Proof of Proposition 1.2.15

Note that I used (i)-(viii) instead of (a)-(h) to label the statements. Everything else should be exactly the same as in the textbook.

Proposition (Basic properties of open and closed sets)

Let (X,d)(X, d) be a metric space.

  1. Let EE be a subset of XX. Then EE is open if and only if E=int(E)E = \int\p{E}. In other words, EE is open if and only if for every xEx \in E, there exists an r>0r > 0 such that B(x,r)EB\p{x, r} \subseteq E.
  2. Let EE be a subset of XX. Then EE is closed if and only if EE contains all its adherent points. In other words, EE is closed if and only if for every convergent sequence (xn)n=m\p{x_n}_{n=m}^\infty in EE, the limit limnxn\lim_{n\to\infty} x_n of that sequence also lies in EE.
  3. For any x0Xx_0 \in X and r>0r > 0, then the ball B(x0,r)B\p{x_0, r} is an open set. The set {xX|d(x,x0)r}\set{x \in X \st d\p{x, x_0} \leq r} is a closed set. (This set is sometimes called the closed ball of radius rr centered at x0x_0.)
  4. Any singleton set {x0}\set{x_0}, where x0Xx_0 \in X, is automatically closed.
  5. If EE is a subset of XX, then EE is open if and only if the complement XE{xX|xE}X \setminus E \coloneqq \set{x \notin X \st x \notin E} is closed.
  6. If E1,,EnE_1, \ldots, E_n is a finite collection of open sets in XX, then E1E2EnE_1 \cap E_2 \cap \cdots \cap E_n is also open. If F1,,FnF_1, \ldots, F_n is a finite collection of closed sets in XX, then F1F2FnF_1 \cup F_2 \cup \cdots \cup F_n is also closed.
  7. If {Eα}αI\set{E_\alpha}_{\alpha \in I} is a collection of open sets in XX (where the index set II could be finite, countable, or uncountable), then the union αIEα{xX|xEα for some αI}\bigcup_{\alpha \in I} E_\alpha \coloneqq \set{x \in X \st x \in E_\alpha \text{ for some } \alpha \in I} is also open. If {Fα}αI\set{F_\alpha}_{\alpha \in I} is a collection of closed sets in XX, then the intersection αIFα{xXxFα for all αI}\bigcap_{\alpha \in I} F_\alpha \coloneqq \set{x \in X \mid x \in F_\alpha \text{ for all } \alpha \in I} is also closed.
  8. If EE is any subset of XX, then int(E)\int\p{E} is the largest open set which is contained in EE; in other words, int(E)\int\p{E} is open, and given any other open set VEV \subseteq E, we have Vint(E)V \subseteq\int\p{E}. Similarly E\overline{E} is the smallest closed set which contains EE; in other words, E\overline{E} is closed, and given any other closed set KEK \supseteq E, KEK \supseteq \cl{E}.
Proof.
  1. "    \implies"

    Assume EE is open. By definition, this means that EE=\bd{E} \cap E = \emptyset.

    First, notice that Eint(E)E \supseteq \int\p{E} by definition of int(E)\int\p{E} (check this!). For the reverse inclusion, let xEx \in E. Recall that int(E),ext(E),E\int\p{E}, \ext\p{E}, \bd{E} partition XX; that is, the three sets are disjoint and their union is all of XX:

    X=int(E)ext(E)E.X = \int\p{E} \cup \ext\p{E} \cup \bd{E}.

    Since xEx \in E, we know that xext(E)x \notin \ext\p{E} (check this!). Because EE=\bd{E} \cap E = \emptyset and xEx \in E, this means that xEx \notin \bd{E}. Thus, xint(E)x \in \int\p{E}.

    This shows that Eint(E)E \subseteq \int\p{E}, so E=int(E)E = \int\p{E} by double-inclusion.

