The useful feature of a distance is that it lets us discuss approximation before we know what the limit is. A sequence of approximate answers may become internally consistent, remain confined to a small region, or admit a convergent subsequence. These are different achievements. Completeness, total boundedness and compactness express their precise relationship.
Motivation
For a sequence of real numbers, being Cauchy is enough to guarantee convergence. The same statement fails in the rational numbers: rational decimal approximations to become arbitrarily close to one another without approaching any rational limit. The distance calculation succeeds; the proposed space does not contain the answer.
A different obstruction appears in an infinite discrete space. Give distinct points distance one. Every Cauchy sequence is eventually constant, so no limits are missing. Nevertheless, an infinite list of distinct points has no convergent subsequence. Completeness does not force a large collection of points to cluster.
The missing condition is finite control at every scale. A bounded interval can be covered by finitely many balls of any prescribed positive radius. An infinite discrete space cannot be covered by finitely many balls of radius one half. The compactness theorem below combines this finite approximation property with completeness. Baire's theorem then uses completeness in another way: to construct a point satisfying countably many dense open conditions simultaneously.
Definitions
Definition 1.1.1
(Metric and relative balls)
A metric on a set is a function such that exactly when , , and . Write . All balls, open sets and closures in this note are relative to unless a larger ambient space is specified.
A sequence converges to if, for every , eventually . It is Cauchy if, for every , there is such that whenever . Convergence names a limit; the Cauchy condition compares only terms of the sequence. The triangle inequality proves that every convergent sequence is Cauchy and that a limit is unique.
Definition 1.1.2
(Completeness, total boundedness and compactness)
A metric space is complete if every Cauchy sequence converges in it. It is totally bounded if, for every , finitely many open -balls cover it. It is compact if every open cover has a finite subcover. A set of centers giving a finite -ball cover is called a finite -net.
Total boundedness implies boundedness. If is empty, it is bounded vacuously. Otherwise choose a finite cover by open balls of radius one, with centers in . For any in the th ball, . Thus one ball about contains . The converse requires special structure, such as finite-dimensional Euclidean space. Do not use it as a general metric-space fact.
A subset is closed if its complement is open. In a metric space this is equivalent to containing the limit of every convergent sequence from the subset. A subset is dense if every nonempty open ball meets . It is nowhere dense if the interior of its closure is empty. A meager set is a countable union of nowhere dense sets. These definitions concern the surrounding space: the rationals are dense and meager in , while is not meager in itself.
Interior, closure and boundary depend on the space
Before using these words in a compactness argument, fix the ambient metric space and a subset . An open ball is always a ball in , with positive radius and centered at the candidate point unless stated otherwise. A set is open when every has some ball contained in . The interior consists of points that admit such a ball contained in . The closure consists of points whose every ball meets . The boundary consists of points whose every ball meets both and . These definitions still apply when is empty or all of .
Why distinguish them? An interior point has a margin: a sufficiently small perturbation stays in . A closure point can be approximated arbitrarily closely by points of , even when it does not itself belong to . At a boundary point no ball decides membership in either direction. Directly from the definitions,
For the first boundary identity, a closure point is not interior precisely when every ball also meets the complement. In particular is open exactly when , and closed exactly when . To see the latter, the complement of is open: if misses , then consists of points with a smaller ball missing . Thus is closed; if is closed, any point outside it has a ball in its open complement, so cannot lie in .
The interior is the largest open set contained in : if and , then a small ball around remains inside ; every open subset of is contained in by definition. Likewise, the closure is the smallest closed set containing : if a closed contains , a point outside has a ball disjoint from , and hence lies outside .
A boundary need not be a thin set. For in , every real interval meets both the rationals and irrationals. Therefore , and ; this boundary has nonempty interior.
From balls to sequences
Proposition 1.1.3
(Sequential test for closure)
For , the point belongs to if and only if there is a sequence in with . Consequently is closed if and only if every sequence in that converges in has its limit in .
Proof
(Proof of the sequential test)
If , for each choose . Then , so . Conversely, if , every ball about contains a sufficiently late , so meets . The closed-set statement follows from . The limit must lie in the stated ambient space: a sequence may converge in a larger space without having a limit in .
