Introduction
The initial motivation for this paper comes from operator theory. It follows from von Neumann’s classical double commutant theorem (see [35 ] ) that if ( Ω , Σ , μ ) (\Omega,\Sigma,\mu) ( Ω , Σ , μ ) is a measure space, f f f is a real-valued function from L ∞ ( μ ) L^\infty(\mu) L ∞ ( μ ) , and M f M_f M f is the operator of multiplication by f f f on L 2 ( μ ) L^2(\mu) L 2 ( μ ) then the double commutant { M f } ′ ′ \{M_f\}^{\prime\prime} { M f } ′′ of M f M_f M f coincides with the closure in the strong operator topology of the algebra generated by M f M_f M f and the identity operator I I I .
De Pagter and Ricker [38 ] extended the above result to the spaces L p ( μ ) L^p(\mu) L p ( μ ) , 1 ≤ p < ∞ 1 \le p < \infty 1 ≤ p < ∞ , over Maharam measure spaces, and showed that on L ∞ [ 0 , 1 ] L^\infty[0,1] L ∞ [ 0 , 1 ] the double commutant of the operator of multiplication by the independent variable consists of the multiplications by Riemann integrable functions, so it is strictly larger than the strongly closed algebra generated by the operator and the identity, which coincides with the algebra of multiplications by functions from C [ 0 , 1 ] C[0,1] C [ 0 , 1 ] .
In view of these results it is natural to consider the following problem. Let K K K be a compact Hausdorff space, let C ( K ) C(K) C ( K ) be the space of all real-valued continuous functions on K K K , and for f ∈ C ( K ) f \in C(K) f ∈ C ( K ) let A f A_f A f be the closure of the algebra generated by the multiplication operator M f M_f M f on C ( K ) C(K) C ( K ) and the identity. What properties of f f f and K K K ensure that { M f } ′ ′ = A f \{M_f\}^{\prime\prime}=A_f { M f } ′′ = A f ?
A significant part of this problem is the following one: which properties of K K K ensure that { M f } ′ ′ = A f \{M_f\}^{\prime\prime}=A_f { M f } ′′ = A f for every f ∈ C ( K ) f \in C(K) f ∈ C ( K ) ? Clearly this property is a topological invariant of K K K , and we say that such a K K K has the double commutant property , or K ∈ D C P K \in DCP K ∈ D C P .
We restrict our attention to the case when K K K is a continuum (a compact metrizable connected space), and only indicate that by virtue of the results of [38 ] the maximal ideal space of algebra L ∞ [ 0 , 1 ] L^\infty[0,1] L ∞ [ 0 , 1 ] is not in D C P \mathcal{DCP} D C P , and that some results concerning metrizable compact spaces with isolated points are contained in [23 Theorem 1.15] .
The results obtained by the author in [23 ] and in [24 ] can be briefly summarized as follows.
If a metrizable compact space without isolated points is in D C P \mathcal{DCP} D C P , then it is a continuum.
Every locally connected continuum is in D C P \mathcal{DCP} D C P .
There are continua that are not in D C P \mathcal{DCP} D C P , e.g. K = cl { ( x , sin ( 1 / x ) ) : 0 < x ≤ 1 } K=\operatorname{cl}\{(x,\sin(1/x)):0<x\leq1\} K = cl {( x , sin ( 1/ x )) : 0 < x ≤ 1 } .
There are arc connected but not locally connected continua, e.g. the broom and the bookcase, that are in D C P \mathcal{DCP} D C P ; it is not known whether every arc connected continuum (in particular the Cantor fan) is in D C P \mathcal{DCP} D C P .
There are continua that are not arc connected but are in D C P \mathcal{DCP} D C P , for example K = ( [ − 1 , 0 ] × [ − 1 , 1 ] ) ∪ cl { ( x , sin ( 1 / x ) ) : 0 < x ≤ 1 } K=([-1,0]\times[-1,1])\cup\operatorname{cl}\{(x,\sin(1/x)):0<x\leq1\} K = ([ − 1 , 0 ] × [ − 1 , 1 ]) ∪ cl {( x , sin ( 1/ x )) : 0 < x ≤ 1 } .
In the current paper, while not arriving at a complete description of the continua in D C P \mathcal{DCP} D C P , we considerably strengthen the results of [23 ] and [24 ] . Below we state the main results; the definitions needed for them are given here and recalled where they are used.
The topological reformulation. We write f ∈ D ( K ) f\in\mathcal{D}(K) f ∈ D ( K ) if { M f } ′ ′ = A f \{M_f\}^{\prime\prime}=A_f { M f } ′′ = A f ; thus K ∈ D C P K\in\mathcal{DCP} K ∈ D C P if and only if D ( K ) = C ( K ) \mathcal{D}(K)=C(K) D ( K ) = C ( K ) . The following notions translate the problem into topology.
Definition 1.1. Let K K K be a compact metrizable space and f ∈ C ( K ) f\in C(K) f ∈ C ( K ) .
(i) A map Ψ \Psi Ψ from K K K into the nonempty subsets of K K K is lower semicontinuous (lsc) if { k : Ψ ( k ) ∩ V ≠ ∅ } \{k:\Psi(k)\cap V\ne\varnothing\} { k : Ψ ( k ) ∩ V = ∅ } is open for every open V ⊆ K V\subseteq K V ⊆ K .
(ii) M f M_f M f denotes the largest lsc map with k ∈ M f ( k ) ⊆ f − 1 ( f ( k ) ) k\in M_f(k)\subseteq f^{-1}(f(k)) k ∈ M f ( k ) ⊆ f − 1 ( f ( k )) for all k ∈ K k\in K k ∈ K (it exists; see Definition 2.1).
(iii) A function g ∈ C ( K ) g\in C(K) g ∈ C ( K ) is f f f -admissible if it is constant on every set M f ( k ) M_f(k) M f ( k ) .
Every function of the form h ∘ f h\circ f h ∘ f is f f f -admissible. The point of the definition is the converse.
Theorem A (Theorem 2.6 ). Let K K K be a compact metrizable space and f , g ∈ C ( K ) f,g\in C(K) f , g ∈ C ( K ) . Then M g ∈ { M f } ′ ′ M_g\in\{M_f\}^{\prime\prime} M g ∈ { M f } ′′ if and only if g g g is f f f -admissible. Consequently f ∈ D ( K ) f\in\mathcal{D}(K) f ∈ D ( K ) if and only if every f f f -admissible function is constant on the fibres of f f f , and K ∈ D C P K\in\mathcal{DCP} K ∈ D C P if and only if this holds for every f ∈ C ( K ) f\in C(K) f ∈ C ( K ) .
The argument for Theorem A goes back to [23 ] , and it is closely related to Ditor’s theory of averaging operators and lower semicontinuous sections [14 ] (Remark 2.7). Operator theory enters only through this theorem; from Section 3 on, all arguments are topological.
Local structure: locally connected points and teeth. Our sufficient conditions are expressed in terms of two kinds of local structure.
Definition 1.2. For a continuum K K K , L ( K ) L(K) L ( K ) denotes the set of points of K K K that have an open locally connected neighbourhood. It is an open subset of K K K .
For a locally connected continuum L ( K ) = K L(K)=K L ( K ) = K ; for the Cantor fan L ( K ) = ∅ L(K)=\varnothing L ( K ) = ∅ . Where L ( K ) L(K) L ( K ) is not dense we use instead a product structure along arcs.
Definition 1.3. A tooth chart of a continuum K K K is a continuous map h : U × [ 0 , 1 ] → K h:U\times[0,1]\to K h : U × [ 0 , 1 ] → K , U U U a nonempty open subset of a compact metrizable space, together with an interval J h = ( 0 , 1 ) J_h=(0,1) J h = ( 0 , 1 ) or J h = ( 0 , 1 ] J_h=(0,1] J h = ( 0 , 1 ] , such that h h h maps U × J h U\times J_h U × J h homeomorphically onto an open subset O h O_h O h of K K K . The arcs h ( { y } × J h ) h(\{y\}\times J_h) h ({ y } × J h ) are the open teeth, and h ( y , 0 ) h(y,0) h ( y , 0 ) is the root of the tooth. Given a family of tooth charts, a point v ∈ K v\in K v ∈ K is a star point if it has arbitrarily small open neighbourhoods S S S such that every p ∈ S ∖ { v } p\in S\setminus\{v\} p ∈ S ∖ { v } is joined to v v v by an arc A ⊆ S A\subseteq S A ⊆ S with A ∖ { v } A\setminus\{v\} A ∖ { v } contained in one open tooth.
When U U U is a clopen subset of the Cantor set and J h = ( 0 , 1 ) J_h=(0,1) J h = ( 0 , 1 ) , the set O h O_h O h is a Cantor box (§ 3 ), an open piece of a Cantor fence C × [ 0 , 1 ] C\times[0,1] C × [ 0 , 1 ] in the sense of [27 ] . The teeth of the Cantor comb are its vertical segments, rooted on the base; the teeth of a cone Cone ( Y ) \operatorname{Cone}(Y) Cone ( Y ) are the segments { y } × ( 0 , 1 ] \{y\}\times(0,1] { y } × ( 0 , 1 ] , all rooted at the vertex, which is a star point. Star points are thus modelled on the top of a fan [6 ] , at which the legs of the fan meet (see [3 ] ). Note, however, that tooth charts require open product neighbourhoods, which a general fan need not have: in the Lelek fan the end points are dense, and no nonempty open set is of the form O h O_h O h (an end point of the fan lying in O h O_h O h would have to be of the form h ( y , 1 ) h(y,1) h ( y , 1 ) with J h = ( 0 , 1 ] J_h=(0,1] J h = ( 0 , 1 ] , so the nonempty open set h ( U × ( 0 , 1 ) ) h(U\times(0,1)) h ( U × ( 0 , 1 )) would contain no end points).
Main results. Our most general sufficient condition is the following; it is proved as Theorem 4.25 together with Theorem 4.15 .
Theorem B. Let K K K be a metrizable continuum with a family of tooth charts, let B B B be the union of the sets O h O_h O h , and let R ⊆ K R\subseteq K R ⊆ K be closed. Assume:
(1) K K K is arc connected, or every arc component P P P of K K K is dense or satisfies P ⊆ cl Int P P\subseteq\operatorname{cl}\operatorname{Int}P P ⊆ cl Int P ;
(2) K = cl L ( K ) ∪ R ∪ B K=\operatorname{cl}L(K)\cup R\cup B K = cl L ( K ) ∪ R ∪ B , and h ( U × { 0 } ) ⊆ R ∪ cl L ( K ) h(U\times\{0\})\subseteq R\cup\operatorname{cl}L(K) h ( U × { 0 }) ⊆ R ∪ cl L ( K ) for every chart h h h ;
(3) every point of ( R ∪ cl L ( K ) ) ∖ ( L ( K ) ∪ B ) (R\cup\operatorname{cl}L(K))\setminus(L(K)\cup B) ( R ∪ cl L ( K )) ∖ ( L ( K ) ∪ B ) has arbitrarily small open neighbourhoods U ′ U' U ′ such that every component of U ′ ∩ T U'\cap T U ′ ∩ T , T T T an open tooth, has an end point in U ′ ∖ B U'\setminus B U ′ ∖ B (an end point of h ( { y } × I ) h(\{y\}\times I) h ({ y } × I ) , I I I an interval, is a point h ( y , α ) h(y,\alpha) h ( y , α ) with α \alpha α an end point of I I I );
(4) every point of R R R lies in cl L ( K ) \operatorname{cl}L(K) cl L ( K ) or is a star point.
Then K ∈ D C P K\in\mathcal{DCP} K ∈ D C P , and so is every image of K K K under an open continuous map.
Condition (3) is a mild local condition at the points outside the teeth; (2) and (4) say that every tooth is rooted either in the closure of L ( K ) L(K) L ( K ) or at a star point. Theorem B contains the following special cases.
With no charts at all, its hypotheses say that L ( K ) L(K) L ( K ) is dense and (1) holds (Theorem 4.7 ); in particular every arc connected continuum with L ( K ) L(K) L ( K ) dense is in D C P \mathcal{DCP} D C P (Theorem 4.4 ). This contains the locally connected continua [23 ] and the broom and the bookcase of [24 ] .
The Cantor comb M = ( C × [ 0 , 1 ] ) ∪ ( [ 0 , 1 ] × { 0 } ) M=(C\times[0,1])\cup([0,1]\times\{0\}) M = ( C × [ 0 , 1 ]) ∪ ([ 0 , 1 ] × { 0 }) , in which L ( M ) L(M) L ( M ) is not dense, and the cone over every nonempty compact metrizable space, in particular the Cantor fan F F F , in which L ( F ) = ∅ L(F)=\varnothing L ( F ) = ∅ , are in D C P \mathcal{DCP} D C P (Theorem 4.22 ); this answers two questions of [24 ] . So are the Cantor ladder and the suspension of every nonempty compact metrizable space (Remark 4.24).
Every smooth fan with a closed set of end points is an open image of the Cantor fan [10 ] , hence in D C P \mathcal{DCP} D C P (Corollary 4.23).
If K × Y ∈ D C P K\times Y\in\mathcal{DCP} K × Y ∈ D C P for some nonempty compact metrizable Y Y Y , then K ∈ D C P K\in\mathcal{DCP} K ∈ D C P (Corollary 4.16).
The Cantor grid is shown to be in D C P \mathcal{DCP} D C P in this paper but is not covered by Theorem B, for any family of tooth charts: the points where its arcs cross lie in no open tooth and are neither in cl L ( G ) = ∅ \operatorname{cl}L(G)=\varnothing cl L ( G ) = ∅ nor star points (Remark 4.26). An additional argument at these points gives the following.
Theorem C (Theorem 4.31 ). The Cantor grid G = ( C × [ 0 , 1 ] ) ∪ ( [ 0 , 1 ] × C ) G=(C\times[0,1])\cup([0,1]\times C) G = ( C × [ 0 , 1 ]) ∪ ([ 0 , 1 ] × C ) is in D C P \mathcal{DCP} D C P , although L ( G ) = ∅ L(G)=\varnothing L ( G ) = ∅ .
The hypotheses of Theorem B cannot simply be dropped. By Proposition 4.36 , no compactification of a ray, or of two disjoint open rays, with a nondegenerate remainder is in D C P \mathcal{DCP} D C P (for instance the sin ( 1 / x ) \sin(1/x) sin ( 1/ x ) continuum or a ray spiralling onto a circle), although adding the disk bounded by the circle gives a continuum in D C P \mathcal{DCP} D C P (Example 4.12 ). Example 4.38 shows that arc connectedness cannot be weakened to the condition that every point can be joined by an arc to L ( K ) L(K) L ( K ) , and Example 4.39, in which L ( K ) L(K) L ( K ) is dense, that the condition on arc components in (1) cannot be dropped. Indecomposability does not help either: by a theorem of Fukaishi and Matsuhashi [18 Theorem 1.1] and Theorem 4.15 , there are indecomposable, even tree-like indecomposable [30 ] , continua not in D C P \mathcal{DCP} D C P (Remark 5.13).
Individual functions. For continua not covered by Theorem B, such as solenoids, the buckethandle continuum and the Lelek fan, we show that many individual functions belong to D ( K ) \mathcal{D}(K) D ( K ) .
Theorem D. Let K K K be a metrizable continuum.
(i) If all arc components of K K K are dense, then every f ∈ C ( K ) f \in C(K) f ∈ C ( K ) that is constant on no arc belongs to D ( K ) \mathcal{D}(K) D ( K ) (Theorem 5.1 ).
(ii) If there are retractions r n r_n r n of K K K onto subsets in D C P \mathcal{DCP} D C P with r n → id K r_n \to\operatorname{id}_K r n → id K uniformly, then φ ∘ r n ∈ D ( K ) \varphi\circ r_n \in\mathcal{D}(K) φ ∘ r n ∈ D ( K ) for all φ ∈ C ( K ) \varphi\in C(K) φ ∈ C ( K ) ; in particular D ( K ) \mathcal{D}(K) D ( K ) is dense in C ( K ) C(K) C ( K ) (Proposition 5.4 ).
Part (i) applies to arc connected continua, solenoids and the buckethandle continuum. Part (ii) applies, through the small retractions onto trees and finite fans constructed by Fugate [20 , 19 ] , to every smooth dendroid and every fan, for instance the Lelek fan (Corollaries 5.6 and 5.7 ); such approximations are what Eberhart and Fugate call approximation from within [16 ] . Density of D ( K ) \mathcal{D}(K) D ( K ) is, however, no evidence that K ∈ D C P K \in\mathcal{DCP} K ∈ D C P : the sin ( 1 / x ) \sin(1/x) sin ( 1/ x ) continuum admits small retractions onto arcs, so D ( K ) \mathcal{D}(K) D ( K ) is dense for it, yet it is not in D C P \mathcal{DCP} D C P (Remark 5.8).
Open problems and organization. The basic question of [24 ] , whether every arc connected continuum is in D C P \mathcal{DCP} D C P , remains open; so does the question whether every dendroid, or the Lelek fan, is in D C P \mathcal{DCP} D C P . These and related questions on retracts, continuous images and inverse limits are collected in § 11 . § 2 proves Theorem A; § 3 derives collapsed levels from local structure; § 5 proves Theorems B and C and contains the examples and counterexamples; § 10 deals with individual functions, including Theorem D , solenoids, the buckethandle continuum and fans.
Notation and terminology. Spaces and functions. Except in the first paragraphs of the Introduction, K K K is a compact metrizable space with a fixed metric d d d . C ( K ) C(K) C ( K ) is the Banach space of real-valued continuous functions on K K K with the supremum norm. For a set A ⊆ K A \subseteq K A ⊆ K , cl A \operatorname{cl} A cl A and Int A \operatorname{Int} A Int A denote its closure and interior. For a measure μ ∈ C ( K ) ∗ \mu\in C(K)^* μ ∈ C ( K ) ∗ , supp μ \operatorname{supp}\mu supp μ is its support. C \mathcal{C} C is the ternary Cantor set in [ 0 , 1 ] [0,1] [ 0 , 1 ] .
Operators. L ( C ( K ) ) L(C(K)) L ( C ( K )) is the algebra of bounded linear operators on C ( K ) C(K) C ( K ) . For f ∈ C ( K ) f \in C(K) f ∈ C ( K ) , M f ∈ L ( C ( K ) ) M_f \in L(C(K)) M f ∈ L ( C ( K )) is the operator of multiplication by f f f . For a set B ⊆ L ( C ( K ) ) \mathcal{B} \subseteq L(C(K)) B ⊆ L ( C ( K )) , B ′ \mathcal{B}' B ′ is its commutant and B ′ ′ = ( B ′ ) ′ \mathcal{B}'' = (\mathcal{B}')' B ′′ = ( B ′ ) ′ its double commutant. A f A_f A f is the closure of the algebra generated by M f M_f M f and the identity I I I ; it is the same in the operator norm and in the strong operator topology [23 Corollary 1.2] . We write f ∈ D ( K ) f \in\mathcal{D}(K) f ∈ D ( K ) if { M f } ′ ′ = A f \{M_f\}'' = A_f { M f } ′′ = A f , and K ∈ D C P K \in\mathcal{DCP} K ∈ D C P if D ( K ) = C ( K ) \mathcal{D}(K) = C(K) D ( K ) = C ( K ) .
Fibres. For f ∈ C ( K ) f \in C(K) f ∈ C ( K ) and t ∈ R t \in\mathbb{R} t ∈ R we write Z t = f − 1 ( t ) Z_t = f^{-1}(t) Z t = f − 1 ( t ) , and Φ f ( k ) = Z f ( k ) \Phi_f(k) = Z_{f(k)} Φ f ( k ) = Z f ( k ) is the fibre of f f f through k ∈ K k \in K k ∈ K . When several functions are involved, the dependence of Z t Z_t Z t on f f f is clear from the context.
Lower semicontinuous maps. A map Ψ \Psi Ψ from K K K into the nonempty subsets of K K K is lower semicontinuous (lsc) if { k : Ψ ( k ) ∩ V ≠ ∅ } \{k : \Psi(k) \cap V \ne\varnothing\} { k : Ψ ( k ) ∩ V = ∅ } is open for every open V ⊆ K V \subseteq K V ⊆ K ; equivalently, for every k 0 k_0 k 0 , every s ∈ Ψ ( k 0 ) s \in\Psi(k_0) s ∈ Ψ ( k 0 ) and every sequence k n → k 0 k_n \to k_0 k n → k 0 there are s n ∈ Ψ ( k n ) s_n \in\Psi(k_n) s n ∈ Ψ ( k n ) (for large n n n ) with s n → s s_n \to s s n → s . For f ∈ C ( K ) f \in C(K) f ∈ C ( K ) , M f M_f M f also denotes the largest lsc map with k ∈ M f ( k ) ⊆ Φ f ( k ) k \in M_f(k) \subseteq\Phi_f(k) k ∈ M f ( k ) ⊆ Φ f ( k ) for all k k k (Definition 2.1). Following [23 ] we use the same letter as for the multiplication operator; the sets M f ( k ) M_f(k) M f ( k ) always appear with an argument, so no confusion should arise.
Admissible functions. A function g ∈ C ( K ) g \in C(K) g ∈ C ( K ) is f f f -admissible if it is constant on every set M f ( k ) M_f(k) M f ( k ) . Every function h ∘ f h \circ f h ∘ f , h ∈ C ( f ( K ) ) h \in C(f(K)) h ∈ C ( f ( K )) , is f f f -admissible. By Theorem 2.6 , g g g is f f f -admissible if and only if M g ∈ { M f } ′ ′ M_g \in\{M_f\}^{\prime\prime} M g ∈ { M f } ′′ ; hence f ∈ D ( K ) f \in D(K) f ∈ D ( K ) if and only if every f f f -admissible function is constant on the fibres of f f f .
Continua. A continuum is a connected compact metrizable space; it is nondegenerate if it has more than one point. An arc is a homeomorphic image of [ 0 , 1 ] [0,1] [ 0 , 1 ] ; its relative interior is the image of ( 0 , 1 ) (0,1) ( 0 , 1 ) . A space is arc connected if any two of its points are joined by an arc, and the arc components of a space are its maximal arc connected subsets. L ( K ) L(K) L ( K ) (Definition 1.2) is the set of points of K K K that have an open locally connected neighbourhood; it is an open subset of K K K .
Dendroids and fans. A dendroid is an arc connected hereditarily unicoherent continuum; any two of its points a , b a,b a , b are joined by a unique arc a b ab ab . A dendroid K K K is smooth if there is a point p ∈ K p \in K p ∈ K (an initial point ) such that a n → a a_n \to a a n → a implies p a n → p a pa_n \to pa p a n → p a in the Hausdorff metric. A fan is a dendroid with exactly one ramification point v v v , its vertex (a point that is the common end point of three arcs, pairwise disjoint except for that point); a fan is smooth if it is smooth with initial point v v v , and a point of a fan is an end point if it is an end point of every arc containing it.
Further notions (crossing points, Cantor boxes) are defined where they are first used.
Lower semicontinuous submaps and the double commutant
Definition 2.1. Let M f \mathcal{M}_f M f be the family of lsc maps Ψ \Psi Ψ with k ∈ Ψ ( k ) ⊆ Φ f ( k ) k \in\Psi(k) \subseteq\Phi_f(k) k ∈ Ψ ( k ) ⊆ Φ f ( k ) for all k k k . Put
M f ( k ) = cl ⋃ Ψ ∈ M f Ψ ( k ) . M_f(k)=\operatorname{cl}\bigcup_{\Psi\in\mathcal{M}_f}\Psi(k). M f ( k ) = cl Ψ ∈ M f ⋃ Ψ ( k ) .
A union of lsc maps is lsc, and so is the closure of an lsc map; since the fibres are closed, M f ∈ M f M_f \in\mathcal{M}_f M f ∈ M f , and M f M_f M f is the largest element of M f \mathcal{M}_f M f . Thus, to show that a point s s s lies in M f ( u ) M_f(u) M f ( u ) , it suffices to find some Ψ ∈ M f \Psi\in\mathcal{M}_f Ψ ∈ M f with s ∈ Ψ ( u ) s \in\Psi(u) s ∈ Ψ ( u ) .
The double commutant. Let C ( K ) C(K) C ( K ) be the Banach space of real-valued continuous functions on K K K . For f ∈ C ( K ) f \in C(K) f ∈ C ( K ) the closure of the algebra generated by M f M_f M f and I I I is the same in the operator norm and in the strong operator topology [23 Corollary 1.2] ; we denote it by A f \mathcal{A}_f A f .
Lemma 2.2 ([23 Lemmas 1.1, 1.8] ). Every T ∈ L ( C ( K ) ) T \in L(C(K)) T ∈ L ( C ( K )) has the form ( T h ) ( k ) = ∫ h d μ k (Th)(k)=\int h\,\mathrm{d}\mu_k ( T h ) ( k ) = ∫ h d μ k , where k ↦ μ k ∈ C ( K ) ∗ k \mapsto\mu_k \in C(K)^* k ↦ μ k ∈ C ( K ) ∗ is weak* continuous and bounded, and conversely. Moreover T M f = M f T TM_f=M_fT T M f = M f T if and only if supp ∣ μ k ∣ ⊆ Φ f ( k ) \operatorname{supp}|\mu_k| \subseteq\Phi_f(k) supp ∣ μ k ∣ ⊆ Φ f ( k ) for all k ∈ K k \in K k ∈ K , and every operator in { M f } ′ ′ \{M_f\}^{\prime\prime} { M f } ′′ is a multiplication operator M g M_g M g , g ∈ C ( K ) g \in C(K) g ∈ C ( K ) .
