Non-Metrizable Manifolds and Contractibility: Mathieu Baillif August 8, 2023

Non-metrizable manifolds and contractibility
Mathieu Baillif
August 8, 2023
Abstract
We investigate whether non-metrizable manifolds in various classes can be homotopy equivalent to a
arXiv:2308.03459v1 [math.GN] 7 Aug 2023
CW-complex (in short: heCWc), and in particular contractible. We show that a non-metrizable manifold
cannot be heCWc if it has one of the following properties: it contains a countably compact non-compact
subspace; it contains a copy of an ω1 -compact subset of an ω1 -tree; it contains a non-Lindelöf closed
subspace functionally narrow in it. (These results hold for more general spaces than just manifolds.) We
also show that the positive part of the tangent bundle of L+ is not heCWc (for any smoothing). These
theorems follow from stabilization properties of real valued maps. On a more geometric side, we also show
that the Prüfer surface, which has been shown to be contractible long ago, has an open submanifold which
is not heCWc. On the other end of the spectrum, we show that there is a non-metrizable contractible
Type I surface.
1 Introduction
By space we mean ‘topological Hausdorff space’, in particular ‘regular’ and ’normal’ imply
Hausdorff. By manifold, or n-manifold if we want to empasize the dimension, is meant a
connected space each of whose points has a neighborhood homeomorphic to the Euclidean
space Rn for some finite n ≥ 1. A surface is a 2-manifold. Every function is assumed continuous
otherwise stated. We use brackets h , i for ordered pairs reserving parenthesis ( , ) for open
intervals. The letters α, β, γ are used exclusively for ordinals, and an ordinal is assumed to be
the set of its predecessors. Undefined terminology used in this introduction is explained either
at its end or in later sections.
This article is concerned with homotopy questions for non-metrizable manifolds and similar
spaces. This is a somewhat exotic subject for which there exists a small but growing literature;
let us cite the book by D. Gauld [18, Chapter 5.2] and the articles by D.Gauld, S. Deo, A.
Gabard and the author [17, 2, 7, 11], for instance. It is worth noting that the methods and
results in these papers (and the present one) belong mainly to general topology (with a small
set theoretic flavor) and have nothing in common with homotopy theory per se.
Acknowledging its relative unpopularity, let us explain why we think homotopy in non-
metrizable manifolds is an interesting subject.1 Milnor showed in [26] that a metrizable manifold
is homotopy equivalent to a CW-complex (in short: heCWc), hence metrizable manifolds that
are contractible coincide with those having vanishing homotopy groups by Whitehead Theorem
(see, e.g., [22, p. 346]). If one releases the metric assumption, this is not true anymore as
there are simple non-metrizable non-contractible manifolds with vanishing homotopy groups
(e.g. the longray described below). Of course, any space with a non-trivial homotopy group is
non-contractible. It was at first not clear to us whether contractible non-metrizable manifolds
1
The true and honest reason is that we happen to like to think about these problems. It might unfortunately be
seen as slightly insuficient.
1
do exist at all. We then noticed that examples were found a long time ago by Calabi and
Rosenlicht: the Prüfer surface and some of its variants are contractible. (The original source is
[10], see also [17].)
While manifolds can be widly different one from another, they share the local properties of
the Euclidean space. In particular, any manifold is locally compact and thus Tychonoff, locally
(path) connected and locally second countable. In addition, metric manifolds are normal (even
perfectly normal and monotonically normal, see e.g. [18, p. 23] for definitions), second countable
and thus Lindelöf and paracompact. Lindelöfness and paracompactness are actually two of the
many properties equivalent to metrizability for manifolds, see [18, Chapter 2]. In metric spaces,
Lindelöfness is equivalent to being ω1 -compact, and compactness to countable compactness.
(Recall that a space is ω1 -compact or has countable extent iff its closed discrete subsets are at
most countable. A countably compact space is ω1 -compact.) Non-metrizable manifolds, on the
contrary, can exhibit some but not all of these features, and the mere existence of non-metrizable
manifolds with such-and-such properties sometimes depends on the axioms of set theory. For
example, perfectly normal manifolds do exist under CH and do not under MA + ¬CH (due to
M.E. Rudin and P. Zenor, see e.g. [18, Chapter 6]), and monotonically normal non-metrizable
manifolds of dimension > 1 do not exist (Z. Balogh and M.E. Rudin, see [9, Cor. 2.3 (e)]).
Given this variety, we thought it might be interesting to see whether contractible non-
metrizable manifolds exist in such-and-such class, or on the other hand if such-and-such topo-
logical property prevents contractibility regardless of the homotopy groups. Some results are
already available, for instance the following theorem was proved by S. Deo and D. Gauld (and
by the author for manifolds):
Theorem 1.1 ([2, Prop. 7.1], [11, Thm 3.4]). A locally metrizable space (in particular, a
manifold) containing a copy of ω1 is non-contractible.
Notice that the proper forcing axiom PFA implies that a countably compact non-compact
subset of a manifold contains a copy of ω1 (due to Balogh, see [8]), an immediate corollary is
that under PFA manifolds with such subsets are non-contractible. We show below that PFA
is actually not needed.
On the other side of the spectrum, the contractible Prüfer surface is neither normal nor
ω1 -compact. (A version of it is realcompact, though, as shown by A. Mardani [25].) It contains
an open (Lindelöf) Euclidean set with a non-metrizable (equivalently, non-Lindelöf) closure. In
a terminology dating to P. Nyikos, it is in the broad class of Type II manifolds (see just below
for a precise definition). In a loose sense, what makes it non-metrizable lies just at the boundary
of some perfectly nice Euclidean open contractible set. We happen to be able to ‘push’ this
non-metrizable stuff inside this Euclidean open set, yielding the contractibility.
The other broad class of non-metrizable manifolds are the Type I manifolds [27, Def. 2.10].
Denote as usual by A the closure of A (in some topological space clear from the context). A
space X is of Type I if and only if X = ∪α∈ω1 Xα , where Xα is open, Xα Lindelöf for each α,
and Xα ⊂ Xβ whenever α < β < ω1 (and of Type II otherwise).
In case of manifolds, each Xα is an open metrizable submanifold. Loosely speaking again,
these manifolds ‘grow slowly’ instead of jumping at once into non-metrizability. In a Type I
space, Lindelöf subsets have Lindelöf closure.
Let us give a summary of our results. The first one (obtained with Peter Nyikos) shows that
Type I contractible manifolds exist.
Theorem 1.2. There is a contractible non-metrizable Type I surface.
The mapping class group of such a manifold, on the other hand, can be quite large, see
Section 7.2 for definitions and details. This surface is obtained by “fattening” an R-special
ω1 -tree and is non-normal. (Trees in this article are to be understood in a set theoretic sense.
Definitions of the relevant notions are recalled in Section 6.) Our other results are on the negative
2
side, we show that spaces in various classes are not contractible, actually not heCWc (which we
recall is short for homotopy equivalent to a CW-complex). We first obtained a generalization
of Theorem 1.1. (Theorems 1.3–1.5 below are stated for manifolds in this introduction, but we
obtained results in larger classes.)
Theorem 1.3. A manifold containing a countably compact non-compact subspace is not heCWc.
Since ω1 -compactess is a weakening of countable compactness, it seemed natural to see

whether the same result holds in this class. What we obtained is something quite weaker and
involves subsets of trees. The order on the tree is essential to our proof.
Theorem 1.4. A manifold containing a copy of an ω1 -compact subset of an ω1 -tree (in partic-
ular, a Suslin tree) is not heCWc.
Recall that the closed longray L≥0 is the product ω1 ×[0, 1) with lexicographic order topology
(not the product topology !). When dealing with it, we often see ω1 as a subset of L≥0 by
identifying α with hα, 0i. L+ is L≥0 with the 0 point removed.
Our next result involves the concept of functionally narrow subspaces in Type I spaces,
which whose definition is recalled in Section 3. Loosely speaking, a closed subspace in a Type
I space X is functionally narrow in X if it cannot be separated by a function X → L≥0 into
two non-Lindelöf pieces, one with a bounded image and the other with an unbounded image.
Functional narrowness is relative to the ambient space, however a closed copy of ω1 in any Type
I space is functionally narrow in it. There are spaces with discrete subspaces functionally narrow
in them as well, see [6, Example 5.12] (and the rest of this paper for more on the subject).
Theorem 1.5. A Type I (non-metrizable) manifold containing a closed non-Lindelöf subspace

functionally narrow in it is not heCWc.
We also show that surfaces in a very interesting class studied extensively by P. Nyikos in
[28] are non-contractible: the (positive part of the) tangent bundle of the longray L+ . (We
write the tangent bundle, but actually there are a lot of non-homeomorphic ones, with distinct
topological properties.)
Theorem 1.6. The positive part of the tangent bundle of L+ for a given smoothing is not
contractible (and thus not heCWc).
(See the preprint [3, Section 6] for related results.) The reason for the non-homotopy equiv-
alence in Theorems 1.3–1.6 relies on stabilization properties of real valued maps on relevant
subspaces. This is basically the same for Theorem 1.1, whose proof is based on the fact that
any function ω1 → R is eventually constant. However, a non-metrizable manifold can be
non-contractible for more ‘geometrical’ reasons: there is a submanifold (with trivial homotopy
groups) of the contractible Prüfer surface which is non-contractible. To understand better
the statement, let us briefly recall that (the separable version of) Prüfer surface is obtained
by adjoining to the open Euclidean plane uncountably many boundary components Rc , each
homeomorphic to R. (This version of Prüfer surface is actually a surface with boundary, see the
end of this introduction.) A precise definition is recalled in Section 5.
Theorem 1.7. The Prüfer surface minus the 0-points in each Rc is a non-contractible (and
hence not heCWc) non-metrizable submanifold of a contractible manifold.
This last example gives us the impression that being heCWc is a somewhat rare occurence
in the world of non-metrizable manifolds: some kind of perfect alignment of uncountably many
pieces seems to be necessary. We are thus inclined to believe that the problems below have
negative answers, but were not able to find any evidence corroborating our feeling.
3
Problem 1.8. Is there a (consistent) example of a contractible ω1 -compact manifold ? More
generally, is there a contractible locally metrizable space containing a non-metrizable ω1 -compact
subspace ?
Theorem 1.4 is far from providing an answer to this problem. For instance, with the help of
♦ one can build an ω1 -compact 2-manifold which mimicks any given ω1 -tree, in particular an
R-special one, see [19, 20]. (Recall that the surface in Theorem 1.2 is build with an R-special
tree.) It is not clear to us whether these manifolds are contractible. Under PFA, a contractible
ω1 -compact manifold cannot be of Type I (see Theorem 2.3 below), but this does not settle the
general case. Another problem, for which we have no information so far:
Problem 1.9. Is there a normal contractible non-metrizable manifold ? A perfectly normal

one, consistently ?
Notice that by Jones Lemma (see e.g. [14, 1.7.12 (c) p.60]), a separable normal manifold is
ω1 -compact if 2ω < 2ω1 , hence Problems 1.8–1.9 might be related.
As said above, Theorems 1.1 and 1.3–1.5 hold for more general spaces than just manifolds.
It might be instructing to note that some amount of non-compactness, first countability and
local Lindelöfness has to be involved to ensure non-contractibility, as the next simple examples
show.
Example 1.10. There are contractible non-metrizable spaces in the following classes:
(i) compact and first countable,
(ii) compact and containing a (non-closed) copy of ω1 ,
(iii) first countable, countably compact and containing a closed copy of ω1 .
Details. For (i), Helly’s space [32, Ex. 107] is compact, first countable and contractible. For
(ii)–(iii), given a space X we let CX = X × [0, 1]/hx, 1i ∼ hy, 1i be the cone over X. CX is
contractible for any space X. For (iii) take Cω1 , which is first countable (a neighborhood basis
of the vertex of the cone is given by the vertex union ω1 × (1 − 1/n, 1) for n ∈ ω, as well known).
Notice that Cω1 is not locally Lindelöf. For (ii) take C(ω1 + 1). It is not first countable (or
even countably tight, for that matter) due to extra point in ω1 + 1.
This paper is organized in somewhat independant sections. In Sections 2 & 3, we prove