    "    \impliedby"

    Assume E=int(E)E = \int\p{E}. By definition, we need to check that EE=\bd{E} \cap E = \emptyset. We have

    (E)E=(E)int(E)=,\bd\p{E} \cap E = \bd\p{E} \cap \int\p{E} = \emptyset,

    where in the last step, we used the fact that int(E),ext(E),E\int\p{E}, \ext\p{E}, \bd{E} partition XX.

  2. "    \implies"

    Assume EE is closed. This means that EE\bd{E} \subseteq E. We know that the set of adherent points of EE satisfies E=EE\cl{E} = E \cup \bd{E} (this is Proposition 1.2.10 in the textbook), so because both EEE \subseteq E and EE\bd{E} \subseteq E, their union EE\cl{E} \subseteq E also.

    "    \impliedby"

    Assume that EE\cl{E} \subseteq E. By Proposition 1.2.10 again, we know EEEE \cup \bd{E} \subseteq E. In particular, EE\bd{E} \subseteq E, so EE is closed by definition,

  3. By part (i), we need to show that for all yB(x0,r)y \in B\p{x_0, r}, there exists s>0s > 0 such that B(y,s)B(x0,r)B(y, s) \subseteq B\p{x_0, r}. (It's a little confusing, but in (i), I replaced EE with B(x0,r)B\p{x_0, r}, x0x_0 with yy, and rr with ss.)

    Let s=rd(x0,y)s = r - d\p{x_0, y} (draw a picture to see where I got this from). Then if zB(y,s)z \in B\p{y, s},

    d(z,x)d(z,y)+d(y,x0)<s+d(x0,y)=r.d\p{z, x} \leq d\p{z, y} + d\p{y, x_0} < s + d\p{x_0, y} = r.

    In other words, d(z,x0)<rd\p{z, x_0} < r, which means zB(x0,r)z \in B\p{x_0, r}. Thus, B(y,s)B(x0,r)B\p{y, s} \subseteq B\p{x_0, r}, so B(x0,r)B\p{x_0, r} is open.

    For brevity, I will denote K(x0,r){xX|d(x,x0)r}K\p{x_0, r} \coloneqq \set{x \in X \st d\p{x, x_0} \leq r}. We'll use (ii): let (xn)n=m\p{x_n}_{n=m}^\infty be a sequence of elements in K(x0,r)K\p{x_0, r} which converges to xXx' \in X. We need to show that xK(x0,r)x' \in K\p{x_0, r}. Since xnK(x0,r)x_n \in K\p{x_0, r}, we have

    d(x,x0)d(x,xn)+d(xn,x0)d(x,xn)+r.d\p{x', x_0} \leq d\p{x', x_n} + d\p{x_n, x_0} \leq d\p{x', x_n} + r.

    Lastly, because xnnxx_n \xrightarrow{n\to\infty} x', we can send nn \to \infty to get d(x,x0)rd\p{x', x_0} \leq r.

    (I used the following fact from 131A: if annaa_n \xrightarrow{n\to\infty} a and anLa_n \leq L for all nn, then aLa \leq L.)

  4. We'll use (ii). Let (xn)n=m\p{x_n}_{n=m}^\infty be a sequence of elements in {x0}\set{x_0} which converges to xXx' \in X. Since xn{x0}x_n \in \set{x_0}, this means that xn=x0x_n = x_0 for all nn, i.e., (xn)n=m\p{x_n}_{n=m}^\infty is a constant sequence. Thus, xnnx0{x0}x_n \xrightarrow{n\to\infty} x_0 \in \set{x_0}, so {x0}\set{x_0} is closed.