Proof. Only the commutation criterion needs a comment: ( T M f h ) ( k ) − ( M f T h ) ( k ) = ∫ ( f − f ( k ) ) h d μ k (TM_fh)(k)-(M_fTh)(k)=\int(f-f(k))h\,\mathrm{d}\mu_k ( T M f h ) ( k ) − ( M f T h ) ( k ) = ∫ ( f − f ( k )) h d μ k , which vanishes for all h h h iff ( f − f ( k ) ) μ k = 0 (f-f(k))\mu_k=0 ( f − f ( k )) μ k = 0 , i.e. iff ∣ μ k ∣ |\mu_k| ∣ μ k ∣ is concentrated on the closed set Φ f ( k ) \Phi_f(k) Φ f ( k ) . □
If g ∈ C ( K ) g \in C(K) g ∈ C ( K ) is constant on the fibres of f f f , then h ( f ( k ) ) : = g ( k ) h(f(k)):=g(k) h ( f ( k )) := g ( k ) is a well-defined function on f ( K ) f(K) f ( K ) with g = h ∘ f g=h\circ f g = h ∘ f , and h h h is continuous: for a closed set F ⊆ R F \subseteq\mathbb{R} F ⊆ R we have h − 1 ( F ) = f ( g − 1 ( F ) ) h^{-1}(F)=f(g^{-1}(F)) h − 1 ( F ) = f ( g − 1 ( F )) , which is compact, being a continuous image of the compact set g − 1 ( F ) g^{-1}(F) g − 1 ( F ) . (Equivalently, f : K → f ( K ) f:K \to f(K) f : K → f ( K ) is a closed continuous surjection and hence a quotient map; see [17 §2.4 and Theorem 3.1.12] .) Thus the functions in C ( K ) C(K) C ( K ) constant on the fibres of f f f are exactly the functions h ∘ f h\circ f h ∘ f , h ∈ C ( f ( K ) ) h \in C(f(K)) h ∈ C ( f ( K )) ; by the Weierstrass theorem
A f = { M g : g is constant on every fibre of f } . (1) \mathcal{A}_f=\{M_g:g\text{ is constant on every fibre of }f\}. \tag*{(1)} A f = { M g : g is constant on every fibre of f } . (1)
Lemma 2.3 ([14 Proposition 3.1] ). Let k ↦ μ k ∈ C ( K ) ∗ k \mapsto\mu_k \in C(K)^* k ↦ μ k ∈ C ( K ) ∗ be weak* continuous. Then k ↦ { k } ∪ supp ∣ μ k ∣ k \mapsto\{k\}\cup\operatorname{supp}|\mu_k| k ↦ { k } ∪ supp ∣ μ k ∣ is lsc.
Proof. Let V V V be open and V ∩ supp ∣ μ k 0 ∣ ≠ ∅ V \cap\operatorname{supp}|\mu_{k_0}| \ne\varnothing V ∩ supp ∣ μ k 0 ∣ = ∅ . Then μ k 0 \mu_{k_0} μ k 0 does not vanish on the functions supported in V V V , so there is h ∈ C ( K ) h \in C(K) h ∈ C ( K ) with supp h ⊆ V \operatorname{supp}h \subseteq V supp h ⊆ V and μ k 0 ( h ) ≠ 0 \mu_{k_0}(h) \ne0 μ k 0 ( h ) = 0 . By weak* continuity
@$references@ μ k ( h ) ≠ 0 \mu_k(h) \ne0 μ k ( h ) = 0 for k k k near k 0 k_0 k 0 , hence supp ∣ μ k ∣ ∩ V ≠ ∅ \operatorname{supp}|\mu_k| \cap V \ne\varnothing supp ∣ μ k ∣ ∩ V = ∅ . The map k ↦ { k } k \mapsto\{k\} k ↦ { k } is lsc, and a union of lsc maps is lsc.
We shall use Michael’s selection theorem in the following form.
Theorem 2.4 (Michael [33, Theorem 1.2]). Let X X X be paracompact, Δ \Delta Δ a metrizable subset of a complete locally convex space E E E , and Θ \Theta Θ an lsc map from X X X into the nonempty subsets of Δ \Delta Δ with compact values. Then there is a continuous φ : X → E \varphi:X \to E φ : X → E with φ ( x ) ∈ conv Θ ( x ) \varphi(x) \in\operatorname{conv}\Theta(x) φ ( x ) ∈ conv Θ ( x ) for all x x x .
We also use the following lemma due to Michael [31 p. 362, Examples 1.3, 1.3*]
Lemma 2.5. If Θ \Theta Θ is lsc, A ⊆ X A \subseteq X A ⊆ X is closed, and s s s is a continuous selection of Θ ∣ A \Theta|_A Θ ∣ A , then the map equal to { s ( x ) } \{s(x)\} { s ( x )} on A A A and to Θ \Theta Θ off A A A is lsc.
The sets M f ( k ) M_f(k) M f ( k ) were introduced in [23 Definition 1.5, Lemma 1.9] , and the argument for the necessity part of the following theorem is contained in the proof of [23 Theorem 1.14] ; the theorem itself was not stated there. We include the short proof for completeness.
Theorem 2.6. Let K K K be a compact metrizable space and f , g ∈ C ( K ) f,g \in C(K) f , g ∈ C ( K ) . Then M g ∈ { M f } ′ ′ M_g \in\{M_f\}'' M g ∈ { M f } ′′ if and only if g g g is f f f -admissible. Consequently f ∈ D ( K ) f \in D(K) f ∈ D ( K ) if and only if every f f f -admissible function is constant on the fibres of f f f .
Proof. Suppose g g g is constant on each M f ( k ) M_f(k) M f ( k ) and let T ∈ { M f } ′ T \in\{M_f\}' T ∈ { M f } ′ have kernels μ k \mu_k μ k . By Lemmas 2.2 and 2.3, k ↦ { k } ∪ supp ∣ μ k ∣ k \mapsto\{k\} \cup\operatorname{supp}|\mu_k| k ↦ { k } ∪ supp ∣ μ k ∣ belongs to M f \mathcal{M}_f M f , hence supp ∣ μ k ∣ ⊆ M f ( k ) \operatorname{supp}|\mu_k| \subseteq M_f(k) supp ∣ μ k ∣ ⊆ M f ( k ) and ( T M g h ) ( k ) = ∫ g h d μ k = g ( k ) ∫ h d μ k = ( M g T h ) ( k ) (TM_gh)(k) = \int gh\,\mathrm{d}\mu_k = g(k)\int h\,\mathrm{d}\mu_k = (M_gTh)(k) ( T M g h ) ( k ) = ∫ g h d μ k = g ( k ) ∫ h d μ k = ( M g T h ) ( k ) .
Conversely, let M g ∈ { M f } ′ ′ M_g \in\{M_f\}'' M g ∈ { M f } ′′ , k 0 ∈ K k_0 \in K k 0 ∈ K and s 0 ∈ M f ( k 0 ) s_0 \in M_f(k_0) s 0 ∈ M f ( k 0 ) . Let E E E be the algebraic dual of C ( K ) C(K) C ( K ) with the weak* topology (a complete locally convex space) and Δ = { δ s : s ∈ K } ⊆ E \Delta= \{\delta_s:s \in K\} \subseteq E Δ = { δ s : s ∈ K } ⊆ E , which is homeomorphic to K K K . The map Θ 0 ( k ) = { δ s : s ∈ M f ( k ) } \Theta_0(k) = \{\delta_s:s \in M_f(k)\} Θ 0 ( k ) = { δ s : s ∈ M f ( k )} is lsc with compact values, and by Lemma 2.5 so is Θ \Theta Θ , equal to { δ s 0 } \{\delta_{s_0}\} { δ s 0 } at k 0 k_0 k 0 and to Θ 0 \Theta_0 Θ 0 elsewhere. Theorem 2.4 gives a continuous φ : K → E \varphi:K \to E φ : K → E with φ ( k ) \varphi(k) φ ( k ) in the closed convex hull of Θ ( k ) \Theta(k) Θ ( k ) . Every such functional is positive with φ ( k ) ( 1 ) = 1 \varphi(k)(1) = 1 φ ( k ) ( 1 ) = 1 , hence a probability measure, and it vanishes on nonnegative functions vanishing on M f ( k ) M_f(k) M f ( k ) ; so supp φ ( k ) ⊆ M f ( k ) ⊆ Φ f ( k ) \operatorname{supp}\varphi(k) \subseteq M_f(k) \subseteq\Phi_f(k) supp φ ( k ) ⊆ M f ( k ) ⊆ Φ f ( k ) and φ ( k 0 ) = δ s 0 \varphi(k_0) = \delta_{s_0} φ ( k 0 ) = δ s 0 . The operator ( T h ) ( k ) = φ ( k ) ( h ) (Th)(k) = \varphi(k)(h) ( T h ) ( k ) = φ ( k ) ( h ) is bounded on C ( K ) C(K) C ( K ) and commutes with M f M_f M f by Lemma 2.2; hence T M g = M g T TM_g = M_gT T M g = M g T and g ( s 0 ) = ( T M g 1 ) ( k 0 ) = ( M g T 1 ) ( k 0 ) = g ( k 0 ) g(s_0) = (TM_g1)(k_0) = (M_gT1)(k_0) = g(k_0) g ( s 0 ) = ( T M g 1 ) ( k 0 ) = ( M g T 1 ) ( k 0 ) = g ( k 0 ) .
The remaining statements follow from Lemma 2.2 and (1).
Remark 2.7. Theorem 2.6 can also be obtained from Ditor’s theory of regular averaging operators and lower semicontinuous sections [14 ] (see also [32 , 39 , 13 ] ). Let P f = { ( k , s ) ∈ K × K : f ( k ) = f ( s ) } P_f = \{(k,s) \in K \times K:f(k) = f(s)\} P f = {( k , s ) ∈ K × K : f ( k ) = f ( s )} and let π 1 , π 2 : P f → K \pi_1,\pi_2:P_f \to K π 1 , π 2 : P f → K be the coordinate projections. One checks with Lemma 2.2 and [14 Lemma 2.1] that the positive operators T ∈ { M f } ′ T \in\{M_f\}' T ∈ { M f } ′ with T 1 = 1 T1 = 1 T 1 = 1 are exactly the operators h ↦ u ( h ∘ π 2 ) h \mapsto u(h \circ\pi_2) h ↦ u ( h ∘ π 2 ) , where u u u is a regular averaging operator for π 1 \pi_1 π 1 , and that k ↦ { k } × M f ( k ) k \mapsto\{k\} \times M_f(k) k ↦ { k } × M f ( k ) is the largest lsc section Λ π 1 \Lambda_{\pi_1} Λ π 1 of π 1 \pi_1 π 1 in the sense of [14 Section 4] (see [14 Proposition 4.2] ). The sufficiency part of Theorem 2.6 then corresponds to [14 Corollary 3.2 and Proposition 4.2(ii)] , and the necessity part to the proof of [14 Theorem 3.4] (cf. [14 Remark 3.6] ), which produces a regular averaging operator whose measures are supported in a prescribed lsc section; the only additional point is that the section must take a prescribed value at one point, which is provided by Michael’s observation.
Standing conventions. From now on we use only the topological formulation. A function f ∈ C ( K ) f \in C(K) f ∈ C ( K ) and an f f f -admissible function g g g are fixed. A level t t t is collapsed if g g g is constant on Z t Z_t Z t , and Σ \Sigma Σ denotes the set of points of K K K whose level is collapsed. Thus f ∈ D ( K ) f \in D(K) f ∈ D ( K ) if and only if, for every f f f -admissible g g g , every level is collapsed.
Collapsed levels from local structure
By Theorem 2.6 , to prove that f ∈ D ( K ) f \in\mathcal{D}(K) f ∈ D ( K ) we must show that, for every f f f -admissible g g g , every level of f f f is collapsed. This is done in two steps. First, in this section, we obtain a large supply of collapsed levels from local conditions and from retractions compatible with f f f . For instance, if L ( K ) L(K) L ( K ) is dense in K K K , every level in f ( D f ) f(D_f) f ( D f ) is collapsed for a dense set D f D_f D f (Lemma 3.10). Second, in § 5 , we show that under additional global assumptions on K K K , such as arc connectedness, all levels are then collapsed; the main tool is Theorem 4.2 , which passes the information from collapsed levels near c c c to the level c c c itself.
All the conditions in the first part of this section work through the same mechanism: they show that M f ( k ) M_f(k) M f ( k ) is large. Either some k k k satisfies M f ( k ) = Z f ( k ) M_f(k)=Z_{f(k)} M f ( k ) = Z f ( k ) (Lemmas 3.1 , 3.2 ), or a point v v v lies in M f ( u ) M_f(u) M f ( u ) for every u u u in the fibre of v v v (Lemma 3.3 and its special cases, Lemmas 3.6 , 3.8). In both cases g g g is constant on the fibre, since g g g is f f f -admissible.
Openness and crossing points.
Lemma 3.1 (cf. [23 Lemma 1.13] ). If U = Int Z t ≠ ∅ U=\operatorname{Int} Z_t \ne\varnothing U = Int Z t = ∅ , then M f ( k ) = Z t M_f(k)=Z_t M f ( k ) = Z t for k ∈ U k\in U k ∈ U ; in particular the level t t t is collapsed.
Proof. The map equal to Z t Z_t Z t on the open set U U U and to { k } \{k\} { k } elsewhere belongs to M f \mathcal{M}_f M f . □
Lemma 3.1 is the special case U = Int Z t U=\operatorname{Int} Z_t U = Int Z t of the following lemma, since Φ f ≡ Z t \Phi_f\equiv Z_t Φ f ≡ Z t on U U U .
Lemma 3.2. Let U ⊂ K U\subset K U ⊂ K be open and assume that Φ f \Phi_f Φ f is lsc at every point of U U U . Then M f ( k ) = Φ f ( k ) M_f(k)=\Phi_f(k) M f ( k ) = Φ f ( k ) for every k ∈ U k\in U k ∈ U ; in particular every level f ( k ) f(k) f ( k ) , k ∈ U k\in U k ∈ U , is collapsed.
Proof. Let Ψ = Φ f \Psi=\Phi_f Ψ = Φ f on U U U and Ψ ( k ) = { k } \Psi(k)=\{k\} Ψ ( k ) = { k } off U U U . At points of the open set U U U , Ψ \Psi Ψ coincides near the point with Φ f \Phi_f Φ f , hence is lsc there; at a point k 0 ∉ U k_0\notin U k 0 ∈ / U we have Ψ ( k 0 ) = { k 0 } \Psi(k_0)=\{k_0\} Ψ ( k 0 ) = { k 0 } and k ∈ Ψ ( k ) k\in\Psi(k) k ∈ Ψ ( k ) for all k k k . So Ψ ∈ M f \Psi\in\mathcal{M}_f Ψ ∈ M f . □
The next lemma is the main tool of this section.
Lemma 3.3. Let R ⊂ K R\subset K R ⊂ K be such that f ∣ R : R → f ( K ) f|_R:R\to f(K) f ∣ R : R → f ( K ) is an open map. Then s ∈ M f ( u ) s\in M_f(u) s ∈ M f ( u ) for every s ∈ R s\in R s ∈ R and every u ∈ Z f ( s ) u\in Z_{f(s)} u ∈ Z f ( s ) . In particular every level in f ( R ) f(R) f ( R ) is collapsed.
Proof. Let Ψ ( k ) = { k } ∪ ( Φ f ( k ) ∩ R ) \Psi(k)=\{k\}\cup(\Phi_f(k)\cap R) Ψ ( k ) = { k } ∪ ( Φ f ( k ) ∩ R ) ; it suffices to show that Ψ \Psi Ψ is lsc, for then Ψ ∈ M f \Psi\in\mathcal{M}_f Ψ ∈ M f and s ∈ Ψ ( u ) ⊆ M f ( u ) s\in\Psi(u)\subseteq M_f(u) s ∈ Ψ ( u ) ⊆ M f ( u ) . Let V V V be open and Ψ ( k 0 ) ∩ V ≠ ∅ \Psi(k_0)\cap V\ne\varnothing Ψ ( k 0 ) ∩ V = ∅ . If k 0 ∈ V k_0\in V k 0 ∈ V , then V V V is a neighbourhood of k 0 k_0 k 0 , and for every k ∈ V k\in V k ∈ V we have k ∈ Ψ ( k ) ∩ V k\in\Psi(k)\cap V k ∈ Ψ ( k ) ∩ V , so Ψ ( k ) ∩ V ≠ ∅ \Psi(k)\cap V\ne\varnothing Ψ ( k ) ∩ V = ∅ . Otherwise there is s ∈ V ∩ R s\in V\cap R s ∈ V ∩ R with f ( s ) = f ( k 0 ) f(s)=f(k_0) f ( s ) = f ( k 0 ) ; since f ∣ R f|_R f ∣ R is open, f ( V ∩ R ) ⊇ ( f ( s ) − δ , f ( s ) + δ ) ∩ f ( K ) f(V\cap R)\supseteq(f(s)-\delta,f(s)+\delta)\cap f(K) f ( V ∩ R ) ⊇ ( f ( s ) − δ , f ( s ) + δ ) ∩ f ( K ) for some δ > 0 \delta>0 δ > 0 , and for every k k k in the open neighbourhood { k : ∣ f ( k ) − f ( k 0 ) ∣ < δ } \{k:|f(k)-f(k_0)|<\delta\} { k : ∣ f ( k ) − f ( k 0 ) ∣ < δ } of k 0 k_0 k 0 there is y ∈ V ∩ R y\in V\cap R y ∈ V ∩ R with f ( y ) = f ( k ) f(y)=f(k) f ( y ) = f ( k ) , i.e. y ∈ Ψ ( k ) ∩ V y\in\Psi(k)\cap V y ∈ Ψ ( k ) ∩ V . □
Remark 3.4. With R = K R=K R = K , Lemma 3.3 gives Corollary 3.9 below in the form M f ( k ) = Φ f ( k ) M_f(k)=\Phi_f(k) M f ( k ) = Φ f ( k ) for all k k k . In the applications below f ∣ R f|_R f ∣ R is even open as a map into R \mathbb{R} R . The point of Lemma 3.3 is that R R R may be small (for instance nowhere dense), and that nothing is assumed about the points u u u .
Definition 3.5. A point v ∈ K v\in K v ∈ K is an openness point of f f f if f ( v ) f(v) f ( v ) is an interior point of f ( U ) f(U) f ( U ) for every neighbourhood U U U of v v v , i.e. f f f is open at v v v .
Lemma 3.6. Let W ⊂ K W\subset K W ⊂ K be open and locally connected, and assume that f f f is not constant on any nonempty open subset of W W W . If v ∈ W v\in W v ∈ W is an openness point of f f f , then v ∈ M f ( u ) v\in M_f(u) v ∈ M f ( u ) for every u ∈ Z f ( v ) u\in Z_{f(v)} u ∈ Z f ( v ) .
Proof. Let O O O be the set of openness points of f f f in W W W ; by Lemma 3.3 it suffices to show that f ∣ O f|_O f ∣ O is open. Let s ∈ O s \in O s ∈ O and let V V V be an open neighbourhood of s s s . Choose a connected open V ′ ⊆ V ∩ W V' \subseteq V \cap W V ′ ⊆ V ∩ W containing s s s , and δ > 0 \delta> 0 δ > 0 with ( f ( s ) − δ , f ( s ) + δ ) ⊆ f ( V ′ ) (f(s)-\delta, f(s)+\delta) \subseteq f(V') ( f ( s ) − δ , f ( s ) + δ ) ⊆ f ( V ′ ) . Let ∣ t − f ( s ) ∣ < δ |t-f(s)| < \delta ∣ t − f ( s ) ∣ < δ . The sets A = V ′ ∩ { f < t } A = V' \cap\{f < t\} A = V ′ ∩ { f < t } and B = V ′ ∩ { f > t } B = V' \cap\{f > t\} B = V ′ ∩ { f > t } are nonempty and open, and V ′ ∖ ( A ∪ B ) = V ′ ∩ Z t V' \setminus(A \cup B) = V' \cap Z_t V ′ ∖ ( A ∪ B ) = V ′ ∩ Z t has empty interior; hence V ′ V' V ′ is the union of the relative closures of A A A and B B B , which therefore meet at a point y ∈ V ′ y \in V' y ∈ V ′ . Then f ( y ) = t f(y) = t f ( y ) = t , and since K K K is locally connected at y y y , every neighbourhood of y y y contains a connected neighbourhood meeting A A A and B B B , so y y y is an openness point. Thus y ∈ O ∩ V y \in O \cap V y ∈ O ∩ V and f ( y ) = t f(y) = t f ( y ) = t , so ( f ( s ) − δ , f ( s ) + δ ) ⊆ f ( V ∩ O ) (f(s)-\delta, f(s)+\delta) \subseteq f(V \cap O) ( f ( s ) − δ , f ( s ) + δ ) ⊆ f ( V ∩ O ) .
Lemma 3.6 strengthens [23 Lemma 1.11] and Lemma 4 of [24 ] : no assumption is made on u u u , and local connectedness is used only near v v v . The same argument works along arcs, without any local connectedness.
Definition 3.7. Let γ ⊆ K \gamma\subseteq K γ ⊆ K be an arc such that f ∣ γ f|_\gamma f ∣ γ is not constant on any nondegenerate subarc. A point v v v of the relative interior γ ∘ \gamma^\circ γ ∘ is a crossing point of f f f on γ \gamma γ if every subarc of γ \gamma γ containing v v v in its relative interior contains points with f < f ( v ) f < f(v) f < f ( v ) and points with f > f ( v ) f > f(v) f > f ( v ) .
Lemma 3.8. If v v v is a crossing point of f f f on an arc γ \gamma γ , then v ∈ M f ( u ) v \in M_f(u) v ∈ M f ( u ) for every u ∈ Z f ( v ) u \in Z_{f(v)} u ∈ Z f ( v ) .
Proof. Let O γ O_\gamma O γ be the set of crossing points on γ \gamma γ ; by Lemma 3.3 it suffices to show that f ∣ O γ f|_{O_\gamma} f ∣ O γ is open. Let s ∈ O γ s \in O_\gamma s ∈ O γ and let V V V be an open neighbourhood of s s s . Choose a subarc J ⊆ V ∩ γ ∘ J \subseteq V \cap\gamma^\circ J ⊆ V ∩ γ ∘ with s s s in its relative interior; then f ( J ) f(J) f ( J ) is an interval containing f ( s ) f(s) f ( s ) in its interior, say ( f ( s ) − δ , f ( s ) + δ ) ⊆ f ( J ) (f(s)-\delta, f(s)+\delta) \subseteq f(J) ( f ( s ) − δ , f ( s ) + δ ) ⊆ f ( J ) . For ∣ t − f ( s ) ∣ < δ |t-f(s)| < \delta ∣ t − f ( s ) ∣ < δ the relatively open sets J ∩ { f < t } J \cap\{f < t\} J ∩ { f < t } and J ∩ { f > t } J \cap\{f > t\} J ∩ { f > t } are nonempty, and J ∩ Z t J \cap Z_t J ∩ Z t has empty interior in J J J ; since J J J is connected, their relative closures meet at some y ∈ J y \in J y ∈ J . Then f ( y ) = t f(y) = t f ( y ) = t , and every subarc of γ \gamma γ containing y y y in its relative interior contains points of J J J near y y y with f < t f < t f < t and with f > t f > t f > t ; so y ∈ O γ ∩ V y \in O_\gamma\cap V y ∈ O γ ∩ V .
The next result was stated in [23 Theorem 1.19(1)] , where its proof was omitted as being very similar to that of [23 Theorem 1.14] ; it is now an immediate consequence of Lemma 3.2 .
Corollary 3.9 ([23 Theorem 1.19(1)] ). If f : K → f ( K ) f : K \to f(K) f : K → f ( K ) is an open map, then f ∈ D ( K ) f \in\mathcal{D}(K) f ∈ D ( K ) .
Proof. If f f f is open, t ↦ f − 1 ( t ) t \mapsto f^{-1}(t) t ↦ f − 1 ( t ) is lsc on f ( K ) f(K) f ( K ) (equivalently, the decomposition of K K K into the fibres of f f f is continuous [29 Theorem 1.2.25] ), so Φ f \Phi_f Φ f is lsc on K K K ; by Lemma 3.2 with U = K U = K U = K , M f = Φ f M_f = \Phi_f M f = Φ f , so every f f f -admissible function is constant on the fibres of f f f .
Let D f D_f D f be the set of points x ∈ K x \in K x ∈ K such that x ∈ Int Z f ( x ) x \in\operatorname{Int} Z_{f(x)} x ∈ Int Z f ( x ) , or Φ f \Phi_f Φ f is lsc at every point of some neighbourhood of x x x , or x ∈ R x \in R x ∈ R for some R ⊆ K R \subseteq K R ⊆ K with f ∣ R f|_R f ∣ R open (this includes the points satisfying the hypotheses of Lemma 3.6 and the crossing points on arcs). By Lemmas 3.1, 3.2 and 3.3, the level f ( x ) f(x) f ( x ) is collapsed for every x ∈ D f x \in D_f x ∈ D f .
Lemma 3.10. If L ( K ) L(K) L ( K ) is dense in K K K , then D f D_f D f is dense in K K K for every f ∈ C ( K ) f \in C(K) f ∈ C ( K ) .
Proof. Let U U U be nonempty and open. There is a nonempty connected open locally connected V ⊆ U ‾ V \subseteq\overline{U} V ⊆ U . If f f f is constant on a nonempty open subset of V V V , that subset lies in D f D_f D f . Otherwise f ( V ) f(V) f ( V ) is a nondegenerate interval; for t t t in its interior the argument in the proof of Lemma 3.6 gives an openness point y ∈ V y \in V y ∈ V , and y y y satisfies the hypotheses of Lemma 3.6 with W = V W = V W = V .
Local retractions and retractions.
Lemma 3.11. Let O ⊆ K O \subseteq K O ⊆ K be open and r : O → K r : O \to K r : O → K continuous with f ∘ r = f f \circ r = f f ∘ r = f on O O O . Then r ( k ) ∈ M f ( k ) r(k) \in M_f(k) r ( k ) ∈ M f ( k ) for every k ∈ O k \in O k ∈ O . In particular, g ∘ r = g g \circ r = g g ∘ r = g on O O O .
Proof. The map equal to { k , r ( k ) } \{k,r(k)\} { k , r ( k )} on O O O and to { k } \{k\} { k } off O O O is lsc and belongs to M f \mathcal{M}_f M f .