Theorems 1.3 and 1.5 involving countable compactness and functionally narrow subspaces; Sec-
tion 4 deals with Theorem 1.6 (about tangent bundles of L+ ) and Section 5 with Theorem 1.7
(about subspaces of Prüfer surface). Then, Section 6 is a bit different as it studies ω1 -trees
and the homotopic properties of their road spaces, generalizing Theorem 1.1 in various ways.
This enables us in particular to prove Theorem 1.4. We also prove that the road space of an
R-special tree is contractible, which gives half the proof of Theorem 1.2. The other half was
given to us by Peter Nyikos who found a way to “fatten” this road space to obtain a surface
homotopy equivalent to it. This part is given in Section 7. This Section contains further results
on the mapping class group of this contractible surface and the Prüfer surface. (We note in
passing that the results of Section 6 appeared in the preprint [5].)
Let us finish this introduction with some definitions for further reference. By cover of a
space, we always mean a cover by open sets. A manifold with boundary M is a connected space
each of whose points has a neighborhood homeomorphic to Rn−1 × R≥0 for some finite n > 0.2
Its boundary ∂M contains the points which do not have a neighborhood homeomorphic to Rn , it
is an n−1-manifold (without boundary) and a closed subset of M . The term club is a shorthand
for closed and unbouded. We use very often the notation ht (x) for h(x, t) and ht for h(·, t) when
2
Whether a manifold possesses a boundary or not is mostly irrelevant for the content of this paper, but we sometimes
allow boundaries to simplify some definitions.
4
h is a function with domain X × [0, 1]. An homotopy between two maps f, g : X → Y is a
function h : X × [0, 1] → Y such that h(·, 0) = f and h(·, 1) = g (in short: h0 = f , h1 = g).
Two spaces X, Y are homotopy equivalent if there are f : X → Y and g : Y → X such that f ◦ g
and g ◦ f are homotopic to idY and idX , respectively. A contraction is an homotopy between
the identity map and a constant map, and a space is contractible if it admits a contraction (or
equivalently if it is homotopy equivalent to a one point space).
We recall briefly the definition of a CW-complex. (Caution: to avoid confusion with the
notation X n usually reserved for the n-th power of the space X in point set topology, we will
denote the n-skeleton by X (n) .) A CW-complex is a space X constructed by starting with
a discrete space X (0) of 0-cells (which are points). We then inductively form the n-skeleton
X (n) by attaching n-cells (which are copies of the open n-disk) eni (i in some index set) via
continuous maps ϕi : S n−1 → X (n−1) ; that is, we identify x ∼ ϕi (x) for x ∈ ∂Din , where Din is
a copy of the closed n-disk and ∂Din ≃ S n−1 its boundary. We endow X (n) with the quotient
topology. The n-cell eni is then the homeomorphic image of Din − ∂Din . X = ∪n∈ω X (n) is given
the weak topology: A ⊂ X is open (or closed) if and only if A ∩ X (n) is open (or closed) for
all n. In particular, X (n) is closed in X. Then, eni is the image under a characteristic map
Φi : Din → X (the composition of ϕi and the inclusion) of Din − ∂Din . For more details, see for
instance Appendix A in [22], from which we borrowed the terminology. Any CW-complex is
hereditarily paracompact (see for instance [16, Thm 1.3.5 & Ex. 1 p. 33]) and hence hereditarily
collectionwise normal (see [14, Theorem 5.1.18 p. 305] for the definition and proof).
2 Countably compact non-metrizable manifolds are

non-contractible
A space is C-closed if any countably compact subspace is closed. The class of C-closed spaces
contains in particular the hereditarily paracompact spaces, the sequential spaces and the regular
spaces with Gδ points, see e.g. [23]. In particular, manifolds and CW-complexes are C-closed.
A cover of a space is a chain cover if it is linearly ordered by the inclusion relation. We may (and
do) assume that chain covers are indexed by regular cardinals and that a member with a strictly
smaller index than another member is strictly contained in it. We will prove the following.
Theorem 2.1. Let X be countably compact and Y be C-closed such that each point in Y has
a Lindelöf (non-necessarily open) neighborhood. Let h : X × [0, 1] → Y be such that ht0 (X) is
compact for some t0 ∈ [0, 1]. Then ht (X) is compact for each t ∈ [0, 1].
Recall that compactness is equivalent to linear compactness, that is, a space S is compact
if and only if any chain cover of it contains S as a member. (To our knowledge, this dates back
to Alexandroff and Urysohn [1].) Hence, if S is a non-compact space, there is a chain cover
U = {Uα : α ∈ κ} of S, with κ a regular cardinal, such that S 6⊂ Uα for each α ∈ κ. Given
such a chain cover U , we say that E ⊂ S is U -unbounded iff E 6⊂ Uα for each α ∈ κ.
Lemma 2.2. Let X be countably compact and non-compact, and let U = {Uα : α ∈ κ} be a
chain cover such that X 6⊂ Uα for each α. Then the following hold.
(a) Given f : X → Y such that f (X) is compact and Y is C-closed, there is c ∈ Y such that
f −1 ({c}) is U -unbounded.
(b) Let En ⊂ X (n ∈ ω) be closed and U -unbounded such that En+1 ⊂ En . Then E = ∩n∈ω En
is closed and U -unbounded.
Proof.
(a) For each α ∈ κ choose xα ∈ X − Uα and set Eα = {xβ : β ≥ α}. Then Eα ∩ Uα = ∅ and
each Eα is U -unbounded. Since Eα is countably compact, f (X) compact and Y C-closed, then
f (Eα ) ⊂ f (X) is compact. Any finite intersection of Eα is nonempty, hence by compactness
5
∩α∈κ f (Eα ) 6= ∅. Take c is this intersection, then f −1 ({c}) ∩ Eα 6= ∅ for each α, which shows
that f −1 ({c}) is U -unbounded.
(b) For each α, ∩n∈ω (En − Uα ) 6= ∅ by countable compactness.
Proof of Theorem 2.1. If X is compact, there is nothing to do. We assume that X is non-
compact. Set A = {t ∈ [0, 1] : ht (X) is compact}. We will show that A is open and closed in
[0, 1]. Since t0 ∈ A, it is non-empty and thus equal to [0, 1].
A is open. Let t ∈ A. Assume that there is no open interval around t in A, hence there
is a sequence tn converging to t such that htn (X) is non-compact for each n ∈ ω. By local
Lindelöfness of Y and compactness of ht (X), there is a Lindelöf V ⊂ Y whose interior contains
ht (X). Since Y is C-closed and htn (X) is countably compact, the latter is closed for each n.
Hence htn (X) 6⊂ V (otherwise it would be compact), choose xn ∈ X such that htn (xn ) 6∈ V .
Take an accumulation point x of the xn ’s, then ht (x) is not in the interior of V , which contains
ht (X), a contradiction.
A is closed. Let tn be a sequence converging to t and assume that htn (X) is compact for each
n. We show that ht (X) is compact as well. Assume it is not the case, hence there is a chain
cover V = {Vα : α ∈ κ} of ht (X) such that ht (X) 6⊂ Vα for each α. Let U be the chain cover of
X given by the inverse images of Vα under ht . Set E0 = X. We assume by induction that En is
closed and U -unbounded, in particular En is countably compact non-compact. Since htn (En ) is
compact, by Lemma 2.2 (a) there is cn ∈ Y such that En+1 = h−1 tn ({cn }) ∩ En is U -unbounded.
By Lemma 2.2 (b), E = ∩n∈ω En is U -unbounded. But htn is constant on E for each n, hence
by continuity ht is also constant on E, taking value c ∈ f (X). Let α be such that c ∈ Vα , then
E ⊂ Uα = h−1 t (Vα ), a contradiction.
Notice that our only use of local Lindelöfness is to ensure that given any compact subspace
of X, there is a (non-necessarily open) neighborhood V of it such that any countably compact
subset of V is compact. Notice also that the proof goes through if one replaces [0, 1] by any
connected sequential space. Theorem 1.3 is an immediate consequence of Theorem 2.1.
Proof of Theorem 1.3. Manifolds are C-closed and each point has a Lindelöf (open) neighbor-
hood. Let thus M be a manifold containing a countably compact non-compact subspace N . Let
W be a CW-complex, and f : M → W , g : W → M be given. Since f (N ) is countably compact
and CW-complexes are paracompact and C-closed, f (N ) is compact, hence so is g ◦ f (N ). It
follows that g ◦ f cannot be homotopic to idM by Theorem 2.1.
We close this section with a consequence of Theorem 2.1 under the locally compact tri-
chotomy axiom LCT due to Eisworth and Nyikos [13]. Recall that a space is ω-bounded iff
any countable set has compact closure. LCT states that a locally compact space is either a
countable union of ω-bounded spaces, or contains an uncountable closed discrete set, or has a
countable subset with non-Lindelöf closure. This axiom holds for instance under PFA or in
models of the form PFA(S)[S], see [30].
Theorem 2.3 (LCT). Let X be locally compact, C-closed, ω1 -compact and non-Lindelöf. If X
is heCWc, then X does not contain a closed Type I subspace.
Proof. Recall that we assume our Type I spaces to be non-Lindelöf. It is well known that in
a locally compact space, each point has an open Lindelöf neighborhood. If Y ⊂ X is closed
and of Type I, the closure (in Y , hence in X) of any Lindelöf subset of Y is Lindelöf. Hence
by LCT Y is a countable union of countably compact subsets Yn , which are thus closed in X.
Since one of the Yn must be non-compact, Theorem 2.1 shows that X is not heCWc.
6
3 A Type I space containing a closed non-Lindelöf
subspace functionally narrow in it is non-contractible
Let X be of Type I, hence X = ∪α∈ω1 Xα with Xα open, Xα Lindelöf and Xα ⊂ Xβ whenever
α < β. The cover by the Xα ’s is said canonical if Xα = ∪β<α Xβ for each limit α. Each chain
cover indexed by ordinals < ω1 can be made canonical by adding members whenever needed.
(Note: our terminology differs slightly from that of Nyikos in [27] who uses the terms canonical
sequence instead of canonical cover.) Two canonical covers agree on a club set of α’s, as easily
seen. A subset of X is bounded if it is contained in some Xα and unbounded otherwise. A map
between Type I spaces is bounded if its range is bounded and unbounded otherwise. A slicer of
(the canonical cover of) X is a map σ : X → L≥0 such that σ(Xα+1 − Xα ) ⊂ [α, α + 1] ⊂ L≥0 . If
X is regular, then X is Tychonoff, each Xα is normal and there is a slicer of X (see [6, Lemmas
2.2–2.3]). The following is proved in [6, Theorem 3.8].
Lemma 3.1.
Let X = ∪α<ω1 Xα be Type I and D ⊂ X be club. Then the following are equivalent.
(a) For each f : X → L≥0 with f (D) unbounded, the set S ⊂ ω1 of α such that f (D − Xα ) ⊂
[α, ω1 ) is club,
(b) For each f : X → L≥0 with f (D) unbounded, f −1 ({0}) ∩ D is bounded in X.
A closed subspace D of X is functionally narrow in X if it satisfies (a) and (b) in the above
lemma. Whether a Type I space X contains a non-Lindelöf subset functionally narrow in it can
be a tricky question: any closed unbounded copy of ω1 is functionally narrow in X, but there
are spaces containing discrete subspaces functionally narrow in them as well, and an Aronszajn
tree (see the definition in Section 6) does not contain any uncountable subspace functionally
narrow in it. What we prove is the following.
Theorem 3.2. A regular Type I space containing a closed non-Lindelöf subset functionally
narrow in it is not heCWc.
For the proof, we need a fact about CW-complexes.
Lemma 3.3. Let W be a CW-complex, and Y = {yα : α ∈ ω1 } be an uncountable subset of W
(all yα are distinct). Then, either there is a cell eni such that eni ∩ Y is uncountable, or there
is an uncountable E ⊂ ω1 and for each α ∈ E, an open Vα ∋ yα such that {Vα : α ∈ E} is a
discrete family.
Recall that a family F of subsets of W is discrete iff any point in W has a neighborhood
which intersects at most one member of F. This implies in particular that for any G ⊂ F,
∪{F : F ∈ G} is closed.
Proof. We can assume that Y ⊂ W (n) for the smallest n such that W (n) ∩ Y is uncountable.
Then, Y ∩ W (n−1) is at most countable, so we can assume that Y ⊂ W (n) − W (n−1) . So each
yα lies in an n-cell. If eni ∩ Y is at most countable for all n-cells eni , take an uncountable subset
E ⊂ ω1 such that each eni intersects Y0 = {yα : α ∈ E} in at most one point. It thus follows that
Y0 is closed discrete. Since any CW-complex is collectionwise normal, it implies in particular
that Y0 can be expanded into a discrete collection of open sets, which gives us the required
Vα .
Proof of Theorem 3.2. Let X = ∪α∈ω1 Xα be a Type I regular space and D ⊂ X be closed,
non-Lindelöf and functionally narrow in X. Let σ : X → L≥0 be a slicer and h : X × [0, 1] → X
be such that h1 = id. We show that ht (D) must be unbounded for each t. Set
τ = inf{t : hs (D) ⊂ X is unbounded ∀s > t}.
7
We first show that hτ must be unbounded on D. This is obvious if τ = 1. If τ < 1, let {tn }n∈ω
be a dense subset of (τ, 1]. By Lemma 3.1 (a) applied to σ ◦ htn , there is a club Sn ⊂ ω1
such that htn (D − Xα ) ⊂ X − Xα for all α ∈ Sn . By continuity, this holds also for hτ and
α ∈ S = ∩n∈ω Sn . Since S is also club, hτ (D) is unbounded. If τ > 0, take a sequence tn ր τ
with htn (D) bounded. By continuity, hτ (D) must also be bounded. This shows that τ = 0,
hence ht is unbounded for each t.
Now, for each α ∈ ω1 choose xα ∈ D − Xα . Let W be a CW-complex, and f : X → W ,
g : W → X be homotopy equivalences. Set Y = {yα = f (xα ) : α ∈ ω1 }. If Y is countable
or there is an uncountable Y ′ ⊂ Y contained in a countable (hence Lindelöf) subcomplex, then
an unbounded subset of D is sent by g ◦ f to a bounded subset of X. Since D is functionally
narrow in X, σ ◦ g ◦ f (D) must be bounded by Lemma 3.1 (recall that σ is a slicer), thus so is
g ◦ f (D). It follows that g ◦ f cannot be homotopic to the identity.
Thus, we can assume that no uncountable subset of Y is contained in a countable subcomplex
of W . Let E ⊂ ω1 and the Vα ∋ yα be given by Lemma 3.3, and partition E into two disjoint
uncountable subsets E0 , E1 . Letting Uα = f −1 (Vα ), by discreteness the Uα are 2-by-2 disjoint
and ∪α∈E0 Uα = ∪α∈E0 Uα . We want to define a continuous χ : X → L≥0 such that χ(xα ) = α
when α ∈ E0 and with value 0 on the complement of ∪α∈E0 Uα . This can be done since we
assumed X to be regular and hence (since of Type I), Tychonoff: set χ to be 0 on X − ∪α∈E0 Uα ,
and inside Uα take any function into [0, α] ⊂ L≥0 that maps xα to α and Uα − Uα to 0. The
inverse image of an interval (a, b) ⊂ L≥0 (resp. [0, b) ⊂ L≥0 ) is then an union of open subsets of
the Uα (resp. together with X − ∪α∈E0 Uα ), and is then open. By definition, χ(D) is unbounded
(because of E0 ), and χ−1 ({0}) ∩ D is unbounded as well (because of E1 ). By Lemma 3.1, D is
not functionally narrow in X, a contradiction.
4 Tangent bundles of L+ are non-contractible

