Ch1. Measure Theory (Ex.01 ~ Ex.18)
1. Prove the following properties of Cantor set \(C = \bigcap_{n=1}^{\infty} C_n\)
(ii) Cantor set \(C\) is (totally) disconnected
proof
① \(\exists N\) s.t. \(|x - y| > \frac{1}{3^N}\)
Let \(x, y \in C\) and \(x < y\). Then, there is an integer \(N \in \mathbb{N}\) such that \(|x - y| > \frac{1}{3^N}\).
If not, it means \(|x - y| \leq
\frac{1}{3^k}\) for all \(k \in
\mathbb{N}\).
Since \(\frac{1}{3^k} \to 0\), for
every \(\varepsilon > 0\) there is
\(K_{\varepsilon}\) such that \(k \geq K_{\varepsilon}\) implies \(\frac{1}{3^k} < \varepsilon\).
When \(\varepsilon \geq |x - y|\), then
\(|x - y| > \frac{1}{3^k}\) for
\(k \geq K_{\varepsilon}\).
This is a contradiction.
② \(\exists z \notin C\) such that \(x < z < y\)
Note that \(\frac{1}{3^N}\) is a
length of a closed interval in \(C_N\).
Then, \(x\) and \(y\) are in distinct
intervals.
So, there is \(z \in [0, 1]\) such that
\(x < z < y\), \(z \notin C_N\), and hence \(z \notin C\).
③ \(A, B\) are separated
Let \(A = C \cap (-\infty, z)\) and
\(B = C \cap (z, \infty)\).
\(\overline{A} = \overline {C \cap (-\infty, z]}
\subset (-\infty, z]\) and \((-\infty,
z] \cap B = \emptyset\) (\(\because b
> z\) for \(\forall b \in
B\)),
so \(\overline{A} \cap B =
\emptyset\).
Similarly, \(\overline{B}
= \overline{C \cap [z, \infty)} \subset [z, \infty)\) and \(A \cap [z, \infty) = \emptyset\) (\(\because a < z\) for \(\forall a \in A\)), so \(A \cap \overline{B} = \emptyset\).
So,\(A, B\) are separated.
④ \(A \cup B = C\), conclusion.
\(x \in A \cup B\): Since \(A \subset C\) and \(B \subset C\), \(x \in C\).
\(\nexists z \in C\): If \(y < z\), \(y \in A\).
If \(y > z\), \(y \in B\). So, \(y \in A \cup B\).
So, \(A \cup B = C\). Since \(C\) is a union of separated sets, \(C\) is disconnected.
(ii) \(C\) is perfect.
proof
① \(x \in I_{k,i}\) for some \(k \in \mathbb{N}\) and \(i = 1, \cdots, 2^k\).
Choose any \(x \in C\), then \(x \in C_k\) for all \(k \in \mathbb{N}\). Each \(C_k\) has \(2^k\) closed intervals of same length (\(\forall k\)): \(C_k = \bigcup_{i=1}^{2^k} I_{k,i}\) where \(m(I_{k,i}) = \frac{1}{3^k}\) and \(I_{k,i} \cap I_{k,j} = \emptyset\) (\(i \neq j\)). Then, \(x \in I_{k,i}\) for some \(i = 1, \cdots, 2^k\).
② \(x\) is a limit point of \(C\)
Let \(y_k\) be an end point of the interval \(I_{k,i}\) such that \(y_k \neq x\) and \(|y_k - x| \leq |y_k' - x|\) (\(y_k'\) is the other end point).
\(r > |y_k - x|\): the open ball \(B_r(x)\) contains \(y_k\).
\(0 < r \leq |y_k - x|\): \(B_r(x)\) contains a point.
Then, \((x, x + r) \subset I_{k,i} \cap B_r(x)\). So \(B_r(x)\) contains \(x_k \in (x, x + r) \subset I_{k,i}\) or \((x - r, x)\).
To sum up, \(B_r(x)\) contains a point of \(C\) not equal to \(x\). So, \(x\) is a limit point of \(C\). This holds for all \(x \in C\), so \(C\) is perfect.
2. The Cantor set \(C\) can also be described in terms of ternary expansions.
(b) The Cantor-Lebesgue function defined on \(C\) is as follows: \(F(x) = \sum b_k 2^{-k}, x = \sum a_k 3^{-k}, b_k = a_k / 2\). In this case, \(a_k\) has 0 or 2. Show that \(F\) is continuous on \(C\).
proof
① \(\delta = \frac{1}{3^N}\) with \(\frac{1}{2^N} < \varepsilon\)
For given \(\varepsilon > 0\), choose \(\delta = \frac{1}{3^N}\) with \(\frac{1}{2^N} < \varepsilon\). For \(x, y \in C\), if \(|x - y| < \delta\), this means \(x\) and \(y\) are in the same interval of \(C_N\). Let \(x = \sum a_{x,k} 3^{-k}\) and \(y = \sum a_{y,k} 3^{-k}\) with a unique binary form.
② \(b_{x,k} = b_{y,k}\) for \(k \leq N\)
Assume that there exists \(k \leq
N\) such that \(a_{x,k} \neq
a_{y,k}\), and let \(K\) be the
first integer with the property: \(a_{x,K} =
0\), \(a_{y,K} = 2\). Then,
\(\sum a_{x,k} 3^{-k} \leq \sum a_{y,k}
3^{-k}\) and \(\sum a_{y,k} 3^{-k} \geq
0\).
So,
\[
\begin{aligned}
|y - x| &= \sum (a_{y,k} - a_{x,k}) 3^{-k} = \sum_{k = K}^{\infty} 2
\cdot 3^{-k} = 2 \cdot 3^{-K} \sum_{k = 0}^{\infty} 3^{-k} = 2 \cdot
3^{-K} \cdot \frac{1}{1 - \frac{1}{3}} = \frac{3}{3^K}.
\end{aligned}
\]
Since \(K \leq N\), \(\delta = \frac{1}{3^N} \leq \frac{1}{3^K} \leq |y
- x|\), which is a contradiction.
So, \(a_{x,k} = a_{y,k}\) for all \(k \leq N\), and consequently \(b_{x,k} = b_{y,k}\) (\(k \leq N\)).
③ \(|x - y| < \delta \implies |F(x) - F(y)| < \varepsilon\)
Then,
\[
\begin{aligned}
|F(x) - F(y)| &= \left| \sum (b_{x,k} - b_{y,k}) 2^{-k} \right| =
\sum_{k = N + 1}^{\infty} |b_{x,k} - b_{y,k}| 2^{-k} \leq \sum_{k = N +
1}^{\infty} 2^{-k} = \frac{1}{2^N}.
\end{aligned}
\]
This holds for all \(\varepsilon >
0\), so \(F\) is continuous on
\(C\).
(c) Prove that \(F: C \to [0,1]\) is surjective (onto).
proof
It is identical to show that there is a unique binary
expansion for every \(y \in
(0,1)\) (\(F(1) = 1\), so \(1 = \sum b_k\), \(F(0) = 0\), so \(0 = \sum b_k\)). For the first step, let
\(b_1 = 1\) and \(y_1 = y_0 - \frac{1}{2}\) if \(\frac{1}{2} \leq y_0 < 1\), and \(b_1 = 0\) and \(y_1 = y_0\) if \(0 \leq y_0 < \frac{1}{2}\).
