A CHAIN CONDITION FOR OPERATORS FROM C(K)-SPACES

KLAAS PIETER HART, TOMASZ KANIA AND TOMASZ KOCHANEK

arXiv:1211.2770v2 [math.FA] 15 Feb 2013

Abstract. We introduce a chain condition (B), defined for operators acting on C(K)-
spaces, which is intermediate between weak compactness and having weakly compactly
generated range. It is motivated by Pelczyński’s characterisation of weakly compact oper-
ators on C(K)-spaces. We prove that if K is extremally disconnected and X is a Banach
space then, for an operator T : C(K) → X, T is weakly compact if and only if T satis-
fies (B) if and only if the representing vector measure of T satisfies an analogous chain
condition. As a tool for proving the above-mentioned result, we derive a topological
counterpart of Rosenthal’s lemma. We exhibit several compact Hausdorff spaces K for
which the identity operator on C(K) satisfies (B), for example both locally connected
compact spaces having countable cellularity and ladder system spaces have this property.
Using a Ramsey-type theorem, due to Dushnik and Miller, we prove that the collection
of operators on a C(K)-space satisfying (B) forms a closed left ideal of B(C(K)).

1. Introduction
The aim of this paper is to study a certain chain condition, denoted (B), defined for
(bounded, linear) operators T : C(K) → X, where K is a compact Hausdorff space, X is
an arbitrary Banach space and C(K) is the Banach space of all scalar-valued continuous
functions on K, equipped with the supremum norm. The main motivation behind our
work comes from a theorem of Pelczyński which asserts that an operator T : C(K) → X
is weakly compact if and only if there is no isomorphic copy of c0 in C(K) on which
T is bounded below. In other words, T fails to be weakly compact provided there is
a sequence of pairwise disjoint open sets {On }∞n=1 in K and a uniformly bounded family
of functions {fn }∞n=1 in C(K) such that the support of fn is contained in On (n ∈ N) and
inf n kT (fn )k > 0. Loosely speaking, Pelczyński’s theorem characterises weakly compact
operators from C(K)-spaces as precisely those which annihilate the sequences of disjointly
supported and uniformly bounded functions in C(K). Our condition (B) is similar in
nature to this requirement.
Let K be a compact Hausdorff space and X be a Banach space. The support, supp(f ),
of a function f ∈ C(K) is defined as the closure of the set {x ∈ K : f (x) 6= 0}. For
any pair of functions f, g ∈ C(K) we write f ≺ g whenever f 6= g and f (x) = g(x) for
each x ∈ supp(f ). The relation ≺ is a strict ordering on C(K). For a bounded and linear
operator T : C(K) → X, the chain condition (B) is defined as follows:

2010 Mathematics Subject Classification. Primary 46B50, 46E15, 37F20; Secondary 54F05, 46B26.
Key words and phrases. Weakly compact operator, chain condition, extremally disconnected space,
vector measure, Rosenthal’s lemma, partition lemma.
1
2 HART, KANIA AND KOCHANEK

(B) for each uncountable ≺-chain F in C(K) with supf ∈F kf k < ∞,

inf{kT f − T gk : f, g ∈ F, f 6= g} = 0.
Any uniformly bounded ≺-chain {fi }i∈I such that kfi − fj k > δ for i, j ∈ I, i 6= j, and
some positive δ, will be called a δ-≺-chain. We also say that a compact Hausdorff space K
satisfies (B) whenever the identity operator on C(K) satisfies (B), equivalently, for each
δ > 0 there is no uncountable δ-≺-chain in C(K).
We start Section 2 with the observations that, for each operator T : C(K) → X, the
condition (B) is weaker than weak compactness (Proposition 2.1), and that these two
properties coincide provided that K is extremally disconnected (i.e. open sets have open
closures), in which case weak compactness may also be characterised by a counterpart of
condition (B) for the representing measure of T (Theorem 2.5). This result is obtained by
refining a classical lemma due to Rosenthal to the extremally disconnected setting (Lemma
2.2) and by applying some vector-measure techniques (Proposition 2.4).
In Section 3 we give examples of compact Hausdorff spaces that satisfy (B). They
include compact Hausdorf spaces which are countable, one-point compactifications of dis-
crete sets (Proposition 3.3), compact Hausdorff spaces which are locally connected and
have countable cellularity (Theorem 3.5) and ladder system spaces (Corollary 3.8).
Like other classical chain conditions for topological spaces or Boolean algebras, this one
carries some combinatorial insight. Using Ramsey-theoretic techniques we prove that the
family of operators on a given C(K)-space satisfying (B) forms a closed left ideal of the
Banach algebra B(C(K)) of all operators on C(K) (Theorem 4.2).
All Banach spaces are assumed to be over either real or complex scalars. By an operator
we understand a bounded linear operator acting between Banach spaces. An operator
T : E → F is bounded below whenever there exists γ > 0 such that kT xk > γkxk for each
x ∈ E; an operator which is bounded below is injective and has closed range. An operator
T : E → F fixes a copy of a Banach space X, if there is a subspace E0 of E isomorphic
to X such that T |E0 is bounded below. The identity operator on a Banach space X is
denoted IX . For a topological space K, we denote its cardinal number by |K|, and define
the cellularity of K, c(K), to be the supremum of the cardinalities of families consisting
of pairwise disjoint open sets in K. A topological space with countable cellularity is said
to satisfy c.c.c. The symbol 1A denotes the indicator function of a set A.

