guide:781736e534

[math] \newcommand{\R}{\mathbb{R}} \newcommand{\A}{\mathcal{A}} \newcommand{\B}{\mathcal{B}} \newcommand{\N}{\mathbb{N}} \newcommand{\C}{\mathbb{C}} \newcommand{\Rbar}{\overline{\mathbb{R}}} \newcommand{\Bbar}{\overline{\mathcal{B}}} \newcommand{\Q}{\mathbb{Q}} \newcommand{\E}{\mathbb{E}} \newcommand{\p}{\mathbb{P}} \newcommand{\one}{\mathds{1}} \newcommand{\0}{\mathcal{O}} \newcommand{\mat}{\textnormal{Mat}} \newcommand{\sign}{\textnormal{sign}} \newcommand{\CP}{\mathcal{P}} \newcommand{\CT}{\mathcal{T}} \newcommand{\CY}{\mathcal{Y}} \newcommand{\F}{\mathcal{F}} \newcommand{\mathds}{\mathbb}[/math]

Proposition

Let [math](\Omega,\A,\p)[/math] be a probability space. Let [math](X_n)_{n\geq 1}[/math] be a sequence of r.v.'s and assume that for all [math]\epsilon \gt 0[/math] we have

[[math]] \sum_{n\geq 1}\p[\vert X_n-X\vert \gt \epsilon] \lt \infty. [[/math]]

Then

[[math]] \lim_{n\to\infty\atop a.s.}X_n=X. [[/math]]

Show Proof

Take [math]\epsilon_k=\frac{1}{k}[/math] for [math]k\in\N[/math] with [math]k\geq 1[/math]. Now with the Borel-Cantelli lemma we get

[[math]] \p\left[\limsup_n\left\{\vert X_n-X\vert \gt \frac{1}{k}\right\}\right]=0, [[/math]]

which implies that [math]\p\left[\bigcup_{k\geq 1}\limsup_n\left\{\vert X_n-X\vert \gt \frac{1}{k}\right\}\right]=0[/math] and hence

[[math]] \p\left[\underbrace{\bigcap_{k\geq 1}\liminf_n\left\{\vert X_n-X\vert\leq \frac{1}{k}\right\}}_{\Omega'}\right]=1. [[/math]]

Moreover, we have that [math]\p[\Omega']=1[/math] and for [math]\omega\in\Omega'[/math] we get that for all [math]k\geq 1[/math] there is [math]n_0(\omega)\in\N\setminus\{0\}[/math] such that for [math]n\geq n_0(\omega)[/math] we get that [math]\vert X_n(\omega)-X(\omega)\vert\leq \frac{1}{k}[/math], i.e. [math]\lim_{n\to\infty}X_n(\omega)=X(\omega)[/math] for [math]\omega\in\Omega'[/math].

■

Example

Let [math](\Omega,\A,\p)[/math] be a probability space. Let [math](X_n)_{n\geq 1}[/math] be a sequence of r.v.'s such that [math]\p[X_n=0]=1-\frac{1}{1+n^2}[/math] and [math]\p[X_n=1]=\frac{1}{1+n^2}[/math]. Then for all [math]\epsilon \gt 0[/math] we get [math]\p[\vert X_n\vert \gt \epsilon]=\p[X_n \gt \epsilon]=\frac{1}{1+n^2}[/math], so it follows

[[math]] \sum_{n\geq 1}\p[\vert X_n\vert \gt \epsilon] \lt \infty, [[/math]]

which implies that [math]\lim_{n\to\infty\atop a.s.}X_n=0.[/math]

Proposition

Let [math](\Omega,\A,\p)[/math] be a probability space. Let [math](X_n)_{n\geq 1}[/math] be a sequence of r.v.'s. Then

[[math]] \lim_{n\to\infty\atop a.s.}X_n=X\Longleftrightarrow \lim_{n\to\infty\atop \p}\sup_{m \gt n}\vert X_m-X\vert=0. [[/math]]

Show Proof

Exercise.

