1 Introduction
A continuum is a compact, connected metric space. A point x in a plane continuum
$X\subset \mathbb R^2$
is buried if x is not in the boundary (or frontier) of any component of
$\mathbb R ^2\setminus X$
. We denote by
$\operatorname {\mathrm {bur}}(X)$
the set of all buried points of X. Note that
$\operatorname {\mathrm {bur}}(X)$
is a
$G_{\delta }$
-set.
Motivation to study buried points comes from complex dynamics. Specifically, one of the difficult problems in complex dynamics asks whether
$\operatorname {\mathrm {bur}}(J(R))$
must be zero-dimensional in case it is punctiform, where
$J(R)$
is the Julia set of a rational function on the sphere
$\mathbb C_\infty $
. The Devaney–Rocha examples of Sierpiński gasket like Julia sets have this property. Curry, Mayer, and Tymchatyn proposed to study it from a purely topological point of view. For more details, see [Reference Curry and Mayer1, Reference Curry, Mayer and Tymchatyn2, Reference Devaney, Rocha and Siegmund4].
In [Reference van Mill and Tuncali9, Proposition 3.1], van Mill and Tuncali prove that the set of buried points of a plane continuum can be a Cantor set. They use this to provide in §4 of their paper an example of a plane Suslinian continuum whose set of buried points is totally disconnected and weakly
$1$
-dimensional. Their example answers in the negative Question 1 of Curry, Mayer, and Tymchatyn [Reference Curry, Mayer and Tymchatyn2], concerning whether or not a set of buried points is zero-dimensional provided it is totally disconnected. Note that if a plane continuum is regular (that is, has a basis of open sets with finite boundaries), then the set of buried points is either empty or zero-dimensional (see [Reference Curry, Mayer and Tymchatyn2, Theorem 3]).
Curry et al. showed that if X is a plane continuum with complementary components whose boundaries are locally connected and
$\operatorname {\mathrm {bur}}(X)$
is Suslinian, then in fact X has to be Suslinian [Reference Curry, Mayer and Tymchatyn2, Theorem 6]. This is the case for the example in [Reference van Mill and Tuncali9, Section 4] (the buried point set in that example is even totally disconnected). However, it is not locally connected. A plane continuum X is locally connected if and only if the boundaries of complementary components are locally connected and form a null-sequence [Reference Whyburn13, VI Theorem 4.4]. Using this fact, we construct in §5 a locally connected plane continuum whose set of buried points is totally disconnected but not zero-dimensional. The set at which the buried points are
$1$
-dimensional is countable. This is sharp according to the following theorem, which we prove in §3.
Theorem A Let X be a Suslinian plane continuum. If
$Y\subset X$
is a totally disconnected Borel set, then Y is zero-dimensional at all but countably many points.
Corollary B Let X a plane continuum such that the boundary of each component of
$\mathbb R^2\setminus X$
is locally connected (e.g. X is locally connected). If
$\,\operatorname {\mathrm {bur}}(X)$
is totally disconnected, then
$\operatorname {\mathrm {bur}}(X)$
is zero-dimensional at all but countably many points.
Countable sets are zero-dimensional, so under the assumptions of Corollary B the buried set is either zero-dimensional or weakly
$1$
-dimensional. There is no almost zero-dimensional, weakly
$1$
-dimensional space [Reference Lipham7, Theorem 1]. Therefore,
$\operatorname {\mathrm {bur}}(X)$
cannot be almost zero-dimensional in the proper sense (cf. [Reference Curry and Mayer1, Question 2.7]). To summarize:
Corollary C Let X be a plane continuum such that each component of
$\mathbb R^2\setminus X$
has locally connected boundary (e.g. X is locally connected).
-
(i) If
$\,\operatorname {\mathrm {bur}}(X)$
is totally disconnected, then
$\operatorname {\mathrm {bur}}(X)$
is at most weakly
$1$
-dimensional. -
(ii) If
$\,\operatorname {\mathrm {bur}}(X)$
is almost zero-dimensional, then
$\operatorname {\mathrm {bur}}(X)$
is zero-dimensional.
It is still unknown whether the Julia set of a rational function may have a buried set which is totally disconnected but not zero-dimensional (cf. [Reference Curry, Mayer and Tymchatyn2, Question 3]).
2 Definitions
All spaces under consideration are assumed to be separable and metrizable.
A space X is Suslinian if each collection of pairwise disjoint nondegenerate continua in X is countable.