2. (B) and weakly compact operators from C(K)-spaces

By a classical result of Pelczyński (see, e.g., [3, Theorem VI.2.15]), every non-weakly
compact operator from a C(K)-space into an arbitrary Banach space X fixes a copy of c0 .
In the case where K is extremally disconnected the Goodner–Nachbin theorem ([8], [13])
says that C(K) is (isometrically) injective, and then Rosenthal’s theorem [15, Theorem
1.3] asserts that any non-weakly compact operator T : C(K) → X fixes a copy of ℓ∞ . The
main result of this section, Theorem 2.5, is in the spirit of these facts. It asserts that weak
compactness of an operator from a C(K)-space, for K extremally disconnected, may be
characterised in terms of the chain conditions, imposed both upon the operator itself and
its representing measure.
 A CHAIN CONDITION FOR OPERATORS FROM C(K)-SPACES 3

Let us begin with the following refinement of Pelczyński’s theorem.

Proposition 2.1. Let K be a compact Hausdorff space, X be a Banach space and let
T : C(K) → X be an operator. Then the following assertions are equivalent
(i) T is weakly compact;
(ii) for each infinite ≺-chain F ⊆ C(K),

inf{kT f − T gk : f, g ∈ F, f 6= g} = 0.

In particular, every weakly compact operator T : C(K) → X satisfies (B).

Proof. For the implication (i) ⇒ (ii) assume contrapositively that

δ = inf{kT f − T gk : f, g ∈ F, f 6= g} > 0.

Let F0 = (fn )n∈N be a ≺-monotone subsequence of F . For each n ∈ N set gn = |fn+1 −

fn |. Then Y = span{gn : n ∈ N} is an isomorphic copy of c0 and kT gn k > δ (n ∈ N).
Consequently, T is bounded below on Y . So T cannot be weakly compact.
To prove (ii) ⇒ (i) assume, for a contradiction, that T is not weakly compact. Appealing
to the proof of [3, Theorem VI.2.15], we may construct a sequence (fn )∞ n=1 of disjointly
supported functions in C(K) of norm at most one such that inf n∈N kT fn > 0. The family
k
 Pn ∞
F = k=1 fk n=1 is then a uniformly bounded infinite ≺-chain for which inf{kT f −
T gk : f, g ∈ F, f 6= g} > 0. 

Given a compact Hausdorff space K and a Banach space X, every operator T : C(K) →
X admits a Riesz-type representation (cf. [3, Chapter 6]) in the following precise sense:
there exists a w ∗ -countably additive vector measure µ : Σ → X ∗∗ (called the representing
measure for T ) on the σ-algebra Σ of all Borel subsets of K such that:
(i) for each x∗ ∈ X ∗ the map Σ ∋ A 7→ µ(A)x∗ is a regular countably additive scalar
measure (and will be denoted x∗ ◦ µ);
(ii) the map xR∗ 7→ x∗ ◦ µ from X ∗ into C(K)∗ is w ∗-to-w ∗ continuous;
(iii) x∗ T (f ) = K f d(x∗ ◦ µ) for each x∗ ∈ X ∗ and f ∈ C(K);
(iv) kT k = kµk(K).
The representing measure µ may be expressed explicitly by the formula µ(A) = T ∗∗ ϕA ,
where ϕA ∈ C(K)∗∗ acts as ϕA (ν) = ν(A) (A ∈ Σ, ν ∈ C(K)∗ ). Equivalently, it may
be defined by the prescription µ(A)x∗ = µx∗ (A), where µx∗ = T ∗ x∗ is the scalar measure
produced by the Riesz theorem applied to the functional x∗ T .
We will need the following topological counterpart of Rosenthal’s lemma (cf. [3, Lemma
I.4.1]). Although its proof is almost the same as the original one, we present it for com-
pleteness and to demonstrate the rôle played by extremal disconnectedness.
Lemma 2.2. Let K be a compact Hausdorff space which is extremally disconnected, and
let (Vn )∞ ∞
n=1 be a sequence of pairwise disjoint open subsets of K. Suppose that (µn )n=1 is
a sequence of scalar Borel measures on K having uniformly bounded variations. Then, for
every ε > 0 there exists a strictly increasing sequence (nk )∞
k=1 of natural numbers such that,
4 HART, KANIA AND KOCHANEK

for each k ∈ N, !
[
|µnk | Vnj < ε.
j6=k

Proof. Let us suppose, with no loss of generality, that |µn |(K) 6 1 for each n ∈ N. Consider
any sequence (Mp )∞ p=1 of pairwise disjoint infinite subsets of N such that N = p Mp . We
S

consider two cases.

