The core uses basic real analysis and the usual classical choice background for successive selections. The optional ordinal example explains its additional set-theoretic notation. No theorem identifying general topological compactness with a sequence condition is assumed.
Motivation
An open-cover definition can look remote from a sequence calculation. In metric spaces, finite small-ball covers bridge the two: repeated refinement traps a subsequence in sets of shrinking diameter. Completeness supplies its limit. Compactness therefore combines a finite approximation property with a convergence property.
Boundedness is too weak for finite approximation in an arbitrary metric space. A whole infinite discrete set can fit inside one large ball while still needing infinitely many tiny balls. Conversely, a bounded interval with a missing endpoint has excellent finite approximations but loses Cauchy limits. These two failures will isolate the exact hypotheses.
Definitions
Definition 10.1.1
(Total boundedness and -nets)
A metric space is totally bounded if, for every , finitely many open balls of radius , with centers in , cover . Their centers form a finite -net. This condition controls all scales. Bounded only means that one finite radius about some point contains the space.
Definition 10.1.2
(Sequential compactness and completeness)
A space is sequentially compact if every sequence has a convergent subsequence with limit in the space. A subsequence uses strictly increasing indices; simply repeating one convenient term is not allowed. It is complete if every Cauchy sequence converges in the space. Compactness retains its open-cover definition.
Definition 10.1.3
(Ball Lebesgue number)
We call a ball Lebesgue number for an open cover if each ball lies in some member of the cover. This is equivalent, up to shrinking constants, to the earlier small-diameter formulation. For example a nonempty set of diameter less than lies in the -ball about any one of its points.
Theorem/proposition
Theorem 10.1.4
(Three equivalent forms of metric compactness)
For a metric space the following are equivalent: compactness; sequential compactness; completeness together with total boundedness. The empty space satisfies all three conditions.
Concept lensAnalytic
Total boundedness through sequences
A metric space is totally bounded exactly when every sequence has a Cauchy subsequence. The refinement argument below gives one direction without using completeness. Conversely, if some admits no finite net, the greedy choice produces terms pairwise at distance at least , and no subsequence of them is Cauchy. The first theorem then reads: compactness means that every sequence has a Cauchy subsequence and every Cauchy sequence converges.
Theorem 10.1.5
(Uniform smallness for every open cover)
Every open cover of a nonempty sequentially compact metric space has a ball Lebesgue number. Combined with total boundedness, this gives a finite subcover.
Theorem 10.1.6
(Heine–Borel in finite-dimensional Euclidean space)
A subset of , , is compact if and only if it is closed and bounded. This is a finite-dimensional Euclidean conclusion, not a closed-bounded criterion for arbitrary metric spaces.
Proof sketch or proof idea
Proof
(Compact implies sequentially compact)
First suppose compact and let be a sequence. Set . These are nonempty nested closed sets. Their intersection is nonempty: otherwise the complements cover , and a finite subcover would force the corresponding finite intersection, equal to one , to be empty. Choose in the intersection. Every ball about meets every tail. Recursively choose strictly increasing with , using the least eligible index at each step. This is a convergent subsequence.
Proof
(Sequential compactness gives finite nets and completeness)
Now assume sequential compactness. If some had no finite -net, choose , then outside the -balls about all earlier choices. Distinct chosen points have distance at least , so no subsequence can converge: convergent sequences are Cauchy. This contradiction proves total boundedness. Also, a Cauchy sequence has a convergent subsequence by assumption, and then the whole sequence converges to that subsequence's limit. Indeed put both a sufficiently late term and a sufficiently late subsequence term within the Cauchy tolerance, and use the triangle inequality. Thus the space is complete.
