Altanis

3.5Compact Subspaces of the Real Line

Updated 20 Jul 2026Chapter (PDF)

[3.5.1]Theorem(Closed Intervals are Compact in Order Topology)#

Let XX be a totally ordered set with the least upper bound property. In its order topology, closed intervals [a,b]X[a, b] \subseteq X are compact subsets of XX.

Proof.

We split this proof into four steps. For setup, let a<bXa < b \in X, and consider the interval [a,b][a, b] with an arbitrary open cover A\mathcal{A} of [a,b][a,b], where the open sets are taken from the subspace topology on [a,b][a, b] (which coincides with the order topology). We show there is a finite subcover of A\mathcal{A}.

  1. First we show that, for any x[a,b)x \in [a, b), there is some y>xy > x such that [x,y][x, y] can be covered by, at most, two elements of A\mathcal{A}. If yy is the immediate successor to xx, then note that [x,y]={x,y}[x, y] = \{x, y\}, so there is certainly one element of A\mathcal{A} that contains xx and one element (not necessarily distinct) of A\mathcal{A} that contains yy. Now suppose yy is not the immediate successor to xx. There is some AAA \in \mathcal{A} such that xAx \in A. Since AA is open, there is some c[a,b]c \in [a, b] such that x[x,c)Ax \in [x, c) \subseteq A. Then choose y(x,c)y \in (x, c), and so [x,y]A[x, y] \subseteq A.

  2. Let CC be the set of points y>ay > a in [a,b][a, b] such that [a,y][a, y] can be covered by finitely many elements of A\mathcal{A}. Then the set CC is nonempty by (1)(1), using x=ax = a (there is definitely some yy such that [a,y][a, y] can be covered by up to two elements of A\mathcal{A}).

  3. Let c=sup(C)c = \sup(C)—we seek to show cCc \in C. Let AAA \in \mathcal{A} be some open set containing cc; then there is some d[a,b]d \in [a, b] such that c(d,c]Ac \in (d, c] \subseteq A. For the sake of contradiction, suppose cCc \notin C. Then there is some point zz of CC such that z(d,c)z \in (d, c), since otherwise dd would be a smaller upper bound for CC. Since zCz \in C, there are finitely many sets from A\mathcal{A} that covers [a,z][a, z], say nn. Then [z,c]A[z, c] \subseteq A, so [a,c]=[a,z][z,c][a, c] = [a, z] \cup [z, c] can be covered by, at most, n+1n + 1 sets, which is finite. Thus cCc \in C.

  4. Finally, we show c=bc = b. Suppose, for the sake of contradiction, that c<bc < b. Then, by (1)(1), there is some y(c,b]y \in (c, b] such that [c,y][c, y] is covered by up to two sets from A\mathcal{A}. Then [a,y]=[a,c][c,y][a, y] = [a, c] \cup [c, y] can be covered by finitely many elements, so yCy \in C, contradicting that cc is an upper bound for CC.

Thus [a,b][a, b] can be covered by finitely many elements from A\mathcal{A}, completing the proof.

[3.5.2]Remark(Intuition Behind Proof)#

This proof is quite long, so we break it down intuitively. Consider any closed interval [a,b][a, b] from the space XX in its order topology with the least upper bound property. Let A\mathcal{A} be an open cover for [a,b][a, b] comprising subsets of [a,b][a, b] that are open in its subspace topology (which coincides with its order topology). First we establish that, for any point x[a,b)x \in [a, b), there is some element of this interval next to xx such that the area between them ([x,y][x, y]) can be covered by finitely many members of A\mathcal{A}. Obviously we exclude bb, since nothing comes after bb. Knowing this, we know that there is some element after aa such that the area between them can be covered by finitely many members of A\mathcal{A}. We let CC be all the elements of [a,b][a, b] that satisfy this, and since it is nonempty, the value c=sup(C)c = \sup(C) is well-defined. For any member of C\mathcal{C} that contains cc, since it is open, there is some nontrivial intersection with CC, say at zCz \in C. By definition of CC, we know [a,z][a, z] can be finitely covered, and since [z,c]A[z, c] \subseteq A, the entire interval [a,c]=[a,z][z,c][a, c] = [a, z] \cup [z, c] can be finitely covered, and so cCc \in C. Finally, to show c=bc = b, we note that if c<bc < b, then there is some y(c,b]y \in (c, b] such that [c,y][c, y] is finitely covered, and so [a,y]=[a,c][c,y][a, y] = [a, c] \cup [c, y] is finitely covered, contradicting that cc is an upper bound for CC. Thus c=bc = b, completing the proof.