A space X is totally disconnected if every two points of X are contained in disjoint clopen sets, and zero-dimensional if X has a basis of clopen sets. Respectively, X is zero-dimensional at
$x\in X$
if the point x has a neighborhood basis of clopen sets (written
$\operatorname {\mathrm {ind}}_x X=0$
, cf. [Reference Engelking3, Problem 1.1.B]).
A space X is almost zero-dimensional if X has a basis of open sets whose closures are intersections of clopen sets [Reference Lipham7, Reference Oversteegen and Tymchatyn12]. Observe that every almost zero-dimensional space is totally disconnected.
The dimensional kernel of a
$1$
-dimensional space X is defined to be the set of points at which X is not zero-dimensional;
$\Lambda (X)=\{x\in X:\operatorname {\mathrm {ind}}_x X=1\}.$
A
$1$
-dimensional space X is weakly
$\mathbf {1}$
-dimensional if
$\Lambda (X)$
is zero-dimensional.
A domain is a connected open subset of the plane or topological
$2$
-sphere. If X is a continuum on the
$2$
-sphere, then each complementary component of X is a simply connected domain. Riemann’s mapping theorem implies that every simply connected domain
$U\subset \mathbb C$
is homeomorphic to the unit disc
$\{z\in \mathbb C:|z|<1\}$
. Carathéodory’s theorem implies moreover that if
$\partial U$
is a simple closed curve then
$\overline U$
is homeomorphic to the closed unit disc
$\{z\in \mathbb C:|z|\leq 1\}$
.
3 Main result
Suppose that X is a continuum and
$Y\subset X$
. For each
$y\in Y$
let
$\mathcal U_y$
be the collection of all open subsets U of X such that
$y\in U$
and
$\partial U\subset X\setminus Y$
. For each
$y\in Y$
put
Lemma 1 If Y is totally disconnected and
$y\in \Lambda (Y)$
, then
$F(y)$
is a nondegenerate subcontinuum of X and
$F(y)\cap Y = \{y\}$
.
Proof Suppose
$F(y)=\{y\}$
is degenerate. We show that Y is zero-dimensional at y. Let V be an arbitrary relatively open subset of Y that contains y. Pick an open subset W of X such that
$W \cap Y = V$
. Since
$F(y) = \{y\} \subset V$
, there is by compactness an element
$U \in \mathcal U_y$
such that
$\overline {U} \subset W$
. But then
$y \in U \cap Y \subset V$
, and
$U \cap Y$
is clopen in Y.
Now suppose that
$y\in Y$
and
$z\in Y\setminus \{y\}$
. Pick by total disconnectedness of Y clopen subsets
$C_0$
and
$C_1 = Y \setminus C_0$
of Y such that
$y \in C_0$
and
$z \in C_1$
. There are disjoint open subsets
$V_0$
and
$V_1$
of X such that
$V_0 \cap Y = C_0$
and
$V_1 \cap Y = C_1$
. Observe that the boundary of
$V_0$
is contained in
$X \setminus Y$
. Since
$V_1 \cap V_0 = \varnothing $
, this shows that
$z \notin F(y).$
Hence,
$F(y)\cap Y = \{y\}$
.
It remains to prove
$F(y)$
is connected. Just suppose
$F(y) = A \cup B$
where A and B are disjoint closed sets with
$y \in A$
. Let U and V be disjoint open neighborhoods of A and B in X, respectively. By compactness, there exists
$W \in \mathcal U_y$
such that
$\overline W \subset U \cup V$
. Then
$W \cap U \in \mathcal U_y$
, so
$B = \varnothing .$
Remark 1 If X is locally connected and
$U\in \mathcal U_y$
, then the connected component of y in U is path-connected and belongs to
$\mathcal U_y$
.
Lemma 2 Let S be a circle in the plane, and
$a_1,a_2,a_3,a_4$
four points of S in cyclic order. Let
$A_i$
,
$i\in \{1,2,3,4\}$
, be pairwise disjoint arcs intersecting S only at an endpoint
$a_i$
, with opposite endpoints
$b_i$
in the bounded component of
$\mathbb R ^2\setminus S$
, denoted
$\operatorname {\mathrm {int}}(S)$
.