Proof
(Finite nets and completeness give convergent subsequences)
Conversely let be complete and totally bounded. Given a sequence, cover the space by finitely many balls of radius ; one contains terms from infinitely many indices. Inside those indices choose an infinite subset lying in one member of a finite radius- cover, and continue. We obtain nested infinite index sets , with the corresponding terms in one ball of radius . Choose strictly increasing. For , both belong to that same ball, so
The subsequence is Cauchy and hence converges. The infinite-index refinement matters: independently choosing one point from each small ball would not necessarily produce a subsequence of the original sequence.
Proof
(Lebesgue numbers recover finite subcovers)
It remains to recover open-cover compactness from sequential compactness. Let be an open cover and suppose it has no ball Lebesgue number. For each positive integer , choose such that is contained in no member of the cover. Pass to a subsequence . Some covering set contains , so for some . For large , both and . Then
contradicting its choice. Thus a uniform exists. Sequential compactness already gave total boundedness; choose a finite -net. Each of its balls lies in the corresponding radius- ball and hence in some . Finitely many such members cover , completing the equivalences.
Proof X-ray
Two finiteness steps play different roles
A Lebesgue number makes one radius work uniformly across a given cover. Total boundedness then reduces the infinitely many possible ball centers to finitely many. Neither step alone yields the finite subcover. In the other direction finite nets produce a Cauchy subsequence, while completeness supplies the point missing from an incomplete domain.
Proof
(Closed and bounded in finite dimensions)
For Heine–Borel, compact subsets are closed because Euclidean space is Hausdorff. They are bounded: a finite radius-one net puts every point within one of finitely many distances of the origin. Conversely, is complete because every coordinate of a Cauchy sequence is real Cauchy and coordinate limits converge in Euclidean norm. A closed subset is complete, since its limits remain in it. A bounded subset is totally bounded: enclose it in a cube and subdivide the cube into finitely many smaller cubes of diameter less than . Choose one point of the subset from every small cube that meets it. These points form an -net with centers in the subset. Completeness plus total boundedness now proves compactness. The grid construction depends on finite dimension.
Worked examples
Worked example 10.1.7
(Complete and bounded can still leave infinitely many separated points)
On an infinite set put for distinct points. It is bounded, and every Cauchy sequence is eventually constant, so it is complete. A ball of radius contains just one point, so no finite such cover exists. An infinite sequence of distinct points has no convergent subsequence, and the singleton open cover has no finite subcover. Total boundedness is the missing condition.
Worked example 10.1.8
(Finite approximation does not restore a missing endpoint)
The interval is totally bounded, by subdividing its ambient unit interval. Yet is Cauchy and approaches zero, which is absent. It has no subsequence converging in , because every real subsequence still tends to zero. The open cover covers but has no finite subcover: finitely many of these intervals still miss sufficiently small positive points. Each failure is visible without merely citing the theorem.
Worked example 10.1.9
(An optional nonmetric boundary: the first uncountable ordinal)
Let be the least uncountable ordinal, and give its order topology. At a positive ordinal , neighborhoods contain a segment ; zero has initial-interval neighborhoods. In the usual choice background, a countable family of countable ordinals has a countable supremum, since their union is countable.
Every sequence has a nondecreasing subsequence: choose a term attaining the least value in the whole sequence, then a term attaining the least value in the remaining tail, and continue. Its supremum is still below . The subsequence converges to , because it eventually exceeds any ; if the supremum is attained it is eventually constant there. Thus the space is sequentially compact. But the open initial intervals , , cover it and no finite subfamily does. It is not compact and therefore not metrizable. The countable-supremum assertion and this order-topology argument are additional ingredients, not consequences of the metric proof.
Common mistakes
Common mistake
(Total boundedness does not make every sequence Cauchy)
Completeness concerns Cauchy sequences, not all sequences. Total boundedness does not say every sequence is Cauchy; an alternating sequence in a two-point space already disproves that. It gives a Cauchy subsequence, which is exactly what compactness needs.
Common mistake
(Nets and limits must lie in the given space)
A finite ball cover may have centers outside a subset being studied. To obtain an intrinsic net, select a point of the subset from each ball that meets it and enlarge the radius, as in the first exercise. Also, convergent subsequences must have their limits in the given space, not merely in an ambient completion.
