Compactness and open covers
Every continuous function on [0, 1] has a largest value, while \displaystyle \frac{1}{x} on (0, 1] has none. What does the closed interval have that the half-open one lacks, stated with open sets alone so that it makes sense in every space?
Where you are 5 in Connectedness and compactness
Why this exists
Every continuous function on [a, b] reaches a largest value (The extreme value theorem), and Open, closed and compact sets in R traced this to sequences: in [a, b] every sequence has a subsequence converging in the set. This lesson states the property with open sets alone: every cover by open sets has a finite part that still covers. Like connectedness (Connectedness), it survives continuous maps, and with the Hausdorff property (Hausdorff spaces and separation) it makes continuous bijections homeomorphisms.
Definition
Throughout, X is a topological space.
Open covers. A cover of A \subseteq X is a family of sets U_i, i \in I, with A \subseteq \bigcup_{i \in I} U_i. It is an open cover if every U_i is open. A subcover is a subfamily U_j, j \in J, with J \subseteq I, that still covers A; it is finite if J is finite.
Compact. X is compact if every open cover of X has a finite subcover. A subset A \subseteq X is compact if it is compact as a space with the subspace topology (Subspace and product topologies).
Every finite space is compact: choose one member of the cover containing each point.
Finite intersection property. A family of sets C_i, i \in I, has the finite intersection property if every finite subfamily has nonempty intersection: C_{i_1} \cap \cdots \cap C_{i_n} \ne \varnothing for all i_1, \dots, i_n \in I.
Sequentially compact. A metric space (X, d) (Metric spaces, a first look) is sequentially compact if every sequence in X has a subsequence converging to a point of X. This is what Open, closed and compact sets in R and Metric spaces, a first look call compact.
Intuition
For 0 < x \le 12 let \displaystyle U_x = (\frac{x}{2}, \frac{3x}{2}). Each x lies in its own U_x, so these open sets cover (0, 12]. But each reaches back only to half its own x, so finitely many of them stop short of 0 (Figure 1). The point the cover needs, and cannot get, is the one the set lacks.
A set that runs off to infinity fails the same way: the intervals (-n, n) cover it with no finite subcover.
Results
Theorem 1 (covering by open sets of X). A \subseteq X is compact if and only if every family of open sets of X whose union contains A has a finite subfamily whose union contains A.
Theorem 2 (closed sets). X is compact if and only if every family of closed subsets of X with the finite intersection property has nonempty intersection. In particular, if X is compact and C_1 \supseteq C_2 \supseteq \cdots are nonempty closed sets, then \bigcap_{n \ge 1} C_n \ne \varnothing.
Theorem 3 (closed subsets). A closed subset of a compact space is compact.
Theorem 4 (compact sets in Hausdorff spaces). Let X be Hausdorff, K \subseteq X compact and x \in X \setminus K. Then there are disjoint open sets U and V with K \subseteq U and x \in V. Hence every compact subset of a Hausdorff space is closed.
Theorem 5 (continuous images). If f : X \to Y is continuous and A \subseteq X is compact, then f(A) is compact. Hence compactness is a topological property.
Theorem 6 (compact to Hausdorff). Let X be compact, Y Hausdorff and f : X \to Y continuous. Then f is closed: f(C) is closed in Y for every closed C \subseteq X. If f is also a bijection, it is a homeomorphism.
Theorem 7 (extreme values). If X is compact and nonempty and f : X \to \mathbb{R} is continuous, there are c, d \in X with f(c) \le f(x) \le f(d) for every x \in X.
Theorem 8 (metric spaces). A metric space is compact if and only if it is sequentially compact. The same holds for a subset with the restricted metric.
Theorem 9 (the line). A set K \subseteq \mathbb{R} is compact if and only if it is closed and bounded. In particular every interval [a, b] is compact.
