Uniqueness of a topological Furstenberg system

The use of topological dynamics to obtain results of a Ramsey-theoretic nature was first implemented by Furstenberg and Weiss in [12], where they provided a dynamical proof of the multidimensional van der Waerden theorem. This was achieved using a topological variant of Furstenberg’s correspondence principle—originally introduced by Furstenberg in his seminal ergodic proof of Szemerédi’s theorem [10]. In this way, they reformulated van der Waerden’s theorem as a statement about multiple recurrence in topological dynamics, which they subsequently proved. Since then, variations of this approach have been successfully adapted to prove a wide range of combinatorial results, including the celebrated polynomial van der Waerden theorem of Bergelson and Leibman [2], as well as the breakthrough result of Moreira [27] on the partition regularity of $\{x,x+y,xy\}$ in the integers.

The essence of this topological correspondence principle is that, given a bounded function $f:{\mathbb{N}}\to{\mathbb{C}}$, the statistical properties of the sequence $(f(n))_{n\in{\mathbb{N}}}$ can be encoded via a topological dynamical system $(X,T)$. In the setting of ergodic theory, the measurable version of Furstenberg’s correspondence, associates to a bounded function $f:{\mathbb{N}}\to{\mathbb{C}}$ some measure preserving system $(X,\mathcal{B},\mu,T)$, provided that some (uniform) Cesáro averages of the sequence $(f(n))_{n\in{\mathbb{N}}}$ exist.

These notions can be extended to more general semigroup settings. Given a semigroup $G$, a topological dynamical $G$-system $(X,(T_{g})_{g\in G})$ is a compact space $X$ together with an action of $G$ on $X$ by continuous transformations, meaning that $T_{g}:X\to X$ is continuous and $T_{gh}=T_{g}T_{h}$ for any $g,h\in G$. We sometimes refer to a topological dynamical $G$-system as simply a system, and, for brevity, we sometimes refer to the system $(X,(T_{g})_{g\in G})$ by $(X,(T_{g}))$ or by $\mathbb{X}$. Note that, when $G$ is a group, the action is automatically by homeomorphsims as $T_{g^{-1}}=T_{g}^{-1}$, for any $g\in G$, and for the identity element $e$ of $G$, $T_{e}$ is the identity map. Throughout this paper, we consider semigroups to which we have added an artificial identity element, if necessary (these objects are also known as monoids). Moreover, all our semigroups are endowed with the discrete topology.

A point $a\in X$ is called transitive if $\{T_{g}a:g\in G\}$ is dense in $X$. A system that admits a transitive point is called transitive and a system with all its points being transitive is called minimal.

For the precise definition of measurable Furstenberg systems we refer the reader to [3, Definition 1.2]. When it is clear from the context, we will omit the term “topological” and refer to a Furstenberg system of $f$.

For a given semigroup $G$ and a bounded function $f:G\to\mathbb{C}$, one can always construct a symbolic Furstenberg system associated to $f$. Indeed, this is the original construction of Furstenberg and Weiss in [12] and it motivates the more abstract Definition 1.1.

To construct this system, we let $K:=\overline{f(G)}$, which is compact by assumption, and consider the space $X:=K^{G}$ endowed with the product topology. Define an action $R_{g}:K^{G}\to K^{G}$ by

$$R_{g}\big((x(u))_{u\in G}\big):=(x(ug))_{u\in G},\quad g\in G.$$

It is straightforward to verify that $R_{gh}=R_{g}R_{h}$ for all $g,h\in G$, so that the right shift $(R_{g})_{g\in G}$ defines an action of $G$ on $K^{G}$. We identify $f$ with the point $(f(g))_{g\in G}\in K^{G}$, and define

$$X_{R,f}:=\overline{\{R_{h}f:h\in G\}}=\overline{\{(f(gh))_{g\in G}:h\in G\}},$$

where the closure is taken in $K^{G}$. Setting $F(x)=x(e)$, for $x\in X_{R,f}$, condition (i) is immediately satisfied, and condition (ii) clearly holds. Hence, $(X_{R,f},(R_{g}))$ is a Furstenberg system of $f$.

To determine when two topological systems are essentially equivalent, we use the notion of topological isomorphism (or topological conjugacy). Before stating our main results, we recall some standard terminology. A factor map between two topological $G$-systems $(X,(T_{g}))$ and $(Y,(S_{g}))$ is a continuous surjection $\phi:X\to Y$ satisfying $\phi\circ T_{g}=S_{g}\circ\phi$, meaning that $\phi(T_{g}(x))=S_{g}(\phi(x))$, for all $x\in X$ and $g\in G$. In this case, we call $(Y,(S_{g}))$ a factor of $(X,(T_{g}))$ or the latter an extension of the former.