We will use the terminology for Type I spaces defined at the beginning of the previous section.
The following theorem shows that a topologically very interesting class of surfaces with trivial
homotopy groups does not contain contractible ones. Given a smoothing of the longray L+ , write
T L+ for the resulting tangent bundle. P. Nyikos has extensively studied these spaces in [28] and
showed that they have very different topological properties depending on the smoothing. We
refer to this paper for precise definitions. Since we will just use properties shared by all tangent
bundles of L+ , we only need to recall that there is a bundle projection π : T L+ → L+ such that
{π −1 ((0, α)) : α ∈ ω1 } is a canonical cover for T L+ and π −1 ((0, α)) is homeomorphic to R2 for
each α, hence T L+ is a Type I surface. (Here and in the remainder of this section, we view ω1
as a subset of L+ whenever convenient.) T L+ contains a copy of L+ (the 0-section). Removing
it disconnects the surface in two homeomorphic subsurfaces T + , T − , none of which contain a
copy of L+ . Depending on the smoothing, they may contain a copy of ω1 . It is important to
note that π −1 ((0, α)) ∩ T + is also homeomorphic to R2 , thus we slightly abuse the notation by
denoting the restriction of π to T + also by π. We say that a function f with domain T + is
almost constant if there is some club C of ω1 such that f is constant on π −1 (C) ⊂ T + .
Theorem 4.1. If h : T + ×[0, 1] → L+ is given such that ht0 (T + ) is Lindelöf for some t0 ∈ [0, 1],
then ht is almost constant for each t ∈ [0, 1].
An immediate corollary is:
Corollary 4.2. Neither T L+ nor T + are heCWc.
Proof. The homotopy groups of T + are all trivial since any compact subset is contained in
π −1 ((0, α)) (which is contractible) for some α. If T + was heCWc it would then be contractible,
which would yield an homotopy between π : T + → L+ and a constant map. Theorem 4.1
8
impedes it, proving the result for T + . The result for T L+ follows from Theorem 1.1 since it
contains a copy of ω1 .
Following [28], call a subset of T + large if it intersects all the fibers above some stationary
subset of ω1 . We are going to use the following two results of P. Nyikos.
Lemma 4.3 ([28, Thm 4.10 & Cor. 4.15]).
(i) If U is a large open set of T + , then U contains π −1 (C) for a club subset C of ω1 .
(ii) If U ⊂ T + is open and contains π −1 (C) for some club C of ω1 , then given f : U → R, there
is a club D of ω1 such that f is constant on π −1 (D) ∩ U .
Actually, Nyikos proves part (ii) only for U = T + , but his proof works as well in our case.
Proof of Theorem 4.1. Let h be as in the statement. Write T + = ∪α∈ω1 Tα+ with Tα+ =
π −1 ((0, α)) homeomorphic to R2 . Set
A = {t ∈ [0, 1] : ht is almost constant}.
We show that A is open and closed in [0, 1]. Since ht0 (T + ) is Lindelöf, it is contained in some
(0, α) ⊂ L+ which is homeomorphic to R, hence t0 ∈ A by Lemma 4.3 (ii). It follows that
A = [0, 1].
A is closed. Let tn ∈ A (n ∈ ω) be a sequence converging to t ∈ [0, 1] and let Cn ⊂ ω1 be
club such that htn is constant on π −1 (Cn ). Then C = ∩n∈ω Cn is club and by continuity ht is
constant on π −1 (C).
A is open. Let t ∈ A and suppose in the way of contradiction that there is a sequence tn 6∈ A
(n ∈ ω) converging to t. Let γ be such that ht (π −1 (C)) is contained in (0, γ) ⊂ L+ for some
club C. If π h−1 −1 −1
tn ((0, γ)) is stationary, by Lemma 4.3 (i) htn ((0, γ + 1)) contains π (Dn ) for
a club Dn , hence by (ii) tn ∈ A. There is thus a club Cn ⊂ ω1 such that h−1 tn ((0, γ)) misses
−1 −1
the fibers above Cn , that is, htn (π (Cn )) ⊂ [γ, ω1 ). By continuity, ht (π (B)) ⊂ [γ, ω1 ) for
B = C ∩ ∩n∈ω Cn , a contradiction.
We believe that the following problem has a negative answer, but cannot back our claim
with any evidence.
Problem 4.4. Is it possible to have a locally trivial fiber bundle projection p : E → L+ with a
contractible total space E ?
5 A non-contractible sub-surface of the Prüfer sur-

face
We first recall the definition of the Prüfer surface. We just sketch the construction and refer to
[18, Ex. I.25–28] for more details. Notice however that our notation is slightly different. Let
H be the half plane {hx, yi ∈ R2 : x ≥ 0}, and H0 contain the points of H with horizontal
coordinate > 0. By Prüferizing H at height c ∈ R, we mean deleting h0, ci ∈ H and replacing
it by a copy Rc of R, with the following topology. Given E ⊂ R, we write Ec for the copy of E
in Rc . The neighborhoods of the non-deleted points in H are unchanged, and a neighborhood
basis of u ∈ Rc is given by all
c
Ta,b,ǫ = (a, b)c ∪ {hx, yi ∈ H0 : ax + c < y < bx + c , x < ǫ},
c
for a, b ∈ Rc , a < u < b and ǫ > 0. In words, Ta,b,ǫ ∩ H0 is the triangular region between the lines
of slopes a, b passing through h0, ci and the vertical line x = ǫ. If A ⊂ R, by Prüferizing at A we
mean Prüferizing at each point of A. If one Prüferizes at A and deletes {0}×(R−A) from H, the
9
resulting space PA is a surface with boundary ∪c∈A Rc . A finite A yields a surface homeomorphic
to H minus finitely many points in the vertical axis. The surface (with boundary) PR obtained
by Prüferizing at each height is usually called the (separable) Prüfer surface and will be denoted
by P. It is known for a long time that PA is contractible for any A ⊂ R, in particular P and its
Moorized version (see below) are both contractible, see [10]. We will show that deleting some
points in the boundary of P yields a non-contractible submanifold. Let 0c denote the 0 point in
Rc ⊂ P and set Z = ∪c∈R {0c }. Then Z is a closed discrete subset of P.
Example 5.1. The separable surface with boundary P − Z has trivial homotopy groups but is
non-contractible (and hence not heCWc).
We first show the following variation, for which we have a simpler proof of the non-contractibility.
Example 5.2. Let d > 0 and Nd = ∪c∈R [−d, d]c . Then the separable surface with boundary
P − Nd has trivial homotopy groups but is non-contractible (and hence not heCWc).
Details. Recall that H0 is the open half plane {hx, yi ∈ R2 : x > 0}. Firstly, it is clear that the
homotopy groups of P − Nd are trivial. Indeed, any compact subset intersects finitely many Rc ,
and is thus contained in a submanifold homeomorphic to H minus finitely many points in its
boundary, which is contractible. Secondly, if d, d′ 6= 0, the map P − Nd → P − Nd′ defined by
′ ′
hx, yi 7→ h dd x, yi for points in H0 and u 7→ dd u for u ∈ Rc is a homeomorphism. We may thus
assume that d = 12 and prove that P − N1/2 is not contractible.
By way of contradiction, assume h : (P − N1/2 ) × [0, 1] → P − N1/2 to be such that h1 = id
and h0 ≡ ∗. We may assume that ∗ ∈ H0 . Write 1c , −1c for the points 1, −1 ∈ Rc ⊂ P − N1/2 .
We are going to use the subsets of H0 depicted in Figure 1. (Strictly speaking, h0, ci belongs to
H and not to H0 , but it should cause no confusion since it also does not belong to any of the
±
subsets we consider.) We allow δ1 to be 0 and δ2 to be equal to +∞. Ic,δ are closed segments
± ±
in H0 , while Ac,δ1 ,δ2 , Bc,δ1 ,δ2 are open (hence dashed lines do not belong to them).
Figure 1: Subsets of H0 .
Notice that in P − N1/2 , for any δ > 0 the following hold:

