A SIMPLE EFIMOV SPACE WITH SEQUENTIALLY-NICE SPACE OF

PROBABILITY MEASURES

TARAS BANAKH AND SAAK GABRIYELYAN

arXiv:2110.09062v1 [math.GN] 18 Oct 2021

Abstract. Under Jensen’s diamond principle ♦, we construct a simple Efimov space K

whose space of nonatomic probability measures Pna (K) is first-countable and sequentially
compact. These two properties of Pna (K) imply that the space of probability measures P (K)
on K is selectively sequentially pseudocompact and the Banach space C(K) of continuous
functions on K has the Gelfand–Phillips property. We show also that any sequence of
probability measures on K that converges to an atomic measure converges in norm, and any
sequence of probability measures on K converging to zero in sup-norm has a subsequence
converging to a nonatomic probability measure.

1. Introduction
For a Tychonoff space K, we denote by P (K) the space of all regular probability measures
on K and by C(K) the space of all real-valued continuous functions on K. Then the space
P (K) endowed with the weak∗ topology induced from the dual space C(K)′ of the Banach
space C(K) is a compact Hausdorff space.
The study of spaces of probability measures is an important direction of research in Measure
Theory and General Topology. For numerous results, open questions and historical remarks
we refer the reader to the survey of Fedorchuk [14] and the recent monograph of Bogachev [4].
Many results in this area describe the interplay between properties of the space P (K) and
topological properties of the compact space K. For example, according to a classical result of
Lindenstrauss [20], a compact space K is Eberlein if and only if so is its space of probability
measures P (K). However, the question of whether P (K) is Corson compact for any Corson
compact K does not depend on ZFC.
The most important topological properties of compact spaces are properties of sequen-
tial type such as first-countability, sequential compactness, countable tightness or selective
sequential pseudocompactness. Following [5], we call a topological space X selectively sequen-
tially pseudocompact Qif for any sequence (Un )n∈ω of nonempty open sets in X there exists a
sequence (xn )n∈ω ∈ n∈ω Un containing a convergent subsequence. Clearly, every infinite
compact selectively sequentially pseudocompact space contains many nontrivial convergent

2010 Mathematics Subject Classification. 03E65; 28A33; 54A35; 54D30.

Key words and phrases. Efimov space, selective sequential pseudocompactness, space of probability mea-
sures, nonatomic probability measure, Gelfand–Phillips property.
1
2 TARAS BANAKH AND SAAK GABRIYELYAN

sequences. For any compact space K we have the following implications.

metrizable sequentially +3 selectively sequentially

compact
KS
pseudocompact


first +3 sequential +3 countably tight
countable

It is easy to show that the space P (K) is metrizable if and only if so is K. However, al-
ready the first-countability of K may not imply even the countable tightness of P (K). Using
the Continuum Hypothesis Haydon [17] constructed a first-countable non-metrizable compact
space K which is the support of some nonatomic probability measure. As it was noticed by
Fedorchuk [14], for the Haydon compact K the tightness of P (K) is uncountable. These
results motivate the inverse question: Assume that P (K) has one of the sequential properties
from the diagram. What can be said about sequential properties of the compact space K?
Clearly, if P (K) is countably tight (sequentially compact or first countable) then so is K.
But what about the case when the compact space P (K) is selectively sequentially pseudo-
compact? Is then also K selectively sequentially pseudocompact? In our main result, under
Jensen’s diamond principle ♦, we answer this question in the negative in a very strong form.
More precisely, under ♦ we construct a simple Efimov space K whose space of probability
measures P (K) contains a dense first-countable sequentially compact subspace and hence is
selectively sequentially pseudocompact. Let us recall that a compact space is called Efimov if
it contains neither nontrivial convergent sequences nor a topological copy of the Stone-Čech
compactification βω of the discrete space ω (we recall the notion of a simple compact space
below in the section Preliminaries). All known examples of Efimov spaces are constructed
only under some additional set-theoretic assumptions: CH [11, 8], ♦ [12, 9], s = ω1 < 2ω1 = c
[13] or cf([s]ω ) = s < 2s < 2c [7]. Yet, the existence of a Efimov space in ZFC is a major
unsolved problem of General and Set-Theoretic Topology, see [21, 16]. Our example of a
Efimov space K is based on a classical construction of a Efimov space due to Fedorchuk and
follows [9].
Our interest to the selective sequential pseudocompactness is motivated by a result of
Drewnowski and Emmanuele [6, Theorem 4.1] which states that if a compact space K is
selectively sequentially pseudocompact (that is, K belongs to the class K′′ in the terminology
of [6]), then the Banach space C(K) has the Gelfand–Phillips property. Recall that a Banach
space E has the Gelfand–Phillips property if every limited set in E is precompact, where
a bounded subset B of E is called limited if each weak∗ null sequence (χn )n∈ω in the dual
space E ∗ converges uniformly on B, that is limn sup{|χn (x)| : x ∈ B} = 0. In [3], we prove
that the selective sequential pseudocompactness of K in this theorem of Drewnowski and
Emmanuele can be weakened to the selective sequential pseudocompactness of any subspace
A ⊆ P (K) that contains K. In particular, if P (K) is selectively sequentially pseudocompact,
then the Banach space C(K) has the Gelfand–Phillips property. Since any compact selectively
sequentially pseudocompact space has many nontrivial convergent sequences, Drewnowski and
Emmanuele ask in [6] whether there exists a compact space K without nontrivial convergent
sequences such that the Banach space C(K) still has the Gelfand–Phillips property. Under
CH, this question was answered negatively by Schlumprecht [23]. These results motivate a
problem posed in [3, Problem 2.7] of finding a Efimov space K whose probability measure
 A SIMPLE EFIMOV SPACE WITH SEQUENTIALLY-NICE SPACE OF PROBABILITY MEASURES 3

space P (K) has in addition a quite strong sequential type property of being a selectively
sequentially pseudocompact space.
Let K be a compact space. The Hahn decomposition theorem states that any probability
measure µ ∈ P (K) can be represented as a convex combination

µ = tµa + (1 − t)µna ,

where t ∈ [0, 1], µa is atomic and µna is nonatomic. Denote by Pa (K) and Pna (K) sub-
spaces of P (K) consisting of all atomic and all nonatomic measures, respectively. It is clear
that δ[K] ⊆ Pa (K) and Pa (K) ∩ Pna (K) = ∅. This observation motivates the problem of
study of (topological) properties of the spaces Pa (K) and Pna (K) and their relationships to
(topological) properties of K and P (K).
The set Pa (K) is dense in P (K) and can be considered also as a subset of the Banach space
ℓ1 (K) as well as a subset of the Banach space c0 (K). Since ℓ1 (K) has the Schur property,
one can naturally ask: Can also Pa (K) have the same property in the sense that every
convergent sequence in Pa (K) is also norm convergent? On the other hand, it is interesting
to understand what happen with sequences (µn )n∈ω converging to zero in the sup-norm of
c0 (K). This motivates the following notion. A sequence (µn )n∈ω of measures on K (not
necessary atomic) is defined to be c0 -vanishing if

lim sup µ({x}) = 0.

n→∞ x∈K

It turns out that our Efimov space K satisfies very strong conditions: (1) every sequence in
P (K) that converges to an atomic measure is norm convergent, and (2) every c0 -vanishing
sequence in P (K) has a subsequence converging to a nonatomic measure. The latter property
of K implies that the space Pna (K) is sequentially compact. In fact, our Efimov space K
is crowded (i.e., it has no isolated points), and therefore the set Pna (K) is dense in P (K).
Moreover, Pna (K) is a first-countable sequentially compact space. The space Pna (K) is
not metrizable (and cannot be metrizable as every metrizable sequentially compact space is
compact and Pna (X) is not compact for a crowded compact space X). On the other hand,
in Example 2.3 we construct a non-metrizable crowded compact space C whose space of
nonatomic probability measures Pna (C) is metrizable.
Now we formulate our main result in which (under ♦) the clause (iv) solves Problem 2.7
from [3].

Theorem 1.1. Under Jensen’s Diamond Principle ♦, there exists a simple crowded Efimov
space K such that
(i) every c0 -vanishing sequence (µn )n∈ω in P (K) has a subsequence converging to a
nonatomic measure µ ∈ Pna (K);
(ii) every sequence in P (K) that converges to an atomic measure converges in norm, so
the weak∗ convergence and the norm convergence coincide on Pa (K);
(iii) the space Pna (K) is non-metrizable, first-countable, Čech-complete, sequentially com-
pact, and the set Pna (K) is dense in P (K);
(iv) the space P (K) is selectively sequentially pseudocompact but not sequentially compact;
(v) the Banach space C(K) has the Gelfand–Phillips property.

