T_a^b(f)-
176 | \vep$. Meanwhile, for every $i=1,\dots,n$,
177 | \[
178 | \int_{x_{i-1}}^{x_i}|f\hp| \ge
179 | \left|\int_{x_{i-1}}^{x_i}f\hp \right| =
180 | |f(x_i) - f(x_{i-1})|,
181 | \]
182 | where the second equality is guaranteed by the absolute continuity. Hence,
183 | $\int_a^b|f\hp|>T_a^b(f)-\vep$ for every $\vep>0$. Thus, $T_a^b(f)=
184 | \int_a^b|f\hp|$.\par
185 | By Lemma 4, $2P_a^b(f) = T_a^b(f) + f(b)-f(a)$. Hence,
186 | \[
187 | P_a^b(f) = \frac{1}{2}\left(\int_a^b|f\hp| + f(b)-f(a)\right)
188 | =\frac{1}{2}\int_a^b (|f\hp|+f\hp) = \int_a^b[f\hp]^+.
189 | \]
190 | \end{proof}
191 |
192 | \paragraph{14.}
193 | \begin{proof}
194 | $\,$\par
195 | (a) Suppose that $f$ and $g$ are absolutely continuous. Then for every
196 | $\vep>0$, there exists some $\delta>0$ such that for all finite
197 | nonoverlapping $\langle(x_n,y_n)\rangle$ with $|x_n-y_n|<\vep$,
198 | \[
199 | \sum |f(x_n)+g(x_n)-f(y_n)-g(y_n)| \le
200 | \sum |f(x_n)-f(y_n)|+|g(x_n)-g(y_n)| \le 2\vep.
201 | \]
202 | Hence, $f+g$ is also absolutely continuous. Since $-g$ is absolutely
203 | continuous as long as $g$ is, so is $f-g$. \par
204 | (b) Suppose that $f$ and $g$ are absolutely continuous. Then they are
205 | bounded, by $M$ for example. Hence for every $\vep>0$, there exists some
206 | $\delta>0$ such that for all finite nonoverlapping $\langle(x_n,y_n)\rangle$
207 | with $|x_n-y_n|<\vep$,
208 | \begin{align*}
209 | &\sum |f(x_n)g(x_n)-f(y_n)g(y_n)| \\
210 | =&\sum |f(x_n)g(x_n)-f(x_n)g(y_n) + f(x_n)g(y_n)-f(y_n)g(y_n)| \\
211 | \le&\sum \{|f(x_n)||g(x_n)-g(y_n)| + |f(x_n)-f(y_n)||g(y_n)|\} \\
212 | \le& M\vep.
213 | \end{align*}
214 | Thus, $fg$ is also absolutely continuous.\par
215 | (c) Since $f$ is continuous on $[a,b]$, $f$ can achieve its minimum in $[a,
216 | b]$. Hence, $|f(x)|\ge m>0$ as $f$ is never zero. Therefore for every $\vep>
217 | 0$, there exists some $\delta>0$ such that for all finite nonoverlapping
218 | $\langle(x_n,y_n)\rangle$ with $|x_n-y_n|<\vep$,
219 | \begin{align*}
220 | \sum\left|\frac{1}{f(x_n)}-\frac{1}{f(y_n)}\right|
221 | = \sum\left|\frac{f(x_n)-f(y_n)}{f(x_n)f(y_n)}\right|
222 | \le \frac{1}{m^2}\sum|f(x_n)-f(y_n)|\le \frac{\vep}{m^2}.
223 | \end{align*}
224 | \end{proof}
225 |
226 | \paragraph{17.}
227 | Part (a) is wrong. It can be fixed if we further require $g$ to be monotone
228 | increasing.
229 | \begin{proof}
230 | $\,$\par
231 | (a) For every $\vep>0$, let $\delta_1$ be the number in the definition of
232 | $F$ corresponding to $\vep$ and $\delta_2$ the number in the definition of
233 | $g$ corresponding to $\delta_1$. Then for every finite nonoverlapping
234 | $\langle (x_n, y_n)\rangle$ with $|x_n-y_n|<\delta_2$, $\sum |g(x_n)-g(y_n)|
235 | < \delta_1$. Since $g$ is monotone increasing, $(g(x_n), g(y_n))$ are
236 | nonoverlapping. Therefore, $\sum|F(g(x_n))-F(g(y_n))|<\vep$. Hence, $F\circ
237 | g$ is absolutely continuous.
238 | \end{proof}
239 |
240 | \paragraph{18.}
241 | \begin{proof}
242 | Without loss of generality, we assume that $g$ is nondecreasing.
243 | Since $mE=0$, for every $\vep>0$, by Proposition 3.15, there exists an open
244 | set $O\supset E$ with $mO<\vep$. Meanwhile, there exists a sequence of
245 | disjoint open intervals $\langle I_n=(a_n,b_n)\rangle$ such that
246 | $\bigcup_{n=1}^\infty I_n=O$ and $l(I_n)<\delta$ where $\delta$ is the
247 | number in the definition of absolute continuity. Then $g[E]\subset
248 | \bigcup_{n=1}^\infty g[I_n\cap[0,1]]$. Since $g$ is continuous, the image of
249 | an interval is still an interval and since $g$ is also nondecreasing, $g[I_n
250 | \cap[0,1]]=(g(a_n\hp), g(b_n\hp))$, where $a_n\hp=\max\{a_n,0\}$ and $b_n\hp
251 | =\min\{b_n,1\}$. Finally,
252 | \[
253 | m(g[E]) \le \sum_{n=1}^\infty m(g[I_n]) = \sum_{n=1}^\infty|g(b_n\hp)-
254 | g(a_n\hp)| \le \vep,
255 | \]
256 | where the last inequality comes from the absolute continuity of $g$. Since
257 | the choice of $\vep$ is arbitrary, $m(g[E])=0$.
258 | \end{proof}
259 |
260 | \paragraph{20.}
261 | \begin{proof}
262 | $\,$\par
263 | (a) For every $\vep>0$, let $\delta=\vep/M$. Then for every $\langle x_n
264 | \rangle_{i=1}^n$ and $\langle y_n\rangle_{i=1}^n$ with $|x_n-y_n|\le\delta$,
265 | \[
266 | \sum_{i=1}^n|f(x_n)-f(y_n)|\le M\sum_{i=1}^n|x_n-y_n| \le \vep,
267 | \]
268 | as $f$ satisfies the Lipschitz condition.\par
269 | (b) Suppose that $f$ is absolute continuous and $|f\hp|$ is bounded by $M$.
270 | Then for every $x$ and $y$ in the interval,
271 | \[
272 | |f(x)-f(y)|=\left|\int_x^y f\hp(t)\rd t\right| \le M|x-y|.
273 | \]
274 | Hence, $f$ satisfies the Lipschitz condition. The converse part has been
275 | proved in (a).\par
276 | (c) It is wrong. A counterexample is $f(x)=\chi_{[0,1]}$, $x\in(-1,1)$
277 | \end{proof}
278 |
279 | \paragraph{21.}
280 | \begin{proof}
281 | $\,$\par
282 | (a) Suppose that $O=\bigcup_{n=1}^\infty(c_n,d_n)$ where $(c_n,d_n)$ are
283 | disjoint. Since $g$ is continuous and increasing, $g\inv(c_n,d_n)$ is still
284 | an open interval, denoting it by $(a_n,b_n)$, and $(a_n,b_n)$ are also
285 | disjoint. Meanwhile, $d_n-c_n=f(a_n)-f(b_n)=\int_{a_n}^{b_n}g\hp $. Hence,
286 | \[
287 | mO = m\left(\bigcup_{n=1}^\infty(c_n,d_n)\right) =
288 | \sum_{n=1}^\infty(d_n-c_n) =
289 | \sum_{n=1}^\infty \int_{a_n}^{b_n}g\hp =
290 | \int_{g\inv[O]}g\hp.
291 | \]\par
292 | (b) Without loss of generality, we assume that $d\notin E$. For every $\vep
293 | >0$, there exists an open set $O\supset E$ with $mO<\vep$. By Part (a),
294 | \[
295 | \int_{g\inv[O]\cap H}g\hp = \int_{g\inv[O]}g\hp = mO < \vep.
296 | \]
297 | Since the choice of $\vep$ is arbitrary, $\int_{g\inv[O]\cap H}g\hp = 0$.
