Read bases, subspaces, connectedness and compactness first. A linear order is not assumed to carry a topology until the order topology is explicitly chosen. The lexicographic example uses ordinary real order completeness, not an unproved classification of ordered spaces.
Motivation
Sometimes continuity requirements are easier to specify than all open sets. We may want precisely the least topology making a family of measurements continuous. Subbases turn these tests into a topology by finite intersections and arbitrary unions.
Orders give another source of tests: whether a point is below or above a specified value. Yet retaining the same order inside a larger space does not automatically retain its subspace topology. We will distinguish these constructions and show how order density and least-upper-bound completeness together prevent topological gaps.
Definitions
Definition 1.4.1
(Subbasis and generated topology)
For a collection of subsets of , include as the intersection of the empty finite subcollection. The topology generated by the subbasis consists of arbitrary unions of finite intersections of . Including the empty intersection avoids requiring that itself cover ; the empty union is empty.
Definition 1.4.2
(Initial topology)
For maps into topological spaces, the initial topology on is generated by all for open in . It is the least topology making every continuous. No injectivity or point separation is assumed.
Definition 1.4.3
(Order topology, convexity and linear continua)
A nonempty linearly ordered has order topology generated by the rays and . Finite intersections give ordinary open intervals, rays at endpoints and the whole space. A subset is order convex if it contains every point between any two of its points. A linear continuum is a densely ordered set with the least-upper-bound property: between any two distinct points is a third, and every nonempty subset bounded above has a supremum in . Endpoints are allowed.
Theorem/proposition
Theorem 1.4.4
(Subbasis and initial universal property)
The generated collection is a topology and is contained in every topology containing . A map into it is continuous exactly when the preimage of every subbasis member is open. Consequently is continuous for the initial topology exactly when every composite after is continuous.
Concept lensAnalytic
Initial topology as convergence of every measurement
In the initial topology a sequence converges to exactly when for every . One direction is continuity of each . Conversely, every neighborhood of contains a finite intersection containing , and each of these finitely many conditions holds eventually, hence all of them together. The same argument works for nets, which determine the topology; for a product it is coordinatewise convergence.
Theorem 1.4.5
(Connected ordered spaces)
Every order topology is Hausdorff. Such a space is connected exactly when its order is a linear continuum. Every order-convex subset of a linear continuum is connected with its subspace topology. For a nonconvex subset, its induced order topology may differ from its ambient subspace topology.
Theorem 1.4.6
(Compact linear continua with endpoints)
A linear continuum is compact in its order topology exactly when it has a least and greatest element. No separability or countable basis is required. In particular connected compact ordered spaces can behave quite differently from metric intervals.
Proof sketch or proof idea
Proof
(Generated topologies and initial maps)
Arbitrary unions of generated unions remain generated unions. For intersections, distribute two unions of finite intersections; each resulting intersection is still finite, giving closure under binary and hence finite intersections. and empty are included by the conventions. Every topology containing must contain these operations, proving minimality. Preimages commute with unions and finite intersections, so checking continuity on suffices. For initial maps, the test becomes openness of , exactly the continuity of all composites.
This also recovers product topology as the initial topology of coordinate projections. If the measurement family separates points and all are Hausdorff, the initial is Hausdorff: distinct points have differing values for one , and disjoint target neighborhoods pull back to disjoint neighborhoods. The map is then an embedding into the product. It is injective by separation, and its inverse onto the image is continuous because the subbasis tests on are precisely restrictions of product-coordinate tests. Without point separation the initial topology can identify the neighborhood behavior of different points.
Proof
(Hausdorff order topologies and connectedness)
For ordered Hausdorffness let . If there is strictly between, the rays below and above separate them. If there is no such , the rays below and above are disjoint neighborhoods of . Thus adjacent elements cause no separation failure, although they do obstruct connectedness.
Indeed an adjacent pair separates all of by those same rays. Hence a connected ordered space must be densely ordered. Suppose a nonempty bounded-above set has no supremum. Its set of upper bounds is nonempty and has no least element. Its complement is , open and nonempty, since cannot have a greatest element. is open too: for each in , choose a smaller upper bound , and the ray above lies in and contains . This partitions into nonempty open , contradiction. Thus connectedness forces order completeness.
