If I have i.i.d. r.v.'s $X_1, X_2,...$ then the tail sigma algebra is defined as $\mathcal{T}:=\cap_n\sigma_n$ where $\sigma_n:=\sigma(X_n,X_{n+1},...)$. From this we get very nice results such as the Kolmogorov 0-1 law. I was wondering if it makes sense to consider limsup and liminfs in this fashion: $\bigcup_n\bigcap_{k\ge n}\sigma_k$ and $\bigcap_n\bigcup_{k\ge n}\sigma_k$. Do these have a sensible meaning and are there similar laws such as the 0-1 law pertaining to each?


Let us define $\mathcal{T}_n=\bigcap_{k=n}^\infty\sigma_k$, so that the tail $\sigma$-algebra is just $\mathcal{T}=\mathcal{T}_1$.

Now, the first thing to notice is that $\sigma_{n+1}\subset\sigma_n$ for all $n$. This means that $\mathcal{T}_n=\bigcap_{k=n}^\infty\sigma_k$ does not actually depend on $n$. So $\mathcal{T}_n$=$\mathcal{T}$ for all $n$. Therefore, $$ \bigcup_{n=1}^\infty\bigcap_{k=n}^\infty\sigma_k = \bigcup_{n=1}^\infty \mathcal{T} = \mathcal{T}. $$

On the other hand, $\sigma_{n+1}\subset\sigma_n$ implies that $\bigcup_{k=n}^\infty \sigma_k=\sigma_n$. Therefore, $$ \bigcap_{n=1}^\infty\bigcup_{k=n}^\infty\sigma_k = \bigcap_{n=1}^\infty \sigma_n = \mathcal{T}. $$

So both of these suggested generalizations are, in fact, just the tail $\sigma$-algebra, $\mathcal{T}$.


Here is some additional information. As you may know, the intersection of $\sigma$-algebras is a $\sigma$-algebra, but the union of $\sigma$-algebras is not, in general. It is therefore common to adopt the following notation. If $\{\mathcal{F}_i\}_{i\in I}$ is a collection of $\sigma$-algebras, then $$ \bigvee_{i\in I}\mathcal{F}_i = \sigma\bigg(\bigcup_{i\in I}\mathcal{F}_i\bigg). $$ With this notation, we may write $$ \mathcal{T} = \bigcap_{n=1}^\infty\bigvee_{k=n}^\infty\sigma(X_k). $$ So in this sense, the tail $\sigma$-algebra is already a limsup. We might then wonder about the $\sigma$-algebra $$ \mathcal{S} = \bigvee_{n=1}^\infty\bigcap_{k=n}^\infty\sigma(X_k). $$ It can be shown that $\mathcal{S}\subset\mathcal{T}$, so that Kolmogorov's $0$-$1$ law is still valid on $\mathcal{S}$. To see this, suppose $A\in\bigcup_n\bigcap_{k\ge n}\sigma(X_k)$. Then there exists $n$ such that $A\in\sigma(X_k)\subset\sigma_k$ for all $k\ge n$. This implies $A\in\mathcal{T}$ and shows that $\bigcup_n\bigcap_{k\ge n}\sigma(X_k)\subset\mathcal{T}$. Since $\mathcal{T}$ is a $\sigma$-algebra, it follows that $$ \mathcal{S} = \sigma\bigg(\bigcup_n\bigcap_{k\ge n}\sigma(X_k)\bigg) \subset\mathcal{T}. $$ The reverse inclusion does not necessarily hold. For an example, let $P$ be any Borel probability measure on $\mathbb{R}$ such that $P(\mathbb{Q})=0$. For $n\in\mathbb{N}$, let $X_{2n}=0$ and $X_{2n+1}=1_{\mathbb{Q}}$. Then $X_j=0$ a.s. for every $j$, so this is indeed an iid sequence. (In general, if $U$ is constant a.s. and $V$ is any random variable, then $U$ and $V$ are independent.)

In this example, $\sigma(X_{2n})=\{\emptyset,\mathbb{R}\}$ and $\sigma(X_{2n+1})=\{\emptyset,\mathbb{Q},\mathbb{Q}^c,\mathbb{R}\}$. Hence, $\mathcal{S}=\{\emptyset,\mathbb{R}\}$ and $\mathcal{T}=\{\emptyset,\mathbb{Q},\mathbb{Q}^c,\mathbb{R}\}$.


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.