Professional Documents
Culture Documents
UNIQUENESS OF HYPERSPACES
ALEJANDRO ILLANES
Abstract. In general topology, given a space X there are several ways to con-
struct a new space K(X) from X. Examples of this situation are: products,
spaces of maps to the real line, group of homeomorphisms, hyperspaces, etc. A
natural question is if we can recover the space X when we know the space K(X).
In order to reach this aim, the space K(X) must be unique. That is, it must come
only from one X. In this paper we survey what has been done on this direction
when the structure K(X) is one of the hyperspaces 2X , Cn (X) or Fn (X).
1. Introduction
A continuum is a nondegenerate compact connected metric space. Given a con-
tinuum X, with metric d, we consider the following hyperspaces of X.
2X = {A ⊂ X : A is nonempty and closed in X},
Cn (X) = {A ∈ 2X : A has at most n components},
Fn (X) = {A ∈ 2X : A has at most n points},
C(X) = C1 (X) .
All the hyperspaces are considered with the Hausdorff metric H [38, Definition
2.1 and Theorem 2.2] defined as H(A, B) = max{max{d(a, B) : a ∈ A}, max{d(b, A) :
b ∈ B}, where d(a, B) = min{d(a, b) : b ∈ B}.
The hyperspace Fn (X) is known as the n-symmetric product of X. The hy-
perspace F1 (X) is an isometric copy of X embedded in each one of the other
hyperspaces.
Given K(X) ∈ {2X , C(X), Cn (X), Fn (X)}, we say that a continuum X has
unique hyperspace K(X) provided that the following implication holds:
In general topology, given a space X there are several ways to construct a new
space K(X) from X. Examples of this situation are: products, spaces of maps to
the real line, group of homeomorphisms, hyperspaces, etc. A natural question is
if we can recover the space X when we know the space K(X). In order to reach
this aim, the space K(X) must be unique. That is, it must come only from one X.
In this paper we survey what has been done on this direction when the structure
K(X) is one of the hyperspaces 2X , Cn (X) or Fn (X).
As we will see, the work on this topic has received much attention recently and
there are still many questions to solve.
Recall that the Hilbert cube is the topological product of countably many inter-
vals [0, 1]. An arc is a continuum homeomorphic to the interval [0, 1].
In general, many continua can share an specific hyperspace. The most general
result in this direction is given by the following classical theorem.
Theorem 1 ([14] and [13] for the case n ≥ 2). If X is a locally connected con-
tinuum, then 2X is homeomorphic to the Hilbert cube. For a continuum X, the
following are equivalent.
(a) X is locally connected and each arc in X has empty interior,
(b) C(X) is homeomorphic to the Hilbert cube,
(c) Cn (X) is homeomorphic to the Hilbert cube for each n.
We start with the simplest continuum, namely the unit interval [0, 1]. Notice
that
C([0, 1]) = {[a, b] : 0 ≤ a ≤ b ≤ 1} .
It is easy to check that the function ϕ : C([0, 1]) → R2 (R2 is the Euclidean
plane) given by ϕ([a, b]) = (a, b) is a homeomorphism between C([0, 1]) and the
triangle T = {(a, b) ∈ R2 : 0 ≤ a ≤ b ≤ 1}, represented in Figure 1.
Thus, we can say that this triangle is a model for C([0, 1]). Observe that the set
of elements in C([0, 1]) that contain 0 (intervals of the form [0, b]) are represented
on an edge of T ; the elements of C([0, 1]) that contain 1 are represented on other
edge of T . The set of singletons F1 ([0, 1]) is represented on the third edge of T (the
diagonal).
C([0,1])
[0,1]
Figure 1.
Next, we analyze the model for the unit circle S 1 in R2 , centered at the origin.
For each subarc A of S 1 , let m(A) be the middle point of A in S 1 and let L(A) be
the length of A. Then define F : C(S 1 ) → R2 by
{
[1 − (L(A)/2π)]m(A), if A ̸= S 1 ,
F (A) =
(0, 0), if A = S 1 .
It is easy to check that ϕ is a homeomorphism between C(S 1 ) and the unit disc.
Thus, the disc is a model for C(S 1 ).
We have shown that C([0, 1]) and C(S 1 ) are homeomorphic. Thus [0, 1] (and
S 1 ) does not have unique hyperspace C(X).
