I am interesting in determining if there is a closed form expression of the CDF and PDF of the maximum on $N$ Gumbel distributions that are independent but not identically distributed.

In particular, I am hoping to calculate $E(max(x_1,...,x_N))$

where $x_1,...,x_N$ are each random variables with $x_i \sim Gumbel(\mu_i,\beta_i)$.

When the variables are distributed iid gumbel, it turns out that the maximum also has a gumbel distribution and the expectation above has a closed form solution. I was wondering if any similar results exist for the inid (independent not identically distributed) case, but where each random variable follows a gumbel distribution?

  • $\begingroup$ Do you have any special structure on the parameters $\mu_i, \beta_i$? $\endgroup$ Sep 21, 2015 at 19:50
  • 1
    $\begingroup$ Hi @kjetilbhalvorsen. Thanks for asking. I would prefer no structure, but it would still be very useful for me if we assumed $\mu_i = 0~ \forall ~ i$ $\endgroup$
    – johneric
    Sep 21, 2015 at 21:21
  • 2
    $\begingroup$ I doubt there's a nice closed form. The maximum CDF has distribution $\exp[-\sum_{i}\exp(-(x-\mu_i)/\beta_i)].$ One idea is to write the inner sum of exponentials as a hypergeometric function, and then look it up in an integral table. Try looking at Meijor G functions. $\endgroup$
    – Alex R.
    Sep 21, 2015 at 23:30

1 Answer 1


The maximum CDF is: $$ \Phi(x)=\exp\bigg[-\sum_i \exp(-\frac{x-\mu_i}{\beta_i})\bigg] $$ which gives the maximum PDF: $$\phi(x) = \frac{\partial\Phi}{\partial x} = \bigg[\sum_i \frac {\exp\big(-\frac{x-\mu_i}{\beta_i}\big)}{\beta_i}\bigg]\Phi (x)$$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.