[3.5.3]Corollary#

Every closed interval in R\bR is compact.

[3.5.4]Theorem(Compact iff. Closed and Bounded)#

A subspace ARnA \subseteq \bR^n is compact if and only if it is closed and bounded under either the square or Euclidean metric.

Proof.

Note that ρ(x,y)d(x,y)nρ(x,y)\rho(x, y) \le d(x, y) \le \sqrt{n} \rho(x, y) for x,yRnx, y \in \bR^n, so we only consider the square metric, since AA is bounded under the Euclidean metric if and only if it is bounded under the square metric.

()(\Longrightarrow): Suppose AA is compact. Since it is a compact subspace of a Hausdorff space, it is automatically closed, so we simply show AA is bounded. Let {Bρ(0n,m)}mZ+\{B_{\rho}(\vb{0}_n, m)\}_{m \in \bZ_+} be an open cover for Rn\bR^n—then a finite subcover of this covers AA. Then a box in Rn\bR^n can contain the entirety of AA, implying it is bounded.

()(\Longleftarrow): Suppose AA is closed and bounded under ρ\rho. Let NRN \in \bR be such that ρ(x,y)N\rho(x, y) \le N for every x,yAx, y \in A. Choose some x0Ax_0 \in A, then let ρ(x0,0n)=b\rho(x_0, \vb{0}_n) = b. Then, for any xAx \in A, note that

ρ(x,0)ρ(x,x0)+ρ(x0,0)=N+b.\rho(x, \vb{0}) \le \rho(x, x_0) + \rho(x_0, 0) = N + b.

Then AA is a closed subset of the compact subspace Bρ(0,N+b)B_{\rho}(\vb{0}, N + b), implying AA is compact.

[3.5.5]Theorem(Extreme Value Theorem)#

Suppose f:XYf: X \to Y is a continuous map from a compact set XX to an ordered set YY in its order topology. Then there are c,dXc, d \in X such that f(c)f(x)f(d)f(c) \le f(x) \le f(d) for every xXx \in X.

Proof.

Consider A=im(f)=f(X)A = \im(f) = f(X), a compact set by continuity. We show that AA contains a maximum MM and minimum mm, which completes the proof when taking cf1({m})c \in f^{-1}(\{m\}) and df1({M})d \in f^{-1}(\{M\}). We show that AA contains a maximum and omit the proof for a minimum, as it is entirely symmetric.

For the sake of contradiction, suppose AA has no maximum. Then note A={(,a)}aA\mathcal{A} = \{(-\infty, a)\}_{a \in A} is a collection of open subsets of XX. For any yAy \in A, note there is some aAa \in A such that y<ay < a, and so y(,a)y \in (-\infty, a). Thus A\mathcal{A} forms an open cover for f(X)f(X).Take a finite subcover {(,ak)}k=1n\{(-\infty, a_k)\}_{k = 1}^n with each akAa_k \in A. Choosing a=max{ak}k=1na = \max\{a_k\}_{k = 1}^n, note that aAa \in A but aa is not covered by this finite subcover, a contradiction. Thus AA has a maximum.

For X,Y=RX, Y = \bR, note this is the Extreme Value Theorem from calculus.

We now prove a theorem about uniform continuity: that a continuous map from a compact metric space to a target metric space is uniformly continuous. In the process, we will use something known as a Lebesgue number and the Lebesgue covering lemma.

[3.5.6]Definition(Distance from Point to Metric Space)#

Suppose (X,d)(X, d) is a metric space, and let AXA \subseteq X is nonempty. Then, for any xXx \in X, we define

d(x,A)=infaA{d(x,a)}.d(x, A) = \inf_{a \in A}\{ d(x, a) \}.

Generally, d(x,A)d(x, A) is the “shortest” distance from a point to any member of a set AA. Contrast this with the diameter of AA, which is defined as the “largest” distance between any two points in AA (that is, the diameter is supx,yA{d(x,y)}\sup_{x, y \in A}\{d(x, y)\}).

[3.5.7]Theorem#

For a metric space (X,d)(X, d) and a fixed, nonempty subset AXA \subseteq X, the function d(,A)d(-, A) is continuous.

