Open sets, compactness and connectedness
Introduced| English | Meaning |
|---|---|
| interior/ɪnˈtɪərɪə/ | The set of points with an open ball around them lying wholly inside the given set. |
| closure/ˈkləʊʒə/ | The set of points for which every open ball centred there meets the given set. |
| connected/kəˈnektɪd/ | The impossibility of separating the whole space into two disjoint nonempty relatively open parts. |
Open depends on the available space
- A point at the end of a permitted range can still have room around it within that range. The set $[0,1)$ is not open in $\mathbb R$, but it is open in $X=[0,2]$, since $[0,1)=X\cap(-1,1)$.
- State the ambient space before computing interior 内部, closure 闭包 or boundary. We will then distinguish finite-cover compactness, connectedness and the stronger requirement of a continuous path.
Neighbourhoods define interior and closure
- In a metric space $X$, the interior is the set of points with an open ball around them lying wholly inside the given set. The closure is the set of points for which every open ball centred there meets the given set. All balls here are balls in $X$.
- Every ball around a boundary point meets both the set and its complement in $X$. Equivalently, boundary is closure minus interior. A set is open when it equals its interior, and closed when it equals its closure; isolated points of the set remain in its closure.
Compute the three sets in the real line
- In $\mathbb R$, the set $[0,1)$ has interior $(0,1)$, closure $[0,1]$ and boundary $\{0,1\}$. It is neither open nor closed. Open and closed are not opposites: the empty set and the whole space are both open and closed.
- The rational set $\mathbb Q$ has empty interior in $\mathbb R$, because every open interval contains an irrational. Its closure and boundary are both $\mathbb R$, because every open interval also contains a rational. A boundary can be much larger than two endpoints.
The closure of $\mathbb Q$ in $\mathbb R$ is
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Every real number is a limit of rationals.
The boundary of $(0,1) \cup (1,2)$ in $\mathbb R$ is
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Closure [0, 2] minus interior (0,1) ∪ (1,2).
Subspace topology changes the neighbourhoods
- If $A\subseteq X\subseteq\mathbb R$, the sets open in $X$ are $X\cap U$ with $U$ open in $\mathbb R$. Also $\operatorname{cl}_X A=X\cap\operatorname{cl}_{\mathbb R}A$. Relative interior is not generally the intersection with the real-line interior.
- In the diagram, take $X=[0,1]$ and $U=[0,1/2)$. Then $U=X\cap(-1,1/2)$ is open in $X$, so its relative interior is all of $U$, including $0$. Its closure in $X$ is $[0,1/2]$ and its relative boundary is only $\{1/2\}$; in $\mathbb R$, the boundary also contains $0$.
In $X = [0, 2]$, the set $[0,1)$ is open because it equals $X$ intersected with an ____ subset of $\mathbb R$.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
For example (−1, 1).
For $A=[0,1/2)$ inside $X=[0,1]$, what is the status of $0$ relative to $A$ in $X$?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
A sufficiently small ball in X around 0 is [0, ε), lying inside A. Thus 0 is an interior point in X. In ℝ, however, every neighbourhood also contains negative points, so 0 is a boundary point there.
For $A=[0,1/2)$ and $X=[0,1]$, compute the interior, closure and boundary of $A$ first in $\mathbb R$ and then in $X$. Explain what changes at zero.
In ℝ, the interior is $(0,1/2)$, the closure is $[0,1/2]$, and the boundary is $\{0,1/2\}$. In X, a small neighbourhood of 0 has the form $[0,\varepsilon)$, so lies inside A. Thus the relative interior is all of A, the relative closure is $[0,1/2]$, and the relative boundary is $\{1/2\}$. Zero stays in the closure but changes from a boundary point in ℝ to an interior point in X. In particular, intersecting X with the real-line interior would incorrectly omit zero.
Closure through sequences
- In a metric space, $p$ belongs to the closure of $A$ exactly when a sequence of points of $A$ converges to $p$. If $p\in A$, the constant sequence is enough; a closure does not discard isolated points.
- For $A=\{1/n:n\geq1\}$ in $\mathbb R$, every $1/n$ is isolated, and $0$ is the only new limit point. Thus the closure is $A\cup\{0\}$, the interior is empty, and the boundary is the whole closure. The set $A$ itself is not closed.
Interior, boundary or outside the closure?
Classify each point using neighbourhoods in the named ambient space.
Compactness means a finite subcover
- A set $K$ is compact when every family of open sets covering it has a finite subfamily still covering it. The open sets may be taken in the ambient space, or equivalently in the relative topology of $K$.
- The intervals $U_n=(1/n,1)$ for $n\geq2$ cover $(0,1)$. Any finite selection has a largest index $N$ and union $(1/N,1)$, which misses $1/(2N)$. Therefore this cover has no finite subcover, proving that $(0,1)$ is not compact.