+ +
1. 1c is the only accumulation point of Ic,δ not in Ic,δ and −1c is the only accumulation point
− −
of Ic,δ not in Ic,δ ,
2. For any δ > 0, A+ 1 1 − 1
c,0,δ ∪ ( 2 , +∞)c is a neighborhood of ( 2 , +∞)c and Ac,0,δ ∪ (−∞, − 2 )c is
a neighborhood of (−∞, − 12 )c .
Claim 1. Assume that there are δ > 0, u, v ∈ R ∪ {±∞} and t0 ∈ [0, 1] such that both ht0 (1c )
and ht0 (−1c ) are in H0 and have horizontal coordinate > δ for each c ∈ (u, v). Then:
(i) ∃n ≥ 1, u+ , v + , a+ , b+ ∈ Q ∩ [0, 1], with t0 < a+ < b+ and u ≤ u+ < v + ≤ v such that both
10
+
ht (1c ) and ht (−1c ) are in the closure of Bc,δ/(n+1),δ/n for each c ∈ [u+ , v + ] and t ∈ [a+ , b+ ],
(ii) ∃m ≥ 1, u− , v − , a− , b− ∈ Q ∩ [0, 1], with t0 < a− < b− and u ≤ u− < v − ≤ v such that both
−
ht (1c ) and ht (−1c ) are in the closure of Bc,δ/(m+1),δ/m for each c ∈ [u− , v − ] and t ∈ [a− , b− ].
Figure 2: Claim 1.
This claim is summarized in Figure 2, where it is to be understood that the left part implies
the right part. Let us show why it implies the non-contractibility of P − N1/2 . Since h0 sends
everything to ∗ ∈ H0 , set δ0 to be half of ∗’s horizontal coordinate. We now use induction on k
and the claim to find nested intervals [uk , vk ] ⊂ [uk−1 , vk−1 ], tk ≥ tk−1 and δk = δk−1 /(n(k)+ 1),
δk′ = δk−1 /(n(k)) for some n(k) ≥ 1 such that the following holds:
+
(i) htk (1c ) and htk (−1c ) are both in Bc,δ ,δ′
for c ∈ [uk , vk ] when k is even,
k k
−
(ii) htk (1c ) and htk (−1c ) are both inBc,δ if c ∈ [uk , vk ] when k is odd.
′
k ,δk
(These two points correspond exactly to (i) and (ii) in the claim.) Let t = supk∈ω tk and
c ∈ ∩k∈ω [uk , vk ]. Then htk (1c ) and htk (−1c ) should converge to ht (1c ) and ht (−1c ), but are
divergent in P − N1/2 since they zigzag around the horizontal at height c while going to the
boundary. (This horizontal line converges to 0c in P, which is not present in P − N1/2 .)
Proof of Claim 1. We prove (i), as (ii) is entirely similar. Let π : P → R be the projection on
the first coordinate on H0 and take constant value 0 on the boundary components Rc . Then π
is easily seen to be continuous. For n, k ≥ 1 and p, q ∈ Q ∩ [0, 1], Let
+ +
Cn,k,p,q = {c ∈ (u, v) : ht (Ic,1/k ) ⊂ A+
c,δ/(n+1),δ/n ∀t ∈ (p, q)}.
+
We show that each c ∈ (u, v) belongs to some Cn,k,p,q . By assumption π ◦ht0 (1c ) and π ◦ht0 (−1c )
1
are > δ. Notice that ( 2 , +∞)c is a connected component of (P − N1/2 ) − H0 , hence 1c must
first enter H0 before potentially entering any other boundary component, and 1c does leave
( 12 , +∞)c since π ◦ ht0 (1c ) = δ. Since A+ 1 1
c,0,1 ∪ ( 2 , +∞)c is a neighborhood of ( 2 , +∞)c , 1c enters
A+ 1
c,0,1 after leaving ( 2 , +∞)c . There is thus an interval (p, q) (with p > t0 ) and n ≥ 1 such
+ + being compact for each η > 0,
that ht (1c ) ∈ Ac,δ/(n+1),δ/n for each t ∈ (p, q). Now, {1c } ∪ Ic,η
+ ) ⊂ A+
if we shrink the interval (p, q) and take η small enough, then ht (Ic,η c,δ/(n+1),δ/n for each
1 +
t ∈ (p, q). Taking k such that k < η, we obtain that c ∈ Cn,k,p,q .
+
Since any interval of R is a Baire space and there are countably many Cn,k,p,q , one of them must
+
be non-meagre. In particular, for some n, k, p, q, Cn,k,p,q is dense in an interval [u′ , v ′ ] ⊂ (u, v).
If w = hx, yi ∈ H0 , set c(w) = y − x. By continuity of h, if w is in the parrallelogram in H0
11
defined by the lines of slope 1 emerging from h0, u′ i and h0, v ′ i, the vertical line at k1 and the
boundary, then w is sent to the closure of A+
c(w),δ/(n+1),δ/n by ht for t ∈ (p, q), as shown on the
lefthandside of Figure 3.
Figure 3: Proof of Claim 1.
By looking at righthandside of the same figure, we see that if we take c ∈ (u′ , v ′ ) and η
− is sent in the closure of B +
small enough, then Ic,η c,δn+1 ,δ/n by ht for t ∈ (p, q). (For those keen
δ
on deciphering formulas, take η such that u′ − 2η > v and η < min( 4(n+1) , k1 ).) Since −1c is
− , this holds also for −1 . To finish the proof, take a closed subinterval
in the closure of Ic,η c
+ + ′ ′
[u , v ] ⊂ (u , v ).
The proof that P − Z is also non-contractible (Example 5.1) is very similar, but with extra
steps. A shortcut would be to answer the following question in the affirmative, but we could
not decide whether it is the case or not.
Questions 5.3. Given d > 0, are P − Nd and P − Z homeomorphic ?
Let us summarize the argument of P − Z’s non-contractibility. Assume as before that
h : (P − Z) × [0, 1] → P − Z is a contraction with h1 = id and h0 ≡ ∗. We also consider 1c , −1c .
Set t+ (c) = sup{t ∈ [0, 1] : ht (1c ) ∈ (0, +∞)c }. Then 0 < t+ (c) and ht+ (c) (1c ) is a member
of (0, +∞)c that we denote by s(c). Define A±,ℓ ±
c,δ1 ,δ2 as Ac,δ1 ,δ2 , but with slope 1/ℓ instead of
1/2. Take ℓ such that 1ℓ < s(c) 2 . Then there are some p, q ∈ Q, p < q < 1, and n ∈ ω such
that ht (1c ) is contained in Ac,0,+∞ if p < t and in A+,ℓ
+,ℓ
c,1/(n+1),1/n if p < t < q. Up to shrinking
+
(p, q), the same holds for Ic,δ if δ is small enough. The same Baire argument as before yields
an interval [u, v] and ℓ, n, p, q such that for each c ∈ [u, v], ht (1c ) ∈ A+,ℓ
c,1/(n+1),1/n if p < t < q.
− ) ⊂ A+,ℓ
Then ht (Ic,η c,1/(n+1),1/n as well for the same t’s and η small enough (depending on ℓ).
+
This shows in particular that ht (−1c ) is in the closure of Bc,1/(n+1),1/n for the same t’s. Now
proceed with the −1c , etc, to find the same zigzaging diverging sequence.
Our examples so far all have a boundary, but we may get rid of it by Mooreizing the
boundary components (see again [18, Ex. I.27] for details). In its classical version, we take a
boundary component Rc obtained by Prüferization and identify x with −x in it. This has the
effect of folding Rc which is thus ‘swallowed’ by the manifold and is not a boundary component
anymore. We can perform this identification in all the boundary components or only in some
of them. If one takes the Prüfer surface and Moorize each boundary component, we obtain a
contractible boundaryless non-metrizable surface. We can also fold around another point than
0c , or use more complicated functions for the identification. For instance, if as above 0c has
been removed from the boundary component Rc , we may fold (0, +∞)c around 1c by using any
homeomorphism between (0, 1]c and [1, +∞)c for the identification, and similarly fold (−∞, 0)c
around −1c .
12
Example 5.4. The separable surface obtained by Mooreizing the boundary components of P−Z,
by folding (0, +∞)c around 1c and (−∞, 0)c around −1c is non-contractible.
The proof of non-contractibility in Example 5.1 works verbatim. Another way to get rid
of boundary components is to glue collars to them, that is, copies of R × [0, 1). Doing it to P
yields a contractible non-separable surface, while doing it to P − Z yields a non-contractible
subsurface of the former. For more on these variations of the Prüfer surface, see [4].
6 ω1 -trees and their road spaces

