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

I am trying to find out the distribution of a matrix exponential which is a function of a random variable. My mathematics background is very limited and I hope I can receive some help from here.

What I have is a real symmetric matrix $M\in R^{n\times n}$ with $m_{pq}=m_{qp}=\kappa c_{pq}={\kappa}c_{qp}$ when $p\neq q$ and $m_{pq}=-k_p$ when $p=q$, where $k_p$ and $c_{pq}$ are real constants, and $\kappa$ is a Gaussian random variable: $\kappa\sim N(0,\sigma)$ with $|\sigma c_{pq}|\ll|k_p|$. The values of $k_p$ are very close to each other, and $M$ is sparse i.e. $m_{pq}, p\ne q$ are nearly zero everywhere. I need to find out the distribution of the components of the matrix exponential $D=e^{iMt}$, where $i=\sqrt{-1}$, and $t$ is a large positive integer.

Because $|\sigma c_{pq}|\ll|k_p|$, my first attempt is to treat the off-diagonal elements of $M$ as small perturbations and use the first order perturbation theory to compute the eigenvalues and eigenvectors as simple linear functions of $\kappa$. Because the matrix exponential can be obtained by some simple manipulation of the eigenvalues and eigenvectors, I thought I could get the distribution of $D$ easily. However, some of the eigenvectors are found to be very sensitive to the perturbations, probably because $M$ is sparse and with nearly equal diagonal elements. Large error is found from this method.

My next attempt is to separate $M$ into two parts: $M=R+S$, where $R$ is a diagonal matrix with $r_{pq}= m_{pq}$ if $p=q$ and $r_{pq}=0$ if $p\neq q$, and $S$ is the matrix that contains the off-diagonal components of $M$. Because $|\kappa c_{pq}|$ is generally small, $D=e^{iMt}=e^{i(R+S)t}\approx (e^{iR}e^{iS})^t$. The distribution of $U=e^{iR}e^{iS}$ is easy to evaluate. Because: $u_{pq}\approx e^{-ik_{p}}$ if $p=q$ and $u_{pq}\approx i\kappa c_{pq}e^{-ik_p}$ if $p\neq q$, the diagonal components of $U$ are constants,and the off-diagonal components of $U$ follows the imaginary Gaussian distribution. However, although it looks simple, I don't know how to evaluate the distribution of $U^t$.

share|improve this question
add 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.