Why does the cover $U_n=(1/n,1)$, $n\geq2$, of $(0,1)$ have no finite subcover?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
The sets are nested as n increases. A finite choice has a largest index N and union (1/N, 1), missing the point 1/(2N) of (0, 1). A cover can be infinite and still have a finite subcover, so the number of sets alone is not a proof.
Heine–Borel has an ambient-space condition
- In Euclidean space $\mathbb R^d$, a subset is compact exactly when it is closed in $\mathbb R^d$ and bounded. For example, $K=\{0\}\cup\{1/n:n\geq1\}$ is closed and bounded in $\mathbb R$, hence compact.
- The phrase “closed and bounded” is insufficient in an arbitrary metric space. Even $(0,1)$ is closed and bounded as a subset of itself, while the preceding cover proves it is not compact. Heine–Borel requires closedness in the stated Euclidean ambient space.
The set $\{0\}\cup\{1/n:n\geq1\}$ is compact in $\mathbb R$.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
It includes its only new accumulation point, 0, so it is closed in ℝ. It is also bounded, so Heine–Borel applies. Deleting 0 would destroy closedness and compactness.
Completeness and total boundedness do different jobs
- A metric space is complete if every Cauchy sequence converges within it. It is totally bounded if, for every $\varepsilon>0$, finitely many radius-$\varepsilon$ balls cover it. For metric spaces, compactness, sequential compactness, and completeness together with total boundedness are equivalent.
- Give $\mathbb N$ the discrete metric: distinct points have distance $1$. It is bounded and complete, since every Cauchy sequence is eventually constant. But radius-$1/2$ balls are singletons, so finitely many cannot cover it. It is not totally bounded or compact.
With distance $1$ between distinct points, $\mathbb N$ is compact because it is complete and bounded.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
It is complete and bounded, but not totally bounded. Balls of radius 1/2 are singletons, and no finite set of them covers ℕ. Completeness plus ordinary boundedness is insufficient.
Compare $(0,1)$ with its usual metric and $\mathbb N$ with the discrete metric of distance $1$. Both are closed and bounded as subsets of themselves. Why is neither compact, and which part of the complete-and-totally-bounded criterion fails?
The interval (0,1) is totally bounded: finitely many small ordinary balls cover it at each fixed radius. But it is incomplete, since the Cauchy sequence 1/n, starting at n = 2, has real limit 0 outside the space. The cover $(1/n,1)$ also has no finite subcover. The discrete space ℕ is complete, since a Cauchy sequence must eventually be constant, and it is bounded with diameter 1. It fails total boundedness: radius-1/2 balls are singletons and finitely many cannot cover ℕ. Thus one example lacks completeness and the other lacks total boundedness. Being closed in oneself does not provide the Euclidean ambient closedness required by Heine–Borel.
Continuous images and extrema
- Continuous images of compact sets are compact. Consequently, a continuous real-valued function on a nonempty compact domain attains both a minimum and a maximum. Compactness alone does not force the image to be an interval.
- On $(0,1)$, the continuous function $f(x)=x$ attains neither its infimum $0$ nor its supremum $1$; $f(x)=1/x$ is unbounded. The missing compactness matters. A minimum or maximum also cannot be attained on an empty domain.
A continuous real-valued function on a nonempty compact domain must
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Its image is a nonempty compact subset of ℝ, so its minimum and maximum belong to that image. The image need not be an interval unless connectedness is also assumed.
Connected means no separation of the whole set
- The term connected 连通的 means the impossibility of separating the whole space into two disjoint nonempty relatively open parts. The parts must cover the whole set; two disjoint open subsets that leave other points out do not constitute a separation.
- In $\mathbb R$, connected subsets are exactly intervals, including singletons and the empty set by convention. The rationals are disconnected: cutting at $\sqrt2$ gives two nonempty relatively open parts, and no rational point lies at the cut to join them.
Which subsets of $\mathbb R$ are connected?
Difficulty 2/5 · 140 coins + a small bonus on first completion today
The connected subsets of ℝ are intervals, including single points.
Connected images fill intermediate values
- A continuous image of a connected set is connected. For a real-valued function, if two values occur, every value between them occurs too. This is the topological content of the intermediate value theorem; the domain need not itself be a real interval.
- If a nonempty domain is both compact and connected, a continuous real-valued image is a closed bounded interval $[m,M]$. Compactness supplies attained extrema, while connectedness fills the values between them. For the compact set $\{0,1\}$, the identity map has image $\{0,1\}$, so compactness alone leaves a gap.