Case 1. First, if there is some p ∈ N for which
!
[
|µk | Vj < ε for each k ∈ Mp ,
j∈Mp
j6=k

then we get the assertion from the induced enumeration of Mp : {n1 < n2 < · · · }.
Case 2. Now suppose that for every p ∈ N there is kp ∈ Mp such that
!
[
(2.1) |µkp | Vj > ε.
j∈Mp
j6=kp

Fix, for a moment, any p ∈ N. Since q Vkq is disjoint from the open set j∈Mp ,j6=kp Vj , so
S S

is its closure. Hence,

[ [ [
(2.2) Vj ⊂ Vn \ Vkq .
j∈Mp ,j6=kp n∈N q∈N

Observe that q Vkq is open, as K is extremally disconnected, whence n Vn \ q Vkq is closed.

S S S

Therefore, by (2.2), we get

[ [ [
(2.3) Vj ⊂ Vn \ Vkq .
j∈Mp n∈N q∈N
j6=kp

Obviously, we have
! !
[ [ [
|µkp | Vkq + |µkp | Vn \ Vkq 6 1,
q∈N n∈N q∈N

so (2.3) and (2.1) imply !

[
|µkp | Vkq 6 1 − ε,
q∈N
and this inequality is valid for every p ∈ N.
Consequently,
S we may repeat the same argument replacing the space K with the clopen
subspace q∈N Vkq (which is extremally disconnected as this property is inherited by open
subspaces) and the sequences (µn )∞ ∞ ∞ ∞
n=1 and (Vn )n=1 with (µkp )p=1 and (Vkp )p=1 , respectively.
By continuing this we would get subsequent upper bounds 1 − 2ε, 1 − 3ε, . . . for some of
 A CHAIN CONDITION FOR OPERATORS FROM C(K)-SPACES 5

the variations |µn |. Since this process has to terminate, we will end up with Case 1, where
the assertion has been proved. 
Proposition 2.3. Let K be a compact Hausdorff space which is extremally disconnected,
and let X be a Banach space. Every operator from C(K) into X which satisfies (B) is
weakly compact.
Proof. It suffices to consider the real space CR (K) since, we prove (arguing by contraposi-
tion) that there is a δ-≺-chain in CR (K) which will also do the job in CC (K).
Assume that T : CR (K) → X is a non-weakly compact operator, and let µ be its repre-
senting measure. Then (see the proof of Theorem 5.5.3 in [1]) there exist a number ε > 0,
a sequence (On )∞ ∗ ∞
n=1 of pairwise disjoint open subsets of K and a sequence (xn )n=1 in the
unit ball of X ∗ such that (x∗n ◦ µ)(On ) > ε for each n ∈ N (recall that x∗n ◦ µ = T ∗ x∗n ).
Since x∗n ◦ µ is a regular measure and K is extremally disconnected, we may assume that
each On is clopen. By Lemma 2.2, we may also assume that
!
[ ε
|x∗n ◦ µ| Oj < for each n ∈ N.
j6=n
2
Let {qn : n ∈ N} be an enumeration of the rational numbers and define At = {n : qn < t}
for t ∈ R. Then {At : t ∈ R} is an increasing chain of infinite subsets of N. For each t,
S
the set Ct = n∈At Ot is clopen, hence its characteristic function, call it ft , is continuous.
The family {Ct : t ∈ R} is a strictly increasing chain of clopen sets, so that s < t readily
implies fs ≺ ft . Clearly kft k = 1 for all t.
Now, let s < t. Then, there is n ∈ At \ As (in fact the set is infinite). Since
[ [
supp(ft − fs ) ⊂ Ok = On ∪ Ok
k∈N k6=n

and ft − fs equals 1 on On , we get

Z
kT (ft ) − T (fs )k > x∗n T (ft − fs ) = (ft − fs ) d(x∗n ◦ µ)
K
Z Z 
= + S (ft − fs ) d(x∗n ◦ µ)
On k6=n Ok
!
[
> (x∗n ◦ µ)(On ) − |x∗n ◦ µ| Ok
k6=n
ε ε
> ε− = (i, j ∈ N, i 6= j)
2 2
which proves that T does not satisfy (B).