24 Alejandro Illanes
F(A) = (1-(L(A)/2p))m(A)
Figure 2.
At this moment we can ask: what happens with the rest of the finite graphs?
(a finite graph is a continuum that can be put as a finite union of arcs such that
every two of them intersect in a finite set). To have an idea of how to answer this
question, we construct the model for the simple triod T which is defined as the
union of three arcs L1 , L2 and L3 , called the legs of T , joined by a point v called
the vertex of T (Figure 3). The hyperspace C(T ) is the union of C(L1 ), C(L2 ),
C(L3 ) and Cv (T ) = {A ∈ C(T ) : v ∈ A}. By the first model we constructed, each
set C(Li ) can be represented as a solid triangle. Notice that the elements of Cv (T )
are uniquely determined by the length of their intersections with each leg. So, they
can be represented by a vector with three coordinates (a, b, c).
T
c
a
v b
Figure 3.
Varying the three lengths a, b and c we obtain a solid cube in R3 . Thus C(T ) is
the union of a solid cube in R3 with three solid triangles, as it is pictured in Figure
4.
In this example we can see that a simple triod generates a cube in the hyperspace
C(X). This situation is more general. For example, in the continuum W illustrated
in Figure 5, the subcontinua of W containing the transversal arc J can be enlarged
in four independent movements (a, b, c and d). Thus, in C(W ), each neighborhood
of J is of dimension at least 4.
Uniqueness of hyperspaces 25
C(T)
T
Figure 4.
W c
a
J J d
b
Figure 5.
Figure 6.
This example illustrates the following idea. If we can see the hyperspace C(G),
for a finite graph G, with at least one ramification point, we can look for the
manifold boundary of the 2-dimensional part of C(G) and we will see from what
finite graph G the model was constructed. That is, we can recover G from C(G).
This is the motivation (and the idea of the proof of the following theorem).
Theorem 2. Finite graphs G, different from an arc and a simple closed curve have
unique hyperspace C(G).
Theorem 2 follows essentially from the work by R. Duda [15], it explicitly ap-
peared in [2, Theorem 1]. In his papers [15], [16] and [17], R. Duda made a very
detailed study of the hyperspace C(G) when G is a finite graph.
For a finite graph G, what we have discussed for C(G) is not applicable to
C2 (G). We will see that it is impossible to recognize the hyperspace F1 (G) from
seeing C2 (G). In order to show this we recall how to construct a model for C2 ([0, 1])
(due to R. M. Schori, [29, Lemma 2.2]).
Let C01 = {A ∈ C2 ([0, 1]) : 0, 1 ∈ A} and C1 = {A ∈ C2 ([0, 1]) : 1 ∈ A}. The
typical elements of C01 are of the form A = [0, a] ∪ [b, 1], where 0 ≤ a ≤ b ≤ 1.
We can define ϕ : C01 → R2 by ϕ(A) = (a, b). Then ϕ is not a function since
ϕ([0, 1]) = ϕ([0, a] ∪ [a, 1]) = (a, a) for each a ∈ [0, 1]. The image of ϕ is the
triangle T in Figure 1. If we identify the diagonal ∆ of T to a point we obtain the
space T /∆ and now ϕ is a well defined homeomorphism between C01 and T /∆. This
proves that C01 is a 2-cell. It is easy to show that the function ψ : C01 × [0, 1] → C1
Uniqueness of hyperspaces 27
B. Hilbert Cubes
There are several ways to identify Hilbert cubes in hyperspaces. We illustrate
this fact in some examples.
Consider the continuum Fω in Figure 7 which is the union of a sequence of arcs
converging to a singleton {v}. Let A = {A ∈ Fω : v ∈ A}. Proceeding as we did
with the simple triod, it is possible to show that A is homeomorphic to the Hilbert
cube.
Fw
Figure 7.
Now, consider the continuum X in Figure 8 which is called the Harmonic fan.
It consists of a sequence of arcs converging to an arc, all of them joined by a
point v. In this case it is also possible to show [18, Theorem 2.2] that the set
A = {A ∈ C(X) : v ∈ A} is homeomorphic to the Hilbert cube.
Figure 8.
The next particular continuum that we consider is the enlarged null comb N
illustrated in Figure 9. In this case it is also possible to prove that the set A =
{A ∈ C(N ) : v ∈ A} is a Hilbert cube (see [8, Theorem 3.1]).