If
$K\subset \mathbb R ^2\setminus (A_2\cup S\cup A_4)$
is a continuum containing
$b_1$
and
$b_3$
, then
$b_2$
and
$b_4$
are in different components of
$\mathbb R^2\setminus (A_1\cup S\cup A_3\cup K).$
Proof We may assume that S is the unit circle in the complex plane
$ \mathbb C$
, and
Let
$\gamma \subset \mathbb C\setminus (A_1\cup S\cup A_3)$
be any arc from
$b_2$
to
$b_4$
. We will show that
$\gamma $
intersects K. This will imply that
$b_2$
and
$b_4$
are in different components of
$\mathbb C\setminus (A_1\cup S\cup A_3\cup K)$
, as all such components are path-connected.
In
$\gamma \cup A_2\cup A_4$
, there is an arc
$\hat \gamma $
intersecting S only at endpoints
$\pm {\mathrm {i}}$
. Let
$\alpha $
and
$\beta $
be the left and right halves of the circle S. By the
$\theta $
-curve theorem [Reference Munkres10, Lemma 64.1],
$\text {int}(S)\setminus \hat \gamma $
is the union of two disjoint domains U and V whose boundaries are
$\alpha \cup \hat \gamma $
and
$\hat \gamma \cup \beta $
, respectively. Let W be an open ball centered at
$-1$
that misses
$\hat \gamma $
. Then
$W\cap \text {int}(S)\subset U\cup V$
. Since
$W\cap \text {int}(S)$
is connected and
$-1\in \partial U$
, we have
$W\cap \text {int}(S)\subset U$
. Thus,
$U\cap A_3\neq \varnothing $
. It follows that
$A_3\setminus \{-1\}\subset U$
. Thus,
$b_3\in U$
. Likewise,
$b_1\in V$
. Now K is a connected subset of
$\text {int}(S)$
meeting both U and V. So
$K\cap \hat \gamma \neq \varnothing $
. Therefore,
$K\cap \gamma \neq \varnothing $
.
We can now prove Theorem A for locally connected X.
Theorem 3 Let X be a locally connected Suslinian plane continuum. If
$Y\subset X$
is a totally disconnected Borel set, then
$\Lambda (Y)$
is countable.
Proof For a contradiction suppose that
$Y\subset X$
is totally disconnected and Borel, and
$Z=\Lambda (Y)$
is uncountable. By Lemma 1 there exists
$\varepsilon>0$
such that for uncountably many z’s,
$\operatorname {\mathrm {diam}}(F(z))\geq 5\varepsilon $
. The set of all z’s with this property is closed in Y, thus it is Borel and contains a Cantor set. Let us just assume that
$Z $
is a Cantor set in Y with the property
$\operatorname {\mathrm {diam}}(F(z))\geq 5 \varepsilon $
for each
$z\in Z$
. We may further assume that Z lies inside of an open ball of radius
$\varepsilon $
centered at one of its points. Let S and
$S'$
be the circles of radii
$\varepsilon $
and
$2\varepsilon $
, respectively, centered at the point. Observe that every
$F(z)$
crosses
$S'$
.
In the claims below, all boundaries are respective to X (not
$\mathbb R ^2$
).
Claim 4 Let U be an open subset of X intersecting Z with
$\partial U\subset (X\setminus Y)\cup S$
. Then for every point
$p\in \mathbb R^2$
there exists a connected open set
$W\subset U$
intersecting Z such that
$p\notin \overline W$
and
$\partial W\subset (X\setminus Y)\cup S$
.
Proof of Claim 4 For each z in the uncountable set
$Z\cap U\setminus \{p\}$
there is a nondegenerate continuum
$K(z)\subset F(z)$
containing z and missing
$S\cup \{p\}$
. Just take any closed neighborhood N of z missing
$S\cup \{p\}$
, and let
$K(z)$
be the component of z in
$F(z)\cap N$
; by the boundary bumping theorem [Reference Nadler11, Theorem 5.4]
$K(z)$
is nondegenerate. Note that
$K(z)\cap Y=\{z\}$
by Lemma 1.
By the Suslinian property of X there exist
such that
$K(z_i)\cap K(z_1)\neq \varnothing $
for each
$i\in \{2,3,4\}$
. Let
$U_i\in \mathcal U_{z_i}$
be pairwise disjoint. By Remark 1, we may assume that each
$U_i$
is path-connected. Since
$F(z_i)$
extends beyond
$S'$
, so does
$U_i$
. Thus, there is an arc
$A_i\subset U_i$
from
$z_i$
to
$a_i\in S'$
that intersects
$S'$
only at
$a_i$
. By a permutation of indices
$i\in \{2,3,4\}$
, we may assume that the
$a_i$
’s are in cyclic order around
$S'$
. Let
$K=K(z_1)\cup K(z_3)$
. Note that
$p\notin K$
.