Proof.

Fix x,yXx, y \in X. Then note

d(x,A)d(x,a)d(x,y)+d(y,a)d(x, A) \le d(x, a) \le d(x, y) + d(y, a)

for each aAa \in A. It follows that

d(x,A)d(x,y)infaA{d(y,a)}=d(y,A)    d(x,A)d(y,A)d(x,y).d(x, A) - d(x, y) \le \inf_{a \in A}\{d(y, a)\} = d(y, A) \implies d(x, A) - d(y, A) \le d(x, y).

Switching the roles of x,yx, y, we arrive at the inequality

d(x,A)d(y,A)d(x,y).|d(x, A) - d(y, A)| \le d(x, y).

For any ε>0\epsilon > 0, choosing δ=ε\delta = \epsilon proves d(,A)d(-, A) is continuous by the εδ\epsilon-\delta characterization of continuity.

[3.5.8]Theorem(Lebesgue Covering Lemma)#

Let (X,d)(X, d) be a compact metric space together with any arbitrary open cover A\mathcal{A}. Then there exists some δ>0\delta > 0, called a Lebesgue number, such that every nonempty subset of XX with diameter less than δ\delta is contained in some member of the cover.

Proof.

Let (X,d)(X, d) be a compact metric space, and let A\mathcal{A} be an open cover of XX. Then take a finite subcover {Ak}k=1n\{A_k\}_{k = 1}^n. For each k{1,,n}k \in \{1, \dots, n\}, define Ck=XAkC_k = X \setminus A_k. Then define a function

f:XRf(x)=1nk=1nd(x,Ck).f: X \to \bR \quad f(x) = \frac{1}{n} \sum_{k = 1}^n d(x, C_k).

We show f(x)>0f(x) > 0 for every xXx \in X. For fixed xXx \in X, there is some kk such that xAkx \in A_k. Since AkA_k is open, we may choose some ε>0\epsilon > 0 such that xB(x,ε)Akx \in B(x, \epsilon) \subseteq A_k. Then d(x,Ck)εd(x, C_k) \ge \epsilon, and so f(x)ε/nf(x) \ge \epsilon/n.

Note ff is a constant multiple of a sum of continuous functions, and so ff is continuous. Thus ff attains a minimum value δ\delta, which is greater than 00 by the previous statement, by the Extreme Value Theorem, which we will show is our desired Lebesgue number.

Let BXB \subseteq X be a nonempty subset whose diameter is less than δ\delta. Choose some x0Bx_0 \in B, then note BB lies in the δ\delta-neighborhood of x0x_0. Thus

δf(x0)d(x0,Cm),\delta \le f(x_0) \le d(x_0, C_m),

where CmC_m is chosen such that it is the maximum of d(x0,Ck)d(x_0, C_k) for each kk. Then the δ\delta-neighborhood of x0x_0 is contained in the element Am=XCmA_m = X \setminus C_m of the covering A\mathcal{A}, meaning BB is as well.

Recall the definition of uniform continuity.

[3.5.9]Definition(Uniform Continuity)#

Suppose (X,dX)(X, d_X) and (Y,dY)(Y, d_Y) are metric spaces. A map f:XYf: X \to Y is said to be uniformly continuous if, for every ε>0\epsilon > 0, there exists some δ>0\delta > 0 such that for every x1,x2Xx_1, x_2 \in X, it follows that

dX(x1,x2)<δ    dY(f(x1),f(x2))<ε.d_X(x_1, x_2) < \delta \implies d_Y(f(x_1), f(x_2)) < \epsilon.
[3.5.10]Theorem(Uniform Continuity Theorem)#

Suppose f:XYf: X \to Y is a continuous map from the compact metric space (X,dX)(X, d_X) to a target metric space (Y,dY)(Y, d_Y). Then ff is uniformly continuous.

Proof.

Fix ε>0\epsilon > 0. Let A={BdY(y,ε/2)}yY\mathcal{A} = \{B_{d_Y}(y, \epsilon/2)\}_{y \in Y} be an open cover for YY, and consider the inverse image f1(A)f^{-1}(\mathcal{A}), which forms an open cover for XX. Since XX forms a compact metric space, there is a Lebesgue number δ>0\delta > 0. For any set {x1,x2}\{x_1, x_2\} whose diameter is less than δ\delta, there is some AAA \in \mathcal{A} such that {x1,x2}A\{x_1, x_2\} \subseteq A. Note A=f1(BdY(y,ε/2))A = f^{-1}(B_{d_Y}(y, \epsilon/2)) for some yYy \in Y, and so f(x1),f(x2)BdY(y,ε/2)f(x_1), f(x_2) \in B_{d_Y}(y, \epsilon/2). Thus dY(f(x1),f(x2))<εd_Y(f(x_1), f(x_2)) < \epsilon for any pair x1,x2Xx_1, x_2 \in X, completing the proof.