Lemma 3.12. Let L L L be a closed subset of K K K , let r : K → L r: K \to L r : K → L be a retraction, and let f = f ∘ r f=f\circ r f = f ∘ r . Then r ( k ) ∈ M f ( k ) r(k)\in M_f(k) r ( k ) ∈ M f ( k ) for every k ∈ K k\in K k ∈ K and M f ∣ L ( l ) ⊆ M f ( l ) M_{f|L}(l)\subseteq M_f(l) M f ∣ L ( l ) ⊆ M f ( l ) for every l ∈ L l\in L l ∈ L . Consequently g = g ∘ r g=g\circ r g = g ∘ r , and g ∣ L g|L g ∣ L is f ∣ L f|L f ∣ L -admissible.
Proof. The first statement is Lemma 3.11 with O = K O=K O = K . Put Ψ ( k ) = { k } ∪ M f ∣ L ( r ( k ) ) \Psi(k)=\{k\}\cup M_{f|L}(r(k)) Ψ ( k ) = { k } ∪ M f ∣ L ( r ( k )) . Since f ( M f ∣ L ( r ( k ) ) ) = { f ( r ( k ) ) } = { f ( k ) } f(M_{f|L}(r(k)))=\{f(r(k))\}=\{f(k)\} f ( M f ∣ L ( r ( k ))) = { f ( r ( k ))} = { f ( k )} , we have Ψ ( k ) ⊆ Φ f ( k ) \Psi(k)\subseteq\Phi_f(k) Ψ ( k ) ⊆ Φ f ( k ) . For V V V open in K K K , { k : M f ∣ L ( r ( k ) ) ∩ V ≠ ∅ } = r − 1 ( { l ∈ L : M f ∣ L ( l ) ∩ ( V ∩ L ) ≠ ∅ } ) \{k:M_{f|L}(r(k))\cap V\ne\varnothing\}=r^{-1}(\{l\in L:M_{f|L}(l)\cap(V\cap L)\ne\varnothing\}) { k : M f ∣ L ( r ( k )) ∩ V = ∅ } = r − 1 ({ l ∈ L : M f ∣ L ( l ) ∩ ( V ∩ L ) = ∅ }) is open, because M f ∣ L M_{f|L} M f ∣ L is lsc on L L L and r r r is continuous; hence Ψ \Psi Ψ is lsc and Ψ ∈ M f \Psi\in\mathcal{M}_f Ψ ∈ M f . For l ∈ L l\in L l ∈ L we get M f ∣ L ( l ) ⊆ Ψ ( l ) ⊆ M f ( l ) M_{f|L}(l)\subseteq\Psi(l)\subseteq M_f(l) M f ∣ L ( l ) ⊆ Ψ ( l ) ⊆ M f ( l ) . □
Corollary 3.13. Let r : K → L r: K \to L r : K → L be a retraction and f = f ∘ r f=f\circ r f = f ∘ r . If f ∣ L ∈ D ( L ) f|L\in\mathcal{D}(L) f ∣ L ∈ D ( L ) , in particular if L ∈ D C P L\in\mathcal{DCP} L ∈ D C P , then f ∈ D ( K ) f\in\mathcal{D}(K) f ∈ D ( K ) . This applies whenever L L L is a locally connected continuum, e.g. an arc or a dendrite (a locally connected continuum containing no simple closed curve).
Proof. By Lemma 3.12 and the assumption, g ∣ L = h ∘ f ∣ L g|L=h\circ f|L g ∣ L = h ∘ f ∣ L for some h ∈ C ( f ( L ) ) h\in C(f(L)) h ∈ C ( f ( L )) . Since f = f ∘ r f=f\circ r f = f ∘ r , we have f ( L ) = f ( K ) f(L)=f(K) f ( L ) = f ( K ) and g = g ∘ r = h ∘ f ∘ r = h ∘ f g=g\circ r=h\circ f\circ r=h\circ f g = g ∘ r = h ∘ f ∘ r = h ∘ f . Locally connected continua are in D C P \mathcal{DCP} D C P by [23 Theorem 1.14] ; see also Corollary 4.8 . □
The converse of Lemma 3.12 fails: a function g g g with g = g ∘ r g=g\circ r g = g ∘ r and g ∣ L g|L g ∣ L f ∣ L f|L f ∣ L -admissible need not be f f f -admissible (Example 5.11 ). In particular, the lemma does not show that D C P \mathcal{DCP} D C P passes from K K K to its retracts.
Cantor boxes
A Cantor box in K K K is an open set B B B with a homeomorphism B ≅ A × J B\cong A\times J B ≅ A × J , A A A a nonempty clopen subset of the Cantor set C C C and J J J an open interval; f x ( τ ) = f ( x , τ ) f_x(\tau)=f(x,\tau) f x ( τ ) = f ( x , τ ) . A point ( x , τ ) ∈ B (x,\tau)\in B ( x , τ ) ∈ B is a crossing point if every subinterval of J J J containing τ \tau τ in its interior contains points where f x < f ( x , τ ) f_x<f(x,\tau) f x < f ( x , τ ) and points where f x > f ( x , τ ) f_x>f(x,\tau) f x > f ( x , τ ) . (Unlike the definition for arcs, f x f_x f x may be constant on nondegenerate subintervals of J J J ; only the behaviour of f x f_x f x near τ \tau τ matters.) This property depends only on f f f along the arc of B B B through ( x , τ ) (x,\tau) ( x , τ ) near this point, so it does not depend on the box; in particular a crossing point of B B B is a crossing point of every smaller box A ′ × J ′ ⊆ B A'\times J'\subseteq B A ′ × J ′ ⊆ B containing it. The two notions of crossing point agree when both apply: if γ \gamma γ is an arc such that f ∣ γ f|_\gamma f ∣ γ is not constant on any nondegenerate subarc and v ∈ γ ∘ ∩ B v\in\gamma^\circ\cap B v ∈ γ ∘ ∩ B , then near v v v the arc γ \gamma γ lies in the arc of B B B through v v v , so v v v is a crossing point of f f f on γ \gamma γ if and only if it is a crossing point of B B B .
Lemma 3.14. Let B = A × J B=A\times J B = A × J be a Cantor box and t ∈ R t\in\mathbb{R} t ∈ R such that for every x ∈ A x\in A x ∈ A , f x f_x f x takes values < t <t < t and > t >t > t on J J J . Then either some ( x , τ ) ∈ B (x,\tau)\in B ( x , τ ) ∈ B with f ( x , τ ) = t f(x,\tau)=t f ( x , τ ) = t is a crossing point, or Z t ∩ B Z_t\cap B Z t ∩ B has nonempty interior.
Proof. If x x x has no crossing point at level t t t , pick a , b ∈ J a,b\in J a , b ∈ J with f x ( a ) < t < f x ( b ) f_x(a)<t<f_x(b) f x ( a ) < t < f x ( b ) , say a < b a<b a < b ; let σ = inf { τ ∈ [ a , b ] : f x ( τ ) > t } \sigma=\inf\{\tau\in[a,b]:f_x(\tau)>t\} σ = inf { τ ∈ [ a , b ] : f x ( τ ) > t } and ρ = sup { τ ∈ [ a , σ ] : f x ( τ ) < t } \rho=\sup\{\tau\in[a,\sigma]:f_x(\tau)<t\} ρ = sup { τ ∈ [ a , σ ] : f x ( τ ) < t } . If $\rho=\sigma this point is a crossing point, so ρ < σ \rho<\sigma ρ < σ and f x ≡ t f_x\equiv t f x ≡ t on [ ρ , σ ] [\rho,\sigma] [ ρ , σ ] . Hence A = ⋃ p , q E p , q A=\bigcup_{p,q}E_{p,q} A = ⋃ p , q E p , q over rationals p < q p<q p < q in J J J , where E p , q = { x ∈ A : f x ≡ t on [ p , q ] } E_{p,q}=\{x\in A:f_x\equiv t\text{ on }[p,q]\} E p , q = { x ∈ A : f x ≡ t on [ p , q ]} is closed. By Baire's theorem some E p , q E_{p,q} E p , q contains a clopen A ′ ≠ ∅ A'\ne\varnothing A ′ = ∅ , and A ′ × ( p , q ) A'\times(p,q) A ′ × ( p , q ) is open. □
Proposition 3.15. Let W ⊆ K W\subseteq K W ⊆ K be open, covered by Cantor boxes, and assume f f f is not constant on any nonempty open subset of W W W . Let O W O_W O W be the set of crossing points in W W W . Then f ∣ O W f|_{O_W} f ∣ O W is open; hence every level in f ( O W ) f(O_W) f ( O W ) is collapsed (Lemma 3.3).
Proof. Let s = ( x 0 , τ 0 ) ∈ O W s=(x_0,\tau_0)\in O_W s = ( x 0 , τ 0 ) ∈ O W and V V V a neighbourhood of s s s . Choose a box A ′ × J ′ ⊆ V ∩ W A'\times J'\subseteq V\cap W A ′ × J ′ ⊆ V ∩ W about s s s and values t 0 = f ( α ) < f ( s ) < f ( β ) t_0=f(\alpha)<f(s)<f(\beta) t 0 = f ( α ) < f ( s ) < f ( β ) with α , β ∈ J ′ \alpha,\beta\in J' α , β ∈ J ′ ; shrinking A ′ A' A ′ , choose δ > 0 \delta>0 δ > 0 every x ∈ A ′ x\in A' x ∈ A ′ and ∣ t − f ( s ) ∣ < δ |t-f(s)|<\delta ∣ t − f ( s ) ∣ < δ satisfy f x ( α ) < t < f x ( β ) f_x(\alpha)<t<f_x(\beta) f x ( α ) < t < f x ( β ) . By Lemma 3.14 and the hypothesis on W W W , there is a crossing point y ∈ A ′ × J ′ y\in A'\times J' y ∈ A ′ × J ′ with f ( y ) = t f(y)=t f ( y ) = t . □
Remark 3.16. Zero-dimensionality of A A A is used in Lemma 3.14 and Proposition 3.15 only through the existence of small clopen sets. Call a box an open set B ⊆ K B\subseteq K B ⊆ K with a homeomorphism B ≅ U × J B \cong U \times J B ≅ U × J , U U U a nonempty open subset of a compact metrizable space and J J J an open interval, and define its crossing points as above, along the arcs { y } × J \{y\}\times J { y } × J . Then Lemma 3.14 and Proposition 3.15 hold for boxes, with “clopen” replaced by “open”: U U U is a Baire space, so some closed set E p , q E_{p,q} E p , q contains a nonempty open U ′ ′ ⊆ U U^{\prime\prime} \subseteq U U ′′ ⊆ U , and U ′ ′ × ( p , q ) U^{\prime\prime}\times(p,q) U ′′ × ( p , q ) is open in K K K . Since for U U U not zero-dimensional the arcs { y } × J \{y\}\times J { y } × J depend on the product structure, in Proposition 3.15 one takes O W O_W O W to be the set of points that are crossing points of some box contained in W W W (each with its given product structure).
Continua in DCP: sufficient conditions and examples
In this section we pass from the local information of Section 3 to global conclusions about K K K . Theorem 4.2 gives sufficient conditions for a single level c c c to be collapsed: (a) the values of g g g on Z c Z_c Z c can be approached through collapsed levels above, resp. below, c c c ; (b) a topological condition on how Z c Z_c Z c is reached from { f > c } \{f>c\} { f > c } and { f < c } \{f<c\} { f < c } , which holds automatically when K K K is arc connected. Combined with Lemma 3.10, this shows that every arc connected continuum K K K with L ( K ) L(K) L ( K ) dense is in D C P DCP D C P (Theorem 4.4); we then replace arc connectedness by a condition on arc components (Theorem 4.7). After examples, which include the earlier results of [23 , 24 ] (Corollary 4.8), we show that D C P DCP D C P passes to open images (Theorem 4.15) and that the Cantor comb, all cones over compact metrizable spaces (in particular the Cantor fan) and the Cantor grid are in D C P DCP D C P , although L ( K ) L(K) L ( K ) need not be dense for them; in the fan and the grid no point has an open locally connected neighbourhood (Theorems 4.22 and 4.31); Theorem 4.25 puts most of these results into one framework. The section ends with counterexamples.
We first fix some notation. Recall that Σ \Sigma Σ is the set of points of K K K whose level is collapsed. For c ∈ f ( K ) c\in f(K) c ∈ f ( K ) put Z c + = Z c ∩ cl { f > c } Z_c^+=Z_c\cap\operatorname{cl}\{f>c\} Z c + = Z c ∩ cl { f > c } and Z c − = Z c ∩ cl { f < c } Z_c^-=Z_c\cap\operatorname{cl}\{f<c\} Z c − = Z c ∩ cl { f < c } , and Z c ↑ = { a ∈ Z c : ∀ Z_c^{\uparrow}=\{a\in Z_c:\ \forall Z c ↑ = { a ∈ Z c : ∀ neighbourhood U U U of a ∃ ε > 0 : [ c , c + ε ) ⊆ f ( U ) } a\ \exists\varepsilon>0:\ [c,c+\varepsilon)\subseteq f(U)\} a ∃ ε > 0 : [ c , c + ε ) ⊆ f ( U )} , and Z c ↓ Z_c^{\downarrow} Z c ↓ similarly with ( c − ε , c ] (c-\varepsilon,c] ( c − ε , c ] . Clearly Z c ↑ ⊆ Z c + Z_c^{\uparrow}\subseteq Z_c^+ Z c ↑ ⊆ Z c + and Z c ↓ ⊆ Z c − Z_c^{\downarrow}\subseteq Z_c^- Z c ↓ ⊆ Z c − ; if Int Z c = ∅ \operatorname{Int}Z_c=\varnothing Int Z c = ∅ then Z c = Z c + ∪ Z c − Z_c=Z_c^+\cup Z_c^- Z c = Z c + ∪ Z c − .
Lemma 4.1. Let a ∈ Z c ↑ a\in Z_c^{\uparrow} a ∈ Z c ↑ and u ∈ Z c u\in Z_c u ∈ Z c , and suppose that there are y n ∈ cl Σ y_n\in\operatorname{cl}\Sigma y n ∈ cl Σ with f ( y n ) > c f(y_n)>c f ( y n ) > c , f ( y n ) → c f(y_n)\to c f ( y n ) → c and g ( y n ) → g ( u ) g(y_n)\to g(u) g ( y n ) → g ( u ) . Then g ( u ) = g ( a ) g(u)=g(a) g ( u ) = g ( a ) . The same holds with Z c ↓ Z_c^{\downarrow} Z c ↓ and f ( y n ) < c f(y_n)<c f ( y n ) < c .
Proof. Choose q n ∈ Σ q_n\in\Sigma q n ∈ Σ so close to y n y_n y n that f ( q n ) > c f(q_n)>c f ( q n ) > c , ∣ f ( q n ) − f ( y n ) ∣ < 1 / n \lvert f(q_n)-f(y_n)\rvert<1/n ∣ f ( q n ) − f ( y n )∣ < 1/ n and ∣ g ( q n ) − g ( y n ) ∣ < 1 / n \lvert g(q_n)-g(y_n)\rvert<1/n ∣ g ( q n ) − g ( y n )∣ < 1/ n ; then f ( q n ) → c f(q_n)\to c f ( q n ) → c and g ( q n ) → g ( u ) g(q_n)\to g(u) g ( q n ) → g ( u ) . Let U m U_m U m be the ball of radius 1 / m 1/m 1/ m about a a a and ε m \varepsilon_m ε m as in the definition of Z c ↑ Z_c^{\uparrow} Z c ↑ . Choose n 1 < n 2 < ⋯ n_1<n_2<\cdots n 1 < n 2 < ⋯ with f ( q n m ) < c + ε m f(q_{n_m})<c+\varepsilon_m f ( q n m ) < c + ε m and a m ∈ U m a_m\in U_m a m ∈ U m with f ( a m ) = f ( q n m ) f(a_m)=f(q_{n_m}) f ( a m ) = f ( q n m ) . Since this level is collapsed, g ( a m ) = g ( q n m ) g(a_m)=g(q_{n_m}) g ( a m ) = g ( q n m ) ; letting m → ∞ m\to\infty m → ∞ gives g ( a ) = g ( u ) g(a)=g(u) g ( a ) = g ( u ) . □
Theorem 4.2. Let K K K be a connected compact metrizable space, f ∈ C ( K ) f\in C(K) f ∈ C ( K ) , g g g f f f -admissible, and let c ∈ f ( K ) c\in f(K) c ∈ f ( K ) with Int Z c = ∅ \operatorname{Int}Z_c=\varnothing Int Z c = ∅ . Assume:
(a) for every u ∈ Z c + u\in Z_c^+ u ∈ Z c + (resp. u ∈ Z c − u\in Z_c^- u ∈ Z c − ) there are y n ∈ cl Σ y_n\in\operatorname{cl}\Sigma y n ∈ cl Σ with f ( y n ) > c f(y_n)>c f ( y n ) > c (resp. f ( y n ) < c f(y_n)<c f ( y n ) < c ), f ( y n ) → c f(y_n)\to c f ( y n ) → c and g ( y n ) → g ( u ) g(y_n)\to g(u) g ( y n ) → g ( u ) ;
(b) Z c ⊆ Y + ∪ Y − Z_c\subseteq Y^+\cup Y^- Z c ⊆ Y + ∪ Y − , where Y + = Z c + Y^+=Z_c^+ Y + = Z c + if Z c ↑ ≠ ∅ Z_c^{\uparrow}\ne\varnothing Z c ↑ = ∅ and Y + = ∅ Y^+=\varnothing Y + = ∅ otherwise, and Y − Y^- Y − is defined similarly.
Then g g g is constant on Z c Z_c Z c .
Proof. If Y + ≠ ∅ Y^+\ne\varnothing Y + = ∅ , fix a ∈ Z c ↑ a\in Z_c^{\uparrow} a ∈ Z c ↑ ; by (a) and Lemma 4.1, g ≡ g ( a ) g\equiv g(a) g ≡ g ( a ) on Z c + Z_c^+ Z c + . Similarly g ≡ g ( b ) g\equiv g(b) g ≡ g ( b ) on Z c − Z_c^- Z c − for b ∈ Z c ↓ b\in Z_c^{\downarrow} b ∈ Z c ↓ if Y − ≠ ∅ Y^-\ne\varnothing Y − = ∅ . If only one of Y ± Y^\pm Y ± is nonempty we are done by (b). If both are nonempty, then { f > c } \{f>c\} { f > c } and { f < c } \{f<c\} { f < c } are nonempty; since K ∖ Z c K\setminus Z_c K ∖ Z c is dense, K = cl { f > c } ∪ cl { f < c } K=\operatorname{cl}\{f>c\}\cup\operatorname{cl}\{f<c\} K = cl { f > c } ∪ cl { f < c } , and by connectedness these closed sets meet, necessarily in Z c + ∩ Z c − Z_c^+\cap Z_c^- Z c + ∩ Z c − . Hence g ( a ) = g ( b ) g(a)=g(b) g ( a ) = g ( b ) . □
Condition (a) holds in particular if every point of Z c + Z_c^+ Z c + is the limit of points u n ∈ Σ u_n\in\Sigma u n ∈ Σ with f ( u n ) > c f(u_n)>c f ( u n ) > c , and similarly for Z c − Z_c^- Z c − (take y n = u n y_n=u_n y n = u n ). This is how (a) is verified in all our results except Theorems 4.22, 4.25 and 4.31 on the Cantor comb and fan, rooted teeth and the Cantor grid, where the more general form is needed (Lemmas 4.21 and 4.30).
If K K K is arc connected, condition (b) holds for every f f f and c c c : if Z c + ≠ ∅ Z_c^+ \ne\varnothing Z c + = ∅ , take an arc from a point with f > c f>c f > c to a point of Z c + Z_c^+ Z c + and let a a a be its first point in Z c Z_c Z c ; then f > c f>c f > c on the part of the arc before a a a , so a ∈ Z c ↑ a \in Z_c^{\uparrow} a ∈ Z c ↑ . Hence Y c ± = Z c ± Y_c^{\pm}=Z_c^{\pm} Y c ± = Z c ± and Z c = Y c + ∪ Y c − Z_c=Y_c^+\cup Y_c^- Z c = Y c + ∪ Y c − .
Remark 4.3. Example 4.38 below explains the role of the sets Z c ↑ , Z c ↓ Z_c^{\uparrow}, Z_c^{\downarrow} Z c ↑ , Z c ↓ in (b): there, at the level c = 0 c=0 c = 0 the fibre E E E meets cl { f > 0 } \operatorname{cl}\{f>0\} cl { f > 0 } , but no point of E E E is approached from above along a connected set, so Z 0 ↑ = ∅ Z_0^{\uparrow}=\varnothing Z 0 ↑ = ∅ , condition (b) fails, and indeed f ∉ D ( K ) f\notin D(K) f ∈ / D ( K ) .
Theorem 4.4. Let K K K be an arc connected metrizable continuum such that the set L ( K ) L(K) L ( K ) of points having an open locally connected neighbourhood is dense in K K K . Then K ∈ D C P K\in DCP K ∈ D C P .
Proof. Let f ∈ C ( K ) f\in C(K) f ∈ C ( K ) , let g g g be f f f -admissible and c ∈ f ( K ) c\in f(K) c ∈ f ( K ) . If Int Z c ≠ ∅ \operatorname{Int}Z_c\ne\varnothing Int Z c = ∅ , use Lemma 3.1. Otherwise (b) holds by the remark above, and (a) holds because D f ⊆ Σ D_f\subseteq\Sigma D f ⊆ Σ is dense (Lemma 3.10) and { f > c } \{f>c\} { f > c } , { f < c } \{f<c\} { f < c } are open. By Theorem 4.2 every level is collapsed. □
Corollary 4.5. Let K K K be a metrizable continuum and K 1 ⊆ K 2 ⊆ ⋯ K_1\subseteq K_2\subseteq\cdots K 1 ⊆ K 2 ⊆ ⋯ locally connected subcontinua of K K K such that K m ⊆ cl Int K m K_m\subseteq\operatorname{cl}\operatorname{Int}K_m K m ⊆ cl Int K m for every m m m , ⋃ m K m \bigcup_m K_m ⋃ m K m is dense in K K K , and every point of K K K is joined by a path in K K K to a point of ⋃ m K m \bigcup_m K_m ⋃ m K m . Then K ∈ D C P K\in DCP K ∈ D C P .
Proof. The set Int K m \operatorname{Int}K_m Int K m is open in K K K and is an open subset of the locally connected space K m K_m K m , hence locally connected; so Int K m ⊆ L ( K ) \operatorname{Int}K_m\subseteq L(K) Int K m ⊆ L ( K ) . Since Int K m \operatorname{Int}K_m Int K m is dense in K m K_m K m and ⋃ m K m \bigcup_m K_m ⋃ m K m is dense in K K K , L ( K ) L(K) L ( K ) is dense in K K K . Each K m K_m K m is arc connected [41 Theorem 5.1, p. 36] , hence so is the increasing union ⋃ m K m \bigcup_m K_m ⋃ m K m ; since every point of K K K is joined to it by a path, hence by an arc, K K K is arc connected. Apply Theorem 4.4. □
Arc components.
Lemma 4.6. Let K K K be a continuum such that every arc component P P P of K K K is either dense in K K K or satisfies P ⊆ cl Int P P\subseteq\operatorname{cl}\operatorname{Int}P P ⊆ cl Int P . Then condition (b) of Theorem 4.2 holds for every f ∈ C ( K ) f\in C(K) f ∈ C ( K ) and every c ∈ f ( K ) c\in f(K) c ∈ f ( K ) with Int Z c = ∅ \operatorname{Int}Z_c=\varnothing Int Z c = ∅ .
Proof. We first show: if Z c + ≠ ∅ Z_c^+\ne\varnothing Z c + = ∅ and Z c ↓ = ∅ Z_c^{\downarrow}=\varnothing Z c ↓ = ∅ , then Z c ↑ ≠ ∅ Z_c^{\uparrow}\ne\varnothing Z c ↑ = ∅ and Z c ↑ ⊆ Z c − Z_c^{\uparrow}\subseteq Z_c^- Z c ↑ ⊆ Z c − .
Let W = { f > c } W=\{f>c\} W = { f > c } . If an arc starting at a point of W W W leaves W W W , its first point outside W W W lies in Z c Z_c Z c and is approached from above along the arc, hence lies in Z c ↑ Z_c^{\uparrow} Z c ↑ . Therefore W W W is a union of arc components of K K K . Let u ∈ Z c ↑ u\in Z_c^{\uparrow} u ∈ Z c ↑ ; then u ∈ cl W ∖ W u\in\operatorname{cl}W\setminus W u ∈ cl W ∖ W , so the arc component P u P_u P u of u u u is disjoint from W W W . If P u P_u P u were dense, the closed set K ∖ W ⊇ P u K\setminus W\supseteq P_u K ∖ W ⊇ P u would be K K K , contradicting W ≠ ∅ W\ne\varnothing W = ∅ . Hence u ∈ cl Int P u u\in\operatorname{cl}\operatorname{Int}P_u u ∈ cl Int P u . Since f ≤ c f\leq c f ≤ c on the open set Int P u \operatorname{Int}P_u Int P u and Int Z c = ∅ \operatorname{Int}Z_c=\varnothing Int Z c = ∅ , there are points v ∈ Int P u v\in\operatorname{Int}P_u v ∈ Int P u arbitrarily close to u u u with f ( v ) < c f(v)<c f ( v ) < c ; thus u ∈ Z c − u\in Z_c^- u ∈ Z c − . Moreover, on an arc in P u P_u P u from such a v v v to u u u the first point of Z c Z_c Z c lies in Z c ↓ Z_c^{\downarrow} Z c ↓ .