A topological isomorphism between two topological $G$-systems $(X,(T_{g}))$ and $(Y,(S_{g}))$ is a homeomorpshim $\phi:X\to Y$ satisfying $\phi\circ T_{g}=S_{g}\circ\phi$, for all $g\in G$. Note that, $\phi$ is a factor map from $(X,(T_{g}))$ to $(Y,(S_{g}))$, but it also follows that $\phi^{-1}\circ S_{g}=T_{g}\circ\phi^{-1}$, and in particular, $\phi^{-1}$ is a factor map from $(Y,(S_{g}))$ to $(X,(T_{g}))$ as well. In this case, we say that $(Y,(S_{g}))$ and $(X,(T_{g}))$ are topologically isomorphic.

We call $(Y,(S_{g}))$ a subsystem of $(X,(T_{g}))$ if $Y$ is a closed invariant subset of $X$, i.e. $T_{g}Y\subset Y$, and $(S_{g})=(T_{g}|_{Y})$, for all $g\in G$. Observe that, a system $(X,(T_{g}))$ is minimal if and only if it has no non-trivial subsystems (i.e. other than $\mathbb{X}$ itself), as orbit closures are $(T_{g})$-invariant and thus give rise to subsystems.

The construction of the maximal system referred to in the previous result is given in Section 2.2, and it naturally involves the Stone-Čech compactification of the underlying semigroup. Section 2 also includes some basic, necessary facts about ultrafilters, other constructions and basic examples of topological Furstenberg systems.

Theorem 1.2 is proved in Section 3, where we also establish its analogue for joint Furstenberg systems of finite families of bounded functions. At the end of that section, we investigate under which conditions on a group $G$ the natural symbolic system constructed via left shifts, denoted by $(\widetilde{X}_{L,f},(L_{g})_{g})$ (see Section 2.1 for the precise construction), is a Furstenberg system of a given bounded function $f:G\to\mathbb{C}$. We address this question in general, although the case of infinite groups is more intricate. In particular, even for finite groups the problem already exhibits interesting features, and we prove the following.

Note that for any semigroup $G$, any set $A\subset G$ can be naturally identified with a function $f_{A}:G\to\{0,1\}$ or a point $a\in\{0,1\}^{G}$. Therefore, we can refer interchangeably to the Furstenberg system of $A$, $f_{A}$ or $a$, and the first question that emerges is when the respective Furstenberg systems of different subsets are isomorphic. We say that two sets $A,B\subset G$ are isomorphic if their Furstenberg systems are topologically isomorphic.

In this direction, we can provide necessary and sufficient conditions for $A\subset G$ and $B\subset G$ to be isomorphic; we refer the reader to Theorem 3.8 for the full result as it requires some technical terminology. As a special case of the main result, we have the following sufficient condition, stated in a somewhat combinatorial flavour.

In view of Theorem 1.2, given a bounded function $f:G\to{\mathbb{C}}$, we may speak of the Furstenberg system of $f$. Of special interest are functions whose Furstenberg systems are minimal. In Section 4.1, we focus on subsets of $G$, which correspond to $\{0,1\}$-valued functions, and show that minimal sets can be realised as a special subclass of sets of visit times of orbits in minimal systems, known as dynamically syndetic sets. This is the context of Theorem 4.5. Moreover, in Theorem 4.8, we provide an analogous characterization for minimal colourings, which correspond to finite-valued functions on $G$ whose Furstenberg systems are minimal. As a consequence, we obtain that if a finite colouring of $G$ is minimal, then each colour is a minimal set.

Finally, in Section 4.2, we relate the minimal sets studied in Section 4.1 to the well-known sets of topological recurrence and provide a characterization of these sets using (minimal) Furstenberg systems.

Acknowledgments. The authors thank their advisor, Joel Moreira, for suggesting this problem and for valuable feedback during the developing of this work. We also thank Rigoberto Zelada for helpful discussions.

Throughout, unless otherwise specified, $G$ denotes a semigroup and $f:G\to{\mathbb{C}}$ a bounded function.