Next, let \(b_2 = 1\) and \(y_2 = y_1 - \frac{1}{2^2}\) if \(\frac{1}{2^2} \leq y_1 < \frac{1}{2}\),
and \(b_2 = 0\) and \(y_2 = y_1\) if \(0 \leq y_1 < \frac{1}{2^2}\). Repeat
this process indefinitely, then \(y = \sum b_k
2^{-k}\).
Using the obtained \(b_k\), \(x \in C\) can be expressed as \(\sum a_k 3^{-k}\) where \(a_k = 2 b_k\).
3. Consider \([0,1]\) and let \(\varphi\) be a fixed real number with \(0 < \varphi < 1\). In the first stage, remove a centrally situated open interval in \([0,1]\) of length \(\varphi\). In the 2nd stage, remove 2 central open intervals of relative length \(\varphi\). Let \(C_{\varphi}\) denote the set after applying the above procedure indefinitely.
(a) Show that \(C_{\varphi}\) is the union of open intervals with total length 1.
proof
1st step, the total length of the bisected interval is \(\varphi\).
At the 2nd step, the total length of bisected intervals is \(2 \cdot \varphi \left( \frac{1 - \varphi}{2}
\right)\).
At the 3rd step, the total length of bisected intervals is \(2^2 \cdot \varphi \left( \frac{1 - \varphi}{2}
\right)^2\).
Then, the total length of the bisected intervals at the \(n\)th step is \(2^{n - 1} \varphi \left( \frac{1 - \varphi}{2}
\right)^{n - 1}\).
So,
\[
\sum_{n=1}^{\infty} 2^{n-1} \varphi \left( \frac{1 - \varphi}{2}
\right)^{n-1} = \varphi \sum_{n=1}^{\infty} (1 - \varphi)^{n-1} =
\frac{\varphi}{1 - (1 - \varphi)} = 1.
\]
(b) Show that \(m(C_{\varphi}) = 0\).
proof
Let \(C_n\) be the remaining closed
intervals at \(n\)th step.
It is obvious that \(m(C_n) = (1 -
\varphi)^n\).
\(C_n \supset C_{\varphi}\) and \(m(C_1) < \infty\), so \(m(C_{\varphi}) = \lim_{n \to \infty}
m(C_n)\).
Since \(0 < (1 - \varphi) < 1\),
\(\lim_{n \to \infty} m(C_n) =
0\).
Thus, \(m(C_{\varphi}) = 0\).
4. Construct a “Cantor-like set” \(\hat{C}\) so that at the \(k\)th stage of the construction, one removes \(2^k\) centrally situated open intervals of length \(l_k\) with \(l_1 + 2l_2 + \cdots + 2^k l_k < 1\) for all \(k\).
(a) Show that \(m(\hat{C}) > 0\): \(1 - \sum_{k=1}^{\infty} 2^k l_k > 0\).
proof
Let \(C_k\) denote the remaining set
after \(k\)th iteration of removal
(\(C_0 = [0,1]\)), and let \(L_k\) denote the union of the
removed open intervals until \(k\)th stage: \(C_0 = C_k \cup L_k\) and \(C_k \cap L_k = \emptyset\).
Since \(m(L_k) = \sum_{j=1}^{k} 2^j
l_j\), \(m(C_k) = 1 - \sum_{j=1}^{k}
2^j l_j\).
Since \(C_{k} \supseteq C\), \(m(\hat{C}) = \lim_{k \to \infty} m(C_k) = 1 -
\sum_{j=1}^{\infty} 2^j l_j\).
The condition says \(\sum_{k=1}^{\infty} 2^k
l_k < 1\), so \(m(\hat{C}) >
0\).
(b) For \(x \in \hat{C}\), show that there exists a sequence \(\{x_n\}\) of points such that \(x_n \in I_n\) where \(I_n\) is an open interval bisected at the \(n\)th removal with \(\lim_{n \to \infty} |I_n| = 0\), and \(\lim_{n \to \infty} x_n = x\). (\(I_n \subset L_k\) for some \(k\))
proof
① Def of \(I_n\) & \(x_n\), \(|x_n - x| \leq m(I_n)\)
Let \(J_n\) be one of the closed intervals in \(C_n\) such that \(x \in C_n\), and let \(x_n\) be a point in \(I_n\). Then \(|x_n - x| \leq m(J_n)\) because both \(x\) and \(x_n\) are in \(J_n\).
② \(\lim_{n \to \infty} |I_n| = 0\)
An open interval of length \(l_1\)
is bisected from \(C_0\) of length 1,
so \(l_1 < 1\). Each closed interval
of \(C_1\) has length less than \(1/2\), so an open interval of length \(l_2\) is bisected from one piece of \(C_1\) with length less than \(1/2\).
Each closed interval of \(C_2\) has a
length \(< 1/4\), so an open
interval of length \(l_3 < 1/4\) is
bisected. Repeating this countably, \(0 <
l_k < \frac{1}{2^n}\). \(|I_n| =
l_n\), so \(\lim_{n \to \infty} |I_n| =
0\).
③ \(\lim_{n \to \infty} x_n = x\)
\(C_n\) has \(2^n\) closed intervals
including \(J_n\), so \(m(J_n) < \frac{1}{2^n}\).
This means \(\lim_{n \to \infty} m(J_n) =
0\). Since \(|x_n - x| <
m(J_n)\), \(\lim_{n \to \infty} |x_n -
x| = 0\).
Thus, \(\lim_{n \to \infty} x_n =
x\).
(c) Show that \(\hat{C}\) is perfect, and contains no open interval.
proof
① perfect
Let \(x_n\) be an end point of \(J_n\) such that \(x_n \neq x\). \(x_n\) is not removed in every iteration, so \(x_n \in \hat{C}\) for all \(n\). \(|x_n - x| \leq |J_n| \to 0\), so \(\lim_{n \to \infty} x_n = x\). Then, every neighborhood of \(x\) contains all but finitely many points of \(\{x_n\}\). So \(x\) is a limit point of \(\hat{C}\), and thus the closed \(\hat{C}\) is perfect.
② no open interval
If \(\exists O \subset \hat{C}\), \(\exists\) at least two points of \(\hat{C}\) in \(O\).
Assume there exists an open interval \(O \subset \hat{C}\). Since \(\hat{C}\) is perfect, \(O\) should contain at least two distinct points of \(\hat{C}\), say \(x < y\), such that \(x, y \in O\). (If \(O\) contains only one point \(x\), \(\exists r\) s.t. \(B_r(x) \cap \hat{C} = \emptyset\), contradiction.)contradiction
There exists \(N \in \mathbb{N}\) such that \(l_N < \frac{1}{2^N} < \frac{1}{4} |x - y|\). Then, \(x, y \in J_N\), \(x \in I_{x,N}\), and \(y \in I_{y,N}\). Let \(x_N\) be a right end point of \(J_{x,N}\), \(y_N\) be a left end point of \(J_{y,N}\), and \(z = \frac{1}{2}(x_N + y_N)\). \(x < z < y\) implies \(z \in O\). Also, \(z\) is in a bisected open interval, so \(z \notin \hat{C}\) and thus \(z \notin O\). This is a contradiction.