We now introduce a counterpart of condition (B) for vector measures. Namely, for a set
algebra Σ, Banach space X, and (finitely) additive vector measure µ : Σ → X, consider
the following property:
6 HART, KANIA AND KOCHANEK

(R) for each uncountable chain (with respect to inclusion) {Ei }i∈I in Σ,
inf{kµ(Ei ) − µ(Ej )k : i, j ∈ I, i 6= j} = 0.
A vector measure µ : Σ → X is called stronglyP additive provided that for every sequence
(En )n=1 ⊂ Σ of pairwise disjoint sets, the series ∞
∞
n=1 µ(En ) is unconditionally convergent
in X.
Proposition 2.4. Let Σ be a σ-algebra of sets and X be a Banach space. A bounded vector
measure µ : Σ → X satisfies (R) if and only if it is strongly additive.
Proof. (⇐) Suppose that µ does not satisfy (R). Then there exists a monotone (with
respect to inclusion) sequence (En )∞
n=1 ⊂ Σ such that kµ(Em ) − µ(En )k > δ for m, n ∈ N,
m 6= n, and for some δ > 0. Put Dn = En+1 \ En provided (En )∞ n=1 is increasing and
∞
Dn = En \ En+1 otherwise. In any case (Dn )n=1 forms a sequence of pairwise disjoint
elements from Σ such that kµ(Dn )k > δ for each n ∈ N. Consequently, µ is not strongly
additive.
(⇒) Now, suppose that µ fails to be strongly additive. Then the Diestel–Faires theorem
(cf. [3, Theorem I.4.2]) produces a closed subspace Y of X and an isomorphism T : ℓ∞ → Y
such that T (en ) = µ(An ) for some pairwise disjoint sets (An )∞ n=1 ⊂ Σ. For any n ∈ N we
−1 ∗
have 1 = ken k 6 kT k · kµ(An )k, hence for some xn in the unit ball of X ∗ we have
x∗n µ(An ) > kT −1k−1 . Since (x∗n µ)∞ n=1 is a uniformly bounded sequence of scalar measures,
Rosenthal’s lemma produces a subsequence (x∗nk µ)∞ k=1 such that
[  1
|x∗nk µ| Anj < kT −1 k−1 for each k ∈ N.
j6=k
2
Let C be an uncountable chain of subsets of N and for each C ∈ C define E(C) =
j∈C Anj . Plainly, {E(C)}C∈C is an uncountable chain of members of Σ. Moreover, for
S

any C1 , C2 ∈ C with C1 ( C2 , and any k ∈ C2 \ C1 , we have

 [ 
kµ(E(C2 )) − µ(E(C1))k = kµ(E(C2 \ C1 ))k = µ Anj
j∈C2 \C1
 [  [ 
(2.4) > x∗nk µ Anj > |x∗nk µ(Ank )| − |x∗nk µ| Anj
j∈C2 \C1 j6=k
1 1
> kT −1 k−1 − kT −1 k−1 = kT −1 k−1 .
2 2
This shows that µ fails to satisfy (R), so the proof is complete. 
We are now prepared to proceed to the main result of this section.
Theorem 2.5. Let K be a compact Hausdorff space which is extremally disconnected, and
X a Banach space, let Σ be the Borel σ-algebra of K and let T : C(K) → X be an operator
with representing measure µ : Σ → X ∗∗ . Then the following assertions are equivalent:
(i) T is weakly compact;
(ii) µ is strongly additive;
 A CHAIN CONDITION FOR OPERATORS FROM C(K)-SPACES 7

(iii) T satisfies (B);

(iv) µ satisfies (R).
Proof. The equivalence of clauses (i) and (ii) is the Bartle–Dunford–Schwartz theorem
(cf. [3, Theorem VI.2.5]) and is valid for any compact Hausdorff space K. The equivalence
of clauses (i) and (iii) follows from Propositions 2.1 and 2.3. The equivalence of (ii) and
(iv) is immediate form Proposition 2.4. 
It is easily seen that the implication (⇐) of Proposition 2.4 holds true for any set algebra
Σ, not necessarily a σ-algebra. However, the situation is not so clear for the implication
(⇒), so the natural question arises: how can one characterise the class of set algebras Σ for
which every vector measure µ : Σ → X satisfying (R) is strongly additive? An inspection of
the proof of Proposition 2.4 suggests that some kind of interpolation property of Σ could do
the job, so we may make our problem more precise. A set algebra Σ has the subsequential
completeness property (SCP) whenever for every sequence (En )∞ n=1 of pairwise disjoint sets
∞
from Σ there is a subsequence (Enk )k=1 with k Enk ∈ Σ. (Haydon constructed a set algebra
S

with (SCP) which is not a σ-algebra; see [11, Proposition 1E].)

Question 2.6. Suppose that a set algebra Σ has (SCP). Is it true that, for any Banach
space X, every vector measure µ : Σ → X satisfying (R) is necessarily strongly additive?
Remark 2.7. The second and third-named authors ([12]) have recently studied the oper-
ator ideal W CG of weakly compactly generated operators, that is, operators whose range is
contained in a weakly compactly generated subspace of their codomain. (A Banach space is
weakly compactly generated whenever it contains a linearly dense weakly compact subset.)
The class W CG contains all weakly compact operators and all operators having separable
range, but in contrast, a weakly compactly generated operator defined on a C(K)-space
need not satisfy (B) (cf. Proposition 2.1).
To see this, consider the ordinal interval K = [0, ω1] equipped with the order topology
which makes it a compact Hausdorff space. Let D = {0} ∪ {α + 1 : α < ω1 }. Define
a mapping ϕ : [0, ω1] → [0, ω1 ] by

