[測度論] DCT 應用 (2) - Simple Fubini-Tonollei Theorem

Theorem:
假設 $\{f_n\}\subset L^1$ 使得 $\sum_{n=1}^\infty\int |f_n| <\infty$。則
(a) $\sum_{n=1}^\infty f_n$ almost everywhere 收斂 。
(b) $\int \sum_{n=1}^\infty f_n = \sum_{n=1}^\infty \int f_n$。

Proof:(a) 令$\{f_n\}\subset L^1$ 使得 $\sum_{n=1}^\infty\int |f_n| <\infty$，則我們有
\[
\int \sum_{n=1}^\infty |f_n| = \sum_{n=1}^\infty\int |f_n|   <\infty
\]故我們知道 $ \sum_{n=1}^\infty |f_n| \in L^1$ 且 $\sum_{n=1}^\infty |f_n|<\infty$ almost everywhere。此表明對 almost every $x$， $\sum_{n=1}^\infty f_n(x) $ (絕對)收斂。令極限
$$ f(x):=\lim_{N\to \infty} \sum_{n=1}^N f_n(x) $$ 故我們僅需證明此 $f \in L^1$。注意到 對任意 $N$， $\lim_N \sum_{n=1}^N f_n \leq \lim_N \sum_{n=1}^\infty |f_n|$ a.e. 故 $f \leq \lim_N \sum_{n=1}^N |f_n|$ a.e. 由於 $\lim_N \sum_{n=1}^N |f_n| \in L^1$ 故 $f \in L^1$。

Proof (b)
要證明 $\int \sum_{n=1}^\infty f_n = \sum_{n=1}^\infty \int f_n$。首先觀察
\[
 \int \sum_{n=1}^N f_n = \sum_{n=1}^N \int  f_n \;\;\;\; (*)
\]由於 $\sum_{n=1}^N f_n \leq \sum_{n=1}^\infty |f_n| \in L^1$ 且 $\sum_{n=1}^N f_n \to f$ a.e. in $L^1$ ，由 Dominated Convergence Theorem 我們有\[
\lim_{N\to \infty} \int \sum_{n=1}^N f_n =  \int \lim_{N\to \infty}\sum_{n=1}^N f_n = \int \sum_{n=1}^\infty f_n\;\;\;\; (**)
\]故對 $(*)$兩邊同取極限並利用上述結果 $(**)$ 可得
\[
\int \sum_{n=1}^\infty f_n = \sum_{n=1}^\infty \int f_n
\]至此證明完畢。$\square$

[測度論] Dominated Convergence Theorem

Theorem: Dominated Convergence Theorem DCT  (real-valued functions)
令 sequence $\{f_n\} \subset L^1$ 滿足
(a) $f_n \to f$ almost everywhere
(b) 存在非負函數 $g \in L^1$ 使得 $|f_n| \leq g$ almost everywhere for all $n$
則\[
f \in L^1
\]且
\[
\lim_n \int f_n =\int f
\]
Proof:  先證 $f \in L^1$：由於 $f_n \to f$ almost everywhere  且 $f_n \in L^1$，可知 $f $ measurable (see Proposition 2.11/2.12 )。由於 $|f_n| \leq g$ almost everywhere 故 $|f| \leq g$ almost everywhere，故 $f \in L^1$。