(d) Show that \(\hat{C}\) is uncountable.
proof
Let us denote the subinterval of \(C_k\) with binary numbers. For example, \(I_{00}, I_{01}, I_{10}, I_{11}\) are the closed subintervals of \(C_2\), \(I_{000}, I_{001}, \cdots, I_{110}, I_{111}\) are the subintervals of \(C_3\), and so on.
\(a = 0.a_1 a_2 a_3 \cdots =
\sum_{k=1}^{\infty} a_k 2^{-k}\) with \(a_k = 0\) or \(1\), then \(\{a\}\) is uncountable set.
Let \(I_n^a\) denote an interval of
\(C_n\) whose subscript matches with
the first \(n\) terms of \(a\) (e.g., \(a =
0.101 \cdots\), then \(I_3^a = I_{101}
\subset C_3\)).
Consider \(\bigcap_{n=1}^{\infty} I_n^a\). Since \(I_n^a \subset I_{n-1}^a\) for all \(n\), \(\bigcap_{n=1}^{\infty} I_n^a\) is not empty. Since \(|I_n^a| < \frac{1}{2^n}\) and \(\lim_{n \to \infty} |I_n^a| = 0\), \(\bigcap_{n=1}^{\infty} I_n^a\) contains only one point, say \(x_a \in \hat{C}\). Thus shows a 1-1 correspondence between \(\hat{C}\) and \(A\), so \(\hat{C}\) is uncountable.
5. \(E\) is a compact set and \(O_n = \{ x : d(x, E) < 1/n \}\) is an open set.
(a) Show that \(m(E) = \lim_{n \to \infty} m(O_n)\).
proof
① \(x \in \overline{A}\) iff \(d(x, A) = 0\)
\(\Rightarrow\)
It is trivial when \(x \in A\), so the following proof covers the case \(x \in A'\). \(x\) is a limit point of \(A\), so for given \(\varepsilon > 0\) there is a point \(a_{\varepsilon} \in A\) such that \(a_{\varepsilon} \neq x\) and \(d(x, a_{\varepsilon}) < \varepsilon\). By the def of \(d(x, A) = \inf \{ d(x, a) : a \in A \}\), \(d(x, A) \leq d(x, a_{\varepsilon}) < \varepsilon\), so \(d(x, A) < \varepsilon\). This holds for all \(\varepsilon > 0\), so \(d(x, A) = 0\).\(\Leftarrow\)
\(d(x, A) = 0\) so \(\inf \{ d(x, a) : a \in A \} = 0\). Then, for given \(\varepsilon > 0\), there is \(a_{\varepsilon} \in A\) such that \(d(x, a_{\varepsilon}) < 0 + \varepsilon = \varepsilon\).
- \(x \notin A\): \(B_{\varepsilon}(x)\) contains a point of
\(A\) for all \(\varepsilon > 0\), so \(x \in A'\).
- \(x \in A\): nothing to
prove.
So, \(x \in \overline{A}\).
② \(\overline{E} = \bigcap O_n\)
- \(x \in E\): \(d(x, E) = 0\) so \(x \in O_n\) for all \(n\). So, \(x \in
\bigcap O_n\).
- \(x \in O_n\) for all \(n\), so \(d(x, E) < \frac{1}{n}\) \(\forall n\). This means \(d(x, E) = 0\), so \(x \in \overline{E}\).
③ \(m(E) = \lim_{n \to \infty} m(O_n)\)
\(E = \overline{E}\), \(O_{n+1} \subset O_n\).
\(E\) is closed, so \(\overline{E} = E\). It is obvious that \(O_n \supset O_{n+1}\) for all \(n\).\(m(O_1) < \infty\), compact.
\(E\) is bounded, so there is an open ball \(B_R(o) \supset E\). For \(x \in O_1\) and \(e \in E\), \(d(x, o) \leq d(x, e) + d(e, o) < 1 + d(e, o)\) (\(\because e \in E'\)) \(\leq 1 + R\) (\(\because e \in B_R(o)\)). So \(|x| < R + 1\) and \(x \in B_{R+1}(o)\). This holds for all \(x \in O_1\), so \(m(O_1) < \infty\).\(\therefore m(E) = \lim_{n \to \infty} m(O_n)\) (Cor 1.3.3)
(b) However, (a) may be false for closed and unbounded \(E\), or open and bounded \(E\).
proof
① \(E\) is unbounded & closed
Let \(E = \mathbb{Z}\). For \(k \in \mathbb{Z}\), \(O_n = \bigcup_{k \in \mathbb{Z}} B_{1/k}(k)\), where \(m(B_{1/k}(k)) = \frac{2}{k}\). So \(m(O_1) = \sum_{k \in \mathbb{Z}} \frac{2}{k} = \infty\), and \(\lim_{n \to \infty} m(O_n) = \infty\). However, \(m(E) = 0\). (Note that \(E\) is closed)
② \(E\) is open & bounded
\(E = \bigcup_{k=1}^{\infty} \left( q_k - \frac{1}{k}, q_k + \frac{1}{k} \right)\), \(m(E)\)
Let \(\{q_k\}\) be an enumeration of \(\mathbb{Q}\) in \([0,1]\) and \(E = \bigcup_{k=1}^{\infty} \left( q_k - \frac{1}{k}, q_k + \frac{1}{k} \right)\). Each \(\left( q_k - \frac{1}{k}, q_k + \frac{1}{k} \right)\) is open, so \(E\) (countable union of open sets) is open. Since \(q_k \in [0,1]\) for all \(k \in \mathbb{N}\), \(E \subset \left( -1, 1 + \frac{1}{1} \right)\).Then,
\[ m(E) \leq \sum_{k=1}^{\infty} m \left( \left( q_k - \frac{1}{k}, q_k + \frac{1}{k} \right) \right) = \sum_{k=1}^{\infty} \frac{2}{k} = 2 \sum_{k=1}^{\infty} \frac{1}{k} = \frac{1}{2}. \]\(\lim_{n \to \infty} m(O_n) \geq 1\)
If \(x \in \mathbb{Q} \cap [0,1]\), it is obvious that \(x \in E\) and \(d(x, E) = 0\). Consider the case \(x \in (\mathbb{R} - \mathbb{Q}) \cap [0,1]\), which is irrational in \((0,1)\). Since \(E\) contains all rationals in \([0,1]\) and \(\mathbb{Q}\) is dense in \(\mathbb{R}\), \(x \in E\) and \(d(x, E) = 0\) (\(\because \mathbb{Q}\) dense). Since \(d(x, E) = 0\) for all \(x \in [0,1]\), \(O_n\) contains \([0,1]\) for all \(n \in \mathbb{N}\). This means \(m(O_n) \geq 1\) for all \(n \in \mathbb{N}\), so \(\lim_{n \to \infty} m(O_n) \geq 1\).\(\therefore m(E) \neq \lim_{n \to \infty} m(O_n)\)
6. Let \(B\) be a ball in \(\mathbb{R}^d\) of radius \(r > 0\). Then, \(m(B) = v_d r^d\) where \(v_d = m(B_1)\) and \(B_1 = \{ x \in \mathbb{R}^d : |x| < 1 \}\).
proof
\(B_r = \{ x \in \mathbb{R}^d : |x| < r \} = r B_1 = \{ r y : y \in \mathbb{R}^d, |y| < 1 \}\). Then \(m(B_r) = r^d m(B_1)\).