We begin with the classical symbolic Furstenberg system construction from the introduction, adding some details that were glossed over before. We let $K:=\overline{f(G)}$, which is a compact subset of $\mathbb{C}$ by assumption, and endow $K^{G}$ with the product topology. Then, $K^{G}$ is Hausdorff and, by Tychonoff’s theorem, compact.

We consider the full right shift action on $K^{G}$, that is $(K^{G},(R_{g})_{g\in G})$, where

$$R_{g}\bigl((x(u))_{u\in G}\bigr)=(x(ug))_{u\in G}.$$

Note that for any $g,h\in G$, we have that

$$R_{g}(R_{h}\bigl((x(u))_{u\in G}\bigr)=R_{g}(x(uh))_{u\in G}=(x(ugh))_{u\in G}=R_{gh}\bigl((x(u))_{u\in G}\bigr).$$

Therefore $(R_{g})$ forms a $G$-action on $K^{G}$. Recall that the product topology on $K^{G}$ has a basis consisting of cylinder sets, that is, sets of the form

$$C(U_{1},\ldots,U_{k};g_{1},\ldots,g_{k})=\{x=(x(g))_{g\in G}\in K^{G}:x(g_{1})\in U_{1},\ldots,x(g_{k})\in U_{k}\},$$

where $k\in\mathbb{N}$, $g_{1},\ldots,g_{k}\in G$, and $U_{1},\ldots,U_{k}\subset K$ are open. When $G$ is a group, it is clear that $R_{g}$ is a bijection on $K^{G}$, for all $g\in G$, with inverse being $R_{g^{-1}}$. To see that for any semigroup $G$ any of these shifts defines a continuous map, simply note that preimages of cylinder sets under a shift are still cylinder sets. This justifies why $(K^{G},(R_{g})_{g\in G})$ is a dynamical $G$-system.

We identify $f$ with a point in $K^{G}$, still denoted by $f$, and define

(2.1)

$$X_{R,f}=\overline{\Big\{R_{h}(f):\ h\in G\Big\}}=\overline{\Big\{(f(gh))_{g\in G}:\ h\in G\Big\}},$$

where the closure is taken in the full shift $(K^{G},(R_{g})_{g\in G})$. In this system, we consider the continuous function $F:X_{R,f}\to\mathbb{C}$, given by $F(y)=y_{e}$, where $e$ denotes the identity element of $G$. Clearly, the $(R_{g})$-orbit of $F$ separates points according to condition (ii) of Definition 1.1. In addition, the point $f=(f(g))_{g\in G}$ is transitive by construction and $F(R_{h}f)=f(h)$ for every $h\in G$. Therefore, $(X_{R,f},(R_{g})_{g})$ is indeed a Furstenberg system of $f$.

When $G$ is a group, another natural symbolic Furstenberg system of $f$ is given by considering the left shift action, defined by $L_{h}(x(g))_{g\in G}=(x(h^{-1}g))_{g\in G}$. In this case, we define $\hat{f}=(f(g^{-1}))_{g\in G}$ and we consider the space

(2.2)

$$X_{L,f}=\overline{\Big\{L_{h}(\hat{f}(g))_{g\in G}:\ h\in G\Big\}}=\overline{\Big\{(f(g^{-1}h))_{g\in G}:\ h\in G\Big\}},$$

and the orbit closure is taken in the full shift $(K^{G},(L_{g})_{g\in G})$. One can readily check that this defines a Furstenberg system of $f$ via the continuous function $F:X_{L,f}\to\mathbb{C}$ given by $F(y)=y_{e}$. Indeed, $(L_{g})$ defines a $G$-action on $X_{L,f}$, the point $\hat{f}$ is transitive for $(X_{L,f},(L_{g}))$ and for any $h\in G$ we see that

$$F(L_{h}\hat{f})=F((\hat{f}(h^{-1}g)_{g\in G}))=\hat{f}(h^{-1})=f(h).$$

What is perhaps less clear is whether the seemingly more natural system with phase space

(2.3)

$$\tilde{X}_{L,f}=\overline{\Big\{L_{h}(f(g))_{g\in G}:\ h\in G\Big\}}=\overline{\Big\{(f(h^{-1}g))_{g\in G}:\ h\in G\Big\}},$$

also defines a Furstenberg system of $f$, via some continuous function $F$ on $\tilde{X}_{L,f}$ (not necessarily the projection onto the coordinate of the identity) that witnesses the dynamics of $f$ along the orbit of $(f(g))$, or some other $(L_{g})$-transitive point in $\tilde{X}_{L,f}$. It turns out that, for most groups, we can find a function for which this system does not define a Furstenberg system. We explore this question in Section 3.4.

