Motivation
A path often has different formulas on different parts of its interval. To prove it continuous, checking each formula separately is necessary, but it is not always enough. We must also know how the pieces sit inside the whole domain. This section makes that distinction precise and gives a useful extension from finite closed covers to locally finite ones.
Definitions
Agreement and the subspace topology
Definition 1.2.1
(Pasted function)
Let , and suppose that each is continuous, where has the subspace topology. If on every overlap , there is a unique function whose restriction to is . Agreement is what makes the definition unambiguous; it does not by itself guarantee continuity.
Recall that a subset of is open in precisely when it is for some open . A similar statement holds for closed sets. Relative openness alone is weaker than openness in .
What a local continuity check must show
Continuity of the restrictions means that, for an open target set , each inverse image is open inside . To prove continuity of the whole map, we need the union of these inverse images to be open in . For a single piece, relative openness does not provide that conclusion: is open in the one-point space but is not open in . The position of a piece in the ambient space matters.
There is an equivalent pointwise way to see the open-cover proof. Given and an open neighborhood of , choose a piece containing . Continuity of gives an open set in with . If is itself open in , then is an open neighborhood of on which the pasted map takes values in . This argument needs only one suitable open piece at each point; it does not need a finite cover or a metric.
By contrast, if is closed, need not be a neighborhood of a boundary point. The closed-cover argument instead tests preimages of closed target sets. Being closed in a closed subspace makes each such preimage closed in , and a finite union remains closed. These two routes use different closure properties: arbitrary unions of open sets are open, whereas arbitrary unions of closed sets can fail to be closed.
Theorem 1.2.2
(Open pasting)
If the are open in , the agreed function is continuous, even for an infinite cover.
Proof
(Unions of open preimages)
For open , each is open in and therefore open in , because is open. Hence is open.
Theorem 1.2.3
(Finite closed pasting)
If , each is closed, and the continuous maps agree on overlaps, then the resulting map is continuous.
Proof
(Finite unions of closed preimages)
For every closed , the set is closed in , hence closed in because is closed. Now is closed as a finite union. Thus preimages of closed sets are closed.
Worked example 1.2.4
(Concatenating paths)
Let be continuous and . Define on and on . The restrictions are continuous, the two closed intervals cover , and both values at equal . Closed pasting proves continuity. Without the endpoint equality, the formula is not even well defined at the join.
Infinite covers and local finiteness
Definition 1.2.5
(Locally finite family)
A family is locally finite if every point of has an open neighborhood meeting only finitely many . This concerns the family, not the number of points in each member. Finite families are locally finite, but some infinite families are too.
Lemma 1.2.6
(Locally finite unions of closed sets)
A locally finite union of closed subsets of is closed.
Proof
(Remove finitely many closed sets)
If is outside the union, choose an open neighborhood meeting only . Then is an open neighborhood of disjoint from every member. Thus the complement of the union is open. The case requires no subtraction.
Proof X-ray
Finitely many obstructions near each point
Notice the order of the proof. For a point outside the union, local finiteness first limits the possible obstructions near to finitely many closed sets. Their union is closed, so removing it from the neighborhood still leaves an open set containing . Every other member already misses that neighborhood, so the new neighborhood avoids the whole union. We do not remove infinitely many closed sets at once: the complement of an infinite union is an infinite intersection of open sets, which need not be open.
The lemma is a neighborhood statement, not an assertion that only finitely many members exist. For instance, the closed sets , , are locally finite in : any bounded interval meets only finitely many of them. Their union is therefore closed. The closed sets behave differently. Every neighborhood of meets infinitely many, and their union is not closed because is a limit point outside it.
Common mistake
(Belonging to finitely many pieces is not local finiteness)
Do not confuse local finiteness with the weaker assertion that each point belongs to only finitely many pieces. The family has that weaker property: every point belongs to at most one member. Yet it is not locally finite at , because every neighborhood of meets infinitely many members. The distinction concerns nearby pieces, not just pieces containing the point.
Theorem 1.2.7
(Locally finite closed pasting)
If a locally finite family of closed subsets covers , and the continuous agree on overlaps, then the pasted function is continuous.
Proof
(Closed preimages form a locally finite family)
For closed , the sets are closed in and form a locally finite family because each is contained in . Their union is , which is closed by the lemma. No separation condition on or is needed.
