Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

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

Suppose that $X$ is a non-negative random variable and there exist constants $A,B$ such that $$\forall t > 0\colon P(X>\frac{1}{t})<Bt $$ and $$\forall t > 0\colon E(\sin(tX))<At$$

I want to show that: $$E(X)<\infty$$

share|cite|improve this question
I changed $\forall t\exists A,B$ to $\exists A,B\forall t$. Do you see why the question won't make sense otherwise? – Hagen von Eitzen Jan 15 '13 at 19:18
up vote 2 down vote accepted

Here is a corrected version of Mihai Nica's argument. This was too long for a comment.

In the following, $E(\xi ; F)$ is a notation for $E(\xi\,1_F)$.

For every $n \geq 1$, we have $$ E[\sin(X/n)\,;\, X \leq n] = E[\sin(X/n)] - E[\sin(X/n)\,;\,X > n] $$

Since $\sin(X/n) \geq -1 $, we get $$ E[\sin(X/n)\,;\, X \leq n] \leq E[\sin(X/n)] + P(X>n) \leq (A+B)/n $$

Using the fact that $n\sin(X/n)\,1_{\{X \leq n\}}$ converges almost surely to $X$ as $n \to \infty$ and stays non-negative, Pierre Fatou's lemma gives finally $$ E(X) \leq \liminf_{n\to\infty}\;E[n\sin(X/n)\,;\,X \leq n] \leq A+B. $$

Note : we can get $E(X) \leq A + \dfrac{B}{\pi}$ in the same way.

share|cite|improve this answer
thanks for your answer – yalda Jan 16 '13 at 6:09

Notice, by linearity of $E$, that: $$E\left(\frac{\sin(tX)}{t}\right)<A$$ holds for every $t$. So its true then that $\liminf_{t\to 0} E\left(\frac{\sin(tX)}{t}\right)<A$. If only we could interchange the lim inf from outside the $E$ to inside, then we would be finished and happy, as $\liminf_{t\to 0} \frac{\sin(tX)}{t} = X$

This can be done using Fatou's Lemma, applied in a clever way. Define $X_n = \left.\frac{\sin(tX)}{t}\right|_{t=\frac{1}{n}}=n\sin{\frac{X}{n}}$ so that $\liminf_{n\to\infty} X_n = X$. Applying Fatou's lemma gives: (Actually there is one little hickup preventing us from applying Fatoe at this point that will be corrected later. Can you spot it?)

$$E(X)\leq \liminf_{n\to\infty}E(X_n) \leq A$$

And we are finished! There is one small problem though: Fatou's lemma does not apply because it is not true that $X_n$ is non-negative! Indeed $X_n$ is the sin of something so it can take both positive and negative values. To get around this obstacle, let us fix $M \in \mathbb{N}$ and restrict our attention to the set $\left\{ X < M \right\}$. On this set, we can be sure that $X_n \geq 0$ for $n > M$ as $X_n = n\sin{\frac{X}{n}}$ and the argument to the $\sin$ is $\leq 1$ and $\sin$ takes positive values here. We can now apply Fatou's lemma to the sequence $X_M, X_{M+1}, \ldots$ to get: $$E(X | X < M)\leq \liminf_{n\to\infty,n\geq M}E(X_n | X<M) \leq A$$ Since this holds for every $M$, and since the sets $\left\{X<M\right\} \uparrow \Omega$ increase to the whole probability space (Indeed, the first given inequality tells us $P(X>M) \to 0$ as $M\to\infty$), we can conclude $E(X) < A$

EDIT: The inequality $E(X_n|X<M) < A$ is wrong. To get around this, instead of fixing $M$, we should look at applying Fatou's lemma to $X_n 1_{X<n}$. See Ju'x's post for details.

share|cite|improve this answer
How do you justify that $E(X_n \mid X < M) \leq A$? – Siméon Jan 15 '13 at 23:14
Woops! That is an oversite on my part! I think its not true that $E(X_n | X< M) < A$. However I think we can fix it up using the first inequality (the one involving $B$). Namely: $E(X_n | X < M ) = E(X_n 1_{X<M}) / P(X<M) < E(X_n) / P(X<M) < A \frac{1}{1-B/M}$ For M large enough, this is $<2A$ – Mihai Nica Jan 15 '13 at 23:37
I think that the main problem lies in the inequality $E(X_n\,1_{X < M}) \leq E(X_n)$. – Siméon Jan 15 '13 at 23:42
Yes you are right. To fix it we should look at $E(X_n 1_{X <n})$ and use $E(X_n 1_{X <n}) < A + B$ as in your post. Thank you for pointing out the error! – Mihai Nica Jan 16 '13 at 2:37
@Mihai thanks for your answer – yalda Jan 16 '13 at 6:10

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.