A free arc in a continuum X is an arc α, with end points a and b such that
α − {a, b} is open in X.
If a locally connected continuum X has a point p for which there exists a compact
connected neighborhood M in X such that M does not contain free arcs of X, by
Uniqueness of hyperspaces 29
Figure 9.
Theorem 1, C(M ) is a neighborhood of the element {p} in C(X) such that C(M )
is homeomorphic to the Hilbert cube.
M
p
Figure 10.
In some cases, when there is a Hilbert cube as the described in the previous
examples, it is possible to prove that X does not have unique hyperspace C(X).
Next, we will describe an idea of the steps to do this.
1. Locate a special point p and a subcontinuum M of C(X) such that M is
homeomorphic to the Hilbert cube and {p} ∈ M.
2. Attach a locally connected continuum D, without free arcs, to X by the point
p to construct the continuum Y = X ∪ D. Then C(Y ) = C(X) ∪ C(D) ∪ Cp (Y ),
where Cp (Y ) = {A ∈ C(Y ) : p ∈ A}.
By Theorem 1, C(D) is a Hilbert cube. Under appropriate conditions Cp (Y )
is (or contains) a Hilbert cube. In this way M ∪ Cp (Y ) ∪ C(D) is the union of
three Hilbert cubes. The Hilbert cube is particularly flexible, this allows to show
that, in appropriate cases, the union of three Hilbert cubes is homeomorphic to
a Hilbert cube. If we get a homeomorphism that does not move the boundary of
M in C(X), then it is possible to construct a homeomorphism between C(X) and
C(Y ). Since D can be chosen from a very large collection of continua, it is possible
to choose D in such a way that X is not homeomorphic to Y . Thus, X does not
have unique hyperspace C(X), see Figure 12.
30 Alejandro Illanes
M
p
Figure 11.
C(X)
M
C(Y) Cp(Y)
C(D) M
Figure 12.
Uniqueness of hyperspaces 31
A smooth fan is a subcontinuum of the cone over the Cantor set. The continuum
Fω and the harmonic fan are examples of smooth fans. C. Eberhart and S. B.
Nadler, Jr. applied the technique described above to show the following theorem.
Theorem 4 ([18]). If X is a smooth fan with infinitely many end points, then X
does not have unique hyperspace C(X).
Problem 5. Can Theorem 4 be extended to the hyperspaces Cn (X) for all n ≥ 2?
C. Dendrites
A dendrite is a locally connected continuum without simple closed curves. Define
D = {X : X is a dendrite with closed set of end points} .
It is known [10] that a dendrite X belongs to D if and only if X does not contain
neither a copy of Fω nor a copy of the enlarged null comb. Figure 13 shows the so
called Gehman dendrite which is a dendrite in class D. For this dendrite, the set
of end points is the Cantor set.
Figure 13.
Dendrites X in class D have the property that they can be approximated by
finite graphs G such that G = clX (int(G)). So, it is not surprising that some of the
results about uniqueness of hyperspaces proved for finite graphs can be extended
to dendrites in class D. We enunciate some of them.
The following theorem was proved in several steps. The case n = 1 was shown
in [22], the case n > 2 in [26] and the case n = 2 needed the papers [24] and [33].
32 Alejandro Illanes
Theorem 7 ([9] for the case n = 1 and [25] for the case n > 2). Continua with
a base of neighborhoods belonging to class D have unique hyperspace Cn (X) for all
n ̸= 2.
The next result was proved using the technique described in the section B. Since
a dendrite X that does not belong to class D contains a copy of Fω or a copy of the
enlarged null comb, there is special point p where a continuum D can be attached
to X obtaining a continuum Y such that Y is not homeomorphic to X and C(Y )
is homeomorphic to C(X).
Theorem 8 ([8]). If a dendrite X does not belong to D, then X does not have
unique hyperspace C(X).
One can suspect that Theorem 6 can be extended to Cn (X) for n ≥ 2. However,
the technique described in section B does not behave so well for these hyperspaces.
Theorem 6 cannot be extended even for n = 2. The (unexpected) example X,
illustrated in Figure 14 is a dendrite containing the extended null comb and having
unique hyperspace C2 (X) (see [20, Example 39]).
