Compactness concerns the existence of limits and finite subcovers. Connectedness concerns a different obstruction: can the space be split into two nonempty open pieces? A compact space can be highly disconnected, and a connected space need not contain a continuous path between every pair of points. These distinctions matter when a continuous observation turns a family of ODE solutions into an interval, or when a local argument must be propagated through a domain.
This note develops the underlying metric-space tools. We first prove the usual connectedness principles, then identify the extra local hypothesis that produces paths. Finally we study finite chains of small steps. Chains can imitate connectedness at every scale without giving genuine connectedness; compactness is the hypothesis that makes that imitation reliable.
Relative topology and separations
Definition 1.4.1
(Separation and connectedness)
Let be a metric space. All open and closed sets in a subset are relative to that subset. A separation of is a decomposition into disjoint nonempty relatively open sets. Since each is the complement of the other, both are also relatively closed. The space is connected if no such separation exists. The empty set and singletons are connected.
A set that is both open and closed is called clopen. Thus is connected exactly when its only clopen subsets are and . This does not say that a connected subset of an ambient space must be open or closed there. For example, and are both connected subsets of .
Common mistake
(Open in the subspace, not in the ambient space)
In with the distance of , the set is open in , since , although it is not open in . Its complement is open in for the same reason, so is disconnected. A proof that tests openness in the ambient space would miss this separation.
We use basic metric continuity, completeness of the real line and the compact-metric equivalence with sequential compactness and complete total boundedness.
Continuous observations and intermediate values
Theorem 1.4.2
(Connectedness is preserved by continuous observations)
A continuous image of a connected space is connected. A subset of is connected exactly when it is an interval, allowing the empty set and one-point intervals. Consequently is connected if and only if every continuous has interval image.
Proof
(Proof of the image and interval statements)
For the image statement, suppose were a separation. Its two inverse images would be disjoint nonempty open sets covering , a contradiction. The target may be any metric space; real-valued observations are a particularly useful special case.
If a subset is not an interval, take with but . The sets and separate it. Conversely suppose an interval has a separation . Choose in different parts, relabeling the parts if necessary. The set is nonempty and closed in . Its supremum belongs to it: approximate the supremum by points of that set and use relative closedness. Since , we have . Relative openness of then gives points of just to the right of , a contradiction. The interval is connected.
Finally, if is disconnected, the function equal to zero on one clopen part and one on the other is continuous but has noninterval image . This proves the converse characterization.
In particular, a continuous real observation takes every intermediate value between any two values it attains on a connected space. This conclusion needs no path between the two original points. Conversely, knowing the conclusion for only one selected observation does not certify connectedness of its domain.
Closures, unions and components
Theorem 1.4.3
(The structure of connected components)
The closure of a connected set is connected. A family of connected sets with a common point has connected union. For each , the union of all connected subsets containing is a connected set , called its component. Components are maximal connected subsets, are closed in , and partition . They need not be open.
Proof
(Proof of the component structure)
If had a separation, both nonempty open parts would meet its dense subset . Their intersections with would separate , a contradiction. For a family sharing a point , a proposed separation of the union places on one side. Each connected member must then lie on that side, leaving the other side empty.
The collection defining is nonempty because it contains , and its members share . The union principle proves connected. Any connected set containing it, or merely meeting it, can be united with it to give a connected set containing ; maximality follows. In particular its closure is connected and contains , so the defining union already contains that closure. Hence is closed. If two components meet, their union is connected and maximality makes them equal. Since every point belongs to its component, the components partition the space.
Common mistake
(Components are closed but need not be open)
Every component is closed, by the closure principle. It need not be open: in every component is a singleton, because between two distinct rationals lies an irrational cut, and is not open in , since every ball about contains other rationals. Openness of components is exactly what the local path hypothesis supplies later in this note.
Definition 1.4.4
(Paths and path components)
A path from to is a continuous map with endpoints . Constant paths, reversing a path and concatenating two paths show that path joining is an equivalence relation. For concatenation, traverse the first path on and the second on ; the matching endpoints guarantee continuity at the junction. Its equivalence classes are path components.
A path image is connected because is connected. The union of all paths starting at one point is therefore connected, so each path component lies in a connected component. Equality requires further hypotheses.
Counterexample mode
A connected set need not supply a joining path
Let
The set is path connected: descend a vertical tooth, travel along the base, and ascend another tooth. Also . Since is dense in the relative space , the closure theorem shows connected.
There is no path from to . Suppose one existed and let be its last time at ; this time exists because the inverse image of is closed in . On a short interval after , continuity keeps the second coordinate above . Its first coordinate lies in . A continuous image of that short time interval is an interval in , hence a singleton: contains no nondegenerate interval. Starting from zero, the first coordinate must stay zero. The only point of with first coordinate zero and height above is , contradicting the last-time choice. Thus the component containing is all of , but its path component is only .
