22 Discrete-Time MC

Now we find the optimal p. Let X1,⋯,XN denote the numbers. Order statistics: X(1)<⋯<X(N).

So P(success)=∑k=1(1−p)Np(1−p)k(1k)≈−pln⁡p, which is maximized when p=e−1.

1 Basics of DTMC

Discrete-Time Markov Chain (DTMC)

{Xn,n∈N0}. State space S discrete. Then Markov Chain satisfies for all n∈N,s∈S P(Xn+1=sn+1|X0=s0,⋯,Xn=sn)=P(Xn+1=sn+1|Xn=sn).

For example, 1d random walk.

Claim (Chapman-Kolmogorov Equation)

∀n,m∈N, pij(m+n)=∑k∈Spik(m)pkj(n). I.e. (Pn+m)ij=(PnPm)ij.

This claim shows that the n -step Pn is actually the matrix product.


A graphical representation of a homogeneous Markov chain:
Pasted image 20241201180808.png

Some questions of interest: given X0=i∈{1,⋯,L−1},

2 Classification of States

Accessible & Intercommunicate States

  • i→j: state j is accessible from i if pij(n)>0 for some n∈N.
  • i↔j: states i,j are intercommunicate if i→j,j→i.

Some Important Values

  • First passage probability fij(n)=P(X1≠j,⋯,Xn−1≠j,Xn=j|X0=i).
  • Return probability fii=∑n=1∞fii(n). (probability of finally coming back)
  • Mean recurrence time ri=∑n=1∞nfii(n). (expected time of coming back)

State Definitions

A state i∈S is called

  1. Recurrent/persistent if fii=1.
    1. Null recurrent if ri=∞.
    2. Positive recurrent if ri<∞.
  2. Transient if fii<1.
Theorem

  1. ∑n=1∞pjj(n)=∞⇔ j is recurrent ⇒∀i∈S s.t. i→j, ∑n=1∞pij(n)=∞.
    Recurrent is equivalent to visiting the state in infinitely many times.
  2. ∑n=1∞pjj(n)<∞⇔ j is transitent ⇒∀i∈S. ∑n=1∞pij(n)<∞ and limn→∞pij(n)=0.

3 Hitting and Recurrence

Suppose S={1,⋯,m}⏟B,transient states∪{m+1,⋯,n}⏟A,absorbing. We can't return from A to B (such A are defined as absorbing states, and B are transient). Then transition matrix can be written as P=[QROS].

Hitting Probability

For i∈B,j∈A, hij=P(enters A through j∈A|X0=i).

Claim

hij=pij+∑k∈Bpikhkj.

In Matrix form, H=(hij) satisfies H=R+QH. Then if (I−Q)−1 exists, H=(I−Q)−1R.

Lemma

Suppose M is a square matrix with limn→∞Mn=0. Then (I−M)−1 exists and (I−M)−1=∑i=0∞Mn.

Back to (I−Q)−1. Since Pn=[Qn∗OSn]⇒pik(n)=qik(n).
k transient ⇒limn→∞pik(n)=0⇒limn→∞Qn=0. So by lemma, (I−Q)−1 exists.

Fundamental Matrix

(I−Q)−1=∑i=0∞Qn is called the fundamental matrix of the absorbing Markov Chain.

Claim (Hitting Time)

Let ti be expected number of steps it takes to hit A given X0=i∈B. Then [t1t2⋮tm]=(I−Q)−1[11⋮1].

4 Stationary Distribution

Stationary Distribution

The row vector π=(πi)i∈S is called a stationary distribution of the Markov Chain with transition probability matrix P, if

  • πi≥0,∀i∈S,∑i∈Sπi=1.
  • πP=π. (i.e. π is the left eigenvector of P with eigenvalue 1.)

If X0∼π, then Xn∼π,∀n∈N.

Does every Markov Chain has a stationary distribution? When it exists, is it unique?

Theorem (Perron-Frobenius)

  1. Every Markov Chain with finite S has a stationary distribution π.
  2. If in addition the chain is irreducible, then π is unique, and πi=ri−1,∀i∈S, where ri=∑i=1∞nfii(n) is the mean recurrence time for state i∈S.