  5. "    \implies"

    Assume EE is open. Suppose (xn)n=m\p{x_n}_{n=m}^\infty is a sequence of elements in XEX \setminus E which converges to x0Xx_0 \in X. We need to show that x0XEx_0 \in X \setminus E. If x0Ex_0 \in E, then because EE is open, there exists r>0r > 0 such that B(x0,r)EB\p{x_0, r} \subseteq E. But because xnnx0x_n \xrightarrow{n\to\infty} x_0, there exists nmn \geq m such that xnB(x0,r)x_n \in B\p{x_0, r}. This means xnE(XE)=x_n \in E \cap \p{X \setminus E} = \emptyset, which is a contradiction.

    "    \impliedby"

    Assume XEX \setminus E is closed. We will show that EE=\bd{E} \cap E = \emptyset. Let x0Ex_0 \in \bd{E}. By definition, this means that for all r>0r > 0, we have B(x0,r)(XE)B\p{x_0, r} \cap \p{X \setminus E} \neq \emptyset. But this is exactly what it means for x0XE=XEx_0 \in \cl{X \setminus E} = X \setminus E.

  6. Let x0E1Enx_0 \in E_1 \cap \cdots \cap E_n. For each 1in1 \leq i \leq n, x0Eix_0 \in E_i, so because EiE_i is open, there exists ri>0r_i > 0 such that B(x0,ri)EiB\p{x_0, r_i} \subseteq E_i. Let r=min{r1,,rn}>0r = \min\set{r_1, \ldots, r_n} > 0 (this is where we use the fact that we have a finite intersection). Then for each 1in1 \leq i \leq n,

    B(x0,r)B(x0,ri)Ei.B\p{x_0, r} \subseteq B\p{x_0, r_i} \subseteq E_i.

    Thus, B(x0,r)E1EnB\p{x_0, r} \subseteq E_1 \cap \cdots \cap E_n, so E1EnE_1 \cap \cdots \cap E_n is open.

    For the second statement, we can use (v) by taking complements. Note that

    X(F1Fn)=(XF1)(XFn).X \setminus \p{F_1 \cup \cdots \cup F_n} = \p{X \setminus F_1} \cap \cdots \cap \p{X \setminus F_n}.

    By (v), XF1,,XFnX \setminus F_1, \ldots, X \setminus F_n are all open sets, so the first statement (where we plug in Ei=XFiE_i = X \setminus F_i) implies that X(F1Fn)X \setminus \p{F_1 \cup \cdots \cup F_n} is open. By (v) again, F1FnF_1 \cup \cdots \cup F_n is closed.

  7. Let x0αIEαx_0 \in \bigcup_{\alpha \in I} E_\alpha. By definition, this means there exists αI\alpha \in I such that x0Eαx_0 \in E_\alpha. Since this EαE_\alpha is open, there exists r>0r > 0 such that

    B(x0,r)EααIEα,B\p{x_0, r} \subseteq E_\alpha \subseteq \bigcup_{\alpha \in I} E_\alpha,

    so αIEα\bigcup_{\alpha \in I} E_\alpha is open.

    As in the proof of (vi), we can prove the second statement by using (v) and taking complements.

    XαIFα=αI(XFα)X \setminus \bigcap_{\alpha \in I} F_\alpha = \bigcup_{\alpha \in I} \p{X \setminus F_\alpha}

    is a union of open sets, hence open by the first statement. By (v), this means αIFα\bigcap_{\alpha \in I} F_\alpha is closed.

  8. First, we show that int(E)\int\p{E} is open. Let x0int(E)x_0 \in \int\p{E} so that there exists r>0r > 0 such that B(x0,r)EB\p{x_0, r} \subseteq E. By (iii), B(x0,r)B\p{x_0, r} is an open set, so given any yB(x0,r)y \in B\p{x_0, r}, there exists s>0s > 0 so that

    B(y,s)B(x0,r)E,B\p{y, s} \subseteq B\p{x_0, r} \subseteq E,

    so yint(E)y \in \int\p{E}. Thus, B(x0,r)int(E)B\p{x_0, r} \subseteq \int\p{E}, so int(E)\int\p{E} is open.