298 | Since $g\hp > 0$ on $H$, $g\inv[O]\cap H$ has measure zero.\par
299 | (c) Since $E$ is measurable, so is $g\inv[E]$. Meanwhile, by Theorem 3,
300 | $g\hp$ is measurable, hence $H$ is also measurable. Therefore, $F$ is
301 | measurable.\par
302 | We may assume without loss of generality that $c,d\notin E$. By Proposition
303 | 3.15, there exists some $G\in G_\delta$ such that $E\subset G\subset(c,d)$
304 | and $m(G\setminus E)=0$. Since $g$ is increasing, $g\inv[G]\cap H = F\cup
305 | (g\inv[G\setminus E]\cap H)$ and by (b), $g[G\setminus E]\cap H$ is of
306 | measure zero. Therefore, $\int_F g\hp=\int_{g\inv[G]\cap H}g\hp$. Namely, it
307 | suffices to show the result for $G\in G_\delta$. \par
308 | Suppose that $G=\bigcap_{n=1}^\infty O_n$ where each $O_n\subset(c,d)$ is
309 | open and $mO_1<\infty$. Without loss of generality, we may assume that
310 | $\langle O_n\rangle$ is decreasing. Then $mG=\lim_{n\to\infty}mO_n$. By (a),
311 | \[
312 | mO_n = \int_{g\inv[O_n]}g\hp = \int_a^b\chi_{O_n}(g(x))g\hp(x)\rd x.
313 | \]
314 | As $\chi_{O_n}(g(x))g\hp(x)$ is bounded by $|g\hp|$,
315 | \[
316 | \lim_{n\to\infty}\int_a^b\chi_{O_n}(g(x))g\hp(x)\rd x =
317 | \int_a^b\chi_G(g(x))g\hp(x)\rd x.
318 | \]
319 | Hence, $mG=\int_{g\inv[G]\cap H}g\hp$, completing the proof.\par
320 | (d) By Problem 3.25, $f\circ g$ is measurable. And since $g\hp$ is
321 | measurable by Theorem 3, $(f\circ g)g\hp$ is also measurable. \par
322 | Let $\langle\varphi_n\rangle$ be an increasing sequence of nonnegative
323 | simple functions which converges to $f$, the existence of which is
324 | guaranteed by Problem 4.4. By the monotone convergence theorem, $\int_c^d f
325 | =\lim\int_c^d\varphi_n$.\par
326 | For each $n$, suppose that $\varphi_n(y)=\sum_{k=1}^ma_k^{(n)}(y)
327 | \chi_{E_k^{(n)}}(y)$. Then
328 | \[
329 | \int_c^d\varphi_n = \sum_{k=1}^ma_k^{(n)}mE_{k}^{(n)}=
330 | \sum_{k=1}^ma_k^{(n)}\int_a^b\chi_{E_k^{(n)}}(g(x))g\hp(x)\rd x=
331 | \int_a^b\varphi_n(g(x))g\hp(x)\rd x,
332 | \]
333 | where the second equality comes from (c). Since $g$ is increasing,
334 | $\langle\varphi_n(g(x))g\hp(x)\rangle$ is an increasing sequence. Hence,
335 | \[
336 | \int_a^b f(g(x))g\hp(x)\rd x =
337 | \lim_{n\to\infty}\int_a^b\varphi_n(g(x))g\hp(x)\rd x.
338 | \]
339 | Thus,
340 | \[
341 | \int_c^d f(y)\rd y =
342 | \lim_{n\to\infty}\int_c^d\varphi_n(y)\rd y =
343 | \lim_{n\to\infty}\int_a^b\varphi_n(g(x))g\hp(x)\rd x =
344 | \int_a^b f(g(x))g\hp(x)\rd x.
345 | \]
346 | \end{proof}
347 |
348 | % end
349 |
350 | \subsection{Convex Functions}
351 | \paragraph{23.}
352 | \begin{proof}
353 | $\,$\par
354 | (a) Suppose that $x_0\in(a,b)$ and $y(x)=m(x-x_0)+\varphi(x_0)$ is a
355 | supporting line. As $[a,b)$ is finite, $\varphi\ge\min\{\varphi(a),y(a),
356 | y(b)\}$. \par
357 | (b) If $\varphi$ is monotone, then the limits exists. If $\varphi$ is not
358 | monotone, then since $D^+\varphi$ is nondecreasing, there exists some $[c,d]
359 | \subset(a,b)$ such that $D^+\varphi\le 0$ on $(a,c)$ and $D^+\varphi\ge 0$
360 | on $(d,b)$. Namely, $\varphi$ is monotone on the $(a,c)$ and $(d,b)$.
361 | Therefore, the limits also exist.\par
362 | Consider a finite interval near the finite endpoint. By (a), the limit can
363 | not be $-\infty$ as $\varphi$ is bounded from below.\par
364 | (c) If $x$ and $y$ are in the interior of $I$, the inequality holds by
365 | definition. By the continuity of $\varphi$, the statement holds for all $x,y
366 | \in I$.
367 | \end{proof}
368 |
369 | \paragraph{24.}
370 | \begin{proof}
371 | Note that the existence of $\varphi^{\prime\prime}$ implies $\varphi$ is
372 | continuously differentiable. Suppose that $\varphi$ is convex on $(a,b)$.
373 | Then $D^+\varphi$ is nondecreasing by Proposition 17, hence $\varphi^{\prime
374 | \prime}(x)\ge 0$ for each $x\in(a,b)$. And the converse of the statement
375 | follows from Proposition 18 immediately.
376 | \end{proof}
377 |
378 | \paragraph{25.}
379 | \begin{proof}
380 | $\,$\par
381 | (a) $\varphi^{\prime\prime}(t)=b^2p(p-1)(a+bt)^{p-2}$ which $\ge 0$ on
382 | $[0,\infty)$ if $p\ge 1$ and $\le 0$ if $0S+T\}=0$. Thus, $S+T\ge \|f+g\|_\infty$ by the definition of
8 | $\esssup$.
9 | \end{proof}
10 |
11 | \paragraph{2.}
12 | \begin{proof}
13 | Put $S=\|f\|_\infty$. Since $S\ge |f|$ a.e.,
14 | \[
15 | \|f\|_p=
16 | \left\{\int_0^1|f|^p\right\}^{1/p} \le
17 | \left\{\int_0^1S^p\right\}^{1/p} = S.
18 | \]
19 | Therefore, $\uplim_{p\to\infty}\|f\|_p \le S$. For the converse part, let
20 | $\vep$ be any positive number. Then the measure $\delta$ of $E=\{t:\,|f(t)|>
21 | S-\vep\}$ is positive. Hence,
22 | \[
23 | \left\{\int_0^1|f|^p\right\}^{1/p} \ge
24 | \left\{\int_E|f|^p\right\}^{1/p} \ge
25 | \delta^{1/p}(S-\vep) \to
26 | S-\vep\quad\text{as $p\to\infty$.}
27 | \]
28 | Hence, $\lowlim_{p\to\infty}\ge S$, completing the proof.
29 | \end{proof}
30 |
31 | \paragraph{3.}
32 | \begin{proof}
33 | \[
34 | \|f+g\|_1 = \int|f+g| \le \int|f|+\int|g| = \|f\|_1+\|g\|_1.
35 | \]
36 | \end{proof}
37 |
38 | \paragraph{4.}
39 | \begin{proof}
40 | For every $M>\|g\|_\infty$, $|g|\le M$ a.e. Hence,
41 | \[
42 | \int|fg|\le M\int|f| = \|f\|_1 M.
43 | \]
44 | Since the choice of $M$ is arbitrary, $\int|fg|\le \|f\|_1\|g\|_\infty$.
45 | \end{proof}
46 | % end
47 |
48 | \subsection{The Minkowski and Hölder Inequalities}
49 | \paragraph{8}
50 | \begin{proof}
51 | $\,$\par
52 | (a) The logarithm function is concave, so
53 | \[
54 | \log(a^p/p+b^q/q) \ge \frac{1}{p}\log a^p + \frac{1}{q}\log b^q = \log ab.
55 | \]
56 | Taking $\exp$ on the both sides yields the inequality. The equality holds
57 | iff $a^p=b^q$. \par
58 | (b) The case where $p=\infty$ has been proved in Problem 4 and the case
59 | where $\|f\|_p=0$ or $\|g\|=0$ is straightforward. Hence, we assume that
60 | $1
0$, there exists some $N$ such that for all $n>N$, $\|f_n-f\|<
94 | \vep$. Hence, for every $n,m>N$, by Minkowski inequality,
95 | \[
96 | \|f_n-f_m\| \le \|f_n-f\|+\|f-f_m\| < 2\vep.
97 | \]
98 | Hence, $\langle f_n\rangle$ is a Cauchy sequence.
99 | \end{proof}
100 |
101 | \paragraph{10.}
102 | \begin{proof}
103 | Suppose $f_n\to f$. Then $M_n=\|f_n-f\|_\infty=\esssup|f_n-f|\to 0$. Let
104 | $E_n=\{x:\,|f_n(x)-f(x)|>M_m\}$, each of which is with measure zero. And
105 | therefore $E=\bigcup_{n=1}^\infty E_n$ is with measure zero. Note that
106 | $\tilde{E}=\{x:\,|f_n(x)-f(x)|0$, there exists some $N$ such that for every $n>N$ and $x
110 | \in\tilde{E}$, $|f_n(x)-f(x)|<\vep$. Since $mE=0$, this implies $\|f_n-f
111 | \|_\infty=\esssup|f_n(x)-f(x)|<\vep$. Hence, $f_n\to f$ in $L^\infty$.