Why it is true
Theorem 1. Proof. The open sets of A are the sets A \cap U with U open in X. Let A be compact and U_i, i \in I, open in X with A \subseteq \bigcup U_i. The sets A \cap U_i form an open cover of A, so finitely many cover A, and then A \subseteq U_{i_1} \cup \cdots \cup U_{i_n}. Conversely, assume the condition and let W_i, i \in I, be an open cover of the space A. Choose U_i open in X with W_i = A \cap U_i. Their union contains A, so some U_{i_1}, \dots, U_{i_n} do, and the sets A \cap U_{i_k} = W_{i_k} cover A. \square
Theorem 2. Proof. For closed sets C_i the sets U_i = X \setminus C_i are open, and by De Morgan’s law (Set identities: distributive and De Morgan laws) the U_i with i \in J cover X exactly when \bigcap_{i \in J} C_i = \varnothing, for every nonempty J \subseteq I. Let X be compact and let the C_i have the finite intersection property. If \bigcap C_i were empty, the U_i would cover X, so would finitely many, and the matching finite intersection would be empty. Conversely, assume the condition and let U_i be an open cover with no finite subcover. Then the closed sets C_i = X \setminus U_i have the finite intersection property, and a point of \bigcap C_i lies in no U_i: a contradiction. For nested sets, the intersection of C_{n_1}, \dots, C_{n_k} is C_N with N the largest index. \square
Theorem 3. Proof. Let C \subseteq X be closed and U_i, i \in I, open in X with C \subseteq \bigcup U_i. Together with the open set X \setminus C they cover X, so finitely many of these sets do. The set X \setminus C contains no point of C, so the finitely many U_i among them cover C. Theorem 1 finishes. \square
Theorem 4. Proof. If K = \varnothing take U = \varnothing, V = X. Otherwise, for each a \in K we have a \ne x, so there are disjoint open sets U_a \ni a and V_a \ni x. The U_a cover K, and by Theorem 1 some U_{a_1}, \dots, U_{a_n} do. Put U = U_{a_1} \cup \cdots \cup U_{a_n} and V = V_{a_1} \cap \cdots \cap V_{a_n}, open as a finite intersection. A point of V lies in every V_{a_k}, hence in no U_{a_k}, so U \cap V = \varnothing. For the last sentence, each x \notin K has a neighborhood V missing K, so X \setminus K is open. \square
Theorem 5. Proof. The restriction f|_A is continuous (Theorem 3(b) of Subspace and product topologies), so it is enough to take A = X. Let V_i, i \in I, be open in Y with f(X) \subseteq \bigcup V_i. The open sets f^{-1}(V_i) cover X, so some f^{-1}(V_{i_1}), \dots, f^{-1}(V_{i_n}) do. Each f(x) then lies in some V_{i_k}, so f(X) \subseteq V_{i_1} \cup \cdots \cup V_{i_n}, and Theorem 1 applies. If f is a homeomorphism, f(X) = Y and f^{-1}(Y) = X, so X is compact exactly when Y is. \square
Theorem 6. Proof. A closed C \subseteq X is compact (Theorem 3), so f(C) is compact (Theorem 5) and closed in Y (Theorem 4). A continuous closed bijection is a homeomorphism (Theorem 5(b) of Continuous maps and homeomorphisms). \square
Theorem 7. Proof. f(X) is compact (Theorem 5) and nonempty. The open sets (-n, n), n \ge 1, increase and cover \mathbb{R}, so one of them covers f(X): it is bounded, and s = \sup f(X) exists. If s \notin f(X), each y \in f(X) has \displaystyle y < s - \frac{1}{n} for some n \ge 1, so the increasing open sets \displaystyle (-\infty, s - \frac{1}{n}) cover f(X), one of them does, and \displaystyle s - \frac{1}{N} is a smaller upper bound. So s = f(d) for some d. Applied to -f, this gives c. \square
Theorem 8. Proof. A subset with the restricted metric carries the subspace topology (Theorem 2 of Subspace and product topologies), so take the whole space X. Call x a cluster point of a sequence (x_n) if every ball B(x, r) holds x_n for infinitely many n. Then x is the limit of a subsequence: choose n_1 < n_2 < \cdots with \displaystyle d(x_{n_k}, x) < \frac{1}{k}.
Let X be compact. If a sequence had no cluster point, every x would have a ball B_x holding x_n for only finitely many n. Finitely many B_x cover X, and together they hold x_n for only finitely many n, which is absurd.