The contents of this section appeared in the preprint [5]. The objects we study are set theoretic
trees and their road spaces, which are obtained by joining consecutive points by a line segment
with a topology that makes it locally embeddable in R2 (the details are given below). These
spaces are good toy models for Type I non-metrizable manifolds.
Recall that a tree T is a partially ordered set such that each point has a well ordered set
of predecessors. We define the height of x ∈ T and of T , the α-th-level Levα (T ), the chains
and antichains in T as usual, see for instance [24, Section II.5] or [29]. A subset E of T is
order-dense iff for any y ∈ T there is y ∈ E with x < y. All trees in this note are endowed with
the order (also called interval) topology. A tree is rooted iff it has a unique minimal element
called the root and binary iff each member has exactly 2 immediate successors. A subtree S
is a subset of T with order restricted to S. Notice that the induced topology on a subtree S
is finer (sometimes strictly) than the one given by the order restricted to S. Both topologies
agree if S is closed in T . We assume that our trees are Hausdorff, that is, if x, y ∈ T are at a
limit level and have the same predecessors, then x = y. (This could be false for a subtree.) An
ω1 -tree has countable levels and height ω1 . A tree is Aronszajn (resp. Suslin) if it has height
ω1 and its chains (resp. chains and antichains) are at most countable. When x ∈ T and α is
an ordinal, write T≥x = {y ∈ T : y ≥ x}, x ↾ α for the unique predecessor of x at level α (if x
is below the α-th level, x ↾ α = x) and T≤α for the subset of elements at level ≤ α. If E ⊂ T ,
set E ↓ = {x ∈ T : ∃y ∈ E with x ≤ y} to be its downward closure. We say that the tree T is
R-special iff there is a strictly increasing (not necessarily continuous) function T → R. Recall
that R-special binary ω1 -trees exist in ZFC.
The road space RT of a tree T is obtained by joining consecutive points by an interval [0, 1],
with 0 glued to the lowest point and 1 to the highest. We extend the order in the obvious way.
When convenient we consider T as a subset of RT . The topology on the interior of the added
intervals is that of (0, 1). For x ∈ T ⊂ RT , in order for RT to be (locally) connected any open
set containing x must contain a small portion of each interval emanating from x. In order for
the space to be locally metrizable (and hence first countable), we take these portions uniformly
as follows. Denote by [0, 1]yx ⊂ RT the interval between the two consecutive points x, y ∈ T .
If A ⊂ [0, 1], then Ayx is understood as the corresponding subset of [0, 1]yx . For singletons we
usually write ayx instead of [ {a}yx . If x ∈ T , denote by s(x) the set containing its immediate
successors and set Wx,n = [0, 1/n)yx . If x ∈ s(z) with z ∈ T , a basis for the neighborhoods
y∈s(x)
of x in RT is given by {Wx,n ∪ (1 − 1/n, 1]xz : n ∈ ω} (see Figure 4, left). A basic neighborhood
of x ∈ T ⊂ RT at a limit level is obtained by choosing some z ∈ T , z < x and n ∈ ω and taking
the segment {y ∈ RT : z < y < x} union each Ww,n for those w ∈ T with z < w ≤ x (see Figure
4, right). This makes RT locally embeddable in R2 (as seen by induction, or by extrapolating
from Figure 4). The induced topology on T ⊂ RT is that of T , and RT is arc connected if and
only if T is rooted.
The first result of this section is the following.

Theorem 6.1. The road space RT of a rooted R-special tree is contractible.
13
Figure 4: Neighboroods in RT .
This gives half of the proof of Theorem 1.2, and is proved in Subsection 6.1. The other
two theorems of this section are generalizations of Theorem 1.1. We depart somewhat from the
manifold theme of this paper and investigate quite precise properties of subspaces, trying to see
whether they prevent global contractibility. The first version is to weaken the assumption in
Theorem 1.1 that the subspace is a copy of ω1 .
Theorem 6.2. Let T be a tree, S ⊂ T be uncountable and ω1 -compact in the subspace topology,
and let X be a locally metrizable space. Then there is no continuous h : S × [0, 1] → X such that
h1 : S → X has uncountable image and h0 is constant. In particular, if X contains a copy of S
then X is not contractible.
Our proof of Theorem 6.2 is entirely topological, for one using forcing see [5]. The first step
is to show that one can assume without loss of generality that T is an Aronszajn tree. Then,
the bulk of our argument consists of showing that many well known properties of Suslin trees
also hold for ω1 -compact subsets of trees of height ω1 . These generalizations are probably part
of the folklore and no more than standard exercices, but we found convenient to gather the
proofs of most of them in Subsection 6.2. Then the ‘real’ proof of this theorem (and the fact
that it implies Theorem 1.4) is contained in Subsection 6.3. The property that is central to our
argument is that any continuous map from S to a locally metrizable space must be constant
on the points above x ∈ S if x is high enough in the tree (see Lemma 6.7 (i) below). This is a
slight generalization of a result of Steprāns [33].
Notice that an uncountable ω1 -compact subset of a tree cannot be metrizable, since ω1 -compactness
is equivalent to Lindelöfness in metrizable spaces (see, e.g., [14, Thm 4.1.15]), and a Lindelöf
subset of a tree with countable levels is countable.
A quick corollary, which is another exhibit of the similarities between ω1 and Suslin trees, is
the following. Its proof is immediate since the road space of a tree satisfying the assumptions
is locally metrizable and a Suslin tree is ω1 -compact, as well known (see Lemma 6.6 below).
Corollary 6.3. Let T be a tree of height ω1 . If T contains an uncountable subset S which is
ω1 -compact in the subspace topology (in particular, if T is a Suslin tree), then its road space RT
is non-contractible.
Our second generalization of Theorem 1.1 is to weaken the local metrizability of the target
space X. First countability alone is not enough, as shown by Example 1.10. What we were able
to prove is the following.
14
Theorem 6.4. Let X be a regular space with Gδ points such that each point has a closed [ℵ0 , ℵ1 ]-
compact neighborhood. Let S be a stationary subset of ω1 endowed with the subspace topology.
Then there is no continuous h : S × [0, 1] → X such that h1 : S → X has an uncountable image
and h0 is constant. In particular, if X contains a copy of S then X is not contractible.
Recall that a space is [ℵ0 , ℵ1 ]-compact if and only if any cover of X of size ≤ ℵ1 has a
countable subcover. This is equivalent to the property that given any uncountable subspace
E of X there is x ∈ X which is a condensation point of E, that is, any neighborhood of x
contains uncountably many points of E. We do not know whether this theorem holds if S is
a Suslin tree but note in passing that our main tool (any continuous map from S to X must
be constant on the points above x ∈ S if x is high enough, see Lemma 6.11 below) does not.
(This should appear in a later paper.) Theorem 6.4 holds if X is locally metrizable. Indeed,
a stationary subset of ω1 is ω1 -compact, ω1 -compactness is preserved by continuous functions
and is equivalent to Lindelöfness in metrizable spaces; hence the image of S × [0, 1] in X is
(hereditarily) Lindelöf.
Corollary 6.3 inspires the following question.
Questions 6.5. Let T be a tree and RT be its road space.
(a) Does the contractibility of RT implies that T is R-special ?
(b) Does the non-contractibility of RT imply that T contains a Suslin subtree ?
Notice that it is not possible for (a) and (b) to both have a positive answer since there are
models of set theory with a non-R-special tree T that does not contain a Suslin subtree, see
[31]. We do not know whether these trees have contractible road spaces.
6.1 Proof of Theorem 6.1

Proof. Let T be a rooted R-special tree with root r and let f : T → R be a strictly increasing
function. By replacing f (x) by supy<x f (y) when x is at limit levels, we may assume that f is
continuous. By composing with a strictly increasing function, we may assume that the range
of f is contained in [0, 1]. We first start to define the contraction as a map h : T × [0, 1] → RT .
First, set h1 (x) = x and then define h such that x travels downwards in RT , starting to move
exactly at time t = f (x) and reaching y < x (y ∈ T ) exactly at time t = f (y). Since f is strictly
increasing, there is time available to cross the interval between consecutive points. For those
who like (less readable) formulas, let x at level α be given. If f (x) ≤ t, we set ht (x) = x. If
β < α and f (x ↾ β) ≤ t ≤ f (x ↾ β + 1), we set ht (x) to be ux↾β+1
x↾β where
t − f (x ↾ β)
u= .
f (x ↾ β + 1) − f (x ↾ β)
Finally, if t ≤ f (r), we set ht (x) = r. It should be clear that h is continuous (on T × [0, 1])
and that h0 (x) = r for all x ∈ T . Notice that h has the property that at time f (x), all of T≥x
is squished onto x. Actually, each point on the tree starts to move exactly when all the points
above it reach it, all at the same time. This enables to extend easily the map to all of RT , a
point in the segment [0, 1]yx does not move until y reaches it, and then it follows it until the end.
This gives the required contraction.
Notice that if one sets jt (x) = h1−t (x) for t ∈ [0, 1], jt (x) = x when t < 0 and jt (x) = r
when t > 1, then j is actually a flow, that is jt ◦ js = jt+s .
15
6.2 A collection of facts on uncountable ω1-compact subsets of
trees
Notations: Given a tree T , if A, B ⊂ T and x ∈ T , A < B means that each member of A
is < each member of B. We denote {x} < A by x < A and T≥x ∩ A by A≥x . Notice that
Levα (A↓ ) = Levα (T ) ∩ A↓ .
Lemma 6.6. Let T be a tree of height ω1 and S ⊂ T be uncountable and endowed with the
subspace topology. Then the following hold.
(a) The subspace topology on S ↓ always agrees with the topology given by the induced order. If
S is closed in T , then the subspace topology agrees with the topology given by the induced order
on S.
(b) If S is ω1 -compact, then it intersects a stationary subset of levels of S ↓ (and of T ).
(c) An antichain is closed discrete in T , and a closed discrete subset of T is an at most countable
union of antichains.
(d) If S is ω1 -compact then so is S ↓ .
(e) If S is ω1 -compact and does not contain an uncountable chain, then S ↓ is Suslin.
(f ) If A ⊂ S ⊂ T , then A is a maximal antichain in S if and only if it is a maximal antichain
in S ↓ .
Proof. (a) and (f) are straightforward. For (b), if S avoids a club set of levels of S ↓ then taking
one point between consecutive avoided levels (when available) yields an uncountable discrete
subset which is closed in S. Item (c) is proved e.g. in [29, Thm 4.11], and (d)–(e) follow
immediately from it.
Lemma 6.7. Let T be a tree of height ω1 and S ⊂ T be uncountable and ω1 -compact in the
subspace topology. Then the following hold.
(a) S ↓ is the disjoint union of an at most countable set, a Suslin tree and at most countably
many copies of ω1 . In particular, S ↓ is an ω1 -tree.
(b) There is α ∈ ω1 such that |S≥x | = |(S ↓ )≥x | = ℵ1 when x is above level α.
(c) If C ⊂ S ↓ is uncountable, there is x ∈ S such that (S ↓ )≥x ⊂ C ↓ and |S≥x | = |(S ↓ )≥x | = ℵ1 .
(d) C ↓ is ω1 -compact for any C ⊂ S ↓ .
(e) If C ⊂ S ↓ is club in S ↓ , then {γ : Levγ (C ↓ ) ⊂ C} is club in ω1 .
(f ) If C ⊂ S ↓ is club in S ↓ , then S ∩ C intersects a stationary subset of levels.
(g) If En ⊂ S ↓ are closed and order-dense in S ↓ for n ∈ ω, then ∩n∈ω En is also closed and
order-dense.
(h) Let E, F ⊂ S ⊂ T be closed in S. If |E ∩ F | ≤ ℵ0 , then |E ↓ ∩ F ↓ | ≤ ℵ0 .
(i) Let f : S → Y be continuous where Y is a metrizable space. Then there is β ∈ ω1 such that
f (S≥x ) is a singleton whenever x is above level β in T . In particular, f has a countable image.
Proof. We will use several times (without acknowledging it explicitly) the fact that an at most
countable intersection of club subsets of ω1 is club.
(a) If S contains an uncountable branch B, then there is some minimal xB ∈ B such that
S≥xB is linearly ordered. Indeed, otherwise S contains an uncountable antichain (take points
branching away from B above each height) and hence an uncountable closed discrete subset
by Lemma 6.6 (c). Since the minimal elements of {x : S≥x ⊂ S is an uncountable branch} is
an antichain, S contains at most countably many disjoint uncountable branches. Notice that a
maximal branch in S ↓ contains an unbounded branch of S. Removing the branches above each
minimal xB in S ↓ , what remains is either countable or a Suslin tree by Lemma 6.6 (e).
(b) follows immediately from (a) and the equivalent statement for Suslin trees.
(c) If C ∩ B is uncountable for some branch B ⊂ S ↓ , we are over. If not, by (a) and (b) we
can assume that S ↓ is a Suslin tree such that |S≥x | = ℵ1 for each x ∈ S ↓ . Then the result is
well known (see, for instance, the claim in Theorem 2.1 in [12]).
16
(d) is immediate by (a) since an uncountable (downward closed) subset of a Suslin tree is a
Suslin tree.
(e) By (d), C ↓ is ω1 -compact. Fix α given by (b) and some β > α. Since each uncountable
maximal branch B of C ↓ is a copy of ω1 , C contains a club set of levels of B. By (a) we
may assume that C ↓ is Suslin. By Lemma 6.6 (f), for each n ∈ ω we may find a countable
antichain An ⊂ C above level β which is maximal in C ↓ and such that An+1 > An . Let
γ = supn∈ω sup{height(x) : x ∈ An }. By construction Levγ (C ↓ ) is the set of limit points of
∪n∈ω An , hence since C is closed Levγ (C ↓ ) ⊂ C. This shows that {γ : Levγ (C ↓ ) ⊂ C} is
unbounded in ω1 , and closedness is obvious.
(f) If C ∩ B is unbounded for some maximal uncountable branch B ⊂ S ↓ , then it is home-
omorphic to a club subset of ω1 . By Lemma 6.6 (b) S ∩ C ∩ B is thus stationary. If C ∩ B is
bounded for each uncountable branch of S ↓ , we may assume by (a) that S ↓ is a Suslin tree in
which C is unbounded. By (c) it follows that S is unbounded in C ↓ as well. By (e) and Lemma
6.6 (b), S intersects C on a stationary set of levels.
(g) Closedness is immediate, and order-density follows immediately by (a) and the fact that
the result holds for Suslin trees and ω1 .
(h) Let E, F be the closures in S ↓ of E, F . If E ∩ F is unbounded in S ↓ , by (f) S ∩ E ∩ F =
(S ∩ E) ∩ (S ∩ F ) = E ∩ F is unbounded, a contradiction. Hence E ∩ F is bounded and thus E
and F are disjoint above some level α. It follows that E and F cannot be both unbounded in
the same uncountable branch. It is well known (see e.g. [34, Thm 6.18] or [12, Thm 2.1]) that
if A, B are disjoint closed sets in a Suslin tree, then A↓ ∩ B ↓ is at most countable. Together
with (a), this shows that |E ↓ ∩ F ↓ | ≤ ℵ0 .
(i) Our proof is a slight adaptation of Steprāns topological proof in [33] that a real valued map
with domain a Suslin tree has a countable image. Denote the distance in Y by dist(·, ·). By (b)
me may assume that S≥x is uncountable for each x. Set D(ǫ) = {x ∈ S : diam(f (S≥x )) ≤ ǫ},
where diam stands for diameter, that is, the supremum of the distances between two points in
a set. Assume for now that D(ǫ) is order-dense in S (and thus in S ↓ ) when ǫ > 0. Let D(ǫ)
denote the closure of D(ǫ) in S ↓ . By (g), D = ∩n∈ω,n>0 D(1/n) is closed and order dense in
S ↓ , hence by (f) S ∩ D intersects stationary many levels above each x ∈ S (since |S≥x | = ℵ1 ).
Moreover, D is upward closed in S. Denote by A the minimal elements of D ∩ S. Then A is an
antichain of S ↓ , let β be the supremum
of the heights
of its members. For each x ∈ S above
level β, the diameter of f S≥x is 0, hence f S≥x = f ({x}) and the lemma is proved.
To finish, we now prove that D(ǫ) is order-dense in S. Suppose that it is not the case and let
x ∈ S be such that diam f (S≥y ) > ǫ for each y > x, y ∈ S. We build inductively antichains
Aαn (n ∈ ω, α ∈ ω1 ) such that the following hold.
• Aαn ⊂ S is maximal above x, that is, in (S ↓ )≥x ,
• Aαn+1 > Aαn > Aβm for each n, m ∈ ω and α > β,
• If u ∈ Aαn , v ∈ Aαn+1 , then dist(f (u), f (v)) ≥ ǫ/4.
Assume that Aαn is defined. Set
E = {z ∈ S : z > Aαn and dist(f (z), f (u)) ≥ ǫ/4, where u is the member of Aαn below z}.
It is enough to see that E is order-dense in (S ↓ )≥x , since then we may put its minimal elements
in Aαn+1 . Let thus w ∈ (S ↓ )≥x , w > u ∈ Aαn . Up to going further up, we may assume that
w ∈ S. If dist(f (w), f (u)) ≥ ǫ/4, then w ∈ E. If not, then dist(f (w), f (u)) < ǫ/4. Choose
v ∈ S≥w such that dist(f (w), f (v)) > ǫ/2 (which exists since we assumed diam f (S≥y ) > ǫ for
each y > x). Then
dist(f (v), f (u)) ≥ dist(f (v), f (w)) − dist(f (w), f (u)) > ǫ/4,
and v ∈ E. If Aγn is chosen for each n ∈ ω and each γ < α, set Aα0 to be an antichain
in S, maximal in (S ↓ )≥x , whose members are all > ∪n∈ω,γ<α Aγn . This defines Aαn for each
17
n ∈ ω, α ∈ ω1 .
Set β(α) to be sup{height(y) : y ∈ ∪n∈ω Aαn }, let C be the closure in ω1 of {β(α) : α ∈ ω1 }, and
C ′ be its derived set (that is its limit points). By construction, if there is y ∈ S≥x whose height
in S ↓ is in C ′ , then f is not continuous at y as there is a sequence of points in S converging to
y whose images are ≥ ǫ/4 apart. But by Lemma 6.6 (b) (and the fact that S≥x is ω1 -compact),
there must be such an y, a contradiction. This shows that D(ǫ) is order-dense and concludes
the proof.