112 | \end{proof}
113 |
114 | \paragraph{11.}
115 | \begin{proof}
116 | Let $\langle f_n\rangle\subset L^\infty$ be absolutely summable. Put $M_n=
117 | \|f_n\|_\infty$ and $A_n=\{t:\,|f_n(t)|>M_n\}$. By the definition of $\|
118 | \cdot\|_\infty$, $mA_n=0$. Hence, $A=\bigcup_{n=1}^\infty A_n$ is of measure
119 | zero. \par
120 | Note that $|f_n(x)|\le M_n$ for every $n$ and $x\in E\setminus A$. Thus, by
121 | the Weierstrass M-test, $\sum_{n=1}^\infty f_n$ converges uniformly. Hence,
122 | on $E\setminus A$, $\sup|\sum_{n=1}^\infty f_n - \sum_{n=1}^N f_n|\to 0$ as
123 | $N\to\infty$. Since $mA=0$, this implies the summability of $\langle f_n
124 | \rangle$.
125 | \end{proof}
126 |
127 | \paragraph{13.}
128 | \begin{proof}
129 | Suppose $\langle f_n\rangle\subset C$ be absolutely summable. Since for
130 | every $x$, $0\le|f_n(x)|\le\|f_n\|$, $\langle f_n\rangle$ is uniformly
131 | convergent on $[0,1]$. Put $s=\sum_{n=1}^\infty f_n$. Since each $f_n$ is
132 | continuous, so is $s$. Therefore, $s\in C$. \par
133 | For every $\vep>0$, there exists some $N$ such that for every $n>N$ and $x
134 | \in[0,1]$, $\left|s(x)-\sum_{k=1}^nf_k(x)\right|<\vep$. Hence, $\|s-
135 | \sum_{k=1}^nf_k\|<\vep$. Thus, $\langle f_n\rangle$ is summable and
136 | therefore $C$ is a Banach space.
137 | \end{proof}
138 |
139 | \paragraph{16.}
140 | \begin{proof}
141 | Since $\|f_n-f\| \ge |\|f_n\|-\|f\||$, $f_n\to f$ in $L^p$ implies $\|f_n\|
142 | \to \|f\|$. For the converse part, note that $2^p(|f_n|^p+|f|^p)-|f_n-f|^p
143 | \ge 0$ and for almost every $x$,
144 | \[
145 | 2^p(|f_n|^p+|f|^p)-|f_n-f|^p \to 2^{p+1}|f|^p.
146 | \]
147 | By Fatou's Lemma,
148 | \begin{align*}
149 | 2^{p+1}\|f\|^p = 2^{p+1}\int|f|^p
150 | &\le \lowlim\int\{2^p(|f_n|^p+|f|^p)-|f_n-f|^p\} \\
151 | &= 2^{p+1}\|f\|^p - \uplim\|f_n-f\|^p.
152 | \end{align*}
153 | Hence, $\uplim\|f_n-f\|^p \le 0$. Since clear that $\lowlim\|f_n-f\|^p\ge0$,
154 | $\lim\|f_n-f\|=0$, i.e., $f_n \to f$ in $L^p$.
155 | \end{proof}
156 |
157 | \paragraph{17.}
158 | I assume that $1/p+1/q=1$.
159 | \begin{proof}
160 | Since $g\in L^p$, $|g|^q$ is integrable on $E=[0,1]$ and therefore for every
161 | $\vep>0$, there exists some $\delta$ such that for every $A\subset E$ with
162 | $mA<\delta$, $\int_A|g|^q<\vep$. Meanwhile, since $f_n(x)\to f(x)$ for
163 | almost every $x$, by Egoroff's Theorem, there exists some $A\subset E$ with
164 | $mA<\delta$ such that $f_ng$ converges to $fg$ uniformly on $E\setminus A$.
165 | \par
166 | From the uniform convergence we conclude
167 | \begin{equation}
168 | \label{eq:6.17.1}
169 | \int_{E\setminus A} fg = \lim_{n\to\infty}\int_{E\setminus A}f_ng.
170 | \end{equation}
171 | Meanwhile, by Hölder inequality,
172 | \begin{align*}
173 | \left|\int_A (f-f_n)g\right| \le \int_A|(f-f_n)g|
174 | &\le\left\{\int_A|f_n-f|^p\right\}^{1/p}\left\{\int_A|g|^q\right\}^{1/q}
175 | \le M\vep^{1/q}.
176 | \end{align*}
177 | Hence, \eqref{eq:6.17.1} can be extended to $E$.\par
178 | For $p=1$, this is not true. $f_n=n\chi_{[0,1/n]}$ and $g=\chi_{[0,1]}$
179 | gives a counterexample.
180 | \end{proof}
181 |
182 | \paragraph{18.}
183 | \begin{proof}
184 | By Minkowski inequality,
185 | \[
186 | \|g_nf_n-gf\| = \|g_n(f_n-f)+(g_n-g)f\| \le \|g_n(f_n-f)\|+\|(g_n-g)f\|.
187 | \]
188 | Fix $\vep>0$. Since $f,g_n,g\in L^p$, $|g_n-g|^p|f|^p$ is integrable and
189 | therefore there exists some $\delta>0$ such that for all subsets with
190 | measure $<\delta$, the integral of over it $<\vep$. Meanwhile, since $g_n\to
191 | g$ a.e., by Egoroff's Theorem, there exists some $A\subset E=[0,1]$ with $mA
192 | <\delta$ such that $g_n\to g$ uniformly on $E\setminus A$ and therefore
193 | there exists some $N_1>0$ such that for all $n>N_1$, $|g_n(x)-g(x)|^p<\vep$
194 | for $x\in E\setminus A$. Thus, for every $n>N_1$,
195 | \begin{align*}
196 | \|(g_n-g)f\|
197 | &=\left\{\int_{E\setminus A}|g_n-g|^p|f|^p\right\}^{1/p} +
198 | \left\{\int_A|g_n-g|^p|f|^p\right\}^{1/p} \\
199 | & \le \sqrt[p]{\vep}\|f\| + \sqrt[p]{\vep} \le (\|f\|+1)\vep.
200 | \end{align*}
201 | Since $|g_n|\le M$, $\|g_n(f_n-f)\|\le M\|f_n-f\|$. And since $f_n\to f$ in
202 | $L^p$, there exists some $N_2>0$ such that for all $n>N_2$, $\|f_n-f\|<
203 | \vep$. Put $N=\max(N_1,N_2)$, then for every $n>N$,
204 | \[
205 | \|g_nf_n-gf\| \le (\|f\|+1+M)\vep.
206 | \]
207 | Hence, $g_nf_n\to gf$ in $L^p$.
208 | \end{proof}
209 | % end
210 |
211 | \subsection{Approximation in $L^p$}
212 | \paragraph{19.}
213 | \begin{proof}
214 | Since $\|T_\Delta f\|\le \|T_\Delta|f|\|$ and $\|f\|=\||f|\|$, we may assume
215 | without loss of generality that $f\ge 0$. For $p>1$, by Jensen's inequality,
216 | \begin{align*}
217 | \|T_\Delta f\|_p^p
218 | &= \sum_{k=1}^m\int_{\xi_{k-1}}^{\xi_k}
219 | \left(\frac{1}{\xi_k-\xi_{k-1}}\int_{\xi_{k=1}}^{\xi_k}f\right)^p \\
220 | &\le \sum_{k=1}^m\int_{\xi_{k-1}}^{\xi_k}
221 | \frac{1}{\xi_k-\xi_{k-1}}\int_{\xi_{k-1}}^{\xi_k}f^p \\
222 | &= \sum_{k=1}^m\int_{\xi_{k-1}}^{\xi_k}f^p \\
223 | &= \int_0^1f^p = \|f\|_p^p.
224 | \end{align*}
225 | \end{proof}
226 | % end
227 |
--------------------------------------------------------------------------------
/real_analysis_3rd/ch7_metric_spaces.tex:
--------------------------------------------------------------------------------
1 | \section{Metric Spaces}
2 | \setcounter{subsection}{6}
3 | \subsection{Compact Metric Space}
4 | \paragraph{27.}
5 | \begin{proof}
6 | If $\rho(F, K) > 0$, then clear that $F\cap K = \varnothing$. For the reverse
7 | direction, consider the function $h(x) = \rho(x, F)= \inf_{y\in F}\rho(x,y)$.
8 | Clear that for $x\in K$, $h(x) \le \rho(K, F)$.
9 |
10 | First, we show that $h$ is continuous. Let $x$ be fixed. For every
11 | $x\hp \in X$
12 | \[
13 | h(x) \le \rho(x, y) \le \rho(x, x\hp) + \rho(x\hp, y),
14 | \quad\forall y\in F.