A continuous real-valued function on a connected domain takes the values −2 and 5. It must also take the value 3.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
The continuous image is connected in ℝ, hence an interval. Since −2 < 3 < 5, the intermediate value 3 lies in the image. Compactness is not needed for this conclusion.
Let $K$ be nonempty, compact and connected, and let $f:K\to\mathbb R$ be continuous. Prove that $f(K)=[m,M]$ for some attained values $m\leq M$. Explain why both topological hypotheses matter.
The image f(K) is nonempty and compact, so it has a minimum m and maximum M, both attained by f. It is also connected, hence an interval in ℝ. It therefore contains every value between m and M, and no value outside those extrema, giving $f(K)=[m,M]$. Without connectedness, the compact domain {0,1} under the identity map has the gapped image {0,1}. Without compactness, the connected domain (0,1) under the identity map has image (0,1), with neither endpoint attained. If m = M, the closed interval reduces to a single point.
Paths give a stronger kind of connectedness
- A space is path connected when every pair of its points can be joined by a continuous map $\gamma:[0,1]\to X$ with the prescribed endpoints. Every path-connected space is connected: a separation would pull back along a joining path to a separation of $[0,1]$.
- The converse fails. A connected set can have points that no path joins. To see why, we must specify the limiting vertical segment in the topologist’s sine curve; the oscillating graph by itself is path connected.
The closed topologist’s sine curve
- Let $G=\{(x,\sin(1/x)):0<x\leq1\}$, and add the segment $V=\{0\}\times[-1,1]$. The resulting set $S=G\cup V$ is the closure of $G$: every height in $[-1,1]$ is approached by graph points with $x\to0$.
- The graph $G$ is path connected as the continuous image of the interval $(0,1]$. Its closure $S$ is connected, because the closure of a connected set is connected. Also $S$ is closed and bounded in $\mathbb R^2$, hence compact. Adding the full limit segment does not produce a path from it into the graph.
The drawing shows finitely many oscillations; the graph continues oscillating towards $x=0$.
Why the vertical segment cannot be left by a path
- Suppose a path starts on $V$ and ends on $G$. Let $t_0$ be the last time its first coordinate is zero; this time exists because the zero-time set is closed in $[0,1]$. After $t_0$, the first coordinate is positive and tends to zero as $t$ decreases to $t_0$.
- By the intermediate value theorem, the first coordinate takes all sufficiently small positive values arbitrarily close to $t_0$. The corresponding sine values reach both $1$ and $-1$ arbitrarily close to that time. The path’s second coordinate therefore cannot be continuous at $t_0$, a contradiction. Thus $S$ is connected but not path connected.
The closure of every path-connected subset of Euclidean space is path connected.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
The graph G of sin(1/x) for 0 < x ≤ 1 is path connected. Its closure adds the vertical segment at x = 0 and is connected but not path connected. Closure preserves connectedness, but need not preserve path connectedness.
For $G=\{(x,\sin(1/x)):0<x\leq1\}$ and $S=G\cup(\{0\}\times[-1,1])$, compare compactness, connectedness and path connectedness. Explain which property closure can destroy.
The graph G is path connected because its parameter interval (0,1] is path connected and the graph map is continuous. It is not compact: it is bounded but omits its limiting vertical segment and is not closed in ℝ². Its closure S includes the whole segment, so is closed and bounded, hence compact. The closure of a connected set is connected, so S is connected. It is not path connected: after a path’s last zero first-coordinate time, continuity and the intermediate value theorem would force sine heights 1 and −1 arbitrarily near that time, contradicting continuity of the second coordinate. Thus closure can destroy path connectedness while preserving connectedness.
Compactness and connectedness are independent
- In $\mathbb R$, $[0,1]$ has both properties; $\{0,1\}$ is compact but disconnected; $(0,1)$ is connected but not compact; and $\mathbb Q$ has neither property. A singleton is both compact and connected.
- The empty set is also compact and connected: the empty subcover suffices, and no separation into two nonempty parts exists. These conventions do not give it an attained extremum, which is why the extreme-value conclusion needs a nonempty domain.
Compact, connected, both or neither?
Separate compactness from connectedness for subsets of ℝ.
Every compact subset of $\mathbb R$ is connected.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
False. {0, 1} is closed and bounded, hence compact, but it is not connected: the connected subsets of ℝ are intervals.
State the space, then choose the theorem
- Interior, closure and boundary use neighbourhoods in the stated space. Compactness concerns all open covers; “closed and bounded” characterises it in Euclidean space, while the general metric criterion is complete and totally bounded.
- Connectedness rules out a separation of the whole set. Path connectedness is stronger. Continuous images preserve compactness and connectedness; a nonempty compact connected domain gives a continuous real-valued image that is a closed bounded interval.