Suppose $(\Omega,\mathcal{F},\mathbb{P})$ is a probability space, where $\mathbb{P}$ is a probabilitiy measure such that each elementary event has the same probability. It would seem natural to me, that the two implications

  1. $\Omega$ countably infinite $ \, \Longrightarrow \, $ $\forall \, \omega \in \Omega \colon \mathbb{P}(\{\omega\})=0$
  2. $\Omega$ uncountably infinite $ \, \Longrightarrow \, $ $\forall \, \omega \in \Omega \colon \mathbb{P}(\{\omega\})=0$

are both correct. Anyway, I am not sure if they are true or how to prove them.

  • 1
    $\begingroup$ Implication 1) leads to $\mathbb P(\Omega)=\sum_{\omega\in\Omega}\mathbb P(\{\omega\})=0$ contradicting that $P(\Omega)=1$. $\endgroup$ – drhab Mar 30 '16 at 10:04
  • $\begingroup$ I forgot an important assumption: each elementary event should have the same probability. $\endgroup$ – user148364 Mar 30 '16 at 10:05
  • 2
    $\begingroup$ In that case it cannot happen that $\Omega$ is countably infinite. $\endgroup$ – drhab Mar 30 '16 at 10:07

Edit: If we additionally assume that each elementary event has the same probability then both assertions are correct. To see this, let $$\Bbb P(\{ω\})=ε$$ for any $ω\in Ω$ and some $ε>0$ fixed. Choose infinite but countably many of the $ω$'s (with index say $I$, so that $I$ is a countably infinite discrete set) and sum up their probabilities $$\sum_{ω\in Ι}\Bbb P({ω})=|I|ε=+\infty$$ contradicting the fact that $\Bbb P(I)$ must be finite (less or equal than $1$).

If we let different elementary events have different probabilities, then neither is correct. Two well known counterexamples of distributions with countable infinite support are the geometric distribution $$\Bbb P(X=k)=p(1-p)^k$$ for any $k\in \Bbb N$ and $0<p<1$ and the Poisson distribution $$\Bbb P(X=k)=e^{-λ}\frac{λ^k}{k!}$$ for any $k\in \Bbb N$ and $λ>0$. The second is also not correct, since you can assign positive probability only to some $ω\in Ω$. For example, let $Ω=[0,1]$ and $\Bbb P(0)=P(1)=1/2$ and $\Bbb P(ω)=0$ else.

Actually what is correct in this case is the following: There are at most countably many $ω \in Ω$ such that $\Bbb P(\{ω\})>0$. This cannot occur when all elementary events have equal probability as requested.

  • $\begingroup$ I think we have a problem in the second case as $\mathbb{P}$ is only $\sigma$-additive. It does not work for uncountable unions. $\endgroup$ – user148364 Mar 30 '16 at 10:21
  • $\begingroup$ @user148364 Yes, you are right. This is false, since it may well be that $P(ω)=0$ for any $ω\in Ω$ if $P$ is atomless. Let me fix this (shame on me now). $\endgroup$ – Jimmy R. Mar 30 '16 at 10:23
  • $\begingroup$ @user148364 I think that I have a proof now that works for any case. Just go back to a countable sum. If any $ω$ has positive probability then this suffices for a contradiction. What do you think? $\endgroup$ – Jimmy R. Mar 30 '16 at 10:45
  • $\begingroup$ You can add to your last sentence that at least one $P(\{\omega\})>0$, hence they are not all equal as requested by the OP, and your first sentence is wrong. $\endgroup$ – Jean-Claude Arbaut Mar 30 '16 at 10:48
  • 1
    $\begingroup$ @Jimmy R. Yes it sounds clear to me. Thanks. $\endgroup$ – user148364 Mar 30 '16 at 11:22

If $\Omega$ is countable and the sets $\{\omega\}$ are measurable then: $$\mathbb P(\Omega)=\sum_{\omega\in\Omega}\mathbb P(\{\omega\})\tag1$$

If secondly $\mathbb P(\{\omega\})=c$ for each $\omega\in\Omega$ then (1) shows that $c=0$ leads to $\mathbb P(\Omega)=0$ hence contradicts $\mathbb P(\Omega)=1$.

If $c>0$ and $\Omega$ is not finite then (1) also leads to a contradiction: $P(\Omega)=+\infty\neq1$.

So the mentioned conditions can only be satisfied if $\Omega$ is finite.

If $\Omega$ is uncountably infinite then again $c>0$ leads to a contradiction.

For any countably infinite $A\subset\Omega$ we find $\mathbb P(A)=\sum_{\omega\in A}\mathbb P(\{\omega\})=+\infty>1$.

In that case we must have $c=0$ wich will not lead necessarily to a contradiction.


Both implications are false. Note that for any $x \in \Omega$, the Dirac measure defined as $$\delta_x(A)=\begin{cases} 0, & x \notin A \\ 1, & x \in A \end{cases}$$ for any measurable set $A \in \mathcal{F}$ is probability measure, regardless of the set $\Omega$.

Answer to the edited question: If you restrict your attention to just the probability measures that give each singleton the same probability, then there is no such measure in the first case, since $$\bigcup_{x \in \Omega} \{x\}=\Omega$$ Probability $P(\Omega)=1$, but if $P(\{x\})=\varepsilon \neq 0$ for all $x \in \Omega$, the sum $\sum_{x \in \Omega} P(\{x\})$ diverges.

To see that second implication is true, observe any infinite countable subset $I \subset \Omega$. Then, again, $$\bigcup_{x \in I} \{x\} \subset \Omega$$ We know that $P(\Omega)=1$, but sum $\sum_{x \in I} P(\{x\})$ diverges if $P(\{x\}) \neq 0$, for all $x \in I$. Hence, $P(\{x\})=0$, for all $x \in \Omega$.

  • $\begingroup$ I do not think that the sum diverges in every case. $\{\omega_0\} \subseteq \Omega$ is a countable subset of $\Omega$ and then the sum is equal to $\mathbb{P}(\{\omega_0\})$. $\endgroup$ – user148364 Mar 30 '16 at 10:38
  • $\begingroup$ @user148364 Yes of course, by countable I meant infinite and countable. $\endgroup$ – Zoran Loncarevic Mar 30 '16 at 10:41
  • $\begingroup$ How would the first implication be true? You would have by $\sigma$-additivity that $P(\Omega)=0$. $\endgroup$ – Jean-Claude Arbaut Mar 30 '16 at 10:50
  • $\begingroup$ @Jean-ClaudeArbaut Well, in the first case, the set of uniform probability measures on countable set $\Omega$ is empty, so if you add the condition that $P$ is such measure to the left side of implication, it is trivially true. $\endgroup$ – Zoran Loncarevic Mar 30 '16 at 10:56
  • $\begingroup$ True, but misleading. $\endgroup$ – Jean-Claude Arbaut Mar 30 '16 at 10:58

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.