15 | \]
16 | Take infimum on the right hand side and we get
17 | \[
18 | h(x) \le \rho(x, x\hp) + h(x\hp)
19 | \quad\Rightarrow\quad
20 | h(x) - h(x\hp) \le \rho(x, x\hp).
21 | \]
22 | Similarly, we have $h(x\hp) - h(x) \le \rho(x, x\hp)$. Thus, $h$ is
23 | continuous.
24 |
25 | Since $K$ is compact and $h$ is continuous, $h$ attains its infimum $c$
26 | at some point $x_0 \in K$. Assume, to obtain a contradiction, that $c = 0$.
27 | Then, for every $\vep > 0$, there is a $y \in F$ s.t. $\rho(x_0, y) < \vep$.
28 | Namely, $x_0$ is a cluster point of $F$. Since $F$ is closed, $x_0 \in F$,
29 | which contradicts with $F\cap K=\varnothing$. Thus, $c > 0$ and therefore
30 | $\rho(F, K) > 0$.
31 | \end{proof}
32 |
33 | \paragraph{28.}
34 | \begin{proof}
35 | $\,$\par
36 | (a) Let $\vep > 0$ be fixed. Since $f$ is uniformly continuous, there exists
37 | some $\delta > 0$ s.t. for every $x_1,x_2\in X$ with $\rho(x_1, x_2) <
38 | \delta$, $\rho(f(x_1), f(x_2)) < \vep$. Let $B\subset X$ be a ball with
39 | radius $\delta/2$. Then $f(B) \subset Y$ is contained by some ball of
40 | radius $\vep$. Hence, we can cover $Y$ with finitely many balls of radius
41 | $\vep$ as long as we can cover $X$ with finitely many balls of radius
42 | $\delta / 2$ and this can be done since $X$ is totally bounded.
43 |
44 | (b) Clear that $X = (0, 1)$ is totally bounded while $Y = (1, \infty)$ is
45 | not. The function $1/x$ a continuous function that maps $X$ onto $Y$. Hence,
46 | the result does not hold.
47 | \end{proof}
48 |
49 |
50 | \paragraph{29.}
51 | \begin{proof}
52 | $\,$\par
53 | (a) First, by the definition of open cover and open sets, $\varphi(x) > 0$.
54 | Meanwhile, since $X$ is compact, it is bounded and, therefore,
55 | $\varphi(x) < \infty$.
56 |
57 | (b) Let $x$ be fixed and $r$ be such that there exists some $O\in \mcal{U}$
58 | with $B_{x, r}\subset O$. Let $s = r - \rho(x, y)$. If $s \le 0$, then
59 | clear that $s \le \varphi(y)$. If $s > 0$, then we have $B_{y, s}
60 | \subset B_{x, r} \subset O$. Hence, $s \le \varphi(y)$. Thus,
61 | $r - \rho(x, y) \le \varphi(y)$. Take supremum on the left hand side and we
62 | get $\varphi(x) - \rho(x, y) \le \varphi(y)$.
63 |
64 | (c) It follows immediately from (b).
65 |
66 | (d) Let $(x_n)$ be a sequence s.t. $\varphi(x_n) \to \vep$. Since $X$ is
67 | sequentially compact, $(x_n)$ has a convergent subsequence. For the sake of
68 | convenience, we assume the subsequence is $(x_n)$ itself. Suppose $x_n
69 | \to x$. Since $\varphi$ is continuous, $\varphi(x) = \lim\varphi(x_n) =
70 | \vep$. Thus, by (a), $\vep > 0$.
71 |
72 | (e) Let $\delta$ be any positive number that is less than $\vep$. For every
73 | $x$, since $\delta < \vep = \inf\varphi$, $\delta < \varphi(x)$ and,
74 | therefore, $B_{x, \delta} \subset O$ for some $O \in \mcal{U}$.
75 | \end{proof}
76 |
77 | \subsection{Baire Category}
78 | \paragraph{31.(a)}
79 | \begin{proof}
80 | Suppose that $F$ is nowhere dense. Since $F$ is closed, $F^c$ is dense.
81 | Therefore, for every point $x\in F$, every neighborhood of $x$ contains a
82 | point of $F^c$. Thus, $F$ contains to open set. For the reverse direction,
83 | since $F$ contains no open set, every neighborhood of every $x\in X$ contains
84 | a point in $F^c$, which implies that $F^c$ is dense.
85 | \end{proof}
86 |
87 | \paragraph{34.}
88 | \begin{proof}
89 | For the first part, it suffices to show that $\partial E$ is nowhere dense
90 | for any set $E \subset X$. Since $\partial E$ is closed and contains no open
91 | set, by 31.(a), it is nowhere dense.
92 |
93 | For the second part, assume, to obtain a contradiction, that $F$ is not
94 | nowhere dense. Then by 31.(a), it contains an nonempty open set $O$. By
95 | 32.(a), $O$ is also meager, which contradicts the Baire category theorem.
96 | \end{proof}
97 |
98 | \paragraph{35.}
99 | \begin{proof}
100 | Let $E$ be a subset set of the complete metric space $X$. If $E$ is residual,
101 | then $E^c = \bigcup_n K_n$ where $K_n$ are nowhere dense sets. Hence,
102 | \[
103 | E = \bigcap K_n^c \supset \bigcap (\cl K_n)^c.
104 | \]
105 | Note that $(\cl K_n)^c$ is a $G_\delta$. Meanwhile, since each $(\cl K_n)^c$
106 | is a dense open set and $X$ is complete, By the theorem of Baire,
107 | $\bigcap (\cl K_n)^c$ is also dense. The proof of the reverse direction is
108 | similar.
109 | \end{proof}
110 |
111 | \paragraph{39.}
112 | \begin{proof}
113 | Let $E_m$ has the same meaning as in the proof of Theorem 32. Let $O :=
114 | \bigcup_m E_m^\circ$. By Prop. 31, $O$ is a dense residual open set. For
115 | every $x\in O$, $x \in E_m^\circ$ for some $m$. Since $E_m^\circ$ is open,
116 | there is a neighborhood $U$ of $x$ which is contained by $E_m^\circ$. Thus,
117 | $\mcal{F}$ is uniformly bounded by $m$ in $U$.
118 | \end{proof}
119 |
120 |
121 | \setcounter{subsection}{9}
122 | \subsection{The Ascoli-Arzel\'{a} Theorem}
123 | \paragraph{47.}
124 | \begin{proof}
125 | Let $x$ be an arbitrary fixed point in $X$ and $(x_n)\subset X$ a sequence
126 | that converges to $x$. Clear that $K = \{x\}\cup(x_n)$ is (sequently)
127 | compact. Hence, $f_n$ converges to $f$ uniformly on $K$. Thus, $f$ is also
128 | continuous on $K$ and therefore at $x$. Namely, $f$ is continuous.
129 | \end{proof}
130 |
131 | \paragraph{49.}
132 | \begin{proof}
133 | Let $x$ be an arbitrary point in $X$. For every $\vep > 0$, since $\mcal{F}$
134 | is equicontinuous, there is an open neighborhood $O$ of $x$ s.t.
135 | $\sigma(f(x), f(y)) < \vep$ for every $y\in O$ and $f\in\mcal{F}$. Now, for
136 | every $f^+\in \mcal{F}^+$, suppose $f_n \to f$. We have
137 | \[
138 | \sigma(f^+(x), f^+(y))
139 | = \sigma\left(\lim_n f_n(x), \lim_m f_m(y)\right)
140 | = \lim_n\lim_m \sigma(f_n(x), f_m(y)) \le \vep,
141 | \]
142 | where the second equality comes from the continuity of $\sigma$ and the last
143 | inequality comes from the equicontinuity of $\mcal{F}$.
144 | \end{proof}
145 |
146 | \paragraph{50.} I assume that the norm on $C[0, 1]$ is the sup norm.
147 | \begin{proof}
148 | Let $\mcal{F} := \{f\in C[0, 1]\mid \|f\|_\alpha \le 1 \}$. First, we show
149 | that $\mcal{F}$ is equicontinuous. By the definition of $\|\cdot\|_\alpha$,
150 | $f\in\mcal{F}$ only if it is bounded by $1$ and $|f(x) - f(y)| \le
151 | |x - y|^\alpha$ for all $x, y \in [0, 1]$. For every $x\in X$ and $\vep > 0$,
152 | consider the $\sqrt[\alpha]{\vep}$-ball $B$ centered at $x$. For every
153 | $y \in B$, $|f(x) - f(y)| \le |x - y|^\alpha < \vep$. Hence, $\mcal{F}$ is
154 | equicontinuous. Let $(f_n)$ be a sequence in $\mcal{F}$. Since $f_n$ is
155 | bounded by $1$, by Corollary 41, $(f_n)$ contains a convergent subsequence.
156 | Thus, $\mcal{F}$ is a (sequently) compact subset of $C[0, 1]$.