An accumulation point of is a point for which every ball around meets . It need not belong to . Equivalently, there is a sequence of distinct points of converging to . To construct one, choose successively within while avoiding and all earlier choices: if earlier choices exist, their distances from have a positive minimum, so take the radius smaller than both half that minimum and ; for the first choice use radius . Conversely, convergence of a sequence of distinct points implies that every ball contains a term different from .
The difference from closure matters. For in , zero is in but is not an accumulation point: the constant sequence witnesses closure, whereas the punctured-ball test fails. For , zero is an accumulation point outside and each is isolated in . In general, is the union of and its accumulation points; a set is closed exactly when it contains all its accumulation points.
The ambient space can change the answer
Worked example 1.1.4
(One set, two ambient spaces)
Take . In , its interior is , its closure is , and its boundary is . Now regard the same as a subset of with the restricted metric. The ball in centered at of radius is , so is interior relative to . Thus the relative interior is , relative closure is , and relative boundary is . The metric did not change on pairs of points of ; the available nearby points did.
More generally, if and , then . Indeed , so either ball meets exactly when the other does. Do not apply the same intersection formula blindly to interiors: in the example above, .
Quick check
(In , find the interior, closure, boundary and accumulation points of . Which answers change if the ambient space is itself?)
Solution · Interior, closure and ambient space
In , the interior is and the closure is . The boundary is : the isolated point is in but no real ball about it lies in . The accumulation points are ; a sufficiently small ball about contains no other point of . In the ambient space , the interior and closure are both and the boundary is empty. Its accumulation points are : the endpoints are not points of , and is still isolated. The distinction between boundary and accumulation points is therefore essential.
Theorem/proposition
Proposition 1.1.5
(Closed subspaces and complete ambient spaces)
A closed subset of a complete metric space is complete with the restricted metric. Conversely, a complete subspace of any metric space is closed in that ambient space.
Theorem 1.1.6
(Metric compactness criterion)
For a metric space , the following are equivalent: is compact; every sequence in has a convergent subsequence with limit in ; is complete and totally bounded.
Theorem 1.1.7
(Baire category theorem)
If is a nonempty complete metric space and are open dense subsets of , then is dense in . Consequently, a nonempty complete metric space cannot be a countable union of nowhere dense subsets of itself.
The second conclusion follows by taking complements of closures. Notice the word “itself”: a complete line in the plane has empty interior in the plane and hence is nowhere dense there. Baire's theorem does not forbid that.
Proof sketch or proof idea
Proof
(Proof of the subspace proposition)
For the first proposition, a Cauchy sequence in a closed subset converges in the complete ambient space, and closedness keeps the limit inside. Conversely, if is in the ambient closure of a complete subspace , choose with . The sequence is Cauchy, hence converges to some . Uniqueness of limits gives .
Proof
(Proof of the metric compactness criterion)
Here is the main mechanism of the compactness criterion. Total boundedness lets us place infinitely many terms of any sequence inside one ball of radius . Cover that infinite set by finitely many balls of radius and keep an infinite part in one of them. Continue and choose indices in increasing order from the nested infinite index sets. Two sufficiently late selected terms lie in the same ball of radius , so their distance is less than . The selected subsequence is Cauchy. Completeness turns it into a convergent subsequence.
To pass from subsequences to open covers, first observe that sequential compactness implies total boundedness. Otherwise, for some , repeatedly choose a point outside the -balls around all earlier points. The resulting sequence has pairwise distances at least and cannot have a convergent subsequence.
Next, any open cover of a sequentially compact metric space has a Lebesgue number: some such that each ball is contained in a cover member. If not, choose for which is contained in no member. A subsequence converges to . Some cover member contains , and eventually along that subsequence, a contradiction. A finite -net now yields a finite subcover.
Finally, compactness implies sequential compactness directly. If a sequence had no convergent subsequence, each point would have a neighborhood containing only finitely many sequence indices. Indeed, infinitely many indices in every neighborhood would let us select a subsequence converging to that point. These neighborhoods cover , and a finite subcover would contain only finitely many indices in total, an impossibility. A Cauchy sequence with a convergent subsequence converges to the same limit, proving completeness as well.
