2
$\begingroup$

Let $\lambda_k,\mu_k\in\mathbb R_{\ge0}$ $(k\ge1)$ be nonnegative real numbers such that $\sum_{k=1}^\infty k\lambda_k<\infty,$ let $S=\mathbb Z_{\ge0}$ be the nonnegative integers, let $T=\mathbb R_{\ge0}$ be the nonnegative real numbers and consider the continuous-time Markov chain $X=(X_t)_{t\in T}$ on $S$ with rates $$Q(n,n+k)=(n+1)\lambda_k\quad(k\ge1),\qquad Q(n,n-k)=(n+1-k)\mu_k\quad(1\le k\le n).$$ (This Markov chain appears in biology as a model of the length of an evolving DNA sequence (Miklós et. al. 2004). A MathOverflow post contains more information about this process.) For example, if $0=\lambda_k=u_k$ for all integers $k\ge2,$ then we recover the linear birth-death process with immigration with birth rate $\lambda_1,$ death rate $\mu_1$ and immigration rate $\lambda_1,$ whose nonzero rates are $$Q(n,n+1)=(n+1)\lambda_1\quad(n\ge0,k\ge1),\qquad Q(n,n-1)=n\mu_1\quad(n\ge1).$$

Now, assume this chain is reversible and irreducible, and let $\nu$ be the stationary measure. Then detailed balance implies $\forall n\in\mathbb Z_{\ge0}~\forall k\in\mathbb Z_{\ge1}\quad \nu_n\cdot(n+1)\lambda_k=\nu_{n+k}\cdot(n+1)\mu_k,$ so that $\nu_n\lambda_k=\nu_{n+k}\mu_k.$ From reversibility and irreducibility it follows that $\nu_i>0$ for all $i\in S,$ so that $\forall k\in\mathbb Z_{\ge1}\quad\lambda_k=0\leftrightarrow\mu_k=0.$

Now, let $A=\{k\in\mathbb Z_{\ge1}:\mu_k\ne0\}~(=\{k\in\mathbb Z_{\ge1}:\lambda_k\ne0\}),$ and let $d=\gcd A.$ If $d\ne1,$ then $d>1,$ and so considering things mod $d$ we see that $X$ is not irreducible, contradiction; thus $\gcd A=d=1.$ And if $1\in A,$ then for all $n\in S$ we have $\nu_n\lambda_1=\nu_{n+1}\mu_1,$ and since $\mu_1\ne0$ we can say $\nu_{n+1}=\nu_n\lambda_1/\mu_1,$ so that $\forall k\in S\quad\nu_k=\nu_0(\lambda_1/\mu_1)^k.$ Normalizing, we have $\forall k\in S\quad\nu_k=(1-\lambda_1/\mu_1)(\lambda_1/\mu_1)^k.$ Note that detailed balance also yields $\forall k\in A\quad\lambda_k/\mu_k=(\lambda_1/\mu_1)^k.$

Now, my question is: What is the stationary distribution, if $1\not\in A?$ I tried the following example: Take $A=\{2,3\}.$ In addition, let ($r$ for ratio) $r_2=\lambda_2/\mu_2,$ $r_3=\lambda_3/\mu_3.$ From detailed balance, we have for all $n\in S$ that $\nu_nr_3=\nu_3=\nu_{n+1}r_2,$ so that $\nu_{n+1}=\nu_nr_3/r_2.$ It follows that $\forall n\in S\quad\nu_n=(1-r_3/r_2)(r_3/r_2)^n.$ (Note that $r_3/r_2=\lambda_1/\mu_1$ if $\{1,2,3\}\subseteq A.$) However, I cannot think of how to generalize this example to arbitrary sets $A.$


Miklós, I., Lunter, G. A., & Holmes, I. (2004). A “long indel” model for evolutionary sequence alignment. Molecular Biology and Evolution, 21(3), 529-540.

$\endgroup$
1
$\begingroup$

The reversibility assumption (and detailed balance) are doing a lot of heavy lifting here. If it holds, then we have $\nu_{n+k} = r_k \nu_n$ for all $n \in \mathbb Z_{\ge 0}$ and all $k \in A$, which essentially tells us $\nu$, we just have to massage it into the right form.

Since $\gcd A = 1$, there is some integer linear combination $\sum_{k \in A} c_k \cdot k = 1$, and it follows from $\nu_{n+k} = r_k \nu_n$ that $$ \nu_{n+1} = \nu_n \prod_{k \in A} r_k^{c_k}. $$ So we have $\nu_n = (1-\rho) \rho^n$, where $\rho = \prod_{k \in A} r_k^{c_k}$. In your example, we can take $c_2 = -1$ and $c_3 = 1$, so $\rho = \frac{r_3}{r_2}$, and we get back your formula. Of course, we can also take $c_2 = 2$ and $c_3 = -1$, and get a formula with $\rho = \frac{r_2^2}{r_3}$. Detailed balance only holds if all of these expressions for $\rho$ are equal; for any $j,k \in A$ we must have $r_j^k = r_k^j$. In that case, we actually just have $r_k = \rho^k$ for all $k \in A$, and the process is not meaningfully different in the end from the $A = \{1\}$ case.

(In other words, assuming detailed balance holds, we can just pick any $k \in A$ and say that $\nu_n = (1-r_k^{1/k}) r_k^{n/k}$, ignoring all other elements of $A$.)

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.