By symmetry, if Z c − ≠ ∅ Z_c^-\ne\varnothing Z c − = ∅ and Z c ↓ = ∅ Z_c^{\downarrow}=\varnothing Z c ↓ = ∅ , then Z c ↑ ≠ ∅ Z_c^{\uparrow}\ne\varnothing Z c ↑ = ∅ and Z c − ⊆ Z c + Z_c^-\subseteq Z_c^+ Z c − ⊆ Z c + . Now recall Z c = Z c + ∪ Z c − Z_c=Z_c^+\cup Z_c^- Z c = Z c + ∪ Z c − . If Z c ↑ Z_c^{\uparrow} Z c ↑ and Z c ↓ Z_c^{\downarrow} Z c ↓ are both nonempty, then Y ± = Z c ± Y^\pm=Z_c^\pm Y ± = Z c ± and (b) holds. If Z c ↓ = ∅ Z_c^{\downarrow}=\varnothing Z c ↓ = ∅ , then either Z c + = ∅ Z_c^+=\varnothing Z c + = ∅ , and Z c = Z c − = Y − Z_c=Z_c^-=Y^- Z c = Z c − = Y − because Z c ↑ ≠ ∅ Z_c^{\uparrow}\ne\varnothing Z c ↑ = ∅ by the symmetric statement; or Z c + ≠ ∅ Z_c^+\ne\varnothing Z c + = ∅ , and Z c ⊆ Z c − = Y − Z_c\subseteq Z_c^-=Y^- Z c ⊆ Z c − = Y − by the statement proved above. The case Z c ↑ = ∅ Z_c^{\uparrow}=\varnothing Z c ↑ = ∅ is symmetric. □
Theorem 4.7. Let K K K be a metrizable continuum such that L ( K ) L(K) L ( K ) is dense in K K K and every arc component P P P of K K K is either dense in K K K or satisfies P ⊆ cl Int P P\subseteq\operatorname{cl}\operatorname{Int}P P ⊆ cl Int P . Then K ∈ D C P K\in DCP K ∈ D C P .
Proof. Let f ∈ C ( K ) f\in C(K) f ∈ C ( K ) , let g g g be f f f -admissible and c ∈ f ( K ) c\in f(K) c ∈ f ( K ) . If Int Z c ≠ ∅ \operatorname{Int}Z_c\ne\varnothing Int Z c = ∅ , use Lemma 3.1. Otherwise (a) holds as in the proof of Theorem 4.4 (Lemma 3.10), and (b) holds by Lemma 4.6. Apply Theorem 4.2. □
Examples
Corollary 4.8. The following continua are in D C P DCP D C P : locally connected metrizable continua [23 Theorem 1.14] ; the broom and the bookcase [24 ] ; the continua of [24 Proposition 7] .
Proof. Locally connected continua are arc connected [41 Theorem 5.1, p. 36] . For the broom L ( K ) L(K) L ( K ) contains the segments { y = x / n , 0 < x } \{y=x/n, 0<x\} { y = x / n , 0 < x } without the origin, for the bookcase it contains the open horizontal segments at the heights 1 / n 1/n 1/ n (but not the bottom segment y = 0 y=0 y = 0 , which is a limit of them) and the vertical sides minus the bottom corners; both are dense and both continua are arc connected. The continua of [24 Proposition 7] satisfy the hypotheses of Corollary 4.5. □
Remark 4.9. Corollary 4.5 is [24 Proposition 7] with fewer hypotheses: condition (4) there (every point of K m K_m K m is a point of local connectedness of K K K ) is not needed, and of condition (1) (K m ⊊ K m + 1 K_m \subsetneq K_{m+1} K m ⊊ K m + 1 ) only the inclusion K m ⊆ K m + 1 K_m \subseteq K_{m+1} K m ⊆ K m + 1 is used. The path condition (6) cannot be omitted: for the closure S S S of { ( x , sin ( 1 / x ) ) : 0 < x ≤ 1 } \{(x,\sin(1/x)):0<x\leq1\} {( x , sin ( 1/ x )) : 0 < x ≤ 1 } and K m K_m K m the part of the curve over [ 1 / m , 1 ] [1/m,1] [ 1/ m , 1 ] , all other conditions hold, but S ∉ D C P S\notin DCP S ∈ / D C P (Example 4.35 ). On the other hand, conditions (1), (2) and (6) alone imply that K K K is arc connected. Hence an example showing that the density conditions (3) or (5) cannot be omitted would be an arc connected continuum outside D C P DCP D C P , and would answer Problem 6.1 in the negative.
Theorem 4.7 contains Theorem 4.4 (one arc component, P = K P=K P = K ) and the following result of [24 ] .
Proposition 4.10 ([24 Proposition 9] ). Let Q = [ − 1 , 0 ] × [ − 1 , 1 ] Q=[-1,0]\times[-1,1] Q = [ − 1 , 0 ] × [ − 1 , 1 ] , C 2 = { ( x , sin ( 1 / x ) ) : 0 < x ≤ 1 } C_2=\{(x,\sin(1/x)):0<x\leq1\} C 2 = {( x , sin ( 1/ x )) : 0 < x ≤ 1 } and K = Q ∪ C 2 K=Q\cup C_2 K = Q ∪ C 2 . Then K ∈ D C P K\in DCP K ∈ D C P , although K K K is not arc connected.
Proof. The arc components of K K K are Q Q Q and C 2 C_2 C 2 ; we have Q = cl Int Q Q=\operatorname{cl}\operatorname{Int}Q Q = cl Int Q , and C 2 C_2 C 2 is open in K K K . Moreover L ( K ) ⊇ ( Q ∖ E ) ∪ C 2 L(K)\supseteq(Q\setminus E)\cup C_2 L ( K ) ⊇ ( Q ∖ E ) ∪ C 2 , where E = { 0 } × [ − 1 , 1 ] E=\{0\}\times[-1,1] E = { 0 } × [ − 1 , 1 ] , which is dense. Apply Theorem 4.7 . □
Example 4.11. Let S S S be the closure of { ( x , sin ( 1 / x ) ) : 0 < x ≤ 1 } \{(x,\sin(1/x)):0<x\leq1\} {( x , sin ( 1/ x )) : 0 < x ≤ 1 } . The cone K = Cone ( S ) K=\operatorname{Cone}(S) K = Cone ( S ) is arc connected (every point is joined to the vertex by a segment), and the image of { ( x , sin ( 1 / x ) ) : 0 < x ≤ 1 } × ( 0 , 1 ) \{(x,\sin(1/x)):0<x\leq1\}\times(0,1) {( x , sin ( 1/ x )) : 0 < x ≤ 1 } × ( 0 , 1 ) is an open locally connected dense subset of K K K . By Theorem 4.4 , K ∈ D C P K\in DCP K ∈ D C P , although S ∉ D C P S\notin DCP S ∈ / D C P . (By Theorem 4.22 below, the cone over every compact metrizable space is in D C P DCP D C P .)
The next two continua from [21 ] are not arc connected.
Example 4.12. Let R = { ( 1 + 1 1 + t ) ( cos t , sin t ) : t ≥ 0 } R=\{(1+\frac{1}{1+t})(\cos t,\sin t):t\geq0\} R = {( 1 + 1 + t 1 ) ( cos t , sin t ) : t ≥ 0 } be a ray spiralling onto the unit circle S 1 S^1 S 1 , let D D D be the closed unit disk and K = D ∪ R K=D\cup R K = D ∪ R (the “spiral with a disk” of [21 Exercise 13.14] ). The arc components of K K K are R R R , which is open in K K K , and D = cl Int D D=\operatorname{cl}\operatorname{Int}D D = cl Int D , and L ( K ) ⊇ R ∪ Int D L(K)\supseteq R\cup\operatorname{Int}D L ( K ) ⊇ R ∪ Int D is dense. By Theorem 4.7 , K ∈ D C P K\in DCP K ∈ D C P . Without the disk, S 1 ∪ R ∉ D C P S^1\cup R\notin DCP S 1 ∪ R ∈ / D C P by Proposition 4.36 below; so adding a disk can produce a continuum in D C P DCP D C P .
Example 4.13. Bing’s double tornado [21 Exercise 13.28] is the union of two disjoint subsets C 1 , C 2 C_1,C_2 C 1 , C 2 of R 3 \mathbb{R}^3 R 3 , each homeomorphic to S 1 × [ 0 , ∞ ) S^1\times[0,\infty) S 1 × [ 0 , ∞ ) , such that each C i C_i C i spirals onto the base circle (the image of S 1 × { 0 } S^1\times\{0\} S 1 × { 0 } ) of the other: cl C i \operatorname{cl}C_i cl C i is the union of C i C_i C i and the base circle of C 3 − i C_{3-i} C 3 − i . As in the sin( 1 / x ) (1/x) ( 1/ x ) continuum, no arc joins C 1 C_1 C 1 and C 2 C_2 C 2 , so these are the arc components. The interior of C i C_i C i is C i C_i C i without its base circle, so C i ⊆ cl Int C i C_i\subseteq\operatorname{cl}\operatorname{Int}C_i C i ⊆ cl Int C i , and L ( K ) ⊇ Int C 1 ∪ Int C 2 L(K)\supseteq\operatorname{Int}C_1\cup\operatorname{Int}C_2 L ( K ) ⊇ Int C 1 ∪ Int C 2 is dense. By Theorem 4.7 , K ∈ D C P K\in DCP K ∈ D C P .
Open images
Lemma 4.14. Let p : K → D p:K\to D p : K → D be an open continuous surjection of compact metrizable spaces, h ∈ C ( D ) h\in C(D) h ∈ C ( D ) and f = h ∘ p f=h\circ p f = h ∘ p . Then M f ( k ) = p − 1 ( M h ( p ( k ) ) ) M_f(k)=p^{-1}(M_h(p(k))) M f ( k ) = p − 1 ( M h ( p ( k ))) for every k ∈ K k\in K k ∈ K .
Proof. Put Ψ ( k ) = p − 1 ( M h ( p ( k ) ) ) \Psi(k)=p^{-1}(M_h(p(k))) Ψ ( k ) = p − 1 ( M h ( p ( k ))) . Then k ∈ Ψ ( k ) ⊆ Φ f ( k ) k\in\Psi(k)\subseteq\Phi_f(k) k ∈ Ψ ( k ) ⊆ Φ f ( k ) , and for V V V open in K K K , { k : Ψ ( k ) ∩ V ≠ ∅ } = p − 1 ( { y : M h ( y ) ∩ p ( V ) ≠ ∅ } ) \{k:\Psi(k)\cap V\ne\varnothing\}=p^{-1}(\{y:M_h(y)\cap p(V)\ne\varnothing\}) { k : Ψ ( k ) ∩ V = ∅ } = p − 1 ({ y : M h ( y ) ∩ p ( V ) = ∅ }) is open because p ( V ) p(V) p ( V ) is open and M h M_h M h is lsc. Hence Ψ ∈ M f \Psi\in\mathcal{M}_f Ψ ∈ M f and Ψ ( k ) ⊆ M f ( k ) \Psi(k)\subseteq M_f(k) Ψ ( k ) ⊆ M f ( k ) .
Conversely, put Ψ ′ ( y ) = cl ⋃ x ∈ p − 1 ( y ) p ( M f ( x ) ) \Psi'(y)=\operatorname{cl}\bigcup_{x\in p^{-1}(y)}p(M_f(x)) Ψ ′ ( y ) = cl ⋃ x ∈ p − 1 ( y ) p ( M f ( x )) . Then y ∈ Ψ ′ ( y ) ⊆ Φ h ( y ) y\in\Psi'(y)\subseteq\Phi_h(y) y ∈ Ψ ′ ( y ) ⊆ Φ h ( y ) . If W ⊆ D W\subseteq D W ⊆ D is open and Ψ ′ ( y 0 ) ∩ W ≠ ∅ \Psi'(y_0)\cap W\ne\varnothing Ψ ′ ( y 0 ) ∩ W = ∅ , there are x 0 ∈ p − 1 ( y 0 ) x_0\in p^{-1}(y_0) x 0 ∈ p − 1 ( y 0 ) and w ∈ M f ( x 0 ) w\in M_f(x_0) w ∈ M f ( x 0 ) with p ( w ) ∈ W p(w)\in W p ( w ) ∈ W . Since M f M_f M f is lsc, M f ( x ) ∩ p − 1 ( W ) ≠ ∅ M_f(x)\cap p^{-1}(W)\ne\varnothing M f ( x ) ∩ p − 1 ( W ) = ∅ for all x x x in a neighbourhood N N N of x 0 x_0 x 0 , and since p p p is open, p ( N ) p(N) p ( N ) is a neighbourhood of y 0 y_0 y 0 ; thus Ψ ′ ( y ) ∩ W ≠ ∅ \Psi'(y)\cap W\ne\varnothing Ψ ′ ( y ) ∩ W = ∅ for y y y near y 0 y_0 y 0 . Hence Ψ ′ ∈ M h \Psi'\in\mathcal{M}_h Ψ ′ ∈ M h , so p ( M f ( k ) ) ⊆ M h ( p ( k ) ) p(M_f(k))\subseteq M_h(p(k)) p ( M f ( k )) ⊆ M h ( p ( k )) and M f ( k ) ⊆ p − 1 ( M h ( p ( k ) ) ) M_f(k)\subseteq p^{-1}(M_h(p(k)) ) M f ( k ) ⊆ p − 1 ( M h ( p ( k ))) . □
Theorem 4.15. Let p : K → D p:K\to D p : K → D be an open continuous surjection of compact metrizable spaces, h ∈ C ( D ) h\in C(D) h ∈ C ( D ) and f = h ∘ p f=h\circ p f = h ∘ p . Then the f f f -admissible functions are exactly the functions g 0 ∘ p g_0\circ p g 0 ∘ p with g 0 g_0 g 0 h h h -admissible, and f ∈ D ( K ) f\in\mathcal{D}(K) f ∈ D ( K ) if and only if h ∈ D ( D ) h\in\mathcal{D}(D) h ∈ D ( D ) . In particular, if K ∈ D C P K\in DCP K ∈ D C P then D ∈ D C P D\in DCP D ∈ D C P .
Proof. By Lemma 4.14, M f ( k ) ⊇ p − 1 ( p ( k ) ) M_f(k)\supseteq p^{-1}(p(k)) M f ( k ) ⊇ p − 1 ( p ( k )) , so an f f f -admissible g g g is constant on the fibres of p p p and, p p p being a closed surjection, g = g 0 ∘ p g=g_0\circ p g = g 0 ∘ p with g 0 ∈ C ( D ) g_0\in C(D) g 0 ∈ C ( D ) ; moreover g 0 g_0 g 0 is constant on p ( M f ( k ) ) = M h ( p ( k ) ) p(M_f(k))=M_h(p(k)) p ( M f ( k )) = M h ( p ( k )) . Conversely, if g 0 g_0 g 0 is h h h -admissible, then g 0 ∘ p g_0\circ p g 0 ∘ p is constant on p − 1 ( M h ( p ( k ) ) ) = M f ( k ) p^{-1}(M_h(p(k)))=M_f(k) p − 1 ( M h ( p ( k ))) = M f ( k ) . If h ∈ D ( D ) h\in\mathcal{D}(D) h ∈ D ( D ) , then g 0 = H ∘ h g_0=H\circ h g 0 = H ∘ h and g = H ∘ f g=H\circ f g = H ∘ f . If f ∈ D ( K ) f\in\mathcal{D}(K) f ∈ D ( K ) and g 0 g_0 g 0 is h h h -admissible, then g 0 ∘ p = H ∘ f = H ∘ h ∘ p g_0\circ p=H\circ f=H\circ h\circ p g 0 ∘ p = H ∘ f = H ∘ h ∘ p , and g 0 = H ∘ h g_0=H\circ h g 0 = H ∘ h since p p p is surjective. □
Corollary 4.16. If Y Y Y is a nonempty compact metrizable space and K × Y ∈ D C P K\times Y\in DCP K × Y ∈ D C P , then K ∈ D C P K\in DCP K ∈ D C P .
Proof. The projection K × Y → K K\times Y\to K K × Y → K is an open continuous surjection; apply Theorem 4.15 . □
The Cantor comb, the Cantor fan and cones. All positive results above for arbitrary f f f assume that L ( K ) L(K) L ( K ) is dense, and by Remark 4.17 open maps cannot remove this assumption. We now show that two classical continua for which L ( K ) L(K) L ( K ) is not dense are in D C P DCP D C P . The Cantor comb is
M = ( C × [ 0 , 1 ] ) ∪ ( [ 0 , 1 ] × { 0 } ) , M=(C\times[0,1])\cup([0,1]\times\{0\}), M = ( C × [ 0 , 1 ]) ∪ ([ 0 , 1 ] × { 0 }) ,
the comb of [24 Problem 10] . The Cantor fan F F F is the cone C × [ 0 , 1 ] / C × { 0 } C\times[0,1]/C\times\{0\} C × [ 0 , 1 ] / C × { 0 } over the Cantor set; we use its planar model, the union of the straight line segments (legs) joining the vertex v = ( 1 2 , 1 ) v=(\frac{1}{2},1) v = ( 2 1 , 1 ) to the points ( x , 0 ) (x,0) ( x , 0 ) , x ∈ C x\in C x ∈ C (Figure 1 ). The points of the base [ 0 , 1 ] × { 0 } [0,1]\times\{0\} [ 0 , 1 ] × { 0 } of M M M over the gaps of C C C have open arcs as neighbourhoods. Every other point of M M M , and every point of F F F , has a Cantor box (see Lemma 4.18(ii) below) in each of its neighbourhoods; since an open subset of a locally connected space is locally connected and a Cantor box is not, such a point has no locally connected open neighbourhood. Hence L ( M ) = ( ( 0 , 1 ) ∖ C ) × { 0 } L(M)=((0,1)\setminus C)\times\{0\} L ( M ) = (( 0 , 1 ) ∖ C ) × { 0 } , which is not dense in M M M , and L ( F ) = ∅ L(F)=\varnothing L ( F ) = ∅ .
Figure 1. The Cantor comb M M M and the Cantor fan F F F , with C C C replaced by the 64 end points of the fifth step of its construction. In M M M the base over the gap ( 1 3 , 2 3 ) (\frac{1}{3},\frac{2}{3}) ( 3 1 , 3 2 ) (blue) is a part of L ( M ) L(M) L ( M ) ; a Cantor box is outlined in red. The roots of the teeth are on the base of M M M and at the vertex v v v of F F F .
The proof for the fan works for all cones. For a nonempty compact metrizable space Y Y Y let Cone ( Y ) = Y × [ 0 , 1 ] / Y × { 0 } \operatorname{Cone}(Y)=Y\times[0,1]/Y\times\{0\} Cone ( Y ) = Y × [ 0 , 1 ] / Y × { 0 } , with vertex v v v , the image of Y × { 0 } Y\times\{0\} Y × { 0 } ; thus F = Cone ( C ) F=\operatorname{Cone}(C) F = Cone ( C ) . We treat the Cantor fan and the cones at once. Let K K K be M M M , and then put Y = C Y=C Y = C , or K = Cone ( Y ) K=\operatorname{Cone}(Y) K = Cone ( Y ) , and define h : Y × [ 0 , 1 ] → K h:Y\times[0,1]\to K h : Y × [ 0 , 1 ] → K and the root set R ⊆ K \mathcal{R}\subseteq K R ⊆ K by
h ( x , τ ) = ( x , τ ) , R = [ 0 , 1 ] × { 0 } for K = M ; h(x,\tau)=(x,\tau),\qquad\mathcal{R}=[0,1]\times\{0\}\quad\text{for }K=M; h ( x , τ ) = ( x , τ ) , R = [ 0 , 1 ] × { 0 } for K = M ;
h h h the quotient map, and R = { v } \mathcal{R}=\{v\} R = { v } for K = Cone ( Y ) K=\operatorname{Cone}(Y) K = Cone ( Y ) . For F F F in its planar model, h ( x , τ ) = ( 1 − τ ) v + τ ( x , 0 ) h(x,\tau)=(1-\tau)v+\tau(x,0) h ( x , τ ) = ( 1 − τ ) v + τ ( x , 0 ) . The arcs T x = h ( { x } × [ 0 , 1 ] ) T_x=h(\{x\}\times[0,1]) T x = h ({ x } × [ 0 , 1 ]) , x ∈ Y x\in Y x ∈ Y , are the teeth of K K K (the legs of F F F ); h ( x , 0 ) ∈ R h(x,0)\in\mathcal{R} h ( x , 0 ) ∈ R is the root of T x T_x T x , and h ( x , 1 ) h(x,1) h ( x , 1 ) is its free end point.
Lemma 4.18. (i) K K K is an arc connected continuum, h h h is continuous, K = h ( Y × [ 0 , 1 ] ) ∪ R K=h(Y\times[0,1])\cup R K = h ( Y × [ 0 , 1 ]) ∪ R , and h h h maps Y × ( 0 , 1 ] Y\times(0,1] Y × ( 0 , 1 ] homeomorphically onto the open set K ∖ R K\setminus R K ∖ R .
(ii) For A A A open in Y Y Y and J J J open in ( 0 , 1 ] (0,1] ( 0 , 1 ] the set h ( A × J ) h(A\times J) h ( A × J ) is open in K K K ; if A ≠ ∅ A\ne\varnothing A = ∅ and J ⊂ ( 0 , 1 ) J\subset(0,1) J ⊂ ( 0 , 1 ) is an open interval, it is a box (Remark 3.16), and a Cantor box if Y = C Y=C Y = C and A A A is clopen. Every point of h ( Y × ( 0 , 1 ) ) h(Y\times(0,1)) h ( Y × ( 0 , 1 )) has arbitrarily small box neighbourhoods h ( A × ( τ − ε , τ + ε ) ) h(A\times(\tau-\varepsilon,\tau+\varepsilon)) h ( A × ( τ − ε , τ + ε )) .
(iii) Every point of R R R has arbitrarily small open neighbourhoods U ′ U' U ′ that are closed downwards: if h ( x , τ ) ∈ U ′ h(x,\tau)\in U' h ( x , τ ) ∈ U ′ , then h ( x , σ ) ∈ U ′ h(x,\sigma)\in U' h ( x , σ ) ∈ U ′ for 0 ≤ σ ≤ τ 0\le\sigma\le\tau 0 ≤ σ ≤ τ .
Proof. (i) For M M M this is clear. For Cone ( Y ) \operatorname{Cone}(Y) Cone ( Y ) , the restriction of the quotient map h h h to the saturated open set Y × ( 0 , 1 ] Y\times(0,1] Y × ( 0 , 1 ] is a bijective quotient map onto the open set K ∖ { v } K\setminus\{v\} K ∖ { v } , hence a homeomorphism; every point is joined to v v v by a tooth. (ii) follows from (i). (iii) For u = ( x 0 , 0 ) ∈ M u=(x_0,0)\in M u = ( x 0 , 0 ) ∈ M take U ′ = M ∩ ( ( x 0 − ε , x 0 + ε ) × [ 0 , ε ) ) U'=M\cap((x_0-\varepsilon,x_0+\varepsilon)\times[0,\varepsilon)) U ′ = M ∩ (( x 0 − ε , x 0 + ε ) × [ 0 , ε )) ; for Cone ( Y ) \operatorname{Cone}(Y) Cone ( Y ) take U ′ = h ( Y × [ 0 , δ ) ) U'=h(Y\times[0,\delta)) U ′ = h ( Y × [ 0 , δ )) , whose complement h ( Y × [ δ , 1 ] ) h(Y\times[\delta,1]) h ( Y × [ δ , 1 ]) is compact; these sets form a base at v v v because Y Y Y is compact. □
For the rest of this subsection f ∈ C ( K ) f\in C(K) f ∈ C ( K ) and an f f f -admissible g g g are fixed, Σ \Sigma Σ is the set of points of K K K whose level is collapsed, and Ω = K ∖ cl Σ \Omega=K\setminus\operatorname{cl}\Sigma Ω = K ∖ cl Σ . Thus Ω \Omega Ω is open, and an open set disjoint from Σ \Sigma Σ is contained in Ω \Omega Ω .
Lemma 4.19. f f f is constant on no nonempty open subset of Ω \Omega Ω , and f f f is constant on every connected subset of Ω \Omega Ω that is contained in a tooth.
Proof. A nonempty open set on which f f f is constant lies in Σ \Sigma Σ by Lemma 3.1 , so it is not contained in Ω \Omega Ω . Suppose f f f is not constant on a connected set J ⊆ Ω J\subseteq\Omega J ⊆ Ω contained in a tooth T x T_x T x ; then J = h ( { x } × I J ) J=h(\{x\}\times I_J) J = h ({ x } × I J ) for an interval I J ⊆ [ 0 , 1 ] I_J\subseteq[0,1] I J ⊆ [ 0 , 1 ] . Choose τ 1 < τ 2 \tau_1<\tau_2 τ 1 < τ 2 in I J I_J I J with f ( h ( x , τ 1 ) ) ≠ f ( h ( x , τ 2 ) ) f(h(x,\tau_1))\ne f(h(x,\tau_2)) f ( h ( x , τ 1 )) = f ( h ( x , τ 2 )) , and a level t t t strictly between these values such that f ( h ( x , ⋅ ) ) f(h(x,\cdot)) f ( h ( x , ⋅ )) is not identically t t t on any nondegenerate subinterval of [ τ 1 , τ 2 ] [\tau_1,\tau_2] [ τ 1 , τ 2 ] (all but countably many t t t qualify). As in the proof of Lemma 3.14 there is τ ∈ ( τ 1 , τ 2 ) \tau\in(\tau_1,\tau_2) τ ∈ ( τ 1 , τ 2 ) such that s = h ( x , τ ) s=h(x,\tau) s = h ( x , τ ) satisfies f ( s ) = t f(s)=t f ( s ) = t and every subinterval of [ τ 1 , τ 2 ] [\tau_1,\tau_2] [ τ 1 , τ 2 ] containing τ \tau τ in its interior contains points where f ( h ( x , ⋅ ) ) < t f(h(x,\cdot))<t f ( h ( x , ⋅ )) < t and points where it is > t >t > t . Since 0 < τ < 1 0<\tau<1 0 < τ < 1 and s ∈ J ⊆ Ω s\in J\subseteq\Omega s ∈ J ⊆ Ω , by Lemma 4.18 (ii) there is a box D = h ( A × ( τ − ε , τ + ε ) ) ⊆ Ω D=h(A\times(\tau-\varepsilon,\tau+\varepsilon))\subseteq\Omega D = h ( A × ( τ − ε , τ + ε )) ⊆ Ω , and s s s is a crossing point of D D D . By the first part f f f is constant on no nonempty open subset of D D D , so by Proposition 3.15 (and Remark 3.16) the level t t t is collapsed, i.e. s ∈ Σ ∩ Ω = ∅ s\in\Sigma\cap\Omega=\varnothing s ∈ Σ ∩ Ω = ∅ , a contradiction. □
Lemma 4.20. R ⊆ cl Σ R\subseteq\operatorname{cl}\Sigma R ⊆ cl Σ . Consequently Ω ⊆ h ( Y × ( 0 , 1 ) ) \Omega\subseteq h(Y\times(0,1)) Ω ⊆ h ( Y × ( 0 , 1 )) .