    Now, assume VEV \subseteq E, where VV is an open set. Then for any x0Vx_0 \in V, there exists r>0r > 0 so that

    B(x0,r)VE,B\p{x_0, r} \subseteq V \subseteq E,

    so x0int(E)x_0 \in \int\p{E}. Thus, Vint(E)V \subseteq \int\p{E}.

    Next, we show that E\cl{E} is a closed set. Let (xn)n=m\p{\cl{x}_n}_{n=m}^\infty be a sequence of elements in E\cl{E} which converges to x0Ex_0 \in E. We need to show that x0Ex_0 \in \cl{E}, which means we need to find a sequence of elements in EE which converges to x0x_0. Since xnE\cl{x}_n \in \cl{E}, we know that for any r>0r > 0, B(x,r)EB\p{\cl{x}, r} \cap E \neq \emptyset. In particular, if r=1nr = \frac{1}{n}, then there exists xnEx_n \in E such that xnB(x,1n)x_n \in B\p{\cl{x}, \frac{1}{n}}. The sequence (xn)n=m\p{x_n}_{n=m}^\infty also converges to x0x_0:

    d(xn,x0)d(xn,xn)+d(xn,x0)1n+d(xn,x0).d\p{x_n, x_0} \leq d\p{x_n, \cl{x}_n} + d\p{\cl{x}_n, x_0} \leq \frac{1}{n} + d\p{\cl{x}_n, x_0}.

    Sending nn \to \infty (as we did in the proof of (iii)), the right-hand side tends to 00, so xnnx0x_n \xrightarrow{n\to\infty} x_0. This shows that x0Ex_0 \in \cl{E}, so E\cl{E} is closed.

    Lastly, if KK is a closed set such that KEK \supseteq E. Let x0Ex_0 \in \cl{E}. Then there exists a sequence (xn)n=m\p{x_n}_{n=m}^\infty of elements in EE such that xnnx0x_n \xrightarrow{n\to\infty} x_0. But because EKE \subseteq K, this means (xn)n=m\p{x_n}_{n=m}^\infty is also a sequence of elements in KK which converges to x0x_0, so x0K=Kx_0 \in \cl{K} = K. Thus, EK\cl{E} \subseteq K. \square

Remark.

It's important in (vi) (but not in (vii)) that the number of open/closed sets is finite. Here are two counter-examples from (R,)\p{\R, \abs{\,\cdot\,}}.

  1. Let En=(1n,1n)E_n = \p{-\frac{1}{n}, \frac{1}{n}}. Then

    n=1En={0},\bigcap_{n=1}^\infty E_n = \set{0},

    which is not an open set. Indeed, these sets are equal since 0En0 \in E_n for all n1n \geq 1, and if x0n=1Enx_0 \in \bigcap_{n=1}^\infty E_n, then x0x_0 satisfies

    1n<x0<1n    x0<1n.-\frac{1}{n} < x_0 < \frac{1}{n} \iff \abs{x_0} < \frac{1}{n}.

    If x00x_0 \neq 0, then if n0n_0 is large enough, we have x0<1n0<x0\abs{x_0} < \frac{1}{n_0} < \abs{x_0}, which is not possible. Thus, it must be the case that x0=0x_0 = 0, which proves that the two sets above are equal.

    (Notice how I didn't mention that {0}\set{0} is closed. In general, a set can be open, closed, both open and closed, or neither open nor closed. In other words, {0}\set{0} being closed doesn't tell you whether or not it's open.)

  2. Fn=[1+1n,11n]F_n = \br{-1 + \frac{1}{n}, 1 - \frac{1}{n}} gives another counter-example (exercise below).

Exercise 1.

Let Fn=[1+1n,11n]F_n = \br{-1 + \frac{1}{n}, 1 - \frac{1}{n}}. Compute n=1Fn\bigcup_{n=1}^\infty F_n and show that it is not closed (which is not the same as showing that it's open).