I have two (or more) independent exponential variables $ X_1 \sim \exp(\lambda_1) $ and $ X_2 \sim \exp(\lambda_2) $. I want to get both the value of $ \min(X_1, X_2) $ and $ \arg\min(X_1, X_2) $. Can I just draw them independently? Is such approach correct?

I.e. it is well-known that $ \min(X_1, X_2) \sim \exp(\lambda_1 + \lambda_2) $ and $ P(X_1 < X_2) = \frac{\lambda_1}{\lambda_1 + \lambda_2} $, but what about the distribution of $ \min(X_1, X_2) $ if we know that, for example, $ X_1 < X_2 $?

  • $\begingroup$ Your assertion about the distribution of $\min\{X_1,X_2\}$ is right if $X_1,X_2$ are independent. ${}\qquad{}$ $\endgroup$ – Michael Hardy Jan 28 '15 at 1:34
  • $\begingroup$ It's not clear what you mean by $\arg\min$ in this case. $\endgroup$ – Michael Hardy Jan 28 '15 at 1:35
  • $\begingroup$ Yes, of course you're right, they should be independent. By $\arg\min$ I meant which one of the two variables $X_1, X_2$ will be the minimal of this two. $\endgroup$ – Andrew Jan 28 '15 at 8:52

I find I am slightly surprised by the answer I get. The event that $\min\{X_1,X_2\}>x$ and the event that $X_1<X_2$ are actually independent of each other. Consequently the conditional distribution of the minimum given that $X_1<X_2$ is the same as the unconditional distribution of the minimum. I'd have expected that if $X_1,X_2$ had the same distribution, but as it is I expected the answer to depend on the ratio of the two $\lambda$s. But it doesn't.

One of the results that you say you already know can be stated by saying $\Pr(\min\{X_1,X_1\}>x)$ $=e^{-(\lambda_1+\lambda_2)x}$, for all $x>0$. The other one says $\Pr(X_1<X_2)=\lambda_1/(\lambda_1+\lambda_2)$.

So now consider \begin{align} & \Pr(x<\min\{X_1,X_2\}\ \&\ X_1<X_2)=\Pr(x<\min=X_1<X_2) \\[8pt] = {} & \int_x^\infty\cdots\,dx_1 = \int_x^\infty\left(\int_{x_1}^\infty\cdots\,dx_2\right)\,dx_1 = \int_x^\infty\left(\int_{x_1}^\infty \lambda_1 e^{-\lambda_1 x_1} \lambda_2 e^{-\lambda_2 x_2} \,dx_2\right)\,dx_1 \\[8pt] = {} & \int_x^\infty\left( \lambda_1 e^{-\lambda_1 x_1} \int_{x_1}^\infty e^{-\lambda_2 x_2} (\lambda_2\,dx_2) \right) dx_1 = \int_x^\infty\left( \lambda_1 e^{-\lambda_1 x_1} e^{-\lambda_2 x_1} \right) dx_1 \\[8pt] = {} & \lambda_1 \int_x^\infty e^{-(\lambda_1+\lambda_2)x_1}\,dx_1 =\lambda_1 \frac{e^{-(\lambda_1+\lambda_2)x}}{\lambda_1+\lambda_2} \\[8pt] = {} & \frac{\lambda_1}{\lambda_1+\lambda_2} \cdot e^{-(\lambda_1+\lambda_2)x} \\[8pt] = {} & \Pr(X_1<X_2)\cdot \Pr(\min>x). \end{align}

  • $\begingroup$ Thank you very much! I've corrected your answer a bit, as $\Pr(\min\{X_1,X_2\}>x)$ $=1 - e^{-(\lambda_1+\lambda_2)x}$, not just $ e^{-(\lambda_1+\lambda_2)x} $ $\endgroup$ – Andrew Jan 28 '15 at 8:54
  • $\begingroup$ You're mistaken. $1-e^{-(\lambda_1+\lambda_2)x}$ is the probability that the minimum is LESS than $x$, not that it is greater than $x$. ${}\qquad{}$ $\endgroup$ – Michael Hardy Jan 28 '15 at 15:51

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.