Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Suppose we have $n$ normal variable $X_1,X_2,\dots,X_n$, with corresponding mean $\mu_1,\dots,\mu_n$ and sd $\sigma_1,\dots,\sigma_n$. What is the probability of $X_1 < X_2 < \dots < X_n$, i.e. $P(X_1 < X_2 < \dots < X_n)$.

Numerical method is also okay. Actually I have thought about how to do the multi-variable integration but I cannot get it done. When $n=2$, it is easy to see that $X_1-X_2$ is also a normal variable so $P(X_1-X_2<0)$ is quite easy to calculate. However, I think when $n=3$, because of $P(X_1-X_2)$ and $P(X_2-X_3)$ are not independent, the result is not easy to calculate. Currently I am thinking about whether we can use some method to approximately calculate the value the that probability. I need to come up with an algorithm to compute that.

Thank you!

share|improve this question
    
The probability that any two are equal is zero, right? So the answer is 1/n! by symmetry. –  Qiaochu Yuan Oct 6 '10 at 14:20
1  
Different $\mu$ and $\sigma$. –  Steve Huntsman Oct 6 '10 at 14:22
    
right, different $mu$ and $delta$ –  user9836 Oct 6 '10 at 14:30
    
Presumably the $X_i$'s are supposed to be independent? –  Andreas Blass Oct 25 '10 at 18:25
add comment

1 Answer 1

up vote 5 down vote accepted

We can use multivariate normal distribution to solve this problem. Denote the difference between consecutive elements with a (n-1)-dimensional random vector $Y=(X_1-X_2,X_2-X_3,\dots, X_{n-1}-X_n)$, what we need to find is $Y<(0,0,...,0)$.

Random vector $Y$ has a multivariate normal distribution.

The mean vector $\mu = (E(X_1-X_2), \dots, E(X_{n-1}-X_n)) = (\mu_1 - \mu_2, \dots, \mu_{n-1} - \mu_n)$.

The covariance matrix $\Sigma = [Cov[X_i-X_{i+1}, X_j-X_{j+1}]], i=1,\dots, n-1, j=1,\dots, n-1$.

For each entry of the matrix,

$Cov_{i,i} = Var(X_i-X_{i+1}) = \sigma_i^2 + \sigma_{i+1}^2$

$Cov_{i,i+1} = Cov_{i+1,i} = E((X_i-X_{i+1})(X_{i+1}-X_{i+2})) - E(X_i-X_{i+1})E(X_{i+1}-X_{i+2})$

$ = E(X_iX_{i+1} + X_{i+1}X_{i+2}-X_{i+1}^2-X_iX_{i+2})-(\mu_i-\mu_{i+1})(\mu_{i+1}-\mu_{i+2})$

Since $X_i, X_{i+1}, X_{i+2}$ are independent, we have the above expression

$= E(X_i)E(X_{i+1})+E(X_{i+1})E(X_{i+2})-E(X_{i+1}^2)-E(X_i)E(X_{i+2}) - (\mu_i-\mu_{i+1})(\mu_{i+1}-\mu_{i+2})$

Apart from $E(X_{i+1}^2)$, other terms are easy to calculate now. For $E(X_{i+1}^2)$, it can be calculate using chi-square distribution of order one. The result turns out to be $\mu_{i+1}^2+\sigma_{i+1}^2$. Plug in the above equation, we have $Cov_{i,i+1}=-\sigma_{i+1}^2$.

Other entries in the covariance matrix is zero.

Finally, use numerical method to calculate the probability. One implementation of the cdf of multivariate normal distribution can be found in http://www.math.wsu.edu/faculty/genz/software/software.html, the MVNDST function.

share|improve this answer
add comment

Your Answer

 
discard

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.