We now prove that the reals are uncountable. Namely, we use the order properties of R\bR, avoiding decimals and binary expansions and the like.

[3.5.11]Definition(Isolated Point)#

Let XX be a topological space. A point xXx \in X is said to be isolated if {x}\{x\} is open. Equivalently, xx is isolated if it is not a limit point of XX, or if there exists an open neighborhood OXO \subseteq X of xx that has no other points.

[3.5.12]Theorem#

Suppose XX is a nonempty, compact, Hausdorff space. If XX has no isolated points, then it is uncountable.

Proof.

First, we claim that for any nonempty open UXU \subseteq X and xXx \in X, there is a nonempty open VUV \subseteq U such that xVx \notin \bar{V}. Choose yUy \in U distinct from xx. (If xUx \in U, yy exists because xx is not isolated; if xUx \notin U, yy exists because UU is nonempty.) Since XX is Hausdorff, there are disjoint open neighborhoods W1W_1 and W2W_2 of xx and yy. Let V=W2UV = W_2 \cap U. Then VUV \subseteq U is nonempty and open (since yVy \in V). Since W1W_1 is an open neighborhood of xx disjoint from VV, xVx \notin \bar{V}.

To show XX is uncountable, let f:Z+Xf: \bZ_+ \to X be any map with f(n)=xnf(n) = x_n. We show ff is not surjective. For n=1n = 1, applying the claim to U=XU = X allows us to choose a nonempty open V1XV_1 \subseteq X such that x1V1x_1 \notin \bar{V}_1. Inductively, given Vn1V_{n-1}, applying the claim to U=Vn1U = V_{n-1} allows us to choose a nonempty open VnVn1V_n \subseteq V_{n-1} such that xnVnx_n \notin \bar{V}_n.

Thus, V1V2\bar{V}_1 \supseteq \bar{V}_2 \supseteq \cdots is a sequence of nonempty closed sets with the finite intersection property. By compactness, nZ+Vn\bigcap_{n \in \bZ_+} \bar{V}_n is nonempty, so it contains some element xx. Since xnVnx_n \notin \bar{V}_n for all nn, xxnx \neq x_n. Thus xf(Z+)x \notin f(\bZ_+), meaning ff is not surjective and XX is uncountable.

[3.5.13]Corollary#

Every nonempty compact subset of R\bR is uncountable.

[3.5.14]Recap#
  1. Let XX be a totally ordered set with the least upper bound property. In its order topology, closed intervals of XX are compact subspaces.

    Intuition.

    Let [a,b]X[a, b] \subseteq X be a closed interval, and let A\mathcal{A} be an open cover for this interval, where it takes open subsets of [a,b][a, b] in its subspace topology. First, we show that for every x[a,b)x \in [a, b), there is some y>xy > x such that [x,y][a,b][x, y] \subseteq [a, b] can be covered by finitely many members of the cover A\mathcal{A} (in fact, by up to two!).

    Then we consider starting at aa. Let CC be the set of all points y>ay > a such that [a,y][a,b][a, y] \subseteq [a, b] can be covered by finitely many members of A\mathcal{A}. We are guaranteed that CC \ne \emptyset, since letting x=ax = a and applying the previous logic proves there is some element in CC. Since it is non-empty and bounded, there is some c=sup(C)c = \sup(C). Let AAA \in \mathcal{A} be a member of the cover such that cAc \in A. By AA being open, it follows there is some d<cd < c such that c(d,c]Ac \in (d, c] \subseteq A. Note that there is some zCz \in C such that d<z<cd < z < c, otherwise dd would be a smaller upper bound than CC. Then [a,c]=[a,z][z,c][a, c] = [a, z] \cup [z, c], where the first part is covered by finitely many members of A\mathcal{A} as per the definition of CC, and the second part is covered by finitely many members of A\mathcal{A} since [z,c]A[z, c] \subseteq A—thus cCc \in C.

