控制收敛定理(Dominated Convergence Theorem)是继Fatou's Lemma与单调收敛定理(Monotone Convergence Theorem)后的又一重要定理,在测度论、条件期望、随机微积分中有诸多重要应用。
在证明前先引入一条引理:
Lemma
对于任意实序列 \(\left\{ a_{n} \right\}^{\infty}_{n=1}\),都有:
\[\bf \liminf\limits_{n \rightarrow \infty} ~ \left( -a_{n} \right) = -\limsup\limits_{n \rightarrow \infty} ~ \left( a_{n} \right)
\]
Proof. (Lemma)
对于: \(\forall N \in \mathbb{N}: \inf\limits_{k \geq N} \left( -a_{k} \right) = \inf\left\{ -a_{N}, -a_{N+1}, -a_{N+2}, \ldots \right\}\),
即,由定义:
\[\forall k \geq N: ~ -a_{k} \geq \inf\limits_{k \geq N} \left( -a_{k} \right)
\]
并且:
\[\forall \epsilon > 0: ~ \exists k_{0} \in \mathbb{N}: ~ -a_{k_{0}} < \inf\limits_{k \geq N} \left( -a_{k} \right) + \epsilon
\]
而以上两个statement等价于:
\[\forall k \geq N: ~ a_{k} \leq -\inf\limits_{k \geq N} \left( -a_{k} \right)
\]
与:
\[\forall \epsilon > 0: ~ \exists k_{0} \in \mathbb{N}: ~ a_{k_{0}} > -\inf\limits_{k \geq N} \left( -a_{k} \right) - \epsilon
\]
可以发现,若将 \(-\inf\limits_{k \geq N} \left( -a_{k} \right)\) 视作一个整体,则恰好满足supremum的定义,即:
\[\forall N \in \mathbb{N}: ~ -\inf\limits_{k \geq N} \left( -a_{k} \right) = \sup\limits_{k \geq N} \left( a_{k} \right)
\]
所以:
\[\forall N \in \mathbb{N}: ~ \sup\limits_{k \geq N} \left( a_{k} \right) + \inf\limits_{k \geq N} \left( -a_{k} \right) = 0
\]
所以:
\[\lim\limits_{N \rightarrow \infty} \left( \sup\limits_{k \geq N}\left( a_{k} \right) + \inf\limits_{k \geq N} \left( -a_{k} \right) \right) ~ = ~ \lim\limits_{N \rightarrow \infty} \sup\limits_{k \geq N}\left( a_{k} \right) + \lim\limits_{N \rightarrow \infty}\inf\limits_{k \geq N} \left( -a_{k} \right) = 0\\
\implies \qquad \qquad \qquad \limsup\limits_{n \rightarrow \infty} \left( a_{n} \right) ~ + ~ \liminf\limits_{n \rightarrow \infty} \left( -a_{n} \right) ~ = ~ 0 \qquad \qquad \qquad \qquad
\]
Dominated Convergence Theorem(控制收敛定理)
令 \(\left\{ f_{n} \right\}^{\infty}_{n=1}\) 为 \(E \in \mathcal{M}\) 上的可测函数序列,\(g\) 为同在 \(E\) 上的勒贝格可积函数(即,\(\int_{E} ~ g ~ dm < \infty\)),满足对于 $\forall n \geq 1: |f_{n}| \leq g ~ $ almost surely。若 $f = \lim\limits_{n \rightarrow \infty} f_{n} ~ $ almost surely,那么:
- \(f\) 在 \(E\) 上可积
- 且:\(\lim\limits_{n\rightarrow\infty} \int_{E} ~ f_{n} ~ dm = \int_{E} ~ f ~ dm = \int_{E} ~ \lim\limits_{n \rightarrow \infty} f_{n} ~ dm\)
Proof. (Dominated Convergence Theorem)
由:\(|f_{n}| \leq g\),有 \(-g \leq f_{n} \leq g \iff 0 \leq f_{n} + g \leq 2g\),那么:
\[\]
1. \(f_{n} + g \geq 0\) 非负且可积,并且逐点收敛于 \(\left( f + g \right)\),对其应用Fatou's Lemma:
