Consider the $Q$-matrix on $\mathbb{N}$ defined by $q(m,m+1)=-q(m,m)=\rho m$ and $q(m,m')=0$ if $|m-m'|>1$, where $\rho$ is a constant. Letting $(P_t)$ be the transition function of this process, to show is that for $k\leq \ell$ $$ P_t(k,\ell) = {\ell-1\choose \ell-k}e^{-k\rho t}(1-e^{-\rho t})^{\ell-k} $$ I have already determined that $p(m,m+1) = 1$ and each stay at state $m$ is $\sim \exp(\rho m)$.

Every try based on only calculations resulted in mess. I think we must use that something in this system is binomial, because the final expression looks like that. To come to there, we can think about the reserved process $(Y_t)$ of $(X_t)$ and transform the system a bit: if $Y_t$ is the number of particles at time $t$, then if we assume that each of the particles dies exponentially with rate $\rho$, then $(Y_t)$ is indeed the reversed process of $(X_t)$. This is because $p(m,m-1) = 1$ for $(Y_t)$ and $p(m,m+1)=1$ for $(X_t)$, and the lengths of periods at all states $m$ have the same distribution at both $(X_t)$ and $(Y_t)$. But I don’t know how to use this further. Bayes rule seems to be essential, but we need to know $P(X_s = k)/ P(X_{s+t} = \ell)$ for this.

Is it doable to calculate $P(X_s = k)/P(X_{s+t} = \ell)$, or would I yet better just calculate $P_t(k,\ell)$ without using the fact that something is binomial?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.