Houston Journal of Mathematics

c 2011 University of Houston

Volume 37, No. 3, 2011

MAPS OF ABSOLUTE RETRACTS FOR TREE-LIKE

CONTINUA

WLODZIMIERZ J. CHARATONIK AND JANUSZ R. PRAJS

Communicated by Charles Hagopian

Abstract. It is shown that the class AR(TL) of absolute retracts for tree-
like continua is invariant with respect to monotone maps. From the past
results it follows that the inverse limit of trees with confluent bonding maps
is an absolute retract for tree-like (and more generally, hereditarily unico-
herent) continua. In this paper we provide the first example of a member of
AR(TL) that is not such a limit.

The study of absolute retracts for hereditarily unicoherent continua (abbr.

AR(HU)) and tree-like continua (abbr. AR(TL)) was initiated by Tadeusz
Maćkowiak [19], and also by David Bellamy according to [19, Added in proof,
p. 183]. More recently it was continued by late Janusz Charatonik and the au-
thors [8], [9], [10], [11], [12], [13], [14]. Except for dendrites, these spaces are
not absolute retracts for all metric spaces. Nevertheless, they manifest regular
properties: they are Kelley continua, they have arc approximation property, and
generalized ε-push property [10]. Strong mapping extension properties similar
to those of ARs, and deep regularity make these absolute retracts subject of
continued interest.
The regular structure of the members of AR(HU) and AR(TL) leads to ex-
pectation that these classes should be invariant with respect to various types
of mappings such as open, monotone or confluent ones. However, proving such
invariant properties turns out difficult because we have no structural characteri-
zation of any of these classes. In this paper we offer the first non-trivial invariance
result for AR(TL). We show that this class is preserved under monotone maps.

2000 Mathematics Subject Classification. 54F15, 54F50, 54C10, 54C55, 54E45.

Key words and phrases. continuum, Kelley continuum, absolute retract, monotone map,
tree-like.
967
968 WLODZIMIERZ J. CHARATONIK AND JANUSZ R. PRAJS

Our proof depends on the homeomorphism extension theorem for the Menger
curve, and on shape theory. We are not aware of any elementary proof of this
result.
We present the following application of the invariance theorem. In [11, Theo-
rem 3.6, p. 94] it was shown that the inverse limit of trees with confluent bonding
maps is an AR(HU). Such inverse limits have been the only known examples of
AR(TL)’s and AR(HU)’s. The problem whether all members of these classes are
inverse limits of trees with confluent bonding maps [11, Problem 1.2, p. 85 and
Question 6.2, p. 108] has been open for several years despite numerous attempts
of solution. Here we solve this problem in the negative. Namely, we define a
Kelley dendroid K that does not have such inverse limit representation but it is a
monotone image of an AR(TL). Thus K is an AR(TL) by the main result of the
paper. Using the continuum K we also answer, in the negative, an older question
whether each Kelley dendroid can be represented as an inverse limit of dendrites
with confluent bonding maps [6, Question 7, p. 77], [11, Question 6.5, p. 108].

1. Preliminaries
All spaces are assumed to be metric and mappings, continuous. Given a space
X and its subspace Y ⊂ X, a mapping r : X → Y is called a retraction if the
restriction r|Y is the identity. Then Y is called a retract of X. The reader is
referred to [2] and [16] for needed information on these concepts.
Let C be a class of compacta, i.e., of compact metric spaces. Following [16, p.
80], we say that a space Y ∈ C is an absolute retract for the class C (abbreviated
AR(C)) if for any space Z ∈ C such that Y is a subspace of Z, Y is a retract of
Z. The concept of an AR space originally had been studied by K. Borsuk [2].
By a continuum we mean a connected compactum. A curve means a 1-
dimensional continuum. A continuum X is said to be unicoherent if the in-
tersection of every two of its subcontinua whose union is X is connected. X is
said to be hereditarily unicoherent if all of its subcontinua are unicoherent. A
hereditarily unicoherent and arcwise connected continuum is called a dendroid. A
locally connected dendroid is called a dendrite. A tree means a graph containing
no simple closed curve.
A continuum is said to be decomposable provided that it can be represented
as the union of two of its proper subcontinua. A continuum is said to be hered-
itarily decomposable provided that each of its subcontinua is decomposable. A
 MAPS OF ABSOLUTE RETRACTS FOR TREE-LIKE CONTINUA 969