Case 1:
$p\in U_1\cup S'\cup U_3$
. Let W be the component of
$z_2$
in
$U_2\cap U\setminus S$
. Then W is a connected open subset of U intersecting Z, with
$p\notin \overline W$
. Note also that W is clopen in
$U_2\cap U\setminus S$
, which means that
$\partial W\subset \partial (U_2\cap U\setminus S)$
. The boundary of any finite intersection of open sets is contained in the union of the individual boundaries. Therefore,
$\partial W\subset \partial U_2 \cup \partial U\cup \partial (X\setminus S)\subset (X\setminus Y)\cup S$
.
Case 2:
$p\notin U_1\cup S'\cup U_3$
. Then certainly
$p\notin A_1\cup S'\cup A_3$
and since
$p\notin K$
we find that
$p\in O=\mathbb R^2\setminus (A_1\cup S'\cup A_3\cup K)$
. By Lemma 2,
$z_2$
and
$z_4$
are in different components of O. Of these two components, at least one does not contain p. Let’s say that V is the component of
$z_2$
in O, and
$p\notin V$
. The component of p in O is an open set missing V, so
$p\notin \overline V$
. Let W be the component of
$z_2$
in
$U_2\cap U \cap V\setminus S$
. Then W is a connected open subset of U intersecting Z, and
$p\notin \overline {W}$
. It remains to show that
$\partial W\subset (X\setminus Y)\cup S$
. Note the following:
-
•
$\partial V\subset A_1\cup S'\cup A_3\cup K$
; -
•
$\overline W$
misses
$A_1\cup A_3$
because
$W\subset U_2$
; -
•
$\overline W$
misses
$S'$
because it lies in the closed disc bounded by S.
Hence,
$\overline {W}\cap \partial V\subset K\setminus \{z_1,z_3\}\subset X\setminus Y.$
We conclude that
$$ \begin{align*} \partial W&\subset \overline W\cap \big[\partial U_2\cup \partial U \cup \partial V\cup \partial (X\setminus S)\big] \\ &\subset \partial U_2\cup \partial U \cup \big(\overline W\cap \partial V\big)\cup \partial (X\setminus S)\\ &\subset (X\setminus Y)\cup S.\end{align*} $$
This completes the proof of Claim 4.
Claim 5 Let U be an open subset of X intersecting Z with
$\partial U\subset (X\setminus Y)\cup S$
. Then there are connected open subsets
$W_1$
and
$W_2$
of U each intersecting Z, with disjoint closures, such that
$\partial W_i\subset (X\setminus Y)\cup S$
.
Proof of Claim 5 Let
$z_1,z_2,z_3,z_4\in Z\cap U$
. Let
$U_i\in \mathcal U_{z_i}$
be pairwise disjoint and path-connected. Let
$A_i$
be an arc in
$U_i$
from
$z_i$
to
$a_i\in S'$
, intersecting
$S'$
only at
$a_i$
. We may assume that the
$a_i$
’s are in cyclic order. Let
$V_i$
be the component of
$z_i$
in
$U_i\cap U\setminus S$
. Assume that
$\overline {V_1}$
and
$ \overline {V_3}$
have a common point p (if they are disjoint then we’re done). Note that
$p\notin U_2\cup U_4$
, so
$p\notin A_2\cup A_4\cup V_2\cup V_4$
.
By Claim 4, there exist
$z_2',z_4'\in Z$
and connected open subsets
$W_2$
and
$W_4$
of
$V_2$
and
$V_4$
whose closures miss p, such that
$z_i'\in W_i$
and
$\partial W_i\subset (X\setminus Y)\cup S$
. For each
$i\in \{2,4\}$
, working within
$V_i$
we can modify
$A_i$
to an arc
$A_i'$
that ends at
$z_i'$
instead of
$z_i$
. The other endpoint of
$A_i'$
is still
$a_i$
, and
$p\notin A_i'$
. Let V be a connected neighborhood of p with closure missing
$\overline {W_2}\cup \overline {W_4}\cup S'\cup A_2'\cup A_4'$
. For each
$i\in \{1,3\}$
let
$B_i$
be an arc in
$V_i$
from
$z_i$
into V. Let
$K=B_1\cup B_3\cup \overline V$
. By Lemma 2,
$z_2'$
and
$z_4'$
are in different components of
$\mathbb R^2\setminus (A_1\cup S'\cup A_3\cup K)$
. The continua
$\overline {W_2}$
and
$\overline {W_4}$
are contained in these components, hence they are disjoint.
