MT Lesson 16: Monotone converge theorem
Published:
In this section, we talk about the the theorems related to the convergence of sequence of integrals.
Monotone converge theorem
THEOREM Suppose $f_n\geq 0$ and $f_n\uparrow f$, where $f_n\in \mathcal{F}$ on $(\Omega,\mathcal{F},\mu)$, then $\int f_nd\mu\uparrow \int f d\mu$.
PROOF Suppose for any fixed $n$, there exists a simple function sequence $g_{n,k}\uparrow f_n$,
\[\begin{matrix} g_{1,1}&g_{1,2}&\cdots&\uparrow&f_1\\ g_{2,1}&g_{2,2}&\cdots&\uparrow&f_2\\ &&&&\vdots\\ &&&&f \end{matrix}\]then $\int f_nd\mu = \lim_k \int g_{n,k}d\mu$, $\forall n$.
Let $g_k=\max_{1\leq n\leq k}g_{n,k}$, we prove that $\{g_k\}$ is a sequence of simple functions converge to $f$.
- $g_k\leq g_{k+1}$
- $g_k\leq f_k$
- $g_k\uparrow f$
Now we have $\int fd\mu = \lim_k \int g_kd\mu$.
$\int f_nd\mu\leq \int fd\mu \Longrightarrow \lim\int f_nd\mu\leq \int fd\mu$. We have to prove $\lim\int f_nd\mu\geq \int fd\mu$. $g_k\leq f_k\Longrightarrow \int f d\mu = \lim_k \int g_kd\mu \leq \lim_k \int f_kd\mu$. $\square$
REMARK if the theorem is $f_n\geq g$, $g$ integrable, then we can consider the sequence $g_n=f_n-f$, and the result still holds.
Construct measure from integrals
Suppose $f\geq 0$, $\mu_f(A)=\int_Afd\mu$, then
- $\mu_f(\emptyset)=0$
- $\mu_f$ is $\sigma$-additive.
OBS Suppose $B\in \mathcal{F}$ and $\mu(B)=0$, then $\int_B fd\mu=0$, $\mu_f(B)=0$.
DEF (Absolute continuity) Suppose $\mu, \nu$ on $(\Omega, \mathcal{F})$, $\mu$ is absolute continuous with respect to $\nu$ ($\mu\ll \nu$) if $\nu(A)=0\Longrightarrow \mu(A)=0$.
Radon-Nykodym Theorem
THM If $\mu \ll \nu$, $\exists g=\frac{d\mu}{d\nu}\geq0$ s.t. $\mu(B)=\int_Bgd\nu$.
It is possible for two measures $\mu, \nu$ that neither $\mu\ll \nu$ nor $\nu \ll \mu$. Assume $f\geq 0$ on $\mathbb{R}$, $\lambda$ to be the Lebesgue measure, $\mu_f(A)=\int_Afd\lambda$,
\[\delta_x(B)= \begin{cases} 1&x\in B\\ 0&x\notin B \end{cases}\]$\forall x \in B$, $\mu(\left{x\right})=0$, $\delta_x(\left{x\right})=1$, thus $\delta_x\ll \mu_f$ is false. $\forall [a,b)\notin \mathbb{R}$, $\delta_x([a,b))=0$, $\mu_f([a,b))>0$, thus $\mu_f \ll \delta_x$ is false.
Uniform integrability
$\int fd\mu\leq \infty$, $\forall \varepsilon>0$, $\exist \delta > 0$, $\mu(A)<\delta\Longrightarrow \mu_f(A)<\varepsilon$.
Fatou’s Lemma
If $f_n\geq g$, where $g$ is integrable, $\int \underline{\lim}f_nd\mu\leq \underline{\lim}\int f_nd\mu$.
If $f_n\leq g$, where $g$ is integrable, $\int \overline{\lim}f_nd\mu\geq \overline{\lim}\int f_nd\mu$.
Dominated convergence theorem
$f_n\to f$, $ | f_n | \leq g$, where $g$ is integrable, then $f$ is integrable and $\int f_nd\mu \to \int fd\mu$. |
PROOF
\[\overline{\lim}\int f_nd\mu\leq \int \overline{\lim}f_nd\mu=\int fd\mu\\=\int \underline{\lim}f_nd\mu\leq \underline{\lim }\int f_nd\mu\leq \overline{\lim} \int f_nd\mu\]Thus all “$\geq$” are “$=$”.