Let X be sequentially compact. First, for each r > 0 finitely many balls of radius r cover X. If not, choose x_1, then x_{n + 1} \notin B(x_1, r) \cup \cdots \cup B(x_n, r) for each n; then d(x_m, x_n) \ge r for m \ne n, and no subsequence converges, since two of its terms within \displaystyle \frac{r}{2} of the limit would be within r of each other. Now suppose an open cover U_i has no finite subcover. For each n \ge 1, X is a union of finitely many balls of radius \displaystyle \frac{1}{n}, so one of them, \displaystyle B(c_n, \frac{1}{n}), is covered by no finitely many U_i. A subsequence c_{n_k} \to c. Some U_i contains c, and then a ball B(c, r) \subseteq U_i. For large k, \displaystyle d(c_{n_k}, c) < \frac{r}{2} and \displaystyle \frac{1}{n_k} < \frac{r}{2}, so by the triangle inequality \displaystyle B(c_{n_k}, \frac{1}{n_k}) \subseteq B(c, r) \subseteq U_i, covered by one member: a contradiction. \square
Theorem 9. Proof. By Theorem 8 with d(x, y) = |x - y|, K is compact exactly when every sequence in K has a subsequence converging in K, which by Theorem 7 of Open, closed and compact sets in R happens exactly when K is closed and bounded. \square
Worked examples
Lamps on a causeway
Grace and Mateo are lighting a causeway that runs 12 meters across a pond from the bank at 0. A lamp may stand at any point x with 0 < x \le 12, on a pole whose height grows with x, and it lights the open stretch \displaystyle U_x = (\frac{x}{2}, \frac{3x}{2}). A lamp on the bank lights U_0 = (-1, 1). (a) Show that no finitely many of the lamps U_x with x > 0 light all of (0, 12]. (b) Find the fewest lamps that light [0, 12].
(a) Given U_{x_1}, \dots, U_{x_n} with all x_k > 0, let m be the smallest x_k. The point \displaystyle \frac{m}{2} \in (0, 12] lies in U_x only if \displaystyle \frac{x}{2} < \frac{m}{2}, that is x < m, so no chosen lamp lights it.
(b) The family U_x, 0 \le x \le 12, is an open cover of [0, 12], compact by Theorem 9, so finitely many lamps suffice. The point 0 lies in no U_x with x > 0, so U_0 is needed. Suppose finitely many lamps light [0, 12], and put p_0 = 12. While p_j \ge 1, the point p_j is lit by some chosen U_x with x > 0; put \displaystyle p_{j + 1} = \frac{x}{2}, its left end. From \displaystyle \frac{x}{2} < p_j < \frac{3x}{2}, \displaystyle \frac{p_j}{3} < p_{j + 1} < p_j. The left ends strictly decrease, so the lamps found are distinct and the process stops, at some p_j < 1. Since \displaystyle p_j > \frac{12}{3^j}, we need 3^j > 12, so j \ge 3: three lamps besides U_0. Four lamps do light [0, 12]:
since \displaystyle \frac{3}{4} < 1, \displaystyle 2 < \frac{9}{4}, \displaystyle \frac{11}{2} < 6 and \displaystyle 12 < \frac{33}{2}. The fewest is four lamps.
Two sequences in the lower limit line
In \mathbb{R}_{\ell} (Bases and subbases) let \displaystyle K_+ = \{0\} \cup \{\frac{1}{n} : n \ge 1\} and \displaystyle K_- = \{0\} \cup \{-\frac{1}{n} : n \ge 1\}. Both are compact in \mathbb{R} (Theorem 9). Decide which is compact in \mathbb{R}_{\ell}.
SolutionK_+ is compact. Let open sets of \mathbb{R}_{\ell} cover K_+ (Theorem 1). One of them, U, contains 0, so it contains a basis element [p, q) \ni 0, hence [0, q) with q > 0. Then \displaystyle \frac{1}{n} \in U for every \displaystyle n > \frac{1}{q}, and the finitely many \displaystyle n \le \frac{1}{q} need one member each.