接著我們證 $\lim_n \int f_n =\int f$：注意到 $|f_n| \leq g$ almost everywhere，故我們有
\[
-g \leq f_n \leq g \text{ almost everywhere}
\]換言之，我們有
\[
f_n + g \geq 0\text{ almost everywhere}
\]與
\[
g-f_n \geq 0 \text{almost everywhere}
\]也就是說 $\{f_n + g\}, \{g-f_n\} \in L^+$。由 Fatou Lemma 我們有
\[
\begin{gathered}
  \int {\mathop {\lim \inf }\limits_n } ({f_n} + g) \leqslant \lim \inf \int {({f_n} + g)} ; \hfill \\
  \int {\mathop {\lim \inf }\limits_n } (g - {f_n}) \leqslant \lim \inf \int {(g - {f_n})}  \hfill \\
\end{gathered} \;\;\; (*)
\]注意到上述積分不等式左方，積分項有以下結果 (利用 limsup or liminf 的性質)，我們有 $\mathop {\lim \inf }\limits_n \left( {{f_n} + g} \right) = g + \mathop {\lim \inf }\limits_n {f_n}$ 以及 $$\mathop {\lim \inf }\limits_n \left( {g - {f_n}} \right) = g + \mathop {\lim \inf }\limits_n \left( { - {f_n}} \right) = g - \mathop {\lim \sup }\limits_n {f_n}
$$同理，對於積分不等式右方，我們有
\[\lim \inf \int {({f_n} + g)}  = \lim \inf \left( {\int {{f_n} + \int g } } \right) = \lim \inf \int {{f_n} + \int g } \]與
\[\lim \inf \int {(g - {f_n})  = \int {g + \lim \inf \left( { - \int {{f_n}} } \right)}  = \int {g - \lim \sup \int {{f_n}} } } \]
將此結果帶入 $(*)$ 我們有
\begin{align*}
  &\left\{ \begin{gathered}
  \int {\mathop {\lim \inf }\limits_n } ({f_n} + g) \leqslant \lim \inf \int {({f_n} + g)} ; \hfill \\
  \int {\mathop {\lim \inf }\limits_n } (g - {f_n}) \leqslant \lim \inf \int {(g - {f_n})}  \hfill \\
\end{gathered}  \right. \hfill \\
   &\Rightarrow \left\{ \begin{gathered}
  \int g  + \int {\mathop {\lim \inf }\limits_n } {f_n} \leqslant \lim \inf \int {{f_n} + \int g } ; \hfill \\
  \int g  - \int {\mathop {\lim \sup }\limits_n {f_n}}  \leqslant \int {g - \lim \sup \int {{f_n}} }  \hfill \\
\end{gathered}  \right. \hfill \\
   &\Rightarrow \left\{ \begin{gathered}
  \int {\mathop {\lim \inf }\limits_n } {f_n} \leqslant \lim \inf \int {{f_n}} ; \hfill \\
   - \int {\mathop {\lim \sup }\limits_n {f_n}}  \leqslant  - \lim \sup \int {{f_n}}  \hfill \\
\end{gathered}  \right. \hfill \\
\end{align*} 注意到 $f = \lim_n f_n = \liminf_n f_n = \limsup_n f_n$ ，而且 $\{x: \lim_n f_n \neq f(x)\}$ 的測度為 $0$，故 $\int f = \int \liminf f_n  = \int \limsup f_n$。 現在我們進一步改寫上式
\[\begin{gathered}
  \left\{ \begin{gathered}
  \int f  \leqslant \lim \inf \int {{f_n}} ; \hfill \\
  \int f  \geqslant \lim \sup \int {{f_n}}  \hfill \\
\end{gathered}  \right. \hfill \\
   \Rightarrow \lim \sup \int {{f_n} \leqslant \int {f \leqslant \lim \inf \int {{f_n}} } }  \hfill \\
\end{gathered} \]亦即 $\int f_n \to \int f$。至此得證。$\square$



Remarks:
$\liminf (a_k + b+k) = \liminf a_k + \liminf b_k$ 不全為然。Counter example?

上述等式成立僅僅 $\lim a_k = a$ 則 $\liminf_k (a_k + b_k) = a+\liminf b_k$。

[測度論] DCT 應用：積分例子(1)

試證下列積分
\[
\lim_{n \to \infty} \int_0^\infty (1+(x/n))^{-n} \sin(x/n)dx = 0
\] Proof: 對任意 $x \in [0,\infty)$而言，令 $f_n(x):=(1+(x/n))^{-n} \sin(x/n)$，則 $f_n \to 0$。另外我們注意到 因為 $\sin(x/n) \leq 1 $ 以及利用 附註的 Claim 可知 $(1+x/n)^{-n} \leq e^{-x}$ 故
\[
f_n(x) = (1+(x/n))^{-n} \sin(x/n) \leq e^{-x}
\]現在我們令 $g(x):=e^{-x} \in L^1([0,\infty),m)$。為此我們觀察
\[
\int_0^\infty e^{-x} dx = 1 < \infty
\]故 $g \in L^1([0,\infty),m)$ 且 $|f_n(x)| \leq g(x)$，故由 Dominanted Convergence Theorem 可知
\begin{align*}
\lim_{n \to \infty} \int_0^\infty (1+(x/n))^{-n} \sin(x/n)dx &=  \int_0^\infty \lim_{n \to \infty}(1+(x/n))^{-n} \sin(x/n)dx\\
&=\int_0^\infty  0 dx = 0
\end{align*}
至此證明完畢。$\square$



Claim: 對任意 $x\in(0,\infty)$ 與 $n \in \mathbb{N}$，我們有 \[
(1+x/n)^{-n} \leq e^{-x}
\] Proof:
注意到 $\log (1+x/n)^{-n} = -n \log(1+x/n) \leq -n \frac{x}{n}=-x$。現在對兩邊同取 $exp()$ 可得
\[
(1+x/n)^{-n} \leq e^{-x}
\]至此得證。$\square$