    Now we show c=bc = b. For the sake of contradiction, suppose c<bc < b. Following the logic of the previous paragraph, there is some AAA \in \mathcal{A} such that cAc \in A, and since AA is open, there is some x(c,b]x \in (c, b] such that [c,x]A[c, x] \subseteq A. Then [a,x]=[a,c][c,x][a, x] = [a, c] \cup [c, x] can be finitely covered, meaning xCx \in C, contradicting that cc is an upper bound. Thus c=bc = b.

  2. As a consequence, taking X=RX = \bR means are closed intervals in R\bR are compact.

  3. Compact iff. Closed and Bounded in Rn\bR^n. A subspace ARnA \subseteq \bR^n is compact if and only if it is closed and bounded under the Euclidean or square metrics.

    Intuition.

    Note that a subspace ARnA \subseteq \bR^n is bounded under the Euclidean metric if and only if it is bounded under the square metric, since

    ρ(x,y)d(x,y)nρ(x,y)\rho(x, y) \le d(x, y) \le \sqrt{n} \rho(x, y)

    for every x,yRnx, y \in \bR^n.

    ()(\Longrightarrow): Suppose AA is compact. Then AA is a compact subspace of the Hausdorff space Rn\bR^n, and so AA is closed. Let {Bρ(0n,m)}mZ+\{B_{\rho}(\vb{0}_n, m)\}_{m \in \bZ_+} be an open cover for Rn\bR^n, then choose a finite subcover that covers AA. Then the union of all members of the finite subcover is contained in a box, and since AA is inside this box, AA is bounded as well.

    ()(\Longleftarrow): Suppose AA is closed and bounded under ρ\rho. Suppose that NRN \in \bR is such that ρ(x,y)N\rho(x, y) \le N for each x,yAx, y \in A. Fix some x0Ax_0 \in A, then let ρ(x0)=b\rho(x_0) = b. Then note ρ(x,0n)ρ(x,x0)+ρ(x0,0n)=N+b\rho(x, \vb{0}_n) \le \rho(x, x_0) + \rho(x_0, \vb{0}_n) = N + b for every xAx \in A, and so ABρ(0n,N+b)A \subseteq B_{\rho}(\vb{0}_n, N + b). Since AA is a closed subset of the compact box, AA is compact as well.

  4. Extreme Value Theorem. Let f:XYf: X \to Y be a continuous map from a topological space XX to a totally ordered set YY in its order topology. If XX is compact, there exist c,dXc, d \in X such that f(c)f(x)f(d)f(c) \le f(x) \le f(d) for every xXx \in X.

    Intuition.

    This theorem essentially states that f(X)f(X) has a minimum and maximum. We prove f(X)f(X) contains its maximum, since the proof that it contains its minimum is analogous. For the sake of contradiction, we suppose f(X)f(X) does not contain a maximum. Then note {(,a)}af(X)\{(-\infty, a)\}_{a \in f(X)} is a set of open subsets of YY. Since f(X)f(X) has no maximum, for each yf(X)y \in f(X), there is some a>yf(X)a > y \in f(X), so this collection forms an open cover for f(X)f(X). Since XX is compact, f(X)f(X) is also compact by continuity of ff. Thus there is a finite subcover

    {(,a1),,(,an)}\{(-\infty, a_1), \dots, (-\infty, a_n)\}

    that covers f(X)f(X). Note a=max{a1,,an}a = \max\{a_1, \dots, a_n\} exists and should be an element of AA, but it is not contained in the finite subcover, a contradiction.

  5. For a metric space (X,d)(X, d) and some nonempty subset AXA \subseteq X, we define the distance of a point xXx \in X from AA by d(x,A)=infaA{d(x,a)}d(x, A) = \inf_{a \in A}\{d(x, a)\}. Colloquially, this is the shortest distance from the point to the set AA. Compare this to the diameter of AA, which is colloquially the largest distance between any two points in AA.

  6. The function d(,A)d(-, A) is continuous.

    Intuition.

    Let x,yXx, y \in X be arbitrary points. Note then that

    d(x,A)d(x,a)d(x,y)+d(y,a)d(x, A) \le d(x, a) \le d(x, y) + d(y, a)

    for every aAa \in A. Notably, we seek to minimize d(y,a)d(y, a), such that

    d(x,A)d(x,y)d(y,a)d(x,y)infaA{d(y,a)}    d(x,A)d(y,A)d(x,y).d(x, A) \le d(x, y) - d(y, a) \le d(x, y) - \inf_{a \in A} \{d(y, a)\} \implies d(x, A) - d(y, A) \le d(x, y).