7. \(\delta = (\delta_1, \delta_2, \dots, \delta_d)\) is a \(d\)-tuple of positive numbers \(\delta_i > 0\). For \(E \subset \mathbb{R}^d\), \(\delta E = \{ (\delta_1 x_1, \dots, \delta_d x_d) : (x_1, \dots, x_d) \in E \}\). If \(E\) is measurable, show that \(\delta E\) is measurable and \(m(\delta E) = \delta_1 \cdots \delta_d m(E)\).
proof
① \(\delta Q\) is a rectangle, and \(|\delta Q| = \delta_1 \cdots \delta_d |Q|\)
\(Q\) is a cube with each side of length \(l\): \(Q = [a_1, a_1 + l) \times \cdots \times [a_d, a_d + l)\). Let \(y = (\delta_1 y_1, \dots, \delta_d y_d) \in \delta Q\), then \(a_i \leq y_i/\delta_i \leq a_i + l\) and thus \(\delta_i a_i \leq y_i \leq \delta_i a_i + \delta_i l\). So, \(\delta Q\) is a rectangle: \(\delta Q = [\delta_1 a_1, \delta_1 a_1 + \delta_1 l) \times \cdots \times [\delta_d a_d, \delta_d a_d + \delta_d l)\) and \(|\delta Q| = \delta_1 \cdots \delta_d l^d = \delta_1 \cdots \delta_d |Q|\).
② \(\delta \bigcup A_i = \bigcup \delta A_i\), \(\delta \bigcap A_i = \bigcap \delta A_i\), \(A \subset B \Leftrightarrow \delta A \subset \delta B\), \((A^c)^{\delta} = (\delta A)^c\)
\(y \in \delta \bigcup A_i \Leftrightarrow
\exists i\) s.t. \(y/\delta \in A_i
\Leftrightarrow \exists i\) s.t. \(y
\in \delta A_i \Leftrightarrow y \in \bigcup \delta A_i\).
\(z \in \delta \bigcap A_i \Leftrightarrow
z/\delta \in \bigcap A_i \Leftrightarrow z \in \delta A_i\) for
all \(i \Leftrightarrow z \in \bigcap \delta
A_i\).
If \(A \subset B\), \(x \in A \Rightarrow \delta x \in \delta
A\). Since \(\delta x \in \delta
B\), \(\delta A \subset \delta
B\).
\(y \in (\delta A)^c \Leftrightarrow y \notin
\delta A \Leftrightarrow y/\delta \notin A \Leftrightarrow y \in (\delta
A^c)\).
③ \(m_k(\delta E) = \delta_1 \cdots \delta_d m_k(E)\) (\(E \subset \mathbb{Q}^d\))
There exists a collection of cubes \(\{ Q_i \}\) such that \(\sum_{i} |Q_i| < m_k(E) + \varepsilon\). Then, \(\sum_{i} |\delta Q_i| = \delta_1 \cdots \delta_d \sum_{i} |Q_i| < \delta_1 \cdots \delta_d (m_k(E) + \varepsilon)\). Since \(\delta E \subset \bigcup \delta Q_i = \delta \bigcup Q_i\), \[m_k(\delta E) \leq \sum_{i} |\delta Q_i| = \delta_1 \cdots \delta_d \sum_{i} |Q_i| < \delta_1 \cdots \delta_d m_k(E) + \varepsilon\]. This holds for all \(\varepsilon > 0\), so \(m_k(\delta E) \leq \delta_1 \cdots \delta_d m_k(E)\).
There exists a collection of cubes \(\{ P_i \}\) such that \(\sum_{i} |P_i| < m_k(\delta E) + \delta_1 \cdots \delta_d \varepsilon\). Let \(\delta^{-1} = (\delta_1^{-1}, \dots, \delta_d^{-1})\). Then, \(\sum_{i} | \delta^{-1} P_i | = \delta_1^{-1} \cdots \delta_d^{-1} \sum_{i} |P_i| < m_k(E) + \varepsilon\). Using \(E = \delta^{-1} (\delta E) \subset \bigcup \delta^{-1} P_i = \delta^{-1} \bigcup P_i\), \[m_k(E) \leq \sum_{i} |\delta^{-1} P_i| = \delta_1^{-1} \cdots \delta_d^{-1} \sum_{i} |P_i| < m_k(\delta E) + \varepsilon\]This holds for all \(\varepsilon > 0\), so \(m_k(E) \leq \delta_1^{-1} \cdots \delta_d^{-1} m_k(\delta E)\).
Combining the two results, \(m_k(\delta E) = \delta_1 \cdots \delta_d m_k(E)\).
④ \(O\) is open, then \(\delta O\) is open
\(x \in O\), then there exists \(r > 0\) such that \(B_r(x) \subset O\). Naturally \(\delta B_r(x) \subset \delta O\). Let \(\delta_m = \min \{ \delta_1, \dots, \delta_d \}\), then \(B_{\delta_m r} (\delta x) \subset \delta B_r(x) \subset \delta O\), so \(\delta x\) is an interior point of \(\delta O\). This holds for all \(x \in O\), so \(\delta O\) is open.
⑤ \(\delta E\) is measurable
For given \(\varepsilon > 0\), there is an open set \(O\) such that \(E \subset O\) and \(m_k(O - E) < \varepsilon\). \[\delta O - \delta E = \delta O \cap (\delta E)^c = \delta O \cap \delta E^c = \delta(O \cap E^c) = \delta(O - E)\]Then, \(m_k(\delta O - \delta E) = m_k(\delta(O - E)) = \delta_1 \cdots \delta_d m(O - E) < \delta_1 \cdots \delta_d \varepsilon\). This holds for all \(\varepsilon > 0\), so \(\delta E\) is measurable.
8. \(L\) is a linear transformation of \(\mathbb{R}^d\) (to \(\mathbb{R}^{d'}\)).
(a)-1. \(E\) is a compact set in \(\mathbb{R}^d\). Then, \(L(E)\) is also compact.
proof
① \(\| L(x) \| \leq M |x|\) for some \(M > 0\), \(x \in \mathbb{R}^d\)
Let \(x = x_1 e_1 + x_2 e_2 + \cdots + x_d e_d\), \(x_i \in \mathbb{R}\) and \(e_i\) is a unit vector. Then \(L(x) = x_1 L(e_1) + \cdots + x_d L(e_d)\) and \[\| L(x) \| \leq |x_1| \| L(e_1) \| + \cdots + |x_d| \| L(e_d) \| \leq (x_1^2 + \cdots + x_d^2)^{1/2} ( \| L(e_1) \|^2 + \cdots + \| L(e_d) \|^2 )^{1/2}\] (\(\because\) Cauchy-Schwartz ineq) \(\triangleq M |x|\).
