# On Set Convergence I

We give the definitions of inner and outer limits for sequences of sets in topological and normed spaces and we provide some important facts on set convergence on topological and normed spaces. We juxtapose the notions of the limit superior and limit inferior for sequences of sets and we outline some facts regarding the Painlevé-Kuratowski convergence of set sequences.

This article will be in three parts. Here we provide some basic definitions and facts which we shall later use to provide a topological characterisation of set convergence and study some interesting properties of the Painlevé-Kuratowski convergence.

In what follows we always consider $\mathcal{X}$ to be a set endowed with a Hausdorff topology $\tau$ which we will denote by $\left( \mathcal{X},\tau\right)$. The topology of the space defines the class of open neighborhoods of points in $\mathcal{X}$:

Definition 1 (Open neighbourhoods). Let $\left(\mathcal{X},\tau\right)$ be a topological space and $x\in\mathcal{X}$. The set of open neighbourhoods of $x$ is defined as:

$\mho\left(x\right):=\left\{V\in\tau|x\in V\right\}$

The topology of $\mathcal{X}$ governs the convergence of sequences of elements in this space.

Definition 2 (Convergence wrt a topology). A sequence $\{x_n\}_{n}\subseteq\mathcal{X}$ is said to converge to some $x\in\mathcal{X}$ with respect to the topology $\tau$ if for every $V\in\mho\left(x\right)$ there is a $N_0\in\mathbb{N}$ such that $x_k\in V$ for all $k\geq N_0$.

We now introduce the notions of cofinal and cofinite sets in partially ordered spaces.

Definition 3 (Cofinal). Let $(\Lambda,\leq)$ be a directed set (i.e. $\leq$ is a preorder). Then the set $\Sigma\subseteq \Lambda$ is called a cofinal subset of $\Lambda$ if for all $\lambda\in\Lambda$ there exists a $\sigma\in\Sigma$ such that $\lambda \leq \sigma$. We denote the cofinal subsets of $\mathbb{N}$ by $\mathcal{N}_\infty^\#$.

Cofinal subsets of $\mathbb{N}$ look like the sets of indices of subsequences: are those subsets of $\mathbb{N}$ with infinitely many points.

We may now prove the following:

Proposition 1. $\mathcal{N}_\infty^\# := \left\{ N\subseteq \mathbb{N},\ N \text{ is infinite}\right\}$

Proof. (1). Let $\Sigma$ be an infinite subset of $\mathbb{N}$ and $n\in\mathbb{N}$ arbitrary. The set $\mathbb{N}_n=\left\{ m\in\mathbb{N},\ m\leq n\right\}$ is finite, so it cannot be $\Sigma\subseteq \mathbb{N}_n$, therefore there is a $\sigma\in\Sigma$ such that $n \leq \sigma$. Thus, $\Sigma$ is cofinal.

(2). Assume that $\Sigma$ is a cofinal subset of $\mathbb{N}$. Let us assume that $\Sigma$ is finite. Then, $\Sigma$ has a maximal element, say $N$. For every $\sigma\in\Sigma$, $N+1\nleq \sigma$. Hence $\Sigma$ is not cofinal. This contradicts our assumption, therefore $\Sigma$ is infinite. ♣

We will now give the following definition:

Definition 4 (Cofinite subset). Let $\Lambda$ be any set. A set $\Phi\subseteq \Lambda$ is called a cofinite subset of $\Lambda$  if $\Phi^c=\Lambda\setminus \Phi$ is finite. Hereinafter, we shall denote the class of  cofinite subsets of $\mathbb{N}$ by $\mathcal{N}_\infty$.

The cofinite subset sets are the ones which eventually look like $\mathbb{N}$ it self, that is, they can written as $\Phi=\Phi_0\cup \{\kappa, \kappa+1, \kappa+2, \ldots\}$, where $\Phi_0$ is some subset of $\mathbb{N}.$

We may now give the definition of the inner limit of a sequence of sets

Definition 5 (Inner limit). Let $\{C_n\}_{n}$ be a sequence of sets in a Hausdorff topological space $\left(\mathcal{X},\tau\right)$. The inner limit of this sequence is defined as:

$\liminf_n C_n = \left\{x \left| {\begin{array}{c} \exists N \in \mathcal{N}_\infty,\ \exists x_v \in C_v\\ v\in N,\ x_v \to x \end{array} } \right. \right\}$

Where the convergence $x_v \to x$ is meant with respect to the topology $\tau$.