By Claim 5, there are two connected open sets
$W_{\langle 0\rangle }$
and
$W_{\langle 1\rangle }$
in
$X\setminus S$
, intersecting Z, with disjoint closures and boundaries in
$(X\setminus Y)\cup S$
. Assuming that
$\alpha \in 2^{<\mathbb N}$
is a finite binary sequence and
$W_{\alpha }$
has been defined, apply Claim 5 to get connected open subsets
$W_{\alpha ^\frown 0}$
and
$W_{\alpha ^\frown 1}$
of
$W_\alpha $
, each intersecting Z, with disjoint closures and boundaries in
$(X\setminus Y)\cup S$
. Their boundaries must meet S because each
$F(z)$
meets
$S'$
. So for every infinite sequence
$\alpha \in 2^{\mathbb N}$
,
is a continuum in X stretching from Z to S. Clearly if
$\alpha \neq \beta $
then
$K_{\alpha }$
and
$K_{\beta }$
are disjoint. Therefore,
$\{K_\alpha :\alpha \in 2^{\mathbb N}\}$
is an uncountable collection of pairwise disjoint nondegenerate subcontinua of X, a contradiction to the Suslinian property of X. Hence,
$\Lambda (Y)$
must have been countable. This completes the proof of Theorem 3.
Theorem A is a direct consequence of Theorem 3 and:
Theorem 6 Every Suslinian plane continuum is contained in a locally connected Suslinian plane continuum.
Proof Let X be a Suslinian plane continuum. Let
$U_0,U_1,U_2,\ldots $
be the connected components of
$\mathbb R^2\setminus X$
with
$U_0$
unbounded. In each domain
$U_m$
, there is a sequence
$S^m_0,S^m_1\ldots $
of disjoint, concentric simple closed curves limiting to the boundary of
$U_m$
. Let
$A^0_0=\varnothing $
, and for each
$m\geq 1$
let
$A^m_0$
be the closed topological disc in
$U_m$
that is bounded by
$S^m_0$
. For each
$n\geq 1$
, let
$A^m_n$
be the closed annulus in
$U_m$
that is bounded by
$S^m_{n-1}$
and
$S^m_{n}$
. In
$A^m_n$
there is a finite collection of arcs
$\mathcal I^m_n$
covering
$\partial A^m_n$
such that each component of
$A^m_n\setminus \bigcup \mathcal I^m_n$
has diameter less than
$1/(m+n)$
. We can easily arrange that the boundaries of these components are simple closed curves, and
$X^m_n=\bigcup \mathcal I^m_n$
is connected. For the reader who’d like more details, one can assume that X is a subset of the Riemann sphere
$\hat {\mathbb C}=\mathbb C\cup \{\infty \}$
, so that all complementary components of X are simply connected. Let
$\mathbb D=\{z\in \mathbb C:|z|<1\}$
. Given a component U of
$\hat {\mathbb C}\setminus X$
, apply the Riemann mapping theorem to get a continuous bijection
$f:\mathbb D\to U$
and transfer the circles
$\{z\in \mathbb C:|z|=1-2^{-n}\}$
to U to get the curves
$S_n$
. Apply uniform continuity on each compact annulus
$ \{z\in \mathbb C : 1-2^{-n} \leq |z| \leq 1-2^{-n-1}\}$
to divide it further into a grid whose cells have images with a small enough diameter. See Figure 2.
Dividing a domain into cells via a conformal mapping of
$\mathbb D$
.

Put
The components of
$\mathbb R^2\setminus X'$
form a null sequence of domains whose boundaries are simple closed curves, so the continuum
$X'$
is locally connected [Reference Whyburn13, VI Theorem 4.4]. Moreover,
$X'$
is Suslinian because
$X' \setminus X$
is a countable union of arcs.
4 Examples
First, we describe a locally connected extension of the continuum in [Reference van Mill and Tuncali9] with the same buried set, which is the totally disconnected and weakly
$1$
-dimensional space by Kuratowski [Reference Kuratowski6], [Reference Engelking3, Exercise 1.2.E].