Proof
(Baire's nested-ball construction)
Take any nonempty open set . Choose a point in and a positive radius so that its closed ball lies in . This is possible because the intersection is nonempty and open; choose the closed-ball radius strictly smaller than an available open-ball radius.
Inductively choose a closed ball of positive radius at most inside , where is the center of . Density supplies a point of intersection; openness supplies the smaller ball. The centers form a Cauchy sequence because every later center lies in , whose diameter is at most . Completeness gives a limit . Closedness and nesting imply for every , hence . Since was arbitrary, the intersection is dense.
Compactness was never used: closed balls need not be compact. The construction requires their radii to tend to zero and the ambient space to be complete.
Proof X-ray
Two selection arguments, two uses of completeness
In the compactness criterion, finite ball covers select a Cauchy subsequence; completeness is used only after that selection. In Baire's theorem, density selects the next center, openness leaves room for a smaller closed ball, and completeness supplies a point in all the nested balls. The balls need not be compact. Keeping these roles separate prevents an unnoticed use of the compactness conclusion inside its own proof.
Finite products: coordinatewise control
Analysis often treats a pair of objects as one point: an initial time and state, or two functions in a coupled equation. For finitely many metric spaces, we can measure the largest coordinate error. This makes coordinatewise arguments precise without assuming that the factors are Euclidean.
Let be nonempty metric spaces for , where is a positive integer. On , put
This is a metric: positivity and symmetry follow coordinatewise, and for each , . Taking the maximum proves the triangle inequality. A ball has the explicit form
A sequence converges to in this metric exactly when each coordinate converges to . One direction follows from . For the other, given , choose an index for each coordinate and take . This last step uses the finiteness of the number of coordinates. The same reasoning shows that a sequence in is Cauchy exactly when each of its coordinate sequences is Cauchy.
Theorem 1.1.8
(Completeness and compactness of finite products)
For nonempty metric spaces with the maximum metric above, the product is complete if and only if every factor is complete, and compact if and only if every factor is compact.
Proof
(Proof for finite products)
If every factor is complete, the coordinates of any Cauchy sequence in converge to points ; coordinatewise convergence makes the original sequence converge to . Conversely, fix a coordinate and choose one point in each other factor. Insert any Cauchy sequence of into coordinate , keeping the other coordinates fixed. The resulting sequence is Cauchy in . If is complete, its limit supplies a limit in .
If every factor is compact, take any sequence in the product. First select a subsequence whose first coordinates converge; from it select one whose second coordinates converge, and continue through the finitely many factors. Every later subsequence retains the earlier coordinate limits. The final subsequence converges in the maximum metric, so the product is sequentially compact and hence compact by the metric compactness criterion above. Conversely, each coordinate projection is continuous because . It is onto because all other factors are nonempty. The continuous image of a compact space is compact: pull an open cover back to the domain, choose a finite subcover, then push the selected sets forward. Hence every factor is compact.
Proof X-ray
Where finite means one common index
For each coordinate, a separate estimate eventually holds; finiteness lets us take the largest of finitely many starting indices. Likewise, successive subsequence selection ends after steps. Neither argument automatically works for infinitely many coordinates under an arbitrarily chosen product metric.
Worked example 1.1.9
(The unit square under the maximum metric)
With the maximum metric on , a point is close to precisely when both coordinates are close to their targets. The square is complete and compact because is both. By contrast, is not complete: insert rational approximations to as the first coordinate and keep the second coordinate zero. The product sequence is Cauchy but has no limit in the product.
Counterexample mode
Infinitely many compact factors under the supremum metric
Claim: a product of compact metric spaces is compact under the supremum of the coordinate distances. Each two-point space is compact with the discrete metric. On the countable product of binary sequences use , which is finite and equals one when the sequences differ. Let have a one in coordinate and zeros elsewhere. These sequences are pairwise distance one, so no subsequence converges: this product metric is not compact. In fact converges coordinatewise to zero, but for every . The metric induces the discrete topology, not the usual product topology, on this countable product. Thus the example does not contradict the finite-product theorem and does not test compactness for the usual product topology.