K_- is not compact. The sets [0, 1) and \displaystyle [-\frac{1}{n}, -\frac{1}{n + 1}), n \ge 1, are open in \mathbb{R}_{\ell} and cover K_-. The intervals are pairwise disjoint and [0, 1) holds no negative number, so \displaystyle -\frac{1}{n} lies in \displaystyle [-\frac{1}{n}, -\frac{1}{n + 1}) and in no other member. Every subcover keeps all infinitely many members.
K_- is closed in \mathbb{R}, so in the finer \mathbb{R}_{\ell}, and bounded: Theorem 9 is special to the line. And x \mapsto -x is not continuous on \mathbb{R}_{\ell}, or it would carry K_+ onto a compact K_- (Theorem 5).
A curve traced once
Let f(t) = (t^3 + t, t^2) for t \in [-1, 2] and C = f([-1, 2]) \subseteq \mathbb{R}^2. Show that C is closed in \mathbb{R}^2, that f : [-1, 2] \to C is a homeomorphism, and find the largest and smallest values of x + y on C.
SolutionThe coordinates are polynomials in t, so f is continuous into \mathbb{R}^2 (Theorems 5 and 6(b) of Subspace and product topologies) and into C (Theorem 3(c) there). [-1, 2] is compact (Theorem 9), so C is compact (Theorem 5) and closed in the Hausdorff space \mathbb{R}^2 (Theorem 4).
If f(s) = f(t), then
and \displaystyle s^2 + st + t^2 + 1 = (s + \frac{t}{2})^2 + \frac{3t^2}{4} + 1 > 0, so s = t: f is a bijection onto C. Disjoint open sets of \mathbb{R}^2 meet C in disjoint open sets of C, so C is Hausdorff, and Theorem 6 makes f a homeomorphism, with no need to solve the cubic t^3 + t = x for the inverse.
On C, x + y = t^3 + t^2 + t, whose derivative 3t^2 + 2t + 1 has discriminant 4 - 12 < 0, so it is positive and x + y increases. The extremes Theorem 7 promises are -1 at f(-1) = (-2, 1) and 14 at f(2) = (10, 4).
Two origins, two compact sets
In the line with two origins X = \mathbb{R} \cup \{0^*\} (Example 6 of Hausdorff spaces and separation), let K = [-1, 1] and K^* = ([-1, 1] \setminus \{0\}) \cup \{0^*\}. Show that K and K^* are compact, neither is closed, and K \cap K^* is not compact.
SolutionBy Theorem 1(b) of Subspace and product topologies, K has the basis of sets (p, q) \cap K and I^*(p, q) \cap K = ((p, q) \setminus \{0\}) \cap K. These are open in [-1, 1], and the first kind alone gives its standard topology, so K is compact by Theorem 9. The map h : X \to X exchanging 0 and 0^* and fixing all else exchanges (p, q) and I^*(p, q) when p < 0 < q and fixes the other basis elements, so it is a homeomorphism, and K^* = h(K) is compact by Theorem 5.
Every basis element containing 0^* is some I^*(p, q), which contains points of (0, \min\{q, 1\}) \subseteq K. So 0^* \in \overline{K} \setminus K (Theorem 3 of Closed sets, closure and limit points); applying h, 0 \in \overline{K^*} \setminus K^*.
K \cap K^* = [-1, 1] \setminus \{0\}. The open sets \displaystyle W_n = (-\infty, -\frac{1}{n}) \cup (\frac{1}{n}, \infty), n \ge 1, cover it and increase, so finitely many of them make W_N, which misses \displaystyle \frac{1}{N + 1}.
Nested compact sets
Let X be Hausdorff and K_1 \supseteq K_2 \supseteq \cdots nonempty compact subsets. Prove that \bigcap_{n \ge 1} K_n is nonempty, and that if U is open and contains it, then K_N \subseteq U for some N.
SolutionProof. Each K_n is closed in X (Theorem 4), so K_n \setminus U = K_n \cap (X \setminus U) is closed in X, hence in K_1 (Theorem 1(c) of Subspace and product topologies). These sets are nested with intersection \big(\bigcap K_n\big) \setminus U = \varnothing. As K_1 is compact, Theorem 2 gives a finite subfamily with empty intersection, which is K_N \setminus U for its largest index N. So K_N \subseteq U. Taking U = \varnothing: if \bigcap K_n were empty, some K_N would be empty. \square
Hausdorff is needed. In \mathbb{N} with the finite-complement topology every subset is compact (one member of a cover misses only finitely many points), and the sets K_n = \{m \in \mathbb{N} : m \ge n\} are nonempty and nested with empty intersection.