hereditarily unicoherent and hereditarily decomposable continuum is called a λ-

dendroid. A continuum is said to be tree-like provided that it is the inverse limit
of an inverse sequence of trees.
Let D0 denote the class of dendrites, D — the class of dendroids, TL — of
tree-like continua, and HU — the class of hereditarily unicoherent ones. Then
D0 ⊂ D ⊂ λD ⊂ TL ⊂ HU.
According to a classical result of K. Borsuk [2, (13.5), p. 138] each dendrite is
an absolute retract for the class of all compacta. Consequently any dendrite D is
an absolute retract for each class C of compacta (abbreviated AR(C)) such that
D ∈ C. More generally, if C1 ⊂ C2 for some classes C1 and C2 of spaces, then
C1 ∩ AR(C2 ) ⊂ AR(C1 ).
In particular, we have
AR(D0 ) = D0 ⊂ AR(D) ∩ AR(λD) ∩ AR(TL) ∩ AR(HU).
We start with recalling the following concept. A continuum X is called Kelley,
or is said to have the property of Kelley, provided that for each point p ∈ X,
for each subcontinuum K of X containing p and for each sequence of points pn
converging to p there exists a sequence of subcontinua Kn of X containing pn and
converging to the continuum K (see e.g. [17, p. 167] or [22, Definition 16.10, p.
538]).
A mapping between continua is called monotone if preimages of points are
connected, equivalently, if preimages of continua are connected. A mapping f :
X → Y between continua X and Y is confluent if for every subcontinuum K of
Y , and for every component C of f −1 (K), we have f (C) = K. Of course each
monotone map is confluent.
We say that a subcontinuum C of a continuum X locally separates X if there
is an open connected set U ⊆ X such that U − C is not connected. Note that we
do not require that C is a subset of U .
In this article we will use the Menger universal curve whose description can be
found in [15, Chapter 1, §11, p. 121]. Here we use its topological characterization
by Anderson [1].

2. Main results
In this section we prove that every monotone image of a member of AR(TL)
is again a member of AR(TL). Let us start with an auxiliary, but quite general
results we will need to prove the main theorem.
970 WLODZIMIERZ J. CHARATONIK AND JANUSZ R. PRAJS

Lemma 2.1. Let K be a non-locally separating subcontinuum of the Menger curve

M , and D a monotone, upper semi-continuous decomposition of K whose quo-
tient, K/D, is one-dimensional. Let D∗ be the decomposition of M that extends
D and has all singleton in M − K as members. Then the decomposition space
X = M/D∗ is homeomorphic to M and the set K/D = K/D∗ ⊂ X does not
locally separate X.
Proof. To show that X is homeomorphic to M , by Anderson’s characterization
of the Menger curve [1], it suffices to show that X is a one-dimensional, locally
connected continuum with no open non-empty subsets embeddable in the plane,
and no locally separating points.
Let q : M → M/D∗ = X be the quotient map. Since X is a continuous im-
age of M , it is locally connected. Note that X is a countable union of closed
S∞
one-dimensional subsets, namely X = q(K) ∪ n=1 q(M − N n1 (K)). Therefore
dim X = 1 by [15, Theorem 1.5.3]. We have Int M K = ∅ and q maps homeomor-
phically M − K onto X − q(K). Consequently, every open set in X contains a
copy of an open set in M . Thus X has no non-empty open subsets embeddable
in the plane.
Now, we show that q(K) does not locally separate X. Indeed, suppose U
is an open, connected set in X. We have U − q(K) = q(q −1 (U − q(K))) =
q(q −1 (U ) − q −1 (q(K))) = q(q −1 (U ) − K). The set K is non-locally separating
and q −1 (U ) is a connected open subset of M . Thus q −1 (U ) − K is connected.
Therefore, its continuous image, U − q(K), is also connected.
Suppose p is a locally separating point of X. Since the points in X − q(K)
have neighborhoods homoeomorphic to open sets in M , it follows p ∈ q(K). This
would imply, however, that q(K) locally separates X, a contradiction. The proof
is complete. 
Remark 2.2. The next result is related to the statement of [20, Corollary 6.5,
p.44]. This statement is clearly false by an example of a two point set in M
mapped to a single point. We suspect that the following captures a correct idea
behind that statement.
Theorem 2.3. Let K and L be two non-locally separating subcontinua of the
Menger curve M , and f : K → L a monotone surjection. Then there is an
extension f ∗ : M → M of f such that f |(M − K) : M − K → M − L is a
homeomorphism.
Proof. Let D = {f −1 (p) | p ∈ L}. Extend D to a decomposition D∗ of the
whole M whose elements are members of D and singletons outside of K. Let
 MAPS OF ABSOLUTE RETRACTS FOR TREE-LIKE CONTINUA 971