Common mistake
(Transferring the metric equivalences to general spaces)
The equivalences are metric results. The ordinal example disproves sequential compactness implying compactness in general spaces. No reverse implication for arbitrary topological spaces is justified by the metric argument.
Finally, closed-and-bounded compactness depends on Euclidean finite grids and cannot be transported to every metric space, even to a complete normed space.
Counterexample mode
A closed bounded ball that is not compact
Claim: in a complete metric space, every closed and bounded subset is compact. Take the space of real sequences with finite and the norm ; it is complete. Its closed unit ball is closed and bounded. Let have in place and elsewhere. Each lies in , and for
So no subsequence of is Cauchy, and is not sequentially compact. The failure is total boundedness: if two of the lay in one open ball of radius , the triangle inequality would give distance less than , so each such ball contains at most one and no finite -net exists. The correct statement replaces boundedness by total boundedness; Heine–Borel holds because in the two coincide.
Summary
Compactness gives cluster points of every sequence. Sequential compactness gives finite nets and completeness, while finite-net refinement constructs Cauchy subsequences. Uniform cover radii then recover open-cover compactness. In Euclidean finite dimension, closedness and boundedness supply the two metric ingredients; outside that setting they must be checked separately.
Exercises
Quick check
(Prove a closed subset of a complete metric space is complete, and any subset of a totally bounded metric space is totally bounded with its induced metric.)
Solution · Answer
A Cauchy sequence in a closed subset is ambient Cauchy, so converges there; closedness puts its limit back in the subset. For total boundedness, take a finite ambient -net. From each ball meeting the subset choose one subset point. Any two points in that ball are less than apart, so the chosen points give a finite intrinsic -net. The empty subset is immediate. No completeness is needed for the second assertion.
Quick check
(For nonempty metric spaces , use . Prove completeness and total boundedness pass to the product, giving another proof that the product of compact metric spaces is compact.)
Solution · Answer
A product Cauchy sequence is Cauchy in both coordinates; coordinate completeness gives a pair limit in the maximum metric. If finite -nets in the factors have sizes , their Cartesian product is an -net of size at most . Thus a product of complete totally bounded spaces has both properties and is compact. For compact factors the theorem supplies the two properties. The maximum metric is the usual finite-product topology because its balls are products of equal-radius balls and finite rectangular neighborhoods contain such balls.
Quick check
(Prove a continuous map from a compact metric space to a metric space is uniformly continuous, using convergent subsequences.)
Solution · Answer
If uniform continuity failed, some would admit pairs with but image distance at least . Compactness gives . The triangle inequality also gives . Continuity makes both image sequences converge to , contradicting their fixed separation. This proof uses metric compactness to extract a subsequence and does not silently assume a continuous map is uniformly continuous on a noncompact domain.
Quick check
(Give an explicit finite -net of and a bound on its number of points. Explain why this does not make every sequence in the interval Cauchy.)
Solution · Answer
Choose an integer and centers , . Every interval point is at distance at most from a center, so these points work. One may take . The sequence alternating between and is not Cauchy, although each constant parity subsequence converges. Finite nets control subsequences, not arbitrary consecutive terms.
Quick check
(Suppose is isometrically embedded as a dense subset of a complete metric space . Prove is compact if and only if is totally bounded. You need not construct the completion itself.)
Solution · Answer
Necessity follows because a compact is totally bounded and this property passes to subsets. Conversely cover by finitely many radius- balls with centers in . Since their union contains the dense subset, the union of their closed radius- balls covers : closure commutes with finite unions and each open-ball closure lies in its closed ball. These closed balls lie in the corresponding open radius- balls, proving total boundedness of . Its assumed completeness now gives compactness. The argument is conditional on a given completion and does not claim an unproved general construction of one.