Quick check
(Let be complete and be incomplete, both nonempty. Under the maximum metric, can be complete? If , what can be said about and ?)
Solution · Coordinate limits and completeness
No: embed a nonconvergent Cauchy sequence of while holding one chosen point of fixed. Any product limit would project to a limit in . Product convergence implies and because each coordinate distance is bounded by the maximum. Conversely, both coordinate convergences imply product convergence by taking the larger of two indices.
Countably many coordinates: a weighted metric
The preceding counterexample leaves a natural question: is there a metric on a countable product for which convergence does mean convergence in each coordinate? We can give later coordinates progressively smaller weights. Truncation keeps even unbounded factor distances under control.
Let , , be nonempty metric spaces and choose a base point in their product . Define
The series is bounded by . It is a metric: symmetry is immediate, a zero sum forces every coordinate distance to be zero, and the triangle inequality follows by summing . The base point is not part of the formula; it is used below to embed an individual factor. We assume the countable choices of points and subsequences made in these arguments are available.
Theorem 1.1.10
(Coordinatewise convergence and countable products)
For the metric , a sequence converges to exactly when for every fixed . It is Cauchy exactly when each coordinate sequence is Cauchy. Moreover, is complete if and only if each is complete, and is compact if and only if each is compact.
Proof
(Proof for countable products)
For any fixed , . Hence a -convergent or -Cauchy sequence has the corresponding property in each coordinate. For the converse, first choose so that the tail is less than half the desired error. The finitely many remaining coordinates can be made simultaneously close enough by taking the largest of their finitely many starting indices. Their weighted sum is then less than the other half. This proves both converse statements; for Cauchy sequences the estimates hold for all pairs of sufficiently late terms.
If every factor is complete, the coordinates of a -Cauchy sequence converge to points , and the first part gives convergence to . Conversely, embed a Cauchy sequence in as the th coordinate and hold all other coordinates at . Completeness of supplies a product limit, whose th coordinate is the required factor limit.
If every factor is compact, start with any sequence in . Select a subsequence converging in coordinate , then from that subsequence select one converging in coordinate , and continue. Take the th term of the th selected subsequence. For each fixed , all diagonal terms from index onward belong to the th subsequence. The th diagonal term has position at least in that subsequence, because every subsequence index map is strictly increasing. These positions therefore tend to infinity. Thus coordinate converges. The first part gives convergence in . Thus is sequentially compact, hence compact by the metric compactness theorem. Conversely, each projection is continuous by the first inequality and is onto because the other coordinates can be fixed at . A continuous image of a compact space is compact, so each factor is compact.
Proof X-ray
Why the diagonal works
The th diagonal term must be taken from a nested sequence of subsequences, not independently from the original sequence. For a fixed coordinate , all sufficiently late diagonal terms then remain in the subsequence already chosen to make that coordinate converge. The weighted tail estimate turns these infinitely many separate limits into one metric limit.
The open sets induced by are the usual product topology: a basic product neighborhood restricts only finitely many coordinates to open sets, leaving the rest unrestricted. Each such neighborhood is -open by the coordinate inequality and finitely many radii. Conversely, inside a -ball about , first make the weighted tail less than half the radius and then restrict finitely many coordinates so their contributions sum to less than the other half. This gives a basic product neighborhood inside the ball.
Worked example 1.1.11
(Binary sequences under the weighted metric)
On with the discrete factor metrics, the earlier sequence has . Every factor is compact, so the whole product is compact under . Under the earlier supremum metric its terms remain distance one from zero and have no convergent subsequence. These are two different metrics and two different topologies on the same set; there is no contradiction.
Quick check
(On with the metric built from ordinary absolute distances, let have value in coordinate and zero elsewhere. Does it converge? Is this product compact?)
Solution · A moving large coordinate
Yes: . Equivalently, each fixed coordinate is eventually zero. The product is not compact: its first-coordinate projection is onto , which is not compact. The large coordinate moving to the right does not make the product compact; compactness concerns every sequence, not just this one.
Open balls, closed balls and closures
Throughout this note, is a metric space, and . We write
- for the open ball, and
- for the closed ball.