Theorem 1.1 will be proved in Section 6 after some preliminary work done in Sections 2–5.
4 TARAS BANAKH AND SAAK GABRIYELYAN

2. Preliminaries
All topological spaces in the paper are Tychonoff. A subset of a topological space is clopen
if it is both closed and open. A topological space X is called crowded if it has no isolated
points. An indexed family of subsets (Fi )i∈I of a set is called disjoint if Fi ∩ Fj = ∅ for any
distinct indices i, j ∈ I. For two sets A, B we write A ⊆∗ B if A \ B is finite. If f : X → Y is
a map and F is a subset of X, we denote by f [F ] the image {f (x) : x ∈ F } of the set F . For a
set A and an indexed family (xα )α∈A of points of a topological space X, we say that (xα )α∈A
converges to a point x ∈ X and write limα∈A xα = x if for every neighborhood Ox ⊆ X the
set {α ∈ A : xα ∈ / Ox } is finite.
The following lemma is Corollary 3.4 in [5]. We give its direct and short proof for the sake
of completeness.
Lemma 2.1. If some dense subspace Y of a topological space X is selectively sequentially
pseudocompact, then X itself is selectively sequentially pseudocompact.
Proof. Let {Un }n∈ω be a sequence of nonempty open subsets of X. Then {Y ∩ Un }n∈ω is a
sequence of nonempty open subsets Q of Y . Since Y is selectively sequentially pseudocompact,
there exists a sequence (xn )n∈ω
Q ∈ n∈ω Y ∩Un containing a convergent subsequence (xnk )k∈ω .
It is clear that (xn )n∈ω ∈ n∈ω Un and (xnk )k∈ω converges in X. Thus X is selectively
sequentially pseudocompact. 
By a measure on a compact space X we understand any regular σ-additive measure µ :
B(X) → [0, ∞) on the σ-algebra B(X) of Borel subsets of X. The regularity of µ means that
for every Borel set B ⊆ X and every ε > 0 there exists a closed subset F ⊆ B of X such
that µ(B \ F ) < ε. A measure µ is a probability measure if µ(X) = 1. The space P (X) of
probability measures on X is endowed with the topology generated by the subbase consisting
of the sets {µ ∈ P (X) : µ(U ) > a} where a ∈ R and U runs over open subsets of X.
By the Riesz Representation Theorem [18, 7.6.1], the space P (X) can be identified with
the subspace
{µ ∈ C ′ (X) : kµk = 1 = µ(1X )}
of the dual Banach space C ′ (X), endowed with the weak∗ topology. Besides the weak∗
topology we shall also use the topology generated by the norm on P (X). For two measures
µ, λ on X their norm-distance can be found by the formula
kµ − λk = sup |µ(B) − λ(B)|.
B∈B(X)

Each element x ∈ X can be identified with the Dirac measure

(
1 if x ∈ B;
δx : B(X) → {0, 1}, δx : B 7→
0 otherwise.
So, X can be identified with the closed subspace {δx : x ∈ X} of P (X).
The convex hull of X in P (X) is denoted by Pω (X) and elements of Pω (X) are called
finitely supported measures on X. The support of a measure µ ∈ P (X) denoted by supp(µ) is
the smallest closed subset F ⊆ K such that µ(X \F ) = 0. The support of a finitely supported
measure µ ∈ Pω (X) is finite and coincides with the set
atom(µ) = {x ∈ X : µ({x}) > 0}
of atoms of µ. The additivity of a probability measure µ ∈ P (X) implies that the set atom(µ)
is at most countable and µ(atom(µ)) ≤ 1. A measure µ ∈ P (X) is called
 A SIMPLE EFIMOV SPACE WITH SEQUENTIALLY-NICE SPACE OF PROBABILITY MEASURES 5

• atomic if µ(atom(µ)) = 1;
• nonatomic if atom(µ) = ∅.
By Pa (X) and Pna (X) we denote the subspaces of P (X) consisting of all atomic measures
and all nonatomic measures on X, respectively. Note that the set Pa (X) of atomic measures
contains Pω (X) and hence is dense in P (X).
Every continuous map f : X → Y between compact spaces induces a continuous map
P f : P (X) → P (Y ) assigning to every µ ∈ P (X) the measure P f (µ) : B(Y ) → [0, 1] defined
by the formula
P f (µ) : B 7→ µ f −1 [B] (B ∈ B).

Lemma 2.2. For every (crowded) compact space X, the set Pω (X) is dense in P (X) and
the set Pna (X) is a (dense) Gδ -set in P (X).
Proof. The density of Pω (X) in P (X) is a well-known fact, but for the convenience of the
reader, we present here a simple proof. Fix any measure µ ∈ P (X) and a neighborhood Oµ
of µ in P (X). By the definition of the topology on P (X), there exist ε > 0 and open sets
U1 , . . . , Un in X such that
\n
{λ ∈ P (X) : λ(Ui ) > µ(Ui ) − ε} ⊆ Oµ .
i=1
Let B be the smallest Boolean algebra of subsets of X containing the family {U1 , . . . , Un },
and let A be the set of atoms of B.PIn each set A ∈ A choose a point xA ∈ A and observe
that the finitely supported measure A∈A µ(A) · δxA belongs to the intersection Oµ ∩ Pω (X),
witnessing that the set Pω (X) is dense in P (X).
Next, we show that the set Pna (X) is (dense) Gδ in P (X). For every natural number n > 0,
consider the continuous map
Ξn : P (X) × X × [ n1 , 1] → P (X), Ξn : (µ, x, t) 7→ (1 − t)µ + tδx ,
and observe that the image
An = Ξ P (X) × X × [ n1 , 1] = µ ∈ P (X) : ∃x ∈ X µ({x}) ≥ n1



is compact and hence closed in P (X). Since Pna (X) = ∞

T
n=1 (P (X) \ An ), the set Pna (X) is
of type Gδ in P (X).
If the space X is crowded, then every set An is nowhere dense in P (X). Indeed, given any
nonempty open set U ⊆ P (X) we can use the density of Pω (X) in P (X) and find a finitely
supported measure µ ∈ U ∩ Pω (X). By the definition of the topology on P (X), there exist a
positive real number ε > 0 and a family of open sets (Ox )x∈supp(µ) in X such that x ∈ Ox for
every x ∈ supp(µ) and
\
{λ ∈ P (X) : λ(Ox ) > µ(Ox ) − ε} ⊆ U.
x∈supp(µ)

Since the space X is Hausdorff, we can additionally assume that the indexed family (Ox )x∈supp(µ)
is disjoint. The absence of isolated points in the crowded space X ensures that for ev-
ery x ∈ supp(µ), the set Ox is infinite and hence contains a finite subset Fx of cardinality
|Fx | > n. Then the finitely supported measure
X X µ({x})
|Fx | δy
x∈supp(µ) y∈Fx
6 TARAS BANAKH AND SAAK GABRIYELYAN

belongs to the set U \ An , witnessing that the closed

T set An is nowhere dense in P (X).
By the Baire Theorem, the Gδ -set Pna (X) = n∈N (P (X) \ An ) is dense in P (X), being a
countable intersection of open dense sets. 
Example 2.3. There is a crowded compact nonmetrizable space C such that the space Pna (C)
is metrizable.
Proof. Let D be an uncountable discrete space. Denote by C = {∞} ∪ (D × 2ω ) the one point
compactification of the space D × 2ω . It is clear that C is crowded and non-metrizable, and
D × 2ω is a metrizable open dense subspace in C. By Theorem 1.4 in [1] and Theorem 4.4 in
[2], the subspace P̂ (D × 2ω ) = {µ ∈ P (C) : µ(D × 2ω ) = 1} of P (C) is metrizable and so is
its subspace Pna (C) ⊂ P̂ (D × 2ω ). 

A compact space X is simple if it is the limit of a well-ordered continuous spectrum

{Xα , παβ : α ≤ β < ω1 } of length ω1 consisting of zero-dimensional compact metrizable
spaces Xα and simple bonding maps παα+1 . A map f : X → Y between topological spaces is
called simple if there exists y ∈ Y such that 1 ≤ |f −1 (y)| ≤ 2, f −1 (y) is nowhere dense in
X and |f −1 (z)| = 1 for every z ∈ Y \ {y}. For more information on inverse spectra, see [10,
§2.5].
According to [19, 7], simple compact spaces do not contain topological copies of βω, so
constructing a simple Efimov space X one should only care about killing all convergent
sequences in X. Another helpful property of simple compact spaces is the uniform regularity
of nonatomic probability measures on such spaces, see [9].
A measure µ ∈ P (X) on a compact space X is called uniformly regular if there exists a

continuous map f : X → Y to a compact metrizable space Y such that µ(F ) = µ f −1 [f [F ]]
for every closed set F ⊆ X.
Lemma 2.4. For every simple compact space X, the space P (X) is first-countable at points
of the set Pna (X).
Proof. By Lemma 4.1 of [9], any nonatomic measure on a simple compact space is uniformly
regular. By (the proof of) Proposition 2 in [22] (see also the statement (v) in [9, p.2605])
the space P (X) is first countable at each uniformly regular measure. Consequently, P (X) is
first-countable at points of the set Pna (X). 

3. Measures on sequentially discrete compact spaces

A topological space X is defined to be sequentially discrete if every convergent sequence in
X is eventually constant. So, X is sequentially discrete if and only if X contains no non-trivial
convergent sequences.
We shall use the following probabilistic characterization of sequentially discrete compact
spaces which can have an independent value.
Proposition 3.1. For a compact space K the following assertions are equivalent:
(i) K is sequentially discrete;
(ii) if a sequence (µn )n∈ω ∈ P (K)ω converges to a measure µ ∈ P (K), then


lim sup sup µn ({x}) − µ({x}) ≤ 0.
n→∞ x∈K
 A SIMPLE EFIMOV SPACE WITH SEQUENTIALLY-NICE SPACE OF PROBABILITY MEASURES 7

Proof. (i)⇒(ii) Suppose for a contradiction that there is a sequence (µn )n∈ω ∈ P (K)ω con-
verging to some measure µ ∈ P (K) such that


(3.1) lim sup sup µn ({x}) − µ({x}) > 0.
n→∞ x∈K

Let A be the set of all numbers c ∈ (0, 1] for which there is a sequence (µn )n∈ω ∈ P (K)ω
converging to some measure µ ∈ P (K) such that


lim sup sup µn ({x}) − µ({x}) ≥ c.
n→∞ x∈K

By our assumption, A is not empty and hence A has the least upper bound sup(A) ∈ (0, 1].
sup(A)
It is clear that (0, sup(A)) ⊆ A. Therefore the inequality 1−sup(A) > sup(A) implies that
′
there exist elements a, a ∈ A ∩ (0, 1) such that 1−a > sup(A) and a < a′ . Choose a sequence
a

(µn )n∈ω ∈ P (K)ω converging to a measure µ ∈ P (K) such that

lim sup µn ({x}) − µ({x}) ≥ a′ > a


(3.2)
n→∞ x∈K

Passing to a subsequence, we can assume additionally that

(3.3) inf sup µn ({x}) − µ({x}) > a.
n∈ω x∈K

Consider the sets F = {x ∈ K : µ({x}) ≥ a} and Fn = {x ∈ K : µn ({x}) ≥ a} for n ∈ ω.