q : M → M/D∗ be the quotient map of this decomposition. By Lemma 2.1 the

space M/D∗ is homeomorphic to M , and the set L1 = K/D ⊂ M/D∗ does not
locally separate M/D∗ . Since f and q|K : K → L1 are quotient maps defined by
the same decomposition of K, there is a homeomorphism h : L1 → L such that
h ◦ (q|K) = f . By [20, Theorem 5.1] there is a homeomorphism h∗ : M/D∗ → M
such that h∗ |K = h. The composition f ∗ = h∗ ◦ q is the required extension of the
map f . 

Theorem 2.4. Monotone image of a member of AR(TL) is a member of AR(TL).

Proof. Let K be an absolute retract for tree-like continua, and let f : K → L

be a monotone map. Then, by [21], L is a tree-like continuum. Suppose that L
is a subcontinuum of a tree-like continuum T . It suffices to show there exists a
retraction r : T → L. Since K and T can be embedded into a Menger curve M as
non-locally separating subcontinua of M , we assume, without loss of generality,
that K and T are such subcontinua of M . Clearly, L ⊂ T is also a non-separating
subcontinuum of M . According to Theorem 2.3, there is a monotone extension
f ∗ of f such that f |(M − K) : M − K → M − L is a homeomorphism.
Recall that tree-like continua are characterized as continua of dimension at
most one with trivial shape, which follows from [4, Theorem 1, p. 74]. Let
S = (f ∗ )−1 (T ), and note that S is a one-dimensional continuum. The pre-images
(f ∗ )−1 (p) for p ∈ T are tree-like, and thus S is a tree-like continuum by [3,
Theorem 9.3, p. 352]. Since K ∈ AR(TL), there is a retraction r0 : S → K.
Define r : T → L by r(p) = p for p ∈ L, and r(p) = f (r0 ((f ∗ |(M − S))−1 (p))) for
p ∈ T − L. The proof of the continuity of r is left to the reader. The map r is a
desired retraction. 

Corollary 2.5. If X is a tree-like continuum in AR(HU), then every monotone

image of X is such.

Proof. It is enough to recall that AR(TL) = TL ∩ AR(HU), see [11, Corollary

4.9]. 

In connection with Theorem 2.4 and Corollary 2.5, we ask the following ques-
tions.

Question 2.6. Is a confluent (open) image of a member of AR(TL) a member

of AR(TL)?

Question 2.7. Is every monotone (cell-like, confluent, open) image of a member

of AR(HU) a member of AR(HU) ?
972 WLODZIMIERZ J. CHARATONIK AND JANUSZ R. PRAJS

Also, we still do not know if every member of AR(HU) is tree-like (see [11,
Question 6.3]). If this is the case, then the answer to Question 2.7 is positive for
monotone and, equivalently, cell-like maps.

3. Applications
A continuum X is called confluently tree-representable if it can be represented
as the inverse limit of trees with confluent bonding mappings; it is called conflu-
ently tree-like if for every ε > 0 there is a confluent ε-mapping of X onto a tree.
It has been shown in [23] that these two notions are equivalent.
In this section we use Theorem 2.4 to show an example of a dendroid in
AR(TL), which is not confluently tree-like. This answers in the negative question
[11, Problem 1.2, p. 85] and also an older question whether every Kelley dendroid
is confluently tree-representable, see [6, Question 7, p. 77].
To prove the properties of the example we use the order in the classical sense of
a point a in a dendroid X defined as the cardinality of the set of arc components
of X − {x} and denoted by ord(x). A path from a to b in a space X is a map
p : [0, 1] → X such that p(0) = a and p(1) = b. Given a point a in a dendroid X,
define a cardinal number

α(a) = min{sup{ord(x) : x ∈ p([0, 1])} :

p is a path from a to an end point in X}.