.
Our proof relies on simple consequences of the properties given in Lemma 6.7, especially (i).
When S is a subset of a tree T , the height of a point of S is to be understood as its height in
T . Recall that S≥x = T≥x ∩ S.
Lemma 6.8. Let X be a space containing a closed metrizable Gδ subset B ⊂ X, S be an
uncountable ω1 -compact subset of a tree of height ω1 and f : S → X be continuous. Then there
is α ∈ ω1 such that either f (S≥x ) ∩ B = ∅ or f (S≥x ) is a singleton whenever x is at height
≥ α.
Proof. By Lemma 6.7 (b), above some level each S≥x is uncountable, we assume for simplicity
that this holds for each x ∈ S. Let Un be open sets such that ∩n∈ω Un = B. By Lemma 6.7 (h),
there is β ∈ ω1 such that for each n we have
↓ ↓
f −1 (B)) ∩ f −1 (X − Un ) − T≤β = ∅.
It follows that if x ∈ S is at level above β, either f (S≥x ) ⊂ B or f (S≥x ) ∩ B = ∅. Let

E = {x ∈ S : Lev(x) ≥ β and f (S≥x ) ⊂ B}. Then E is an upward closed subspace of S, in
particular it is an ω1 -compact subspace of T . By Lemma 6.7 (i), there is α ≥ β such that f is
constant on S≥x whenever x ∈ E is at level ≥ α. This proves the lemma.
Corollary 6.9. Let X be a space and U ⊂ X be open such that U is contained in a metrizable
open V ⊂ X. Let S be an uncountable ω1 -compact subset of a tree of height ω1 . Let h :
S × [0, 1] → X be continuous. Then there is α ∈ ω1 such that for each t ∈ [0, 1] and each x ∈ S
above level α, either h−1
t (U ) ∩ S≥x = ∅, or ht is constant on S≥x .
Proof. Let {tn : n ∈ ω} be a countable dense subset of [0, 1]. Set B = U . Since B is closed in
the metrizable subset V , it is a Gδ . The previous lemma shows that there is some α such that
when x is above level α, either h−1tn (B) ∩ S≥x = ∅ or htn is constant on S≥x for each n ∈ ω.
The result follows by continuity.
Proof of Theorem 6.2. We first show that we may assume that T is an Aronszajn tree. Indeed,
consider S ↓ ⊂ T ; if some level of S ↓ is uncountable, then S is not ω1 -compact (by Lemma 6.6
(c) & (d)). If S ↓ contains an uncountable branch E, then E is ω1 -compact in the subspace
topology and thus homeomorphic to a stationary subset of ω1 , and we may apply Theorem 6.4
(whose proof given later is independent from this one).
Let h : S × [0, 1] → X be continuous such that h0 (x) = u0 ∈ X and h1 : S → X has uncountable
image. The set of x ∈ S such that S≥x has uncountable image under h1 is uncountable and
downward closed, hence by Lemma 6.7 (c) there is x such that the image of S≥z under h1 is
uncountable for each z ≥ x. Up to replacing S ↓ by (S ↓ )≥x we assume that this holds for all
z ∈ S ↓ . For x ∈ S, set
τ (x) = sup{t : ht is constant on S≥x }.
18
Then τ is an increasing map S → R, however τ is a priori neither continuous nor strictly
increasing. We will show that there is a closed unbounded C ⊂ ω1 such that τ is strictly
increasing on the subspace of members of S at levels belonging to C. Such a subspace is
uncountable (and thus unbounded) in S by Lemma 6.7 (f). Hence as a partially ordered space
S contains an R-special tree and thus (at least) an uncountable antichain (see for instance [29,
Thm 4.29]), a contradiction with the fact that S is ω1 -compact.
It is enough to show that for each x ∈ S, there is α such that τ (y) > τ (x) whenever y > x is at
level ≥ α. Indeed, since the levels of S ↓ are countable by Lemma 6.7 (a) and τ is increasing, a
simple induction provides C. Let x ∈ S be fixed. By continuity, hτ (x) is constant on S≥x with
value u = hτ (x) (x) and thus τ (x) < 1 since h1 (S≥x ) is uncountable. (While it is not needed,
notice that h restricted to the subspace S≥x × [τ (x), 1] ’contracts’ all of S≥x to the point u.)
Since X is locally metrizable, we may choose an open U ∋ u such that B = U is contained in
an open metrizable set. Let α be given by Corollary 6.9. We may assume that α > height(x).
Assume that there is y at level above α such that for each t > τ (x), ht is not constant on S≥y .
By definition of α, this implies that ht (S≥y ) ∩ U = ∅ and in particular that ht (y) 6∈ U for each
t > τ (x). But this contradicts the continuity of h since hτ (x) (y) = u ∈ U . Hence, ht is constant
on S≥y for at least one t > τ (x) and thus τ (y) > τ (x). This finishes the proof.
Theorem 1.4 follows almost immediately.
Proof of Theorem 1.4. Suppose that W is a CW-complex and f : M → W , g : W → M are such

that g ◦ f is homotopic to idM . If f (S) meets uncountably many open cells, then S contains an
uncountable closed discrete subset (see for instance [16, Theorem 1.5.19]) which contradicts its
ω1 -compactness. Hence f (S) meets the interior of at most countably many cells and is therefore
Lindelöf. It follows that g ◦ f (S) is contained in a Lindelöf hence metrizable submanifold of M .
By Lemma 6.7 (i), g ◦ f (S) is constant on S≥x if x is above some fixed level α in T . Since S≥x
is uncountable for some x, Theorem 6.2 yields a contradiction.