157 | \end{proof}
158 |
159 |
160 |
161 |
162 |
163 |
164 |
165 |
166 |
--------------------------------------------------------------------------------
/real_analysis_3rd/ch8_topological_spaces.tex:
--------------------------------------------------------------------------------
1 | \section{Topological Spaces}
2 | \subsection{Fundamental Notions}
3 | \paragraph{3.}
4 | \begin{proof}
5 | If $A$ is open, then clear that $x \in A \subset A$. For the converse, let
6 | $E = \bigcup_{x\in A}O_x$ where $x\in O_x \subset A$. Since $O_x\subset A$,
7 | $E \subset A$. Meanwhile, for every $x\in A$, $x \in O_x\subset E$. Hence,
8 | $A = E$. Thus, $A$ is open since $E$ is the union of open sets.
9 | \end{proof}
10 |
11 | \paragraph{7.}
12 | \begin{proof}
13 | $\,$\par
14 | (a) We argue by contradiction. Assume $x \in F^c$. Since $F$ is closed, $F^c$
15 | is open and therefore there is neighborhood $O$ of $x$ s.t. $O \cap F
16 | \ne \varnothing$, which contradicts the fact that $x$ is a cluster point.
17 | Thus, $x \in F$.
18 |
19 | (b) Let $y := f(x)$ and $y_n := f(x_n)$. Let $O$ be an arbitrary
20 | neighborhood of $y$. Since $f$ is continuous, there is a neighborhood $U$ of
21 | $x$ s.t. $f(U) \subset O$. Since $x = \lim x_n$, there is an integer $N$ s.t.
22 | $x_n \in U$ for every $n > N$. Hence, $y_n = f(x_n) \in f(U) \subset O$ for
23 | all $n > N$. Thus, $y_n \to y$.
24 |
25 | (c) The previous argument, \textit{mutatis mutandis}, yields the result.
26 | \end{proof}
27 |
28 | \paragraph{10.}
29 | \begin{proof}
30 | $\,$\par
31 | (a) Suppose that both $A_1$ and $A_2$ are open.
32 | Let $f_1 := f|_{A_1}$, $x\in A$ and $y = f(x)$. We may assume without
33 | loss of generality that $x \in A_1$. For every neighborhood $O$ of $y$, since
34 | $f_1$ is continuous, there is a neighborhood $A_1\cap U$ s.t. $f(A_1\cap U)
35 | \subset O$ where $U$ is an open set of $X$. Since $A_1\cap U$ is still open,
36 | $f$ is continuous at $x$. Thus, $f$ is continuous.
37 |
38 | (b) Let $f(x) = 0$ on $A_1 = (-1, 0)$ and $f(x) = 1$ on $A_2 = [0, 1]$.
39 | \end{proof}
40 |
41 | \subsection{Bases and Contability}
42 | \paragraph{11.}
43 | \begin{proof}
44 | $\,$\par
45 | (a) Suppose for every $B\in\mcal{B}$ containing $x$, there is a $y\in
46 | B\cap E$. Let $O$ be an arbitrary open set containing $x$. By the definition
47 | of bases, there is a base set $B$ s.t. $x \in B \subset \mcal{B}$. Thus,
48 | $O\cap E \supset B\cap E \ne \varnothing$. Namely, $x \in \cl E$. The
49 | reversed direction follows immediately from the definition of cluster points.
50 |
51 | (b) If there is sequence in $E$ that converges to $x$, clear that $x\in
52 | \cl E$. Now we show the reverse. Let $\mcal{B}_x = (B_k)$ be a countable base
53 | at $x$ and $S_n = \bigcap_{k=1}^n B_k$. Choose $x_n \in S_n$. For every
54 | open set $O$ containing $x$, there exists a $B_m$ s.t. $x \in B_m
55 | \subset O$. Hence, for every $n > m$, $x_n \in S_n\subset S_m
56 | \subset S_m \subset O$. Thus, $x_n \to x$.
57 |
58 | (c) It follows immediately from (b).
59 | \end{proof}
60 |
61 | \paragraph{13.}
62 | \begin{proof}
63 | By the construction of $\mcal{B}$ and Prop. 5, $\mcal{B}$ is a base for some
64 | topology on $X$. Let $\mcal{T}$ be a topology containing $\mcal{C}$. Since
65 | $X \in \mcal{T}$ and $\mcal{T}$ is closed under finite intersection,
66 | $\mcal{T} \supset \mcal{B}$. Thus, $\mcal{T}$ contains the topology generated
67 | by $\mcal{B}$. Since the choice of $\mcal{T}$ is arbitrary, we conclude that
68 | $\mcal{B}$ is a base for the weakest topology containing $\mcal{C}$.
69 | \end{proof}
70 |
71 |
72 | \paragraph{16.}
73 | \begin{proof}
74 | Let $\mcal{B}$ be a countable base for the topology on $X$ and
75 | $\mcal{U}$ an open cover of $X$. For each $B\in\mcal{B}$, if there is
76 | some $U \in \mcal{U}$ with $B \subset U$, then pick that $U$. This yields
77 | a at most countable subset $\mcal{V}$ of $\mcal{U}$. Now we show that
78 | $\mcal{V}$ covers $X$. For every $x\in X$, since $\mcal{U}$ covers $X$, there
79 | is a $U\in\mcal{U}$ s.t. $x\in U$. Meanwhile, since $\mcal{B}$ is a base,
80 | there is a $B \in \mcal{B}$ s.t. $x\in B \subset U$. Hence, by our
81 | construction, there is a $U\hp \in \mcal{U}$ containing $x$ is picked. Thus,
82 | $x$ is covered by $\mcal{V}$.
83 | \end{proof}
84 |
85 | \subsection{The Separation Axioms and Continuous Real-Valued Functions}
86 | \paragraph{18.}
87 | \begin{proof}
88 | $\,$\par
89 | (a) If $x \ne y$. then $d(x, y) > 0$. It suffices to choose the open balls
90 | of radius $d(x, y) / 2$ centered at $x$ and $y$ respectively.
91 |
92 | (b) Let $O_1 = \{x\mid \rho(x, F_1) < \rho(x, F_2) \}$ and $O_2 = \{
93 | x\mid \rho(x, F_1) > \rho(x, F_2)\}$. Clear that $O_1$ and $O_2$ are
94 | disjoint. Since $F_1$ and $F_2$ are closed and disjoint, $\rho(F_1, F_2)
95 | > 0$. Hence, for every $x\in F_1$, $\rho(x, F_1) = 0$ and $\rho(x, F_2)
96 | > 0$. Therefore, $F_1 \subset O_1$. Similarly, $F_2 \subset O_2$. Meanwhile,
97 | for every $x\in O_1$, the open ball centered at $x$ and of radius
98 | $(\rho(x, F_2) - \rho(x, F_1))/3$ is contained in $O_1$. Therefore, $O_1$ is
99 | open and similarly for $O_2$. Thus, $X$ is normal.
100 | \end{proof}
101 |
102 | \paragraph{20.}
103 | \begin{proof}
104 | If $f$ is continuous, then clear that $\{x\mid f(x) < a\} =
105 | f\inv((-\infty, a))$ and $\{x\mid f(x) > a\} = f\inv((a, \infty))$ are open.
106 | Now we show the reverse. For every open interval $(a, b)$, $f\inv((a, b))
107 | = f\inv((a, \infty))\cap f\inv((-\infty, a))$ is open. Meanwhile, every open
108 | set $O$ in $\mathbb{R}$ is a union $\bigcup I_\alpha$of open intervals. Thus,
109 | $f\inv(O) = \bigcup f\inv(I_\alpha)$ is open. Namely, $f$ is continuous.
110 | The second results follows immediately from $\{f \ge a\}$ is closed iff
111 | $\{f < a\}$ is open.
112 | \end{proof}
113 |
114 | \paragraph{23.}
115 | It seems that in (a), the Hausdorff condition is not necessary.
116 | \begin{proof}
117 | $\,$\par
118 | (a) First, suppose that $X$ is normal. Let $G = O^c$. $F$ and $G$
119 | are disjoint closed set and, therefore, there are disjoint open sets $U$ and
120 | $V$ s.t. $F\subset U$ and $G\subset V$. To show $\cl U\subset O$, note that
121 | $O = G^c \supset V^c \supset U$, since $U$ and $V$ are disjoint. Since $V^c$
122 | is closed, $V^c \supset \cl U$. Thus, $F\subset U$ and $\cl U \subset O$.
123 |
124 | For the reverse, let $F_1$ and $F_2$ be two disjoint closed sets. Then
125 | $F_1^c$ is an open set containing $F_2$ and therefore there exists an open
126 | $U_2$ s.t. $F_2 \subset U_2$ and $\cl U_2\subset F_1^c$. Note that $\cl U_2$
127 | and $F_1$ are again disjoint closed sets. Hence, similarly, we can find an
128 | open $U_1$ s.t. $F_1\subset U_1$ and $\cl U_1 \subset (\cl U_2)^c$. Thus,
129 | $X$ is normal.