α + 1, if α ∈ D,
ϕ(α) =
α, if α ∈ [0, ω1] \ D.
Plainly, ϕ is continuous, hence the operator Cϕ : C[0, ω1 ] → C[0, ω1] defined by Cϕ f = f ◦ϕ
is bounded.
We note that T = IC[0,ω1 ] − Cϕ maps the Schauder basis {1[0,α] }06α6ω1 of C[0, ω1 ] onto
the set {1{α} }α∈D ∪ {0}. Consequently, the range of T is isomorphic to c0 (ω1 ), which is
a weakly compactly generated Banach space, so that T ∈ W CG (C[0, ω1 ]). On the other
hand, T maps the 1-≺-chain {1[0,α] }α∈D onto {1{α} }α∈D , so T fails (B).
It is natural to ask whether the condition (B) is the same as a seemingly weaker similar
statement (Bwo ) in which the δ-≺-chains are assumed to be well-ordered. This is, however,
not the case, as we shall see using the following observation.
Proposition 2.8. Let K be a compact Hausdorff space satisfying c.c.c. Then the identity
map IC(K) satisfies (Bwo ).
8 HART, KANIA AND KOCHANEK

Proof. Let K be a compact Hausdorff space which does not satisfy the condition (Bwo ).
With no loss of generality, we may assume that, for some δ > 0, there exists an uncountable,
well-ordered δ-≺-chain {f
α }α<ω1 ⊂ C(K). For each ordinal α < ω1 set gα = |fα+1 − fα |.
−1
The family {gα (δ, ∞) : α < ω1 } consists of uncountably many pairwise disjoint open
sets, so K fails the c.c.c. condition. 
Corollary 2.9. The condition (Bwo ) is strictly weaker than (B).
Proof. In the light of Proposition 2.8, it is sufficient to exhibit an example of a compact
Hausdorff space which is c.c.c., yet fails (B).
Let K be the double arrow space, that is, K is the set ((0, 1] × {0}) ∪ ([0, 1) × {1})
endowed with the order topology arising from the lexicographic order. It is well known
that K is a first-countable, non-metrisable, compact Hausdorff space which is separable
(hence also c.c.c.), see [6, 3.10.C]. We note that for each 0 < b < 1 the order-interval
Ib := (0, 1), (b, 0) is clopen in K, which means that its characteristic function fb = 1Ib
is continuous. Consequently, {fb }0<b<1 is an uncountable 1-≺-chain, so K fails (B), as
desired. 
3. Compact Hausdorff spaces enjoying (B)
In this section we analyse certain compact spaces to show that the identity operator on
a C(K)-space can satisfy (B). We note that there is no loss of generality in considering
only ≺-chains consisting of non-negative functions. Indeed, if f ≺ g then |f | ≺ |g|. In the
remainder of this section, we shall therefore assume that every ≺-chain has this property.
For a ≺-chain F , we denote by A(F, f ) the set of predecessors of f , that is, A(F, f ) =
{g ∈ F : g ≺ f }; furthermore we set
[
Z(F, f ) = {x ∈ K : f (x) 6= 0} \ {supp(g) : g ∈ A(F, f )}.
Lemma 3.1. Let K be a compact Hausdorff space, let δ > 0 and let F ⊂ C(K) be a
δ-≺-chain. Then, for each f ∈ F , Z(F, f ) is nonempty; in particular there is x ∈ Z(F, f )
such that f (x) > δ.
Proof. This is clear if f has a direct predecessor g in F ; in that case
Z(F, f ) = {x ∈ K : x ∈
/ supp(g) and f (x) > 0}.
As kf − gk > δ there must be x ∈ K \ supp(g) with f (x) > δ; hence x belongs to Z(F, f ).
In case where f does not have a direct predecessor, we take an increasing and cofi-
nal (possibly transfinite) sequence {gα : α < θ} in A(F, f ). For each α we pick xα ∈
supp(gα+1 ) \ supp(gα ) such that gα+1 (xα ) > δ. Note that this implies that f (xα ) > δ for
all α; therefore f (x) > δ for all x ∈ L, where
\
L= {xβ : β > α}.
α<θ

On the other hand, if g ≺ f , then g ≺ gα for some α and then g(xβ ) = 0 for β > α and
hence g(x) = 0 for x ∈ L. This yields that the intersection L ∩ supp(g) is empty for all
g ∈ A(F, f ) and hence that L ⊆ Z(F, f ). 
 A CHAIN CONDITION FOR OPERATORS FROM C(K)-SPACES 9