The proof is almost exactly the same as that of Theorem 6.2 once we have an equivalent of
Corollary 6.9 in this context. This is given by Corollary 6.12 below. Lemma 6.11 below plays
the role of Lemma 6.7 (i). We first state the following easy fact.
Lemma 6.10. Let S ⊂ ω1 be stationary, endowed with the subspace topology. Then an at most
countable family of club subsets of S has a club intersection.
Proof. A direct proof is straightforward, but notice that since S is an ω1 -compact subspace of
the tree ω1 , the result is also a consequence of Lemma 6.7 (f)–(g).
Lemma 6.11. Let S be a stationary subset of ω1 endowed with the subspace topology. If Y
is regular, [ℵ0 , ℵ1 ]-compact and has Gδ points, then any continuous f : S → Y is eventually
constant, that is, there is α ∈ ω1 such that f (β) = f (α) for each β ≥ α, β ∈ S.
Proof. We start by showing that there is some c ∈ Y such that f −1 ({c}) is club in S. If f (S) is
countable, then this is immediate. We thus assume that f (S) is uncountable. Since Y is [ℵ0 , ℵ1 ]-
compact, f (S) has a condensation point c ∈ Y . Since Y is regular and has Gδ points, we may
choose open sets Un ∋ c, n ∈ ω, such that ∩n∈ω Un = ∩n∈ω Un = {c}. Since c is a condensation
point, f −1 (Un ) is club in S for each n, hence f −1 ({c}) = f −1 (∩n∈ω Un ) = ∩n∈ω f −1 (Un ) is club
in S by Lemma 6.10.
Now, since f −1 (Y − Un ) is closed, it must be bounded, otherwise it intersects f −1 ({c}). It
follows that f −1 (Y − {c}) = ∪n∈ω f −1 (Y − Un ) is bounded in S, say by α. Hence f is constant
on S above α.
19
Corollary 6.12. Let S be a stationary subset of ω1 endowed with the subspace topology. Let
Y be a regular space with Gδ points. Let U ⊂ Y be open such that U is [ℵ0 , ℵ1 ]-compact. Let
h : S × [0, 1] → Y be continuous. Then there is α ∈ ω1 such that for each t ∈ [0, 1] either
h−1
t (U ) ∩ [α, ω1 ) ∩ S = ∅, or ht is constant on [α, ω1 ) ∩ S.
Proof. Fix a countable dense subset {tn : n ∈ ω} ⊂ [0, 1]. By Lemmas 6.10–6.11 we may fix
α such that for each n either [α, ω1 ) ∩ S ∩ h−1
tn (U ) = ∅ or htn is constant above α. The result
follows by continuity.
The proof of Theorem 6.4 can now be done exactly along the same lines as that of Theorem
6.2, we thus only give a sketch. Set τ = sup{t : ht is eventually constant}. Then hτ is
eventually constant and τ < 1 since h1 has uncountable image. Fix α minimal in S such that
hτ is constant above α, take an open U containing hτ (α) such that U is [ℵ0 , ℵ1 ]-compact. By
Corollary 6.12 this contradicts the continuity of h.
7 Surfaces from trees

The goal of this section is to define a surface MT homotopy equivalent to the road space RT of
a given rooted binary ω1 -tree, and to check some additional properties involving the mapping
class group. We recall that we assume that our trees are Hausdorff. (Everything in this section
could be done starting from any rooted ω1 -tree, but for simplicity we shall consider only binary
ones.)
7.1 A surface homotopy equivalent to the road space of a tree

This construction was found and explained to us by Peter Nyikos in 2006. We fix a binary
ω1 -tree T . Let us define the manifold MT . As a set, √ MT = ∪x∈T Mx , where each piece Mx is
the union of a right isoceles triangle with sides 1, 1, 2 and two rectangles with sides 1, 12 glued
together as in Figure 5 (dashed boundaries do not belong to Mx ). The topology in the interior
Figure 5: The ‘pieces’ Mx .
of Mx is the usual topology. We shall need an arbitrary total order ⊳ on T . If y, y ′ are the
two immediate ≤-successors
√ of x in T and y ⊳ y ′ , we glue My and My′ on the top of Mx (after
multiplying them by 1/ 2) as in Figure 6 below (we put the ⊳-bigger on the right), which gives
a topology for points in the bottom boundary of Mx for x at successor levels.
If now p ∈ M belongs to the bottom boundary of Mx for x at a limit level α of T , a
neighborhood of p is given by a half ball in Mx of diameter ǫ centered at p, together with a
‘zigzag’ running between some level γ < α and α which follows p in MT , that is, the union of
open strips of width ǫ in My for all y < x in levels ≥ γ (see Figure 7), choosing the strip that
points eastwards or westwards according to ⊳. Then, to make it open, remove the part that
lies in the bottom boundary of My , for y ∈ Levγ (T ).
20
Figure 6: Gluing successor levels
Figure 7: Neighborhoods at limit levels
This gives MT a structure of Type I surface without boundary with canonical cover given by
Mα = ∪β<α ∪x∈Levβ (T ) Mx . (Strictly speaking, there is a remaining boundary in M0 , 0 the root
of T , but we remove it by adding an extra open strip below it.) Since a countable increasing
union of 2-discs is homeomorphic to a 2-disc (the famous M. Brown theorem, here in dimension
2, see for instance [18, Section 3.1]), by induction Mα is homeomorphic to a 2-disc, and MT is
a surface. T and RT embed in MT as seen in Figure 8 (where Rx = R ∩ Mx ). It is clear that all
homotopy groups of MT are trivial, because the image of a compact set by a continuous map
will be contained in some Mα . Notice that T is closed in MT but RT is not: if x is at a limit
level in T , then at least one of the left or right half part of the bottom boundary of Mx is in
the closure of the segments in RT emanating from the points of T below x.
Figure 8: T and RT in MT .
Theorem 7.1. (P. Nyikos) The inclusion RT ⊂ MT is a homotopy equivalence.

This theorem is a direct consequence of the following lemma (see Exercise 4 p. 18 in [22]):
Lemma 7.2. There is a deformation retraction in the weak sense from MT to RT , i.e. there
is a homotopy h : MT × [0, 1] → MT with h0 = id, h1 (MT ) = RT and ht (RT ) = RT for all
t ∈ [0, 1].
21
The geometric features of MT will be used in the proof. It is interesting to note that it is
not possible to obtain a (real) deformation retraction (i.e. ht ↾ RT = idRT for each t), in fact
there is even no retraction h : MT → RT (i.e. with h ↾ RT = idRT ), since RT is not closed in
MT .
Proof of Lemma 7.2. The homotopy we shall describe is the same in every piece Mx of MT ,
and acts symmetrically on Mx , we thus only need to explain what happens in its left part (see
Figure 9). Let us first describe how the homotopy works globally. The points in Rx will stay in
it, those on the segment between x and x + 1/2 (see Figure 9, left, for the definition) will be all
contracted to the point x while the segment between x + 1/2 and the next level will eventually
cover all of Rx . Meanwhile, the south-west and north-east ‘sides’ will shrink at unit speed to
Rx , and the triangle will be completely contracted to x (see Figure 9, on the right).
Figure 9: The homotopy globally seen.
Let us now be more specific and describe the trajectories of the points under the homotopy.
First, the triangle. The points at height x + 1/2 start moving at once, while the points in the
triangle start moving when the points at height x + 1/2 ‘reach them’. Then, they go at speed 12
parallel to Rx until√they hit the bottom or the vertical line above x, and then they head directly
toward x at speed 22 (see Figure 10, on the left). Hence, h1 (p) = x for all points in the triangle.
Points in the rectangle have one velocity component parallel to Rx which is proportional to the
distance to the next level (that is, between 0 and 21 ), the other component is at speed 12 and goes
perpendicular to Rx . The first component starts immediately, but the second one only when
the south-west or the north-east side reaches it (see Figure 10, on the right). The homotopy ht
Figure 10: Trajectories under ht .
described above is continuous with h0 = id and h1 (MT ) = RT .
Theorem 1.2 follows immediately:
Proof of Theorem 1.2. Let T be a binary R-special ω1 -tree, and set M = MT . Then, by Theo-
rems 7.1 and 6.1, M is a Type I contractible surface.
22
Figure 11: Lemma 7.3.
The tree T embeds in MT as a closed subset. If T is Q-special (that is, there is a strictly
increasing function T → Q), then T is normal in some models of set theory, for instance under
MA + ¬CH, see e.g. [29]. We might hope that normality ‘lifts up’ to MT , but this is never
the case if T is Aronszajn.
Lemma 7.3. If T is an Aronszajn tree, then MT is not normal.
The proof is reminiscent of Lemma 4.9 in [28].
Proof. For each x ∈ T , let x− , x+ ∈ MT be defined as on the lefthandside of Figure 11. Define
E ± = {x± : x ∈ T }, then E + , E − are closed subsets of MT and have empty intersection. Let
U + ⊃ E + , U − ⊃ E − be open. For each x ∈ T , choose Ux+ , Ux− to be basic connected open
subsets of U + , U − containing respectively x+ and x− . If x is at limit level α, we may assume
that Ux+ , Ux− are union of strips as in Figure 7, hence there is γ(x) < α such that y + ∈ Ux+ and
y − ∈ Ux− , where y = x ↾ γ(x). By the pressing down lemma for ω1 -trees (see e.g. [21, p. 154]),
there is γ ∈ ω1 and S ⊂ T meeting a stationary set of levels such that γ(x) = γ for each x ∈ S.
Since Levγ (T ) is countable, there is some y ∈ T at level γ such that y + ∈ Ux+ and y − ∈ Ux−
for x in an uncountable set S ′ ⊂ T . It follows that both V + = ∪x∈S ′ Ux+ and V − = ∪x∈S ′ Ux−
are unbounded connected subsets of U + and U − , respectively. Since T is Aronzsajn and S ′
is uncountable, there are x, y, z in S ′ with x < y, x < z. Since T is Hausdorff there is w at
maximal level such that w < y, w < z. By connectedness, either V + intersects V − below the
level of w, or they intersect in Mw , as shown on the righthandside of Figure 11. This shows
that U + ∩ U − 6= ∅, hence MT is not normal.
7.2 Isotopies and the mapping class group