Proof. The comb. Let V V V be an open interval of the base over a gap of C C C ; it is open in M M M , connected and locally connected. If f f f is constant on a nonempty open subset of V V V , that subset lies in Σ \Sigma Σ by Lemma 3.1 . Otherwise f ( V ) f(V) f ( V ) is a nondegenerate interval and, as in the proof of Lemma 3.10, V V V contains an openness point of f f f , whose level is collapsed by Lemma 3.6 (with W = V W=V W = V ). So Σ \Sigma Σ meets every such V V V , and R = cl L ( M ) ⊆ cl Σ \mathcal{R}=\operatorname{cl}L(M)\subseteq\operatorname{cl}\Sigma R = cl L ( M ) ⊆ cl Σ .
The cones. Suppose v ∈ Ω v\in\Omega v ∈ Ω . By Lemma 4.18 (iii) there is δ > 0 \delta>0 δ > 0 such that the open set S = h ( Y × [ 0 , δ ) ) S=h(Y\times[0,\delta)) S = h ( Y × [ 0 , δ )) is contained in Ω \Omega Ω . For each x ∈ Y x\in Y x ∈ Y the set h ( { x } × [ 0 , δ ) ) h(\{x\}\times[0,\delta)) h ({ x } × [ 0 , δ )) is connected, contained in Ω \Omega Ω and in the tooth T x T_x T x , and contains v v v ; by Lemma 4.19 , f ≡ f ( v ) f\equiv f(v) f ≡ f ( v ) on it. Hence f f f is constant on the nonempty open set S ⊆ Ω S\subseteq\Omega S ⊆ Ω , contrary to Lemma 4.19 . So v ∈ cl Σ v\in\operatorname{cl}\Sigma v ∈ cl Σ .
Finally K ∖ R = h ( Y × ( 0 , 1 ] ) K\setminus\mathcal{R}=h(Y\times(0,1]) K ∖ R = h ( Y × ( 0 , 1 ]) .
Lemma 4.21. Let c ∈ f ( K ) c\in f(K) c ∈ f ( K ) . Every u ∈ Z c + u\in Z_c^{+} u ∈ Z c + admits a sequence y n y_n y n as in condition (a) of Theorem 4.2 . The same holds for Z c − Z_c^{-} Z c − .
Proof. The statement for Z c − Z_c^{-} Z c − follows by applying the statement for Z − c + Z_{-c}^{+} Z − c + to − f -f − f , which has the same fibres, the same M f M_f M f , the same admissible functions and the same Σ \Sigma Σ .
If u ∈ cl ( Σ ∩ { f > c } ) u\in\operatorname{cl}(\Sigma\cap\{f>c\}) u ∈ cl ( Σ ∩ { f > c }) , take y n ∈ Σ ∩ { f > c } y_n\in\Sigma\cap\{f>c\} y n ∈ Σ ∩ { f > c } with y n → u y_n\to u y n → u . Otherwise there is an open neighbourhood U U U of u u u with U ∩ { f > c } ⊆ Ω U\cap\{f>c\}\subseteq\Omega U ∩ { f > c } ⊆ Ω ; then U ′ ∩ { f > c } U'\cap\{f>c\} U ′ ∩ { f > c } is a nonempty open subset of Ω \Omega Ω for every open U ′ ⊆ U U'\subseteq U U ′ ⊆ U containing u u u , because u ∈ cl { f > c } u\in\operatorname{cl}\{f>c\} u ∈ cl { f > c } .
u u u is not in R \mathcal{R} R . Otherwise choose an open neighbourhood U ′ ⊆ U U'\subseteq U U ′ ⊆ U of u u u that is closed downwards (Lemma 4.18 (iii)), and put W ′ = U ′ ∩ { f > c } W'=U'\cap\{f>c\} W ′ = U ′ ∩ { f > c } . By Lemma 4.20 , W ′ ∩ R = ∅ W'\cap\mathcal{R}=\varnothing W ′ ∩ R = ∅ ; take p = h ( x , τ ) ∈ W ′ p=h(x,\tau)\in W' p = h ( x , τ ) ∈ W ′ , τ > 0 \tau>0 τ > 0 . The set { σ ∈ ( 0 , 1 ] : h ( x , σ ) ∈ W ′ } \{\sigma\in(0,1]:h(x,\sigma)\in W'\} { σ ∈ ( 0 , 1 ] : h ( x , σ ) ∈ W ′ } is open in ( 0 , 1 ] (0,1] ( 0 , 1 ] ; let I I I be its component containing τ \tau τ , and α = inf I < τ \alpha=\inf I<\tau α = inf I < τ . By Lemma 4.19 , f ≡ f ( p ) > c f\equiv f(p)>c f ≡ f ( p ) > c on h ( { x } × cl I ) h(\{x\}\times\operatorname{cl}I) h ({ x } × cl I ) . The point e = h ( x , α ) e=h(x,\alpha) e = h ( x , α ) lies in U ′ U' U ′ , since U ′ U' U ′ is closed downwards, and f ( e ) > c f(e)>c f ( e ) > c , so e ∈ W ′ e\in W' e ∈ W ′ . If α = 0 \alpha=0 α = 0 , then e ∈ R e\in\mathcal{R} e ∈ R , contrary to W ′ ∩ R = ∅ W'\cap\mathcal{R}=\varnothing W ′ ∩ R = ∅ . If α > 0 \alpha>0 α > 0 , then an interval about α \alpha α lies in the open set { σ : h ( x , σ ) ∈ W ′ } \{\sigma:h(x,\sigma)\in W'\} { σ : h ( x , σ ) ∈ W ′ } , contrary to α = inf I \alpha=\inf I α = inf I .
u = h ( x u , τ u ) u=h(x_u,\tau_u) u = h ( x u , τ u ) with 0 < τ u ≤ 1 0<\tau_u\leq1 0 < τ u ≤ 1 (this includes the free end points). Choose 0 < ε < τ u 0<\varepsilon<\tau_u 0 < ε < τ u and an open A u ∋ x u A_u\ni x_u A u ∋ x u in Y Y Y such that U 0 = h ( A u × J u ) ⊆ U U_0=h(A_u\times J_u)\subseteq U U 0 = h ( A u × J u ) ⊆ U , where J u = ( τ u − ε , τ u + ε ) ∩ ( 0 , 1 ] J_u=(\tau_u-\varepsilon,\tau_u+\varepsilon)\cap(0,1] J u = ( τ u − ε , τ u + ε ) ∩ ( 0 , 1 ] ; U 0 U_0 U 0 is open by Lemma 4.18 (ii). Put W = U 0 ∩ { f > c } W=U_0\cap\{f>c\} W = U 0 ∩ { f > c } , a nonempty open subset of Ω \Omega Ω with u ∈ cl W u\in\operatorname{cl}W u ∈ cl W .
Let x ∈ A u x\in A_u x ∈ A u and let I I I be a component of the open set { σ ∈ J u : h ( x , σ ) ∈ W } \{\sigma\in J_u:h(x,\sigma)\in W\} { σ ∈ J u : h ( x , σ ) ∈ W } . By Lemma 4.19 , f f f is constant, > c >c > c , on h ( { x } × cl I ) h(\{x\}\times\operatorname{cl}I) h ({ x } × cl I ) . A boundary point σ \sigma σ of I I I in J u J_u J u would therefore satisfy h ( x , σ ) ∈ W h(x,\sigma)\in W h ( x , σ ) ∈ W , and an interval about σ \sigma σ would lie in { σ ′ ∈ J u : h ( x , σ ′ ) ∈ W } \{\sigma'\in J_u:h(x,\sigma')\in W\} { σ ′ ∈ J u : h ( x , σ ′ ) ∈ W } , contrary to the maximality of I I I ; so I = J u I=J_u I = J u . Hence W = h ( A W × J u ) W=h(A_W\times J_u) W = h ( A W × J u ) , where A W = { x ∈ A u : h ( x , τ u ) ∈ W } A_W=\{x\in A_u:h(x,\tau_u)\in W\} A W = { x ∈ A u : h ( x , τ u ) ∈ W } is open in Y Y Y , and f ( h ( x , σ ) ) = φ ( x ) : = f ( h ( x , τ u ) ) > c f(h(x,\sigma))=\varphi(x):=f(h(x,\tau_u))>c f ( h ( x , σ )) = φ ( x ) := f ( h ( x , τ u )) > c for x ∈ A W x\in A_W x ∈ A W , σ ∈ J u \sigma\in J_u σ ∈ J u . Moreover x u ∈ cl A W x_u\in\operatorname{cl}A_W x u ∈ cl A W , and φ \varphi φ , defined by the same formula on A u A_u A u , is continuous with φ ( x u ) = c \varphi(x_u)=c φ ( x u ) = c . For x ∈ A W x\in A_W x ∈ A W put
β ( x ) = inf { s ∈ [ 0 , τ u ] : f ( h ( x , σ ) ) = φ ( x ) for all σ ∈ [ s , τ u ] } , e x = h ( x , β ( x ) ) . \beta(x)=\inf\{s\in[0,\tau_u]:f(h(x,\sigma))=\varphi(x)\text{ for all }\sigma\in[s,\tau_u]\},\qquad e_x=h(x,\beta(x)). β ( x ) = inf { s ∈ [ 0 , τ u ] : f ( h ( x , σ )) = φ ( x ) for all σ ∈ [ s , τ u ]} , e x = h ( x , β ( x )) .
Then β ( x ) ≤ τ u − ε \beta(x)\leq\tau_u-\varepsilon β ( x ) ≤ τ u − ε , and f ( h ( x , σ ) ) = φ ( x ) f(h(x,\sigma))=\varphi(x) f ( h ( x , σ )) = φ ( x ) for σ ∈ [ β ( x ) , τ u ] \sigma\in[\beta(x),\tau_u] σ ∈ [ β ( x ) , τ u ] ; in particular f ( e x ) = φ ( x ) > c f(e_x)=\varphi(x)>c f ( e x ) = φ ( x ) > c .
We have e x ∈ cl Σ e_x\in\operatorname{cl}\Sigma e x ∈ cl Σ : if β ( x ) = 0 \beta(x)=0 β ( x ) = 0 , then e x ∈ R ⊆ cl Σ e_x\in\mathcal{R}\subseteq\operatorname{cl}\Sigma e x ∈ R ⊆ cl Σ by Lemma 4.20 ; if β ( x ) > 0 \beta(x)>0 β ( x ) > 0 and e x ∈ Ω e_x\in\Omega e x ∈ Ω , then by Lemma 4.19 f ≡ f ( e x ) = φ ( x ) f\equiv f(e_x)=\varphi(x) f ≡ f ( e x ) = φ ( x ) on a neighbourhood of e x e_x e x in T x T_x T x , contrary to the definition of β ( x ) \beta(x) β ( x ) . The function β \beta β is lower semicontinuous on A W A_W A W , because for 0 ≤ s ≤ τ u 0\leq s\leq\tau_u 0 ≤ s ≤ τ u the set { x ∈ A W : β ( x ) ≤ s } = { x ∈ A W : f ( h ( x , σ ) ) = φ ( x ) for all σ ∈ [ s , τ u ] } \{x\in A_W:\beta(x)\leq s\}=\{x\in A_W:f(h(x,\sigma))=\varphi(x)\text{ for all }\sigma\in[s,\tau_u]\} { x ∈ A W : β ( x ) ≤ s } = { x ∈ A W : f ( h ( x , σ )) = φ ( x ) for all σ ∈ [ s , τ u ]} is closed in A W A_W A W . Since A W A_W A W is open in the compact metrizable space Y Y Y , it is a Baire space, and the points of continuity of β \beta β are dense in A W A_W A W .
Let x ∈ A W x\in A_W x ∈ A W be a point of continuity of β \beta β , and 0 < η < ε 0<\eta<\varepsilon 0 < η < ε . There is an open neighbourhood A ′ ′ ⊆ A W A''\subseteq A_W A ′′ ⊆ A W of x x x with β < β ( x ) + η \beta<\beta(x)+\eta β < β ( x ) + η on A ′ ′ A'' A ′′ . The set O = h ( A ′ ′ × ( ( β ( x ) + η , τ u + ε ) ∩ ( 0 , 1 ] ) ) O=h(A''\times((\beta(x)+\eta,\tau_u+\varepsilon)\cap(0,1])) O = h ( A ′′ × (( β ( x ) + η , τ u + ε ) ∩ ( 0 , 1 ])) is open and contains h ( x , τ u ) h(x,\tau_u) h ( x , τ u ) , and f ( h ( x ′ , σ ) ) = φ ( x ′ ) f(h(x',\sigma))=\varphi(x') f ( h ( x ′ , σ )) = φ ( x ′ ) on O O O : for σ ≥ τ u \sigma\geq\tau_u σ ≥ τ u because h ( x ′ , σ ) ∈ W h(x',\sigma)\in W h ( x ′ , σ ) ∈ W , and for σ < τ u \sigma<\tau_u σ < τ u because σ > β ( x ′ ) \sigma>\beta(x') σ > β ( x ′ ) . So r ( h ( x ′ , σ ) ) = h ( x ′ , β ( x ) + η ) r(h(x',\sigma))=h(x',\beta(x)+\eta) r ( h ( x ′ , σ )) = h ( x ′ , β ( x ) + η ) defines a continuous map r : O → K r:O\to K r : O → K with f ∘ r = f f\circ r=f f ∘ r = f on O O O , and by Lemma 3.11 g ( h ( x , τ u ) ) = g ( h ( x , β ( x ) + η ) ) g(h(x,\tau_u))=g(h(x,\beta(x)+\eta)) g ( h ( x , τ u )) = g ( h ( x , β ( x ) + η )) . Letting η → 0 \eta\to0 η → 0 , g ( h ( x , τ u ) ) = g ( e x ) g(h(x,\tau_u))=g(e_x) g ( h ( x , τ u )) = g ( e x ) .
Choose points of continuity x n x_n x n of β \beta β with x n → x u x_n\to x_u x n → x u and put y n = e x n y_n=e_{x_n} y n = e x n . Then y n ∈ cl Σ y_n\in\operatorname{cl}\Sigma y n ∈ cl Σ , f ( y n ) = φ ( x n ) > c f(y_n)=\varphi(x_n)>c f ( y n ) = φ ( x n ) > c , f ( y n ) → φ ( x u ) = c f(y_n)\to\varphi(x_u)=c f ( y n ) → φ ( x u ) = c and g ( y n ) = g ( h ( x n , τ u ) ) → g ( u ) g(y_n)=g(h(x_n,\tau_u))\to g(u) g ( y n ) = g ( h ( x n , τ u )) → g ( u ) .
Theorem 4.22. The Cantor comb M M M and the cone Cone ( Y ) \operatorname{Cone}(Y) Cone ( Y ) over every nonempty compact metrizable space Y Y Y are in D C P \mathcal{DCP} D C P . In particular the Cantor fan F F F is in D C P \mathcal{DCP} D C P , although L ( M ) L(M) L ( M ) is not dense in M M M and L ( F ) = ∅ L(F)=\varnothing L ( F ) = ∅ .
Proof. Let K K K be M M M or Cone ( Y ) \operatorname{Cone}(Y) Cone ( Y ) , f ∈ C ( K ) f\in C(K) f ∈ C ( K ) , g g g f f f -admissible and c ∈ f ( K ) c\in f(K) c ∈ f ( K ) . If Int Z c ≠ ∅ \operatorname{Int} Z_c\ne\varnothing Int Z c = ∅ , the level c c c is collapsed by Lemma 3.1. Otherwise condition (a) of Theorem 4.2 holds by Lemma 4.21, and (b) holds because K K K is arc connected (see the remark preceding Theorem 4.4 ). Hence every level is collapsed.
Corollary 4.23. Every smooth fan whose set of end points is closed is in D C P \mathcal{DCP} D C P .
Proof. Such a fan is either an arc, which is locally connected, or an open image of F F F [10 Theorem 8] ; apply Theorem 4.15 .
Remark 4.24. (1) Theorem 4.22 answers Problem 10 of [24 ] and the question, raised there, whether the Cantor fan is in D C P \mathcal{DCP} D C P .
(2) The proof follows each plateau of f f f on the arcs of a box towards the root of the tooth; the free end points of the teeth are never needed. What is needed is that the roots lie in cl Σ \operatorname{cl}\Sigma cl Σ for every f f f : for the comb this holds because R = cl L ( M ) \mathcal{R}=\operatorname{cl}L(M) R = cl L ( M ) , for the cones because of the argument at the vertex in Lemma 4.20.
(3) For cones, Y Y Y need not be connected, nor in D C P \mathcal{DCP} D C P , nor contain any arc. Besides the Cantor fan, Theorem 4.22 applies, e.g., to the cones over solenoids, over the buckethandle continuum and over the pseudo-arc; for these L ( Y ) L(Y) L ( Y ) is not dense, so Theorem 4.4 does not apply. (Since Cone ( Y ) ∖ { v } ≅ Y × ( 0 , 1 ] \operatorname{Cone}(Y)\setminus\{v\}\cong Y\times(0,1] Cone ( Y ) ∖ { v } ≅ Y × ( 0 , 1 ] , we have L ( Cone ( Y ) ) ∖ { v } = h ( L ( Y ) × ( 0 , 1 ] ) L(\operatorname{Cone}(Y))\setminus\{v\}=h(L(Y)\times(0,1]) L ( Cone ( Y )) ∖ { v } = h ( L ( Y ) × ( 0 , 1 ]) , so L ( Cone ( Y ) ) L(\operatorname{Cone}(Y)) L ( Cone ( Y )) is dense if and only if L ( Y ) L(Y) L ( Y ) is. For the harmonic fan Cone ( { 0 } ∪ { 1 / n : n ∈ N } ) \operatorname{Cone}(\{0\}\cup\{1/n:n\in\mathbb{N}\}) Cone ({ 0 } ∪ { 1/ n : n ∈ N }) , L ( Y ) L(Y) L ( Y ) is dense and Theorem 4.4 applies as well; compare Example 4.11 .) By Corollary 4.16, products behave differently: for the sin ( 1 / x ) \sin(1/x) sin ( 1/ x ) continuum S S S of Example 4.11 , Cone ( S ) ∈ D C P \operatorname{Cone}(S)\in\mathcal{DCP} Cone ( S ) ∈ D C P but S × [ 0 , 1 ] ∉ D C P S\times[0,1]\notin\mathcal{DCP} S × [ 0 , 1 ] ∈ / D C P . By Theorem 4.15 , open images of cones are in D C P \mathcal{DCP} D C P .
(4) The same proof, with R = [ 0 , 1 ] × { 0 , 1 } \mathcal{R}=[0,1]\times\{0,1\} R = [ 0 , 1 ] × { 0 , 1 } and, at the points of the top bar, neighbourhoods closed upwards, shows that the Cantor ladder ( C × [ 0 , 1 ] ) ∪ ( [ 0 , 1 ] × { 0 , 1 } ) (C\times[0,1])\cup([0,1]\times\{0,1\}) ( C × [ 0 , 1 ]) ∪ ([ 0 , 1 ] × { 0 , 1 }) is in D C P \mathcal{DCP} D C P ; similarly, with two vertices, the suspension of any nonempty compact metrizable space is in D C P \mathcal{DCP} D C P .
Rooted teeth
The proofs of Lemmas 4.19–4.21 use only the following structure, which also underlies Theorems 4.4 and 4.7 .
Recall from Definition 1.3 the notions of a tooth chart h : U × [ 0 , 1 ] → K h:U\times[0,1]\to K h : U × [ 0 , 1 ] → K with h ( U × J h ) = O h h(U\times J_h)=O_h h ( U × J h ) = O h open, of open teeth h ( { y } × J h ) h(\{y\}\times J_h) h ({ y } × J h ) and their roots h ( y , 0 ) h(y,0) h ( y , 0 ) , and of star points.
Theorem 4.25. Let K K K be a metrizable continuum with a family of tooth charts, let B B B be the union of the sets O h O_h O h , and let R ⊆ K \mathcal{R}\subseteq K R ⊆ K be closed. Assume:
(1) K K K is arc connected, or every arc component P P P of K K K is dense or satisfies P ⊆ cl Int P P\subseteq\operatorname{cl}\operatorname{Int}P P ⊆ cl Int P ;
(2) K = cl L ( K ) ∪ R ∪ B K=\operatorname{cl}L(K)\cup\mathcal{R}\cup B K = cl L ( K ) ∪ R ∪ B , and h ( U × { 0 } ) ⊆ R ∪ cl L ( K ) h(U\times\{0\})\subseteq\mathcal{R}\cup\operatorname{cl}L(K) h ( U × { 0 }) ⊆ R ∪ cl L ( K ) for every chart h h h ;
(3) every point of ( R ∪ cl L ( K ) ) ∖ ( L ( K ) ∪ B ) (\mathcal{R}\cup\operatorname{cl}L(K))\setminus(L(K)\cup B) ( R ∪ cl L ( K )) ∖ ( L ( K ) ∪ B ) has arbitrarily small open neighbourhoods U ′ U' U ′ such that every component of U ′ ∩ T U'\cap T U ′ ∩ T , T T T an open tooth, has an end point in U ′ ∖ B U'\setminus B U ′ ∖ B (an end point of h ( { y } × I ) h(\{y\}\times I) h ({ y } × I ) , I I I an interval, is a point h ( y , α ) h(y,\alpha) h ( y , α ) with α \alpha α an end point of I I I );
(4) every point of R \mathcal{R} R lies in cl L ( K ) \operatorname{cl}L(K) cl L ( K ) or is a star point.
Then K ∈ D C P K\in\mathcal{DCP} K ∈ D C P .
Proof. Fix f ∈ C ( K ) f\in C(K) f ∈ C ( K ) and an f f f -admissible g g g , and let Σ \Sigma Σ and Ω = K ∖ cl Σ \Omega=K\setminus\operatorname{cl}\Sigma Ω = K ∖ cl Σ be as before.
(i) L ( K ) ⊆ cl Σ L(K)\subseteq\operatorname{cl}\Sigma L ( K ) ⊆ cl Σ : every nonempty open subset of L ( K ) L(K) L ( K ) contains a nonempty connected open locally connected set, which contains a point of D f ⊆ Σ D_f\subseteq\Sigma D f ⊆ Σ by the proof of Lemma 3.10.
(ii) As in Lemma 4.19, f f f is constant on no nonempty open subset of Ω \Omega Ω and on every connected subset of Ω \Omega Ω contained in an open tooth: the crossing point s = h ( y , τ ) s=h(y,\tau) s = h ( y , τ ) found there has 0 < τ < 1 0<\tau<1 0 < τ < 1 , so it has a box neighbourhood h ( U ′ ′ × ( τ − δ , τ + δ ) ) ⊆ Ω h(U'' \times(\tau-\delta,\tau+\delta)) \subseteq\Omega h ( U ′′ × ( τ − δ , τ + δ )) ⊆ Ω , and Proposition 3.15 with Remark 3.16 applies.
(iii) R ⊆ cl Σ \mathcal{R} \subseteq\operatorname{cl}\Sigma R ⊆ cl Σ , hence Ω ⊆ B \Omega\subseteq B Ω ⊆ B : points of R ∩ cl L ( K ) \mathcal{R} \cap\operatorname{cl}L(K) R ∩ cl L ( K ) are in cl Σ \operatorname{cl}\Sigma cl Σ by (i); if a star point v v v were in Ω \Omega Ω , then, as for the vertex in Lemma 4.20 , f f f would be constant, equal to f ( v ) f(v) f ( v ) , on a star neighbourhood S ⊆ Ω S \subseteq\Omega S ⊆ Ω , contrary to (ii). By (2), Ω ⊆ B \Omega\subseteq B Ω ⊆ B .