Boundaries
- Hausdorff matters. Without it compact sets need not be closed, and their intersections and nested sequences misbehave (Examples 4 and 5).
- Closed and bounded is special to \mathbb{R}^n. It fails in \mathbb{R}_{\ell} (Example 2) and in general metric spaces (Metric spaces, a first look).
- Compactness depends on the topology. A finer topology has fewer compact sets: apply Theorem 5 to the identity map. Without compactness a continuous bijection need not be a homeomorphism (Example 4 of Continuous maps and homeomorphisms).
- Sequences. In metric spaces compactness is sequential compactness (Theorem 8). In general spaces neither implies the other ([0, 1]^{[0, 1]} is compact but not sequentially compact; beyond this lesson). Exercise 14 is what survives.
Where this goes next
Compactness in products and the Heine-Borel theorem proves that finite products of compact spaces are compact, that closed bounded subsets of \mathbb{R}^n are compact, and the Lebesgue number lemma. In Classification of surfaces a closed surface is a compact connected surface.
Exercises
Covers and the finite intersection property
- Mateo lights a second causeway, 20 meters long, where a lamp at x, 0 < x \le 20, lights \displaystyle (\frac{2x}{3}, \frac{4x}{3}) and the bank lamp lights (-2, 2). (a) Show that no finitely many lamps with x > 0 light (0, 20]. (b) Find the fewest lamps that light [0, 20], with positions that work.
- For each family of closed sets, decide whether it has the finite intersection property, find its intersection, and say whether Theorem 2 shows the space is not compact. (a) [n, \infty), n \ge 1, in \mathbb{R}; (b) \displaystyle (0, \frac{1}{n}], n \ge 1, in (0, 1]; (c) \displaystyle [-\frac{1}{n}, \frac{1}{n}], n \ge 1, in \mathbb{R}.
- Which subsets of \mathbb{R} are compact? For each that is not, give an open cover with no finite subcover. (a) \{x : x^3 - x^2 - 6x \le 0\}; (b) \displaystyle \{x \in [0, 2\pi] : \sin x \ge \frac{1}{2}\}; (c) \{e^{-x} : x \ge 0\}; (d) \{x \in \mathbb{Q} : x^2 \le 2\}.
Other topologies
- The ray topology on \mathbb{R} has as open sets \varnothing, \mathbb{R} and the rays (a, \infty), a \in \mathbb{R}. (a) Show that a nonempty A \subseteq \mathbb{R} is compact if and only if it has a least element. (b) Which are compact: [3, 5), (3, 5], \mathbb{N}, \mathbb{Z}, \displaystyle \{\frac{1}{n} : n \ge 1\}, \displaystyle \{-\frac{1}{n} : n \ge 1\}? (c) Is [3, 5) closed? Which hypothesis of Theorem 4 fails?
- On \mathbb{N} let the open sets be \mathbb{N} and every subset not containing 0. (a) Show that this is a topology and find its closed sets. (b) Show that A is compact if and only if 0 \in A or A is finite. (c) Find a compact set that is not closed, and a closed set with a subset that is neither compact nor closed.
Maps and extreme values
- Explain why f(x, y) = x^2 y has a largest and a smallest value on S^1, and find them and where they occur.
- Explain why some point of P = \{(t, t^2) : -2 \le t \le 2\} is nearest to (0, 2), and find the distance and the nearest points.
- Let X = [0, 1) \cup [2, 3] and f(t) = t for t < 1, f(t) = t - 1 for t \ge 2. Show that f : X \to [0, 2] is a continuous bijection but not a homeomorphism. Which hypothesis of Theorem 6 fails?
Proofs
- Prove that a union of finitely many compact subsets of X is compact.
- Let \mathcal{B} be a basis for X. Prove that X is compact if and only if every cover of X by members of \mathcal{B} has a finite subcover.
