3
$\begingroup$

I'm struck with an exercise. I tried, but the results don't seem to fit to those proposed.

Exercise: Two players play the following game. The one who begins draws two cards from a deck of 40 cards (10 cards per suit):

  • if they are both clubs the player wins;
  • if they are of the same suit but not clubs, the player shuffles the cards and start again;
  • else the player shuffles the cards and let the other player play.

1) Model the game with a Markov Chain.

2) What is the probability that who started wins?

My attempt: If $X_n$ is the player in the current turn, we have a three state MC whose state space is $\{A,\, B,\, \text{exit}\}$. The exit state is the one reached when the game ends. The transition matrix is

$$P=\begin{pmatrix}9/52 & 10/13 & 3/52 \\ 10/13 & 9/52 & 3/52 \\ 0 & 0 & 1 \end{pmatrix}$$

since the probability of staying is $\frac{3 \cdot 9}{4 \cdot 39}$ and so on.

For the second question I thought I could use the Strong Markov Property (the one which states that a Markov Chain can start afresh in a Stopping Time) by using the last time I see A (or B) before the game ends. Starting from that point, the probability is just the jump from A (or B) to exit times two (to consider both the A and the B case).

What's wrong with this last point?

$\endgroup$
  • $\begingroup$ $40$ cards or $52$? $\endgroup$ – Henry Jun 30 '15 at 21:54
  • $\begingroup$ 40. It's an Italian deck :) $\endgroup$ – Phugo Jun 30 '15 at 21:55
  • 1
    $\begingroup$ Would you, please, explain the meaning of the states $A,B,exit$? $\endgroup$ – zoli Jun 30 '15 at 22:01
  • 1
    $\begingroup$ I only read this quick but usually, the ``last time A happens before B'' types of variables are not stopping times. Edit : the time variable you take is definitely not a stopping time $\endgroup$ – hHhh Jul 1 '15 at 15:18
  • 1
    $\begingroup$ @Phugo intuitively a stopping time is a time at which you can stop knowing only the past and present. In your case you need to know what happens after $T$ to determine the value of $T$. see en.wikipedia.org/wiki/Stopping_time $\endgroup$ – hHhh Jul 1 '15 at 18:37
2
$\begingroup$

First, I wouldn't modelise the markov chain like that , I would consider four states 1 = "player A play", 2 = "player B play", 3 = "player A had won" and 4 = "player B had won"

The transition matrix would be

$$M = \begin{pmatrix} 9/52 & 10/13 & 3/52 & 0 \\ 10/13 & 9/52 & 0 & 3/52 \\ 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 1 \end{pmatrix}$$

Then the probability to be at each state after n step knowing that we started in state 1) is given by

$$\pi^{(n)} = (1,0,0,0)M^n$$

And the probability we're looking for is

$$p = \lim_{n\to +\infty} \pi_3^{(n)}$$

To calculate $M^n$, you can diagonalize it, and it should give you the answer (I didn't do the calcul)

$\endgroup$
  • $\begingroup$ Thank you! It's more or less the same procedure outlined in my handout, but I was really interested in understanding why the method I proposed is not alright. Probably I didn't understand what's the Markov Property is about and I want to know where I have been wrong. Thank you, again :) $\endgroup$ – Phugo Jun 30 '15 at 22:30
1
$\begingroup$

To solve question 2 you don't need Markov chains, only first step analysis. In the first step there are 3 possibilities:

  1. player $A$ wins,
  2. the game starts over,
  3. the deck goes to player $B$.

Letting $p_A$ be the probability that player $A$ is the eventual winner, and taking these 3 cases into account gives equation (1) below.

With $p={10\choose 2}/{40\choose 2}$ we get $$p_A=p\cdot 1 +3p\cdot p_A+(1-4p)\cdot(1-p_A).\tag 1$$ Solving (1) gives $p_A={43\over 83}= 0.51808$.

$\endgroup$

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.