1
$\begingroup$

The above symmetric property isn't referring to the double stochastic property

\begin{bmatrix} 0.2&0.8&0&0&0\\ 0.2&0.2&0.6&0&0\\ 0&0.4&0.2&0.4&0\\ 0&0&0.6&0.2&0.2\\ 0&0&0&0.8&0.2 \end{bmatrix}

The stationary distribution is given as (1/16,1/4,3/8,1/4,1/16).

Is it true that for an irreducible and aperiodic markov chain, if we can label the first row to last row (state 1 to 5) and also alternatively (state 5 to 1) and we get the same transition matrix, we will have a symmetric stationary distribution i.e. (a,b,c,b,a)?

$\endgroup$
1
$\begingroup$

More generally let $\pi$ be a permutation matrix. The act of permuting rows an columns is equivalent to $\pi P \pi$. Let $\mu$ be the stationary distribution of $P$, and suppose $\pi P\pi=P$. Define $\mu'=\pi^T\mu$. Then:

$$\mu^T=\mu^T P = \mu^T \pi P\pi=\mu'^TP\pi$$

Multiply both sides on the right by $\pi^{-1}=\pi^T$ ($\pi$ is orthonormal afterall). This gives:

$$\mu'^T=\mu^T\pi^T=\mu'^TP.$$

Since the Markov chain $P$ is assumed to be irreducible and aperiodic, it has a unique stationary distribution, which allows us to conclude $\mu'=\mu$. Thus if $P$ is left invariant under permutations of its rows and columns by $\pi$, this implies $\mu=\pi\mu$, i.e. $\mu$ is invariant under $\pi$.

If you look at the individual coefficients, you get $\mu_i=\mu_{\pi^{-1}(i)}$.

$\endgroup$
  • $\begingroup$ did that mean my inference that we will have a "symmetric" stationary distribution is true for all Markov chain P when it is assumed to be irreducible and aperiodic $\endgroup$ – Joseph Nov 1 '17 at 8:01
  • $\begingroup$ @Joseph: Right, I added a line to clarify that. In your case you have the reverse permutation, which implies $\mu_i=\mu_{n-i}$, since $\pi^{-1}(i)=n-i$. $\endgroup$ – Alex R. Nov 1 '17 at 17:26

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.