② \(L\) is continuous
For \(x, y \in \mathbb{R}^d\) with \(|x - y| < \delta = \varepsilon / M\), \[\| L(x) - L(y) \| = \| L(x - y) \| \leq M |x - y| < \varepsilon.\] This holds for all \(\varepsilon > 0\), so \(L\) is (uniformly) continuous.
③ \(L(E)\) is compact, conclusion
Since \(L\) is continuous, \(L(E)\) is compact. (PMA 4.1A)
(a)-2. \(E\) is \(F_\sigma\), then \(L(E)\) is also \(F_\sigma\).
proof
① \(F = \bigcup_{n=1}^{\infty} K_n\), where \(F\) is closed set and \(K_n\) is a compact set in \(\mathbb{R}^d\)
\(\overline{B_n(0)}\) is closed and
bounded, so it is compact. Let \(K_n = F \cap
\overline{B_n(0)}\), which is an intersection of closed and
compact sets. So, \(K_n\) is
compact.
When \(x \in \bigcup K_n\), there is
\(N \in \mathbb{N}\) such that \(x \in K_N\). Since \(K_N \subset F\), \(x \in F\). \(\forall x \in F\) : there is \(N \in \mathbb{N}\) with \(N \geq |x|\), which means \(x \in \overline{B_N(0)}\). Since \(x \in K_N\), \(x
\in \bigcup K_n\). Thus, \(F = \bigcup
K_n\).
② \(L(E)\) is \(F_\sigma\)
Let \(E = \bigcup_{n=1}^{\infty} F_n\), where \(F_n\) is closed. Then, \(L(E) = L( \bigcup F_n ) = \bigcup L(F_n)\). Each \(F_n\) can be written as \(F_n = \bigcup_{k=1}^{\infty} K_{n,k}\) where \(K_{n,k}\) is compact. Then, \(L(F_n) = \bigcup_{k=1}^{\infty} L(K_{n,k})\) and thus, \(L(E) = \bigcup_{n=1}^{\infty} \bigcup_{k=1}^{\infty} L(K_{n,k})\). \(L(K_{n,k})\) is compact, which is closed as well. Thus, \(L(E)\) is an \(F_\sigma\) set.
(b)-1. For a cube \(Q \subset \mathbb{R}^d\), \(L(Q) \subset Q'\) for some cube \(Q'\) in \(\mathbb{R}^n\).
proof
① \(\|L(x) - L(y)\| \leq \sqrt{d} M \ell\)
Let \(Q\) be a cube of length \(\ell\) in \(\mathbb{R}^d\) and let \(x\) be a corner of the \(Q\). For \(y \in Q\), \(|y - x| \leq \sqrt{d} \ell\). So, \(\|L(x) - L(y)\| \leq M |x - y| \leq \sqrt{d} M \ell\).
② \(L(Q) \subset Q'\) with length \(2 \sqrt{d} M \ell\)
Let \(Q'\) be a cube in \(\mathbb{R}^n\) with length \(2 \sqrt{d} M \ell\) centered at \(L(x)\). The minimum distance between \(L(x)\) and the exterior of the \(Q'\) is \(\sqrt{d} M \ell\). Since \(\|L(x) - L(y)\| \leq \sqrt{d} M \ell\) for all \(y \in Q\), \(L(Q) \subset Q'\). Let \(|Q'| = (2 \sqrt{d} M)^n |Q| \triangleq c |Q|\).
(b)-2. \(m(A) = 0\), then \(m(L(A)) = 0\).
proof
For given \(\varepsilon > 0\), there is a collection of cubes \(\{ Q_i \} \subset \mathbb{R}^d\) such that \(\sum_{i} |Q_i| < \varepsilon\). Let \(Q_i'\) be a cube in \(\mathbb{R}^n\) with the same property in ①. Since \(A \subset \bigcup Q_i\), \(L(A) \subset \bigcup L(Q_i) \subset \bigcup Q_i'\) and \(m_k(L(A)) \leq \sum_{i} |Q_i'| = c \sum_{i} |Q_i| < c \varepsilon\). This holds for all \(\varepsilon > 0\), so \(m_k(L(A)) = m(L(A)) = 0\).
(c) \(E\) is measurable, then \(L(E)\) is measurable.
proof
\(E\) is measurable, so it can be written as \(E = F \cup (E - F)\), where \(F\) is a set of \(F_\sigma\) and \(m(E - F) = 0\). We’ve shown that \(L(F)\) is a countable union of closed sets, which is measurable. Since \(m(E - F) = 0\), \(m(L(E - F)) = 0\) (by (b)-2). Then, \(L(E)\) is a union of measurable sets. Thus, \(L(E)\) is measurable.
11. Let \(A\) be a subset of \([0,1]\) which consists of all numbers without the digit ‘4’ appearing in their decimal expansion. Show that \(m(A) = 0\).
proof
For any \(a \in [0,1]\), we’ll
express as \(a = 0.a_1 a_2 a_3 \dots\),
where \(a_i = 0,1, \dots, 9\) (\(i \in \mathbb{N}\)).
At the first step, the number \(a\)
with \(0.4 \leq a < 0.5\) will be
removed, so \(m(A_1) =
\frac{9}{10}\).
At the second step, the number with \(0.a_1 4
\leq a < 0.a_1 5\) (\(a_1 =
0,1,2,3,5,6,7,8,9\)) will be removed.
So \(m(A_2) = \frac{9}{10} \times \frac{9}{10}
= \left( \frac{9}{10} \right)^2\).
After \(n\)th step, \(m(A_n) = \left( \frac{9}{10}
\right)^n\).
Since \(\lim_{n \to \infty} m(A_n) =
0\), \(m(A_1) < \infty\), and
\(\bigcap_{n} A_n = A\), \(m(A) = 0\).
12. In \(\mathbb{R}^n\), show that
(a) an open disc \(B\) is not a disjoint (countable) union of open rectangles.
proof
Assume \(B = \bigcup_{i=1}^{\infty}
R_i\), where \(R_i\) is an open
rectangle and \(R_i \cap R_j =
\emptyset\) for \(i \neq
j\)
Consider \(x \in \partial R_i\)
\(x\) is a limit point of \(R_i\), so \(B_r(x)\) contains a point \(y \in R_i\) for all \(r > 0\)
Since \(x \in B\), this means \(x \in \overline{B}\)
When \(x \in (\overline{R_i} \cap
B)\), \(x\) is a point of \(B\) but not in \(R_i\)
Then, there is some \(R_N\) such that
\(x \in R_N\)
\(R_N\) is open, so there is \(r > 0\) such that \(B_r(x) \subset R_N\)
Since \(B_r(x)\) contains a point of
\(R_i\) (\(\because\) \(x\) is a limit point of \(R_i\)), \(R_i
\cap R_N\) is not empty
This leads to a contradiction
(b) an open connected set \(\Omega\) is the disjoint union of open rectangles iff \(\Omega\) itself is an open rectangle.
proof
\(\Leftarrow\): \(R = R \cup \emptyset\). Let \(\Omega = R\).
\(\Rightarrow\) (\(\Omega = R_1\))
\(\Omega = \bigcup_{i=1}^{\infty} R_i\), where \(R_i\) is open rectangle and \(R_i \cap R_j = \emptyset\) for all \(i \neq j\). Let \(R_1\) be a non-empty subset (wlog): \(R_1 \subset \Omega\).