By considering the symbolic right shift system described above, it follows that in any semigroup $G$, any subset $A\subset G$ can be realised as

$$A=\{g\in G:T_{g}a\in U\},$$

where $(X,(T_{g}))$ is a Furstenberg system of $A$ (or, equivalently, of $f_{A}:G\to\{0,1\}$ defined via $f(g)=1\iff g\in A$), $a\in X$ is a $(T_{g})$- transitive point and $U\subset X$ is a non-empty open set.

Indeed, any transitive $G$-system $(X,(T_{g}))$ that admits a continuous function whose orbit separates points (as in condition (ii) of Definition 1.1) is a Furstenberg system of some bounded function $f:G\to{\mathbb{C}}$. Given a transitive point $a\in X$ and a bounded continuous function $F\in C(X)$ whose orbit separates points, we can simply consider $f:G\to{\mathbb{C}}$ given by $f(g)=F(T_{g}a)$, for each $g\in G$. If the system is not transitive but admits such a function, then passing to an orbit closure yields a subsystem that is a Furstenberg system of $f:G\to\mathbb{C}$. In that sense, any system $(X,(T_{g}))$ is a wasteful and generalised Furstenberg system of some bounded function $f:G\to{\mathbb{C}}$.

If $f:{\mathbb{N}}\to\{1,\ldots,r\}$ is a periodic function, say $f(n)=n\pmod{r}$, then its Furstenberg system is isomorphic to the rotation $({\mathbb{T}},R_{1/r})$.

If $f:G\to\{1,\ldots,r\}$ is a random colouring, or more precisely, a sequence of independent and uniformly distributed random variables, namely $\mathbb{P}(\{f(g)=i\})=1/r$, then almost surely, the Furstenberg system of $f$ is the full shift on $r$-symbols, $(\{1,\ldots,r\}^{G},(R_{g}))$.

Furstenberg systems of multiplicative functions have lead to interesting connections between ergodic theory/topological dynamics and number theory. For example, the well-known Chowla conjecture can be reformulated as saying that the topological Furstenberg system of the Liouville function $\lambda:{\mathbb{N}}\to\{-1,1\}$ is a Bernoulli shift $(\{-1,1\}^{\mathbb{N}},\sigma)$. For many deep results, fascinating connections with number theory, and interesting examples of diverse behaviours of sequences, through Furstenberg systems we refer the reader to [6, 7, 8, 9, 17, 23].

One may also consider dynamical systems $(X,(T_{g}))$ with anti-actions of the acting semigroup $G$ by continuous transformations, by requiring that $T_{gh}=T_{h}T_{g}$, for all $g,h\in G$. Observe that, when $G$ is a group, an anti-action of $G$ is simply an action of $G^{-1}=\{g^{-1}:g\in G\}$, defined by $(T^{\prime}_{g^{-1}})_{g\in G}=(T_{g})_{g\in G}$. Of course, when $G$ is abelian any $G$ action is also an anti-action and vice versa. Replacing actions with anti- actions in Definition 1.1 we may speak of the Furstenberg anti-stystem of a function and we remark that all the results we present for actions, can be carried out for anti-actions as well. For completeness we include without proof the following simple observation.

Recall that a compactification of a topological space $X$ is a compact Hausdorff space $Y$ that contains $X$ as a dense subspace. Discrete spaces are completely regular, and so we can consider their Stone-Čech compactification, which is essentially the maximal one. More precisely, given a discrete semigroup $G$, we denote by $\beta G$ its Stone-Čech compactification, which is a compactification with the universal property that any bounded (continuous) function $f:G\to{\mathbb{C}}$ (indeed, any function $f:G\to K$, where $K$ is compact Hausdorff) extends uniquely to a continuous function $\beta f:\beta G\to{\mathbb{C}}$. In fact, there exists a continuous function $i_{G}:G\to\beta G$ such that $f=\beta f\circ i_{G}$.

In this setting, $\beta G$ can be described as the space of ultrafilters on $G$, endowed with the topology whose basis consists of the sets $\overline{A}=\{p\in\beta G:A\in p\}$ for $A\subset G$, making it compact Hausdorff.

Moreover, $G$ can be embedded densely into $\beta G$ by identifying each $g\in G$ with the principal ultrafilter at $g$, which we also denote by $g$, defined via $g=\{A\subset G:g\in A\}.$ Under this identification, $i_{G}(g)=g$ for all $g\in G$.