Remark 2 Kuratowski’s space K is the graph of a particular function
$f:C\to [-1,1]$
defined on the middle-thirds Cantor set C. It has a simple algebraic formula based on binary representations of members of C (see §4 of [Reference van Mill and Tuncali9]). Clearly K is totally disconnected, and it is weakly
$1$
-dimensional by the following. Let
$C_1$
be the countable set consisting of
$0$
and all right endpoints of the intervals
removed from
$[0,1]$
to obtain C. For every
$x\in C_1$
, it happens that K is
$1$
-dimensional at
$\langle x,f(x)\rangle $
. On the other hand, f is continuous at each point of
$C_0=C\setminus C_1$
. Hence, if
$x\in C_0$
, then K is zero-dimensional at
$\langle x,f(x)\rangle $
. Therefore,
and K is weakly
$1$
-dimensional.
See Figure 3 for an illustration of K and its closure in
$C\times [-1,1]$
.
The closure of K consists of the points of K and vertical arcs through points of
$\Lambda (K)$
.

Example 1 There exists a locally connected plane continuum whose buried set is totally disconnected and
$1$
-dimensional.
Proof Let Z be the plane continuum from [Reference van Mill and Tuncali9] which has
$\operatorname {\mathrm {bur}}(Z)=K$
. By construction,
$\mathbb R^2 \setminus Z $
is a countable union of disjoint connected open sets with simple closed curves as boundaries. One complementary component of Z, say
$W_0$
, is unbounded and the others, say
$W_1, W_2, \ldots $
, form a sequence of bounded domains. Let
$\mathcal I_n$
, for
$n \geq 1$
, be a finite set of arcs in
$\overline {W_n}$
such that the collection
$\mathcal V_n$
of components of
$W_n \setminus \bigcup \mathcal I_n$
has mesh less than
$1/n$
. We can easily arrange that the boundary of every element of
$\mathcal V_n$
is a simple closed curve, and
$Z_n=\partial W_n\cup \bigcup \mathcal I_n$
is connected. Here one can even apply Carathéodory’s theorem to get
$Z_n$
from a grid in the closed unit disc. As in Theorem 6,
is a locally connected continuum. Clearly,
$\operatorname {\mathrm {bur}}(Z) \subset \operatorname {\mathrm {bur}}(Z')$
. For the other inclusion, note that points in
$Z' \setminus Z$
have finite graph neighborhoods in
$Z'$
, so are not in
$\operatorname {\mathrm {bur}}(Z')$
. Now suppose
$z\in Z\setminus \operatorname {\mathrm {bur}}(Z)$
. Then
$z\in \partial W_n$
for some n, and there exists
$V\in \mathcal V_n$
such that
$z \in \partial V$
. So
$z\notin \operatorname {\mathrm {bur}}(Z')$
. We conclude that
$\operatorname {\mathrm {bur}}(Z') = \operatorname {\mathrm {bur}}(Z)=K$
.
The next example shows that Theorem A is false outside of the plane. By a dendroid, we shall mean a hereditarily unicoherent, arcwise-connected continuum.
Example 2 There exists a Suslinian dendroid whose endpoint set is totally disconnected and
$1$
-dimensional at every point.
We briefly indicate the construction of such an example below. For further details, see [Reference Lipham8, Section 5.3].
Proof
Begin with a Lelek function
$\varphi :C\to [0,1]$
whose positive graph
is
$1$
-dimensional at every point [Reference Dijkstra and van Mill5]. Let
The quotient of L that is obtained by shrinking
$C\times \{0\}$
to a point is a continuum known as the Lelek fan. We will identify arcs in
$L\setminus E$
to produce a Suslinian continuum (a dendroid, in fact) that homeomorphically contains E.
For each
$n=1,2,3,\ldots $
let
$\mathscr C_n$
be the natural partition of C into
$2^n$
pairwise disjoint intervals of length
$3^{-n}$
. For each
$i=1,\ldots ,2^n-1$
define
Put
$\langle x,0\rangle \sim \langle x',0\rangle $
for all
$x,x'\in C$
. If
$y>0$
then define
$\langle x,y\rangle \sim \langle x',y\rangle $
if there exists
$n,i$
such that
$x,x'\in A_{n,i}$
belong to the same member of
$\mathscr C_n$
, and
$y\leq i/2^n$
. It is possible to see that
$\sim $
is an equivalence relation, the equivalence classes under
$\sim $
form an upper semicontinuous decomposition of L, and the quotient
$D=L/\sim $
is a Suslinian dendroid with endpoint set E.
Remark 3 The endpoint set of a smooth dendroid is always
$G_{\delta }$
; therefore, E is Borel and Polish.