Assume that \(R_1 \neq \Omega\)
(\(\exists j > 1\) s.t. \(R_j \neq \emptyset\)).
\(\Omega = R_1 \cup \left(
\bigcup_{j=2}^{\infty} R_j \right)\).
Since \(R_i \cap R_j = \emptyset\)
(\(\forall i \neq j\)), \(R_1 \cap \left( \bigcup_{j=2}^{\infty} R_j \right)
= \emptyset\).
Then \(\Omega\) is a union of two
non-empty disjoint open sets, so \(\Omega\) is disconnected.
This is a contradiction.
Thus, \(R_1 = \Omega\).
13. The following deals with \(G_\delta\) and \(F_\sigma\).
(a) Show that a closed set is \(G_\delta\), and an open set is \(F_\sigma\).
proof
① \(O_n = \{ x : d(x, F) < \frac{1}{n} \}\) is open
Let \(y \in O_n\) and \(\varepsilon = \frac{1}{n} - d(y, F) >
0\)
For \(z \in B_\varepsilon(y)\), \(d(z, F) \leq d(z, y) + d(y, F) < \varepsilon +
d(y, F) = \frac{1}{n}\)
This means \(z \in O_n\), and this
holds for all \(z \in
B_\varepsilon(y)\)
So, \(B_\varepsilon(y) \subset
O_n\)
Since this holds for all \(y \in O_n\),
\(O_n\) is open
② \(F = \bigcap_{n=1}^{\infty} O_n \sim G_\delta\)
\(d(y, F) = 0\) for \(y \in F\), so \(y
\in O_n\) for all \(n\) and
\(y \in \bigcap_{n=1}^{\infty}
O_n\)
So, \(F \subset \bigcap O_n\)
If \(z \in \bigcap O_n\): \(d(z, F) < \frac{1}{n}\) for all \(n\)
Since \(d(z, F) = 0\), \(z \in F = F\)
Thus, \(\bigcap O_n = F\)
③ \(U = F_\sigma\) for open set \(U\) \(\therefore F^c = \bigcup O_n^c\)
(b) Give an example of \(F_\sigma\) which is not \(G_\delta\).
proof
\(\mathbb{Q}\), the set of all rationals, can be represented as a countable union of closed sets, so \(\mathbb{Q}\) is \(F_\sigma\).
① \(\mathbb{R} - \mathbb{Q}\)
The set of all irrationals is \(\mathbb{R}
- \mathbb{Q} = \mathbb{R} - \bigcup \{ q_i \}\)
\(= \mathbb{R} \cap \bigcap ( \{ q_i \}^c ) =
\mathbb{R} \cap \bigcap ( \mathbb{R} \setminus \{ q_i \}
)\)
Each \(\mathbb{R} \setminus \{ q_i
\}^c\) is dense in \(\mathbb{R}\) (\(\because q_i\) is a limit point of \(\mathbb{R} \setminus \{ q_i \}^c\))
So \(\mathbb{R} - \mathbb{Q}\) is a
countable intersection of open and dense subsets
② \(\mathbb{Q}\)
Assume that \(\mathbb{Q} = \bigcap
O_n\) for some open sets \(O_n\)
Naturally, \(\mathbb{Q} \subset O_n\)
for all \(n\)
The point \(x \in \mathbb{R} \setminus
\mathbb{Q}\) is a limit point of \(\mathbb{Q}\)
Since \(O_n^c \cap \mathbb{Q} =
\emptyset\), \(x\) is a limit
point of \(\mathbb{Q} \subset
\mathbb{R}\)
This implies \(O_n\) is dense in \(\mathbb{R}\), and \(\mathbb{Q}\) is a countable intersection of
open & dense sets, which isn’t empty
③ concl.
Then, Baire’s thm (PMA ex 3-22) says \((\mathbb{R} - \mathbb{Q}) \cap \mathbb{Q}\)
is not empty
However, \((\mathbb{R} - \mathbb{Q}) \cap
\mathbb{Q} = \emptyset\), which is a contradiction
(c) Give an example of a Borel set which is not \(G_\delta\) nor \(F_\sigma\).
proof
① \(\mathbb{Q} \cup (\mathbb{R} - \mathbb{Q})\) is a Borel set
Since a closed subset is a Borel set, \(\{
q_i \}\) is a Borel set for all \(q_i
\in \mathbb{Q}\)
The Borel \(\sigma\)-algebra is closed
under countable union, so \(\bigcup \{ q_i
\}\) is a Borel set
Since Borel \(\sigma\)-algebra is
closed under complement, \(\mathbb{Q}^c =
\mathbb{R} - \mathbb{Q}\) is a Borel set
② \(A = (\mathbb{Q} \cap (0,1)) \cup ((\mathbb{R} - \mathbb{Q}) \cap (2,3))\) is not \(G_\delta\), \(F_\sigma\)
1) \(A \neq F_\sigma\)
Assume that \(A = \bigcup F_n\),
where \(F_n\) is a closed set in \(\mathbb{R}\)
Then \((\mathbb{R} - \mathbb{Q}) \cap (2,3) =
A \cap (2,3) = \bigcup (F_n \cap (2,3))\)
The rationals in \((2,3)\) is then
\((2,3) \cap A^c = \bigcap (F_n^c \cap
(2,3))\)
Which is a countable intersection of open sets \(F_n^c \cap (2,3)\)
Each \(F_n \cap (2,3)\) contains all
rationals in \((2,3)\) and possibly
some irrationals in \((2,3)\)
So \(F_n \cap (2,3)\) is dense in \((2,3)\)
Let \(q_n\) be a rational in \((2,3)\)
\(q_n\) is a limit point of \(F_n \cap (2,3)\)
So \((F_n \cap (2,3)) - \{ q_n \}\) is
also dense in \((2,3)\)
Then \(\bigcap (F_n \cap (2,3)) \cap \bigcap
\{ q_n \}^c = \bigcap (F_n \cap (2,3) - \{ q_n \})\) is not empty
by Baire’s thm
Since \(\bigcap \{ q_n \}^c\) does not
contain any rationals in \((2,3)\) and
\(\bigcap (F_n \cap (2,3))\) is all
rationals in \((2,3)\)
Their intersection should be empty, which is a contradiction
2) \(A \neq G_\delta\)
Also assume that \(A = \bigcap
O_n\), where \(O_n\) is an open
set in \(\mathbb{R}\)
Then, the rationals in \((0,1)\) can be
expressed as \(\mathbb{Q} \cap (0,1) = A \cap
(0,1) = \bigcap (O_n \cap (0,1))\)
Which is a countable intersection of open sets
\(O_n \cap (0,1)\) contains all
rationals in \((0,1)\) and possibly
some irrationals in \((0,1)\)
So \(O_n \cap (0,1)\) is dense in \((0,1)\)
Let \(q_n\) be a rational in \((0,1)\)
Then \(O_n \cap (0,1) - \{ q_n \}\) is
also open and dense in \((0,1)\)
Then \(\bigcap (O_n \cap (0,1)) \cap \bigcap
\{ q_n \}^c\) is not empty by Baire’s thm
\(\bigcap \{ q_n \}^c\) does not have
any rational in \((0,1)\), so \(\bigcap (O_n \cap (0,1)) = \emptyset\),
which is contradiction
14. The outer Jordan content \(J_*(E)\) in \(\mathbb{R}\) is defined as \(J_*(E) = \inf \sum_{i=1}^{n} |I_i|\) where inf is taken over every finite covering \(E \subset \bigcup_{i=1}^{n} I_i\) by intervals \(I_i\).