Concept lensStructural
Pasting as a property of the cover
Say is coherent with a cover when is open as soon as every is open in . Coherence gives pasting for every target, since . Conversely, test the indicator map of into with open sets : its restrictions are continuous exactly when each is relatively open. Open covers and locally finite closed covers are coherent; the singleton cover of is not.
Worked example 1.2.8
(A function with infinitely many pieces)
For each integer , let and put . At a common endpoint , both neighboring formulas give . Each is continuous, and every real point has a small neighborhood meeting only finitely many intervals . The pasted function is therefore continuous, despite having infinitely many pieces. Local finiteness supplies the missing justification; it is not necessary to find a finite cover of the entire line.
Worked example 1.2.9
(An accumulation point needs a separate argument)
For let , and let . Together these closed sets cover . Put and, on every , put . Each restriction is continuous: the denominator is nonzero on . Adjacent intervals share one endpoint and use the same formula there, while has no intersection with any . Thus the pieces agree and define a function.
The family is locally finite at every : a neighborhood bounded away from can meet only finitely many intervals. It is not locally finite at , because every neighborhood of meets infinitely many . We cannot invoke locally finite closed pasting on the whole interval. Nevertheless the function is continuous at , since
Given , the choice works. At a point the ordinary formula is continuous on in the subspace topology; alternatively, a small neighborhood meets finitely many pieces and finite closed pasting applies there. Hence is continuous everywhere. This example shows that local finiteness is sufficient for the theorem, not necessary for every individual pasted function.
Now keep the same cover but set every interval function equal to and the function on equal to . The restrictions are continuous and still agree on all overlaps. Yet for every while , so continuity at fails; directly, the inverse image of the open set is , which is not open in . Agreement controls shared endpoints; it does not control values along pieces that accumulate at a point outside every one of them.
A practical checklist
First identify the topology on each piece and verify continuity as a map from that subspace. Next check agreement on all overlaps; for disjoint pieces agreement is automatic but tells us nothing about their limiting behavior. Then ask whether the cover is open, finite closed, or locally finite closed. Any one of these hypotheses gives a suitable pasting theorem. If none applies, the map is not automatically discontinuous: check the troublesome points directly, as at above.
Common mistake
(Two misreadings of the pasting hypotheses)
For an open cover, a common error is to replace “the pieces are open in ” with “each inverse image is open in its piece.” The latter is merely the definition of continuity of a restriction and also holds in the singleton counterexample below. For a closed cover, another error is to infer that infinitely many closed preimages have closed union. Local finiteness provides precisely the extra neighborhood control used in the lemma.
Closures, compact sets and locally finite sums
Local finiteness is used far beyond the pasting theorem. Three further facts recur whenever a construction is assembled from infinitely many local pieces, for example in partitions of unity.
Lemma 1.2.10
(Closures of a locally finite family)
If is locally finite in , then so is , and
Proof
(An open set meets a set exactly when it meets its closure)
If is open and , pick in the intersection. Then is a neighborhood of a closure point of , so . The converse is clear. Hence an open neighborhood meeting only finitely many also meets only finitely many , which proves the first claim.
For the equality, each is contained in , so the right side is always contained in the left. Conversely, the family is a locally finite family of closed sets, so its union is closed by the lemma above. That closed set contains every , hence contains the closure of their union.
Without local finiteness the equality fails: for in , every equals , but the closure of the union also contains .
Proposition 1.2.11
(Compact sets meet finitely many members)
If is locally finite in and is compact, then only finitely many meet . In particular, a locally finite family of nonempty subsets of a compact space is finite.
Proof
(Cover the compact set by good neighborhoods)
For each choose an open neighborhood meeting only finitely many members. The sets cover , so finitely many of them, say , already cover it. A member meeting meets some , and each meets only finitely many members. A finite union of finite sets of indices is finite. For the second claim take ; every nonempty member meets .
This explains the earlier examples. The closed unit interval is compact, so the infinite family cannot be locally finite in ; the failure has to occur somewhere, and it occurs at . On the other hand is infinite and locally finite in , which is not compact.
For a function , its support is .
Proposition 1.2.12
(Locally finite sums are continuous)
Let , , be continuous, and suppose the family is locally finite. Then at every only finitely many are nonzero, and is a continuous function on .
Proof
(On each good neighborhood the sum is finite)
Each has an open neighborhood meeting only . Every other vanishes on , so on the sum equals , a finite sum of continuous functions; for it is zero. Thus the restriction of to each such open set is continuous. These open sets cover , and open pasting gives the continuity of .
