3
$\begingroup$

I'm confused about the difference between the support of a discrete random variable and the atoms of its probability distribution.

Suppose I have a discrete random variable $X$ defined on the probability space $(\Omega, \mathcal{F}, P)$ with support $\mathcal{A}\subseteq \mathbb{R}$. I am told that the distribution of $X$ is $H$ with $c$ atoms, $a_1,\ldots,a_c$.

Question: Is $\{a_1,\ldots,a_c\}=\mathcal{A}$?

My thought is the following: they are different because, while $\mathcal{A}$ could contain also a point $b$ at which $P(X=b)=0$, $P(X=a_i)>0$ for $i=1,\ldots,c$.

Is this correct?

$\endgroup$
2
$\begingroup$

Below I describe a discrete probability distribution whose support is bigger than its set of atoms.

A point $x$ is a member of the "support" of the probability distribution of $X$ precisely if for every open neighborhood $G$ of $x$, we have $\Pr(X\in G)>0$.

Now suppose $$ X = \begin{cases} 1/2 & \text{with probability }1/2, \\[6pt] 1/3 \text{ or } 2/3 & \text{with equal probabilities totalling }1/4, \\[6pt] 1/4 \text{ or } 3/4 & \text{with equal probabilities totalling }1/8, \\ & (\text{We skipped $2/4$ since it's not in lowest terms.}) \\[6pt] 1/5,\ 2/5,\ 3/5,\text{ or }4/5 & \text{with equal probabilities totalling }1/16, \\[6pt] 1/6\text{ or }5/6 & \text{with equal probabilities totalling }1/32, \\[6pt] 1/7,\ 2/7,\ 3/7,\ 4/7,\ 5/7,\text{ or }6/7 & \text{with equal probabilities totalling }1/64, \\[6pt] \text{and so on.} \end{cases} $$

Then the probability distribution of $X$ is discrete since the probabilities of the atoms add up to $1$, i.e. 100 percent of the probability is in point masses. The set of atoms is just the set of all rational numbers between $0$ and $1$.

But the support is the whole set $[0,1]$, which contains $0$ and $1$ (which are not atoms) and every irrational number between $0$ and $1$. The reason for that is for that every open interval about each such number, the probability that $X$ falls in that interval is positive.

$\endgroup$

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.