Corollary 3.2. Every compact Hausdorff space which is countable satisfies (B).
For an order type α we denote by α∗ the reverse of α.
Proposition 3.3. Let K be the one-point compactification of a discrete space Γ, let δ > 0,
and let a δ-≺-chain F ⊂ C(K) be given. Then there are α, β 6 ω such that the order type
of F is α + β ∗
Proof. Let F be a δ-≺-chain. Split F into F1 = {f : f (∞) = 0} and F2 = {f : f (∞) > 0}.
(Recall that we are assuming that F consists of non-negative functions only.) Note that
f ≺ g whenever f ∈ F1 and g ∈ F2 .
For f ∈ F1 set Υf = {x : f (x) > δ}; this set is finite. If f ≺ g in F1 then Υf is a proper
subset of Υg . This implies that the order type of F1 is at most ω.
Likewise, for f ∈ F2 set Υf = {x : f (x) = 0}, which is again a finite set. In this case
Υf contains properly Υg whenever f ≺ g in F2 . It follows that F2 is order-isomorphic to a
subset of {−n : n ∈ ω} ⊆ Z. 
We have now arrived at one of the main results of this section. We shall prove that local
connectedness is a sufficient condition for absence of uncountable δ-≺-chains of functions
on spaces satisfying c.c.c. A topological space is locally connected if each point has a
neighbourhood basis consisting of connected sets. The disjoint union of finitely many
copies of the unit interval is an easy example of a (linearly ordered) compact space which
is locally connected, but not connected.
Theorem 3.4. Let K be a compact Hausdorff space which is locally connected and let
δ > 0. The cardinality of any δ-≺-chain in K does not exceed the cellularity of K.
Proof. Let F = {fi }i∈I ⊂ C(K) be a δ-≺-chain for some δ > 0. The sets Z(F, fi ) (i ∈ I)
are non-empty and pairwise disjoint. It is enough to prove that each Z(F, fi ) (i ∈ I) is
open, as this will immediately yield the inequality |I| 6 c(K).
Fix i ∈ I and x ∈ Z(F, fi ); our aim is to prove that x lies in the interior of Z(F, fi ).
Choose an open connected neighbourhood U ⊆ K of x such that f (y) > 12 f (x) for each
y ∈ U. We claim that U ∩ supp(fj ) = ∅ for each fj ≺ fi , which means that U ⊆ Z(F, fi ).
Suppose that this is not the case, that is, U ∩ supp(fj ) 6= ∅ for some fj ≺ fi . Because
supp(fj ) is the closure of the open set V = {w ∈ K : fj (w) 6= 0}, there must be y ∈
U ∩ supp(fj ) such that fj (y) 6= 0. Since U is connected, x ∈ U \ supp(fj ) and y ∈ U ∩ V ,
the set U intersects the boundary of V ; let z be an element of this intersection. Then,
z ∈ U, so fi (z) 6= 0. On the other hand, z ∈ supp(fj ), yet z does not belong to V , so
0 = fj (z) = fi (z), as fj ≺ fi ; a contradiction. 
Corollary 3.5. Every locally connected compact Hausdorff space which satisfies c.c.c. also
satisfies (B).
Proof. This is a direct reformulation of Theorem 3.4 for spaces having countable cellularity.

Remark 3.6. (i) Compact spaces which are connected need not be locally connected;
a standard example of such a space (which is also not path-connected) is the so-called
10 HART, KANIA AND KOCHANEK

topologist’s closed sine curve, that is, the graph of sin(1/x), x ∈ (0, 1] with the interval
{0} × [−1, 1] adjoined, endowed with the relative Euclidean topology. Nevertheless, it is
easily seen from the very definition of the order topology that every connected linearly
ordered compact space is also locally connected.
Evidently, the unit interval satisfies the assumptions of Corollary 3.5. The existence
of other linearly ordered connected (hence locally connected) examples of compact spaces
satisfying c.c.c. is equivalent to the negation of the Souslin hypothesis, SH, which is con-
sistent with and independent of ZFC. Namely, under ¬SH, one may consider the two-point
compactification of a Souslin line (recall that a Souslin line is a non-separable, linearly or-
dered, connected space without end-points which satisfies c.c.c.). Consistently, there may
exist non-homeomorphic Souslin lines.
(ii) There are consistent examples of non-metrisable, locally connected compact spaces
which are hereditarily Lindelöf and hereditarily separable (hence also c.c.c.) (see [7] and
[9]). Curiously, there may be no non-trivial convergent sequences in spaces satisfying the
assumptions of Corollary 3.5 as shown by van Mill under CH ([16]). Assuming Jensen’s
diamond principle ♦, one has an example of such a space which is moreover one-dimensional
(cf. [10]).
We refer to [5, Chapter b-11] for an exposition concerning local connectedness and further
examples.
Corollary 3.5 is optimal in the sense that there may exist uncountable ≺-chains, yet for
any δ > 0 there may be no uncountable δ-≺-chains.
Example 3.7. Let ∆ ⊆ [0, 1] be the ternary Cantor set. For each d ∈ ∆ set fd : [0, 1] →
[0, 1] by fd (x) = dist(x, ∆) · 1[d,1] (x). Each function fd (d ∈ ∆) is continuous and the family
{fd }d∈∆ forms an uncountable ≺-chain in C[0, 1]. On the other hand, it is not hard to see
that it is not a δ-≺-chain for any positive δ.
Now, we exhibit another example of a compact, totally disconnected Hausdorff space
which enjoys (B). This is a well-known construction in point-set topology, where it is
sometimes called a ladder system space.
We equip the ordinal number ω1 with a topology as follows:
• we declare zero and each countable successor number to be an isolated point;
• for each non-zero limit ordinal λ we choose a set {αn,λ : n < ω} ⊆ λ of order type
ω consisting of successor numbers and cofinal in λ (a ladder ); then we define basic
open neighbourhoods of λ to be of the form Uλ,m = {λ} ∪ {αn,λ : n > m} (m < ω).
Of course, the topology on ω1 depends on the choice of ladders but in any case the space
ω1 topologised in this manner is first countable, locally compact and Hausdorff.
Theorem 3.8. Let K be the one-point compactification of a ladder system space on ω1 ,
and let δ > 0. Then every δ-≺-chain in C(K) is countable. In particular, K satisfies (B).
Proof. We model the proof after that of Proposition 3.3: we take a δ-≺-chain F and divide
it into two sets: F1 = {f ∈ F : f (ω1 ) = 0} and F2 = {f ∈ F : f (ω1 ) > 0}.
 A CHAIN CONDITION FOR OPERATORS FROM C(K)-SPACES 11

