I'm a bit lost here, the question goes as follows:

Suppose that $X_k$ are i.i.d. and follow an exponential distribution with parameter $\lambda$.

Define $F_n(x) := \frac{1}{n}\sum_{i=1}^n(X_{k}\leq x) $ for x $\geq0$

Question :Show that $F_n(x)$ converges in probability to $ 1−e ^{−\lambda x} $ Does it also converge in $L^{1}$ norm?

I think to show that $F_n(x)$ converges in probability we need to show that:


Therefore we can state that:

$F_n(x)$$\overset{p}{\to} 1−e ^{−\lambda x} $


$$\lim_{n\to\infty}P \left(\left| \frac{1}{n}\sum_{i=1}^n\mathbb{1}(X_{k}\leq x)- \left(1-e^{-\lambda x}\right)\right|>\varepsilon\right)=0.$$

But here is where I get stuck in terms of figuring out the correct way. Could I use the CLT?
I was under the impression that the CLT is for convergence in distribution whereas convergence in probability is stronger (except for the case where it converges to a constant which is not the case here.)

Alternatively I thought about using The weak law of large numbers and Chebychev,but then I still need to prove $L^p$

I already know that convergence in probability will not mean convergence in $L^p$ but again how would I show this?

Kind regards

  • 3
    $\begingroup$ Did you mean to take an average? As stated the sum obviously diverges (I'm assuming here you're summing indicator random variables). $\endgroup$ – dsaxton Oct 3 '16 at 21:20
  • $\begingroup$ As other comments indicate, you need to fix your definition of $F_n(x)$ as an average. Then, you need to apply what you know about the law of large numbers. Will you apply the "weak law" or the "strong law"? $\endgroup$ – Michael Oct 4 '16 at 0:17
  • $\begingroup$ I find it unusual that the first comment gets 1 upvote while the second gets three. $\endgroup$ – Michael Oct 4 '16 at 0:18
  • $\begingroup$ Yes , there was a mistake in the questions as the 1/n was omitted , I have changed my question to reflect this @SangchulLee $\endgroup$ – jerreyz Oct 4 '16 at 21:03
  • 1
    $\begingroup$ You might like to know that a much stronger result is true. You are proving convergence pointwise in $x$, in probability. The Glivenko-Cantelli theorem says that you get convergence uniformly in $x$, almost surely. $\endgroup$ – Nate Eldredge Oct 6 '16 at 17:00

As stated above to state convergence in probability we can simply apply the law of large numbers using Chebychev's inequality:

$P(| \frac{1}{n}\sum_{i=1}^n\mathbb{1}(X_{k}\leq x)- (1-e^{-\lambda x})|>\varepsilon)≤\frac{\sigma^2 (F_n(x))}{\varepsilon^2}$

where the second term reduces to: $\frac{1}{n\lambda^2\varepsilon^2} \rightarrow0 $ for $\lim_{n\to\infty}$


If you know the weak law of large numbers, then it is immediate.

Otherwise, we can argue directly. We can prove that $$ \lim_{n \to + \infty}\mathbb E \left[ \left(F_n(x)-F(x)\right)^2\right]=0,$$ where $F(x)=1-e^{ \lambda x}$. Indeed, note that $$\left(F_n(x)-F(x)\right)^2 = \frac 1 {n^2 } \sum_{j=1}^n \left(\mathbf 1 \left\{X_j \leqslant x \right\} -F(x)\right)^2+ \frac 2{n^2 } \sum_{1 \leqslant i \lt j \leqslant n }\left(\mathbf 1 \left\{X_i \leqslant x \right\}-F(x) \right)\left(\mathbf 1 \left\{X_j \leqslant x \right\} -F(x)\right).$$ The expectation of the second sum term is $0$ because the random variables $\left(\mathbf 1 \left\{X_i \leqslant x \right\}-F(x) \right)$ and $\left(\mathbf 1 \left\{X_j \leqslant x \right\}-F(x) \right)$ are centered and independent. Since $X_j$ has the same distribution as $X_1$, $\left(\mathbf 1 \left\{X_j \leqslant x \right\}-F(x) \right)^2$ has the same distribution as $\left(\mathbf 1 \left\{X_1\leqslant x \right\}-F(x) \right)^2$. In particular, these random variables share the same expectation. We thus have $$ \mathbb E \left[\left(F_n(x)-F(x)\right)^2 \right] = \frac 1 {n } \mathbb E\left[\left(\mathbf 1 \left\{X_1 \leqslant x \right\} -F(x)\right)^2\right],$$ which shows that $F_n(x)\to F(x)$ in $ \mathbb L^2$ for any $x \in\mathbb R$.


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.