Suppose it is Friday (Fr), and you are asked to determine the probability that it will rain tomorrow on Saturday (Sa). You will need to know whether it rained yesterday, Thursday (Th), and two days ago on Wednesday (We). Given the information in the problem, you can make the following table:
$$\begin{array} {c|c}
\text{Rained on} & \text{Will Rain on} \\
\begin{array} {ccc}
\text{We} & \text{Th} & \text{Fr} \\ \hline
Y & Y & Y \\
Y & Y & N \\
Y & N & Y \\
Y & N & N \\
N & Y & Y \\
N & Y & N \\
N & N & Y \\
N & N & N \\
\end{array}&
\begin{array} {c}
\text{Sa} \\ \hline
0.8 \\
0.4 \\
0.6 \\
0.4 \\
0.6 \\
0.4 \\
0.6 \\
0.2 \\
\end{array}
\end{array}$$
Now this is a transition table , but the output states are not the same as the input states. The input states are the rain of 3 days; the input states are the rain of 1 day. So to make a transition matrix, we need the output states to be the same as the input states: 3 days to 3 days, {We, Th, Fr} to {Th, Fr, Sa}.
$$\begin{array} {c|c}
\text{Rained on} & \text{Will Rain on Th Fr Sa } \\
\begin{array} {ccc}
\text{We} & \text{Th} & \text{Fr} \\ \hline
Y & Y & Y \\
Y & Y & N \\
Y & N & Y \\
Y & N & N \\
N & Y & Y \\
N & Y & N \\
N & N & Y \\
N & N & N \\
\end{array}&
\begin{array} {c}
\text{YYY} & \text{YYN} & \text{YNY} & \text{YNN} &
\text{NYY} & \text{NYN} & \text{NNY} & \text{NNN} \\ \hline
0.8 & 0.2 \\
& & 0.6 & 0.4 \\
& & & & 0.6 & 0.4 \\
& & & & & & 0.6 & 0.4 \\
0.6 & 0.4 \\
& & 0.6 & 0.4 \\
& & & & 0.6 & 0.4 \\
& & & & & & 0.2 & 0.8 \\
\end{array}
\end{array}$$
Each blank entry is an impossible transition. For example, it's impossible to transition from a Friday where it rained 3 days in a row to a Saturday where it was clear 3 days in a row (the top right corner). So your final transition matrix is:
$$\begin{bmatrix}
0.8 & 0.2 & 0 & 0 & 0 & 0 & 0 & 0 \\
0 & 0 & 0.4 & 0.6 & 0 & 0 & 0 & 0 \\
0 & 0 & 0 & 0 & 0.6 & 0.4 & 0 & 0 \\
0 & 0 & 0 & 0 & 0 & 0 & 0.4 & 0.6 \\
0.6 & 0.4 & 0 & 0 & 0 & 0 & 0 & 0 \\
0 & 0 & 0.4 & 0.6 & 0 & 0 & 0 & 0 \\
0 & 0 & 0 & 0 & 0.6 & 0.4 & 0 & 0 \\
0 & 0 & 0 & 0 & 0 & 0 & 0.2 & 0.8 \\
\end{bmatrix}$$
The "notation" (if it can even be called that) that your guide is using is replacing the yes/no sequence with a number (So YYY is 0, NNN is 7, it is quite confusing). $P_{0,0}$ is the probability that it rained for the past 3 days and tomorrow will also be a day where it has rained for 3 days.
There are different types of Markov chains, which you will be able to recognize by their transition matrix (and/or by their transition graph). These transition matrices have the property that their columns add up to one and all entries are between 0 and 1. I will discuss here two of the most common types. (Here I will use the convention that the transition matrix has columns adding to one and multiplying by distribution vectors on the right).
A regular Markov Chain is one in which for our transition matrix $T$, there exists some $n$ such that $T^{n}$ has all nonzero positive entries. Regular Markov chains have the very nice property that as $n\to\infty$ you will have some stable distribution, $y$, such that $T y = y$, and furthermore that $\lim_{n\to\infty}T^n x = y$ regardless what $x$ is.