Conversely let be a linear continuum with a supposed separation . Choose points in its two sides and label them , with in and in . Let be the supremum of , which exists and lies in . If lies in , it is not . Openness and order density then give a point of strictly between and sufficiently near , contradicting the supremum. If lies in , it is not . Openness gives with ; take at least if necessary. The supremum property supplies a point of above , contradicting . Both cases fail, proving connectedness. The required one-sided points exist because a basic order neighborhood contains a small interval on the needed side, and density puts a point in that interval.
For a convex subset , its order rays are intersections of ambient rays, so its order topology is no finer than the subspace topology. Conversely, intersect an ambient ray with . If its endpoint belongs to this is a relative order ray. If not, convexity prevents from having points on both sides of that missing endpoint, so the intersection is either all of or empty. Thus the topologies agree. A nonempty convex subset of a linear continuum is itself densely ordered and order complete for subsets bounded above within it: an ambient supremum lies between one member and a given upper bound, hence stays in . It is therefore connected by the theorem just proved.
Proof
(Compactness of linear continua with endpoints)
For compactness let be the endpoints and take an open cover. A singleton space is immediate. Otherwise let contain those for which has a finite subcover, and let . The set is nonempty because is covered. If , an open set containing contains for some by density and the order-neighborhood definition; then belongs to , contradiction. Hence .
Choose a cover member containing . Some has . The supremum property gives in above . If already belongs to there is nothing to prove at this step; otherwise and a finite cover of , together with , covers . Thus in either case belongs to . If , openness and density give with contained in . Its addition to a finite cover of puts in , contradiction. Therefore and the cover has a finite subcover of the whole space. This proof makes no countability assumption about the order or cover.
Conversely, any nonempty ordered space that has no least element is covered by all rays above , for in . A finite subfamily has union the ray above its least selected endpoint and misses that endpoint, so it cannot cover. The reversed argument excludes lack of a greatest element in a compact ordered space. This proves endpoint necessity without needing the linear-continuum assumptions.
Proof X-ray
One supremum sweep, three hypotheses
The connectedness and compactness proofs run the same sweep. Take the supremum of the points already reached, either the points of in or the points for which is finitely covered, and show that the sweep can neither stop before nor stop at before the end. Completeness is used only to produce inside ; the rational order shows what happens without it. Density is used to find a point strictly between and a nearby bound, so that an open interval around contains new points of the set; an adjacent pair is exactly where that step would fail. Openness of and , or of the cover members, turns membership of into a whole interval or .
In the compactness proof the least element starts the sweep and the greatest element is where it is allowed to stop. Without endpoints, the cover by rays in the converse argument has no finite subcover, so the endpoint hypothesis is necessary as well as sufficient.
Worked examples
Worked example 1.4.7
(The same inherited order can have two different topologies)
Let inside . Its ambient subspace topology isolates zero. In its induced order, zero is least but has no immediate successor; every order neighborhood of zero contains some initial portion of , so zero is not isolated. The map sending zero to one and every other point to itself is an order isomorphism onto [), hence a homeomorphism for the order topologies by their ray subbases. Thus the induced order topology is connected, while the ambient subspace topology is disconnected. Nonconvexity of is exactly why the two constructions need not agree.
Counterexample mode
Dense order without completeness: the rationals
Claim: a densely ordered set is connected in its order topology. The rationals are densely ordered, since lies between , and their order topology is the usual subspace topology, because rational rays and intervals are the intersections of real ones with rational endpoints. Put
No rational has square two, so and are disjoint and cover , with and . For positive , the rational satisfies and . Hence has no greatest element (use , or 1 when ) and has no least element. It follows that is the union of the rays below its members and the union of the rays above its members, so both are open, and is disconnected. The bounded set has no rational supremum: the missing hypothesis is exactly the least-upper-bound property of a linear continuum.
Worked example 1.4.8
(The lexicographic square is compact and connected but not separable)
Order lexicographically, comparing first coordinates before second coordinates. It is densely ordered: use an intermediate second coordinate in one fiber, or an intermediate first coordinate between distinct fibers. For a nonempty subset , let be the real supremum of its first-coordinate projection. If has elements in fiber , let be the supremum of their second coordinates; otherwise put . Then is its supremum. It bounds , and any smaller pair is exceeded either by a point with larger first coordinate, or by one in fiber with second coordinate closer to . In the empty-fiber case leaves no smaller second-coordinate pair in that fiber. Hence is a linear continuum with endpoints , so compact and connected.