By the closure lemma, it suffices to know that the sets themselves form a locally finite family. What does not suffice is that each point lies in only finitely many supports. On let be the tent function supported on , zero at both endpoints and equal to at the midpoint. Every point lies in at most two supports, and the sum is defined everywhere, with . Yet equals at every midpoint, and these midpoints converge to , so is discontinuous at . The supports are not locally finite at . The finite partitions of unity in the unit on Tietze extension and uniform approximation are the special case of a finite family; the proposition is what allows infinitely many cutoffs once their supports are locally finite.
Common mistakes
Why hypotheses matter
Counterexample mode
Arbitrary closed covers do not paste
Claim: if closed pieces cover the space, the restrictions are continuous, and they agree on overlaps, then the pasted map is continuous. Cover by its singleton subsets, all closed. Define and for . Every restriction to a singleton is continuous and the overlap condition is automatic, but is discontinuous at : the inverse image of is , not open in . The family of singletons is not locally finite at . Thus one cannot replace “finite closed” by “arbitrary closed” in the theorem. The repair is to require the closed family to be finite or, more generally, locally finite.
An open cover and a closed cover are two different sufficient mechanisms. A piece need not be open and closed at once. Conversely, a particular pasted function can be continuous even if its cover satisfies neither hypothesis; the theorems give sufficient conditions, not a characterization of all continuous maps.
Exercises
Exercises with solutions
Quick check
(Let with closed. A continuous is zero on . Show that its extension by zero to is continuous.)
Solution · Answer
The zero map on is continuous and agrees with on the intersection. Finite closed pasting applies.
Quick check
(Is locally finite in ? Is locally finite there?)
Solution · Answer
Yes for the first: a bounded neighborhood meets only finitely many such intervals. No for the second: every neighborhood of contains infinitely many , although none of those sets contains .
Quick check
(Let with open. Suppose continuous functions on and agree on their intersection. Prove that their common function is continuous at a point of .)
Solution · Answer
For an open target set containing the common value, the inverse image under the restriction to is open in , hence open in , and is a neighborhood of the point contained in the global inverse image of . The same reasoning applies at every point of the cover.
Quick check
(Define by for and for . Prove continuity and compute .)
Solution · Answer
Both restrictions are continuous on closed intervals and agree at , so is continuous. Solving the two equations gives .
Quick check
(Let for . Show directly that every is closed but is not closed. Explain why this does not contradict the locally finite union lemma.)
Solution · Answer
Each singleton is closed in . The sequence converges to , which is not among the , so the union does not contain all its limit points and is not closed. Every neighborhood of contains infinitely many of its members; thus the family fails the local finiteness hypothesis exactly where the missing limit occurs.
Quick check
(Cover by and . Define and on their respective domains. Explain why they determine a continuous function using open pasting, including what happens on the overlap.)
Solution · Answer
Both restrictions are continuous and agree on , so they determine the function on the whole line. Both and are open and cover , so the open pasting theorem applies. In particular, a point in the overlap may be checked using either restriction; agreement ensures the two choices give the same values.
Quick check
(On for , consider the restrictions . Do they paste to a function? What changes if every restriction is instead ?)
Solution · Answer
The first family does not define a function: at the shared endpoint , but . The second family agrees on every overlap. Its pieces are closed and locally finite, so closed pasting proves continuity. The resulting function is the identity. This comparison separates the work of agreement, which gives a function, from the work of a pasting theorem, which gives continuity.
Quick check
(Let and be continuous with for all . Show that the map equal to on and to on is continuous.)
Solution · Answer
The sets and are closed in : each is the preimage of a closed interval under the continuous projection to . They cover the product and intersect in , where the two maps agree. Finite closed pasting proves continuity. This is the step used to compose homotopies one after the other.
Quick check
(Let be locally finite in and let carry the subspace topology. Show that is locally finite in . Is the family locally finite in ? In ?)
Solution · Answer
For , an open in meeting only finitely many gives the relatively open , which meets only if meets . For the intervals: in every point has the neighborhood , which meets only the intervals with , finitely many. In the family is not locally finite at ; this is forced, because is compact and the family is infinite and consists of nonempty sets.
Quick check
(For , , compute and in . Which hypothesis of the closure lemma fails?)
Solution · Answer
Each is closed, so , while . The family is not locally finite at : every neighborhood of contains for all large and so meets all with large index. Here the members are even nested, so each point of belongs to infinitely many of them.