130 |
131 | (b) We index the sequence by $n$ instead of $r$. Suppose that $N$ is the
132 | smallest integer s.t. $r = p2^{-N} < 1$. By (a), we may find a open set
133 | $U_{N}$ s.t. $F\subset U_N$ and $\cl U_N\subset O$. Now, $U_N$ is again an
134 | open set containing $F$ and therefore we can find $U_{N+1}$ s.t.
135 | $F\subset U_{N+1}$ and $\cl U_{N+1} \subset U_N$. Proceed iteratively and
136 | we get the required sequence.
137 |
138 | (c) Clear that $0 \le f \le 1$, $\equiv 0$ on $F$ and $f\equiv 1$ on $O^c$.
139 | Hence, it suffices to show the continuity. For every $x \in X$ and $\vep> 0$,
140 | choose $r_1, r_2 \in \mathbb{Q}$ such that
141 | \[
142 | f(x) - \vep < r_1 < f(x) < r_2 < f(x) + \vep.
143 | \]
144 | Let $U = U_{r_2} \setminus \cl U_{r_1}$. Clear that $U$ is open. Meanwhile,
145 | \[
146 | f(x) < r_2 \quad\Rightarrow\quad
147 | \inf\{r\mid x \in U_r\} < r_2 \quad\Rightarrow\quad
148 | \exists r < r_2 \text{ s.t. } x \in U_r.
149 | \]
150 | Hence, $x \in U_{r_2}$. If $x \in \cl U_{r_1}$, then $x \in U_r$ for all
151 | $r > r_1$ since $U_r \supset \cl U_{r_1}$ and, therefore, $f(x) \le r_1$.
152 | Contradiction. Thus, $x \notin U_{r_1}$. Hence, $U$ is an open neighborhood
153 | of $x$. For every $y \in U$, clear that $f(x) < r_2$. Also, $y \notin
154 | \cl U_{r_1}$ implies that $f(y) \ge r_1$. Hence, $f(U)\subset (f(x) - \vep,
155 | f(x) + \vep)$. Thus, $f$ is continuous.
156 |
157 | (d) If $X$ is normal, then clear that the function described in (c) satisfies
158 | the requirements. For the reverse, let $O_1 = \{x\mid f(x) < 1/2\}$ and
159 | $O_2 = \{x \mid f(x) > 1/2\}$. Clear that $O_1$ and $O_2$ are disjoint and
160 | since $f$ is continuous, they are open. Meanwhile, as $f\equiv 0$ on $A$
161 | and $f\equiv 1$ on $B$, $O_1$ contains $A$ and $O_2$ contains $B$. Thus,
162 | $X$ is normal.
163 | \end{proof}
164 |
165 | \paragraph{24.}
166 | The function in (f) should be $g = \varphi k / (1 - |\varphi k|)$.
167 | \begin{proof}
168 | $\,$\par
169 | (a) Obvious.
170 |
171 | (b) Clear that $B$ and $C$ are disjoint. Since $f$ is continuous, so is
172 | $h$ and therefore $B$ and $C$ are closed. By Urysohn's lemma, there exists
173 | continuous $0 \le h_{1, B}, h_{1, C} \le 1$ s.t. $h_{1, B} \equiv 1$ (resp.
174 | $h_{1, C} \equiv 1$) on $B$ (resp. $C$) and vanishes on $C$ (resp. $B$). Let
175 | $h_1 = (-h_{1, B} + h_{1, C}) / 3$. Then $h_1 \equiv -1/3$ on $B$ and
176 | $h_1 \equiv 1/3$ on $C$. Meanwhile, $h_1$ is continuous and $|h_1| \le 1/3$.
177 | Let $x \in A$. If $h(x) \le -1/3$, then $|h(x) - h_1(x)| = |h(x) + 1/3| <
178 | 2/3$. Similarly for $x$ with $h(x) \ge 1/3$. If $-1/3 < h(x) < 1/3$, then
179 | $|h(x) - h_1(x)| < |h(x)| + |h_1(x)| \le 2/3$. Thus, $|h - h_1| < 2/3$.
180 |
181 | (c) Suppose that we have constructed $h_n$. Let $s_n = \sum_{i=1}^n h_i$. Let
182 | \[
183 | B_n = \{x \mid h(x) - s(x) < -2^n/3^{n+1}\}
184 | \quad\text{and}\quad
185 | C_n = \{x \mid h(x) - s(x) > -2^n/3^{n+1}\}.
186 | \]
187 | The previous argument, \textit{mutatis mutandis}, yields a continuous
188 | function $h_{n+1}$ s.t. $|h_{n+1}| < 2^n/3^{n+1}$ and $|h - s_n - h_{n+1}|
189 | = |h - s_{n+1}| < 2^{n+1} / 3^{n+1}$ for all $x \in A$.
190 |
191 | (d) By the Weierstrass $M$-test, $h_n$ is uniformly summable. Hence,
192 | $k = \sum_{n=1}^\infty h_n$ is continuous as each $h_n$ is. Clear that $|k|
193 | \le 1$. Moreover, by the estimation in (c), $h = k$ on $A$.
194 |
195 | (e) By Urysohn's lemma, there is a continuous function $\varphi$ on $X$ s.t.
196 | $\varphi \equiv 1$ on $A$ and $\varphi \equiv 0$ on $\{x\mid k(x) = 1\}$.
197 |
198 | (f) Let $g = \varphi k / (1 - |\varphi k|)$. By (e), $g$ is well-defined on
199 | entire $X$. Also, $g$ is continuous and for $x \in A$
200 | \[
201 | g(x) = \frac{\varphi(x) k(x)}{1 - |\varphi(x) k (x)|}
202 | = \frac{1\times h(x)}{1 - |1\times h(x)|}
203 | = \frac{\frac{f}{1 + |f|}}{1 - \frac{|f|}{1 + |f|}}
204 | = f.
205 | \]
206 | \end{proof}
207 |
208 | \paragraph{26.}
209 | \begin{proof}
210 | Let $\mcal{J}$ be the topology generated by $\mcal{F}$. Since every
211 | $f \in \mcal{F}$ is continuous, $\mcal{J} \subset \mcal{T}$. Let $O \in
212 | \mcal{T}$. For every $x \in O$, there is a continuous function $f\in\mcal{F}$
213 | s.t. $f(x) = 1$ and vanishes on $O^c$. Namely,
214 | \[
215 | U_x := f\inv (\mathbb{R}\setminus\{0\}) \in \mcal{J}
216 | \]
217 | is an open set with $x \in U_x \subset O$. Clear that $O = \bigcup_x U_x$.
218 | Thus, $O \in \mcal{J}$. Namely, $\mcal{T} = \mcal{J}$.
219 | \end{proof}
220 |
221 |
222 | \subsection{Connectedness}
223 | \paragraph{32.}
224 | \begin{proof}
225 | Assume that $G$ is not connected and let $(O_1, O_2)$ be a separation of $G$.
226 | Since each $G_\alpha$ is connected, $G_\alpha$ is contained by exactly
227 | one of $O_1$ and $O_2$ since otherwise $(G_\alpha\cap O_1, G_\alpha\cap O_2)$
228 | would be a separation of $G_\alpha$. Thus, there are two of $\{G_\alpha\}$
229 | s.t. one is contained by $O_1$ and the other contained by $O_2$ and hence
230 | they have no point in common. Contradiction.
231 | \end{proof}
232 |
233 | \paragraph{33.}
234 | \begin{proof}
235 | Assume that $B$ is not connected and let $(O_1, O_2)$ be a separation of $B$.
236 | Since $A$ is connected, either $A\cap O_1$ or $A\cap O_2$ is empty. Assume
237 | without loss of generality that $A\cap O_2 = \varnothing$. Then, $O_2
238 | \subset \partial A$ and therefore has empty interior. Contradiction.
239 | \end{proof}
240 |
241 | \paragraph{35.}
242 | \begin{proof}
243 | $\,$\par
244 | (a) If $X$ is not connected, then we choose two points $x, y$ from each set
245 | of a separation. Clear that they can not be connected by some arc since the
246 | image of $[0, 1]$ under a continuous map is still connected.
247 |
248 | (b) Assume that $X$ is not connected and let $(O_1, O_2)$ be a separation.
249 | Since each of $X_1=\{(x, y) \mid x = 0, -1\le y \le 1\}$ and $X_2\{(x, y)
250 | \mid y = \sin x, 0 < x \le 1\}$ is connected, they are contained in, say,
251 | $O_1$ and $O_2$ respectively. Clear that any neighborhood of $(0, 0)$
252 | contains points in $X_2$ and therefore $O_1\cap O_2 \ne \varnothing$.
253 | Contradiction.
254 |
255 | Now we show that $X$ is not arcwise connected. (TODO)
256 |
257 | (c) Let $x \in G$ and $H$ be the points of $G$ that can be connected to $x$
258 | by a polygonal arc. For every $y \in H$, since $G$ is open, there is an open
259 | ball $B$ centered at $y$ with $B \subset G$. Clear that we can connect every
260 | $z \in B$ by the arc connecting $x$ and $y$, and the segment from $y$ to $z$.
