Let $(\Omega, \mathscr{A}, \mathbb{R})$ be a probability space.

We define an atom of probability as an event $A \in \mathscr{A}$ such that $\mathbb{P}(A)>0$ and for all events $D \subset A$, $\mathbb{P}(D) = 0$ or $\mathbb{P}(D)=\mathbb{P}(A)$.

What my textbook states about atoms is (and I understand the proof of both of these statements): if $A$ and $B$ are atoms, then either $\mathbb{P}(A \cap B)=\mathbb{P}(A)=\mathbb{P}(B)$ or $\mathbb{P}(A \cap B)=0$, and there can be at most countably many disjoint (up to an event of zero probability) atoms. I'm not sure if this is useful for my question, but I stated it here anyway.

Now, for my question: prove that $\Omega$ can be written as a union $$ \Omega = (\bigcup_{n=0}^{+\infty}A_{n}) \cup X,$$ where $\{A_{n}\}_{n \in \mathbb{N}}$ is a sequence of atoms (does not specify what their intersections are), and $X$ is an event with the following property: for any $p \in (0, P(X))$, there exists an event $B \subset X$ such that $P(B)=p$.

My idea was to take a maximal family of disjoint (again, up to an event of zero probability) atoms, which is at most countable, $\{A_{n}\}_{n \in \mathbb{N}}$, and consider $X := \Omega \setminus \cup_{n=0}^{+\infty}A_{n}$, and prove the required property for that set.

If $Y \subset X$ is an atom, then, seeing as how my family is maximal, $Y$ must non-trivially intersect some $A_{i}$ from my sequence, and therefore $\mathbb{P}(Y\cap A_{i})=\mathbb{P}(A_{i})>0$, but also since $Y \subset X$, $Y \cap A_{i}= \emptyset$, so $\mathbb{P}(Y \cap A_{i})=0$, which is a contradiction, so $X$ cannot contain any atoms. Is this useful for what I'm supposed to prove? I can't seem to see where to go from here, or if this is even the right first step.

EDIT: I'm not even 100% sure if I can take a maximal family without proving it exists in some way, so this possibly isn't the right way to go.

EDIT 2: I've managed to work out that because $X$ cannot contain any atoms, for every $p \in (0, \mathbb{P}(X))$, there must exist an event $B$ such that $\mathbb{P}(B)<p$, therefore $0$ is a limit point of the set of $p \in (0, \mathbb{P}(X))$ for which there exists an event $B$ for which $\mathbb{P}(B) = p$, but this still doesn't give me the whole solution.

  • 1
    $\begingroup$ Please clarify : your notation $AB$ is what everyone else denotes by $A\cap B$, i.e. the intersection ? $\endgroup$ – Ewan Delanoy Apr 24 '17 at 7:19
  • $\begingroup$ Yes, in my environment (and I think probability theory as well) $AB = A \cap B$. $\endgroup$ – Matija Sreckovic Apr 24 '17 at 9:39
  • 1
    $\begingroup$ Thank you for clarifying this point. Personally I've never seen the notation $AB$ used to mean intersection in any probability textbook, and I would advise you to change all your $AB$'s to $A\cap B$ in your text. $\endgroup$ – Ewan Delanoy Apr 24 '17 at 9:50
  • $\begingroup$ @EwanDelanoy: For what it's worth, there are popular books that use $AB$; Sheldon Ross's books, for instance. But I don't much like it either. $\endgroup$ – Nate Eldredge Apr 26 '17 at 15:57
  • $\begingroup$ @NateEldredge Indeed, $AB$ is already used to mean the product of two random variables $A$ and $B$. $\endgroup$ – Ewan Delanoy Apr 26 '17 at 16:29

Everything you said is OK, you just need a few complements and simplifications.

To justify the existence of a maximal disjoint set of atoms, you can use Zorn's lemma : let $P$ be the poset whose elements are the sets of disjoint atoms, and whose order is defined by inclusion. Then every chain in $P$ has an upper bound in $P$ (the union of the chain), so the hypothesis of Zorn's lemma is satisfied.

So there is a maximal disjoint set of atoms by Zorn's lemma, the set is necessarily countable as explained in the OP, call it $(A_n)_{n\in{\mathbb N}}$ and let $X=\Omega \setminus \bigcup_{n\in{\mathbb N}}A_n$. Then $X$ is measurable, and atomless as explained in the OP.

Now, let $p\in (0,{\mathbb{P}}(X))$. Let $${\mathscr{A}}_{p,X}=\lbrace U\in {\mathscr{A}} | U \subseteq X, {\mathbb{P}}(U) \leq p \rbrace$$.

Say that two elements $U_1$ and $U_2$ in ${\mathscr{A}}_{p,X}$ are linked iff $U_1\subseteq U_2$. I call $L=U_2\setminus U_1$ the link between $U_1$ and $U_2$. If $L$ is either empty or has positive measure, we say that $U_1$ and $U_2$ are strongly linked, and denote this by $U_1 \leq_{p,X} U_2$.

It is easily checked that $\leq_{p,X}$ defines a partial ordering on ${\mathscr{A}}_{p,X}$. Also, if $C$ is any $\leq_{p,X}$-chain in ${\mathscr{A}}_{p,X}$, since the links between two (distinct) successive elements are all disjoint, $C$ is necessarily countable. The union of $C$ will be in ${\mathscr{A}}_{p,X}$, so it will be an upper bound for $C$. By Zorn's lemma, there is a $\leq_{p,X}$-maximal element, denote it by $M$. We know that $\delta=p-{\mathbb{P}}(M)$ is nonnegative. If $\delta>0$, by the property called "Edit 2" in the OP, there is a $V_1\subseteq X\setminus U$ with $0<{\mathbb{P}}(V_1)<\delta$, but then the set $M'=M\cup V_1$ would contradict the maximality of $M$, so $\delta=0$ and hence $P(M)=p$, which finishes the proof.

  • $\begingroup$ I'm confused about the definition of linked elements. If $X \in L$, that is $X \in U_{2} \setminus U_{1}$, then $X \in U_{2}$, therefore $X \in V_{2}$. But since $U_{2} \setminus U_{1} = V_{1} \setminus V_{2}$, then $X \in V_{1} \setminus V_{2}$, therefore $X \notin V_{2}$, which is a contradiction. So by this if I'm not mistaken $L$ can't be anything other than $\emptyset$. Did you mean to write $L = U_{1} \setminus U_{2}$ or is the relation meant to be trivial in this way? $\endgroup$ – Matija Sreckovic Apr 26 '17 at 19:55
  • $\begingroup$ @MatijaSreckovic Answer to your first question : my initial answer was indeed incorrect as stated. Please check my corrected version. Answer to your 2nd question ; it is not true than $P(L)$ is always $0$. Consider for example, $\Omega=[0,1]$, $({\mathscr A},{\mathbb P})=$ Lebesgue measure, $p=\frac{2}{3},U_1=[0,\frac{1}{3}], U_2=[0,\frac{2}{3}],$. Then $L$ has probability $\frac{1}{3}$ not zero. $\endgroup$ – Ewan Delanoy Apr 27 '17 at 7:49
  • $\begingroup$ The second comment was meant in accordance with your first definition, but yeah, it's pointless with the fixed definition of linked sets. $\endgroup$ – Matija Sreckovic Apr 27 '17 at 19:25

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.