(iv) Condition (a) of Theorem 4.2 . Let u ∈ Z c u \in Z^c u ∈ Z c (Z c Z^c Z c is handled by passing to − f -f − f ). If u ∉ cl ( Σ ∩ { f > c } ) u \notin\operatorname{cl}(\Sigma\cap\{f>c\}) u ∈ / cl ( Σ ∩ { f > c }) , fix an open N ∋ u N \ni u N ∋ u with N ∩ { f > c } ⊆ Ω ⊆ B N \cap\{f>c\} \subseteq\Omega\subseteq B N ∩ { f > c } ⊆ Ω ⊆ B . Since u ∈ cl ( N ∩ { f > c } ) ⊆ cl Ω u \in\operatorname{cl}(N \cap\{f>c\}) \subseteq\operatorname{cl}\Omega u ∈ cl ( N ∩ { f > c }) ⊆ cl Ω and L ( K ) L(K) L ( K ) is open and disjoint from Ω \Omega Ω , u ∉ L ( K ) u \notin L(K) u ∈ / L ( K ) . If u ∈ B u \in B u ∈ B , then u = h ( x u , τ u ) u=h(x_u,\tau_u) u = h ( x u , τ u ) with τ u ∈ J h \tau_u \in J_h τ u ∈ J h , and the second case of the proof of Lemma 4.21 applies in the chart h h h , with J u = ( τ u − ε , τ u + ε ) ∩ J h J_u=(\tau_u-\varepsilon,\tau_u+\varepsilon)\cap J_h J u = ( τ u − ε , τ u + ε ) ∩ J h . (There the openness of the sets h ( A × J ) h(A\times J) h ( A × J ) now follows from O h O_h O h being open and h h h being a homeomorphism on U × J h U\times J_h U × J h ; Baire’s theorem is applied to an open subset of U U U ; and the continuity of h h h on U × [ 0 , 1 ] U\times[0,1] U × [ 0 , 1 ] makes the relevant sublevel sets of β \beta β closed and puts each e x e_x e x in the closure of its plateau.) The end points e x e_x e x of the plateaus lie in cl Σ \operatorname{cl}\Sigma cl Σ , either as roots, which lie in R ∪ cl L ( K ) ⊆ cl Σ \mathcal{R}\cup\operatorname{cl}L(K)\subseteq\operatorname{cl}\Sigma R ∪ cl L ( K ) ⊆ cl Σ , or by (ii). Otherwise u ∈ ( R ∪ cl L ( K ) ) ∖ ( L ( K ) ∪ B ) u\in(\mathcal{R}\cup\operatorname{cl}L(K))\setminus(L(K)\cup B) u ∈ ( R ∪ cl L ( K )) ∖ ( L ( K ) ∪ B ) by (2). Take U ′ ⊆ N U'\subseteq N U ′ ⊆ N as in (3), put W ′ = U ′ ∩ { f > c } W'=U'\cap\{f>c\} W ′ = U ′ ∩ { f > c } , a nonempty open subset of Ω \Omega Ω , and let p ∈ W ′ p\in W' p ∈ W ′ lie on the open tooth T T T . Let J J J be the component of U ′ ∩ T U'\cap T U ′ ∩ T and I ⊆ J I\subseteq J I ⊆ J the component of W ′ ∩ T W'\cap T W ′ ∩ T containing p p p . By (ii), f ≡ f ( p ) > c f\equiv f(p)>c f ≡ f ( p ) > c on cl I \operatorname{cl}I cl I , so a boundary point of I I I in J J J would lie in W ′ W' W ′ ; hence I = J I=J I = J . The end point e ∈ U ′ ∖ B e\in U'\setminus B e ∈ U ′ ∖ B of J J J given by (3) satisfies f ( e ) > c f(e)>c f ( e ) > c , so e ∈ W ′ ⊆ B e\in W'\subseteq B e ∈ W ′ ⊆ B , a contradiction.
Finally, levels with interior are collapsed by Lemma 3.1 , and condition (b) holds by (1) (the remark preceding Theorem 4.4 , or Lemma 4.6). Theorem 4.2 applies to every level. □
Remark 4.26. (1) Without charts (B = ∅ B=\varnothing B = ∅ ) and with R = K ∖ L ( K ) \mathcal{R}=K\setminus L(K) R = K ∖ L ( K ) , Theorem 4.25 is Theorems 4.4 and 4.7 . The comb and the cones are covered with the charts of Lemma 4.18 , with R = C × { 0 } ⊆ cl L ( M ) \mathcal{R}=C\times\{0\}\subseteq\operatorname{cl}L(M) R = C × { 0 } ⊆ cl L ( M ) , resp. R = { v } \mathcal{R}=\{v\} R = { v } , a star point; the Cantor ladder with R = C × { 0 , 1 } ⊆ cl L \mathcal{R}=C\times\{0,1\}\subseteq\operatorname{cl}L R = C × { 0 , 1 } ⊆ cl L . In a cone, points of cl L ∖ L \operatorname{cl}L\setminus L cl L ∖ L may lie in B B B (e.g. the leg over 0 in the harmonic fan); this is why (3) is required only outside B B B . The suspension of a nonempty compact metrizable space is covered with R \mathcal{R} R the two vertices, both star points.
(2) The Cantor grid below satisfies (1)–(3), with the nodes as R \mathcal{R} R and the charts ( x , τ ) ↦ ( x , a + τ ( b − a ) ) (x,\tau)\mapsto(x,a+\tau(b-a)) ( x , τ ) ↦ ( x , a + τ ( b − a )) over the gaps ( a , b ) (a,b) ( a , b ) (and their horizontal analogues), but the nodes are neither in cl L ( G ) = ∅ \operatorname{cl}L(G)=\varnothing cl L ( G ) = ∅ nor star points. In fact no family of tooth charts makes G G G satisfy Theorem 4.25 . Since G G G is one-dimensional, each O h ≅ U × J h O_h\cong U\times J_h O h ≅ U × J h has dim U = 0 \dim U=0 dim U = 0 , so O h O_h O h contains no simple triod; as every neighbourhood of a node contains one, no node lies in B B B . If v ∈ G v\in G v ∈ G is not a node, every sufficiently small neighbourhood S S S of v v v meets other segments parallel to the one containing v v v , and their points cannot be joined to v v v by an arc in S S S ; and every neighbourhood of a node contains other nodes, which lie in no open tooth. Hence G G G has no star points. Thus (4) forces R ⊆ cl L ( G ) = ∅ \mathcal{R}\subseteq\operatorname{cl}L(G)=\varnothing R ⊆ cl L ( G ) = ∅ , and then (2) fails at the nodes. (Heuristically, along a line through a node f f f may vary like the Cantor function.) The main step of the proof for the grid, Lemma 4.29 , shows that nevertheless Ω \Omega Ω contains no node; with this in place of (4) the proof above becomes the proof of Theorem 4.31 .
The Cantor grid. In the comb and the cones every arc of a box lies in a tooth whose root is in cl Σ \operatorname{cl}\Sigma cl Σ for every f f f (Lemma 4.20 ). We now give a continuum in D C P \mathcal{DCP} D C P in which, as in the fan, no point has an open locally connected neighbourhood, but in which both ends of every such arc are points of a Cantor set of nodes , about which nothing of this kind is known a priori.
Recall that C C C is the ternary Cantor set. Let E \mathcal{E} E be the set of end points of the bounded components (gaps) of [ 0 , 1 ] ∖ C [0,1]\setminus C [ 0 , 1 ] ∖ C , and ϑ \vartheta ϑ the Cantor function; thus ϑ ( E ) \vartheta(\mathcal{E}) ϑ ( E ) is the set Δ \Delta Δ of dyadic rationals in ( 0 , 1 ) (0,1) ( 0 , 1 ) . (We write ϑ \vartheta ϑ rather than the usual F F F , which denotes the Cantor fan.) The Cantor grid is
G = ( C × [ 0 , 1 ] ) ∪ ( [ 0 , 1 ] × C ) . G=(C\times[0,1])\cup([0,1]\times C). G = ( C × [ 0 , 1 ]) ∪ ([ 0 , 1 ] × C ) .
The points of C × C C \times C C × C are the nodes of G G G . For a gap ( a , b ) (a,b) ( a , b ) and x , y ∈ C x,y \in C x , y ∈ C the arcs { x } × [ a , b ] \{x\} \times[a,b] { x } × [ a , b ] and [ a , b ] × { y } [a,b] \times\{y\} [ a , b ] × { y } are the (vertical and horizontal) segments of G G G . Every point of G G G that is not a node lies in the relative interior of exactly one segment, and the sets ( C ∩ I ) × ( a , b ) (C \cap I) \times(a,b) ( C ∩ I ) × ( a , b ) and ( a , b ) × ( C ∩ I ) (a,b) \times(C \cap I) ( a , b ) × ( C ∩ I ) , I I I an open interval with end points not in C C C , are Cantor boxes in G G G (with vertical, resp. horizontal, arcs). Put κ = ϑ × ϑ : G → [ 0 , 1 ] 2 \kappa= \vartheta\times\vartheta: G \to[0,1]^2 κ = ϑ × ϑ : G → [ 0 , 1 ] 2 .
Lemma 4.27. (i) G G G is an arc connected planar continuum and L ( G ) = ∅ L(G) = \varnothing L ( G ) = ∅ .
(ii) Let z = ( x , y ) z = (x,y) z = ( x , y ) with x , y ∈ C ∖ ( E ∪ { 0 , 1 } ) x,y \in C \setminus(\mathcal{E} \cup\{0,1\}) x , y ∈ C ∖ ( E ∪ { 0 , 1 }) . Then κ \kappa κ maps every neighbourhood of z z z in G G G onto a neighbourhood of κ ( z ) \kappa(z) κ ( z ) .
(iii) If σ , τ ∈ ( 0 , 1 ) ∖ Δ \sigma,\tau\in(0,1) \setminus\Delta σ , τ ∈ ( 0 , 1 ) ∖ Δ , then κ − 1 ( σ , τ ) \kappa^{-1}(\sigma,\tau) κ − 1 ( σ , τ ) is a single node, and its coordinates lie in C ∖ E C \setminus\mathcal{E} C ∖ E .
Proof. (i) Every vertical line { x } × [ 0 , 1 ] \{x\} \times[0,1] { x } × [ 0 , 1 ] , x ∈ C x \in C x ∈ C , meets every horizontal line [ 0 , 1 ] × { y } [0,1] \times\{y\} [ 0 , 1 ] × { y } , y ∈ C y \in C y ∈ C , and every point of G G G lies on one of these arcs. Every nonempty open subset of G G G contains a Cantor box, whose components are arcs that are not open in it; so no point has a locally connected open neighbourhood. On the other hand, G G G is locally connected at every node ( x , y ) (x,y) ( x , y ) : for ε > 0 \varepsilon> 0 ε > 0 the set G ∩ ( ( x − ε , x + ε ) × ( y − ε , y + ε ) ) G \cap((x-\varepsilon,x+\varepsilon) \times(y-\varepsilon,y+\varepsilon)) G ∩ (( x − ε , x + ε ) × ( y − ε , y + ε )) is arc connected, since each of its points is joined to ( x , y ) (x,y) ( x , y ) within it by a vertical and a horizontal straight line interval contained in G G G .
(ii) There are x ′ < x < x ′ ′ x' < x < x'' x ′ < x < x ′′ in C C C , arbitrarily close to x x x , with ϑ ( x ′ ) < ϑ ( x ) < ϑ ( x ′ ′ ) \vartheta(x') < \vartheta(x) < \vartheta(x'') ϑ ( x ′ ) < ϑ ( x ) < ϑ ( x ′′ ) , because x ∈ C x \in C x ∈ C is not an end point of a gap; similarly y ′ < y < y ′ ′ y' < y < y'' y ′ < y < y ′′ . If ϑ ( x ′ ) < σ < ϑ ( x ′ ′ ) \vartheta(x') < \sigma< \vartheta(x'') ϑ ( x ′ ) < σ < ϑ ( x ′′ ) and ϑ ( y ′ ) < τ < ϑ ( y ′ ′ ) \vartheta(y') < \tau< \vartheta(y'') ϑ ( y ′ ) < τ < ϑ ( y ′′ ) , choose ξ ∈ C ∩ [ x ′ , x ′ ′ ] \xi\in C \cap[x',x''] ξ ∈ C ∩ [ x ′ , x ′′ ] with ϑ ( ξ ) = σ \vartheta(\xi) = \sigma ϑ ( ξ ) = σ (possible because ϑ \vartheta ϑ is constant on the gaps) and η ∈ [ y ′ , y ′ ′ ] \eta\in[y',y''] η ∈ [ y ′ , y ′′ ] with ϑ ( η ) = τ \vartheta(\eta) = \tau ϑ ( η ) = τ ; then ( ξ , η ) ∈ C × [ 0 , 1 ] ⊆ G (\xi,\eta) \in C \times[0,1] \subseteq G ( ξ , η ) ∈ C × [ 0 , 1 ] ⊆ G and κ ( ξ , η ) = ( σ , τ ) \kappa(\xi,\eta) = (\sigma,\tau) κ ( ξ , η ) = ( σ , τ ) .
(iii) ϑ − 1 ( σ ) \vartheta^{-1}(\sigma) ϑ − 1 ( σ ) is either a point of C ∖ E C \setminus\mathcal{E} C ∖ E or the closure of a gap, and the latter happens exactly when σ ∈ Δ \sigma\in\Delta σ ∈ Δ . □
For the rest of this subsection f ∈ C ( G ) f \in C(G) f ∈ C ( G ) and an f f f -admissible g g g are fixed, Σ \Sigma Σ is the set of points of G G G whose level is collapsed, and Ω = G ∖ cl Σ \Omega= G \setminus\operatorname{cl}\Sigma Ω = G ∖ cl Σ . Thus Ω \Omega Ω is open, and an open set disjoint from Σ \Sigma Σ is contained in Ω \Omega Ω .
Lemma 4.28. f f f is constant on no nonempty open subset of Ω \Omega Ω , and f f f is constant on every connected subset of Ω \Omega Ω that is contained in a segment.
Proof. The proof of Lemma 4.19 applies, with the tooth replaced by a segment, say { x } × [ a , b ] \{x\} \times[a,b] { x } × [ a , b ] , and with τ 1 , τ 2 ∈ ( a , b ) \tau_1,\tau_2 \in(a,b) τ 1 , τ 2 ∈ ( a , b ) . The only change is in the choice of the Cantor box: the crossing point s s s lies in the relative interior of the segment, so it is not a node and has a Cantor box neighbourhood B ⊆ Ω B \subseteq\Omega B ⊆ Ω with vertical arcs. □
Lemma 4.29. Ω \Omega Ω contains no node.
Proof. Suppose it does. Since the nodes with both coordinates in C ∖ ( E ∪ { 0 , 1 } ) C \setminus(\mathcal{E} \cup\{0,1\}) C ∖ ( E ∪ { 0 , 1 }) are dense in C × C C \times C C × C , there are x 1 < x 2 x_1 < x_2 x 1 < x 2 and y 1 < y 2 y_1 < y_2 y 1 < y 2 in C ∖ E C \setminus\mathcal{E} C ∖ E such that S = G ∩ ( [ x 1 , x 2 ] × [ y 1 , y 2 ] ) S = G \cap([x_1,x_2] \times[y_1,y_2]) S = G ∩ ([ x 1 , x 2 ] × [ y 1 , y 2 ]) is contained in Ω \Omega Ω and contains such a node in its interior. Since x i , y i ∉ E x_i,y_i \notin\mathcal{E} x i , y i ∈ / E , a gap meeting ( x 1 , x 2 ) (x_1,x_2) ( x 1 , x 2 ) is contained in it, and similarly for ( y 1 , y 2 ) (y_1,y_2) ( y 1 , y 2 ) . Put Q = [ ϑ ( x 1 ) , ϑ ( x 2 ) ] × [ ϑ ( y 1 ) , ϑ ( y 2 ) ] Q = [\vartheta(x_1),\vartheta(x_2)] \times[\vartheta(y_1),\vartheta(y_2)] Q = [ ϑ ( x 1 ) , ϑ ( x 2 )] × [ ϑ ( y 1 ) , ϑ ( y 2 )] ; then κ ( S ) = Q \kappa(S) = Q κ ( S ) = Q .
Step 1: f = ψ ∘ κ f = \psi\circ\kappa f = ψ ∘ κ on S S S for some ψ ∈ C ( Q ) \psi\in C(Q) ψ ∈ C ( Q ) . A fibre of κ ∣ S \kappa|_S κ ∣ S is a node, a segment, or the union of the four segments bounding a rectangle [ a , b ] × [ a ′ , b ′ ] [a,b] \times[a',b'] [ a , b ] × [ a ′ , b ′ ] with ( a , b ) (a,b) ( a , b ) , ( a ′ , b ′ ) (a',b') ( a ′ , b ′ ) gaps. By Lemma 4.28 and continuity f f f is constant on every segment contained in S S S , hence on every fibre of κ ∣ S \kappa|_S κ ∣ S . Since κ ∣ S \kappa|_S κ ∣ S is a closed surjection onto Q Q Q , ψ \psi ψ is well defined and continuous.
Step 2: ψ \psi ψ is constant on no nonempty open subset of Q Q Q , and on no nondegenerate segment { σ } × J ⊆ Int Q \{\sigma\} \times J \subseteq\operatorname{Int} Q { σ } × J ⊆ Int Q or J × { τ } ⊆ Int Q J \times\{\tau\} \subseteq\operatorname{Int} Q J × { τ } ⊆ Int Q with σ ∈ Δ \sigma\in\Delta σ ∈ Δ , resp. τ ∈ Δ \tau\in\Delta τ ∈ Δ . If ψ \psi ψ is constant on a nonempty open set V ⊆ Q V \subseteq Q V ⊆ Q , we may assume V ⊆ Int Q V \subseteq\operatorname{Int} Q V ⊆ Int Q ; then κ − 1 ( V ) \kappa^{-1}(V) κ − 1 ( V ) is open in G G G , nonempty (since ϑ ( C ) = [ 0 , 1 ] \vartheta(C) = [0,1] ϑ ( C ) = [ 0 , 1 ] ) and contained in S S S (a point of G G G whose image lies in Int Q \operatorname{Int} Q Int Q has x 1 < x < x 2 x_1 < x < x_2 x 1 < x < x 2 and y 1 < y < y 2 y_1 < y < y_2 y 1 < y < y 2 ), and f f f is constant on it, contrary to Lemma 4.28. If ψ \psi ψ is constant on { σ } × J \{\sigma\} \times J { σ } × J with σ = ϑ ( a ) = ϑ ( b ) \sigma= \vartheta(a) = \vartheta(b) σ = ϑ ( a ) = ϑ ( b ) , ( a , b ) (a,b) ( a , b ) a gap in ( x 1 , x 2 ) (x_1,x_2) ( x 1 , x 2 ) , then f f f is constant on ( a , b ) × ( C ∩ ϑ − 1 ( Int J ) ) (a,b) \times(C \cap\vartheta^{-1}(\operatorname{Int} J)) ( a , b ) × ( C ∩ ϑ − 1 ( Int J )) , a nonempty open subset of G G G contained in S ⊆ Ω S \subseteq\Omega S ⊆ Ω , contrary to Lemma 4.28 . The horizontal case is symmetric.
Step 3. Let O ψ O_\psi O ψ be the set of points q ∈ Int Q q \in\operatorname{Int} Q q ∈ Int Q such that ψ \psi ψ maps every neighbourhood of q q q onto a neighbourhood of ψ ( q ) \psi(q) ψ ( q ) , and let R R R be the set of nodes z ∈ S z \in S z ∈ S with coordinates in C ∖ E C \setminus E C ∖ E and κ ( z ) ∈ O ψ \kappa(z) \in O_\psi κ ( z ) ∈ O ψ . We claim:
( ∗ ) (*) ( ∗ ) if V ′ ⊆ Int Q V' \subseteq\operatorname{Int} Q V ′ ⊆ Int Q is an open disc and t t t is an interior point of ψ ( V ′ ) \psi(V') ψ ( V ′ ) , then there is z ∈ R z \in R z ∈ R with κ ( z ) ∈ V ′ \kappa(z) \in V' κ ( z ) ∈ V ′ and f ( z ) = t f(z) = t f ( z ) = t .
Let A = V ′ ∩ { ψ < t } A = V' \cap\{\psi< t\} A = V ′ ∩ { ψ < t } and B = V ′ ∩ { ψ > t } B = V' \cap\{\psi> t\} B = V ′ ∩ { ψ > t } ; both are nonempty and open. Put M = V ′ ∩ cl A ∩ cl B M = V' \cap\operatorname{cl} A \cap\operatorname{cl} B M = V ′ ∩ cl A ∩ cl B . A point of V ′ ∖ ( cl A ∪ cl B ) V' \setminus(\operatorname{cl} A \cup\operatorname{cl} B) V ′ ∖ ( cl A ∪ cl B ) would have a neighbourhood on which ψ ≡ t \psi\equiv t ψ ≡ t , which is excluded by Step 2. Hence V ′ ∖ M V' \setminus M V ′ ∖ M is the union of the disjoint open sets V ′ ∖ cl B ⊇ A V' \setminus\operatorname{cl} B \supseteq A V ′ ∖ cl B ⊇ A and V ′ ∖ cl A ⊇ B V' \setminus\operatorname{cl} A \supseteq B V ′ ∖ cl A ⊇ B , so the relatively closed set M M M separates V ′ V' V ′ between a point of A A A and a point of B B B . Since V ′ V' V ′ is homeomorphic to R 2 \mathbb{R}^2 R 2 , it is connected, locally connected and contractible with respect to the circle [26 §57, I, Theorem 9(i)] ; hence M M M contains a set K 0 K_0 K 0 , closed in V ′ V' V ′ and connected, which separates these points [26 §57, III, Theorem 1] . In particular K 0 K_0 K 0 is not a point. Let p ∈ K 0 p \in K_0 p ∈ K 0 and let D ⊆ V ′ D \subseteq V' D ⊆ V ′ be a closed disc about p p p with K 0 ⊈ D K_0 \not\subseteq D K 0 ⊆ D . Then X = cl K 0 X = \operatorname{cl} K_0 X = cl K 0 (closure in R 2 \mathbb{R}^2 R 2 ) is a continuum and X ∩ D = K 0 ∩ D X \cap D = K_0 \cap D X ∩ D = K 0 ∩ D is a nonempty proper closed subset of X X X . By the boundary bumping theorem [34 Theorem 5.7] the component K 1 K_1 K 1 of K 0 ∩ D K_0 \cap D K 0 ∩ D containing p p p meets cl ( X ∖ D ) ⊆ X ∖ Int D \operatorname{cl}(X \setminus D) \subseteq X \setminus\operatorname{Int} D cl ( X ∖ D ) ⊆ X ∖ Int D ; so K 1 K_1 K 1 is a nondegenerate continuum contained in K 0 ∩ D K_0 \cap D K 0 ∩ D .
Let Λ \Lambda Λ be the union of the lines { σ } × R \{\sigma\} \times\mathbb{R} { σ } × R and R × { τ } \mathbb{R} \times\{\tau\} R × { τ } , σ , τ ∈ Δ \sigma,\tau\in\Delta σ , τ ∈ Δ . If K 1 ⊆ Λ K_1 \subseteq\Lambda K 1 ⊆ Λ , then by Baire’s theorem one of the countably many closed sets K 1 ∩ ℓ K_1 \cap\ell K 1 ∩ ℓ , ℓ \ell ℓ such a line, has nonempty interior U U U in K 1 K_1 K 1 . If cl U = K 1 \operatorname{cl} U = K_1 cl U = K 1 , then K 1 ⊆ ℓ K_1 \subseteq\ell K 1 ⊆ ℓ ; otherwise, by [34 Theorem 5.7] , the component of cl U \operatorname{cl} U cl U containing a point of U U U meets cl ( K 1 ∖ cl U ) \operatorname{cl}(K_1 \setminus\operatorname{cl} U) cl ( K 1 ∖ cl U ) , so it is a nondegenerate continuum contained in cl U ⊆ ℓ \operatorname{cl} U \subseteq\ell cl U ⊆ ℓ . Either way M ⊆ { ψ = t } M \subseteq\{\psi= t\} M ⊆ { ψ = t } contains a nondegenerate segment of a line in Λ \Lambda Λ , contrary to Step 2. Hence there is q ′ ∈ K 1 q' \in K_1 q ′ ∈ K 1 with both coordinates outside Δ \Delta Δ . Every disc about q ′ q' q ′ is connected and meets A A A and B B B , so q ′ ∈ O ψ q' \in O_\psi q ′ ∈ O ψ . By Lemma 4.27 (iii) κ − 1 ( q ′ ) \kappa^{-1}(q') κ − 1 ( q ′ ) is a single node z z z with coordinates in C ∖ E C \setminus E C ∖ E ; it lies in S S S because q ′ ∈ Int Q q' \in\operatorname{Int} Q q ′ ∈ Int Q . Thus z ∈ R z \in R z ∈ R , κ ( z ) = q ′ ∈ V ′ \kappa(z) = q' \in V' κ ( z ) = q ′ ∈ V ′ and f ( z ) = ψ ( q ′ ) = t f(z) = \psi(q') = t f ( z ) = ψ ( q ′ ) = t , which proves ( ∗ ) (*) ( ∗ ) .
By Step 2 and ( ∗ ) (*) ( ∗ ) , R ≠ ∅ R \ne\varnothing R = ∅ . Moreover f ∣ R f|_R f ∣ R is open: let s ∈ R s \in R s ∈ R and let V V V be a neighbourhood of s s s in G G G . The coordinates of s s s are different from 0 0 0 and 1 1 1 because κ ( s ) ∈ Int Q \kappa(s) \in\operatorname{Int} Q κ ( s ) ∈ Int Q , so by Lemma 4.27 (ii) there is an open disc V ′ ⊆ κ ( V ) ∩ Int Q V' \subseteq\kappa(V) \cap\operatorname{Int} Q V ′ ⊆ κ ( V ) ∩ Int Q about κ ( s ) \kappa(s) κ ( s ) , and since κ ( s ) ∈ O ψ \kappa(s) \in O_\psi κ ( s ) ∈ O ψ , ψ ( V ′ ) \psi(V') ψ ( V ′ ) contains an interval ( f ( s ) − δ , f ( s ) + δ ) (f(s) - \delta, f(s) + \delta) ( f ( s ) − δ , f ( s ) + δ ) . For t t t in this interval ( ∗ ) (*) ( ∗ ) gives z ∈ R z \in R z ∈ R with f ( z ) = t f(z) = t f ( z ) = t and κ ( z ) ∈ V ′ ⊆ κ ( V ) \kappa(z) \in V' \subseteq\kappa(V) κ ( z ) ∈ V ′ ⊆ κ ( V ) ; since z z z is the only point of G G G mapped to κ ( z ) \kappa(z) κ ( z ) , z ∈ V z \in V z ∈ V . By Lemma 3.3 every level in f ( R ) f(R) f ( R ) is collapsed, i.e. R ⊆ Σ R \subseteq\Sigma R ⊆ Σ . But R ⊆ S ⊆ Ω R \subseteq S \subseteq\Omega R ⊆ S ⊆ Ω , a contradiction. □ \square □
Lemma 4.30. Let c ∈ f ( G ) c \in f(G) c ∈ f ( G ) . Every u ∈ Z c + u \in Z_c^+ u ∈ Z c + admits a sequence y n y_n y n as in condition ( a ) (a) ( a ) of Theorem 4.2 . The same holds for Z c − Z_c^- Z c − .