    Interchanging the role of x,yx, y yields d(x,A)d(y,A)d(x,y)|d(x, A) - d(y, A)| \le d(x, y), implying d(,A)d(-, A) is Lipschitz and thus continuous.

  7. Lebesgue Covering Lemma. Suppose (X,d)(X, d) is a compact metric space together with an open cover A\mathcal{A}. Then there exists some δ>0\delta > 0, called a Lebesgue number, such that every set whose diameter is less than δ\delta is contained in one member of the cover.

    Intuition.

    Since XX is compact, it admits a finite subcover {Ak}k=1n\{A_k\}_{k = 1}^n, where each AkAA_k \in \mathcal{A}. Let Ck=XAkC_k = X \setminus A_k for each kk, then define the function f:XRf: X \to \bR by

    f(x)=1nk=1nd(x,Ck).f(x) = \frac{1}{n} \sum_{k = 1}^n d(x, C_k).

    We demonstrate that f(x)>0f(x) > 0 for every xXx \in X. For a fixed xXx \in X, there is some kk such that xAkx \in A_k. Since AkA_k is open, there is some ε>0\epsilon > 0 such that xB(x,ε)Akx \in B(x, \epsilon) \subseteq A_k. Thus d(x,Ck)εd(x, C_k) \ge \epsilon and so f(x)ε/nf(x) \ge \epsilon/n. Since XX is compact, ff attains a minimum value δ>0\delta > 0 (where δ>0\delta > 0 since f(x)>0f(x) > 0) by EVT. We show this is our desired Lebesgue number.

    Consider any set BB whose diameter is δ\delta. Then there is some x0Bx_0 \in B such that BB(x0,δ)B \subseteq B(x_0, \delta). Then note

    δf(x0)d(x,Cm),\delta \le f(x_0) \le d(x, C_m),

    where mm is chosen to maximize the value of d(x,Ck)d(x, C_k) for each kk. Then B(x0,δ)AmB(x_0, \delta) \subseteq A_m, where Am=XCmA_m = X \setminus C_m, implying BAmAB \subseteq A_m \in \mathcal{A}, completing the proof.

  8. Uniform Continuity Theorem. Suppose f:XYf: X \to Y is a continuous map from a compact metric space (X,dX)(X, d_X) to a target metric space (Y,dY)(Y, d_Y). Then ff is uniformly continuous, meaning that for any ε>0\epsilon > 0, there is some δ>0\delta > 0 such that

    dX(x1,x2)<δ    dY(f(x1),f(x2))<εd_X(x_1, x_2) < \delta \implies d_Y(f(x_1), f(x_2)) < \epsilon

    for every x1,x2Xx_1, x_2 \in X.

    Intuition.

    Fix ε>0\epsilon > 0. Let A={BdY(y,ε/2)}yY\mathcal{A} = \{B_{d_Y}(y, \epsilon/2)\}_{y \in Y} be an open cover of f(X)f(X) from open subsets of YY. Then note f1(A)f^{-1}(\mathcal{A}) forms an open cover of XX. Since XX forms a compact metric space, there is a Lebesgue number δ\delta. Let x1,x2Xx_1, x_2 \in X be such that dX(x1,x2)<δd_X(x_1, x_2) < \delta. Then {x1,x2}\{x_1, x_2\} is a set whose diameter is less than δ\delta, meaning {x1,x2}A\{x_1, x_2\} \subseteq A for some AAA \in \mathcal{A}. This implies {x1,x2}f1(BdY(y,ε/2))\{x_1, x_2\} \subseteq f^{-1}(B_{d_Y}(y, \epsilon/2)) for some yYy \in Y, and so {f(x1),f(x2)}BdY(y,ε/2)\{f(x_1), f(x_2)\} \subseteq B_{d_Y}(y, \epsilon/2). Thus dY(f(x1),f(x2))<εd_Y(f(x_1), f(x_2)) < \epsilon, completing the proof.

  9. For any topological space XX, a point xXx \in X is said to be isolated if {x}\{x\} is open. Equivalently, xx is isolated if an open neighborhood of xx exists that comprises only xx, or if xx is not a limit point of XX.

