Without using the continuous mapping theorem, I want to show that, given $\{X_n\}$ is a sequence of random variables converging in probability to $1$, $\{1/X_n\}$ converges in probability to $1$.

The place where I am stuck is: how do I know that probabilistically, $1/X_n$ converges? Because after I know that if $1/X_n\overset{p}\to a$ where $a$ is a real number, then I can use the fact that if $X_n$ and $Y_n$ converge in probability to $x$ and $y$ respectively, then $X_nY_n\overset{p}\to xy$. Then taking $Y_n=1/X_n$ the result will be immediate.

But how do I know that the new sequence $1/X_n$ converges at all? I would love to use results in real analysis but I am sure they cannot be applied here.

  • $\begingroup$ suppose $P(X_n=1)=p_n$ and $P(X_n=0)=1-p_n$ for $p_n=1-\frac{1}{n}$. Then $X_n$ converges in probability to $1$, but $\frac{1}{X_n}$ is not properly defined, hence cannot converge in probability to $1$. $\endgroup$ – drhab Jan 24 '15 at 13:51
  • $\begingroup$ The question is from Rohatgi. I am not sure a book as famous as his will make this grievous mistake. Anyhow, suppose an additional information that $1/X_n$ is bounded is given. This is an assumption I am making up. Can we try this then? $\endgroup$ – Landon Carter Jan 24 '15 at 13:54
  • $\begingroup$ You received nice answers. When it comes to my comment: see the remark in the answer of @saz. $\endgroup$ – drhab Jan 24 '15 at 14:14

We have to show that $$\mathbb{P} \left( \left| \frac{1}{X_n} - 1 \right|> \varepsilon \right) \to 0 \qquad \text{as} \, \, n \to \infty$$ for any $\varepsilon>0$. To this end, we note that

$$\left| \frac{1}{X_n} - 1 \right|> \varepsilon \iff |X_n-1|> \varepsilon \cdot |X_n|.$$

This implies

$$\begin{align*} \left\{ \left| \frac{1}{X_n} - 1 \right|> \varepsilon \right\} & \subseteq \left\{|X_n-1|> \varepsilon \cdot |X_n|; |X_n| \geq \frac{1}{2} \right\} \cup \left\{|X_n-1|> \varepsilon \cdot |X_n|; |X_n|< \frac{1}{2} \right\} \\ &\subseteq \left\{|X_n-1|> \frac{\varepsilon}{2} \right\} \cup \left\{|X_n| < \frac{1}{2} \right\}. \end{align*}$$


$$\mathbb{P} \left( \left| \frac{1}{X_n} - 1 \right|> \varepsilon \right) \leq \mathbb{P} \left( |X_n-1|> \frac{\varepsilon}{2} \right) + \mathbb{P}\left(|X_n|< \frac{1}{2} \right).$$

The first term on the right-hand side converges to $0$ as $n \to \infty$ since $X_n \to 1$ in probability. For the second one, we note that

$$\mathbb{P} \left( |X_n| < \frac{1}{2} \right) \leq \mathbb{P} \left( |X_n-1|> \frac{1}{2} \right) \to 0$$

as $n \to \infty$. This finishes the proof.

Remark: Since $X_n(\omega)$ might equal $0$, we have to use the convention $\frac{1}{0} = \infty$. This means that the random variable $\frac{1}{X_n}$ takes values in $\mathbb{R} \cup \{\infty\}$.


You are given $\displaystyle \lim_{n \to \infty} P(\{|X_n - 1| \ge \epsilon\}) = 0$ for every $\epsilon > 0$. If $0 < \epsilon < 1$ then $$\left| \frac{1}{X_n} - 1 \right| \ge \epsilon \iff -\epsilon \le \frac{1}{X_n} - 1 \le \epsilon \iff \frac{1}{1 + \epsilon} \le X_n \le \frac{1}{1 - \epsilon}$$ and $$ \frac{1}{1 + \epsilon} \le X_n \le \frac{1}{1 - \epsilon} \iff -\frac{\epsilon}{1 + \epsilon} \le X_n - 1 \le \frac{\epsilon}{1 - \epsilon}.$$ But $\displaystyle \frac{\epsilon}{1 + \epsilon} < \frac{\epsilon}{1 - \epsilon}$, so that $$-\frac{\epsilon}{1 + \epsilon} \le X_n - 1 \le \frac{\epsilon}{1 - \epsilon} \implies \left| {X_n} - 1\right| \le \frac{\epsilon}{1 - \epsilon}.$$ Thus $$P\left( \left\{ \left| \frac{1}{X_n} - 1 \right| \ge \epsilon \right\} \right) \le P \left( \left\{ |X_n - 1| \ge \frac{\epsilon}{1 - \epsilon} \right\} \right) \to 0.$$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.