Accordingly, the outer limit of a sequence of sets is defined as:

Definition 6 (Outer limit). Let $\{C_n\}_{n}$ be a sequence of sets in a Hausdorff topological space $\left(\mathcal{X},\tau\right)$. The inner limit of this sequence is defined as:

$\limsup_n C_n = \left\{ x \left| {\begin{array}{c} \exists N \in \mathcal{N}_\infty^\#,\ \exists x_v \in C_v\\ v\in N,\ x_v \to x \end{array} } \right. \right\}$

where, again, the convergence $x_v \to x$ is meant with respect to the topology $\tau$.

Figure 1. The inner and the outer limits of a constant sequence of sets – each element of which is an open set U – is the closure of U.

If $\mathcal{X}$ is a normed space then specific conclusions can be drawn exploiting the well known properties of the norm and  the norm-balls. We introduce the notion of the point-to-set distance mapping.

Definition 7 (Point-to-set distance). The point-to-set distance on $X$ is a mapping $d:X\times 2^X \to [0,+\infty]$ defined as

$d\left(x,C\right):=\inf_y\left\{\|x-y\|;\ y\in C\right\}$

The limit inferior – which should not be confused with the inner limit – and the limit superior – which, again, is not the same as the outer limit – of a sequence of real numbers will be of high importance in what follows. We give the following definition:

Definition 8 (Limit inferior, limit superior). The limit inferior of a sequence $\{a_n\}_{n}\subseteq \mathbb{R}$ is defined as:

$\liminf_n a_n = \lim_{n\to\infty}\left[ \inf_{k\geq n} a_k \right]$

Accordingly, the limit superior of $\{a_n\}_{n}$ is:

$\limsup_n a_n = \lim_{n\to\infty}\left[ \sup_{k\geq n} a_k \right]$

Notice that the above definition does not assume the existence of a topology. The limit inferior of a sequence of elements or subsets of a space $\mathcal{X}$ (which does not need to be endowed with any topology) is given as follows

Definition 9.a (Limit inferior of a seq. of sets). Let $\mathcal{X}$ be a set and $\{A_n\}_{n}$ be a sequence of sets in $\mathcal{X}$. The limit inferior of  this sequence is defined to be the set:

$\mathrm{D{-}liminf}_{n\to\infty} C_n = \left\{ x \left| {\begin{array}{c} \exists N \in \mathcal{N}_\infty,\ x \in C_v\\ \forall v\in N \end{array} }\right. \right\}$

Figure 2. The limit inferior of a sequence of sets.

What is the same, we may define the limit superior of a sequence of sets as:

Definition 9.b (Limit superior of a seq. of sets).  Let $\mathcal{X}$ be a set and $\{A_n\}_{n}$ be a sequence of sets in $\mathcal{X}$. The limit superior of  this sequence is defined to be the set:

$\mathrm{D{-}limsup}_{n\to\infty} C_n = \left\{ x \left| {\begin{array}{c} \exists N \in \mathcal{N}_\infty^\#,\ x \in C_v\\ \forall v\in N \end{array} }\right. \right\}$

FIgure 3. The limit superior of a sequence of sets.

The limit inferior and the limit superior are exactly the inner and the outer limits when the space $\mathcal{X}$ is endowed with the discrete topology, i.e. the topology of the power set of $\mathcal{X}$, $\tau=2^\mathcal{X}$. In all other cases, the inner and outer limit yield quite different results that the limits inferior and superior. In all cases it holds:

$\mathrm{D{-}liminf}_{n\to\infty} C_n \subseteq \liminf_{n\to\infty} C_n$

Consider for example the case of $\mathbb{R}$ with the usual topology and the sequence of sets:

$C_n=\left\{ \begin{array}{c} \mathbb{Q},\ n\text{ is odd}\\ \mathbb{R}\setminus\mathbb{Q},\ n\text{ is even} \end{array} \right.$

Then, $\mathrm{D{-}liminf}_{n\to\infty} C_n = \varnothing$ while $\liminf_n C_n = \mathbb{R}^n$.