When connectedness does produce paths
Definition 1.4.5
(Local path connectedness)
Call locally path connected if for each and each open neighborhood of , there is an open path-connected with . The requirement concerns arbitrarily small neighborhoods, not merely the existence of paths somewhere in the whole space.
Theorem 1.4.6
(Local path connectedness and open Euclidean domains)
In a locally path-connected space, connected components equal path components and are open. Moreover, every connected open subset of is polygonally connected: any two points can be joined by a path consisting of finitely many straight segments inside it.
Proof
(Proof of the local path criterion)
Fix a path component . For each , a path-connected open neighborhood of lies in , so is open. Every other path component is likewise open, making its complement open. Thus is clopen. A connected component meeting cannot meet its complement, while the preceding inclusion puts in that connected component. They are equal, proving the first statement.
For an open Euclidean set , small balls around each point lie in and are convex, hence path connected. To obtain the stronger polygonal conclusion, fix and let consist of points reachable from by finitely many segments in . It is open: append a segment in a sufficiently small ball. Its complement is open by the same argument, since reaching any point of such a ball would let us reach its center. If is connected and nonempty, . This proof explicitly uses openness to fit those small balls.
Proof X-ray
The local hypothesis is where the path enters
Connectedness forbids a clopen separation; it does not construct segments. Local path connectedness first makes every path component open. Only then can connectedness force a single path component. In an open Euclidean domain, convex balls provide the local pieces and a finite concatenation provides the path. For a general connected subset of Euclidean space, those balls may leave the subset, so the same proof is unavailable.
Small chains and the role of compactness
Definition 1.4.7
(Small chains)
For , an -chain from to is a finite list with . A zero-step chain is allowed. Chain joining is an equivalence relation. Its classes are open: every -ball about a reachable point stays in its class. Their complements are unions of other classes, so every class is also closed.
Theorem 1.4.8
(Compact connectedness from arbitrarily fine chains)
Let be a nonempty compact metric space. It is connected if and only if every pair of points can be joined by an -chain for every . At each fixed scale there are only finitely many chain classes. Moreover, if is the -chain class of , then
These are connected components, not necessarily path components.
Proof
(Proof of the chain characterization)
A connected space, compact or not, cannot have more than one chain class because the classes are clopen. Conversely, a separation of compact gives nonempty disjoint compact parts at positive distance . To see positivity, if their distance were zero, choose points from the two parts with distances tending to zero and extract a convergent subsequence from one compact part. The paired points have the same limit in the other closed part, a contradiction. A chain with steps less than cannot pass from to . This proves the criterion. A finite -net shows there are finitely many classes, since each of its balls lies in the class of its center.
For the component formula, the are nested nonempty compact clopen sets. Every connected subset containing lies in every , so . We prove connected. If were a separation, would be disjoint nonempty compact sets at positive distance . Their open -neighborhoods have distance at least from each other.
Eventually . Otherwise choose an increasing sequence of indices and points of outside this open union. Compactness of , nestedness and closedness give a subsequential limit in every and outside , contradicting . Yet each contains both and . For large with , a chain in cannot cross between and . Such a chain must exist: any two points of one chain class can be joined, and all intermediate points of the joining chain remain in that class. This contradiction proves connected. It contains , so by maximality, giving equality.
The compactness step did more than prevent sequences from escaping. It converted a separation of the limiting intersection into a separation visible at one finite scale. Without that step, finer and finer chains can keep finding bridges farther away.
Worked examples
Worked example 1.4.9
(An annulus is path connected but not convex)
Consider . Move each chosen point radially to the circle of radius , join the resulting points by a circular arc, then reverse the second radial segment. All pieces stay in the annulus, giving a continuous path. Thus the annulus is connected. It is not convex: and belong to it, while their midpoint does not. The general open-domain theorem also guarantees a polygonal path, even though this particular construction used an arc.
Worked example 1.4.10
(Rational chains at every scale do not make a continuum)
Let . For rational and any , choose an integer with . The arithmetic subdivision , , is an -chain entirely in . Nevertheless the irrational gives a separation by and .
Indeed every component is a singleton: between any two distinct rational points lies an irrational cut that separates any subset containing them. No singleton is open in . Rational grids give finite nets, so the space is totally bounded. It is not compact: rational approximations to the irrational form a Cauchy sequence with no limit in it. All-scale chain joining alone did not repair the missing compactness.
Worked example 1.4.11
(Completeness alone still permits bridges at infinity)
In let
Both graphs are closed, and their union is closed in the complete Euclidean plane, hence complete. Each graph is closed in and is the complement of the other, so both are relatively open and they separate . But their distance is zero: at the same large , the vertical gap is .