- Prove that two disjoint compact subsets A, B of a Hausdorff space lie in disjoint open sets.
- Let (X, d) be a nonempty compact metric space and f : X \to X satisfy d(f(x), f(y)) < d(x, y) whenever x \ne y. Prove that f has exactly one fixed point.
- In the cocountable topology on \mathbb{R} (Exercise 14 of Hausdorff spaces and separation), prove that a subset is compact if and only if it is finite.
- Prove that every infinite subset of a compact space has a limit point.
- Let X be compact, Y Hausdorff, and f : X \to Y a continuous surjection. Prove that V \subseteq Y is open if and only if f^{-1}(V) is open in X.
Answers
- (a) If m is the least chosen x, the point \displaystyle \frac{2m}{3} is lit only by lamps with x < m. (b) 5. Only the bank lamp lights 0. As in Example 1, each p \ge 2 lit by \displaystyle (\frac{2x}{3}, \frac{4x}{3}) gives a left end p' with \displaystyle \frac{p}{2} < p' < p, so the chain from 20 stays above \displaystyle \frac{20}{2^j} and needs 2^j > 10, j \ge 4. Lamps at 0, 2.7, 5.1, 10, 18 work: (-2, 2), (1.8, 3.6), (3.4, 6.8), \displaystyle (\frac{20}{3}, \frac{40}{3}), (12, 24), each left end below the previous right end.
- (a) Yes: finite intersections are [N, \infty). Intersection \varnothing. So \mathbb{R} is not compact. (b) Yes: the sets are closed in (0, 1], being \displaystyle (0, 1] \cap [0, \frac{1}{n}], and finite intersections are \displaystyle (0, \frac{1}{N}]. Intersection \varnothing. So (0, 1] is not compact. (c) Yes, intersection \{0\}: no conclusion.
- (a) x(x - 3)(x + 2) \le 0 gives (-\infty, -2] \cup [0, 3]: not bounded, not compact; cover by (-n, n), n \ge 1. (b) \displaystyle [\frac{\pi}{6}, \frac{5\pi}{6}]: compact. (c) (0, 1]: not closed, not compact; the sets \displaystyle (\frac{1}{n}, 2) cover it, and finitely many make \displaystyle (\frac{1}{N}, 2), missing \displaystyle \frac{1}{N}. (d) Not closed, not compact. The increasing open sets \displaystyle \{x : |x^2 - 2| > \frac{1}{n}\} cover it, as x^2 \ne 2 for rational x, but the N-th misses decimal truncations q of \displaystyle \sqrt{2} with \displaystyle 2 - q^2 \le \frac{1}{N}.
- (a) If m = \min A, a member containing m is \mathbb{R} or a ray (a, \infty) with a < m, and contains A. If A has no least element and s = \inf A is finite, the increasing rays \displaystyle (s + \frac{1}{n}, \infty) cover A, but each misses points of A; if A is unbounded below, use (-n, \infty). (b) Compact: [3, 5), \mathbb{N}, \displaystyle \{-\frac{1}{n}\}. Not compact: (3, 5], \mathbb{Z}, \displaystyle \{\frac{1}{n}\}. (c) No: closed sets are \varnothing, \mathbb{R} and the (-\infty, a]. Not Hausdorff: nonempty open sets meet.
- (a) A union containing \mathbb{N} is \mathbb{N}, otherwise it misses 0; a finite intersection with a member other than \mathbb{N} misses 0. Closed sets: \varnothing and every set containing 0. (b) If 0 \in A, a member containing 0 is \mathbb{N}. If 0 \notin A, the open sets \{a\}, a \in A, cover A, and only a finite A has a finite subcover. (c) \{1\} is compact, not closed. \{0, 2, 4, 6, \dots\} is closed; its subset \{2, 4, 6, \dots\} is neither.