(a) Prove that \(J_*(E) = J_*(\overline{E})\) for every set \(E\) in \(\mathbb{R}\).
proof
① \(J_*(E) \leq J_*(\overline{E})\) for \(E \subset \overline{E}\) and thus \(J_*(E) \leq J_*(\overline{E})\)
Let \(\widetilde{C_1} = \{ \sum |I_i| : E
\subset \bigcup I_i \}\) and \(\widetilde{C_2} = \{ \sum |I_i| : \overline{E}
\subset \bigcup I_i \}\)
Then \(J_*(E) = \inf \widetilde{C_1}\)
and \(J_*(\overline{E}) = \inf
\widetilde{C_2}\)
Since \(E \subset \overline{E}\), the
covering of \(\overline{E}\) also
covers \(E\)
If \(C \in \widetilde{C_2}\), then
\(C \in \widetilde{C_1}\)
This means \(\widetilde{C_2} \subset
\widetilde{C_1}\)
Thus, \(\inf \widetilde{C_2} =
J_*(\overline{E}) \geq \inf \widetilde{C_1} = J_*(E)\)
② \(J_*(E) \geq J_*(\overline{E})\)
Let \(I = \bigcup_{i=1}^{n} I_i\)
covers \(E\): \(E \subset I\)
Since \(E \subset I\) implies \(\overline{E} \subset \overline{I} = \bigcup
\overline{I_i}\)
Let \(S_E = \sum_{i=1}^{n} |I_i|: E \subset
\bigcup I_i\)
Since \(|\overline{I_i}| = |I_i|\),
\(\sum |I_i| = \sum |\overline{I_i}|\)
and thus \(S_E \leq
S_{\overline{E}}\)
So, \(\inf S_E = J_*(E) \leq J_*(\overline{E})
= \inf S_{\overline{E}}\) (\(\because
\sum |I_i| \leq S_{\overline{E}}\))
(b) Exhibit a countable subset \(E \subset [0,1]\) such that \(J_*(E) = 1\) and \(m_k(E) = 0\).
proof
Let \(E = \mathbb{Q} \cap [0,1]\),
the rationals between \(0\) and \(1\)
Since \(E\) is dense in \([0,1]\), \(\overline{E} = [0,1]\)
Then, \(J_*(E) = J_*([0,1]) = 1\),
since \([0,1]\) is the minimum interval
that covers \([0,1]\)
For each rational \(q_n\), \(m_k(\{ q_n \}) = 0\)
So, \(m_k(E) = \sum_{n} m_k(\{ q_n \}) =
0\)
15. \(m_k^R(E)\) is defined as follows: \(m_k^R(E) = \inf \sum_{i} |R_i| : E \subset \bigcup_{i} R_i\). Show that \(m_k^R(E) = m_k(E)\).
proof
① \(m_k^R(E) \leq m_k(E)\)
Let \(C_R = \{ \sum |R_i| : E \subset
\bigcup R_i \}\) and \(C_Q = \{ \sum
|Q_i| : E \subset \bigcup Q_i \}\).
Consider a covering by cubes: \(E \subset
\bigcup Q_i\).
A cube is a kind of rectangle, so \(\bigcup
Q_i\) can be viewed as a covering by rectangles.
Then \(\sum |Q_i|\), which is an
element of \(C_Q\), is also an element
of \(C_R\): \(C_Q \subset C_R\).
Thus, \(\inf C_R = m_k^R(E) \leq m_k(E) = \inf
C_Q\).
② \(m_k^R(E) \geq m_k(E)\)
\(\sum |R_i| \leq m_k^R(E) + \varepsilon\), by def.
For given \(\varepsilon > 0\), there exists a covering of \(E \subset \bigcup R_i\) such that \(\sum |R_i| \leq m_k^R(E) + \varepsilon\).\(R_i = \bigcup Q_i^{(m)}\), so \(\sum |R_i| = \sum |Q_i^{(m)}|\)
Using the similar process in thm 14.4, a rectangle can be represented as a countable union of almost disjoint closed cubes: \(R_i = \bigcup Q_i^{(m)}\).
So, \(m_k(R_i) = \sum m_k(Q_i^{(m)}) = \sum |Q_i^{(m)}| = \sum |R_i|\).\(m_k(E) \leq \sum |Q_i^{(m)}|\), and \(m_k(E) \leq m_k^R(E)\), conclusion.
The covering \(\bigcup Q_i^{(m)}\) also covers \(E\) (\(\because E \subset \bigcup R_i = \bigcup Q_i^{(m)}\)), so \(m_k(E) \leq \sum |Q_i^{(m)}|\).
Then, \(m_k(E) \leq \sum |Q_i^{(m)}| = \sum |R_i| \leq m_k^R(E) + \varepsilon\).
This holds for all \(\varepsilon > 0\), so \(m_k(E) \leq m_k^R(E)\).
③ \(\therefore m_k^R(E) = m_k(E)\)
16. (Borel-Cantelli Lemma) \(\{ E_k \}\) is a countable family of measurable sets in \(\mathbb{R}^d\) such that \(\sum m(E_k) < \infty\). Let \(E = \limsup E_k = \bigcap_{n=1}^{\infty} \bigcup_{k=n}^{\infty} E_k\).