In particular, α(e) = 1 for every end point e in X, and α(x) = 2 if x is not an end
point, but a path from a to an end point does not pass through a ramification
point. It is known [7, Theorem 3.1] that light confluent maps do not increase the
order ord(x) in dendroids. The following proposition is an immediate consequence
of this fact.

Proposition 3.1. If f : X → Y is a light confluent map between dendroids X

and Y , then α(f (x)) ≤ α(x) for every x in X.

Example 3.2. There is a Kelley dendroid X such that X is an absolute retract

for tree-like continua, and X is not confluently tree like. Consequently, X is not
confluently tree representable.

Proof. Denote by C the standard ternary Cantor set in [0, 1] and let Y = ([0, 1]×
{0})∪(C×[0, 1]). In the interval [0, 1]×{0} shrink the closure of every component
of [0, 1]−C to a point to define a quotient map f : Y → f (Y ). Note that X = f (Y )
is a dendroid (see Figure 1).
 MAPS OF ABSOLUTE RETRACTS FOR TREE-LIKE CONTINUA 973

Figure 1

Let D0 = [0, 13 ] and D2 = [ 23 , 1]. In general, for every finite 0,2-sequence σ

denote by Dσ the portion of the unit interval composed of all numbers whose
ternary representation starts with σ. For example D02 = [ 92 , 39 ] and D202 =
20 21
[ 27 , 27 ]. Let fn be the quotient map on Y defined by the condition that two
different points in Y are identified if and only if they have the same second
coordinates, and their first coordinates are in the same set Dσ for some sequence
σ of length n. Note that this decomposition is upper semi-continuous, the quotient
space is a tree, and each equivalence class has diameter less than or equal 3−n .
Thus fn is a (1/3n )-mapping. The proof that fn ’s are confluent maps is left to
the reader.
Using the maps fn we see that Y is confluently tree-like, and thus it is conflu-
ently tree-representable by [23]. Consequently, Y is in AR(HU) by [11, Theorem
974 WLODZIMIERZ J. CHARATONIK AND JANUSZ R. PRAJS

3.6]. Since X is a monotone image of Y , it is a member of AR(TL) by Theo-

rem 2.4.
Now we show that X is not confluently tree-representable (equivalently, not
confluently tree-like). Assume on the contrary there is a confluent map g : X → T
from X onto a tree T such that:
(1) the image g(f ([0, 1] × {0})) is not a one point set;
(2) the images g(f ({x} × [0, 1])) are not one point sets for x ∈ C;
(3) the images g(f ({0} × [0, 1])), g(f ({ 29 } × [0, 1])), g(f ({ 79 } × [0, 1])), and
g(f ({1} × [0, 1])) are pairwise disjoint.
Then g is the composition g = g0 ◦ g1 of a monotone map g1 and a light map
g0 by Whyburn’s factorization theorem. By the composition factor property for
confluent maps [5, IV, p. 214], the map g0 is confluent. The continuum g1 (X)
is a dendroid, by [5, Corollary 2]. Denote by B the arc f ([0, 1] × {0}) and note
that the image g1 (B) is an arc, because the restricted mapping g1 |B is monotone,
and monotone image of an arc is an arc. The arc g1 (B) contains all ramification
points of g1 (X) by [7, Corollary 3.3]. Since the map g1 does not identify any of
the arcs f ({x} × [0, 1]) to a point by (2), the arc g1 (B) is nowhere dense in g1 (X).
For every x ∈/ g1 (B) we have α(x) ≤ 2, and thus the set {x ∈ m(X) : α(x) ≤ 2}
is dense in g1 (X). Consequently, by Proposition 3.1 the set {x ∈ T : α(x) ≤ 2}
is dense in T . Since only n-ods have this property among trees, T is an n-od for
some natural number n. and the image of the arc g1 (B) under the mapping g0
must be a single point, contrary to assumption (1). 

Our example shows that confluently tree representable dendroids form a proper
subclass of the class of Kelley dendroids. We ask the following.

Question 3.3. Is every Kelley dendroid a monotone image of a confluently tree

representable dendroid?

A positive answer to the previous question would solve the following problem,
which is fundamental to the subject of this article.

Question 3.4. Is every Kelley dendroid a member of AR(TL)?