- S^1 = e([0, 1]) with e(s) = (\cos 2\pi s, \sin 2\pi s) continuous, so S^1 is compact (Theorems 9 and 5) and Theorem 7 applies. On S^1, f = y - y^3 with y \in [-1, 1]; the derivative 1 - 3y^2 vanishes at \displaystyle y = \pm\frac{\sqrt{3}}{3}, and f = 0 at y = \pm 1. Largest \displaystyle \frac{2\sqrt{3}}{9} at \displaystyle (\pm\frac{\sqrt{6}}{3}, \frac{\sqrt{3}}{3}), smallest \displaystyle -\frac{2\sqrt{3}}{9} at \displaystyle (\pm\frac{\sqrt{6}}{3}, -\frac{\sqrt{3}}{3}).
- P is the image of [-2, 2] under t \mapsto (t, t^2), so compact, and the distance to (0, 2) is continuous: Theorem 7. Its square is t^2 + (t^2 - 2)^2 = u^2 - 3u + 4 with u = t^2 \in [0, 4], least at \displaystyle u = \frac{3}{2}, value \displaystyle \frac{7}{4} (the ends give 4 and 8). Distance \displaystyle \frac{\sqrt{7}}{2}, at \displaystyle (\pm\frac{\sqrt{6}}{2}, \frac{3}{2}).
- \displaystyle [0, 1) = X \cap (-1, \frac{3}{2}) and \displaystyle [2, 3] = X \cap (\frac{3}{2}, 4) are open in X, so f is continuous by the pasting lemma (Theorem 3(d) of Subspace and product topologies). It maps [0, 1) onto [0, 1) and [2, 3] onto [1, 2], a bijection. The inverse is not continuous: [0, 2] is connected and X is not (Theorem 4 of Connectedness). X is not compact: 1 \notin X is a limit point of X, so X is not closed in \mathbb{R} (Theorem 9).
- Proof. Let K_1, \dots, K_n be compact and let open sets of X cover their union. By Theorem 1 each K_j is covered by a finite subfamily F_j, and F_1 \cup \cdots \cup F_n is finite and covers the union. \square
- Proof. Members of \mathcal{B} are open, so compactness gives the condition. Conversely, let U_i be an open cover. Each x lies in some U_i, and some B \in \mathcal{B} has x \in B \subseteq U_i. These basis elements cover X, so finitely many B_1, \dots, B_n do, and choosing U_{i_k} \supseteq B_k gives a finite subcover. \square
- Proof. If B = \varnothing take X and \varnothing. Each b \in B is outside A, so Theorem 4 gives disjoint open U_b \supseteq A and V_b \ni b. Finitely many V_{b_1}, \dots, V_{b_n} cover B. Then U = U_{b_1} \cap \cdots \cap U_{b_n} \supseteq A is open and misses V = V_{b_1} \cup \cdots \cup V_{b_n} \supseteq B. \square
- Proof. Let g(x) = d(x, f(x)). By the triangle inequality |g(x) - g(y)| \le d(x, y) + d(f(x), f(y)), at most 2d(x, y), so g is continuous (Theorem 4 of Continuous maps and homeomorphisms) and by Theorem 7 has a least value at some c. If f(c) \ne c, then g(f(c)) = d(f(c), f(f(c))) < d(c, f(c)) = g(c), a contradiction. Two fixed points x \ne y would give d(x, y) = d(f(x), f(y)) < d(x, y). \square
- Proof. Finite sets are compact. Let A be infinite, choose distinct a_1, a_2, \ldots \in A and let D = \{a_k : k \ge 1\}. Each U_n = (\mathbb{R} \setminus D) \cup \{a_n\} has countable complement D \setminus \{a_n\} (Countable sets), so it is open. The U_n cover \mathbb{R}, but U_1 \cup \cdots \cup U_N misses a_{N + 1} \in A. \square
- Proof. Suppose A \subseteq X has no limit point. Then each x \in X has an open U_x \ni x with U_x \cap A \subseteq \{x\}. Finitely many U_{x_1}, \dots, U_{x_n} cover X, so A \subseteq \{x_1, \dots, x_n\} is finite. \square
- Proof. If V is open, so is f^{-1}(V). If f^{-1}(V) is open, then C = f^{-1}(Y \setminus V) is closed, so f(C) is closed (Theorem 6), and f(C) = Y \setminus V because f is surjective (Theorem 1 of Continuous maps and homeomorphisms). So V is open. Such an f is a quotient map (Quotient spaces, or gluing). \square