Figure 14.
Problem 9 ([20, Problem 9]). Characterize those dendrites X with unique hyper-
space C2 (X).
D. Framed Continua
Trying to give a unified theory of uniqueness in locally connected continua, in
[20] the following notions were introduced.
Uniqueness of hyperspaces 33
∪
A locally connected continuum X is almost framed provided that {J ⊂ X : J
is a free arc in X} is dense in X. Let
G(X) = {p ∈ X : p has a neighborhood K in X such that K is a finite graph} .
Clearly, a locally connected continuum X is almost framed if and only if G(X)
is dense in X.
We say that X is framed if it is not a simple closed curve, is almost framed and
has a base of neighborhoods B such that for each U ∈ B, U ∩ G(X) is connected.
It is not hard to prove that finite graphs, dendrites in class D and locally class-D
dendrites are framed continua ([20, Theorem 6]). Figures 15 and 16 show almost
framed continua that are not local dendrites, in both continua, the small squares
are divided infinitely many times. The continuum in Figure 16 is framed and the
one in Figure 18 is not framed.
Figure 15.
Figure 16.
The following theorem generalizes Theorems 2, 6 and 7.
Theorem 10 ([20, Theorems 29, 30 and 35]). Framed continua different from an
arc have unique hyperspace Cn (X) for all n ∈ N.
34 Alejandro Illanes
Figure 17.
By Theorem 11, continua X in class (a) does not have unique hyperspace Cn (X)
for any n ∈ N; and by Theorem 10, framed continua have unique hyperspace
Cn (X) for each n ∈ N. Thus, with respect to unique hyperspace Cn (X), the
gray zone is continua in class (c). Theorem 12 gives a sufficient condition to have
unique hyperspace Cn (X). Surprisingly, this condition is not necessary since in
[23, Example 38] it was proved that the continuum X in Figure 18 has unique
hyperspace C(X).
Hence, the following problem remains open.
Problem 13. Characterize locally connected continua X which have unique hyper-
space Cn (X).
Figure 18.
Uniqueness of hyperspaces 35
X Y
Figure 19.
Question 18. Let X be a compactification of the real line. Does X have unique
hyperspace C2 (X)?
F. Type λ continua
A continuum X which is irreducible between two points p and q is said to be
of type λ provided that there is a monotone (with connected fibers) mapping g :
X → [0, 1] such that g(p) = 0, g(q) = 1 and intX (g −1 (t)) = ∅ for each t ∈ [0, 1]; the
fibers g −1 (t) are called layers of X. The fiber g −1 (t) is called a layer of cohesion
provided that g −1 (t) ⊂ clX ((t, 1]), if t < 1 and g −1 (t) ⊂ clX ([0, t)), if 0 < t.
Theorem 19 ([6, Theorem 4.1]). Let X be a continuum of type λ, each layer of
X is of cohesion, the set of degenerate layers is dense in X and both end layers of
X are nondegenerate, then X has unique hyperspace.
36 Alejandro Illanes
Figure 20.
G. Indecomposable continua
A continuum is said to be indecomposable provided that it is not the union of two
proper subcontinua. A hereditarily indecomposable continuum is a continuum such
that all its subcontinua are indecomposable. A continuum is called arc continuum
if each of its proper subcontinua is an arc.
It is known [38, Theorem 14.9] that hyperspaces 2X and Cn (X) are arcwise
connected. For hereditarily indecomposable continua the arc structure of 2X and
Cn (X) reflects many properties of the continuum X. In particular, this arc struc-
ture allows to prove the following result.
metrizable boundary), then X has unique hyperspace C(X) (among all the Haus-
dorff continua). A. Peláez [46] extended this result to all hyperspaces 2X and
Cn (X), still asking that X is rim metrizable. Recently, the author [34] has proved
Theorem 20 for Hausdorff continua without any metrizability assumption.
Problem 21. Find more classes of continua X having unique hyperspace 2X
Other class of continua for which the structure of the subcontinua is relatively
easy are the indecomposable arc continua. Examples of these spaces are the Buck-
ethandle (Figure 21) and the solenoids (see [38, Definition 61.1], for descriptions).
Figure 21.
H. C-Determined continua
A class of continua G is said to be C-determined, provided that if X and Y are
continua in class G and C(X) is homeomorphic to C(Y ), then X is homeomorphic
to Y .
