Created by Knut M. Synstadfrom the Noun Project

Joint Markov Chain

Definition (Joint Markov Chain)

Let SS be a state space, and PP a transition matrix over SS. Consider two independent MCs {Xi}i=0\mboxMarkov(λ,P)   {Yi}i=0\mboxMarkov(μ,P)\{X_{i}\}_{i=0}^{\infty}\sim \mbox{Markov}(\lambda,P) \ \ \ \{Y_{i}\}_{i=0}^{\infty}\sim \mbox{Markov}(\mu,P)For each n0n\ge0 we define Wn=(Xn,Yn)W_{n}=(X_{n},Y_{n})

Lemma (Wn\mboxMarkov(θ~,P~)W_{n}\sim \mbox{Markov}(\tilde\theta,\tilde P))

{Wi}i=0\{W_{i}\}_{i=0}^\infty is a MC with the state space S2S^{2}, initial distribution θ~\tilde\theta and transition matrix P~\tilde P: θ~(i,j)=λiμj\mboxforanyi,jSP~(i,j),(i,j)=Pi,iPj,j\mboxforanyi,ij,jS\begin{align*} &\tilde\theta_{(i,j)}=\lambda_{i}\mu_{j}&\mbox{for any }i,j\in S\\ &\tilde P_{(i,j),(i',j')}=P_{i,i'}P_{j,j'}&\mbox{for any }i,i'j,j'\in S \end{align*}

Lemma (Irreducibility of the JMC)

If PP is irreducible and aperiodic, then P~\tilde P is irreducible

Lemma (Invariant Distribution of the JMC)

Let π\pi be an Invariant Distribution for PP. Define π~(i,j)=πiπj\mboxfori,jS\tilde\pi_{(i,j)}=\pi_{i}\pi_{j} \mbox{ for }i,j\in SThen π~\tilde\pi is invariant for P~\tilde P.

Lemma (Coupling lemma)

Let T=inf{n0:Xn=Yn}T=\inf\{n\ge0:X_{n}=Y_{n}\}. Define for each n0n\ge0 \begin{align*} {X_{n}'=\left\{\begin{array}\ X_{n} & \mbox{if }n<T\\ Y_n&\mbox{if }n\ge T\end{array}\right.} \ \ \ \ \ {Y_{n}'=\left\{\begin{array}\ Y_{n} & \mbox{if }n<T\\ X_n&\mbox{if }n\ge T\end{array}\right.} \end{align*} Let Wn=(Xn,Yn)W_{n}'=(X_{n}',Y_{n}') for n0n\ge0

  1. {Wn:n0}\{W_{n}':n\ge 0\} have the same distribution as JMC {Wn:n0}\{W_{n}:n\ge 0\}
  2. Consequently, for any integer nn, supjSP(Xn=j)P(Yn=j)P(T>n)\sup_{j\in S}|P(X_{n}=j)-P(Y_{n}=j)|\le P(T>n)

Linked from