The additivity of the measure µ implies that these sets are finite of cardinality ≤ 1/a. The
condition (3.3) ensures that for every n ∈ ω, the set Fn is not empty.
S
Claim 3.2. The union C := F ∪ n∈ω Fn is a compact subset of K.
Proof. Indeed, since F is finite, to prove the claim it suffices to show that every open neigh-
borhood U ⊆ K of the set F contains all but finitely many sets Fn . Observe that for every
x ∈ K \ U , we have µ({x}) < a and, by the regularity of µ and K, there exists a closed
neighborhood Nx ⊆ K of x such that µ(Nx ) < a and Nx ∩ F =S∅. By the compactness of
K \ U , there exist a finite subset E ⊆ K \ U such that K \ U ⊆ x∈E Nx . Then the set
Oµ = {λ ∈ P (K) : λ(K \ Nx ) > 1 − a for all x ∈ E}
is an open neighborhood of µ in P (K). Since the sequence (µn )n∈ω converges to µ, there
exists m ∈ ω such that µk ∈ Oµ for all k ≥ m. We show that Fk ⊆ U for every k ≥ m.
Indeed, assuming that Fk 6⊆ U for some k ≥ m, we can find a point z ∈ Fk \ U ⊆ K \ U . Since
{Nx : x ∈ E} covers K \ U , there exists x ∈ E such that z ∈ Nx . It follows from µk ∈ Oµ
that µk (K \ Nx ) > 1 − a and hence µk ({z}) ≤ µk (Nx ) < a, which contradicts the definition
of the set Fk ∋ z. This contradiction shows that every neighborhood of F contains all but
finitely many sets Fn . Thus C is compact. 

Being countable, the compact space C is metrizable. Since the space K is sequentially
discrete, the compact metrizable subspace C of K is discrete and hence finite. Therefore
F is an open subset of C, and hence Fn ⊆ F for all sufficiently large numbers n. By the
Pigeonhole Principle, for some subset F ′ ⊆ F the set J = {n ∈ ω : Fn = F ′ } is infinite.
Choose any point z ∈ F ′ . Since µ({z}) ≥ a, we can find a unique measure µ′ ∈ P (K) such
that µ = aδz + (1 − a)µ′ . By the same reason, for every n ∈ J, there exists a unique measure
µ′n ∈ Pa (K) such that µn = aδz + (1 − a)µ′n .
8 TARAS BANAKH AND SAAK GABRIYELYAN

The convergence (µn )n∈ω → µ implies the convergence (µ′n )n∈J → µ′ . For every n ∈ J,
(3.3) implies
 
sup µ′n ({x}) − µ′ ({x}) = sup 1−a
1 a


µn ({x}) − aδz ({z}) − µ({x}) + aδz ({z}) ≥ 1−a .
x∈K x∈K
a a
Therefore, the number 1−a belongs to A which is impossible since 1−a > sup(A). This
contradiction finishes the proof of the implication (i)⇒(ii).
(ii)⇒(i) If K is not sequentially discrete, then K contains a sequence of points (zn )n∈ω
that converge to some point z ∈ K \ {zn }n∈ω . Then the sequence of Dirac measures (δzn )n∈ω
converges to δz in the space P (K), however



lim sup sup δzn ({x}) − δz ({x}) ≥ lim sup δzn ({zn }) − δx ({zn } = 1 6≤ 0.
n→∞ x∈K n→∞


We recall that a sequence of measures (µn )n∈ω on a Tychonoff space X is c0 -vanishing if
lim sup µn ({x}) = 0.
n→∞ x∈X

The following lemma will be used in the proof of our main theorem.
Lemma 3.3. Let X be a sequentially discrete compact space such that every c0 -vanishing se-
quence (λn )n∈ω of probability measures on X has a subsequence that converges to a nonatomic
measure on X. Then for any sequence (µn )n∈ω ∈ P (X)ω that converges to a measure
µ ∈ P (X), we have limn→∞ supx∈X |µn ({x}) − µ({x})| = 0.
Proof. To derive a contradiction, assume that limn∈ω supx∈X |µn ({x}) − µ({x})| =6 0 for some
sequence (µn )n∈ω ∈ P (K)ω that converges to a measure µ ∈ P (K). Replacing (µn )n∈ω by a
suitable subsequence, we can additionally assume that
• for every x ∈ X, the limit µ∞ (x) = limn∈ω µn ({x}) exists, and
• the limit a = limn∈ω supx∈X |µn ({x}) − µ({x})| exists and hence is strictly positive.
Claim 3.4. For every x ∈ X we have µ∞ (x) ≤ µ({x}).
Proof. Suppose for a contradiction that µ∞ (x) > µ({x}) for some x ∈ X. Choose any real
number a with µ∞ (x) > a > µ({x}). Then µ(X \ {x}) = 1 − µ({x}) > 1 − a, and hence the
set
U = {λ ∈ P (X) : λ(X \ {x}) > 1 − a}
is an open neighborhood of the measure µ in P (X). Since limn∈ω µn = µ and limn∈ω µn ({x}) =
µ∞ (x) > a, there exists n ∈ ω such that µn ({x}) > a and µn ∈ U . The latter inclusion implies
that µn (X \ {x}) > 1 − a and hence µn ({x}) = 1 − µn (X \ {x}) < 1 − (1 − a) = a, which is a
desired contradiction completing the proof of the claim. 
P P
Claim 3.4 ensures that x∈X µ∞ (x) ≤ x∈X µ({x}) ≤ 1 which implies that
X
µ∞ = µ∞ (x)δx
x∈X
is a well-defined atomic measure on X. For every n ∈ ω, consider the measures
µ+
n = max{µn − µ∞ , 0} and µ−
n = max{µ∞ − µn , 0}
and observe that µn − µ∞ = µ+ −
n − µn .
 A SIMPLE EFIMOV SPACE WITH SEQUENTIALLY-NICE SPACE OF PROBABILITY MEASURES 9

By Claim 3.4 and Proposition 3.1, we have

0 ≤ lim sup µ+
n ({x}) = lim sup max{(µn ({x}) − µ∞ ({x}), 0}
n∈ω x∈X n∈ω x∈X

≤ lim sup max{(µn ({x}) − µ({x}), 0} = 0,

n∈ω x∈X

which means that the sequence of measures (µ+

n )n∈ω in c0 -vanishing.

Claim 3.5. The sequence (µ− n )n∈ω is c0 -vanishing.

P
Proof. Since x∈X µ∞ (x) ≤ 1, for every ε > 0, there exists a finite set F ⊆ X such that
supx∈X\F µ∞ (x) < ε. Since, for every x ∈ F , the sequence (µn ({x}))n∈ω converges to µ∞ (x),
there exists m ∈ ω such that supn≥m maxx∈F |µn ({x}) − µ∞ (x)| < ε. Then for every n ≥ m
and every x ∈ F , we have
µ−
n ({x}) = max{0, µ∞ (x) − µn ({x})} ≤ |µn ({x}) − µ∞ (x)| < ε.

Also for every x ∈ X \ F we have

µ−
n ({x}) = max{0, µ∞ (x) − µn ({x})} ≤ µ∞ ({x}) < ε.

Therefore, supn≥m supx∈X µ− −

n ({x}) ≤ ε and the sequence (µn )n∈ω is c0 -vanishing. 
Claim 3.6. Each norm-bounded c0 -vanishing sequence of measures (λn )n∈ω on X contains a
subsequence that converges to a nonatomic measure.
Proof. We lose no generality assuming that λn (X) 6= 0 for every n ∈ ω. Since supn∈ω λn (X) =
supn∈ω kλn k < ∞, we can find an infinite subset I ⊆ ω such that the sequence (λn (X))n∈I
converges to some real number a ≥ 0. If a = 0, then (λn (X))n∈I converges to the zero
measure, which is nonatomic.
If a > 0, then kλλnn k )n∈I is a c0 -vanishing sequence of atomic probability measures. By
the assumption of the lemma, there exists an infinite set J ⊆ I such that the subsequence
λn


kλn k n∈J converges to some nonatomic probability measure λ∞ . Since kλn k → a 6= 0, it
follows that the sequence (λn )n∈J converges to the nonatomic measure aλ∞ . 

By Claim 3.6, for the c0 -vanishing sequence (µ+ n )n∈ω there exists an infinite set I
+ ⊆ ω
+ +
such that the subsequence (µn )n∈I + converges to some nonatomic measure µ . Since the
sequence (µ− +
n )n∈I + is c0 -vanishing, there exists an infinite set I ⊆ I such that the sequence
− −
(µn )n∈I converges to some nonatomic measure µ on X. Then the sequence (µn − µ∞ )n∈I =
(µ+ − + −
n − µn )n∈I converges to the nonatomic sign-measure µ − µ . On the other hand, this
sequence converges to the measure µ − µ∞ . Therefore, the measure µ − µ∞ = µ+ − µ− is
nonatomic and hence
µ({x}) = µ∞ ({x}) for every x ∈ X.
Now the c0 -vanishing property of the sequences (µ+ −
n )n∈ω and (µn )n∈ω guarantees that
 
lim sup sup |µn ({x}) − µ({x})|= lim sup sup µn ({x}) − µ∞ ({x})
n∈ω x∈X n∈ω x∈X
−
= lim sup sup max{µ+
n ({x}), µn ({x})} = 0.
n∈ω x∈X


10 TARAS BANAKH AND SAAK GABRIYELYAN

Lemma 3.7. Let X be a sequentially discrete compact space such that every c0 -vanishing se-
quence (λn )n∈ω of probability measures on X has a subsequence that converges to a nonatomic
measure on X. Then any sequence (µn )n∈ω ∈ P (X)ω that converges to an atomic measure
µ ∈ Pa (X) converges to µ in norm.
P
Proof. To show that limn∈ω kµn − µk = 0, fix any ε > 0. Since kµk = x∈atom(µ) µ({x}) = 1,
there exists a finite set F ⊆ atom(µ) such that x∈F µ({x}) > 1 − 41 ε. Let µ = λ + ν where
P

λ = x∈F µ({x})δx and ν = µ − λ. It follows that kλk = µ(F ) = x∈F µ({x}) > 1 − 14 ε and
P P

hence kνk = µ(X \ F ) < 41 ε.

