Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Assume that $X_1,\cdots,X_n$ are independent r.v., not necessarily iid,

Let $S_n=X_1+\cdots+X_n$, suppose that $S_n$ converges in probability, how can we show that $S_n$ converges a.s.?

share|cite|improve this question
I am not sure whether the following would help. If $A$ is the set of $\omega$ for which $S_n(\omega)$ converges, by Kolmogorov 0-1 law, $P(A)=0$ or $1$. – Ashok Dec 10 '11 at 7:49
See Theorem 5.3.4 in Kai Lai Chung, A course in probability theory, 3rd ed., Academic Press, 2001. – Did Oct 31 '12 at 20:53
up vote 5 down vote accepted

We can use Ottaviani's inequality: if we put $M_k:=\max_{1\leqslant i\leqslant k}|S_i|$ and $S_{k,n}:=\sum_{i=k+1}^nX_i$, then for all $\varepsilon >0$ we have $$\min_{1\leqslant k\leqslant n}P(|S_{k,n}|\leqslant\varepsilon)P(|M_n|>2\varepsilon)\leqslant P(|S_n|>\varepsilon).$$

Put $A_m:=\sup_{k\in\mathbb N^*}|S_{m+k}-S_m|$ and $A:=\inf_{m\in\mathbb N^*}A_m$. We have $$\{\{S_n\}\mbox{ doesn't converge}\}=\{A\neq 0\}\subset\bigcup_{\varepsilon\in\mathbb Q^+}\bigcap_{m\in\mathbb N^*}\{A_m>\varepsilon\}$$ and $$\{A_m>\varepsilon\}=\bigcup_{r\in\mathbb N^*}\left\{\sup_{1\leqslant k\leqslant r}|S_{m+k}-S_m|>\varepsilon\right\}.$$ Ottaviani's inequality gives $$\min_{1\leqslant k\leqslant r}P(|S_{m+r}-S_{m+k}|\leqslant\varepsilon)P\left(\max_{1\leqslant k\leqslant r}|S_{m+k}-S_m|>2\varepsilon\right)\leqslant P(|S_{r+m}-S_m|>\varepsilon).$$ We fix $\delta>0$; we can choose $N=N(\varepsilon,\delta)$ such that for $0\leqslant k\leqslant r$ and $m\geqslant N$, $P(|S_{m+r}-S_{m+k}|>\varepsilon)\leqslant\delta$. We get $$P\left(\max_{1\leqslant k\leqslant r}|S_{m+k}-S_{m+r}|>\varepsilon\right)\leqslant\frac{\delta}{1-\delta},$$ therefore $P(A_m>\varepsilon)\leqslant\frac{\delta}{1-\delta}$. We have $$0\leqslant P\left(\bigcap_{p\in\mathbb N^*}(A_p>\varepsilon)\right)\leqslant P(A_m>\varepsilon)\leqslant\frac{\delta}{1-\delta},$$ so for all $\varepsilon\in\mathbb Q^+$, we have $P\left(\bigcap_{p\in\mathbb N^*}(A_p>\varepsilon)\right)=0$, hence $P(A\neq 0)=0$ and we have the almost sure convergence.

share|cite|improve this answer
great.....I tried without Ottavani.....Ottavani is necessary else we get a bound of N$\delta$ and cant control the probabilities – user24367 Feb 17 '13 at 0:45

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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