Transition Matrix
2 Markov Chains
2.1 Basic Principles.
2.1.1 The Transition Matrix.
A stochastic process is a mathematical model of a situation in the real world that evolves in time in a
probabilistic fashion, i.e. we don't seem to be able to completely predict the future. The situation in the
real world is often called a system. For the time being we model the system in discrete time as opposed
to continuous time. For example, this might be because we observe the system at discrete time
intervals. We let n = 0 be the initial time that we observe the system, n = 1 the next time and so on.
We are interested in calculating probabilities that the system is in various states at various times.
Mathematically, we describe the system by a sequence of random variables X0, X1, ..., Xn where
X0 = the state of the system at time n = 0,
X1 = the state of the system at time n = 1,
. . .Xn = the state of the system at time n
At any particular time the system is in one of s possible states that we number 1, 2, …, s.
Example 1. An office copier is either in
1. good condition (G or 1),
2. poor condition (P or 2), or
3. broken (B or 3).
So the possible states of the system are G = 1, P = 2 and B = 3. At the start of each day we check the
condition of the copier. If we do this for a sequence of four days, we might observe that today it is G,
tomorrow it is G, the next day it is P and the fourth day it is B. In this case we would have X0 = 1, X1 = 1,
X2 = 2 and X3 = 3. Unfortunately, we can't predict the state of the copier in the future.
One of the most basic probabilities that one might want to know is the probability of a sequence of
observations x0, x1, ..., xn. We would denote this probability by
Pr{X0 = x0, X1 = x1, ..., Xn = xn}
, For example, in the context of Example 1 we might want to know the probability that today it is G,
tomorrow it is G, the next day it is P and the fourth day it is B. This would be denoted by
Pr{ X0 = 1, X1 = 1, X2 = 2, X3 = 3}.
For a general stochastic process, one might provide the values of Pr{ X0 = x0. X1 = x1........., Xn = xn} for
each sequence x0, x1,...., xn. This is a lot of data to provide. Things are a lot easier if we make the
following two simplifying assumptions. First the probability of the state being something at one
observation depends only on the previous observation. This can be described by the equation
(1) Pr{Xn+1 = xn+1 | X0 = x0, X1 = x1,...., Xn = xn} = Pr{Xn+1 = xn+1 | Xn = xn}
This is called the Markov assumption and a stochastic process satisfying this assumption is called a
Markov process.
The second assumption is that the quantity appearing on the right of (1) depends only on xn and xn+1 and
not on n, i.e.
(2) Pr{Xn+1 = j | Xn = i} = pij
A Markov process statisfying (2) is called stationary or time-homogeneous. A stochastic process
satisfying (1) and (2) is called a Markov chain.
As an example of what (1) implies, suppose we wanted to know the probability that tomorrow the
copier is broken. Consider the following two scenarios. First, we are told the copier is in poor
condition today. Then the probability the copier is broken tomorrow would be Pr{ X2 = 3 | X1 = 2}
assuming today is time n = 1 and tomorrow is time n = 2. On the other hand suppose we are told both
the copier was in good condition yesterday and is in poor condition today. Then the probability the
copier is broken tomorrow would be Pr{X2 = 3 | X0 = 1 X1 = 2} assuming yesterday is time n = 0. The
Markov assumption would imply that knowing the copier was in good condition yesterday didn't add
any information to the fact that it is in poor condition today, i.e.
Pr{X2 = 3 | X0 = 1, X1 = 2} = Pr{X2 = 3 | X1 = 2}.
The probabilities pij appearing in (2) are called transition probabilities and the matrix
P=
is called the transition matrix. Corresponding to the transition matrix is a transition diagram where the
states are represented by circles and the transitions are represented by arrows from one circle to another
labeled by the transition probabilities.
Example 1 (continued). In the context of Example 1 suppose the transition matrix is
P=