  10. Suppose XX is a nonempty, compact, Hausdorff space. If XX has no isolated points, then XX is uncountable.

    Intuition.

    The intuition behind this proof is to construct a sequence of nested, open sets V1V2V_1 \supseteq V_2 \supseteq \cdots with the finite intersection property, such that xnVnx_n \in V_n but xnVn+1x_n \notin \bar{V}_{n + 1} for every nZ+n \in \bZ_+, where xnXx_n \in X. Then, taking the closure of each VkV_k (note that if VkVk+1V_k \subseteq V_{k + 1}, then VkVk+1\bar{V}_k \subseteq \bar{V}_{k + 1}), we yield a sequence of nested, closed sets with the finite intersection property. Thus xkZ+Vkx \in \bigcup_{k \in \bZ_+} \bar{V}_k, but xxnx \ne x_n for each nn, completing the proof that XX is uncountable.

    Let XX be a nonempty, compact, Hausdorff space with no isolated points. Let UXU \subseteq X be a nonempty, open set and let xXx \in X be arbitrary. We seek to show there is a nonempty, open subset VUV \subseteq U such that xVx \notin \bar{V}. Choose a point yUy \in U such that xyx \ne y—if xUx \in U, then UU necessarily contains a distinct point yy to prevent {x}\{x\} from being open (since XX has no isolated points); otherwise if xUx \notin U, then yy is guaranteed since UU is nonempty. By Hausdorffness of XX, there exists open neighborhoods W1,W2XW_1, W_2 \subseteq X of xx and yy repsectively. Then V=W2UV = W_2 \cap U is a nonempty, open subset of UU. Note xVx \notin \bar{V} since W1W_1 is an open neighborhood of XX that doesn't intersect W2W_2, so xx is not a limit point of VV.

    Consider the subset {xn}nZ+X\{x_n\}_{n \in \bZ_+} \subseteq X. Let V1=XV_1 = X, and let Vk+1VkV_{k + 1} \subseteq V_k be nonempty and open such that xVk+1x \notin V_{k + 1} (existence is proved by the last paragraph). Then the closure of each VkV_k forms a nested sequence of closed sets with the finite intersection property, so there is some xkZ+Vkx \in \cup_{k \in \bZ_+} \bar{V}_k. Note that, for each nZ+n \in \bZ_+, xnVnx_n \notin \bar{V}_n and so xxnx \ne x_n. Thus xx is an element of XX outside this subset, making XX uncountable.

  11. As a direct corollary, it follows that R\bR is uncountable.

3.5.1Exercises#

[3.5.15]Problem#

Suppose XX is a totally ordered set in its order topology. Show that if every closed interval of XX is compact, then XX has the least upper bound property.

Solution.

Let AXA \subseteq X be an arbitrary nonempty, bounded set. Let aAa \in A be arbitrary, and let bb be an arbitrary upper bound. Then consider the interval [a,b]X[a, b] \subseteq X. For the sake of contradiction, suppose AA has no supremum. Then define an open cover for XX by assigning a set to each point x[a,b]x \in [a, b] and forming a collection of sets from it:

  1. If xx is not an upper bound: choose some aAa' \in A such that x<ax < a', then choose the open set (,a)(-\infty, a').

  2. If xx is an upper bound: choose some upper bound bb' such that b<xb' < x, then choose the open set (b,)(b', \infty).

Note these steps are justified since AA has no supremum (i.e., x[a,b]x \in [a, b] cannot be the largest element of AA or the least upper bound). Then there is a finite subcover whose union contains [a,b][a, b]: that is,

[a,b][k=1m(,ak)][k=1n(bk,)][a, b] \subseteq \left[ \bigcup_{k = 1}^m (-\infty, a_k) \right] \cup \left[ \bigcup_{k = 1}^n (b_k, \infty) \right]

for some akAa_k \in A for each k{1,,m}k \in \{1, \dots, m\} and for some upper bound bkb_k of AA for each k{1,,n}k \in \{1, \dots, n\}. We have that bmin=min{bk}k=1nb_{\text{min}} = \min\{b_k\}_{k = 1}^n is the smallest upper bound of AA contained in [a,b][a, b], and by convexity of [a,b][a, b], we note bminb_{\text{min}} is the supremum for AA, a contradiction.