Let Lλ denote the ladder associated to λ, and set L+ λ = Lλ ∪ {λ}. Note that a closed
subset of ω1 is compact if and only if it is covered by finitely many sets of the form L+λ and
possibly a finite number of isolated points.
We claim that F1 is countable. For f in F1 let Υf = {x ∈ ω1 : f (x) > δ}. Each Υf
is compact, hence countable. Lemma 3.1 implies that if g ≺ f , then Z(F, g) ∩ Υf 6= ∅.
As the sets Z(F, f ) and Z(F, g) are disjoint, this shows that A(F, f ) is countable for any
f ∈ F1 . Also, if g ≺ f in F1 , then Υg is a proper subset of Υf .
It suffices to show that F1 has countable cofinality. We assume, seeking of a contradiction,
S in F1 . Then the sets Υfα form a strictly
that {fα : α < ω1 } is a strictly increasing sequence
increasing sequence as well so that the union α<ω1 Υfα is uncountable.
Given α < ω1 let βα > α be such that
S
• γ6α Υfγ ⊆ βα , and
• there is a γ < βα such that x > α for some x ∈ Υfγ .
Next let C be a closed and unbounded set such that for each γ ∈ C we have βα < γ
whenever α < γ.
Let γ ∈ C be such that C ∩ γ has order type ω 2. We show that Υfγ is not compact. To
this end, we note that, by the above two items there is a point in Υfγ between η and ξ,
whenever η and ξ are consecutive elements of C ∩ γ. This shows that the order type of
Υfγ ∩ γ is at least ω 2 .
Let H be a finite subset of Λ. We show that the set Υfγ \ λ∈H L+
S
λ is infinite. This
follows from the following three observations
• if λ ∈ H ∩ γ, then L+ λ is bounded below by γ;
• Υfγ \ L+γ is cofinal in γ (because of the order types of both sets); and
• if λ ∈ H is larger than γ then L+ λ ∩ γ is finite.
Thus we see that Υfγ is not compact.
For f ∈ F2 we consider the compact set Υf = {x ∈ X : f (x) = 0}. Arguing similarly
as before we conclude that F2 is countable, yet the order of F2 is reversed in the following
sense: if f ≺ g in F2 , then Υf ⊇ Υg and there is x ∈ Υf \ Υg such that g(x) > δ. 
A compact Hausdorff space is Eberlein if it is homeomorphic to a weakly compact subset
of a Banach space. Countable compact Hausdorff spaces, the one-point compactification of
a discrete space (Proposition 3.3) as well as the unit interval (cf. Remark 3.6 (i) are classical
examples of Eberlein compact spaces. A space K is Eberlein if and only if the Banach
space C(K) is weakly compactly generated (cf. [2, Theorem 2]). Every weakly compactly
generated Banach space is necessarily Lindelöf in its weak topology. Wage observed that
no ladder system space K is Eberlein ([17]), while Pol proved that the Banach space C(K)
is Lindelöf in its weak topology ([14]). In the light of these observations, we raise the
following question.
Question 3.9. Does the class of compact Hausdorff spaces satisfying (B) contain all
Eberlein compact spaces? More generally, does every compact Hausdorff space K for
which the Banach space C(K) is Lindelöf in its weak topology satisfy (B)?
12 HART, KANIA AND KOCHANEK

4. The left ideal structure of operators satisfying (B)

Let [A]2 stand for the set of all 2-element subsets of a given set A. The Dushnik and Miller
partition lemma ([4]) asserts that, for every infinite regular aleph ℵα , and any colouring
c : [ℵα ]2 → {0, 1}, at least one of the following conditions holds true:
(i) there is a set A ⊂ ℵα with |A| = ℵα and [A]2 ⊂ c−1 {0};
(ii) for every set A ⊂ ℵα with |A| = ℵα there are a ∈ A and a set B ⊂ A such that
|B| = ℵα and {a, b} ∈ c−1 {1} for each b ∈ B.
Proposition 4.1. Let K be a compact Hausdorff space and let X be a Banach space. If
operators T, U : C(K) → X satisfy (B), then so does T + U.
Proof. Assume, in search of a contradiction, that there are δ > 0 and a ≺-chain {fi }i∈I in
C(K), where I has cardinality ℵ1 , such that supi∈I kfi k < ∞ and
(*) k(T + U)(fi ) − (T + U)(fj )k > δ (i, j ∈ I, i 6= j).
Set
I = {i, j} ∈ [I]2 : kU(fi ) − U(fj )k 6 δ/4


and consider the partition [I]2 = I ∪ ([I]2 \ I).