cf. (Set Theory) Show that \[E = \{ x \in \mathbb{R}^d : x \in E_k \text{ for infinitely many } k \} = \bigcap_{n=1}^{\infty} B_n.\]
proof
① (let \(B_n = \bigcup_{k=n}^{\infty} E_k\)) \(\bigcap_{n=1}^{\infty} B_n \subset E\)
\(x \in E^c\), then \(x \in E_k\) for (at most) finitely many
\(k\)
Let \(K-1\) be the maximum of such
\(k\)’s, then \(x \in E_k^c\) for all \(k \geq K\)
Then, \(x \in \bigcap_{k=K}^{\infty} E_k^c =
(\bigcup_{k=K}^{\infty} E_k)^{c} = B_K^{c}\)
Since \(x \in B_K^c\), \(x \in \bigcap_{n=1}^{\infty} B_n^c\)
Thus, in reverse, \(\bigcap_{n=1}^{\infty} B_n
\subset E\)
② \(E \subset \bigcap_{n=1}^{\infty} B_n\)
\(x \in (\bigcap_{n=1}^{\infty} B_n)^{c} =
\bigcup_{n=1}^{\infty} B_n^c\)
Then, there is a positive integer \(M\)
such that \(x \in B_M^c =
\bigcap_{k=M}^{\infty} E_k^c\)
Since \(x \in E_k^c\) for all \(k \geq M\), \(x
\in E_k\) for at most finitely many \(k\)
Thus, \(x \notin E\) and \(E \subset \bigcap_{n=1}^{\infty} B_n\)
③ \(\therefore E = \bigcap_{n=1}^{\infty} B_n\)
(a) Show that \(E\) is measurable
proof
\(E_k\) is measurable for all \(k\), so \(B_n =
\bigcup_{k=n}^{\infty} E_k\) is a countable union of measurable
sets, which is measurable
Thus, \(E = \bigcap_{n=1}^{\infty}
B_n\) is a countable intersection of measurable sets, which is
measurable
(b) Show that \(m(E) = 0\)
proof
Since \(E \subset B_n\), \(m(E) \leq m(B_n)\) by monotonicity
\(B_n = \bigcup_{k=n}^{\infty} E_k\),
so \(m(B_n) \leq \sum_{k=n}^{\infty}
m(E_k)\) by sub-additivity
Since \(\sum_{k=1}^{\infty} m(E_k)\) is
finite (converges), there is a positive integer \(N_\varepsilon\) such that \(n \geq N_\varepsilon\) implies \(\sum_{k=n}^{\infty} m(E_k) <
\varepsilon\) for given \(\varepsilon
> 0\) (Cauchy criterion)
Then, \(\sum_{k=n}^{\infty} m(E_k) <
\varepsilon\)
So, \(m(E) \leq m(B_n) \leq
\sum_{k=n}^{\infty} m(E_k) < \varepsilon\)
Since \(m(E) < \varepsilon\) holds
for all \(\varepsilon > 0\), \(m(E) = 0\)
17. Let \(\{f_n\}\) be a sequence of measurable functions on \([0,1]\) with \(|f_n(x)| < \infty\) a.e. \(x\). Show that there is a sequence \(\{c_n\}\) of positive real numbers such that \(\lim_{n \to \infty} \frac{f_n(x)}{c_n} = 0\) a.e. \(x\).
proof
① \(E_{n,\infty} = \bigcup_{k=n}^{\infty} E_{nk}\) (fixed \(n\))
Let \(E_{nk} = \{ x \in [0,1] : |f_n(x)|
> \frac{k}{n} \}\) and \(E_{n,\infty} = \{ x \in [0,1] : |f_n(x)| = \infty
\}\)
If \(x \in E_{n,\infty}\), \(|f_n(x)| > \frac{k}{n}\) for all \(k \in \mathbb{N}\), which leads to \(x \in E_{nk}\)
If \(|f_n(x)| > \frac{k}{n}\) for
all \(k \in \mathbb{N}\), \(|f_n(x)| \geq \frac{k}{n}\) for all \(k \in \mathbb{N}\), \(\lim_{k \to \infty} \frac{k}{n} =
\infty\)
This means \(x \in E_{n,\infty}\)
Thus, \(E_{n,\infty} = \bigcup_{k=1}^{\infty}
E_{nk}\)
② \(m(E_{n,\infty}) = \lim_{k \to \infty} m(E_{nk})\)
When \(x \in E_{n,k+1}\), \(|f_n(x)| > \frac{k+1}{n} >
\frac{k}{n}\)
Thus, \(x \in E_{nk}\) and \(E_{n,k+1} \subset E_{nk}\)
\(m(E_{n,1}) < \infty\) (\(\because E_{n,1} \subset [0,1]\))
Thus, \(m(E_{n,\infty}) = \lim_{k \to \infty}
m(E_{nk})\) (\(= 0\): \(|f_n(x)| < \infty\) a.e.)
③ \(\exists k\) s.t. \(m(E_{nk}) < \frac{1}{2^n}\), let \(c_n\) be such \(k\)
Assume that there is no such \(k \in
\mathbb{N}\) that \(m(E_{nk}) <
\frac{1}{2^n}\)
In other words, \(m(E_{nk}) \geq
\frac{1}{2^n}\) for all \(k \in
\mathbb{N}\)
Then, \(\lim_{k \to \infty} m(E_{nk}) =
m(E_{n,\infty}) = \frac{1}{2^n}\), which is a contradiction
(\(\because m(E_{n,\infty}) =
0\))
Let \(c_n\) be such \(k\), and let \(E_n = \{ x \in [0,1] : \frac{|f_n(x)|}{c_n} > 1
\} = E_{nc_n}\)
④ \(m(E) = 0\), where \(E = \limsup_{n \to \infty} E_n\)
\(E_n = \{ x : f_n(x) > c_n \} \cup \{ x
: -f_n(x) > c_n \}\), which is a union of measurable sets
(\(\because f_n\) is a measurable
function)
We’ve defined \(m(E_n) <
\frac{1}{2^n}\), so \(\sum_{n=1}^{\infty} m(E_n) < 1\), which
is finite
Then, \(m(E) = 0\) by Borel-Cantelli
lemma
⑤ \(\lim_{n \to \infty} \frac{f_n(x)}{c_n} = 0\), conclusion
\(E = \{ x \in [0,1] : x \in E_n \text{ for
infinitely many } n \}\)
So \(x \in E^c\) implies \(x \in E_n\) for at most finitely many \(n\)
This means there is an integer \(N\)
such that \(n \geq N\) implies \(x \notin E_n\)
Then, \(n \geq N\) implies \(\left| \frac{f_n(x)}{c_n} \right| \leq
1\)
Since \(\frac{f_n(x)}{c_n} \to 0\),
\(\lim_{n \to \infty} \frac{f_n(x)}{c_n} =
0\)
18. Show that every measurable function is the limit of a sequence of continuous functions a.e.
proof
① continuous \(f_n\) on \(E\)
Let \(f\) be a measurable function
on \(E\) with \(m(E) < \infty\)
Let \(F_n\) be a closed subset of \(\mathbb{R}^d\) such that \(F_n \subset E\), \(m(E - F_n) < \frac{1}{2^n}\), and \(f\) is continuous on \(F_n\) (by Lusin’s thm 1.4-(7))
Let \(f_n(x)\) be a continuous function
on \(E\) such that \(f_n(x) = f(x)\) for all \(x \in F_n\) (Tietze’s extension thm PMA ex
4-(7))
② \(B := \{ x \in E : f_n(x) \not\to f(x) \} = \limsup_{n \to \infty} B_n\)
Let \(B_n = E - F_n\)
For \(x \in B\), assume that \(x \in B_n\) for at most finitely many \(n\)’s
Then there exists an integer \(N\) such
that \(x \notin B_n\) for all \(n \geq N\)
This means \(x \in F_n\) for all \(n \geq N\) (\(\because E = B_n \cup F_n\), \(B_n \cap F_n = \emptyset\))
So \(f_n(x) = f(x)\) for \(n \geq N\) and \(\lim_{n \to \infty} f_n(x) = f(x)\)
This means \(x \notin B\), which is a
contradiction
So, \(B = \{ x \in E : x \in B_n \text{ for infinitely many } n \}\) and thus \(B = \limsup_{n \to \infty} B_n\)
③ \(m(B) = 0\), conclusion
Since \(m(B_n) < \frac{1}{2^n}\),
which leads to \(\sum_{n=1}^{\infty} m(B_n)
< 1\) (finite), \(B = \limsup_{n}
B_n\) has measure zero
Thus, \(\lim_{n \to \infty} f_n(x) =
f(x)\) for a.e. \(x \in E\)