To solve, noting that $Ty = y$ and that $y$ is a distribution, meaning that the sum of the entries of $y$ will equal one, you are given the system of equations:
$\begin{cases} y_1 + y_2 + \dots + y_k = 1\\ T_{1,1}y_1 + T_{1,2}y_2 + \dots + T_{1,k}y_k = y_1 \\ \vdots \\ T_{k,1}y_1 + T_{k,2}y_2 + \dots + T_{k,k}y_k = y_k\end{cases}$
This is a system of $k+1$ equations in $k$ unknowns, and as it so happens, one of the lines will be redundant (the rows are not linearly independent due to the stochastic property). Using your favorite method, solve the system to obtain the stable distribution $y$.
The long term matrix, $\lim_{n\to\infty}T^n$, having the property that any initial distribution (in particular distributions with only one nonzero state) will eventually reach the stable distribution, it follows that it will appear as every column being a copy of $y$.
The other type of Markov Chain you are likely to see is an absorbing markov chain. An absorbing markov chain is one that has an absorbing state, a state that once entered cannot be left. I.e. when looking at the transition matrix, there will be a one along the main diagonal. The rows/columns with a one are the absorbing states. The final distribution in these cases will depend on what the initial distribution is, however there will be a longterm trend for the transition matrix itself.
Reorganizing the matrix into its standard form by swapping columns and rows (making sure that the order states appear in the columns is the same as the order they appear for the rows), you will have a matrix in the form
$$T=\left[\begin{array}{c|c}
I & S\\ \hline 0 & R\end{array}\right]$$
and the long term matrix will be:
$$\lim_{n\to\infty} T^n = \left[\begin{array}{c|c}
I & S(I-R)^{-1}\\ \hline 0 & 0\end{array}\right]$$
See examples of absorbing markov chains in action in some of my previous answers Chance of winning simple dice game and Sniper probability question
For an example of a regular markov chain, suppose we have a fleet of taxicabs and three major hubs in the city for them to go. We have midtown, downtown, and the airport. Each taxi will transition from one location to another or stay at the same spot once every 15 minutes.
Suppose our transition diagram is given as:
Our transition matrix is then:
$$\begin{bmatrix} .4 & 0 & .2\\ .3 & 0 & .7\\ .3 & 1 & .1\end{bmatrix}$$
This doesn't have nothing but nonzero numbers, so we might be hesitant to think it falls under the first case, but tedious arithmetic aside, you will see that $T^3$ has all strictly positive entries. (This is seen easier from the transition diagram from a graph theoretic sense by noting that there exists a length 3 path from any state to any other state).
Thus, we solve the system:
$$\begin{cases} y_1+y_2+y_3 = 1\\ .4y_1 + 0 + .2y_3 = y_1\\ .3y_1 + 0 + .7y_3 = y_2\\ .3y_1 + 1y_2 + .1y_3 = y_3\end{cases}$$
Rearranging and setting up a matrix to rowreduce, we arrive at:
$$\left[\begin{array}{ccc|c}
1 & 1 & 1 & 1\\
-.6 & 0 & .2 & 0\\
.3 & -1 & .7 & 0\\
.3 & 1 & -.9 & 0\end{array}\right]$$
which row reduces to
$$\left[\begin{array}{ccc|c}
1 & 0 & 0 & .15625\\
0 & 1 & 0 & .375\\
0 & 0 & 1 & .46875\\
0 & 0 & 0 & 0\end{array}\right]$$
Indeed, checking $Ty$ it works out correctly.
Thus, after a "long time", about an eighth of the taxis will be at midtown, about one third of the taxis will be at the airport, and about half will be at downtown.
Our longterm matrix is thus three columns identical to $y$.
Best Answer
Definitely you need to have $W_{21}=W_{31}$, so as to make sure that the probability for the transition from state II to state I doesn't depend on whether state II was an instance of state $2$ or an instance of state $3$.
But as I see it, that's all you need.
Ler $a=W_{11}$, and let $b=W_{21}=W_{31}$.
Then the Markov chain for the new states has transition matrix $$ \pmatrix { a & 1-a \cr b & 1-b } $$