s23-probability-theory/inputs/lecture_10.tex

339 lines
14 KiB
TeX
Raw Normal View History

2023-07-05 17:53:41 +02:00
\lecture{10}{2023-05-09}{}
2023-05-10 18:56:36 +02:00
First, we will prove some of the most important facts about Fourier transforms.
We consider $(\R, \cB(\R))$.
\begin{notation}
By $M_1 (\R)$ we denote the set of all probability measures on $\left( \R, \cB(\R) \right)$.
\end{notation}
2023-07-12 15:25:25 +02:00
For all $\bP \in M_1(\R)$ we define $\phi_{\bP}(t) = \int_{\R} e^{\i t x}\bP(\dif x)$.
2023-05-10 18:56:36 +02:00
If $X: (\Omega, \cF) \to (\R, \cB(\R))$ is a random variable, we write
$\phi_X(t) \coloneqq \bE[e^{\i t X}] = \phi_{\mu}(t)$,
where $\mu = \bP X^{-1}$.
\begin{refproof}{inversionformula}
2023-07-28 03:45:37 +02:00
We will prove that the limit in the RHS of \yaref{invf}
2023-05-10 18:56:36 +02:00
exists and is equal to the LHS.
Note that the term on the RHS is integrable, as
\[
2023-07-12 15:25:25 +02:00
\lim_{t \to 0} \frac{e^{-\i t b} - e^{-\i t a}}{- \i t} \phi(t) = a - b
2023-07-06 00:36:26 +02:00
\]
2023-05-10 18:56:36 +02:00
and note that $\phi(0) = 1$ and $|\phi(t)| \le 1$.
% TODO think about this
We have
\begin{IEEEeqnarray*}{rCl}
2023-07-12 15:25:25 +02:00
&&\lim_{T \to \infty} \frac{1}{2 \pi} \int_{-T}^T \int_{\R} \frac{e^{-\i t b}- e^{-\i t a}}{-\i t} e^{\i t x} \dif t \bP(\dif x)\\
2023-07-28 22:06:08 +02:00
&\overset{\yaref{thm:fubini}}{=}& \lim_{T \to \infty} \frac{1}{2 \pi} \int_{\R} \int_{-T}^T \frac{e^{-\i t b}- e^{-\i t a}}{-\i t} e^{\i t x} \dif t \bP(\dif x)\\
2023-07-12 15:25:25 +02:00
&=& \lim_{T \to \infty} \frac{1}{2 \pi} \int_{\R} \int_{-T}^T \frac{e^{\i t (b-x)}- e^{\i t (x-a)}}{-\i t} \dif t \bP(\dif x)\\
2023-07-12 16:09:48 +02:00
&=& \lim_{T \to \infty} \frac{1}{2 \pi} \int_{\R} \underbrace{\int_{-T}^T \left[ \frac{\cos(t (x-b)) - \cos(t(x-a))}{-\i t}\right] \dif t}_{=0 \text{, as the function is odd}} \bP(\dif x) \\
&& + \lim_{T \to \infty} \frac{1}{2\pi} \int_{\R}\int_{-T}^T \frac{\sin(t ( x - b)) - \sin(t(x-a))}{-t} \dif t \bP(\dif x)\\
2023-07-12 15:25:25 +02:00
&=& \lim_{T \to \infty} \frac{1}{\pi} \int_\R \int_{0}^T \frac{\sin(t(x-a)) - \sin(t(x-b))}{t} \dif t \bP(\dif x)\\
2023-07-28 22:06:08 +02:00
&\overset{\substack{\yaref{fact:sincint},\text{DCT}}}{=}&
2023-07-12 16:09:48 +02:00
\frac{1}{\pi} \int -\frac{\pi}{2} \One_{x < a} + \frac{\pi}{2} \One_{x > a}
- (- \frac{\pi}{2} \One_{x < b} + \frac{\pi}{2} \One_{x > b}) \bP(\dif x)\\
&=& \frac{1}{2} \bP(\{a\} ) + \frac{1}{2} \bP(\{b\}) + \bP((a,b))\\
2023-05-10 18:56:36 +02:00
&=& \frac{F(b) + F(b-)}{2} - \frac{F(a) - F(a-)}{2}
\end{IEEEeqnarray*}
\end{refproof}
\begin{fact}
2023-07-14 22:07:36 +02:00
\label{fact:sincint}
2023-05-10 18:56:36 +02:00
\[
2023-07-12 15:25:25 +02:00
\int_0^\infty \frac{\sin x}{x} \dif x = \frac{\pi}{2}
2023-07-06 00:36:26 +02:00
\]
2023-05-10 18:56:36 +02:00
where the LHS is an improper Riemann-integral.