■

Example

Let [math](Y_n)_{n\geq 1}[/math] be iid r.v.'s such that [math]\p[Y_n\leq X]=1-\frac{1}{1+X}[/math] for [math]X\geq 0[/math] and [math]n\geq 1[/math]. Take [math]X_n=\frac{Y_n}{n}[/math] and let [math]\epsilon \gt 0[/math]. Then

[[math]] \p[\vert X_n\vert \gt \epsilon]=\p[\vert Y_n\vert \gt n\epsilon]=\frac{1}{1+n\epsilon}\xrightarrow{n\to\infty}0, [[/math]]

and thus [math]\lim_{n\to\infty\atop \p}X_n=0[/math]. Moreover, we have

[[math]] \p\left[\sup_{m\geq n}\vert X_m\vert \gt \epsilon\right]=1-\p\left[\sup_{m\geq n}\vert X_n\vert\leq \epsilon\right]=1-\prod_{m\geq n}^\infty\left(1-\frac{1}{1+m\epsilon}\right), [[/math]]

but [math]\prod_{m\geq n}^\infty\left(1-\frac{1}{1+m\epsilon}\right)=0[/math]. Hence [math]\p[\sup_{m\geq n}\vert X_n\vert \gt \epsilon]\not\rightarrow 0[/math] as [math]n\to\infty[/math] and therefore [math](X_n)_{n\geq 1}[/math] doesn't converge a.s. to [math]X[/math].

Lemma

Let [math](\Omega,\A,\p)[/math] be a probability space. Let [math](X_n)_{n\geq 1}[/math] be a sequence of r.v.'s. Then [math]\lim_{n\to\infty\atop \p}X_n=X[/math] if and only if for very subsequence of [math](X_n)_{n\geq 1}[/math], there exists a further subsequence which converges a.s.

Show Proof

If [math]\lim_{n\to\infty\atop\p}X_n=X[/math], then any of its subsequences also converge in probability. We already know that there exists a subsequence which converges a.s. Conversely, if [math]\lim_{n\to\infty\atop\p}X_n=X[/math], then there is an [math]\epsilon \gt 0[/math], some [math]n_k\in\N[/math] and a [math]\nu \gt 0[/math] such that for all [math]k\geq 1[/math] we get

[[math]] \p[\vert X_{n_k}-X\vert \gt \epsilon] \gt \nu [[/math]]

and therefore we cannot extract a subsequence from [math](X_{n_k})_{k\geq 1}[/math] which would converge a.s.

■

Proposition

Let [math](\Omega,\A,\p)[/math] be a probability space. Let [math](X_n)_{n\geq 1}[/math] be a sequence of r.v.'s and [math]g:\R\to\R[/math] a continuous map. Moreover, assume that [math]\lim_{n\to\infty\atop\p}X_n=X[/math]. Then

[[math]] \lim_{n\to\infty\atop \p}g(X_n)=g(X). [[/math]]

Show Proof

Any subsequence [math]g((X_{n_k})_{k\geq 1})[/math] and [math](X_{n_k})_{k\geq 1}[/math] converges in probability. So it follows that there exists a subsequence [math](X_{m_k})_{k\geq 1}[/math] of [math](X_{n_k})_{k\geq 1}[/math] such that

[[math]] \lim_{n\to\infty\atop a.s.} X_n=X\text{and}\lim_{k\to\infty\atop a.s.}g(X_{m_k})=g(X) [[/math]]

because [math]g[/math] is continuous. Now with the previous lemma we get that

[[math]] \lim_{n\to\infty\atop \p}g(X_n)=g(X). [[/math]]

■

Proposition

Let [math](\Omega,\A,\p)[/math] be a probability space. Let [math](X_n)_{n\geq 1}[/math] and [math](Y_n)_{n\geq 1}[/math] be sequences of r.v.'s such that [math]\lim_{n\to\infty\atop \p}X_n=X[/math] and [math]\lim_{n\to\infty\atop\p} Y_n=Y[/math]. Then

• [math]\lim_{n\to\infty\atop\p} X_n+Y_n=X+Y[/math]
• [math]\lim_{n\to\infty\atop \p}X_n\cdot Y_n=X\cdot Y[/math]

Show Proof

We need to show both points.

• Let [math]\epsilon \gt 0[/math]. Then [math]\vert X_n-X\vert\leq \frac{\epsilon}{2}[/math] and [math]\vert Y_n-Y\vert\leq \frac{\epsilon}{2}[/math] implies that [math]\vert (X_n+Y_n)-(X+Y)\vert\leq \epsilon[/math], and thus we get
  [[math]] \p[\vert X_n+Y_n-(X+Y)\vert \gt \epsilon]\leq \p\left[\vert X_n-X\vert \gt \frac{\epsilon}{2}\right]+\p\left[\vert Y_n-Y\vert \gt \frac{\epsilon}{2}\right]. [[/math]]
• We apply proposition 8.4 to the continuous map [math]g(X)=X^2[/math]. Hence we get
  [[math]] 2X_nY_n=(X_n+Y_n)^2-X_n^2-Y_n^2. [[/math]]

■

General references

Moshayedi, Nima (2020). "Lectures on Probability Theory". arXiv:2010.16280 [math.PR].