261 | Thus, $H$ is open. Now, we show that $K = G\setminus H$ is open. Let $y$ be
262 | a point in $K$ and $B$ a small open ball centered at $y$ that is contained in
263 | $G$. Assume, to obtain a contradiction, that $B \cap H \ne \varnothing$. Then
264 | the previous argument, \textit{mutatis mutandis}, show that $y$ can be
265 | connected to $x$. Contradiction. Thus, $H$ is both open and closed in $G$.
266 | Since $G$ is connected, $H = G$. Thus, $G$ is arcwise connected.
267 | \end{proof}
268 |
269 |
270 |
271 |
272 |
273 |
274 |
275 |
276 |
277 |
278 |
279 |
280 |
281 |
282 |
283 |
284 |
--------------------------------------------------------------------------------
/real_analysis_3rd/ch9_compact_and_locally_compact_spaces.tex:
--------------------------------------------------------------------------------
1 | \section{Compact and Locally Compact Spaces}
2 | \subsection{Compact Spaces}
3 |
4 | \paragraph{2.}
5 | I further assume that $X$ is Hausdorff.
6 | \begin{proof}
7 | Assume, to obtain a contradiction, that $F_n = K_n\setminus O \ne\varnothing$
8 | for every $n$. Since $F_n \supset F_{n+1}$, $\{F_n\}$ is collection of closed
9 | subsets of compact set $K_1$ with finite intersection property. Hence,
10 | $\bigcap F_n$ is nonempty, contradicting with $\bigcap K_n \subset O$.
11 | \end{proof}
12 |
13 | \paragraph{3.}
14 | \begin{proof}
15 | Let $F$ be a closed set and $x \notin F$. Since $X$ is Hausdorff, for every
16 | $y \in F$, there are two disjoint open sets $U_y$ and $O_y$ s.t. $y \in U_y$
17 | and $x \in O_y$. Since $X$ is compact, so is $F$. Note that $\{U_y\}$ is an
18 | open cover for $F$. Hence, it has a finite subcover $\{U_{y_i}\}_{i=1}^n$.
19 | Let $U = \bigcup U_{y_i}$ and $O = \bigcap O_{y_i}$. Clear that they are
20 | disjoint open sets s.t. $F \subset U$ and $x \in O$.
21 | \end{proof}
22 |
23 | \paragraph{6.}
24 | \begin{proof}
25 | Let $\vep > 0$ be fixed. Since $\mcal{F}$ is equicontinuous, for every
26 | $x \in X$, there is a neighborhood $O_x$ s.t. for every $x\hp \in X$,
27 | $\sigma(f(x), f(x\hp)) < \vep$. Clear that $\{O_x\}$ is an open cover and
28 | since $X$ is compact, it has a finite subcover $\{O_{x_i}\}_{i=1}^m$.
29 |
30 | For $x_i$, since $f_n(x_i) \to f(x_i)$, there is an integer $N_i$ s.t. for
31 | every $n > N$, $\sigma(f_n(x_i), f(x_i)) < \vep$. Since $\sigma(f_n(x_i),
32 | f_n(x)) < \vep$ holds for all $n$, $\sigma(f(x_i), f(x)) \le \vep$.
33 | Hence, for every $x \in O_{x_i}$ and $n > N$,
34 | \[
35 | \sigma(f_n(x), f(x))
36 | \le \sigma(f_n(x), f_n(x_i)) + \sigma(f_n(x_i), f(x_i))
37 | + \sigma(f(x_i), f(x))
38 | < 3\vep.
39 | \]
40 | Let $N = \max N_{x_i}$ and we get the desired result.
41 | \end{proof}
42 |
43 | \subsection{Countable Compactness and the Bolzano-Weierstrass Property}
44 | \paragraph{9.}
45 | \begin{proof}
46 | $\,$\par
47 | (a) It follows immediately from the definition and Problem 8.20.
48 |
49 | (b) For every $\alpha \in \mathbb{R}$,
50 | \[
51 | f + g < \alpha
52 | \quad\text{iff}\quad
53 | f < \alpha - g
54 | \quad\text{iff}\quad
55 | \exists q\in \mathbb{Q} \text{ s.t. } f < q,\, q < \alpha - g.
56 | \]
57 | Hence,
58 | \[
59 | \{f + g < \alpha\}
60 | = \bigcup_{q \in \mathbb{Q}} \{f < q\} \cap \{g < \alpha - q\},
61 | \]
62 | which is open. Thus, $f + g$ is also upper semicontinuous.
63 |
64 | (c) Since $(f_n)$ is a decreasing sequence, we can write $f(x) = \inf_n
65 | f_n(x)$. Hence, for every $\alpha \in \mathbb{R}$, $f < \alpha$ iff there
66 | exists some $n$ s.t. $f_n(x) < \alpha$. Hence,
67 | \[
68 | \{f < \alpha\} = \bigcup_n \{f_n < \alpha\},
69 | \]
70 | which is open. Thus, $f$ is also upper semicontinuous.
71 |
72 | (d) Note that $(f_n - f)$ is a decreasing sequence of upper semicontinuous
73 | functions that converges to $0$. Hence, by Dini's theorem, the convergence is
74 | uniform.
75 |
76 | (e) Suppose that $x \in \{f < \alpha\}$. Let $\vep$ be a positive real
77 | number. Since $f_n \to f$ uniformly, there is an integer $n$ s.t.
78 | $|f(y) - f_n(y)| < \vep$ for all $y \in X$. Meanwhile, since $f_n$ is upper
79 | semicontinuous, there is a $\delta > 0$ s.t. for every $y$ in the
80 | $\delta$-ball $B$ centered at $x$, $f_n(y) < f_n(x) + \vep$. Hence, for every
81 | $y \in B$,
82 | \begin{align*}
83 | f(y) = f(y) - f_n(y) + f_n(y) - f_n(x) + f_n(x) - f(x) + f(x)
84 | \le 3\vep + f(x).
85 | \end{align*}
86 | Thus, for sufficiently small $\vep > 0$, we have $B \subset \{f < \alpha\}$.
87 | Namely, $\{f < \alpha\}$ is open whence $f$ is upper semicontinuous.
88 | \end{proof}
89 |
90 | \paragraph{10.}
91 | \begin{proof}
92 | $\,$\par
93 | (i. $\Rightarrow$ iii.) Let $f$ be a bounded continuous real-valued function
94 | and $M := \sup f < \infty$. Let $F_n = \{f \ge M - 1/n\}$. Since $f$ is
95 | continuous, $F_n$ is closed. Note that $(F_n)$ is a countable family of
96 | closed sets with finite intersection property. Hence, $\bigcap F_n =
97 | \{f \ge M\}$ is nonempty as $X$ is countably compact. Namely, the maximum
98 | can be attained.
99 |
100 | (iii. $\Rightarrow$ ii.) Let $f$ be a continuous function and assume, to
101 | obtain a contradiction, that $f$ is unbounded. Then the function
102 | $-1/(|f| + 1)$ is a continuous bounded function whose maximum can not be
103 | attained. Contradiction.
104 |
105 | (ii. $\Rightarrow$ i.) Assume, to obtain a contradiction, that $X$ does not
106 | have the Bolzano-Weierstrass property, that is, there is a sequence $(x_n)$
107 | in $X$ that has no cluster point. Then $F := \{x_n\}_{n=1}^\infty$ is closed.
108 | Define $f: F \to \mathbb{R}$ by $f(x_n) = n$. Note that $f$ is continuous on
109 | $F$ and by Tietze's extension theorem, it can be continuously extended to
110 | $X$. However, $f$ is unbounded, contradicting (ii.). Thus $X$ has the
111 | Bolzano-Weierstrass and, therefore, is countably compact.
112 | \end{proof}
113 |
114 | \subsection{Products of Compact Spaces}
115 | \paragraph{13.}
116 | \begin{proof}
117 | Let $E$ be a closed and bounded set in $\mathbb{R}^n$. Then it is contained
118 | in some closed cube $K = \prod_{i=1}^n[a_i, b_i]$. By Tychonoff's theorem,
119 | $K$ is compact. Thus, $K$, a closed subset of a compact set, is also compact.
120 | \end{proof}
121 |
122 | \paragraph{15.}
123 | \begin{proof}
124 | Let $X = \prod_{n=1}^\infty$ be the product of sequentially compact spaces
125 | $(X_n)$. Let $(x_n)$ be a sequence in $X$. Since $X_1$ is sequentially
126 | compact, we may choose a subsequence $x_n^1$ of $x_n$ s.t. the first
127 | coordinate of $(x_n^1)$ converges to some $x^1$. Similarly, from $(x_n^1)$ we
128 | may choose a subsequence $x_n^2$ whose second coordinate converges to some
129 | $x^2$. Proceed inductively and we get a sequence of sequence. Finally,
130 | consider the sequence $(x_n^n)$. Since each coordinate converges and we are
131 | dealing with the product topology, $x_n^n$ converges to $(x_1, x_2, \dots)$.
