# Specific expectation of non-negative random variable

Let $$T$$ a random variable of non-negative integers. Suppose that there is $$n \in \mathbb{N}$$ and there is $$\alpha>0$$ such that for all $$k \in \mathbb{N}$$ we have $$\mathbb{P}(T >kn) \le (1-\alpha)^k.$$ Prove that $$\mathbb{E}[T] \le \frac{n}{\alpha}$$.

I tried to use the sum formula for expectation of nonnegative integer random variables $$\mathbb{E}[T] = \sum_{i \ge 0}{\mathbb{P}(T > i)},$$ but I can't get a '$$\le$$' innequalty.

Recall from real analysis that $$\sum_{k=1}^{\infty} (1-a)^k = \frac{1-a}{a}\quad \text{for}\quad 0 Consider the random variable $$n^{-1} T$$. We can split its expectation up as follows: $$E[n^{-1}T] = E[n^{-1} T\cdot \mathbf{1}_{\{T\ge n\}}] + E[n^{-1} T\cdot \mathbf{1}_{\{0\le T Observe that $$E[n^{-1} T\cdot \mathbf{1}_{\{0\le T< n\}}]\le 1\cdot P(0\le T On the other hand, $$S:=\lceil n^{-1} T\rceil\cdot \mathbf{1}_{\{T\ge n\}}$$ is a non-negative integer-valued random variable and $$n^{-1} T\cdot \mathbf{1}_{\{0\le T< n\}}\le S.$$ By its definition, note that the following hold:
• $$S>0$$ if and only if $$T\ge n$$,
• $$S>k$$ if and only if $$T>kn$$ for $$k=1,2,\dots$$.
Therefore, using the expectation identity for non-negative integer-valued random variables, \begin{align*} E[S]&= \sum_{k=0}^{\infty} P(S>k) = \sum_{k=0}^{\infty} P(S>k) = P(S>0) + \sum_{k=1}^{\infty} P(S>k)\\ &=P(T\ge n)+ \sum_{k=1}^{\infty} P(T>kn) \le P(T\ge n) + \sum_{k=1}^{\infty} (1-a)^k\\&= P(T\ge n) + \frac{1-a}{a}. \end{align*} Combining the above three inequalities, we have $$E[n^{-1}T] \le [1-P(T\ge n)] + P(T\ge n) + \frac{1-a}{a} = \frac{1}{a}.$$ Complete the proof by multiplying both sides of the above inequality by $$n$$.
• Thanks!!! I forget to say that $\alpha \in [0,1]$. But, I got the idea! Dec 20 '20 at 17:53
• No problem. We should add the $a=1$ case, although it is not hard. In this case, $P(T>n) = 0$, so $P(T\le n) = 1$. Therefore, $E[T] \le n P(T\le n) = n$, as desired. Dec 20 '20 at 17:57
• @GuilhermeCosta Actually, I believe the $a=1$ case is covered by my answer. (But certainly doesn't require that full argument, evidenced by the shorter solution in the comments.) For $a=0$, it implies that $P(T>x)=1$ for all $x>0$. Therefore, $E[T]\ge x\cdot P(T>x) = x$ for all $x>0$. This implies that $E[T]$ is unbounded, and we write $E[T]=\infty$. Although dividing by $0$ is undefined, if we formally set $\infty=1/0$, then the statement is true in the $a=0$ case as well. Dec 20 '20 at 20:36