By Lemma 3.3, limn∈ω supx∈X |µn ({x}) − µ({x})| = 0, and hence there exists m ∈ ω such
that
ε
sup max |µn ({x}) − µ({x})| < 4|F | for every n ≥ m.
n≥m x∈F
P
For every n ≥ m, write the measure µn as µn = λn + νn where λn = x∈F µn ({x})δx and
νn = µn − λn . Observe that
X X
ε 1 1 1 1


kλn k = µn (F ) = µn ({x}) > µ({x}) − 4|F | = µ(F ) − 4 ε > 1 − 4 ε − 4 ε = 1 − 2 ε
x∈F x∈F

and hence kνn k = νn (X \ F ) < 12 ε. Then for every n ≥ m, we have

kµn − µk = kλn + νn − λ − νk ≤ kνn k + kνk + kλn − λk
X
= kνn k + kνk + |µn ({x}) − µ({x})| < 21 ε + 41 ε + |F | · 1
4|F | ε = ε,
x∈F

which means that limn∈ω kµn − µk = 0. 

In the proof of Theorem 1.1 we shall need some results on the preservation of the c0 -
vanishing property by projections παω1 : 2ω1 → 2α , παω1 : x 7→ x↾α . For every α ∈ ω1 , the
projection παω1 induces a continuous map P παω1 : P (2ω1 ) → P (2α ) between the corresponding
spaces of probability measures.
Lemma 3.8. For any nonatomic measure µ ∈ Pna (2ω1 ) there exists an ordinal α ∈ ω1 such
that for any ordinal β ∈ [α, ω1 ) the measure P πβω1 (µ) ∈ P (2β ) is nonatomic.
Proof. Since µ is nonatomic and regular, for every n ∈ ω and x ∈ 2ω1 , there exists a neigh-
borhood Ox,n ⊆ 2ω1 of x such that µ(Ox,n ) < 21n . We lose no generality assuming that Ox,n is
of basic form {y ∈ 2ω1 : y↾Fx,n = x↾Fx,n } for some finite set Fx,n ⊆ ω1 . By the compactness of
2ω1 for every n ∈ ω, there exists a finite set En ⊆ 2ω1 such that ω1 =
S
S 2 S x∈En Ox,n . We claim
that any countable ordinal α containing the countable set n∈ω x∈En Fx,n has the desired
property: for any ordinal β ∈ [α, ω1 ) the measure λ = P πβω1 (µ) is nonatomic. To derive a
contradiction, assume that λ({a}) > 0 for some a ∈ 2β . Find n ∈ ω such that 21n < λ({a}).
Choose any point b ∈ 2ω1 such that πβω1 (b) = a and find x ∈ En such that b ∈ Ox,n . Then

(πβω1 )−1 ({a}) = {z ∈ 2ω1 : z↾β = b↾β } ⊆ {z ∈ 2ω1 : z↾Fx,n = b↾Fxn } = Ox,n
and hence

−1
λ({a}) = µ πβω1 1


({a}) ⊆ µ(Ox,n ) < 2n ,
which contradicts the choice of n. 
 A SIMPLE EFIMOV SPACE WITH SEQUENTIALLY-NICE SPACE OF PROBABILITY MEASURES 11

ω
Lemma 3.9. For every c0 -vanising sequence of measures (µn )n∈ω ∈ P (2ω1 ) there ex-
ists a countable ordinal α such that for any ordinal β ∈ [α, ω1 ) the sequence of measures
(P πβω1 (µn ) n∈ω is c0 -vanishing.

Proof. Write each measure µn as µn = (1 − tn )λn + tn νn for some tn ∈ [0, 1] and measures
λn ∈ Pa (2ω1 ) and νn ∈ Pna (2ω1 ). By Lemma 3.8, there exists a countable ordinal α ∈ ω1 such
that for every β ∈ [α, ω1 ) and every n ∈ ω the measure P πβω1 (νn ) is nonatomic. Replacing
α by a larger countable ordinal, if necessary,
S we can additionally assume that the projection
παω1 is injective on the countable set A = n∈ω atom(λn ).
We claim that the ordinal α has the required property. Indeed, fix any ordinal β ∈ [α, ω1 )
and for every n ∈ ω denote the measure P πβω1 (µn ) by µβn . It is clear that µβn = (1−tn )λβn +tn νnβ
where λβn = P πβω1 (λn ) and νnβ = P πβω1 (νn ). The choice of the ordinal α ensures that the
measure νnβ is nonatomic. The injectivity of the projection παω1 on A implies the injectivity of
the projection πβω1 on A. Then we can choose a function f : 2β → 2ω1 such that πβω1 ◦ f (y) = y
for every y ∈ 2β and f ◦ πβω1 (x) = x for every x ∈ A.
Observe that for every y ∈ πβω1 [A], we have

µβn ({y}) = (1 − tn )λβn ({y}) + tn νnβ ({y}) = (1 − tn )λn ({f (y)}) + 0 = µn ({f (y)})
and for every y ∈ 2β \ πβω1 [A], we have

µβn ({y}) = (1 − tn )λβn ({y}) + tn νnβ ({y}) = 0 + 0.

This implies
sup µβn ({y}) = sup µn ({x})
y∈2β x∈2ω1

and hence
lim sup µβn ({y}) = lim sup µn ({x}) = 0,
n→∞ y∈2β n→∞ x∈2ω1

which means that the sequence (P πβω1 (µn ))n∈ω = (µβn )n∈ω is c0 -vanishing. 

4. A key Lemma
In the following lemma we use some notations related to the binary tree 2<ω := n∈ω 2n .
S

Here 2 stands for the ordinal 2 = {0, 1}. The unique element ∅ of 20 is the root of the tree
2<ω . For an element s ∈ 2<ω , we denote by |s| the unique number n ∈ ω such that s ∈ 2n .
The number |s| is called the length of s. For two elements s, t ∈ 2<ω , we write s ≤ t if |s| ≤ |t|
and s = t↾|s| . Also we write s < t if s ≤ t and s 6= t. Let ↑s := {t ∈ 2<ω : s ≤ t} be the upper
set of s in the tree 2<ω (so ↑s consists of all possible extensions of the function s). For i ∈ 2,
let sˆi be the unique element of ↑s such that |sˆi| = |s| + 1 and sˆi(|s|) = i. So, sˆ0 and sˆ1
are immediate successors of s in the tree 2<ω .
Lemma 4.1. Let X be a zero-dimensional compact metrizable space, (µn )n∈ω ∈ P (X)ω be
a c0 -vanishing sequence of probability measures that converge to a measure µ ∈ P (X). For
every z ∈ atom(µ) there exist a subsequence (µnk )k∈ω of the sequence (µn )n∈ω and a family
(Xs )s∈2<ω of clopen subsets of X \ {z} such X∅ = X \ {z} and for every s ∈ 2<ω the following
conditions are satisfied:
(i) Xsˆ0 ∪ Xsˆ1 = Xs and Xsˆ0 ∩ Xsˆ1 = ∅;
12 TARAS BANAKH AND SAAK GABRIYELYAN

(ii) for any clopen neighborhood U ⊆ X of z and any ε > 0 there exists q ∈ ω such that
1
µnk (U ∩ Xs ) ≤ µ(U \ {z}) + 2|s|
+ε for any q ≥ m.
Proof. Fix a neighborhood base {Un }n∈ω at the point z consisting of clopen subsets of X such
that Un+1 ⊆ Un for every n ∈ ω.
We shall inductively construct increasing number sequences (nk )k∈ω , (mk )k∈ω and a se-
quence (Sk )k∈ω of compact subsets of X \{z} such that for every k ∈ N the following inductive
conditions are satisfied:
(ak ) Sk−1 ∩ Umk =  ∅;
(bk ) supn≥nk min |µn (Umk ) − µ(Umk )|, supx∈X µn ({x}) < 61k µ({z}) ≤ 61k ;

(ck ) Sk ⊆ Umk and µnk (Sk ) > µnk (Umk \ {z}) − 61k µ({z}).
We start the inductive construction letting m0 = 0 and S0 = ∅ and choosing n0 ∈ ω such
that the condition (b0 ) is satisfied. Assume that for some k ∈ N, we have constructed a
compact subset Sk−1 of X \ {z} and numbers nk−1 , mk−1 . As Sk−1 is a compact subset of
X \ {z}, there is a number mk > mk−1 such that Umk ∩ Sk−1 = ∅. Since limn∈ω µn = µ
and limn∈ω supx∈X µn ({x}) = 0, there exists a number nk > nk−1 satisfying the inductive
condition (bk ). Using the regularity of the measure µnk , choose a compact set Sk ⊆ Umk \ {z}
satisfying the condition (ck ). This completes the inductive step.
Observe that for every k ∈ N, the inductive conditions (bk ) and (ck ) imply
1
µnk (Sk ) > µnk (Umk \ {z}) − 6k
µ({z}) = µnk (Umk ) − µnk ({z}) − 61k µ({z})
(4.1) 1
µ({z}) − 61k µ({z}) − 61k µ({z}) ≥ 1 − 63k µ({z}).