[3.5.16]Problem#

Let (X,d)(X, d) be a metric space, and let AXA \subseteq X be nonempty.

  1. Show d(x,A)=0d(x, A) = 0 if and only if xAx \in \bar{A}.

  2. Suppose AA is compact. Show that d(x,A)=d(x,a)d(x, A) = d(x, a) for some aAa \in A.

  3. Define the ε\epsilon-neighborhood of AA in XX to be the set

    U(A,ε)={xX:d(x,A)<ε}.U(A, \epsilon) = \{x \in X: d(x, A) < \epsilon\}.

    Show that U(A,ε)=aAB(a,ε)U(A, \epsilon) = \bigcup_{a \in A} B(a, \epsilon).

Solution.
  1. ()(\Longrightarrow): Suppose d(x,A)=0d(x, A) = 0, meaning infaA{d(x,a)}=0\inf_{a \in A}\{d(x, a)\} = 0. Let ε>0\epsilon > 0 be arbitrary, then consider the open neighborhood B(x,ε)B(x, \epsilon) of xx. Since d(x,A)=0d(x, A) = 0, there is some aAa \in A such that d(x,a)<εd(x, a) < \epsilon. Thus aB(x,ε)a \in B(x, \epsilon), meaning B(x,ε)AB(x, \epsilon) \cap A \ne \emptyset. Since every arbitrary basic open neighborhood of xx intersects AA, it follows that xAx \in \bar{A}.

    ()(\Longleftarrow): Suppose xAx \in \bar{A}. Fix some nZ+n \in \bZ_+, then let ε=1/n\epsilon = 1/n. Note then that B(x,ε)AB(x, \epsilon) \cap A \ne \emptyset by definition of A\bar{A}, so there is some anB(x,ε)Aa_n \in B(x, \epsilon) \cap A. Thus d(x,an)<εd(x, a_n) < \epsilon, and as we let nn \to \infty, we note ε0\epsilon \to 0. Thus d(x,A)=0d(x, A) = 0.

  2. Define f:ARf: A \to \bR by f(a)=d(x,a)f(a) = d(x, a), with xXx \in X fixed. Let a1,a2Aa_1, a_2 \in A be arbitrary, then note

    d(x,a2)d(x,a1)+d(a1,a2)    d(x,a2)d(x,a1)d(a1,a2).d(x, a_2) \le d(x, a_1) + d(a_1, a_2) \implies d(x, a_2) - d(x, a_1) \le d(a_1, a_2).

    Reversing the roles of a1,a2a_1, a_2, we arrive at d(x,a2)d(x,a1)d(a1,a2)|d(x, a_2) - d(x, a_1)| \le d(a_1, a_2), implying d(x,)d(x, -) is Lipschitz and thus continuous. Since ff is continuous and AA is compact, we have that f(A)={d(x,a):aA}f(A) = \{d(x, a): a \in A\} takes on a minimum value by the Extreme Value Theorem. Thus there is some aAa \in A such that f(a)=d(x,a)=infaA{d(x,a)}=d(x,A)f(a) = d(x, a) = \inf_{a \in A}\{d(x, a)\} = d(x, A).

  3. Suppose xU(A,ε)x \in U(A, \epsilon), and so d(x,A)<εd(x, A) < \epsilon. Then there is some aAa \in A such that d(x,a)<εd(x, a) < \epsilon, and so xBd(a,ε)aABd(a,ε)x \in B_d(a, \epsilon) \subseteq \bigcup_{a \in A} B_d(a, \epsilon). Conversely, suppose xaABd(a,ε)x \in \bigcup_{a \in A} B_d(a, \epsilon). Then, for some aAa \in A, we have that d(x,A)d(x,a)<εd(x, A) \le d(x, a) < \epsilon, so xU(A,ε)x \in U(A, \epsilon).

[3.5.17]Problem#

Suppose (X,d)(X, d) is a connected metric space. Show that if XX has more than one point, then XX is uncountable.

Proof.

For fixed xXx \in X, define f:XRf: X \to \bR by f(y)=d(x,y)f(y) = d(x, y). We have that ff is continuous, and since XX is connected, the set f(X)f(X) is connected as well. Note connected subsets of R\bR coincide with intervals or rays, which are uncountable if they have at least one element, and so f(X)f(X) is uncountable (since XX has at least one element). Note that Xf(X)|X| \ge |f(X)|, so XX is uncountable as well.