For any , choose with that gap below . Within either graph, a finite chain can follow its continuous parametrized arc to the point with first coordinate ; uniform continuity on the finite parameter interval supplies sufficiently short steps. One more step crosses the gap. Thus has a single chain class at every scale, despite being disconnected and complete. It is noncompact because it is unbounded. Compactness cannot be replaced merely by completeness in the theorem.
Worked example 1.4.12
(Compact binary coordinates have singleton components)
Let with
Diagonal selection makes every coordinate of a subsequence eventually constant; the remaining distance tail is at most after the first coordinates are fixed. Hence every sequence has a convergent subsequence, so is compact. A coordinate map is continuous because implies . Two different points are separated by a clopen coordinate condition, so every connected component is a singleton. None is open: changing a sufficiently late bit produces an arbitrarily close different point.
More precisely, the -chain classes are exactly the sets with a fixed prefix of bits. A chain cannot change any earlier bit, since each step is strictly shorter than . Conversely, for two points with that prefix, change finitely many later bits one at a time, then jump across a remaining tail of total distance less than . This gives a finite chain. Intersecting the decreasing prefix classes recovers exactly one point, illustrating the component formula without suggesting that a finite scale already identifies every component.
Practice
Quick check
(Show that a continuous nonzero real function on a connected space has constant sign. Deduce that a map with invertible derivative everywhere on a connected open has constant sign of . Does this force global injectivity?)
Solution · Guided solution
The image is an interval missing zero, so it lies wholly in either positive or negative numbers. For the map , continuity of makes its determinant continuous, and invertibility makes it nonzero, proving the sign claim. Global injectivity does not follow: on , the map has determinant but repeats its values when increases by . Constant local orientation is not a global inverse theorem.
Quick check
(If is differentiable on an open Euclidean set and everywhere, prove is constant on each connected component. Give a disconnected domain on which it is not globally constant.)
Solution · Guided solution
An open Euclidean set is locally path connected. Hence each connected component is open in , and therefore open in the ambient Euclidean space. It is a connected open set, so the polygonal theorem applies to it. Along each segment, the one-variable derivative of the restricted function is applied to the segment direction, hence zero. The mean-value theorem makes the function constant on that segment; a finite concatenation gives constancy throughout the component. On , the function zero on the negative half-line and one on the positive half-line is differentiable with zero derivative, but is not globally constant.
Quick check
(For nonempty metric spaces , prove their product with metric is connected if and only if both factors are connected.)
Solution · Guided solution
If the product is connected, both coordinate projections are continuous and onto, so both factors are connected. Conversely fix and . For each , the set is connected: its two pieces are continuous images of the connected factors and meet at . These sets all contain , and their union is the whole product. The common-point union theorem proves connectedness. Nonempty factors ensure the onto projections and the chosen points exist.
Quick check
(Prove that a decreasing sequence of nonempty compact connected subsets of a metric space has nonempty connected intersection. Show that dropping compactness can leave a nonempty disconnected intersection, using the sets below.)
Solution · Guided solution
Choose . Compactness of supplies a convergent subsequence; nestedness and closedness put its limit in every , so the intersection is nonempty and compact. If it separated into compact parts at distance , their disjoint -neighborhoods would contain all sufficiently late : otherwise compactness and nestedness would produce a point of outside those neighborhoods. Each meets both parts and would then be separated by these two open neighborhoods, a contradiction.
The displayed are closed, nonempty, nested and path connected: ascend a vertical ray to height at least , cross the horizontal strip, then descend if needed. Their intersection is exactly the two vertical rays , which are separated by their first coordinates. The connecting strip escaped upward. Closedness without compactness did not keep the limiting connection.
Quick check
(Let and . Prove is compact and path connected, but not locally path connected at . Explain why this does not contradict the local path-connectedness theorem.)
Solution · Guided solution
The set is compact, so is a finite union of compact products. Paths can descend to the base, travel along it and ascend a vertical segment, proving path connectedness. Let and . Any relatively open set with contains for sufficiently large . If were path connected, it would be connected, so its first-coordinate image would be an interval in containing zero and . That is impossible because contains no such interval. Thus no open path-connected neighborhood of fits inside . The theorem assumes local path connectedness; its converse, that global path connectedness supplies this local property, was never asserted.
Exact boundaries
The note establishes foundational connectedness principles, components, local path criteria and the compact small-chain characterization. It does not classify all connected compact metric spaces, give an algorithm to identify components from a finite sample, or infer path connectedness from a compact connected limit. The chain-intersection statement concerns connected components only. Those limits explain precisely why the stronger path conclusions in later applications need extra hypotheses.