> µ(Umk ) − 6k
S
Claim 4.2. For every k ∈ N, the compact set Sk can be represented as a union t∈2k Sk,t of
2k pairwise disjoint compact sets such that
max µnk (Sn,t ) − 21k · µnk (Sk ) < 64k .
 
(4.2)
t∈2k

Proof. Identify 2k with the finite ordinal |2k | = {0, . . . , |2k | − 1} via the bijective map 2k →
Pk−1
{0, . . . , |2k | − 1}, t 7→ i
i=0 t(i)2 . Since the compact space Sk is zero-dimensional, it is
homeomorphic to a subspace of the real line. This allows us to identify Sk with a compact
subset of the real line. For a real number r, let ↓r = Sk ∩ (−∞, r] ⊆ Sk ⊆ R. For every t ∈ 2k ,
let
st = inf{s ∈ Sk : µnk (↓s) ≥ t+1
2k nk
µ (Sk )}.
The regularity of the measure µnk ensures that
t+1 t+1 (bk ) t + 1 1
k
µnk (Sk ) ≤ µnk (↓st ) ≤ k µnk (Sk ) + µnk ({st }) < k
µnk (Sk ) + k µ({z})
2 2 2 6
(4.1) t + 1 µn (Sk ) t+2
< µnk (Sk ) + k k < k µnk (Sk ),
2k 6 (1 − 3/6k ) 2
which implies st < st+1 if t < 2k . Let r−1 , r2k −1 ∈ R \ Sk be any real numbers such that
↓r−1 = ∅ and ↓r2k −1 = Sk . Since the space Sk ⊆ R is zero-dimensional, for every t ∈ 2k − 1 we
can choose a real number rt ∈ R \ Sk such that st < rt < st+1 and µnk (↓rt ) < µnk (↓st ) + 61k .
Then
t+1 t+1 1 1 2
|µnk (↓rt ) − µ (Sk )|
2k nk
≤ |µnk (↓rt ) − µnk (↓st )| + |µnk (↓st ) − µ (Sk )|
2k nk
< 6k
+ 6k
= 6k
.
 A SIMPLE EFIMOV SPACE WITH SEQUENTIALLY-NICE SPACE OF PROBABILITY MEASURES 13

k
t ∈ R \ Sk that for every t ∈ 2 the subspace Sk,t = Sk ∩ (↓rt ) \ (↓rt−1 ) is
It follows from rS
compact and Sk = t∈2k Sk,t . It is clear that the family (Sk,t )t∈2k consists of pairwise disjoint
sets.
Finally, observe that

1 t+1
|µnk (Sk,t) − µ (Sk )|
2k nk
= |µnk (↓rt ) − µnk (↓rt−1 ) − µ (Sk ) + 2tk µnk (Sk )|
2k nk
≤ |µnk (↓rt ) − t+1
µ (Sk )|
2k nk
+ |µnk (↓rt−1 ) − 2tk µnk (Sk )| < 62k + 62k = 64k .

Now we continue the proof of the lemma. For every s ∈ 2<ω , consider the set
[ [
Ts = {Sk,t : t ∈ 2k , t↾|s| = s},
k≥|s|

where S0,0 = S0 = ∅. Taking into account that the subsets Sk,t of Unk−1 are compact and the
sequence {Uk }k∈ω is a neighborhood base at z, we conclude that the set Ts is closed in X \{z}.
The definition of the sets Ts implies Tsˆ0 ∪ Tsˆ1 ⊆ Ts . To show that Tsˆ0 ∩ Tsˆ1 = ∅ for any
s ∈ 2<ω , it suffices to note that the sequence (Sk )k∈ω is disjoint by the inductive conditions
(ak ) and (ck ) and, for every k ∈ ω, the family {Sk,t }t∈2k is disjoint by the construction.
Now, by induction on the binary tree 2<ω , we shall find a family (Xs )s∈2<ω of clopen subsets
of X \ {z} such that X∅ = X \ {z} (the base of induction) and

Ts ⊆ Xs , Xsˆ0 ∪ Xsˆ1 = Xs , Xsˆ0 ∩ Xsˆ1 = ∅ for every s ∈ 2<ω .

Assume that for some s ∈ 2<ω the set Xs has been constructed. Since X is a zero-
dimensional compact metrizable space, the space Xs is zero-dimensional and being Lindelöf
it is strongly zero-dimensional in the sense that any disjoint closed sets in Xs can be separated
by clopen neighborhoods, see [10, Theorem 6.2.7]. Since Tsˆ0 and Tsˆ1 are two closed disjoint
sets in Ts ⊆ Xs , there are clopen sets Xsˆ0 and Xsˆ1 such that Tsˆ0 ⊆ Xsˆ0 , Tsˆ1 ⊆ Xsˆ1 ,
Xsˆ0 ∪ Xsˆ1 = Xs , and Xsˆ0 ∩ Xsˆ1 = ∅. This completes the inductive step.
By the inductive construction, the family (Xs )s∈2<ω satisfies the condition (i) of the lemma.
To verify the condition (ii), fix s ∈ 2<ω , ε > 0 and a clopen neighborhood U of z in X. Find
q ≥ |s| such that

3 4 ε ε
(4.3) Umk−1 ⊆ U, 6k
+ 3k
< 2 and |µnk (U ) − µ(U )| < 2 for every k ≥ q.

We claim that the number m has the property required in (ii). Indeed, if t ∈ ↑s, then
Sk,t ⊆ Ts ⊆ Xs , and if t 6∈ ↑s, then the disjointness of Tt↾s and Ts and the construction of Xs
imply Sk,t ∩ Xs = ∅ and therefore
[ [ [
(4.4) Sk ∩ Xs = (Sk,t ∩ Xs ) = (Sk,t ∩ Xs ) = Sk,t .
t∈2k t∈2k ∩↑s t∈2k ∩↑s
14 TARAS BANAKH AND SAAK GABRIYELYAN

Note also that by (ck ) and (4.3), Sk ⊆ Umk ⊆ U . Now, for any k ≥ q, we have

µnk (U ∩ Xs ) ≤ µnk (U \ Sk ) + µnk (Sk ∩ Xs )

(4.4) X
= µnk (U \ Sk ) + µnk (Sk,t )
t∈2k ∩↑s
X
= µnk (U \ Sk ) + µnk (Sk ) − µnk (Sk,t )
t∈2k \ ↑s
(4.2)
1
< µnk (U ) − (2k − 2k−|s| ) µ (Sk ) − 64k


2k nk
(4.1)
1 3 4



< µnk (U ) − 1 − 2|s|
µ(Umk ) − 6k
− 3k
(4.3)
ε 1 3 4


< µ(U ) + 2 − µ(Umk ) + 2|s|
+ 6k
+ 3k
(4.3)
1 1
< µ(U \ Umk ) + ε + 2|s|
≤ µ(U \ {z}) + 2|s|
+ ε,

as desired. 

5. An implication of ♦
In this section we apply Jensen’s diamond principle ♦ to producing a special enumeration
of the set (P (2ω1 ))ω by countable ordinals.
Jensen’s diamond principle is the following statement (see [15]):
(♦) there exists a transfinite sequence (xα )α∈ω1 ∈ α∈ω1 2α such that for every x ∈ 2ω1 ,
Q
the set {α ∈ ω1 : x↾α = xα } is stationary in ω1 .
A set S ⊆ ω1 is stationary if S has nonempty intersection with any closed unbounded subset
of ω1 . By [15], ♦ implies CH and follows from the Gödel Constructibility Axiom V = L
(which is consistent with ZFC).
We shall apply ♦ to produce a nice enumerations of elements of the limit spaces of sub-
continuous ω1 -spectra.
Let λ be any ordinal. A λ-spectrum if an indexed family Σ = (Xα , παβ : α ≤ β < λ)
consisting of sets Xα and functions παβ : Xβ → Xα such that παγ = πβγ ◦ παβ for any ordinals
α ≤ β ≤ γ in λ. The spectrum Σ is defined to be subcontinuous if for every limit ordinal β < λ
and any distinct elements x, x′ ∈ Xβ there exists an ordinal α < β such that παβ (x) 6= παβ (x′ ).

Proposition 5.1. Let Σ = (Xα , παβ : α ≤ β ≤ ω1 ) be a subcontinuous (ω1 + 1)-spectrum such

that for every α < ω1Qthe set Xα has cardinality |Xα | ≤ c. Under ♦ there exists a transfinite
sequence (xα )α∈ω1 ∈ α∈ω1 Xα such that for every x ∈ Xω1 , the set {α ∈ ω1 : xα = παω1 (x)}
is stationary in ω1 .