For each , the order interval between and is a nonempty open vertical segment. These uncountably many segments are disjoint. A countable proposed dense set has only countably many first coordinates; choosing another yields one segment it misses. Thus is not separable and in particular is not a compact metric space. Its topology is not the ordinary product topology on the square, which has a countable dense set.
Worked example 1.4.9
(Initial topology can be much weaker than a desired metric)
On , take the initial topology of all evaluation maps to . Its neighborhood tests restrict only finitely many sampled values, giving pointwise convergence. The shrinking spikes from the Ascoli chapter converge to zero at every sample and therefore in this topology, but their uniform distance from zero stays one. Thus the initial pointwise topology differs from the uniform metric topology. Specifying which measurements must be continuous does not automatically specify every stronger convergence requirement.
Common mistakes
Common mistake
(A subbasis need not be closed under intersections)
A subbasis is not required to be closed under finite intersections. Those intersections are generated first; arbitrary unions are taken afterwards. A basis check and a subbasis continuity check are different statements.
Common mistake
(Order topology uses open rays)
The order topology uses open rays. Replacing them with half-open intervals specifies another topology and needs a separate analysis. Merely knowing a linear order does not mandate all its familiar Euclidean properties.
Common mistake
(Inherited order and inherited topology can differ)
For nonconvex subspaces the inherited order and inherited topology are distinct operations. The missing ambient interval near zero in the first example becomes no order gap once zero serves as the new endpoint.
Summary
Subbases generate the least topology satisfying given open-set tests, yielding initial universal properties and product embeddings. Order-ray tests make linear orders Hausdorff. Dense order and least-upper-bound completeness together characterize connectedness, and endpoints make linear continua compact, even without countable metric behavior.
Exercises
Quick check
(If all are constant, compute their initial topology. If they separate points and the targets are Hausdorff, prove the initial space is Hausdorff.)
Solution · Answer
Every open target set pulls back to or empty under a constant map, so the generated topology is indiscrete. For separated points choose one with different values, disjoint target neighborhoods and their inverse images. These are disjoint initial-open neighborhoods, giving Hausdorffness. Point separation, not mere existence of many measurements, is the extra condition.
Quick check
(In the first example, describe an order-basic neighborhood of zero and a subspace-basic neighborhood isolating it. Why is the order-isomorphism to [) not an ambient subspace homeomorphism?)
Solution · Answer
An initial ray below any in contains zero and ; no such ray isolates zero. In the real subspace, intersection with gives . Under the order isomorphism that singleton maps to , not open in [). Thus the same bijection fails continuity of its inverse when the source is assigned the stronger ambient subspace topology.
Quick check
(Prove a convex subset of a linear continuum has the least-upper-bound property for its subsets bounded above within it. Where would nonconvexity break the proof?)
Solution · Answer
If is nonempty in and bounded above by in , choose in . Its ambient supremum lies in . Convexity puts that whole interval, hence , in . It is still the least upper bound among -elements. Without convexity the supremum could be an omitted ambient point between retained endpoints, so this containment argument fails.
Quick check
(Give a linear continuum lacking an endpoint that is not compact, with an explicit open cover lacking finite subcover. Prove any nonempty compact ordered space must have both endpoints.)
Solution · Answer
In , the relative open sets , at least two, cover the space but no finite subfamily covers points sufficiently near zero. If a nonempty ordered space has no least element, the rays above all possible endpoints cover it, but any finite such union misses its least selected endpoint. Thus compactness forces a least element. Reversing the order forces a greatest. This necessity holds even beyond linear continua.
Quick check
(Prove an order isomorphism is a homeomorphism for order topologies. Then show a strictly increasing real function need not be continuous by giving a jump example.)
Solution · Answer
The inverse image of a ray under an order isomorphism is the corresponding ray at the inverse endpoint. The subbasis test proves continuity, and the same argument applies to its inverse. For a strictly increasing map to , take for and for at least zero. The preimage of is [), which is not open. Strict order preservation without surjectivity onto the codomain order does not supply the order-isomorphism argument.