132 | \end{proof}
133 |
134 |
135 | \subsection{Locally Compact Spaces}
136 | \paragraph{18.}
137 | \begin{proof}
138 | For every $x \in K$, since $X$ is locally compact, there is an open set
139 | $O_x$ with $\cl O_x$ compact. Note that $\{O_x\}_{x\in K}$ is an open cover
140 | for the compact set $K$. Hence, it has a finite subcover
141 | $\{O_{x_i}\}_{i=1}^n$. Then, $O = \bigcup_{i=1}^n O_{x_i}$ is an open set
142 | containing $K$ whose closure is compact.
143 | \end{proof}
144 |
145 | \paragraph{19.}
146 | \begin{proof}
147 | $\,$\par
148 | (a) Since $X$ is a locally compact Hausdorff space and $K$ is compact, there
149 | exists an open set $V$ with compact closure s.t. $V\supset K$. Since
150 | $\cl V$ is compact, it is normal. Therefore, by Urysohn's lemma, there is
151 | a continuous function $f: \cl V \to [0, 1]$ s.t. $f\equiv 1$ on $K$ and
152 | $f\equiv 0$ on $\partial V$. Extend $f$ to $X$ by setting $f \equiv 0$
153 | outsides $\cl V$. Then $f$ is continuous and $f \equiv 1$ on $K$. Meanwhile,
154 | since $\supp f\subset \cl V$, it is also compact.
155 | \end{proof}
156 |
157 | \paragraph{24.}
158 | Assume that $X$ is also Hausdorff.
159 | \begin{proof}
160 | $\,$\par
161 | (a) Clear that if $F$ is closed, so is $F\cap K$ for each closed compact
162 | $K$. For the reverse, we show that $F^c$ is open.
163 | Let $x \notin F$. Since $X$ is locally compact, there is a neighborhood $U$
164 | of $x$ whose closure is compact. If $F\cap\cl U = \varnothing$, then we are
165 | done. If $F\cap \cl U \ne \varnothing$, then by the hypothesis, it is closed.
166 | Therefore, $U\setminus(F\cap\cl U)$ is again an open neighborhood of $x$.
167 | Since $X$ is a locally compact Hausdorff space, we can find an open
168 | neighborhood $V$ with $\cl V \subset U\setminus(F\cap\cl U)$. In both cases,
169 | $F^c$ is open.
170 |
171 | (b) Suppose that for each closed compact $K$, $F\cap K$ is closed. For every
172 | $x \in \cl F$, since $X$ is first-countable, there exists a sequence $(x_n)
173 | \subset F$ which converges to $x$. Then $E := \{x\}\cup\{x_n\}_{n=1}^\infty$
174 | is closed and compact. Thus, by the hypothesis, $F\cap E$ is also closed,
175 | which implies that $x \in F$. Hence, $F$ is closed.
176 | \end{proof}
177 |
178 | \paragraph{26.}
179 | Assume that $X$ is Hausdorff.
180 | \begin{proof}
181 | Let $x$ be an arbitrary point in $X$ and $V$ any neighborhood of $x$. Since
182 | $X$ is a locally compact Hausdorff space, we may choose a open neighborhood
183 | $U_1$ of $x$ whose closure is compact and contained by $V$. Since $O_1$ is
184 | dense, $U_1\cap O_1$ is a nonempty neighborhood of $x$. Then, choose a
185 | neighborhood $U_2$ of $x$ s.t. $\cl U$ is compact and $\cl U \subset
186 | U_1\cap O_1$. Proceed inductively and we get a sequence $(U_n)$ s.t.
187 | $\cl U_n$ is compact and $\cl U_{n+1} \subset O_n\cap U_n$.
188 |
189 | Since $(\cl U_n)$ is a nested sequence of compact sets, $\bigcap \cl U_n$
190 | is nonempty. Choose $x_* \in \bigcap\cl U_n$. For every $O_n$, $x_*
191 | \in \cl U_{n+1} \subset O_n$. Thus, $x_* \in V\cap\bigcap_{n=1}^\infty O_n$.
192 | Namely, $\bigcap_{n=1}^\infty O_n$ is dense.
193 | \end{proof}
194 |
195 | \paragraph{29.}
196 | \begin{proof}
197 | $\,$\par
198 | (a) Let $F$ be a closed subset of a locally compact space $X$. For every
199 | $x\in F\subset X$, there is a neighborhood $U$ of $x$ whose closure is
200 | compact in $X$. Then, $U\cap F$ is also compact. Thus, $F$ is locally
201 | compact.
202 |
203 | (b) Let $O$ be an open subset of a locally compact Hausdorff space $X$. For
204 | every $x\in O \subset X$, there is a neighborhood $U$ of $x$ whose closure is
205 | compact in $X$ and contained by $O$. Note that $\cl U$ is also compact in
206 | $O$. Thus, $O$ is locally compact.
207 | \end{proof}
208 |
209 |
210 |
211 |
212 |
213 |
214 |
215 |
216 |
217 |
218 |
219 |
220 |
221 |
222 |
223 |
224 |
--------------------------------------------------------------------------------
/real_analysis_3rd/def.tex:
--------------------------------------------------------------------------------
1 | \documentclass[12pt, a4paper]{article}
2 |
3 | \usepackage[margin=1in]{geometry}
4 | \usepackage{
5 | color,
6 | clrscode,
7 | amssymb,
8 | amsmath,
9 | listings,
10 | xcolor,
11 | supertabular,
12 | multirow,
13 | mathtools,
14 | mathrsfs,
15 | amsthm,
16 | systeme,
17 | amsfonts
18 | }
19 | \definecolor{bgGray}{RGB}{36, 36, 36}
20 | \usepackage[
21 | colorlinks,
22 | linkcolor=bgGray,
23 | anchorcolor=blue,
24 | citecolor=green
25 | ]{hyperref}
26 |
27 | \newcommand{\vep}{\varepsilon}
28 |
29 | \newenvironment{solution}
30 | {\begin{proof}[Solution]}
31 | {\end{proof}}
32 | \newcommand{\rd}{\mathrm{d}}
33 | \newcommand{\inv}{^{-1}}
34 | \newcommand{\hp}{^\prime}
35 | \newcommand{\mcal}{\mathcal}
36 | \newcommand{\ubar}[1]{\text{\b{$#1$}}}
37 |
38 |
39 | \DeclareMathOperator*\lowlim{\underline{lim}}
40 | \DeclareMathOperator*\uplim{\overline{lim}}
41 | \DeclareMathOperator*\esssup{ess\,sup}
42 | \DeclareMathOperator\sgn{sgn}
43 | \DeclareMathOperator{\cl}{cl}
44 | \DeclareMathOperator{\supp}{supp}
45 | \def\upint{\mathchoice
46 | {\mkern13mu\overline{\vphantom{\intop}\mkern7mu}\mkern-20mu}%
47 | {\mkern7mu\overline{\vphantom{\intop}\mkern7mu}\mkern-14mu}%
48 | {\mkern7mu\overline{\vphantom{\intop}\mkern7mu}\mkern-14mu}%
49 | {\mkern7mu\overline{\vphantom{\intop}\mkern7mu}\mkern-14mu}%
50 | \int}
51 | \def\lowint{\mkern3mu\underline{\vphantom{\intop}\mkern7mu}\mkern-10mu\int}
52 |
53 |
--------------------------------------------------------------------------------
/real_analysis_3rd/main.pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/Engineev/solutions/4e33274fe1ed9e46fd0e6671c57cb589704939bd/real_analysis_3rd/main.pdf
--------------------------------------------------------------------------------
/real_analysis_3rd/main.tex:
--------------------------------------------------------------------------------
1 | \input{def.tex}
2 |
3 | \title{Real Analysis}
4 | \author{Yunwei Ren}
5 | \date{}
6 |
7 | \begin{document}
8 | \maketitle
9 | \tableofcontents
10 |
11 | \newpage
12 | \setcounter{section}{2}
13 | \input{ch3_lebesgue_measure.tex}
14 | \newpage
15 | \input{ch4_the_lebesgue_integral.tex}
16 | \newpage
17 | \input{ch5_differentiation_and_integration.tex}
18 | \newpage
19 | \input{ch6_the_classical_banach_spaces.tex}
20 | \newpage
21 | \input{ch7_metric_spaces}
22 | \newpage
23 | \input{ch8_topological_spaces}
24 | \newpage
25 | \input{ch9_compact_and_locally_compact_spaces}
26 | \newpage
27 | \setcounter{section}{10}
28 | \input{ch11_measure_and_integration}
29 | \newpage
30 | \input{ch12_measure_and_outer_measure}
31 |
32 |
33 | \end{document}
34 |
--------------------------------------------------------------------------------