Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Let $x_{1},x_{2},\ldots, x_{n}$ be zero mean Gaussian random variables with covariance matrix $\Sigma=(\sigma_{ij}^{2})_{1\leq i,j\leq n}$. In other words, $$ \mathrm{cov}(x_i,x_j)=\sigma_{ij}^{2} $$ for all $i,j\in \{1,2,\ldots,n\}$.

Let $$ m=\max \{x_{i}:i=1,\ldots,n\} $$ be the maximum of the $x_{i}$. Is it known what is the distribution of $m$? Can we at least compute its mean and variance?

Are there asymptotic results as $n\to\infty$, at least for the weak correlation case ($\sigma_{ij}\ll \sigma_{ii}$ for $i\neq j$)? For instance, assume that $\sigma_{ii}=1$ and $\sigma_{ij}=\sigma_{ij}(n)\to 0$ as $n\to\infty$ for $i\neq j$.

Thanks!

share|improve this question
    
There is no closed form, in general, and the moments can be quite difficult to calculate. Monte Carlo methods are often used in this context. Question: Is there any more structure in the problem you're actually interested in, for example, a certain pattern for the covariance matrix? –  cardinal Apr 24 '11 at 15:04
    
Actually, I'm interested in asymptotics as $n\to\infty$ and in the case where there is correlation but it is very small. In other words, $\sigma_{ij}<<\sigma_{ii}$ for $i\neq j$. –  ght Apr 24 '11 at 15:13
    
Ok. That's a good start. What is your definition of $\sigma_{ij} \ll \sigma_{ii}$ for $i \neq j$? For example, is it something like $\sigma_{ii} = 1$ and $|\sigma_{ij}| < \rho$ for all $i \neq j$? Or, something even stronger? Is there a "base" problem you are trying to generalize from? Maybe that would help pin things down a bit more. –  cardinal Apr 24 '11 at 15:24
    
In my case $\sigma_{ii}=1$ and $\sigma_{ij}=\sigma_{ij}(n)\to 0$ as $n\to\infty$ for $i\neq j$. –  ght Apr 24 '11 at 15:27
    
I know this sounds pedantic, but can you explain the notation further? I think any answer would depend on this information. I can imagine $\sigma_{ij}(n) \to 0$ meaning several things. If you're observing a sequence of random variables, then for fixed $(i,j)$, $\sigma_{ij}$ is constant. If you're observing, say, the rows of a triangular array as your "sequence", then the correlations could be changing even for fixed $(i,j)$. Also, do you know the rate at which the correlations are converging to zero, e.g., $O(n^{-1/2})$? –  cardinal Apr 24 '11 at 15:33
show 1 more comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Browse other questions tagged or ask your own question.