Note that the LHS is not Lebesgue-integrable.
It follows that
\begin{IEEEeqnarray*}{rCl}
2023-07-13 00:06:30 +02:00
\lim_{T \to \infty} \int_0^T \frac{\sin(t(x-a))}{t} \dif t &=&
2023-05-10 18:56:36 +02:00
\begin{cases}
2023-07-13 00:06:30 +02:00
- \frac{\pi}{2} &\text{if }x < a,\\
0 &\text{if }x = a,\\
\frac{\pi}{2}&\text{if } x > a.
2023-05-10 18:56:36 +02:00
\end{cases}
\end{IEEEeqnarray*}
\end{fact}
\begin{theorem} % Theorem 3
\label{thm:lec10_3}
2023-07-12 16:39:58 +02:00
Let $\bP \in M_1(\R)$ such that $\phi_\bP \in L^1(\lambda)$.
2023-05-10 18:56:36 +02:00
Then $\bP$ has a continuous probability density given by
\[
2023-07-12 16:39:58 +02:00
f(x) = \frac{1}{2 \pi} \int_{\R} e^{-\i t x} \phi_{\bP}(t) \dif t.
2023-07-06 00:36:26 +02:00
\]
2023-05-10 18:56:36 +02:00
\end{theorem}
\begin{example}
\begin{itemize}
2023-07-12 16:39:58 +02:00
\item Let $\bP = \delta_{0}$.
2023-05-10 18:56:36 +02:00
Then
\[
2023-07-13 00:06:30 +02:00
\phi_{\bP}(t) = \int e^{\i t x} \delta_0(\dif x) = e^{\i t 0 } = 1
2023-07-06 00:36:26 +02:00
\]
2023-05-10 18:56:36 +02:00
\item Let $\bP = \frac{1}{2} \delta_1 + \frac{1}{2} \delta_{-1}$.
Then
\[
2023-07-12 16:39:58 +02:00
\phi_{\bP}(t) = \frac{1}{2} e^{\i t} + \frac{1}{2} e^{- \i t} = \cos(t)
2023-07-06 00:36:26 +02:00
\]
2023-05-10 18:56:36 +02:00
\end{itemize}
\end{example}
\begin{refproof}{thm:lec10_3}
2023-07-12 15:25:25 +02:00
Let $f(x) \coloneqq \frac{1}{2 \pi} \int_{\R} e^{ - \i t x} \phi(t) \dif t$.
2023-05-10 18:56:36 +02:00
\begin{claim}
If $x_n \to x$, then $f(x_n) \to f(x)$.
\end{claim}
\begin{subproof}
2023-07-12 16:39:58 +02:00
Suppose that
$e^{-\i t x_n} \phi(t) \xrightarrow{n \to \infty} e^{-\i t x} \phi(t)$
for all $t$.
Since
2023-05-10 18:56:36 +02:00
\[
|e^{-\i t x} \phi(t)| \le |\phi(t)|
2023-07-06 00:36:26 +02:00
\]
2023-07-12 16:39:58 +02:00
and $\phi \in L^1$,
we get $f(x_n) \to f(x)$
2023-05-10 18:56:36 +02:00
by the dominated convergence theorem.
\end{subproof}
We'll show that for all $a < b$ we have
\[
2023-07-12 16:39:58 +02:00
\bP\left( (a,b] \right) = \int_a^b f(x) \dif x.\label{thm10_3eq1}
2023-07-06 00:36:26 +02:00
\]
2023-05-10 18:56:36 +02:00
Let $F$ be the distribution function of $\bP$.
2023-07-28 03:45:37 +02:00
It is enough to prove \yaref{thm10_3eq1}
2023-05-10 18:56:36 +02:00
for all continuity points $a $ and $ b$ of $F$.
We have
\begin{IEEEeqnarray*}{rCl}
2023-07-28 22:06:08 +02:00
\rhs &\overset{\text{Fubini}}{=}& \frac{1}{2 \pi} \int_{\R} \int_{a}^b e^{-\i t x} \phi(t) \dif x \dif t\\
2023-07-12 15:25:25 +02:00
&=& \frac{1}{2 \pi} \int_\R \phi(t) \int_a^b e^{-\i t x} \dif x \dif t\\
&=& \frac{1}{2\pi} \int_{\R} \phi(t) \left( \frac{e^{-\i t b} - e^{-\i t a}}{- \i t} \right) \dif t\\
&\overset{\text{dominated convergence}}{=}& \lim_{T \to \infty} \frac{1}{2\pi} \int_{-T}^{T} \phi(t) \left( \frac{e^{-\i t b} - e^{- \i t a}}{- \i t} \right) \dif t
2023-05-10 18:56:36 +02:00
\end{IEEEeqnarray*}
2023-07-28 03:45:37 +02:00
By the \yaref{inversionformula},
the RHS is equal to $F(b) - F(a) = \bP\left( (a,b] \right)$.