The operation of the semigroup $G$ naturally extends to $\beta G$, which then becomes a right topological semigroup, $(\beta G,\cdot)$. For completeness, if $p,q\in\beta G$, then

$$p\cdot q=\{A\subset G:\{g\in G:\{h\in G:gh\in A\}\in q\}\in p\}.$$

Given $p\in\beta G$ fixed we denote by $\lambda_{p}:\beta G\to\beta G$, the map $\sigma_{p}:q\mapsto q\cdot p$, which is continuous. We will further need to consider limits along ultrafilters which are defined as follows. Let $q\in\beta G$ and $f:G\to{\mathbb{C}}$ be a bounded function in $G$. Then,

$$q-\lim_{g\in G}f(g)=x$$

if for any open neighborhood $U$ of $x$, it holds that $\{g\in G:f(g)\in U\}\in q$. It is well-known that for functions taking values on compact Hausdorff spaces, these limits always exist and are unique (but not independent of the ultrafilter $q$). Moreover, as we explain in the proof of Theorem 3.2, it actually holds that $\beta f(q)=q-\lim_{g\in G}f(g)$, for any $q\in\beta G$.

Now, let $f:G\to{\mathbb{C}}$ be a bounded function. We proceed to construct the maximal generalised Furstenberg system of $f$. This choice of terminology is justified by Theorem 1.2, as this system is maximal with respect to the partial ordering induced by topological factor maps.

Consider the continuous function $\beta f:\beta G\to{\mathbb{C}}$ described above. Note that, for any $g\in G$, it holds that $\sigma_{g}\beta f(e)=\beta f(g)=f(g)$, where $g$ also denotes the principal ultrafilter at $g$ and $e$ denotes the principal ultrafilter at the identity element, $e$, of $G$. Since $G$ is densely embedded in $\beta G$ it follows that $\{\sigma_{g}(e):g\in G\}$ is dense in $\beta G$. Therefore, $(\beta G,(\sigma_{g}))$ is a dynamical $G$-system with $e$ being a transitive point and $\sigma_{g}\beta f(e)=f(g)$, for any $g\in G$. By definition, $(\beta G,(\sigma_{g}))$ is a generalised Furstenberg system of $f$.

We also explain an alternative way to essentially recover the symbolic Furstenberg system of $f$, this time with phase space in ${\mathbb{C}}^{\beta G}$. Again, we consider the continuous extension $\beta f:\beta G\to{\mathbb{C}}$ and the dynamical $G$-system $(\beta G_{f},(\sigma_{g}))$, where $\beta G_{f}$ is the orbit closure of $\{\sigma_{g}(\beta f):g\in G\}$. Note also that $\beta G_{f}\subset C(\beta G)$, because we endow $C(\beta G)$ with the product topology, which is the topology of pointwise convergence and therefore,

$$q-\lim_{g\in G}\sigma_{g}\beta f(p)=p-\lim_{g\in G}q-\lim_{h\in G}\sigma_{gh}\beta f=p\cdot q-\lim_{g\in G}\sigma_{g}\beta f=\beta f(p\cdot q)=\beta f\circ\sigma_{q}(p),$$

for each $p,q\in\beta G$ and $\beta f\circ\sigma_{q}$ is continuous. Let $F\in C(\beta G_{f})$ be the evaluation at $e$, the principal ultrafilter at the identity element of $G$, namely $F(G)=G(e)$, for $G\in\beta G_{f}$. The $(\sigma_{g})$-orbit of $F$ separates points because $G$ is densely embedded in $\beta G$ and hence, any two continuous functions (in $\beta G_{f}$) that agree on $G$ will be identical. Then, notice that $F(\sigma_{g}(\beta f))=\beta f(g)=f(g)$, for any $g\in G$. Thus, indeed, $(\beta G_{f},(\sigma_{g}))$ is a Furstenberg system of $f$.

In this section we will prove one of our main results, Theorem 1.2. Throughout, we will consider semigroups $G$ and bounded functions $f:G\to{\mathbb{C}}$. However, we invite the reader to notice that all we will demand from $f(G)$ (for the proofs to be carried out) is that it is contained in a compact Hausdorff space.