A chainable continuum is a continuum X such that for each ε > 0, there exists
a finite sequence of open sets U1 , . . . , Un such that: (a) X = U1 ∪ . . . ∪ Un , (b)
diameter(Ui ) < ε for all i and (c) Ui ∩ Uj ̸= ∅ if and only if |i − j| ≤ 1. The
continuum X is said to be circle-like provided that it satisfies similar conditions
but with (c’) Ui ∩ Uj ̸= ∅ if and only if |i − j| ≤ 1 or {i, j} = {1, n}. Answering a
38 Alejandro Illanes
question by S. B. Nadler, Jr. [44, Question 0.62], the author showed that the class
of chainable continua is not C-determined. If X and Y are the examples in Figure
22, then in [27] it was shown that C(X) and C(Y ) are homeomorphic.
Figure 22.
Changing the pattern of the copies of the sin( x1 )-continua in this example, it is
possible to obtain uncountable many examples similar to the one given in Figure
22 [12].
The idea of the Example 24 is as follows, take a continuum X as that in Figure 23.
Then C(X) and C(Y ) are homeomorphic, and they are homeomophic to C(Z),
where Z is as in Figure 23. Since Z was constructing by cutting out an arc in Y
and the cutting can be made in countably many places, it is possible to construct
countably many continua Z with the same properties.
Question 25 ([6, Question 3.11]). Does there exist a chainable continuum X and
an uncountable family {Zα : α ∈ J} of pairwise non-homeomorphic chainable
continua such that C(X) is homeomorphic to C(Zα ) for each α ∈ J?
Question 26. Are there two non-homeomorphic chainable continua X and Y such
that C2 (X) and C2 (Y ) are homeomorphic?
The trick suggested in Figure 23 does not work for circle-like continua. So, the
following problem remains open.
Figure 23.
Theorem 28. ([5, Theorems 5.3 and 6.6]) (a) The class of arcwise connected
circle-like continua is C-determined. (b) The class of homogeneous circle-like con-
tinua is C-determined.
A dendroid is an arcwise connected continuum such that the intersection of any
two of its subcontinua is connected. It is known [45, Theorem 10.10] that the class
of the dendrites coincides with the class of locally connected dendroids. A fan is a
dendroid with exactly one ramification point.
By Theorem 7, most smooth fans do not have unique hyperspace C(X).
Figure 24.
Consider the map ϕ : F3 ([0, 1]) → R2 given by ϕ(A) = (min A, max A). Then ϕ
is a continuous function whose image is the triangle T in Figure 1. Given (a, b) ∈ T ,
the fiber ϕ−1 ((a, b)) is the set {{a, b, c} : a ≤ c ≤ b}. In the case that a < b, the
set {{a, b, c} : a ≤ c ≤ b} is a simple closed curve since c runs on the interval [a, b]
and {a, a, b} = {a, b, b}. In the case that a = b, ϕ−1 ((a, b)) = {{a, b, c} : a ≤ c ≤
b} = {{a}}. Thus, to obtain a model for F3 ([0, 1]) we need to put a circle on each
point (a, b) ∈ T such that a < b and a one-point-set in the points of the diagonal
of T . This can be realized by taking the revolution body that can be obtained by
rotating T around its diagonal. Therefore, F3 ([0, 1]) is a 3-cell.
Notice that the set of singletons F1 ([0, 1]) is represented as the rotation axis
and the elements of the set {{x} ∈ F3 ([0, 1]) : 0 < x < 1} are in the interior
(as manifold) of the 3-cell. Thus, from the topological point of view, they are
undistinguishable from many other elements in F3 ([0, 1]). This implies that it is
Uniqueness of hyperspaces 41
Figure 25.
not possible to identify (topologically) the set F1 ([0, 1]) when we observe hyperspace
F3 ([0, 1]).
Fortunately, for n ≥ 4, it is possible to recognize the subset F1 (G) from Fn (G)
for some continua G. To illustrate how this can be achieved, we suppose that G is
a finite graph. Let
F = {A ∈ Fn (G) : A has a neighborhood in Fn (G) that is a 2n-cell} .
In [11, Corollary 4.4] it is proved that
F = {A ∈ Fn (G) : A has exactly n points and A does not have
ramification points of G} .
