Unpacking the definition of a discrete random variable with an example

57 Views Asked by At

I am taking a course on Markov chains which is intended to be accessible without a rigorous understanding of probability measure, but I'm not grasping even basic definitions and it is hindering my progress. Specifically, I want to precisely understand the definition of a discrete random variable. The following is an excerpt from Markov Chains by Norris. Below I will detail my confusion.

Let $I$ be a countable set. Each $i\in I$ is called a state and $I$ is called the state space. We say that $\lambda=(\lambda_i:i\in I)$ is a measure on $I$ if $0\leq \lambda_i <\infty$ for all $i\in I$. If in addition the total mass $\sum_{i\in I} \lambda_i$ equals $1$, then we call $\lambda$ a distribution. We work throughout with a probability space ($\Omega, \mathcal{F}, \mathbb{P})$. Recall that a random variable $X$ with values in $I$ is a function $X:\Omega\to I$. Suppose we set $$\lambda_i = \mathbb{P}(X=i)=\mathbb{P}(\{\omega:X(\omega)=i\}).$$

My goal is to write down an example of a discrete probability space and a random variable. Let's say that $I=\{a,b,c\}$ with $\lambda_i=1/3$. So, the probability of chooseing $a$ from this set is 1/3. How could you express this in the language of some probability space $(\Omega, \mathcal{F},\mathbb{P})$?

What would $\Omega$ be? According to wikipedia, it is the state space. I would have thought that $I$ is the state space, and yet the book is distinguishing between $\Omega$ and $I$. My understanding is that $\mathcal{F}$ is a sigma algebra on $\Omega$. I can read the definition of a sigma algebra, but I don't have good intuition for the role it is playing here. Somehow, a measure on some function $X:\Omega\to I$ can capture the notion that choosing $a$ has probability 1/3.

My hope is that this simple example, perhaps accompanied by some comments on how to interpet this formalism, will be enough to get me on my way.

1

There are 1 best solutions below

0
On BEST ANSWER

If you were just going to deal with one random variable, you could take $\Omega = I$, but typically there are going to be lots of them, and $\Omega$ has to be rich enough to handle all of them. For example, if you were dealing with $5$ random variables $X_1$ to $X_5$, each taking values in the state space $I = \{a,b,c\}$, then you might make $\Omega = I^5$, and each $X_i$ would be one of the coordinate maps $(x_1, \ldots, x_5) \to x_i$.

The $\sigma$-algebra $\mathcal F$ tells you what subsets of $\Omega$ you can assign probabilities to. At this stage, it can be all the subsets of $\Omega$. When you deal with continuous random variables or infinite collections of random variables, this will not be possible because of the existence of non-measurable sets.