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.

I want to find the normalization constant for the following pdf

$$ f(x_1,x_2,\ldots,x_n) = C\cdot\Pi_{m=1}^n\exp\left\{-\frac{1}{2a^2}(x_m-\sum_{l=1}^L\alpha_lx_{m-l})^2+y_m\sum_{l=0}^L\beta_lx_{m-l}\right\} $$ where $n>L$ and $x_l=0$ for $l\leq0$. Namely, I want to find $C$ for which

$$ \int_{\mathbb{R}^n}f(x_1,x_2,\ldots,x_n)dx_1\ldots dx_n = 1. $$

Any idea?

share|improve this question
    
I think you mean $x_l=0$ for $l\le0$? –  joriki Dec 15 '12 at 7:11
    
@joriki: Yes, you right. Thx –  Josh Dec 15 '12 at 12:02
1  
An idea would be to write the argument of the exponential in a matrix form, complete the squares and diagonalize the matrix. But it won't be pretty. –  Marek Dec 15 '12 at 12:10
    
@Marek: Yes, I have tried that, and it's really not pretty. I thought that maybe there is a nice trick that I misses. –  Josh Dec 15 '12 at 13:05
1  
Then I'd suggest to transform from the index $l$ to the index $k=m-l$ so you don't have to deal with arithmetic inside the indices. –  joriki Dec 15 '12 at 18:46

1 Answer 1

up vote 1 down vote accepted

From your comment, it seems that you may not be aware that the outer product is equivalent to a sum in the exponent. Using $\exp(a)\exp(b)=\exp(a+b)$ and transforming the indices as suggested in a comment, you can rewrite your function as

$$ f(x_1,x_2,\ldots,x_n) = C\exp\left\{\sum_{m=1}^n\left(-\frac{1}{2a^2}\left(x_m-\sum_{k=m-L}^{m-1}\alpha_{m-k}x_k\right)^2+y_m\sum_{l=0}^L\beta_{m-k}x_k\right)\right\}\;. $$

You can see from this representation that the matrix of the quadratic form (whose determinant you need) is a band matrix with bandwidth $2L+1$. The diagonal entries are constant ($1+\sum_i\alpha_i^2$) up to $m=n-L$, but decrease for greater $m$ (because successive $\alpha_i^2$ go missing) down to $1$ for $m=n$. Because of this asymmetry, I doubt that you'll find a nice closed form for the determinant.

share|improve this answer
    
Yes, my previous comment is not clear (I was aware that the outer product is equivalent to a sum in the exponent). Anyway, it really looks that I will not find a nice solution. Thx. –  Josh Dec 15 '12 at 19:31

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.