Proof. Assuming ♦, we obtain a transfinite sequence (yα )α∈ω1 ∈ α∈ω1 2α such that for every
Q
y ∈ 2ω1 , the set {α ∈ ω1 : yα = y↾α } is stationary in ω1 .
Identifying ω1 with the ordinalQ ω · ω1 , we conclude that ♦ implies the existence of a
transfinite sequence (zα )α∈ω1 ∈ α∈ω1 (2ω )α such that for any z ∈ (2ω )ω1 the set {α ∈ ω1 :
zα = z↾α } is stationary in ω1 .
 A SIMPLE EFIMOV SPACE WITH SEQUENTIALLY-NICE SPACE OF PROBABILITY MEASURES 15

For every α ∈ ω1 , the set Xα has cardinality |Xα | ≤ |2ω | and hence admits an injective
function fα : Xα → 2ω . For every β ∈ ω1 , consider the function
gβ : Xβ → (2ω )β , gβ : x 7→ (fα ◦ παβ )α∈β .
Claim 5.2. For every limit ordinal β ∈ ω1 the function gβ : Xβ → (2ω )β is injective.
Proof. Given any distinct elements x, x′ ∈ Xβ , use the subcontinuity of the spectrum Σ and
find an ordinal α < β such that παβ (x) 6= παβ (x′ ). The injectivity of the function fα : Xα → 2ω
implies that fα (παω1 (x)) 6= fα(παω1 (x′ )). Let prβα : (2ω )β → 2ω , prβα : x 7→ x(α), be the
projection of (2ω )β onto the α-th coordinate. It follows from
prβα ◦ gβ (x) = fα ◦ παβ (x) 6= fα ◦ παβ (x′ ) = prβα ◦ gβ (x′ )
that gβ (x) 6= gβ (x′ ), which means that the function gβ : Xβ → (2ω )β is injective. 
Q
Let (xα )α∈ω1 ∈ α∈ω1 Xα be any sequence such that for every α ∈ ω1 with zα ∈ gα [Xα ] we
have gα (xα ) = zα . We claim that the transfinite sequence (xα )α∈ω1 has the required property.
Fix any x ∈ Xω1 and consider the element z = gω1 (x) ∈ (2ω )ω1 . The choice of the transfinite
sequence (zα )α∈ω1 ensures that the set S = {α ∈ ω1 : z↾α = zα } is stationary in ω1 .
To see that the set {α ∈ ω1 : παω1 (x) = xα } is stationary, it suffices to show that it contains
the stationary set S ∩ ω1′ , where ω1′ is the closed unbounded set of all limit countable ordinals.
Given any limit ordinal β ∈ S, observe that zβ = z↾β = gω1 (x)↾β = (fα ◦ παω1 (x))α∈β =
gβ (πβω1 (x)) ∈ gβ [Xβ ] and hence zβ = gβ (xβ ). On the other hand,
zβ = z↾β = gω1 (x)↾β = gβ (πβω1 (x)).
Since gβ (xβ ) = zβ = gβ (πβω1 (x)), the injectivity of gβ implies xβ = πβω1 (x). Therefore,
S ∩ ω1′ ⊆ {α ∈ ω1 : παω1 (x) = xα } and the set {α ∈ ω1 : παω1 (x) = xα } is stationary. 
Now we apply Proposition 5.1 to subcontinuous λ-spectra in the category Comp of compact
Hausdorff spaces and their continuous maps. A λ-spectrum in the category Comp is a λ-
spectrum (Xα , παβ : α ≤ β < λ) consisting of compact Hausdorff spaces Xα and continuous
bounding maps παβ . An example of such spectrum is the (ω + 1)-spectrum (2α , παβ : α ≤ β ≤
ω1 ), where the cubes 2α are endowed with the Tychonoff product topology and the projection
maps παβ : 2β → 2α , παβ : x 7→ x↾α , are continuous.
A functor F : Comp → Comp in the category Comp is defined to be subcontinuous if
for any ordinal λ and any subcontinuous λ-spectrum (Xα , παβ : α ≤ β < λ) in the category
Comp, the spectrum (F Xα , F παβ : α ≤ β < λ) is subcontinuous. By [24, 2.3.3], the functor
of probability measures P : Comp → Comp is subcontinuous and so is the functor (·)ω :
Comp → Comp of taking the countable power. Then the composition P ω : Comp → Comp
of the functors P and (·)ω is a subcontinuous functor, too. This functor assigns to each
compact Hausdorff space X the countable power P ω (X) of the space P (X). To any continuous
function f : X → Y between compact Hausdorff space the function P ω assigns the continuous
function P ω f : P ω (X) → P ω (Y ), P ω : (µn )n∈ω 7→ (P f (µn ))n∈ω .
Corollary 5.3. Under ♦ there exists a transfinite sequence (µα )α∈ω1 ∈ α∈ω1 P ω (2α ) such
Q
that for every µ ∈ P ω (2ω1 ), the set {α ∈ ω1 : µα = P ω παω1 (µ)} is stationary in ω1 .
Proof. By the subcontinuity of the functor P ω , the (ω1 + 1)-spectrum (P ω (2α ), P ω παβ : α ≤
β ≤ ω1 ) is subcontinuous. For every countable ordinal α, the compact space P ω (2ω ) is
16 TARAS BANAKH AND SAAK GABRIYELYAN

metrizable and henceQhas cardinality ≤ c. Applying Proposition 5.1, we obtain a transfinite

sequence (µα )α∈ω1 ∈ α∈ω1 P ω (2α ) with the desired property. 

6. Proof of Theorem 1.1

Consider the (ω1 + 1)-spectrum (2α , παβ : α ≤ β ≤ ω1 } consisting of the Cantor cubes 2α
and projections παβ : 2β → 2α , παβ : x 7→ x↾α.
Assume ♦. By Corollary 5.3, there exists a transfinite sequence
Y
ω
(µαn )n∈ω α∈ω1 ∈ P (2α )


α∈ω1

ω
such that for any sequence of measures (µn )n∈ω ∈ P (2ω1 ) the set
{α ∈ ω1 : µαn = P παω1 (µn ) for every n ∈ ω}
is stationary in ω1 .
Let Λ be the set of all nonzero limit ordinals α ∈ ω1 such that (µαn )n∈ω is a sequence of
pairwise distinct Dirac measures that converges to the Dirac measure µα0 ∈ Xα ⊆ P (Xα ).
Let Ω be the set of all nonzero limit ordinals α ∈ ω1 such that limn∈ω supx∈2α µαn ({x}) = 0.
Observe that the sets Λ and Ω are disjoint.
Now we shall inductively construct families of sets (Iα )α∈ω1 , (Xα )α∈ω1 , (Aα )α∈ω1 , (Bα )α∈ω1
such that for every α ∈ ω1 , the following conditions are satisfied:
(1) if α ≤ ω, then Xα = Aα = 2α , Bα = ∅ and Iα = ω;
(2) if the ordinal α is limit, then Xα = γ<α (πγα )−1 [Xγ ];
T

(3) Iα is an infinite subset of ω;

(4) Xα , Aα , Bα are closed crowded subspaces of 2α and Xα 6= ∅;
(5) Xα = Aα ∪ Bα and |Aα ∩ Bα | ≤ 1;
(6) Xα+1 = (Aα × {0}) ∪ (Bα × {1}) ⊆ Xα × 2 ⊆ 2α+1 ;
(7) πγα [Xα ] = Xγ for every γ < α;
if α ∈ Λ and n∈ω supp(µαn ) ⊆
S
(8) S SXα , then α
α ⊆
n∈ω supp(µ 2n+1 ) Aα and n∈ω supp(µ2n+2 ) ⊆ Bα ;
α
S
(9) if α ∈ Ω and
ω n∈ω supp(µn ) ⊆ Xα , then for any sequence of measures (λn )n∈ω ∈
P (Xα+ω ) with P παα+ω (λn ) = µαn for all n ∈ ω, every accumulation point of the
sequence (λn )n∈Iα in P (Xα+ω ) is a nonatomic measure on the space Xα+ω .
To start the inductive construction, for every α ≤ ω, put Xα = Aα = 2α , Bα = ∅ and
Iα = ω. Assume that for some limit ordinal α and all γ < α we have constructed nonempty
closed subspacesTXγ ⊆ 2γ that have no isolated points and satisfy the inductive condition
(7). Put Xα = γ<α (πγα )−1 [Xγ ] and observe that Xα is crowded and satisfies the inductive
condition (7).
To define the other sets we consider the following three cases.
/ Λ ∪ Ω or n∈ω supp(µαn ) 6⊆ Xα . In this case put
S
Case 1: Assume that α ∈
Iα+n = ω, Xα+n = Aα+n = Xα × {0}n and Bα+n = ∅
for every n ∈ ω. It is easy to see that the families (Iβ )β<α+ω , (Xβ )β<α+ω , (Aβ )β<α+ω ,
(Bβ )β<α+ω satisfy the inductive conditions (1)–(9).
 A SIMPLE EFIMOV SPACE WITH SEQUENTIALLY-NICE SPACE OF PROBABILITY MEASURES 17