2023-05-10 18:56:36 +02:00
\end{refproof}
However, Fourier analysis is not only useful for continuous probability density functions:
2023-07-28 03:45:37 +02:00
\begin{theorem}[Bochner's formula for the mass at a point]
\yalabel{Bochner's Formula for the Mass at a Point}{Bochner}{bochnersformula} % Theorem 4
2023-05-10 18:56:36 +02:00
Let $\bP \in M_1(\lambda)$.
Then
\[
2023-07-13 00:38:46 +02:00
\forall x \in \R .~ \bP\left( \{x\} \right) = \lim_{T \to \infty} \frac{1}{2 T} \int_{-T}^T e^{-\i t x } \phi(t) \dif t.
2023-07-06 00:36:26 +02:00
\]
2023-05-10 18:56:36 +02:00
\end{theorem}
\begin{refproof}{bochnersformula}
We have
\begin{IEEEeqnarray*}{rCl}
2023-07-28 22:06:08 +02:00
\rhs &=& \lim_{T \to \infty} \frac{1}{2 T} \int_{-T}^T e^{-\i t x} \int_{\R} e^{\i t y} \bP(\dif y) \\
2023-07-13 00:06:30 +02:00
&\overset{\text{Fubini}}{=}&
\lim_{T \to \infty} \frac{1}{2 T} \int_\R \int_{-T}^T
e^{-\i t (y - x)} \dif t \bP(\dif y)\\
&=& \lim_{T \to \infty} \frac{1}{2 T} \int_\R \int_{-T}^T
\cos(t(y-x)) + \underbrace{\i \sin(t (y-x))}_{\text{odd}}
\dif t \bP(\dif y)\\
&=& \lim_{T \to \infty} \frac{1}{2T}\int_{\R}
\int_{-T}^T \cos(t(y - x)) \dif t \bP(\dif y)\\
&=& \lim_{T \to \infty} \frac{1}{2T}\int_{\R}
2T \sinc(T(y-x))
\footnote{$\sinc(x) = \begin{cases}
\frac{\sin(x)}{x} &\text{if } x \neq 0,\\
1 &\text{otherwise.}
\end{cases}$} \bP(\dif y)\\
&\overset{\text{DCT}}{=}& \int_{\R}\lim_{T \to \infty}
\sinc(T(y-x)) \bP(\dif y)\\
&=& \bP(\{x\}).
2023-05-10 18:56:36 +02:00
\end{IEEEeqnarray*}
\end{refproof}
\begin{theorem} % Theorem 5
\label{thm:lec_10thm5}
Let $\phi$ be the characteristic function of $\bP \in M_1(\lambda)$.
Then
\begin{enumerate}[(a)]
2023-07-20 12:58:53 +02:00
\item $\phi(0) = 1$, $|\phi(t)| \le 1$, $\phi(-t) = \overline{\phi(t)}$
and $\phi(\cdot )$ is continuous.
2023-05-10 18:56:36 +02:00
\item $\phi$ is a \vocab{positive definite function},
i.e.~
\[\forall t_1,\ldots, t_n \in \R, (c_1,\ldots,c_n) \in \C^n ~ \sum_{j,k = 1}^n c_j \overline{c_k} \phi(t_j - t_k) \ge 0
\]
2023-07-13 00:06:30 +02:00
Equivalently, the matrix $(\phi(t_j- t_k))_{j,k}$ is positive definite.
2023-05-10 18:56:36 +02:00
\end{enumerate}
\end{theorem}
\begin{refproof}{thm:lec_10thm5}
Part (a) is obvious.