We begin by observing that any system that is isomorphic to a Furstenberg system of some function will also be its Furstenberg system. More precisely, if $\mathbb{X}=(X,(T_{g}))$ is a (generalised/wasteful) Furstenberg system of $f$ and $\mathbb{Y}=(Y,(S_{g}))$ is topologically isomorphic to $\mathbb{X}$, then $\mathbb{Y}$ is also a (generalised/wasteful) Furstenberg system of $f$. We leave the easy details of this fact to the interested reader.

We proceed to prove the first part of Theorem 1.2, namely that Furstenberg systems associated to the same function are isomorphic.

Having established that Furstenberg systems are isomorphic, we may speak of the Furstenberg system associated with a given function $f\colon G\to{\mathbb{C}}$, choosing a representative among these isomorphic systems.

An almost obvious consequence of this is that the Furstenberg system of $f:G\to{\mathbb{C}}$ is a subsystem of any wasteful Furstenberg system of it. Indeed, if $(X,(T_{g})_{g})$ is a wasteful Furstenberg system, where $F(T_{g}a)=f(g)$ for some continuous function $F$ and some $a\in X$, we can recover the Furstenberg system of $f$, by restricting to the subsystem $X^{\prime}=\overline{\{T_{g}a:g\in G\}}$. On the other hand, we can trivially add isolated points to the Furstenberg system of a function $f$, and recover a wasteful system, making it obviously non-unique.

The situation becomes more interesting when considering generalised Furstenberg systems, namely when the function $F$ witnessing the behaviour of $f$ is not required to separate distinct orbits. We show that the Furstenberg system can be recovered as a factor of any generalised system. Moreover, there exists a maximal Furstenberg system which contains all generalised Furstenberg systems as its factors. This is the the content of the second part of Theorem 1.2.

We emphasise that generalised Furstenberg systems associated with the same function need not be isomorphic. In particular, there exist functions with a generalised Furstenberg system that is genuinely distinct from (i.e. not isomorphic to) their Furstenberg system and from their maximal Furstenberg system. Indeed, if $f$ is periodic and $G=\mathbb{Z}$, the associated Furstenberg system is finite. A generalised Furstenberg system can then be obtained by taking the direct product of the (unique) Furstenberg system with an irrational rotation, yielding an infinite system. In contrast, the maximal system is non-metrizable, and therefore these notions cannot coincide. The following diagram illustrates how the maximal system controls, through factor maps, all Furstenberg systems, while the (generalised) Furstenberg system depends on the specific function.

In analogy with Definition 1.1, in order to find a dynamical model for the statistics of multiple functions simultanesouly, we can consider the joint Furstenberg system of a finite set of bounded functions over a semigroup.

Note that, in Definition 3.3, we may consider a multi-valued function $F:X\to{\mathbb{C}}^{\ell}$ defined via $F(x)=(F_{1}(x),\ldots,F_{\ell}(x))$, and rewrite condition (ii) as the property that the $G$-orbit of $F$ separates points.

As the prototypical example, we consider the symbolic right shift joint Furstenberg system of a finite family of functions. If $G$ is a semigroup, and $f_{1},\ldots,f_{\ell}:G\to{\mathbb{C}}$ are bounded, then we consider $f\in({\mathbb{C}}^{\ell})^{G}$ defined via $f(g)=(f_{1}(g),\ldots,f_{\ell}(g))$ and in analogy with the symbolic system constructed in Section 2.1, we consider $(X_{R,f},(R_{g}))$ to be the orbit closure of $f$ under $(R_{g})$, only this time, the phase space is a subspace of $({\mathbb{C}}^{\ell})^{G}$. With the continuous function $F:X_{R,f}\to C^{\ell}$ defined by $F(x)=(x_{1}(0),\ldots,x_{\ell}(0))$, we recover a joint Furstenberg system of $f_{1},\ldots,f_{\ell}$. In an analogous manner, the maximal joint Furstenberg system of $f_{1},\ldots,f_{\ell}$ can be defined using the Stone-Čech compactification of $G$, $(\beta G,(\sigma_{g}))$.

As we drew attention to in the beginning of this section, nothing special about the structure of the image of single functions $f:G\to{\mathbb{C}}$ was used in the proofs of the Section 3.1, other than the fact that ${\mathbb{C}}$ is Hausdorff and that $\overline{f(G)}$ is compact, since $f$ was assumed bounded. The same assumptions hold when considering $f=(f_{1},\ldots,f_{\ell})$. Namely, $f(G)\subset{\mathbb{C}}^{\ell}$ is a pre-compact set of a Hausdorff space, and thus all the results proved for a single function extend in the case of finite tuples of functions.