Proof. If u ∈ cl ( Σ ∩ { f > c } ) u \in\operatorname{cl}(\Sigma\cap\{f > c\}) u ∈ cl ( Σ ∩ { f > c }) , take y n ∈ Σ ∩ { f > c } y_n \in\Sigma\cap\{f > c\} y n ∈ Σ ∩ { f > c } with y n → u y_n \to u y n → u . Otherwise there is an open neighbourhood U U U of u u u with U ∩ { f > c } ⊆ Ω U \cap\{f > c\} \subseteq\Omega U ∩ { f > c } ⊆ Ω .
u u u is not a node. Otherwise let U ′ ⊆ U U' \subseteq U U ′ ⊆ U be the trace on G G G of an open square ( x 0 − ε , x 0 + ε ) × ( y 0 − ε , y 0 + ε ) (x_0 - \varepsilon, x_0 + \varepsilon) \times(y_0 - \varepsilon, y_0 + \varepsilon) ( x 0 − ε , x 0 + ε ) × ( y 0 − ε , y 0 + ε ) about u = ( x 0 , y 0 ) u = (x_0,y_0) u = ( x 0 , y 0 ) , and W = U ′ ∩ { f > c } W = U' \cap\{f > c\} W = U ′ ∩ { f > c } , a nonempty open subset of Ω \Omega Ω . By Lemma 4.29 there is no node, so a point p ∈ W p \in W p ∈ W lies, say, in the interior of a vertical segment { x } × [ a , b ] \{x\} \times[a,b] { x } × [ a , b ] . Let I I I be the component of W ∩ ( { x } × ( a , b ) ) W \cap(\{x\} \times(a,b)) W ∩ ({ x } × ( a , b )) containing p p p . By Lemma 4.28 f = f ( p ) > c f = f(p) > c f = f ( p ) > c on cl I \operatorname{cl} I cl I . An end point of I I I lying in U ′ U' U ′ would therefore belong to W W W , which is impossible if it lies in { x } × ( a , b ) \{x\} \times(a,b) { x } × ( a , b ) (maximality of I I I ) or is a node (Lemma 4.29 ). Hence I = { x } × ( y 0 − ε , y 0 + ε ) I = \{x\} \times(y_0 - \varepsilon,y_0 + \varepsilon) I = { x } × ( y 0 − ε , y 0 + ε ) , so y 0 ∈ ( a , b ) y_0 \in(a,b) y 0 ∈ ( a , b ) , contradicting y 0 ∈ C y_0 \in C y 0 ∈ C . For a horizontal segment we obtain x 0 ∈ ( a , b ) x_0 \in(a,b) x 0 ∈ ( a , b ) in the same way.
u u u lies in a segment. Say u = ( x u , τ u ) u = (x_u,\tau_u) u = ( x u , τ u ) with x u ∈ C x_u \in C x u ∈ C and τ u \tau_u τ u in a gap ( a , b ) (a,b) ( a , b ) . The proof of the case u = h ( x u , τ u ) u = h(x_u,\tau_u) u = h ( x u , τ u ) of Lemma 4.21 applies, with Lemma 4.28 in place of Lemma 4.19 , the tooth T x T_x T x replaced by the segment { x } × [ a , b ] \{x\} \times[a,b] { x } × [ a , b ] (and h ( x , σ ) h(x,\sigma) h ( x , σ ) by ( x , σ ) (x,\sigma) ( x , σ ) ), and the following changes: the Cantor box A u × ( τ u − ε , τ u + ε ) A_u \times(\tau_u-\varepsilon,\tau_u+\varepsilon) A u × ( τ u − ε , τ u + ε ) is chosen with [ τ u − ε , τ u + ε ] ⊆ ( a , b ) [\tau_u-\varepsilon,\tau_u+\varepsilon] \subseteq(a,b) [ τ u − ε , τ u + ε ] ⊆ ( a , b ) ; in the definition of β \beta β the interval [ 0 , τ u ] [0,\tau_u] [ 0 , τ u ] is replaced by [ a , τ u ] [a,\tau_u] [ a , τ u ] ; and if β ( x ) = a \beta(x)=a β ( x ) = a , then e x e_x e x is a node, so e x ∉ Ω e_x \notin\Omega e x ∈ / Ω by Lemma 4.29 . Points of horizontal segments are treated in the same way, and Z c − Z_c^{-} Z c − is symmetric. □
Theorem 4.31. The Cantor grid G G G is in D C P DCP D C P , although L ( G ) = ∅ L(G)=\varnothing L ( G ) = ∅ .
Proof. Let f ∈ C ( G ) f \in C(G) f ∈ C ( G ) , let g g g be f f f -admissible and c ∈ f ( G ) c \in f(G) c ∈ f ( G ) . If Int Z c ≠ ∅ \operatorname{Int} Z_c \ne\varnothing Int Z c = ∅ , the level c c c is collapsed by Lemma 3.1 . Otherwise condition (a) of Theorem 4.2 holds by Lemma 4.30 , and (b) holds because G G G is arc connected (see the remark preceding Theorem 4.4 ). Hence every level is collapsed. □
Necessary conditions and counterexamples.
Proposition 4.33 (cf. [23 Theorem 1.15] ). Let K K K be a compact metrizable space and let u , v ∈ K u,v \in K u , v ∈ K be non-isolated points lying in different components of K K K . Then D ( K ) ≠ C ( K ) D(K) \ne C(K) D ( K ) = C ( K ) . In particular a compact metrizable space without isolated points is in D C P DCP D C P only if it is connected.
Proof. Components of a compact space coincide with quasi-components, so there is a clopen W W W with u ∈ W u \in W u ∈ W , v ∉ W v \notin W v ∈ / W . Let f = d ( ⋅ , u ) f=d(\mathord{\cdot},u) f = d ( ⋅ , u ) on W W W and f = − d ( ⋅ , v ) f=-d(\mathord{\cdot},v) f = − d ( ⋅ , v ) on K ∖ W K \setminus W K ∖ W . Every fibre other than Z 0 = { u , v } Z_0=\{u,v\} Z 0 = { u , v } lies in W W W or in K ∖ W K \setminus W K ∖ W . Since u u u is the limit of points k n ∈ W ∖ { u } k_n \in W \setminus\{u\} k n ∈ W ∖ { u } , and M f ( k n ) ⊆ W M_f(k_n) \subseteq W M f ( k n ) ⊆ W , lower semicontinuity gives M f ( u ) ⊆ W M_f(u) \subseteq W M f ( u ) ⊆ W ; similarly M f ( v ) ⊆ K ∖ W M_f(v) \subseteq K \setminus W M f ( v ) ⊆ K ∖ W . So g = χ W g=\chi_W g = χ W is constant on every M f ( k ) M_f(k) M f ( k ) but not on Z 0 Z_0 Z 0 , so f ∉ D ( K ) f \notin D(K) f ∈ / D ( K ) . □
Example 4.35 ([23 Example 1.16] ). Let S S S be the closure of { ( x , sin ( 1 / x ) ) : 0 < x ≤ 1 } \{(x,\sin(1/x)) : 0 < x \le1\} {( x , sin ( 1/ x )) : 0 < x ≤ 1 } and f ( x , y ) = x f(x,y)=x f ( x , y ) = x . All fibres except E = { 0 } × [ − 1 , 1 ] E=\{0\}\times[-1,1] E = { 0 } × [ − 1 , 1 ] are singletons. If k ∈ E k \in E k ∈ E and k n → k k_n \to k k n → k are points of the curve, then M f ( k n ) = { k n } M_f(k_n)=\{k_n\} M f ( k n ) = { k n } , so M f ( k ) = { k } M_f(k)=\{k\} M f ( k ) = { k } . Hence every g ∈ C ( S ) g \in C(S) g ∈ C ( S ) is f f f -admissible, and f ∉ D ( S ) f \notin D(S) f ∈ / D ( S ) .
The same argument applies to every compactification of a ray, and of two disjoint rays each of which is open in it.
Proposition 4.36. Let K K K be a continuum containing a dense subset U U U which is the disjoint union of one or two open subsets R 1 , R 2 R_1,R_2 R 1 , R 2 of K K K , each homeomorphic to [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) , and suppose that A = K ∖ U A=K\setminus U A = K ∖ U is not a single point. Then K ∉ D C P K \notin DCP K ∈ / D C P .
Proof. Let r i : [ 0 , ∞ ) → R i r_i:[0,\infty)\to R_i r i : [ 0 , ∞ ) → R i be homeomorphisms and define f = 0 f=0 f = 0 on A A A , f ( r 1 ( t ) ) = 1 / ( 1 + t ) f(r_1(t))=1/(1+t) f ( r 1 ( t )) = 1/ ( 1 + t ) and f ( r 2 ( t ) ) = − 1 / ( 1 + t ) f(r_2(t))=-1/(1+t) f ( r 2 ( t )) = − 1/ ( 1 + t ) . Since R 1 R_1 R 1 and R 2 R_2 R 2 are open, f f f is continuous on U U U ; and a point a ∈ A a \in A a ∈ A has, for every T > 0 T>0 T > 0 , a neighbourhood disjoint from the compact set r 1 ( [ 0 , T ] ) ∪ r 2 ( [ 0 , T ] ) r_1([0,T])\cup r_2([0,T]) r 1 ([ 0 , T ]) ∪ r 2 ([ 0 , T ]) , on which ∣ f ∣ < 1 / ( 1 + T ) |f|<1/(1+T) ∣ f ∣ < 1/ ( 1 + T ) . So f f f is continuous, and its fibres are the points of U U U and the set A A A . Let Φ ∈ M f \Phi\in\mathfrak{M}_f Φ ∈ M f ; then Φ ( k ) = { k } \Phi(k)=\{k\} Φ ( k ) = { k } for k ∈ U k \in U k ∈ U . If a ∈ A a \in A a ∈ A and b ∈ Φ ( a ) b \in\Phi(a) b ∈ Φ ( a ) , b ≠ a b \ne a b = a , take disjoint open sets V ∋ b V \ni b V ∋ b and V ′ ∋ a V' \ni a V ′ ∋ a ; by lower semicontinuity some neighbourhood N ⊆ V ′ N \subseteq V' N ⊆ V ′ of a a a satisfies Φ ( k ) ∩ V ≠ ∅ \Phi(k)\cap V \ne\varnothing Φ ( k ) ∩ V = ∅ for k ∈ N k \in N k ∈ N , which fails for k ∈ N ∩ U k \in N \cap U k ∈ N ∩ U . Hence M f ( k ) = { k } M_f(k)=\{k\} M f ( k ) = { k } for all k k k , every g ∈ C ( K ) g \in C(K) g ∈ C ( K ) is f f f -admissible, and a g g g that is not constant on A A A shows f ∉ D ( K ) f \notin D(K) f ∈ / D ( K ) . □
Remark 4.37. Proposition 4.36 applies to the sin ( 1 / x ) \sin(1/x) sin ( 1/ x ) continuum, to a ray spiralling onto a circle, to a simple triod with a ray spiralling around it [21 Fig. 10.13] , and to Nadler’s spiral continua [21 Fig. 10.12] . It does not apply to a line both of whose ends accumulate on the same set A A A (no continuous f f f is then injective on the line), e.g. two sin ( 1 / x ) \sin(1/x) sin ( 1/ x ) curves approaching one segment from both sides and joined to each other by an arc; we do not know whether such continua are in D C P \mathcal{DCP} D C P .
Example 4.38. Let K = S ∪ A K=S\cup A K = S ∪ A , where S S S is as in Example 4.35 and A = [ − 1 , 0 ] × { − 1 } A=[-1,0]\times\{-1\} A = [ − 1 , 0 ] × { − 1 } , and let f ( x , y ) = x f(x,y)=x f ( x , y ) = x . The set L ( K ) L(K) L ( K ) , which contains the curve and A A A without the point ( 0 , − 1 ) (0,-1) ( 0 , − 1 ) , is dense, and every point of K ∖ L ( K ) = E K\setminus L(K)=E K ∖ L ( K ) = E can be joined by an arc (along E E E and A A A ) to a point of L ( K ) L(K) L ( K ) . Nevertheless the fibres of f f f other than E E E are singletons, and as in Example 4.35 M f ( k ) = { k } M_f(k)=\{k\} M f ( k ) = { k } for all k k k , so K ∉ D C P K\notin\mathcal{DCP} K ∈ / D C P . Thus in Theorem 4.4 arc connectedness cannot be replaced by the condition that every point of K K K can be joined by an arc to L ( K ) L(K) L ( K ) (condition (6) of [24 Proposition 7] without the other conditions).
In Proposition 4.10 the size of the square matters.
Example 4.39. With C 2 C_2 C 2 and E E E as in Proposition 4.10, let Q ′ = [ − 1 , 0 ] × [ 0 , 1 ] Q'=[-1,0]\times[0,1] Q ′ = [ − 1 , 0 ] × [ 0 , 1 ] , let S S S be the closure of C 2 C_2 C 2 (so S = C 2 ∪ E S=C_2\cup E S = C 2 ∪ E ), and K = Q ′ ∪ S K=Q'\cup S K = Q ′ ∪ S . The arc components are C 2 C_2 C 2 and Q ′ ∪ E Q'\cup E Q ′ ∪ E , and Q ′ ∪ E ⊈ cl Int ( Q ′ ∪ E ) = Q ′ Q'\cup E\not\subseteq\operatorname{cl}\operatorname{Int}(Q'\cup E)=Q' Q ′ ∪ E ⊆ cl Int ( Q ′ ∪ E ) = Q ′ , while L ( K ) ⊇ Int Q ′ ∪ C 2 L(K)\supseteq\operatorname{Int}Q'\cup C_2 L ( K ) ⊇ Int Q ′ ∪ C 2 is dense. Let f ( x , y ) = x f(x,y)=x f ( x , y ) = x . For 0 < x ≤ 1 0<x\leq1 0 < x ≤ 1 the fibre Z x Z_x Z x is a single point of C 2 C_2 C 2 ; since every point of E E E is a limit of such points, M f ( k ) = { k } M_f(k)=\{k\} M f ( k ) = { k } for k ∈ E k\in E k ∈ E . For k ∈ Q ′ ∖ E k\in Q'\setminus E k ∈ Q ′ ∖ E , M f ( k ) ⊆ { x } × [ 0 , 1 ] M_f(k)\subseteq\{x\}\times[0,1] M f ( k ) ⊆ { x } × [ 0 , 1 ] . The function g = max ( 0 , − y ) g=\max(0,-y) g = max ( 0 , − y ) on S S S , g = 0 g=0 g = 0 on Q ′ Q' Q ′ , is continuous (both definitions give 0 0 0 on Q ′ ∩ S = { 0 } × [ 0 , 1 ] Q'\cap S=\{0\}\times[0,1] Q ′ ∩ S = { 0 } × [ 0 , 1 ] ) and constant on every M f ( k ) M_f(k) M f ( k ) , but g ( 0 , − 1 ) = 1 ≠ 0 = g ( 0 , 0 ) g(0,-1)=1\ne0=g(0,0) g ( 0 , − 1 ) = 1 = 0 = g ( 0 , 0 ) although both points lie in Z 0 = E Z_0=E Z 0 = E . Thus f ∉ D ( K ) f\notin D(K) f ∈ / D ( K ) and K ∉ D C P K\notin\mathcal{DCP} K ∈ / D C P .
Individual functions in D ( K ) D(K) D ( K )
For continua such as solenoids and the buckethandle continuum we do not know whether they are in D C P \mathcal{DCP} D C P , but we can show that many individual functions belong to D ( K ) D(K) D ( K ) . The main result, Theorem 5.1, covers every function that is constant on no arc, on a continuum all of whose arc components are dense. Retractions onto locally connected continua give further functions, with no assumption on L ( K ) L(K) L ( K ) (Corollary 3.13); we apply this to radial functions on smooth dendroids and on cones. We then collect what these results give for solenoids, the buckethandle continuum and fans with the property of Kelley.
Theorem 5.1. Let K K K be a metrizable continuum all of whose arc components are dense, and let f ∈ C ( K ) f\in C(K) f ∈ C ( K ) be constant on no arc. Then f ∈ D ( K ) f\in D(K) f ∈ D ( K ) .
Proof. We may assume that K K K is not a point. Every nonempty open set U U U meets a dense, nondegenerate arc component, hence contains an arc (an initial piece of an arc leaving a point of U U U ); so all fibres have empty interior. Let min f < t < max f \min f<t<\max f min f < t < max f . A dense arc component contains points with f < t f<t f < t and points with f > t f>t f > t ; let γ \gamma γ be an arc joining two such points. Since f f f is not constant on subarcs of γ \gamma γ , the relative closures of γ ∩ { f < t } \gamma\cap\{f<t\} γ ∩ { f < t } and γ ∩ { f > t } \gamma\cap\{f>t\} γ ∩ { f > t } cover γ \gamma γ and meet at a point of the relative interior of γ \gamma γ , which is a crossing point at level t t t ; by Lemma 3.8 every t ∈ ( min f , max f ) t\in(\min f,\max f) t ∈ ( min f , max f ) is collapsed. If u ∈ Z c + u\in Z_c^+ u ∈ Z c + then c < max f c<\max f c < max f , and u u u is the limit of points u n u_n u n with c < f ( u n ) < max f c<f(u_n)<\max f c < f ( u n ) < max f ; similarly for Z c − Z_c^- Z c − . Thus (a) of Theorem 4.2 holds, (b) holds by Lemma 4.6, and Theorem 4.2 applies to every level. □ \square □
Remark 5.2. Theorem 5.1 contains [23 Theorem 1.19(2)] (arc connected K K K and totally disconnected fibres), whose proof was omitted there. Besides arc connected continua, it applies to solenoids and the buckethandle continuum; see below.
Retractions. Recall from § 1 the notions of a dendroid and of a smooth dendroid. Charatonik and Eberhart [12 Theorem 10] proved that K K K is smooth with initial point p p p if and only if K K K admits a metric d d d that is radially convex with respect to p p p : x ∈ p y x \in py x ∈ p y , x ≠ y x \ne y x = y imply d ( p , x ) < d ( p , y ) d(p,x) < d(p,y) d ( p , x ) < d ( p , y ) .
Proposition 5.3. Let K K K be a smooth dendroid with initial point p p p , d d d a metric on K K K that is radially convex with respect to p p p , and ρ = d ( p , ⋅ ) \rho=d(p,\cdot) ρ = d ( p , ⋅ ) . Then φ ∘ ρ ∈ D ( K ) \varphi\circ\rho\in\mathcal{D}(K) φ ∘ ρ ∈ D ( K ) for every φ ∈ C ( ρ ( K ) ) \varphi\in C(\rho(K)) φ ∈ C ( ρ ( K )) .
Proof. Let R = max ρ R=\max\rho R = max ρ and choose e e e with ρ ( e ) = R \rho(e)=R ρ ( e ) = R . By radial convexity ρ \rho ρ is strictly increasing along the arc p e pe p e , so it maps p e pe p e homeomorphically onto [ 0 , R ] = ρ ( K ) [0,R]=\rho(K) [ 0 , R ] = ρ ( K ) ; let γ : [ 0 , R ] → p e \gamma:[0,R]\to pe γ : [ 0 , R ] → p e be the inverse. Then r = γ ∘ ρ r=\gamma\circ\rho r = γ ∘ ρ is a retraction of K K K onto the arc p e pe p e (cf. [28 Theorem 1] ) and ρ ∘ r = ρ \rho\circ r=\rho ρ ∘ r = ρ , hence f ∘ r = f f\circ r=f f ∘ r = f for f = φ ∘ ρ f=\varphi\circ\rho f = φ ∘ ρ . Apply Corollary 3.13. □
Note that φ \varphi φ is arbitrary; it may be constant on intervals or behave like the Cantor function, which is where the lemmas of § 3 are most difficult to apply. (For a cone Cone ( Y ) \operatorname{Cone}(Y) Cone ( Y ) the same argument, with the retraction ( y , t ) ↦ ( y 0 , t ) (y,t)\mapsto(y_0,t) ( y , t ) ↦ ( y 0 , t ) onto a single tooth, gives every function of t t t alone; but by Theorem 4.22 every cone is in D C P \mathcal{DCP} D C P anyway.)
Retractions that approximate the identity give a dense set of functions in D ( K ) \mathcal{D}(K) D ( K ) .
Proposition 5.4. Let K K K be a compact metrizable space, and let r n r_n r n be retractions of K K K onto closed subsets L n ∈ D C P L_n\in\mathcal{DCP} L n ∈ D C P such that r n → id K r_n\to\operatorname{id}_K r n → id K uniformly. Then φ ∘ r n ∈ D ( K ) \varphi\circ r_n\in\mathcal{D}(K) φ ∘ r n ∈ D ( K ) for every φ ∈ C ( K ) \varphi\in C(K) φ ∈ C ( K ) and every n n n ; in particular, D ( K ) \mathcal{D}(K) D ( K ) is dense in C ( K ) C(K) C ( K ) .
Proof. For φ ∈ C ( K ) \varphi\in C(K) φ ∈ C ( K ) put f = φ ∘ r n f=\varphi\circ r_n f = φ ∘ r n . Then f ∘ r n = f f\circ r_n=f f ∘ r n = f and L n ∈ D C P L_n\in\mathcal{DCP} L n ∈ D C P , so f ∈ D ( K ) f\in\mathcal{D}(K) f ∈ D ( K ) by Corollary 3.13. Since φ \varphi φ is uniformly continuous, φ ∘ r n → φ \varphi\circ r_n\to\varphi φ ∘ r n → φ uniformly. □
Corollary 5.5. Let K = lim ( X n , f n ) K=\lim(X_n,f_n) K = lim ( X n , f n ) , where X 1 ⊆ X 2 ⊆ ⋯ X_1\subseteq X_2\subseteq\cdots X 1 ⊆ X 2 ⊆ ⋯ are compact metrizable spaces in D C P \mathcal{DCP} D C P and each f n : X n + 1 → X n f_n:X_{n+1}\to X_n f n : X n + 1 → X n is a retraction. Then D ( K ) \mathcal{D}(K) D ( K ) is dense in C ( K ) C(K) C ( K ) .
Proof. Let d m d_m d m be a metric on X m X_m X m bounded by 1 1 1 and d ( x , y ) = ∑ m 2 − m d m ( x m , y m ) d(x,y)=\sum_m 2^{-m}d_m(x_m,y_m) d ( x , y ) = ∑ m 2 − m d m ( x m , y m ) on K K K . For x ∈ X n x\in X_n x ∈ X n put e n ( x ) m = f m ∘ ⋯ ∘ f n − 1 ( x ) e_n(x)_m=f_m\circ\cdots\circ f_{n-1}(x) e n ( x ) m = f m ∘ ⋯ ∘ f n − 1 ( x ) for m < n m<n m < n and e n ( x ) m = x e_n(x)_m=x e n ( x ) m = x for m ≥ n m\ge n m ≥ n ; this is a thread because f m f_m f m is the identity on X m X_m X m . Thus e n : X n → K e_n:X_n\to K e n : X n → K is an embedding with π n ∘ e n = id \pi_n\circ e_n=\operatorname{id} π n ∘ e n = id , and r n = e n ∘ π n r_n=e_n\circ\pi_n r n = e n ∘ π n is a retraction of K K K onto e n ( X n ) ≅ X n e_n(X_n)\cong X_n e n ( X n ) ≅ X n . Since r n ( y ) r_n(y) r n ( y ) and y y y have the same first n n n coordinates, d ( r n ( y ) , y ) ≤ 2 − n d(r_n(y),y)\le2^{-n} d ( r n ( y ) , y ) ≤ 2 − n . Apply Proposition 5.4 . □
Such inverse limits occur, for instance, in the construction of smooth dendroids in [36 proof of Theorem 9.2] .
By a theorem of Fugate [20 Theorem 2] , for every smooth dendroid X X X and every ε > 0 \varepsilon>0 ε > 0 there are a tree (a finite dendrite) T ⊆ X T\subseteq X T ⊆ X and a retraction r : X → T r:X\to T r : X → T that moves no point of X X X as much as ε \varepsilon ε . Trees are locally connected, hence in D C P \mathcal{DCP} D C P (Corollary 4.8 ), so Proposition 5.4 gives the following.
Corollary 5.6. Let X X X be a smooth dendroid. Then D ( X ) \mathcal{D}(X) D ( X ) is dense in C ( X ) C(X) C ( X ) .
For fans smoothness is not needed: by [19 Theorem 1] , for every fan X X X and every ε > 0 \varepsilon>0 ε > 0 there are a finite fan Y ⊆ X Y\subseteq X Y ⊆ X and a retraction r : X → Y r:X\to Y r : X → Y with d ( x , r ( x ) ) < ε d(x,r(x))<\varepsilon d ( x , r ( x )) < ε for all x ∈ X x\in X x ∈ X . Finite fans are locally connected, so Proposition 5.4 again applies.
Corollary 5.7. Let X X X be a fan. Then D ( X ) \mathcal{D}(X) D ( X ) is dense in C ( X ) C(X) C ( X ) .
For smooth fans the retractions can be written down directly, and they can be chosen compatibly (Lemma 5.10 below). A fan is smooth if and only if it embeds in the Cantor fan [6 Theorem 9] , [15 Corollary 4] (cf. the proof of [29 Theorem 6.9.3] ).
Corollary 5.9. Let X X X be a smooth fan. Then there are retractions r n r_n r n of X X X onto finite unions of legs of X X X such that r n → id X r_n \to\operatorname{id}_X r n → id X uniformly, and φ ∘ r n ∈ D ( X ) \varphi\circ r_n \in\mathcal{D}(X) φ ∘ r n ∈ D ( X ) for every φ ∈ C ( X ) \varphi\in C(X) φ ∈ C ( X ) and every n n n . In particular, D ( X ) \mathcal{D}(X) D ( X ) is dense in C ( X ) C(X) C ( X ) .