Our goal is to prove the following theorem.
Theorem 7.4.
(1) The mapping class group of the Prüfer surface P (which is contractible) has cardinality ≥ c.
(2) There is a contractible Type I surface with mapping class group of cardinality ≥ c.
Recall that two homeomorphisms are isotopic if there is a homotopy ht between them which
is an homeomorphism for each t ∈ [0, 1]. The mapping class group MG(M ) of a topological
space X is the group of homeomorphisms of X quotiented by the isotopy classes. Theorem 7.4
shows a sharp contrast between metrizable and non-metrizable surfaces: two homeomorphisms
23
of a metrizable surface are homotopic if and only if they are isotopic (except for some trivial
exceptions). To our knowledge this is due to D.B.A Epstein [15].
We shall prove (2) in detail, and then explain why (1) holds as well. We first recall
some definitions from [27]. Given a non-metrizable Type I manifold M with canonical cover
hMα : α ∈ ω1 i, the tree of non-metrizable-component boundaries Υ(M ) is defined as the set of
topological boundaries ∂C such that C is a non-metrizable connected component of M − Mα for
some α ∈ ω1 , with the following order: σ ≥ τ if σ is a subset of a component whose boundary
is τ . (Of course, Υ(M ) depends on the choice of the canonical cover, but this is not inconse-
quential for our considerations.) Recall that an ω1 -tree T is normalized if T≥x is uncountable
for each x ∈ T . Since we only take non-metrizable components, Υ(M ) is normalized, and if one
starts with a normalized T , then Υ(MT ) = T for the surface MT constructed in Section 7.1.
Given a tree T , we say that a map T → T is an automorphism of T if it is an order preserving
bijection. A map f : M → M of a Type I manifold M (with some given canonical cover) is
a Υ-automorphism if it is a homeomorphism of M such that for all σ ∈ Υ(M ), f (σ) = τ for
some τ ∈ Υ(M ), and if the induced map fb : Υ(X) → Υ(X) is an automorphism. Being a Υ-
automorphism is a strong assumption, however it is not difficult to see that any homeomorphism
of a Type I manifold induces an Υ-automorphism of Υ(M ) for a well choosen canonical cover.
Lemma 7.5. Let M = ∪α∈ω1 Mα be a Type I manifold. If f, g : M → M are Υ-automorphisms
such that fb 6= gb, then f and g are non isotopic.
We shall not give the shortest proof of this lemma but rather an argument which can be
easily adapted for the Prüfer surface.
Proof. Let ht be a homotopy such that h0 = f , h1 = g. Let σ ∈ Υ(M ) be such that fb(σ) 6= b g(σ).
b b
We will show that ht cannot be an isotopy. Since f , gb are automorphisms, f (τ ) 6= b
g (τ ) for all
τ ≥ σ in Υ(M ), and both fb(σ) and gb(σ) belong to the same level γ of Υ(M ). By definition
of Υ(M ), there is some α such that fb(τ ) and b g(τ ) are in different connected components of
M − Mα for all τ > σ. Thus, for τ ≥ σ and all y ∈ f (τ ) ⊂ M , there is an open (connected)
interval Iy ⊂ [0, 1] such that ht (y) ∈ Mα for t ∈ Iy . Now, since Υ(M ) is normalized, for all
β ∈ ω1 with β > γ, we may choose τβ in the β-level of Υ(M ) and yβ in f (τβ ). Since [0, 1]
is second countable, there is an interval J contained in Iyβ for uncoutably many β. Thus, for
t ∈ J, ht sends an unbounded and therefore non-metrizable subset of M inside Mα , which is
metrizable. Hence, ht cannot be a homeomorphism, and ht is not an isotopy.
Proof of Theorem 7.4. We first prove (2). Let T be any rooted binary ω1 -tree and MT ⊃ T be
as in Section 7.1. Take now copies Ti ⊂ Mi of T, MT for each i ∈ Z. Build the manifold M ω by
gluing all these Mi on a open disc as in Figure 12 below. Let T ω be ∪i∈Z Ti together with a new
root x as shown in Figure 12. For m ∈ Z let ϕm : M ω → M ω be the homeomorphism acting
by ‘rotating’ Mi to Mi+m . It is clear that ϕm is a Υ-automorphism of M ω . We now define
M0
M1 M−1
M2 M−2
Figure 12: M ω .
24
M = M ω,ω with copies Mjω of M ω for j ∈ Z in the same way. Again, Υ(M ) = T ω,ω , where T ω,ω
is the ω1 -tree obtained from the Tiω by adding a common root. For s ∈ ZZ , let φs : M → M
be given by φs ↾ Mjω = ϕs(j) : Mjω → Mjω and φs (u) = u if u 6∈ ∪j∈Z Mjω . It is clear that φs
is an Υ-automorphism. By Lemma 7.5, s 7→ [φs ] ∈ MG(M ) is one–to–one, hence MG(M ) has
cardinality at least c. If T is a rooted binary R-special tree, then M is contractible.
We now briefly explain how to prove (1). We use the notations of Section 5 for everything
related to the Prüfer surface P. Given a ∈ R, let fa : P → P send all of Rc to Rc+a pointwise,
and be defined by hx, yi 7→ hx, y + ai in H0 . Then each fa is a homeomorphism. For each c ∈ R
take the 0-point 0c ∈ Rc . The proof of Lemma 7.5 can be easily adapted to show that if h is an
homotopy between fa and fb (with a 6= b), then there is some t such that ht sends 0c inside H0
for an uncountable subset A ⊂ R of c. Recall that {0c : c ∈ A} is closed discrete for any A.
Since H0 is hereditarily separable, the image of these points cannot be closed discrete, so ht is
not an homeomorphism and h is not an isotopy. This shows that the map R → MG(P) defined
by a 7→ [fa ] is one-to-one.
The construction described in the proof of Point (2) but using the first octant in the square of
the longray instead of MT yields an ω-bounded surface with mapping class group of cardinality
≥ c, see [18, Thm 5.31].
References
[1] P. Alexandroff and P. Urysohn. Mémoire sur les espaces topologiques compacts. Number 14
in Proceedings of the section of mathematical sciences. Koninklijke Nederlandse Akademie
van Wetenschappen te Amsterdam, 1929.
[2] M. Baillif. The homotopy classes of continuous maps between some non-metrizable mani-
folds. Topology Appl., 148(1–3):39–53, 2005.
[3] M. Baillif. Homotopy in non-metrizable ω-bounded surfaces. preprint
arXiv:math/0603515v2, 2006.
[4] M. Baillif. Some steps on short bridges: non-metrizable surfaces and CW-complexes.
Journal of Homotopy and Related Structures, 7(2):153–164, 2012.
[5] M. Baillif. (non-)contractible road spaces of trees, 2019. arXiv preprint 1908.09901.
[6] M. Baillif. Relative (functionally) Type I spaces and narrow subspaces. In preparation,
2022.
[7] M. Baillif, S. Deo, and D. Gauld. The mapping class group of powers of the long ray and
other non-metrisable surfaces. Topology Appl., 157(8):1314–1324, 2010.
[8] Z. Balogh. On compact Hausdorff spaces of countable tightness. Proc. Amer. Math. Soc.,
105:755–764, 1989.
[9] Z. Balogh and M.E. Rudin. Monotone normality. Topology Appl., 47(2):115–127, 1992.
[10] E. Calabi and M. Rosenlicht. Complex analytic manifolds without countable base. Proc.
Amer. Math. Soc., 4:335–340, 1953.
[11] S. Deo and D. Gauld. Eventually constant spaces and nonmetrizable homology spheres. J.
Ind. Math. Soc., Special centenary Volume (1907–2007):167–175, 2007.
[12] K. J. Devlin and S. Shelah. Souslin properties and tree topologies. Proc. London Math.
Soc., 39:237–252, 1979.
[13] T. Eisworth and P. Nyikos. Antidiamonds principles and topological applications. Trans.
Amer. Math. Soc., 361:5695–5719, 2009.
[14] R. Engelking. General topology. Heldermann, Berlin, 1989. Revised and completed edition.
25
[15] D.B.A Epstein. Curves on 2-manifolds and isotopies. Acta Math., 115:83–107, 1966.
[16] R. Fritsch and R.A. Piccinini. Cellular structures in topology, volume 19 of Cambridge
studies in advanced mathematics. Cambridge University Press, 1990.
[17] A. Gabard. A separable manifold failing to have the homotopy type of a CW-complex.
Arch. Math., 90:267–274, 2008.
[18] D. Gauld. Non-metrisable manifolds. Springer-Verlag, New York-Berlin, 2014.
[19] S. Greenwood. Constructing type I nonmetrisable manifolds with given Υ-trees. Topology
Appl., 123(1):91–101, 2002. Proceedings of the Janos Bolyai Mathematical Society 8th
International Topology Conference (Gyula, 1998).
[20] S. Greenwood and P. Nyikos. ω1 -compactness in Type I manifolds. Topology Appl., 48(1–
3):165–171, 2005. Proceedings of the Janos Bolyai Mathematical Society 8th International
Topology Conference (Gyula, 1998).
[21] K.P. Hart. More remarks on Souslin properties and tree topologies. Top. App., 15:151–158,
1983.
[22] A. Hatcher. Algebraic topology. Cambridge University Press, Cambridge, 2002. (A free
electronic version is available).
[23] M. Ismail and P. Nyikos. On spaces in which countably compact sets are closed, and
hereditary properties. Topology Appl., 11(3):281–292, 1980.
[24] K. Kunen. Set theory, an introduction to independance proofs (reprint). Number 102 in
Studies in Logic and the Foundations of Mathematics. North-Holland, Amsterdam, 1983.
[25] A. Mardani. Topics in the general topology of non-metric manifolds. The University of
Auckland. Ph.D Thesis, 2014.
[26] J. Milnor. On spaces having the homotopy type of a CW-complex. Trans. Am. Math. Soc.,
90:272–280, 1959.
[27] P. Nyikos. The theory of nonmetrizable manifolds. In Handbook of Set-Theoretic Topology
(Kenneth Kunen and Jerry E. Vaughan, eds.), pages 633–684. North-Holland, Amsterdam,
1984.
[28] P. Nyikos. Various smoothings of the long line and their tangent bundles. Adv. Math.,
93(2):129–213, 1992.
[29] P. Nyikos. Various topologies on trees. In P.R. Misra and M. Rajagopalan, editors, Pro-
ceedings of the Tennessee Topology Conference, pages 167–198. World Scientific, 1997.
[30] P. Nyikos and L. Zdomskyy. Locally compact, ω1 -compact spaces, 2017. ArXiv preprint
1712.03906, submitted.
[31] C. Schlindwein. How special is your Aronszajn tree? eprint arXiv:math/0312445, 2003.
[32] L. A. Steen and J. A. Seebach Jr. Counterexamples in topology. Springer Verlag, New
York, 1978.
[33] J. Steprāns. Trees and continuous mapping into the real line. Topology Appl., 12:181–185,
1981.
[34] S. Todorčević. Trees and linearly ordered sets. In Handbook of Set-Theoretic Topology
(Kenneth Kunen and Jerry E. Vaughan, eds.), pages 235–293. North-Holland, Amsterdam,
1984.
26

Non-Metrizable Manifolds and Contractibility: Mathieu Baillif August 8, 2023

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Non-Metrizable Manifolds and Contractibility: Mathieu Baillif August 8, 2023

Uploaded by

Copyright:

Available Formats

Non-metrizable manifolds and contractibility

Theorem 1.2. There is a contractible non-metrizable Type I surface.

Since ω1 -compactess is a weakening of countable compactness, it seemed natural to see

Theorem 1.5. A Type I (non-metrizable) manifold containing a closed non-Lindelöf subspace

Problem 1.9. Is there a normal contractible non-metrizable manifold ? A perfectly normal

This paper is organized in somewhat independant sections. In Sections 2 & 3, we prove

2 Countably compact non-metrizable manifolds are

τ = inf{t : hs (D) ⊂ X is unbounded ∀s > t}.

4 Tangent bundles of L+ are non-contractible

A = {t ∈ [0, 1] : ht is almost constant}.

5 A non-contractible sub-surface of the Prüfer sur-

Notice that in P − N1/2 , for any δ > 0 the following hold:

Figure 3: Proof of Claim 1.

6 ω1 -trees and their road spaces

The first result of this section is the following.

6.1 Proof of Theorem 6.1

6.3 Proof of Theorem 6.2

It follows that if x ∈ S is at level above β, either f (S≥x ) ⊂ B or f (S≥x ) ∩ B = ∅. Let

Theorem 1.4 follows almost immediately.

Proof of Theorem 1.4. Suppose that W is a CW-complex and f : M → W , g : W → M are such

6.4 Proof of Theorem 6.4

7 Surfaces from trees

7.1 A surface homotopy equivalent to the road space of a tree

Figure 5: The ‘pieces’ Mx .

Figure 7: Neighborhoods at limit levels

Theorem 7.1. (P. Nyikos) The inclusion RT ⊂ MT is a homotopy equivalence.

Figure 9: The homotopy globally seen.

Figure 10: Trajectories under ht .

described above is continuous with h0 = id and h1 (MT ) = RT .

Theorem 1.2 follows immediately:

7.2 Isotopies and the mapping class group

You might also like