For part (b) we have:
\begin{IEEEeqnarray*}{rCl}
2023-07-12 15:25:25 +02:00
\sum_{j,k} c_j \overline{c_k} \phi(t_j - t_k) &=& \sum_{j,k} c_j \overline{c_k} \int_\R e^{\i (t_j - t_k) x} \bP(\dif x)\\
&=& \int_{\R} \sum_{j,k} c_j \overline{c_k} e^{\i t_j x} \overline{e^{\i t_k x}} \bP(\dif x)\\
&=& \int_{\R}\sum_{j,k} c_j e^{\i t_j x} \overline{c_k e^{\i t_k x}} \bP(\dif x)\\
2023-05-10 18:56:36 +02:00
&=& \int_{\R} \left| \sum_{l} c_l e^{\i t_l x}\right|^2 \ge 0
\end{IEEEeqnarray*}
\end{refproof}
2023-07-28 03:45:37 +02:00
\begin{theorem}[Bochner's theorem]
\yalabel{Bochner's Theorem for Positive Definite Functions}{Bochner's Theorem}{thm:bochner}%
The converse to \yaref{thm:lec_10thm5} holds, i.e.~any
$\phi: \R \to \C$ satisfying (a) and (b) of \yaref{thm:lec_10thm5}
2023-07-06 00:36:26 +02:00
must be the Fourier transform of a probability measure $\bP$
2023-05-10 18:56:36 +02:00
on $(\R, \cB(\R))$.
\end{theorem}
2023-07-28 03:45:37 +02:00
Unfortunately, we won't prove \yaref{thm:bochner} in this lecture.
2023-05-10 18:56:36 +02:00
\begin{definition}[Convergence in distribution / weak convergence]
2023-07-07 17:42:38 +02:00
\label{def:weakconvergence}
2023-07-13 00:38:46 +02:00
We say that $\bP_n \in M_1(\R)$ \vocab[Convergence!weak]{converges weakly} towards $\bP \in M_1(\R)$ (notation: $\bP_n \implies \bP$), iff
2023-05-10 18:56:36 +02:00
\[
2023-07-11 23:38:47 +02:00
\forall f \in C_b(\R)~ \int f \dif\bP_n \to \int f \dif\bP.
2023-05-10 18:56:36 +02:00
\]
Where
\[
C_b(\R) \coloneqq \{ f: \R \to \R \text{ continuous and bounded}\}
\]
In analysis, this is also known as $\text{weak}^\ast$ convergence.
\end{definition}
\begin{remark}
2023-07-11 23:38:47 +02:00
This notion of convergence makes $M_1(\R)$ a separable metric space.
We can construct a metric on $M_1(\R)$ that turns $M_1(\R)$ into a complete
2023-05-10 18:56:36 +02:00
and separable metric space:
Consider the sets
\[
2023-07-11 23:38:47 +02:00
\{\bP \in M_1(\R): \forall i=1,\ldots,n ~ \int f \dif\bP - \int f_i \dif\bP < \epsilon \}
2023-05-10 18:56:36 +02:00
\]
for any $f,f_1,\ldots, f_n \in C_b(\R)$.
These sets form a basis for the topology on $M_1(\R)$.
More of this will follow later.
\end{remark}
\begin{example}
\begin{itemize}
\item Let $\bP_n = \delta_{\frac{1}{n}}$.
2023-07-11 23:38:47 +02:00
Then $\int f \dif\bP_n = f(\frac{1}{n}) \to f(0) = \int f d \delta_0$
2023-05-10 18:56:36 +02:00
for any continuous, bounded function $f$.
Hence $\bP_n \to \delta_0$.
\item $\bP_n \coloneqq \delta_n$ does not converge weakly,
as for example
\[
2023-07-11 23:38:47 +02:00
\int \cos(\pi x) \dif\bP_n(x)
2023-05-10 18:56:36 +02:00
\]
does not converge.
\item $\bP_n \coloneqq \frac{1}{n} \delta_n + (1- \frac{1}{n}) \delta_0$.
Let $f \in C_b(\R)$ arbitrary.
Then
\[
2023-07-11 23:38:47 +02:00
\int f \dif\bP_n = \frac{1}{n}(n) + (1 - \frac{1}{n}) f(0) \to f(0)
2023-05-10 18:56:36 +02:00
\]
since $f$ is bounded.
Hence $\bP_n \implies \delta_0$.
\item $\bP_n \coloneqq \frac{1}{\sqrt{2 \pi n}} e^{-\frac{x^2}{2n}}$.
2023-07-13 00:06:30 +02:00
This ``converges'' towards the $0$-measure,
which is not a probability measure.
Hence $\bP_n$ does not converge weakly.
2023-05-10 18:56:36 +02:00
(Exercise) % TODO
\end{itemize}
\end{example}
\begin{definition}
We say that a series of random variables $X_n$
\vocab[Convergence!in distribution]{converges in distribution}
2023-07-15 02:00:04 +02:00
to $X$ (notation: $X_n \xrightarrow{\text{d}} X$), iff
2023-05-10 18:56:36 +02:00
$\bP_n \implies \bP$, where $\bP_n$ is the distribution of $X_n$
and $\bP$ is the distribution of $X$.
