the neighborhood contains none of its terms. Thus is discontinuous. A constant sequence at in converges both to and to , however: every neighborhood of is . Uniqueness of limits is not used in the theorem.
Quick check
(Let be first countable. Prove that is closed if and only if every convergent sequence of points of has each of its limits in .)
Solution · Guided solution
If is closed and converges to , then the open neighborhood of eventually contains , a contradiction. Conversely, if is not closed, choose and a decreasing countable local base at . Choose . The same local-base argument proves , contradicting the stated property. In non-Hausdorff spaces the phrase “each of its limits” matters.
Quick check
(Suppose is a bijection and both and are first countable. Show that is a homeomorphism exactly when both and preserve convergent sequences and their specified limits.)
Solution · Guided solution
Homeomorphisms and their inverses are continuous, so they preserve such limits. Conversely, apply the theorem to with domain and to with domain . Both maps are continuous, hence is a homeomorphism. First countability of is used for the inverse, not for the forward map.
Common mistakes
- "Open means not closed." Both can hold: and are always clopen; in the discrete topology every set is clopen. In the indiscrete topology on a set with at least two points, a nonempty proper subset is neither.
- Testing continuity on arbitrary sets by images. Continuity concerns preimages of open sets. Images of open sets need not be open: the constant map sends to a point.
Common mistake
(Requiring infinite intersections)
The axioms allow only finite intersections. In , is not open.
Common mistake
(Forgetting (B2))
Not every cover of is a basis. In the family satisfies (B1), but the point lies in the intersection and no member of the family fits inside around . Unions of its members are not closed under finite intersection, so they do not form a topology.
Common mistake
(Assuming sequences always suffice)
See the countable-complement counterexample; they suffice in first-countable spaces, in particular metric spaces.
Summary
A topology is a family of open sets closed under arbitrary unions and finite intersections. A basis satisfying (B1) and (B2) generates a topology consisting of exactly the unions of basis elements, and continuity need only be tested on basis elements. Subspaces inherit open sets by intersection and the subspace topology is characterized by the property that a map into it is continuous iff its composite with the inclusion is. Sequence criteria hold in metric spaces but fail in general.
Exercises
Quick check
(Show that the collection of all subsets such that or contains all but finitely many elements is the cofinite topology, and that the collection of all *finite* subsets together with and is not a topology.)
Solution · Answer
The first family consists of and sets with finite complement, which is the definition of cofinite. For the second family, let consist of the finite subsets of together with ( is finite). Each singleton lies in , but is the set of even numbers, which is infinite and different from , so it is not in . Axiom 2 fails.
Quick check
(Let in . Verify (B1) and (B2), and show the generated topology is strictly finer than the standard topology.)
Solution · Answer
(B1): . (B2): if then , and this interval is nonempty since it contains . Each open interval is a union of intervals with , so is open in the generated topology; hence the topology is finer. It is strictly finer because is open in it but not in the standard topology: no interval lies inside .
Quick check
(Let be open in . Prove that a subset of is open in (subspace topology) if and only if it is open in .)
Solution · Answer
If is open in then with open in , so is an intersection of two open sets of , hence open in . Conversely if is open in , then is open in .
Quick check
(Let be continuous and . Prove that the restriction is continuous.)
Solution · Answer
We have where is the inclusion, continuous by the subspace-and-restriction proposition. A composite of continuous maps is continuous.
Quick check
(Let be the set with topology . List all closed sets and compute the closure of .)
Solution · Answer
Open sets are , so closed sets are the complements . The closure of is the smallest closed set containing . Closed sets containing are and ; the smallest is .