A well known property of $\mathrm{D{-}liminf}$ is stated as follows:

Proposition 2. The limit inferior of a sequence of sets is:

$\mathrm{D{-}liminf}_{n\to\infty} C_n = \bigcup_{n=1}^{\infty} \bigcap_{m=n}^{\infty} C_m$

Proof. First, let us define $B_k=\bigcap_{j\geq n}C_j$. Then the right hand side of equation we want to prove is written as $\bigcup_{n\in\mathbb{N}}B_n$

(1). Assume that $x\in C_i$ for all but finitely many indices i. Then, there is a $M\in\mathbb{N}$ so that for all $m\geq M$ it is $x\in C_m$. We notice that if $k\geq M$ then $x\in C_j$ for all $j\geq k$. Therefore,  $x\in\bigcap_{j\geq k}C_j=B_k$

If on the other hand $k\leq M$, then we can find $\hat{k}:=\max\left\{k,M\right\}$ so that $x\in\bigcap_{j\geq \hat{k}}C_j=B_{\hat{k}}$

Thus, for arbitrary index $k\in\mathbb{N}$, there is always a $\hat{k}\in\mathbb{N}$ such that $x\in B_{\hat{k}}$ which means that $x\in\bigcup_{n\in\mathbb{N}}B_n$.

(2). Let us assume that  $x\in \bigcup_{n=1}^\infty B_n,$ but there are infinitely many indices $i$, such that $x\notin C_i$. Let $\{s_j\}_{j}\subseteq \mathbb{N}$ be a strictly increasing sequence such that $x\notin C_{s_j}$ – note that $s_j\geq j$. For any $k\in\mathbb{N}$ we have $x\in C_{s_k}\supseteq \bigcap_{s_j \geq s_k}C_{s_j} \supseteq \bigcap_{j\geq k} C_j=B_k$ which means that $x\notin B_k$. This holds true for all $k\in\mathbb{N}$, thus $x\notin \bigcup_{n=1}^\infty B_n$ which contradicts our initial assumption. This completes the proof. ♣

In a similar fashion we can prove the following fact regarding the limit superior:

Proposition 3. The limit superior of a sequence of sets is:

$\mathrm{D{-}limsup}_{n\to\infty} C_n = \bigcap_{n=1}^{\infty}\bigcup_{m=n}^{\infty} C_m$

The proof of proposition 3 goes along the same lines of proposition 2. We mentioned the notions of the limits superior and inferior which are not topological notions for the sake of completeness. These notions are encountered in measure theory, but are of little use in convex and variational analysis.

We will close this post with an example of a sequence of sets

$C_{\nu} = [\tfrac{1}{\nu}, \tfrac{1}{\nu+1}]$

Then, by definition the limit superior $\mathrm{D{-}limsup}_\nu C_\nu = \varnothing$ because we cannot find a sequence of integers $N\subseteq \mathbb{N}$ so that $x_{\nu} \in C_{\nu}$ and also – to avoid any misunderstanding – notice that $0\notin C_{\nu}$ for all $\nu\in \mathbb{N}$.

Figure 4. Sequence of sets which converges to the empty set in the discrete topology (using the limit superior), but converges to {0} using the definitions of inner and outer limits.

However, this sequence converges to $\{0\}$; we can see that by taking an element $x_n\in C_n$, e.g,. the centre of each set and notice that $x_{\nu}\to 0$.

### 2 responses

1. […] post comes as a sequel of On Set Convergence I where we introduced some necessary notions that are useful for studying how sets converge in a […]

Like

2. […] elements of that sequence and many other interesting properties. * Read first the previous posts: On Set Convergence I and […]

Like

mathbabe

Exploring and venting about quantitative issues

Look at the corners!

The math blog of Dmitry Ostrovsky

The Unapologetic Mathematician

Mathematics for the interested outsider

Almost Sure

A random mathematical blog

Mathematix

Mathematix is the mathematician of the village of Asterix