Case 2: Assume that α ∈ Λ and n∈ω supp(µαn ) ⊆ Xα . In this case the definition of the set Λ
S
ensures that the sequence (µαn )n∈ω consists S
of pairwise distinct Dirac
S measures that converge to
the Dirac measure µα0 in 2α . Since the sets n∈ω supp(µα2n+1 ) and n∈ω supp(µα2n+2 ) are closed
and disjoint in the zero-dimensional metric space Xα \supp(µα0 ), we can use [10, Theorem 6.2.7]
(as in the proof of Lemma 4.1) and find two disjoint clopen sets Vα , Wα in Xα \ supp(µα0 ) such
that
[ [
supp(µα2n+1 ) ⊆ Vα , supp(µα2n+2 ) ⊆ Wα , Vα ∩ Wα = ∅, Vα ∪ Wα = Xα \ supp(µα0 ).
n∈ω n∈ω
Since the space Xα \ supp(µα0 ) is crowded, so are its clopen subspaces Vα and Wα .
Consider the subsets Aα := Vα ∪ supp(µα0 ), Bα := Wα ∪ supp(µα0 ) of 2α , and observe that
they are closed in 2α and have no isolated points. Define
Xα+1 = (Aα × {0}) ∪ (Bα × {1}) ⊆ Xα × 2.
Set Iα := ω. For every ordinal n ∈ [1, ω), put
Iα+n := ω, Xα+n = Aα+n := Xα+1 × {0}n−1 ⊆ Xα × 2n and Bβ+n := ∅.
It is easy to see that the families (Iβ )β<α+ω , (Xβ )β<α+ω , (Aβ )β<α+ω and (Bβ )β<α+ω satisfy
the inductive conditions (1)–(9).
Case 3: Assume that α ∈ Ω and n∈ω supp(µα0 ) ⊆ Xα . In this case the definition of the set
S
Ω ensures that limn∈ω supx∈2α µαn (x) = 0. Since P (Xα ) is a compact metrizable space, there
exists an infinite set Jα ⊆ ω such that the sequence (µαn )n∈Jα converges to some measure
µ ∈ P (Xα ). If the measure µ is nonatomic, then, for every ordinal n < ω, we put
Iα+n = Jα , Xα+n = Aα+n = Xα × {0}n and Bα+n = ∅,
and observe that the families (Iβ )β<α+ω , (Xβ )β<α+ω , (Aβ )β<α+ω and (Bβ )β<α+ω satisfy the
inductive conditions (1)–(9).
Now we assume that the measure µ has atoms. Let Tµ = {x ∈ Xα : µ({x}) > 0} be the set
of all atoms of the measure µ. The additivity of the probability measure µ ensures that the
set Tµ is at most countable. Since the compact metrizable space Xα is crowded, the Gδ -subset
Gα := Xα \ Tµ of Xα is dense in Xα (by the Baire Theorem) and also crowded.
The set Tµ is at most countable and hence admits a well-order  such that for every a ∈ Tµ
the set ↓a = {x ∈ Tµ : x  a} is finite.
Using Lemma 4.1 inductively, for every a ∈ Tµ we can choose an infinite subset Jα,a of Jα
and a family (Z(a,s) )s∈2<ω of clopen subsets of Xα \{a} such that Z(a,∅) = Xα \{a}, Jα,a ⊆ Jα,b
for any b ∈ Tµ with b  a and for every s ∈ 2<ω the following conditions are satisfied:
(†) Z(a,sˆ0) ∪ Z(a,sˆ1) = Z(a,s) and Z(a,sˆ0) ∩ Z(a,sˆ1) = ∅;
(‡) for any clopen neighborhood U ⊆ Xα of a and each ε > 0 there exists an m ∈ ω such
1
that for every n ∈ Jα,a with n ≥ m we have µαn (U ∩ Z(a,s) ) < µ(U \ {a}) + 2|s| + ε.
∗
Choose any infinite set Iα ⊆ Jα such that Iα ⊆ Jα,a for every a ∈ Tµ .
Fix a bijective function ξ : ω → Tµ × 2<ω such that for any a ∈ Tµ and s < t in 2<ω we
have ξ −1 (a, s) < ξ −1 (a, t) (to construct such bijection it is sufficient to consider a partition of
ω onto |Tµ | infinite sets, and on every set of the partition to consider a copy of the bijection
Pk−1
2<ω → ω, t 7→ i=0 t(i)2i ). For every n ∈ ω, let γn : Gα → 2 be the characteristic function
of the clopen subset Zξ(n) ∩ Gα in Gα .
Now, for every n ∈ ω, we define the sets Xα+n , Aα+n and Bα+n as follows:
• Xα+n is the closure of the set {(z, γ0 (z), . . . , γn−1 (z)) : z ∈ Gα } in Xα × 2n ;
18 TARAS BANAKH AND SAAK GABRIYELYAN

• Aα+n is the closure of the set {(z, γ0 (z), . . . , γn−1 (z)) : z ∈ Gα \ Zξ(n) } in Xα × 2n ;
• Bα+n is the closure of the set {(z, γ0 (z), . . . , γn−1 (z)) : z ∈ Gα ∩ Zξ(n) } in Xα × 2n .
Since Zξ(n) ∩ Gα is a clopen set in Gα , the spaces Xα+n , Aα+n , Bα+n are crowded, being
closures of the the topological copies of the crowded spaces Gα , Gα \ Zξ(n) and Gα ∩ Zξ(n) .
This means that the condition (4) is satisfied.
Next, we show that the condition (5) is satisfied. For every n ∈ ω, the equality Xα+n =
Aα+n ∪ Bα+n holds by the definitions of the sets Xα+n , Aα+n , Bα+n . To see that |Aα+n ∩
Bα+n | ≤ 1, take any points (x0 , t0 ), (x1 , t1 ) ∈ Aα+n ∩ Bα+n ⊆ Xα+n ⊆ Xα × 2n , where
x0 , x1 ∈ Xα and t0 , t1 ∈ 2n . Find a ∈ Tµ and s ∈ 2<ω such that (a, s) = ξ(n). Taking into
account that all Z(a,s) are clopen in Xα \ {a}, the condition (†) implies
x0 , x1 ∈ παα+n (Aα+n ) ∩ παα+n (Bα+n ) ⊆ Gα \ Zξ(n) ∩ Gα ∩ Zξ(n) ⊆ {a},
where the closure is taken in the compact space Xα . Therefore x0 = a = x1 . Assuming that
t0 6= t1 , we can find an ordinal k ∈ n such that t0 (k) 6= t1 (k). Let ξ(k) = (b, t) for some
b ∈ Tµ and t ∈ 2<ω . If b 6= a, then either Zξ(k) or Xα \ Zξ(k) is a neighborhood of a in
Xα . By the definition of γk , in the first case we get t0 (k) = t1 (k) = 1, and in the second
case we get t0 (k) = t1 (k) = 0. But this contradicts the inequality t0 (k) 6= t1 (k), and hence
b = a. By the choice of the function ξ, either t ≤ s or t and s are incomparable in the tree
2<ω . If t ≤ s, then Zξ(n) ⊆ Zξ(k) and hence γk (Gα ∩ Zξ(n) ) = {1} and t0 (k) = t1 (k) = 1. If
s and t are incomparable, then Zξ(n) ∩ Zξ(k) = ∅ and γk (Gα ∩ Zξ(n) ) = {0}, which implies
t0 (k) = t1 (k) = 0. In both cases we get a contradiction with the choice of k. Therefore,
|Aα+n ∩ Bα+n | ≤ 1 and the condition (5) holds.
The conditions (6) and (7) follow from the definition of the spaces Xα+n+1 , Aα+n , Bα+n
and the inclusions γn (Gα \ Zξ(n) ) ⊆ {0} and γn (Gα ∩ Zξ(n) ) ⊆ {1} holding for all n ∈ ω.
Finally, we check the inductive condition (9). Choose a sequence of measures (λn )n∈ω ⊆
(P (Xα+ω ))ω such that P παα+ω (λn ) = µαn for all n ∈ ω. Let λ ∈ P (Xα+ω ) be an accumulation
point of the sequence (λn )n∈Iα . The convergence of the sequence (µαn )n∈Jα to the measure
µ and the continuity of the map P παα+ω : P (Xα+ω ) → P (Xα ) imply that P παα+ω (λ) = µ.
We should prove that the measure λ is nonatomic. To derive a contradiction, assume that
λ({b}) > 0 for some b ∈ Xα+ω . Consider the point a = παα+ω (b) and conclude that µ({a}) =
λ((παα+ω )−1 (a)) ≥ λ({b}) > 0 and hence a ∈ Tµ . By the regularity of the measure µ, there
exists a clopen neighborhood U ⊆ Xα of a such that µ(U \ {a}) < 14 λ({b}). Find l ∈ ω
such that 22l < 14 λ({b}). By the property (‡) and almost inclusion Iα ⊆∗ Jα,a , there exists an
m ∈ ω such that for every s ∈ 2l and every n ∈ Iα with n ≥ m we have
1 1
(6.1) µαn (U ∩ Za,s ) < µ(U \ {a}) + 2|s|
+ 2l
< 24 λ({b}).
Since limn∈ω supx∈2α µαn ({x}) = 0, we can replace m by a larger number, if necessary, and
assume additionally that µαn ({a}) < 41 λ({b}) for all n ≥ m.
Consider the continuous map
γ : Gα → Xα+ω ⊆ Xα × 2ω , γ : x 7→ x, (γn (x))n∈ω ,

and observe that the image γ[Gα ] is dense in Xα+ω . For every s ∈ 2l , consider the closed set
Ws = γ[Gα ∩ Z(a,s) ] in Xα+ω . We claim that the family {Ws }s∈2l is disjoint. Indeed, suppose
for a contradiction that for some s0 , s1 ∈ 2l , the intersection Ws0 ∩ Ws1 contains a point, say
x. Since Xα+ω is metrizable, there are two sequences {zn0 }n∈ω and {zn1 }n∈ω in Gα ∩ Z(a,s0 ) and
Gα ∩ Z(a,s1 ) , respectively, such that limn γ(zn0 ) = x = limn γ(zn1 ). Set m := ξ −1 (a, s0 ). Since,
 A SIMPLE EFIMOV SPACE WITH SEQUENTIALLY-NICE SPACE OF PROBABILITY MEASURES 19

by (†), Z(a,s0 ) ∩ Z(a,s1 ) = ∅, the definition of γm implies that γm (zn0 ) = 1 and γm (zn1 ) = 0 for
every n ∈ ω, and hence limn γ(zn0 ) 6= limn γ(zn1 ). This contradiction proves the claim. Then
the density of γ[Gα ] implies that the finite family {Ws }s∈2l covers the compact space Xα+ω .
This fact and the claim imply that {Ws }s∈2l is a disjoint clopen cover of Xα+ω .
Find a unique s ∈ 2l such that the clopen set Ws := γ(Gα ∩ Z(a,s) ) contains the point b.
Then the set W := Ws ∩ (παα+ω )−1 (U ) is a clopen neighborhood of the point b in Xα+ω . Since
λ is an accumulating point of the sequence (λn )n∈Iα , there exists n ∈ Iα with n ≥ m such
that λn (W ) > 34 λ({b}).
Taking into account that Z(a,s) is a closed subset of Xα \ {a}, we conclude that
Ws ⊆ (παα+ω )−1 {a} ∪ Z(a,s) and W ⊆ (παα+ω )−1 {a} ∪ (U ∩ Z(a,s) ) .

Now observe that

µαn (U ∩ Z(a,s) ) = µαn ((U ∩ Z(a,s) ) ∪ {a}) − µαn ({a})
> λn (παα+ω )−1 ((U ∩ Z(a,s) ) ∪ {a}) − 41 λ({b}) ≥ λn (W ) − 41 λ({b}) > 24 λ({b}),

which contradicts (6.1). This finishes the inductive construction.

