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

I didn't find a demonstration on the web of how to obtain the exponential probability distribution without using a Poisson distribution, and I don't manage to do it myself.

I am trying to understand how to get this pdf with the only assumption that for each interval dt, the probability of an event is constant. I guess the demo is pretty immediate, but I am missing it.

share|cite|improve this question
What do you mean with "obtain distribution on the web"? – Ilya Jul 27 '11 at 16:15
He has not been able to find an answer online. – Emre Jul 27 '11 at 16:19
I edited to make it clearer. – WhitAngl Jul 27 '11 at 16:27
up vote 2 down vote accepted

The question is not clear, but maybe you'll find the following useful.

For $t > 0$ fixed, let $E_k$, for $k=1,\ldots,n$, denote the event of $0$ events in the time interval $I_k = [\frac{{(k - 1)t}}{n},\frac{{kt}}{n}]$. Since $I_k$ has length $t/n$, $$ {\rm P}(E_k^c) = \lambda \frac{{ t}}{n} + o\bigg(\frac{1}{n}\bigg) \;\; {\rm as} \;\; n \to \infty. $$ Hence $$ {\rm P}(E_k) = 1 - \lambda \frac{{ t}}{n} + o\bigg(\frac{1}{n}\bigg) \;\; {\rm as} \;\; n \to \infty. $$ Now, by independence, the probability $p_t$ of $0$ events in the time interval $[0,t]$ is given by $$ p_t = {\rm P}(E_1 ) \cdots {\rm P}(E_n ) = \bigg(1 - \frac{{\lambda t}}{n} + o\bigg(\frac{1}{n}\bigg)\bigg)^n . $$ Letting $n \to \infty$, this gives $$ p_t = e^{ - \lambda t} . $$ This probability corresponds to ${\rm P}(X > t)$, where $X$ is exponential with parameter $\lambda$.

EDIT: Note that $n o(1/n) \to 0$ as $n \to \infty$. Using this, it indeed follows that $$ \bigg(1 - \frac{{\lambda t}}{n} + o\bigg(\frac{1}{n}\bigg)\bigg)^n \to e^{ - \lambda t} . $$

share|cite|improve this answer
For the standard approach, see… – Shai Covo Jul 27 '11 at 17:09
Thanks, that's exactly the kind of demonstration I was hoping for :) – WhitAngl Jul 27 '11 at 17:13
Note that the $O(1/n^2)$'s of the original answer have been replaced by $o(1/n)$'s. – Shai Covo Jul 27 '11 at 18:58

I still not sure that understood correctly your condition with $dt$, but there is a usual motivation. We want $\tau\in\mathcal E$ not to have memory, i.e. $$ P(\tau\geq t+s|\tau \geq t) = P(\tau\geq s) $$ for all $s,t\geq 0$. By the formula of the conditional probability this leads to the fact $$ P(\tau\geq t+s) = P(\tau\geq s)P(\tau\geq t) $$ which clearly implies that the density is $\lambda \mathrm e^{-\lambda t}$.

I can make a guess that you also mean the property $$ \frac{P(\tau \in dt|\tau\geq t)}{dt} = c. $$ Again, using the formula $\displaystyle{P(A|B) = \frac{P(A\cap B)}{P(B)}}$ we obtain $$ \frac{P(\tau \in dt)}{dt} = cP(\tau\geq t), $$ i.e. $$ f(t) = c(1-F(t)) $$ where $f(t)$ is a density function and $F(t)$ is a distribution function. You only need now to recall that $F' = f$ and solve an ODE.

share|cite|improve this answer

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.