And, in Lemma 4.5 of [11], it is shown that
C = {A ∈ F1 (G) : A does not contain a ramification point of G}
= {A ∈
/ F : A has a basis B of neighborhoods in Fn (G) such that
U ∩ F is arcwise connected for each U ∈ B} .
Therefore, C can be topologically distinguished in Fn (G). Since clFn (G) (C) =
F1 (G), we have that F1 (G) can be distinguished in Fn (G) and G has unique hy-
perspace Fn (G).
Theorem 34 ([11]). Finite graphs G have unique hyperspace Fn (G) for every n ∈
N.
Theorem 35 ([7], [23] and [30]). Dendrites X in class D have unique hyperspace
Fn (X).
References
[1] G. Acosta, On compactifications of the real line and unique hyperspace, Topology Proc. 25
(2000), 1–25.
[2] G. Acosta, Continua with unique hyperspace, Continuum theory (Denton, TX, 1999, 33-49
Lecture Notes in Pure and Appl. Math. 230, Dekker, New York, 2002.
[3] G. Acosta, Continua with almost unique hyperspace, Topology Appl. 117 (2002), 175–189.
[4] G. Acosta, On smooth fans and unique hyperspace, Houston J. Math. 30 (2004), 99–115.
[5] G. Acosta, Homogeneous circle-like continua are C-determined, Topology Proc. 30, (2006),
1–23.
[6] G. Acosta, J. J. Charatonik and A. Illanes, Irreducible continua of type λ with almost unique
hyperspace, Rocky Mountain J. Math. 31 (2001), 745–772.
Uniqueness of hyperspaces 43
[32] A. Illanes, A model for the hyperspace C2 (S 1 ), Questions Answers Gen. Topology 22 (2004),
117–130.
[33] A. Illanes, Dendrites with unique hyperspace C2 (X), II. Topology Proc. 34 (2009), 77–96.
[34] A. Illanes, Hereditarily indecomposable Hausdorff continua have unique hyperspaces 2X and
Cn (X), Publ. Inst. Math. (Beograd) (N.S.) 89 (103) (2011), 49–56.
[35] A. Illanes, Models of hyperspaces, preprint.
[36] A. Illanes, S. Macı́as and S. B. Nadler, Jr., Symmetric Products and Q-manifolds, Geometry
and Topology in Dynamics, Contemporary Math. Series of Amer. Math. Soc. 246, 1999,
Providence, RI, 137–141.
[37] A. Illanes and Jorge Martı́nez-Montejano, Compactifications of [0, ∞) with unique hyperspace
Fn (X), Glas. Mat., III. Ser. 44 (2009), 457–478.
[38] A. Illanes and S. B. Nadler, Jr., Hyperspaces: Fundamentals and Recent Advances, Mono-
graphs and Textbooks in Pure and Applied Mathematics, 216, Marcel Dekker, Inc., New
York and Basel, 1999.
[39] I. Lončar, Non-metric rim-metrizable continua and unique hyperspace, Publ. Inst. Math.,
Nouv. Sér. 73 (87) (2003), 97–113.
[40] S. Macı́as, Hereditarily indecomposable continua have unique hyperspace 2X , Bol. Soc. Mat.
Mex., III. Ser. 5, (1999), 415–418.
[41] S. Macı́as, On the hyperspaces Cn (X) of a continuum X, II, Topology Proc. 25 (2000),
255–276.
[42] V. Martı́nez-de-la-Vega, Dimension of n-fold hyperspaces of graphs, Houston J. Math. 32
(2006), 783–799.
[43] Verónica Martı́nez-de-la-Vega and Norberto Ordoñez, Embedding hyperspaces, to appear in
Topology Appl.
[44] S. B. Nadler, Jr., Hyperspaces of sets: A text with research questions, Monographs and
Textbooks in Pure and Applied Mathematics, Vol. 49, Marcel Dekker, Inc., New York and
Basel, 1978.
[45] S. B. Nadler, Jr., Continuum Theory, Monographs and Textbooks in Pure and Applied
Mathematics, 158, Marcel Dekker, Inc., New York and Basel, 1992.
[46] A. Peláez, On the uniqueness of the hyperspaces 2X and Cn (X) of rim-metrizable continua,
Topology Proc. 30 (2006) 565–576.