The names suggest that is the closure of . This is true in with the Euclidean metric. It is false in a general metric space. Several standard proofs, including the nested-ball proof of the Baire category theorem, only need the part of that guess that is always true. This note separates out that part and shows exactly where the rest fails.
The closed ball is a closed set
Proposition 1.1.12
(Closed balls are closed)
For every and , the set is closed in .
Proof
(Two proofs that a closed ball is closed)
Proof via the triangle inequality. We show that the complement is open. Let , so , and put . Let . The triangle inequality gives , so . Thus , and the complement is open.
Proof via continuity. Define by . Applying the triangle inequality twice gives , so is continuous. We have , and is closed in . A continuous preimage of a closed set is closed, so is closed.
Both arguments depend on the non-strict inequality . The second argument shows this most clearly. The interval is closed in , but is not. So the same reasoning tells us nothing about whether is closed.
What is always true
Proposition 1.1.13
(The closure of an open ball lies in the closed ball)
For every and , .
Proof
(Proof of the inclusion)
If then , so . The closure is the smallest closed set containing . By the preceding proposition, is a closed set containing . Hence .
Intuition from goes wrong at the reverse inclusion. A point with always lies in . To lie in , however, must be a limit of points whose distance from is strictly less than . In we find such points by moving from a short way along the segment toward . A general metric space has no segments, and there may be no other points near at all.
Worked counterexample: a two-point space
Worked example 1.1.14
(Open ball, closed ball and closure in a two-point space)
Let with . Define and .
This is a metric. Positivity and symmetry are immediate. For the triangle inequality , the only nontrivial case is . Then , so differs from at least one of and . Hence the right-hand side is at least .
Take the center and the radius .
- Open ball. , because is not less than .
- Closed ball. .
- Closure of the open ball. The point is not in . The ball is an open neighborhood of that does not meet . Hence is already closed, and .
Therefore . The inclusion of the preceding proposition is strict here.
Common mistake
(The closed ball need not be the closure)
It is tempting to read as the closure of , but in a general metric space one only knows , and the two-point space shows that this inclusion can be strict. Without an additional hypothesis, a proof may use the facts that is closed and contains , but must not assume equality with .
Nested closed sets and Cantor's intersection theorem
The Baire construction produced a point by nesting closed balls whose radii tend to zero. That mechanism deserves its own statement, because it is in fact a characterization of completeness. For a nonempty subset of write , a number in .
A closure does not enlarge a diameter: . Indeed, if , choose and in . Then , and the triangle inequality gives , so . The reverse inequality holds because .
Theorem 1.1.15
(Cantor's intersection theorem)
A metric space is complete if and only if the following holds: whenever are nonempty closed subsets of with , the intersection is nonempty. When this happens, the intersection consists of exactly one point.
Proof
(Proof of Cantor's intersection theorem)
Suppose is complete and let be as stated. Choose . If , both and lie in by nesting, so . Since these diameters tend to zero, the sequence is Cauchy and converges to some . For a fixed , the terms with lie in the closed set , hence so does their limit. Thus . If also lies in the intersection, then for every , so .
Conversely, assume the nested-set property and let be Cauchy. Put . These sets are nonempty, closed and decreasing. Given , the Cauchy condition gives with for , so , and the same bound holds for every later index. Hence , and there is a point in every . For , both and lie in , so . Therefore , and is complete.
Each hypothesis is needed, and each failure has a simple witness. In , the closed sets are nested and nonempty but have empty intersection; their diameters are infinite. In the complete discrete space with distance one between distinct points, the closed sets have diameter one and empty intersection. The open intervals have diameters tending to zero and empty intersection; they are not closed. Finally, in the sets are closed in , nonempty and shrinking, yet their intersection is empty, as the theorem predicts for an incomplete space.
Compactness removes the diameter condition altogether. If is compact and are nonempty closed sets with empty intersection, the open complements cover . A finite subcover exists, and since the complements increase, a single is all of . This says , a contradiction. So in a compact space every decreasing sequence of nonempty closed sets has a nonempty intersection, whatever the diameters.