After completing the inductive construction, consider the closed subspace
\
K= {x ∈ 2ω1 : x↾α ∈ Xα }
α∈ω1
of the compact space 2ω1 . Below we show that the compact space K satisfies all conditions
of Theorem 1.1.
Claim 6.1. The compact space K is simple and crowded.
Proof. The space K is simple by (4)–(7). To show that K is crowded, fix a point z ∈ K and
an arbitrary basic neighborhood U = {x ∈ K : x↾F = z↾F } of z, where F ⊆ ω1 . Choose
a countable ordinal α ∈ ω1 such that F ⊆ α. The space Xα is crowded (by (4)) and hence
contains a point xα ∈ Xα such that xα ∈ παω1 (U ) \ {παω1 (z)}. Then any x ∈ (παω1 )−1 (xα ) ∩ K
belongs to U \ {z}, witnessing that K is crowded. 
Claim 6.2. The space K is sequentially discrete.
Proof. To derive a contradiction, assume that K contains a nontrivial convergent sequence.
Then there exists a sequence {µn }n∈ω ⊆ K ⊆ P (K) of pairwise distinct Dirac
measures,
converging to the Dirac measure µ0 . The choice of the transfinite sequence (µαn n∈ω )α∈ω1
guarantees that the set
S := {α ∈ ω1 : µαn = P παω1 (µn ) for all n ∈ ω}
is stationary in ω1 , and hence it contains an infinite limit ordinal α ∈ S such that the function
ω → 2α ⊆ P (2α ), n 7→ P παω1 (µn ), is injective.
S Then the ordinal α belongs S to the set Λ and
the inductive condition (8) guarantees that n∈ω supp(µα2n+1 ) ⊆ Aα and n∈ω supp(µα2n+2 ) ⊆
Bα . Since the
ω1
sets Aα × {0} and Bα × {1} have disjoint closures in Xα+1 , the sequence
P πα+1 (µn ) n∈ω is divergent, which contradicts the convergence of the sequence (µn )n∈ω . 
Claim 6.3. The space K is a simple crowded Efimov space.
Proof. By a result of Koppelberg [19] (whose topological proof is given by Dow [7, Propo-
sition 1]), the compact space K, being simple, does not admit continuous surjective maps
onto [0, 1]ω1 . Since βω admits a continuous surjective map onto [0, 1]ω1 (because [0, 1]ω1 is
20 TARAS BANAKH AND SAAK GABRIYELYAN

separable), the Tietze–Urysohn extension theorem implies that K contains no copies of βω.
This result and Claims 6.1 and 6.2 imply that K is a simple crowded Efimov space. 
Claim 6.4. Every c0 -vanishing sequence (µn )n∈ω in P (K) has a subsequence that converges
to some nonatomic measure.
Proof. By Lemma 3.9, the c0 -vanishing property of (µn )n∈ω implies the existence of an ordinal
υ ∈ ω1 such that for every countable ordinal α ≥ υ the sequence of
measures (P παω1 (µn ))n∈ω
is c0 -vanishing. By the choice of the transfinite sequence (µαn )n∈ω α∈ω1 , the set
S = {α ∈ ω1 : µαn = P παω1 (µn ) for all n ∈ ω}
is stationary and hence contains a limit ordinal α ∈
S such that α ≥ υ. By the choice of
υ, the sequence of measures (µαn )n∈ω = P παω1 (µn ) n∈ω is c0 -vanishing and hence α ∈ Ω.
ω1
Let µ ∈ P (K) be any accumulation point of the sequence (µn )n∈Iα . Then P πα+ω (µ) is an
ω1

accumulation point of the sequence P πα+ω (µn ) n∈Iα . Since the ordinal α belongs to the set
ω1
Ω, the inductive condition (9) guarantees that the measure P πα+ω (µ) is nonatomic and so is
the measure µ. By Lemma 2.4, the space P (K) is first-countable at µ. Then there exists an
infinite set J ⊆ Iα such that the sequence (µk )k∈J converges to µ. 
Claim 6.4 and Lemma 3.7 imply
Claim 6.5. Every sequence in P (K) that converges to an atomic measure µ ∈ Pa (K) con-
verges to µ in norm.
Claim 6.6. The space Pna (K) is non-metrizable, first-countable, Čech-complete, sequentially
compact, and the set Pna (K) is dense in P (K).
Proof. The first-countability, Čech-completeness, and density of Pna (K) in P (K) follows from
Lemmas 2.4 and 2.2, respectively. Claim 6.4 implies that the space Pna (K) is sequentially
compact.
To show that Pna (K) is non-metrizable, suppose for a contradiction that Pna (K) is metriz-
able. Then Pna (K) being also sequentially compact must be compact. Since, by Lemma 2.2,
Pna (K) is dense in P (K) we obtain that Pna (K) = P (K), a contradiction. Thus Pna (K) is
not metrizable. 
Claim 6.7. The space P (K) is selectively sequentially pseudocompact but not sequentially
compact.
Proof. The space P (K) is selectively sequentially pseudocompact by Lemma 2.1 and Claim
6.6. Since K is a closed non-sequentially compact subspace of P (K), the space P (K) is also
non-sequentially compact. 
Claim 6.8. The Banach space C(K) has the Gelfand–Phillips property.
Proof. Corollary 2.2 of [3] states that if P (K) is selectively sequentially pseudocompact,
then C(K) has the Gelfand–Phillips property. It remains to note that P (K) is selectively
sequentially pseudocompact by Claim 6.7. 

7. Acknowledgements
The first author expresses his sincere thanks to Grzegorz Plebanek for an inspiring discusion
on the construction of a simple Efimov space from his joint paper with Mirna Džamonja [9].
 A SIMPLE EFIMOV SPACE WITH SEQUENTIALLY-NICE SPACE OF PROBABILITY MEASURES 21

References
[1] T. Banakh, The topology of spaces of probability measures, I: functors Pτ and P̂ . Matematychni Studii 5
(1995), 65–87; (English transl.: arxiv.org/pdf/1112.6161.pdf).
[2] T. Banakh, The topology of spaces of probability measures, II: barycenters of probability Radon mea-
sures and metrization of the functors Pτ and P̂ . Matematychni Studii 5 (1995), 88-106; (English transl.:
arxiv.org/pdf/1206.1727.pdf).
[3] T. Banakh, S. Gabriyelyan, Banach spaces with the (strong) Gelfand–Phillips property, preprint
(arxiv.org/abs/2110.07991).
[4] V.I. Bogachev, Weak convergence of measures, AMS, Mathematical survays and monographs v. 234, 2018.
[5] A. Dorantes-Aldama, D. Shakhmatov, Selective sequential pseudocompactness, Topology Appl. 222 (2017),
53–69.
[6] L. Drewnowski, G. Emmanuele, On Banach spaces with the Gelfand–Phillips property, II, Rend. Circ. Mat.
Palermo 38 (1989), 377–391.
[7] A. Dow, Efimov spaces and the splitting number, Topology Proc. 29:1 (2005), 105–113.
[8] A. Dow, R. Pichardo-Mendoza, Efimov spaces, CH, and simple extensions, Topology Proc. 33 (2009),
277–283.
[9] M. Džamonja, G. Plebanek, On Efimov spaces and Radon measures, Topology Appl. 154:10 (2007), 2063–
2072.
[10] R. Engelking, General Topology, Heldermann Verlag, Berlin, 1989.
[11] V.V. Fedorchuk, A bicompactum whose infinite closed subsets are all n-dimensional, (Russian) Mat. Sb.
(N.S.) 96(138) (1975), 41–62 (in Russian).
[12] V.V. Fedorchuk, Completely closed mappings, and the compatibility of certain general topology theorems
with the axioms of set theory (Russian) Mat. Sb. (N.S.) 99 (141) (1976), 3–33.
[13] V.V. Fedorčuk, A compact space having the cardinality of the continuum with no convergent sequences,
Math. Proc. Cambridge Phil. Soc. 81 (1977), 177–181.
[14] V.V. Fedorchuk, Probability measures in Topology, Uspehi Mat. Nauk 46 (1991) 41–80 (in Russian);
English transl.: Russian Math. Surveys 46 (1991), 45–93.
[15] R. Jensen, The fine structure of the constructible hierarchy. With a section by Jack Silver. Ann. Math.
Logic 4 (1972), 229–308.
[16] K.P. Hart, Efimov’s problem, in: Open Problems in Topology, II (E. Pearl, ed.), 171–177, Elsevier, 2007.
[17] R. Haydon, On dual L1 -spaces and injective bidual Banach spaces, Isr. J. Math. 31 (1972), 142–152.
[18] H. Jarchow, Locally Convex Spaces, B.G. Teubner, Stuttgart, 1981.
[19] S. Koppelberg, Minimally generated Boolean algebras, Order, 5:4 (1989), 393–406.
[20] J. Lindenstrauss, Weakly compact sets, their topological properties and Banach spaces they generate, Proc.
Symp. Infinite Dim. Topology, 1967, Annals of Math. Studies, Princeton Univ. Press, 1972.
[21] P.J. Nyikos, Classic problems—25 years later. I. Proceedings of the Spring Topology and Dynamical
Systems Conference (Morelia City, 2001). Topology Proc. 26:1 (2001/02), 345–356.
[22] R. Pol, Note on the spaces of regular probability measures whose topology is determined by countable
subsets, Pacific J. Math. 100 (1982), 185–201.
[23] T. Schlumprecht, Limited sets in Banach spaces, Dissertation, Univ. Munich, 1987.
[24] A. Teleiko, M. Zarichnyi, Categorical Topology of compact Hausdorff spaces, VNTL Publ., Lviv, 1999.

Ivan Franko National University of Lviv (Ukraine) and Jan Kochanowski University in Kielce
(Poland)
Email address: t.o.banakh@gmail.com

Department of Mathematics, Ben-Gurion University of the Negev, Beer-Sheva, P.O. 653,

Israel
Email address: saak@math.bgu.ac.il