Proof. Regard X X X as a subcontinuum of F = Cone ( C ) F = \operatorname{Cone}(C) F = Cone ( C ) and write [ x , t ] [x,t] [ x , t ] for the image of ( x , t ) ∈ C × [ 0 , 1 ] (x,t) \in C \times[0,1] ( x , t ) ∈ C × [ 0 , 1 ] . Since X X X is not an arc, its ramification point is the vertex v v v of F F F , the only ramification point of F F F ; hence for every y ∈ X y \in X y ∈ X the arc v y vy v y is an initial segment of a leg of F F F . Thus X = { [ x , t ] : t ≤ ℓ ( x ) } X = \{[x,t] : t \le\ell(x)\} X = {[ x , t ] : t ≤ ℓ ( x )} , where ℓ ( x ) = max { t : [ x , t ] ∈ X } \ell(x) = \max\{t : [x,t] \in X\} ℓ ( x ) = max { t : [ x , t ] ∈ X } ; the function ℓ : C → [ 0 , 1 ] \ell: C \to[0,1] ℓ : C → [ 0 , 1 ] is upper semicontinuous because X X X is closed.
Let P n \mathcal{P}_n P n be a partition of C C C into finitely many clopen sets of diameter less than 1 / n 1/n 1/ n . In each P ∈ P n P \in\mathcal{P}_n P ∈ P n choose x P x_P x P with ℓ ( x P ) = max P ℓ \ell(x_P) = \max_P \ell ℓ ( x P ) = max P ℓ (the maximum exists since ℓ \ell ℓ is upper semicontinuous and P P P is compact), and define r n [ x , t ] = [ x P , t ] r_n[x,t] = [x_P,t] r n [ x , t ] = [ x P , t ] for x ∈ P x \in P x ∈ P . This is well defined at v v v , continuous because x ↦ x P x \mapsto x_P x ↦ x P is locally constant, and maps X X X into X X X because t ≤ ℓ ( x ) ≤ ℓ ( x P ) t \le\ell(x) \le\ell(x_P) t ≤ ℓ ( x ) ≤ ℓ ( x P ) . It is the identity on the legs { [ x P , t ] : t ≤ ℓ ( x P ) } \{[x_P,t] : t \le\ell(x_P)\} {[ x P , t ] : t ≤ ℓ ( x P )} , P ∈ P n P \in\mathcal{P}_n P ∈ P n , whose union L n L_n L n is a finite union of arcs with a common end point, hence a locally connected continuum, so L n ∈ D C P L_n \in\mathcal{DCP} L n ∈ D C P by Corollary 4.8 . By uniform continuity of the quotient map C × [ 0 , 1 ] → F C \times[0,1] \to F C × [ 0 , 1 ] → F , r n → id X r_n \to\operatorname{id}_X r n → id X uniformly. Apply Proposition 5.4 . □
The retractions of Corollary 5.9 can be chosen compatibly, which represents every smooth fan as an inverse limit of the kind considered in Corollary 5.5 .
Lemma 5.10. Let X X X be a smooth fan. Then X X X is homeomorphic to lim ( L n , s n ) \lim(L_n,s_n) lim ( L n , s n ) , where each L n L_n L n is a finite union of arcs with a common end point, L n ⊆ L n + 1 L_n \subseteq L_{n+1} L n ⊆ L n + 1 , and s n : L n + 1 → L n s_n : L_{n+1} \to L_n s n : L n + 1 → L n is a retraction.
Proof. We use the notation of the proof of Corollary 5.9, choosing the partitions nested: every member of P n + 1 \mathcal{P}_{n+1} P n + 1 is contained in a member of P n \mathcal{P}_n P n . We choose the points x P x_P x P inductively. If P ∈ P n P \in\mathcal{P}_n P ∈ P n and P ′ ∈ P n + 1 P' \in\mathcal{P}_{n+1} P ′ ∈ P n + 1 is the member with x P ∈ P ′ ⊆ P x_P \in P' \subseteq P x P ∈ P ′ ⊆ P , put x P ′ = x P x_{P'} = x_P x P ′ = x P ; this is admissible since ℓ ( x P ) = max P ℓ ≥ max P ′ ℓ \ell(x_P) = \max_P \ell\geq\max_{P'} \ell ℓ ( x P ) = max P ℓ ≥ max P ′ ℓ . For the other members of P n + 1 \mathcal{P}_{n+1} P n + 1 choose x P ′ x_{P'} x P ′ as before.
Then L n ⊆ L n + 1 L_n \subseteq L_{n+1} L n ⊆ L n + 1 . Moreover r n ∘ r n + 1 = r n r_n \circ r_{n+1} = r_n r n ∘ r n + 1 = r n : if x ∈ P ′ ⊆ P x \in P' \subseteq P x ∈ P ′ ⊆ P with P ′ ∈ P n + 1 P' \in\mathcal{P}_{n+1} P ′ ∈ P n + 1 , P ∈ P n P \in\mathcal{P}_n P ∈ P n , then x P ′ ∈ P x_{P'} \in P x P ′ ∈ P , so r n ( r n + 1 [ x , t ] ) = r n [ x P ′ , t ] = [ x P , t ] = r n [ x , t ] r_n(r_{n+1}[x,t]) = r_n[x_{P'},t] = [x_P,t] = r_n[x,t] r n ( r n + 1 [ x , t ]) = r n [ x P ′ , t ] = [ x P , t ] = r n [ x , t ] . Hence s n = r n ∣ L n + 1 s_n = r_n|_{L_{n+1}} s n = r n ∣ L n + 1 is a retraction of L n + 1 L_{n+1} L n + 1 onto L n L_n L n , and by induction r n ∘ r m = r n r_n \circ r_m = r_n r n ∘ r m = r n for m ≥ n m \geq n m ≥ n .
Define h : X → lim ( L n , s n ) h : X \to\lim(L_n,s_n) h : X → lim ( L n , s n ) by h ( y ) = ( r n ( y ) ) n h(y) = (r_n(y))_n h ( y ) = ( r n ( y ) ) n ; this is a thread because r n ∘ r n + 1 = r n r_n \circ r_{n+1} = r_n r n ∘ r n + 1 = r n . The map h h h is continuous, and injective because r n ( y ) → y r_n(y) \to y r n ( y ) → y . Let ( z n ) (z_n) ( z n ) be a thread and m ≥ n m \geq n m ≥ n . Since r n = r n ∘ r n + 1 ∘ ⋯ ∘ r m − 1 r_n = r_n \circ r_{n+1} \circ\cdots\circ r_{m-1} r n = r n ∘ r n + 1 ∘ ⋯ ∘ r m − 1 and this composition agrees with s n ∘ ⋯ ∘ s m − 1 s_n \circ\cdots\circ s_{m-1} s n ∘ ⋯ ∘ s m − 1 on L m L_m L m , we get r n ( z m ) = z n r_n(z_m) = z_n r n ( z m ) = z n . If y y y is a limit point of ( z m ) (z_m) ( z m ) in X X X , continuity of r n r_n r n gives r n ( y ) = z n r_n(y) = z_n r n ( y ) = z n for every n n n , h ( y ) = ( z n ) h(y) = (z_n) h ( y ) = ( z n ) . Thus h h h is a continuous bijection of compact metrizable spaces, hence a homeomorphism.
Corollary 5.9 applies to the Lelek fan, to which Corollary 5.15 below does not apply (see Problem 6.2). As Remark 5.8 shows, density of D ( X ) \mathcal{D}(X) D ( X ) gives no evidence that X ∈ D C P X \in\mathcal{DCP} X ∈ D C P .
Example 5.11. The converse of Lemma 3.12 fails. Let S S S and E E E be as in Example 4.35 , A = [ − 1 , 0 ] × { 0 } A = [-1,0] \times\{0\} A = [ − 1 , 0 ] × { 0 } , K = S ∪ A K = S \cup A K = S ∪ A , and let r : K → S r : K \to S r : K → S be the identity on S S S and map A A A to ( 0 , 0 ) (0,0) ( 0 , 0 ) . Put f ( x , y ) = x f(x,y) = x f ( x , y ) = x on S S S and f = f ∣ S ∘ r f = f|_S \circ r f = f ∣ S ∘ r , so f = 0 f = 0 f = 0 on A A A and Z 0 = E ∪ A Z_0 = E \cup A Z 0 = E ∪ A . By Example 4.35 , g S ( x , y ) = y g_S(x,y) = y g S ( x , y ) = y is f ∣ S f|_S f ∣ S -admissible. In K K K , the set A ∖ { ( 0 , 0 ) } A \setminus\{(0,0)\} A ∖ {( 0 , 0 )} is open and contained in Z 0 Z_0 Z 0 , so M f ( k ) = Z 0 M_f(k) = Z_0 M f ( k ) = Z 0 for k ∈ A ∖ { ( 0 , 0 ) } k \in A \setminus\{(0,0)\} k ∈ A ∖ {( 0 , 0 )} by Lemma 3.1 . Hence g = g S ∘ r g = g_S \circ r g = g S ∘ r is not f f f -admissible, although g = g S ∘ r g = g_S \circ r g = g S ∘ r and g ∣ S = g S g|_S = g_S g ∣ S = g S . Attaching an arc to a fibre enlarges M f M_f M f .
Solenoids and the buckethandle continuum. Recall that a solenoid is the inverse limit of circles with bonding maps z ↦ z n i z \mapsto z^{n_i} z ↦ z n i , n i ≥ 2 n_i \ge2 n i ≥ 2 (see [34 2.8] for the case n i = p n_i=p n i = p ), and the buckethandle continuum is the inverse limit of copies of [ 0 , 1 ] [0,1] [ 0 , 1 ] with the tent map t ↦ 1 − ∣ 2 t − 1 ∣ t \mapsto1-\lvert2t-1\rvert t ↦ 1 − ∣ 2 t − 1 ∣ [34 2.9] . The tent map is open, and by [22 Theorem 232] the same continuum is obtained from any sequence of maps of [ 0 , 1 ] [0,1] [ 0 , 1 ] onto itself each of which is the union of two homeomorphisms of complementary subintervals onto $[0,1]. (The buckethandle is often called the Knaster continuum; it should not be confused with Knaster’s hereditarily indecomposable continuum of 1922, which is the pseudo-arc {citep}[34 p. 229] and contains no arcs.) Both are indecomposable, and in both the arc components coincide with the composants: for solenoids because every nondegenerate proper subcontinuum is an arc ({citep}[34 2.16] for $n_i=p$; the general case is the same), and for the buckethandle see {citep}[40 p. 268] . Since composants are dense {citep}[34 5.20(a)] `, all arc components are dense.
Hence, by Theorem 5.1 , on a solenoid and on the buckethandle continuum every function that is constant on no arc belongs to D ( K ) \mathcal{D}(K) D ( K ) ; for solenoids see also Proposition 5.14 below.
Remark 5.12. The density of arc components is not special to these examples. Call a continuum K K K confluently locally connected-like if for every ε > 0 \varepsilon>0 ε > 0 there is a confluent ε \varepsilon ε -mapping of K K K onto a locally connected continuum; this holds, for instance, for every inverse limit of locally connected continua with confluent (e.g. open, or monotone) bonding maps. By [11 Corollary 2.26] , every such continuum is arc Kelley, and in particular all its arc components are dense. Hence, on every such continuum, every function that is constant on no arc belongs to D ( K ) \mathcal{D}(K) D ( K ) (Theorem 5.1 ), and condition (b) of Theorem 4.2 holds for every f f f and every level with empty interior (Lemma 4.6); since levels with interior are collapsed (Lemma 3.1 ), whether K ∈ D C P K \in\mathcal{DCP} K ∈ D C P depends only on condition (a), as for arc connected continua. By a theorem of Oversteegen and Prajs [37 Corollary 3.5] (see also [11 Section 3] ), the nondegenerate atriodic continua of this kind are exactly the solenoids and the Knaster type continua.
Remark 5.13. Indecomposability alone does not place a continuum in D C P \mathcal{DCP} D C P . By a theorem of Fukaishi and Matsuhashi [18 Theorem 1.1] , for every continuum X X X there are an indecomposable continuum Y ⊇ X Y \supseteq X Y ⊇ X and an open retraction r : Y → X r:Y \to X r : Y → X (whose fibres are Cantor sets). If Y Y Y were in D C P \mathcal{DCP} D C P , so would be its open image X X X (Theorem 4.15 ). Taking for X X X the sin ( 1 / x ) \sin(1/x) sin ( 1/ x ) continuum (Example 4.35), we obtain an indecomposable continuum that is not in D C P \mathcal{DCP} D C P ; more generally, every continuum not in D C P \mathcal{DCP} D C P is an open retract of an indecomposable continuum not in D C P \mathcal{DCP} D C P . Since the sin ( 1 / x ) \sin(1/x) sin ( 1/ x ) continuum is arc-like, by [30 Theorem 3.2] the continuum Y Y Y can moreover be chosen tree-like, so there are tree-like indecomposable continua outside D C P \mathcal{DCP} D C P .
Proposition 5.14. Let K K K be a solenoid, let f ∈ C ( K ) f \in C(K) f ∈ C ( K ) , let I I I be the union of the interiors of the fibres of f f f , and let TF \operatorname{TF} TF be the set of points having a Cantor box neighbourhood on each of whose arcs { x } × J \{x\}\times J { x } × J the function f f f is constant. If TF ⊆ cl I \operatorname{TF} \subseteq\operatorname{cl} I TF ⊆ cl I , then f ∈ D ( K ) f \in\mathcal{D}(K) f ∈ D ( K ) .
Proof. Levels with interior are collapsed. Let Int Z c = ∅ \operatorname{Int} Z_c=\varnothing Int Z c = ∅ ; (b) holds by Lemma 4.6. For (a) let u ∈ Z + u \in Z^+ u ∈ Z + and U U U a neighbourhood of u u u . If U ∩ { f > c } U \cap\{f>c\} U ∩ { f > c } meets I I I we get a point with f > c f>c f > c on a collapsed level. Otherwise W ′ = U ∩ { f > c } W'=U \cap\{f>c\} W ′ = U ∩ { f > c } is a nonempty open set covered by Cantor boxes (every point of a solenoid has Cantor box neighbourhoods) and disjoint from cl I ⊇ TF \operatorname{cl} I \supseteq\operatorname{TF} cl I ⊇ TF , so f f f is nowhere locally constant on W ′ W' W ′ and some box in W ′ W' W ′ has an arc on which f f f is not constant. That arc carries crossing points (at all but countably many levels of its image), which are collapsed by Proposition 3.15 . Apply Theorem 4.2 . □ \square □
Fans with the property of Kelley. A continuum X X X with metric d d d has the property of Kelley if for every ε > 0 \varepsilon>0 ε > 0 there is δ > 0 \delta>0 δ > 0 such that whenever a , b ∈ X a,b\in X a , b ∈ X , d ( a , b ) < δ d(a,b)<\delta d ( a , b ) < δ , and A A A is a subcontinuum of X X X containing a a a , there is a subcontinuum B B B of X X X containing b b b with Hausdorff distance H ( A , B ) < ε H(A,B)<\varepsilon H ( A , B ) < ε ; see [9 ] . An inverse limit of continua with the property of Kelley with confluent (in particular, open) bonding maps has the property of Kelley [22 Theorem 219] ; for fans there is a converse. (Corollary 5.15 below). Corollary 4.23 covers the smooth fans with a closed set of end points. For fans with the property of Kelley, Theorem 4.15 gives the following.
Corollary 5.15. Let X X X be a fan with the property of Kelley. Then X = lim ( X n , f n ) X = \lim(X_n, f_n) X = lim ( X n , f n ) with X n X_n X n locally connected fans and f n f_n f n open [9 Theorem 3] , and with π n : X → X n \pi_n : X \to X_n π n : X → X n the projections, the functions h ∘ π n h \circ\pi_n h ∘ π n , h ∈ C ( X n ) h \in C(X_n) h ∈ C ( X n ) , n ∈ N n \in\mathbb{N} n ∈ N , belong to D ( X ) D(X) D ( X ) and are dense in C ( X ) C(X) C ( X ) .
Proof. The projections π n \pi_n π n are surjective [34 Exercise 2.14] and open ([29 Corollary 2.1.10] ; we recall the argument): the sets π m − 1 ( U ) \pi_m^{-1}(U) π m − 1 ( U ) , m ≥ n m \ge n m ≥ n , U U U open in X m X_m X m , form a base of X X X , and π n ( π m − 1 ( U ) ) = f n , m ( U ) \pi_n(\pi_m^{-1}(U)) = f_{n,m}(U) π n ( π m − 1 ( U )) = f n , m ( U ) is open, f n , m = f n ∘ ⋯ ∘ f m − 1 f_{n,m} = f_n \circ\cdots\circ f_{m-1} f n , m = f n ∘ ⋯ ∘ f m − 1 being open. Since X n X_n X n is a locally connected continuum, h ∈ D ( X n ) h \in D(X_n) h ∈ D ( X n ) by Corollary 4.8 , so h ∘ π n ∈ D ( X ) h \circ\pi_n \in D(X) h ∘ π n ∈ D ( X ) by Theorem 4.15 . Since π n = f n ∘ π n + 1 \pi_n = f_n \circ\pi_{n+1} π n = f n ∘ π n + 1 , these functions form an algebra; it contains the constants and separates the points of X X X , so it is dense in C ( X ) C(X) C ( X ) by the Stone–Weierstrass theorem. □
Open problems
Problem 6.1. Is every arc connected metrizable continuum in D C P DCP D C P ? By the remark preceding Theorem 4.4 , condition (b) of Theorem 4.2 always holds for arc connected K K K , so only condition (a) has to be verified. By Theorem 5.1 the difficulty comes only from functions that are constant on some arc, and by Lemma 3.1 only from their levels with empty interior.
A weaker question: is every arc connected metrizable continuum in which the points having a locally connected open neighbourhood or a Cantor box neighbourhood are dense in D C P DCP D C P ? This would contain Theorems 4.4 and 4.31 , and Theorem 4.22 for the comb and the Cantor fan. The restriction to Cantor boxes matters: with boxes as in Remark 3.16 the question would be equivalent to the problem, since for an arc connected continuum Y Y Y every point of Y × ( 0 , 1 ) Y \times(0,1) Y × ( 0 , 1 ) has a box neighbourhood, so a positive answer would give Y × [ 0 , 1 ] ∈ D C P Y \times[0,1] \in DCP Y × [ 0 , 1 ] ∈ D C P and hence Y ∈ D C P Y \in DCP Y ∈ D C P by Corollary 4.16. For continua with tooth charts, Theorem 4.25 reduces the question to showing that the roots lie in cl Σ \operatorname{cl}\Sigma cl Σ . A test case is the dendroid of [21 Example 7.15] : C × [ 0 , 1 ] C \times[0,1] C × [ 0 , 1 ] with the two end points of each gap of C C C in C × { 0 } C \times\{0\} C × { 0 } identified, whose roots form an arc.
Problem 6.2. Is every dendroid in D C P DCP D C P ? Every smooth dendroid? Every fan? No dendroid outside D C P DCP D C P is known. By Theorem 4.22 and Corollary 4.23, the Cantor comb, the Cantor fan and all smooth fans with a closed set of end points are in D C P DCP D C P . A natural test case is the Lelek fan, the unique smooth fan with a dense set of end points [8 , 5 ] . Since its set of end points is dense, it does not have the property of Kelley [10 Theorem 6] ; in particular it is neither a retract nor an open image of the Cantor fan [10 Theorems 6, 8] , and Corollary 5.15 does not apply to it. On the other hand, every Cantor fan contained in the Lelek fan is a retract of it [2 Theorem 3.1] , so by Corollary 3.13 and Theorem 4.22 every function on the Lelek fan that factors through such a retraction belongs to D ( X ) D(X) D ( X ) , where X X X is the Lelek fan, and by Corollary 5.9 D ( X ) D(X) D ( X ) is dense in C ( X ) C(X) C ( X ) ; by Corollaries 5.6 and 5.7 the same holds for every smooth dendroid and every fan; by Remark 5.8 , however, this is no evidence for D C P DCP D C P . Since its end points are dense and no end point of a fan lies in a Cantor box (every point of a box is interior to an arc of the box), it contains no Cantor boxes, and Proposition 3.15 does not apply. A convenient explicit model for constructing test functions is given in [4 Theorem 4.34] : if 0 < r < 1 < ρ 0 < r < 1 < \rho 0 < r < 1 < ρ and r k = ρ ℓ r^k = \rho^\ell r k = ρ ℓ only for k = ℓ = 0 k = \ell= 0 k = ℓ = 0 , then the set of all sequences ( x , a 1 x , a 1 a 2 x , a 1 a 2 a 3 x , … ) ∈ [ 0 , 1 ] N (x,a_1x,a_1a_2x,a_1a_2a_3x,\ldots) \in[0,1]^{\mathbb{N}} ( x , a 1 x , a 1 a 2 x , a 1 a 2 a 3 x , … ) ∈ [ 0 , 1 ] N with a i ∈ { r , ρ } a_i \in\{r,\rho\} a i ∈ { r , ρ } is a Lelek fan, the union of straight segments from the origin of the Hilbert cube; it is the generalized inverse limit of [ 0 , 1 ] [0,1] [ 0 , 1 ] with the single set-valued bonding function whose graph is the union of the segments y = r x y=rx y = r x and y = ρ x y=\rho x y = ρ x in [ 0 , 1 ] 2 [0,1]^2 [ 0 , 1 ] 2 , and it is mapped into itself by the shift. Every coordinate function is strictly increasing along each leg and hence in D ( X ) D(X) D ( X ) by Theorem 5.1 ; a function outside D ( X ) D(X) D ( X ) , if one exists, must be constant on some arcs. The Lelek fan also arises in complex dynamics: for 0 < λ < 1 / e 0 < \lambda< 1/e 0 < λ < 1/ e the Julia set J λ J_\lambda J λ of z ↦ λ e z z \mapsto\lambda e^z z ↦ λ e z , together with the point ∞ \infty ∞ , is homeomorphic to the Lelek fan, with ∞ \infty ∞ as its top [1 ] (see also [5 ] ). Thus the question is whether { M f } ′ ′ = A f \{M_f\}^{\prime\prime}=\mathcal{A}_f { M f } ′′ = A f for every multiplication operator M f M_f M f on C ( J λ ∪ { ∞ } ) C(J_\lambda\cup\{\infty\}) C ( J λ ∪ { ∞ }) .
By [7 ] , a dendroid is smooth if and only if it is the inverse limit of finite dendrites with bonding maps that are monotone relative to the points of a thread. These bonding maps cannot be chosen open, nor even confluent, in general: locally connected continua have the property of Kelley (see [11 p. 1105] ), so by [22 Theorem 219] an inverse limit of dendrites with confluent (for instance open, or monotone in the usual sense) bonding maps has the property of Kelley, which the Lelek fan does not. Thus the first question of Problem 6.4 does not reach the Lelek fan (the second one does, by Lemma 5.10), and it is natural to ask whether D C P \mathcal{DCP} D C P is preserved under inverse limits with bonding maps that are monotone relative to the points of a thread.
Problem 6.3. Is a retract of a continuum in D C P \mathcal{DCP} D C P in D C P \mathcal{DCP} D C P ? Is a continuous image of a continuum in D C P \mathcal{DCP} D C P in D C P \mathcal{DCP} D C P ? By [10 Theorem 6] , every fan with the property of Kelley is a retract of the Cantor fan, so by Theorem 4.22 a positive answer to the first question would put all of them in D C P \mathcal{DCP} D C P . By Kuperberg’s theorem [25 Theorem 3.5] (see also [10 Theorem 12] ) the continuous images of the Cantor fan are exactly the uniformly pathwise connected continua; a positive answer to the second question would put all of them in D C P \mathcal{DCP} D C P . Example 5.11 shows that Lemma 3.12 alone does not settle the first question. By Theorem 4.15 both questions have a positive answer for open retractions and open maps.
Problem 6.4. Is D C P \mathcal{DCP} D C P preserved under inverse limits with open bonding maps? A positive answer would give, via Corollary 5.15, that all fans with the property of Kelley are in D C P \mathcal{DCP} D C P ; since the bonding maps in the definitions of solenoids and of the buckethandle continuum (Section 5) are open, it would also put all solenoids and the buckethandle continuum in D C P \mathcal{DCP} D C P . The same question can be asked for inverse limits whose bonding maps are retractions, as in Corollary 5.5; for these D ( K ) \mathcal{D}(K) D ( K ) is always dense in C ( K ) C(K) C ( K ) . By Lemma 5.10, every smooth fan, in particular the Lelek fan, is such a limit of finite unions of arcs with a common end point.
Problem 6.5. Find a characterization of D C P \mathcal{DCP} D C P in terms of the topology of K K K alone, that is, one that does not require verifying f ∈ D ( K ) f\in\mathcal{D}(K) f ∈ D ( K ) for every f ∈ C ( K ) f\in C(K) f ∈ C ( K ) separately. In particular, is the converse of Theorem 4.7 true: if L ( K ) L(K) L ( K ) is dense and K ∈ D C P K\in\mathcal{DCP} K ∈ D C P , is every arc component of K K K either dense or contained in the closure of its interior? This holds in all examples we know: the sin ( 1 / x ) \sin(1/x) sin ( 1/ x ) continuum, Examples 4.38 and 4.39, the union of { 0 } × [ − 1 , 1 ] \{0\}\times[-1,1] { 0 } × [ − 1 , 1 ] with two sin ( 1 / x ) \sin(1/x) sin ( 1/ x ) curves approaching it from both sides, and a ray spiralling onto a circle (the first and the last two are not in D C P \mathcal{DCP} D C P by Proposition 4.36). A positive answer would give a purely topological characterization of D C P \mathcal{DCP} D C P for continua with L ( K ) L(K) L ( K ) dense.
Acknowledgment. Several results of this paper, among them the topological reformulation in Section 2, the sufficient condition for collapsed levels in Section 4 (Theorem 4.2) and the theorems on the Cantor comb, the Cantor fan and the Cantor grid, were obtained in an extended interaction between the author and Claude, an AI system developed by Anthropic. Claude also prepared drafts of the text. A separate instance of Claude, given only the manuscript, reviewed the proofs. The author has checked all arguments and is fully responsible for the contents of the paper.
Community College of Philadelphia, 1700 Spring Garden St., Philadelphia, PA 19130, USA
Email address: akitover@ccp.edu