Closed and bounded subsets of Euclidean space
The warning that “closed and bounded” does not mean compact in a general metric space leaves the Euclidean statement to be proved. Give the Euclidean metric and compare it with the maximum metric of the finite-product section:
Hence the two metrics have the same Cauchy sequences and the same convergent sequences. Since is complete, the finite-product theorem makes complete, and therefore is complete as well.
Theorem 1.1.16
(Heine–Borel theorem)
A subset of with the Euclidean metric is compact if and only if it is closed and bounded.
Proof
(Proof of the Heine–Borel theorem)
Suppose is closed and bounded, and nonempty (the empty set is compact). It is complete, because a closed subset of the complete space is complete. For total boundedness, choose with and let . Pick a positive integer with and cut the cube into closed subcubes of side . Any two points of one subcube are at Euclidean distance at most . For each subcube that meets , choose one point of in it. The open -balls about these finitely many points of cover , because every point of lies in some subcube that meets . Thus is complete and totally bounded, hence compact by the metric compactness criterion.
Conversely, a compact is complete by the same criterion, and a complete subspace is closed in its ambient space, so is closed in . It is totally bounded and therefore bounded.
The proof shows exactly where Euclidean structure enters: a bounded set sits in a cube, and a cube can be cut into finitely many pieces of any prescribed small diameter. The infinite discrete space has no such subdivision, which is why the theorem does not transfer to it.
Two consequences of Baire's theorem
Baire's theorem is most often used in the contrapositive form: if a complete space were a countable union of small sets, it would be meager in itself, which is impossible. A point is isolated in if is open, that is, if some ball equals .
Corollary 1.1.17
(Complete spaces without isolated points are uncountable)
A nonempty complete metric space with no isolated points is uncountable.
Proof
(Proof by singletons)
A finite nonempty metric space has all its points isolated: take a radius smaller than every positive distance from the point. So suppose, toward a contradiction, that is countably infinite. Each singleton is closed. Its interior is empty, since an open ball contained in would equal and make isolated. Thus every singleton is nowhere dense, and is a countable union of nowhere dense subsets of itself. This contradicts Baire's theorem.
In particular is uncountable, and so is every nonempty closed subset of that has no isolated points in itself, such as the Cantor set: a closed subset of a complete space is complete. A set is called a set if it is a countable intersection of open sets. Open sets are , and so is the set of irrationals, for an enumeration of .
Corollary 1.1.18
(The rationals are not a set)
There is no sequence of open sets with .
Proof
(Proof by combining two families of dense open sets)
Suppose with each open. Every contains , so it is dense. Each is also open and dense. The countable family of dense open sets has intersection . Baire's theorem says this intersection is dense in the nonempty complete space , a contradiction.
Both corollaries separate two notions of size. The rationals are dense, yet they are a countable union of nowhere dense singletons; the irrationals are dense and their complement is meager. Density alone cannot tell these sets apart, while Baire's theorem can.
Worked examples
Worked example 1.1.19
(A complete bounded space that is not compact)
Let with for . A Cauchy sequence is eventually constant, since its late terms must have distance less than . Thus is complete and bounded. Each ball of radius contains one point, so is not totally bounded. The singleton cover also shows directly that it is not compact.
Worked example 1.1.20
(A totally bounded space with a missing limit)
The interval with ordinary distance is totally bounded by subdivision into finitely many short intervals. The sequence is Cauchy but has no limit in . It is therefore not complete and not compact. Its closure repairs the missing boundary points and is compact. Taking a closure inside an incomplete ambient space would not automatically repair every missing limit.
Worked example 1.1.21
(Countably many generic conditions)
Enumerate the rationals as . Each is open and dense, so Baire's theorem says that their intersection, the irrational numbers, is dense. Apply the same reasoning within , however, and the intersection is empty. This proves that with ordinary distance is not complete. The failure belongs to the space, not to the density of the individual sets.
A useful further warning concerns changing metrics. On , let . It induces the usual topology, but the sequence is -Cauchy and has no finite real limit. Completeness can change under a topologically equivalent metric. Compactness cannot, because its definition uses only open sets. Uniformly equivalent metrics do preserve completeness, but mere topological equivalence is weaker.