References
[1] R. D. Anderson, One-dimensional continuous curves and a homogeneity theorem, Ann. of
Math. (2) 68 (1958), 1 – 16.
[2] K. Borsuk, Theory of Retracts, PWN, Warszawa 1967.
[3] K. Borsuk, Theory of Shape, PWN, Warszawa 1975.
 MAPS OF ABSOLUTE RETRACTS FOR TREE-LIKE CONTINUA 975

[4] J. H. Case and R. E. Chamberlin, Characterizations of tree-like continua, Pacific J. Math.

10 (1960), 73–84.
[5] J. J. Charatonik, Confluent mappings and unicoherence of continua, Fund. Math. 56
(1964), 213 – 220.
[6] J. J. Charatonik and W. J. Charatonik, Fans with the property of Kelley, Topology Appl.
29 (1988), 73 – 78.
[7] J. J. Charatonik and W. J. Charatonik, Open mappings increasing order, Proc. Amer.
Math. Soc. 125 (1997), 3725–3733.
[8] J. J. Charatonik, W. J. Charatonik and J. R. Prajs, Kernels of hereditarily unicoherent
continua and absolute retracts, Proceedings of the Spring Topology and Dynamical Systems
Conference (Morelia City, 2001), Topology Proc. 26 (2001/02), 127–145.
[9] J. J. Charatonik, W. J. Charatonik and J. R. Prajs, Gate continua, absolute terminal
continua and absolute retracts, Houston J. Math. 28 (2002), no. 1, 89–117.
[10] J. J. Charatonik, W. J. Charatonik and J. R. Prajs, Arc property of Kelley and absolute
retracts for hereditarily unicoherent continua, Colloq. Math. 97 (2003), 49–65.
[11] J. J. Charatonik, W. J. Charatonik and J. R. Prajs, Hereditarily unicoherent continua and
their absolute retracts, Rocky Mountain J. Math. 34 (2004), 83–110.
[12] J. J. Charatonik, W. J. Charatonik and J. R. Prajs, Atriodic absolute retracts for heredi-
tarily unicoherent continua, Houston J. Math. 30 (2004), 1069–1087.
[13] J. J. Charatonik, and J. R. Prajs, Generalized ε-push property for certain atriodic continua,
Houston J. Math. 31 (2005), 441–450.
[14] J. J. Charatonik and J. R. Prajs, AANR spaces and absolute retracts for tree-like continua,
Czechoslovak Math. J. 55(130) (2005), 877–891.
[15] R. Engelking, Dimension Theory, North-Holland Mathematical Library, 19. North-Holland
Publishing Co., Amsterdam-Oxford-New York; PWN—Polish Scientific Publishers, War-
saw, 1978.
[16] Hu, Sze-tsen Theory of retracts, Wayne State University Press, Detroit 1965.
[17] A. Illanes and S. B. Nadler, Jr., Hyperspaces: Fundamentals and Recent Advances. Mono-
graphs and Textbooks in Pure and Applied Mathematics, 216. New York: Marcel Dekker,
Inc., 1999.
[18] A. Lelek and D. R. Read, Compositions of confluent mappings and some other classes of
functions, Colloq. Math. 29 (1974), 101–112.
[19] T. Maćkowiak, Retracts of hereditarily unicoherent continua, Bull. Acad. Polon. Sci. Sr.
Sci. Math. 28 (1980), 177–183.
[20] J. C. Mayer, L. G. Oversteegen and E. D. Tymchatyn, The Menger curve. Characterization
and extension of homeomorphisms of non-locally-separating closed subsets, Dissertationes
Math. (Rozprawy Mat.), 252 (1986), 1–45.
[21] T. B. McLean, Confluent images of tree-like curves are tree-like Duke Math. J. 39 (1972),
465–473.
[22] S. B. Nadler, Jr., Hyperspaces of Sets. Monographs and Textbooks in Pure and Applied
Mathematics, 49. New York: Marcel Dekker, Inc., 1978.
[23] L. G. Oversteegen and J. R. Prajs, On confluently graph-like compacta, Fund. Math. 178
(2003), 109–127.
976 WLODZIMIERZ J. CHARATONIK AND JANUSZ R. PRAJS

Received March 27, 2008

Department of Mathematics and Statistics, Missouri University of Science and

Technology, Rolla, MO 65409

Department of Mathematics, California State University, Sacramento

University of Opole, Institute of Mathematics, Ul. Oleska 48, 45-052 Opole,

Poland
E-mail address: wjcharat@[Link]
E-mail address: prajs@[Link]