\[\liminf\limits_{n \rightarrow \infty} \int_{E} ~ \left( f_{n} + g \right) ~ dm \geq \int_{E} ~ \liminf\limits_{n \rightarrow \infty} \left( f_{n} + g \right) ~ dm = \int_{E} ~ \left( f + g \right) ~ dm
\]
则:
\[\liminf\limits_{n \rightarrow \infty} \left( \int_{E} ~ f_{n} ~ dm + \int_{E} ~ g ~ dm \right) ~ = ~ \liminf\limits_{n \rightarrow \infty} \int_{E} ~ f_{n} ~ dm + \int_{E} ~ g ~ dm ~ \geq ~ \int_{E} ~ f ~ dm + \int_{E} ~ g ~ dm\\
\implies \qquad \qquad \qquad \qquad \qquad \liminf\limits_{n \rightarrow \infty} \int_{E} ~ f_{n} ~ dm ~ \geq ~ \int_{E} ~ f ~ dm \qquad \qquad \qquad \qquad \qquad
\]
\[\]
2. \(2g - \left( f_{n} + g \right) = g - f_{n} \geq 0\) 非负且可积,并且逐点收敛于 \(\left( g - f \right)\),对其应用Fatou's Lemma:
\[\liminf\limits_{n \rightarrow \infty} \int_{E} ~ \left( g - f_{n} \right) ~ dm \geq \int_{E} \liminf\limits_{n \rightarrow \infty} \left( g - f_{n} \right) ~ dm
\]
\[\begin{align*}
\implies \qquad \qquad \int_{E} \liminf\limits_{n \rightarrow \infty} \left( g - f_{n} \right) ~ dm ~ & \leq ~ \liminf\limits_{n \rightarrow \infty} \int_{E} \left( g - f_{n} \right) ~ dm \qquad \qquad \\
& = ~ \liminf\limits_{n \rightarrow \infty} \left( \int_{E} g ~ dm - \int_{E} f ~ dm \right) \qquad \qquad \\
& = ~ \int_{E} g ~ dm + \liminf\limits_{n \rightarrow \infty} \int_{E} \left( -f_{n} \right) ~ dm \qquad \qquad
\end{align*}
\]
由引理所证:
\[\liminf\limits_{n \rightarrow \infty} \int_{E} ~ \left( -f_{n} \right) ~ dm ~ = ~ -\limsup\limits_{n \rightarrow \infty} \int_{E} ~ f_{n} ~ dm
\]
则:
\[\int_{E} ~ \liminf\limits_{n \rightarrow \infty} \left( g - f_{n} \right) ~ dm ~ = ~ \int_{E} ~ \left( g - f \right) ~ dm ~ = ~ \int_{E} ~ g ~ dm - \int_{E} ~ f ~ dm ~ \leq ~ \int_{E} ~ g ~ dm - \limsup\limits_{n \rightarrow \infty} \int_{E} ~ f_{n} ~ dm\\
\implies \qquad \qquad \qquad \qquad \limsup\limits_{n \rightarrow \infty} \int_{E} ~ f_{n} ~ dm \leq \int_{E} ~ f ~ dm \qquad \qquad \qquad \qquad
\]
又由 1. 中所得:
\[\liminf\limits_{n \rightarrow \infty} \int_{E} ~ f_{n} ~ dm ~ \geq ~ \int_{E} ~ f ~ dm
\]
以及显然:
\[\limsup\limits_{n \rightarrow \infty} \int_{E} ~ f_{n} ~ dm \geq \liminf\limits_{n \rightarrow \infty} \int_{E} ~ f_{n} ~ dm
\]
综上则有:
\[\liminf\limits_{n \rightarrow \infty} \int_{E} ~ f_{n} ~ dm ~ \leq ~ \limsup\limits_{n \rightarrow \infty} \int_{E} ~ f_{n} ~ dm ~ \leq ~ \int_{E} ~ f ~ dm ~ \leq ~ \liminf\limits_{n \rightarrow \infty} \int_{E} ~ f_{n} ~ dm
\]
由于当且仅当序列上极限等于下极限时,序列存在极限,且极限值、上极限值、下极限值三者相等,最终有:
\[\liminf\limits_{n \rightarrow \infty} \int_{E} ~ f_{n} ~ dm ~ = ~ \limsup\limits_{n \rightarrow \infty} \int_{E} ~ f_{n} ~ dm ~ = ~ \lim\limits_{n \rightarrow \infty} \int_{E} ~ f_{n} ~ dm ~ = ~ \int_{E} ~ f ~ dm
\]