Common mistakes
Common mistake
(Assuming the limit before proving completeness)
When proving completeness, do not assume the limit exists in the chosen space and then verify convergence. First obtain it from an ambient complete space or construct it, then prove membership. When proving noncompactness, an unbounded sequence works in Euclidean spaces, but boundedness alone tells little in arbitrary metric spaces.
Common mistake
(Closed and bounded is not compact in general)
“Closed and bounded” characterizes compact subsets of , not arbitrary metric spaces. The infinite discrete example satisfies both properties in itself. Likewise, a countable intersection of dense open sets need not be open. For the irrational numbers, every neighborhood contains rationals, so their dense intersection has empty interior.
Baire's theorem gives existence, not a numerical algorithm for locating a point. Its nested choices can depend on arbitrary dense sets. The theorem is especially effective when a contradiction would make a complete space meager; it should not be interpreted as a probability statement about a randomly selected point.
Summary
Completeness keeps Cauchy limits inside a space. Total boundedness supplies finite approximations at every scale. Together they give compactness in metric spaces, and in Euclidean space they reduce to the Heine–Borel test “closed and bounded.” Cantor's intersection theorem recasts completeness as the survival of shrinking nested closed sets. Baire's theorem extracts a different consequence of completeness by nesting balls inside dense open conditions; it shows, for instance, that complete spaces without isolated points are uncountable and that is not a set. In each proof, identify the property responsible for the step: finite selection, existence of a limit, or retention of that limit in a closed set.
Exercises
Quick check
(Show that a compact subset of a metric space is closed. Explain why this does not require the ambient space to be complete.)
Solution · Guided solution
Let be compact and . If were in its closure, choose with . Compactness gives a subsequence converging to . That subsequence also converges to , contradicting uniqueness. Thus no point outside belongs to its closure. Only compactness of and the metric's uniqueness of limits were needed.
Quick check
(Prove that every totally bounded metric space has a countable dense subset.)
Solution · Guided solution
For each positive integer , select a finite -net . Their union is countable. Given and , choose with ; some point of lies within of . The union is therefore dense. This proves separability, but the converse is false: with the discrete metric is countable and not totally bounded.
Quick check
(Let be closed subsets of a nonempty complete metric space and suppose . Show that at least one has nonempty interior.)
Solution · Guided solution
If every had empty interior, each would be nowhere dense because it is already closed. Baire's theorem forbids their union from being the whole nonempty complete space. Closedness matters: is the union of its rational and irrational subsets, both of which have empty interior.
Quick check
(For , show that is totally bounded but incomplete. Identify its completion up to isometry.)
Solution · Guided solution
The map is an isometry onto with ordinary distance. That interval is totally bounded but omits the limits of sequences approaching either endpoint. Its completion is . The two added points correspond to the two directions of escape on the real line; they are distinct points, not one shared infinity.
Quick check
(Let be any function on a metric space. Show that the set of points at which is continuous is a set. Deduce that no function is continuous exactly at the rational points.)
Solution · Guided solution
Define the oscillation , a number in . Put . Each is open: if and , then , so . Next, is continuous at exactly when . If is continuous at and , choose with on ; then . If , a ball on which the image has diameter less than gives there. Hence the continuity set is . A function continuous exactly on would make a set, which the corollary above forbids. The opposite situation does occur: the function equal to at a rational in lowest terms with , and to at irrationals, is continuous exactly at the irrationals.
Quick check
(Show that no metric on that induces its usual topology is complete.)
Solution · Guided solution
Nowhere density depends only on the open sets. In the usual topology of , each singleton is closed and has empty interior, because every open interval around contains other rationals. A complete metric inducing this topology would therefore make a nonempty complete metric space without isolated points, and the corollary above would make it uncountable. This is false. Compare the example: changing the metric can destroy completeness without changing the topology, but no change of metric can make complete.
Quick check
(Let be nonempty compact subsets of a metric space , which need not be complete. Prove that . Does the conclusion survive if the are only closed and bounded?)
Solution · Guided solution
Each is closed in , so is a closed subset of the compact space . The nested-set statement for compact spaces, applied inside , gives a common point. Completeness of is not needed: the argument uses only the compactness of . For closed bounded sets the answer is no: in the discrete space , the sets are closed, bounded and nested with empty intersection.