\end{definition}
2023-07-13 00:38:46 +02:00
It is easy to see, that this is equivalent to $\bE[f(X_n)] \to \bE[f(X)]$
for all $f \in C_b(\R)$.
2023-05-10 18:56:36 +02:00
\begin{example}
Let $X_n \coloneqq \frac{1}{n}$
and $F_n$ the distribution function, i.e.~$F_n = \One_{[\frac{1}{n},\infty)}$.
Then $\bP_n = \delta_{\frac{1}{n}} \implies \delta_0$
which is the distribution of $X \equiv 0$.
But $F_n(0) \centernot\to F(0)$.
\end{example}
2023-05-23 17:10:43 +02:00
\begin{theorem} % Theorem 1
\label{lec10_thm1}
2023-07-15 02:00:04 +02:00
$X_n \xrightarrow{\text{d}} X$ iff
2023-05-10 18:56:36 +02:00
$F_n(t) \to F(t)$ for all continuity points $t$ of $F$.
\end{theorem}
2023-07-14 22:07:36 +02:00
% \begin{proof}\footnote{This proof was not done in the lecture,
% but can be found in the official notes from lecture 13}
% ``$\implies$''
% Suppose $\mu_n \implies \mu$.
% Let $F_n$ and $F$ denote the respective density functions.
% Fix a continuity point $x_0 \in \R$ of $F$.
% We'll show
% \[
% \limsup_{n \to \infty} F_n(x_0) \le F(x_0) + \epsilon
% \]
% and
% \[
% \liminf_{ \to \infty} F_n(x_0) \ge F(x_0) - \epsilon
% \]
% for all $\epsilon > 0$.
% Fix some $\epsilon > 0$.
% Choose $\delta > 0$ such that $F(x_0 + \delta) < F(x_0) + \epsilon$
% and define
% \[
% g(x) \coloneqq \begin{cases}
% 1 &\text{if } x \le x_0,\\
% 1 - \frac{1}{\delta}(x - x_0)&
% \text{if } x \in (x_0, x_0 + \delta],\\
% 0 &\text{if } x \ge x_0 + \delta.
% \end{cases}
% \]
% Since $g$ is continuous and bounded, we have
% \[
% \int g \dif \mu_n \to \int g \dif \mu.
% \]
% It is clear that $\One_{(-\infty, x_0]} \le g$.
% Hence
% \[
% F_n(x_0) = \int \One_{(-\infty, x_0]} \dif \mu_n \le \int g \dif \mu_n.
% \]
% It follows that
% \begin{IEEEeqnarray*}{rCl}
% \limsup_{n} F_n(x_0)
% &\le& \limsup_n \int g \dif \mu_n\\
% &=& \lim_n \int g \dif \mu_n\\
% &=& \int g \dif \mu\\
% &\overset{g \le \One_{(-\infty, x + \delta]}}{=}& F(x + \delta)\\
% &=& F(x) + \epsilon.
% \end{IEEEeqnarray*}
% The assertion about $\liminf_{n \to \infty} F_n(x_0)$
% follows by a similar argument.
%
% ``$\impliedby$''
% Assume that $F_n(x) \to F(x)$ at all continuity points of $F$.
% We need to show
% \[
% \fgrall g \in C_b(\R) .~\int g \dif \mu_n \to \int g \dif \mu.
% \]
% Let $C$ denote the set of continuity points of $f$.
% We apply measure theoretic induction:
% \begin{itemize}
% \item For $g = \One_{(a,b]}$, $a< b \in C$,
% we have
% \[\int g \dif \mu_n = F_n(b) - F_n(a) \to F(b) - F(a) = \int g \dif \mu.\]
% \item For $g = \sum_{i} \alpha_i \One_{(a_i, b_i]}$,
% $a_i < b_i \in C$,
% we get $\int g \dif \mu_n \to \int g \dif \mu$
% by the same argument.
% \item % TODO continue from Lec13 page 21 (iii)
% \end{itemize}
%
% \end{proof}
2023-07-28 03:45:37 +02:00
\begin{theorem}[Levy's continuity theorem]
\yalabel{Levy's Continuity Theorem}{Levy}{levycontinuity}
2023-05-23 17:10:43 +02:00
% Theorem 2
2023-07-15 02:00:04 +02:00
$X_n \xrightarrow{\text{d}} X$ iff
2023-05-10 18:56:36 +02:00
$\phi_{X_n}(t) \to \phi(t)$ for all $t \in \R$.
\end{theorem}
We will assume these two theorems for now and derive the central limit theorem.
The theorems will be proved later.