If assertion (i) from the partition lemma holds (where α = 1 and c−1 {0} = [I]2 \ I), then
we get a contradiction with the fact that U satisfies (B).
If assertion (ii) is valid (where c−1 {1} = I), then there is an i0 ∈ I and a set B ⊂ I with
|B| = ℵ1 , such that {i0 , j} ∈ I for each j ∈ B. Then for every j, k ∈ B we have
kU(fj ) − U(fk )k 6 kU(fi0 ) − U(fj )k + kU(fi0 ) − U(fk )k 6 δ/2,
hence our assumption (*) implies that kT (fj ) − T (fk )k > δ/2 for all j, k ∈ B, j 6= k, which
contradicts the condition (B) for T . 
Theorem 4.2. For every compact Hausdorff space K the set of all operators T ∈ B(C(K))
satisfying (B) forms a closed left ideal of the Banach algebra B(C(K)).
Proof. From the very definition of (B) it is evident that if (Tn )∞ n=1 ⊂ B(C(K)) norm
converges to some T ∈ B(C(K)) and each Tn satisfies (B), then T does as well.
Now, suppose that T ∈ B(C(K)) satisfies (B) and let S ∈ B(C(K)). If for some δ > 0
there were an uncountable chain {fi }i∈I with kST (fi ) − ST (fj )k > δ for all i, j ∈ I, i 6= j,
then for all such i, j we would also have kT (fi ) − T (fj )k > δ/kSk; a contradiction. Finally,
an appeal to Proposition 4.1 completes the proof. 
We have already noticed that for K extremally disconnected the family of operators
T : C(K) → C(K) which satisfy (B) coincides with the two-sided ideal of weakly compact
operators. Examples in Section 3 demonstrate that in some particular cases every operator
on a C(K)-space may satisfy (B), which means that the left ideal of operators satisfying
(B) is again a two-sided ideal, the improper ideal B(C(K)). This raises the following open
question.
Question 4.3. Is the set of operators on a C(K)-space which satisfy (B) always a right,
and hence a two-sided ideal of B(C(K))?
 A CHAIN CONDITION FOR OPERATORS FROM C(K)-SPACES 13

Acknowledgements. The authors would like to express their gratitude to N.J. Laustsen
and J.M. Lindsay for careful reading of preliminary versions of this paper.
References
1. F. Albiac and N. J. Kalton, Topics in Banach Space Theory, Graduate Texts in Mathematics 233,
Springer 2006.
2. D. Amir and J. Lindenstrauss, “The structure of weakly compact sets in Banach spaces”, Ann. of
Math. 88 (1968), 35–46.
3. J. Diestel and J. J. Uhl, Vector Measures. With a foreword by B.J. Pettis, Mathematical Surveys,
no. 15, American Mathematical Society, Providence, R.I. 1977.
4. B. Dushnik, E. W. Miller, “Partially ordered sets”, Amer J. Math. 63 (1941), 600–610.
5. Encyclopedia of General Topology, ed. K. P. Hart, J. Nagata, and J. E. Vaughan, Elsevier (2004).
6. R. Engelking, General Topology, 2nd edition, Sigma Ser. Pure Math., Vol. 6, Heldermann, Berlin
(1989).
7. V. V. Filippov, “Perfectly normal bicompacta” (Russian), Dokl. Akad. Nauk SSSR 189 (1969) 736–739;
English translation in Soviet Math. Dokl. 10 (1969) 1503–1507.
8. D. B. Goodner, “Projections in normed linear spaces”, Trans. Amer. Math. Soc. 69 (1950), 89–108.
9. K. Kunen, “Locally connected hereditarily Lindelöf compacta”. Topology Appl. 158 (2011), no. 18,
2473–2478.
10. K. Kunen and J. E. Hart, “One-dimensional locally connected S-spaces”, Topology Appl. 156 (2009),
no. 3, 601–609.
11. R. Haydon, “A non-reflexive Grothendieck space that does not contain l∞ ”, Israel J. Math. 40 (1981),
65–73.
12. T. Kania and T. Kochanek, “The ideal of weakly compactly generated operators acting on a Banach
space”, to appear in J. Operator Theory, http://arxiv.org/abs/1206.5424v3.
13. L. Nachbin, “A theorem of the Hahn–Banach type for linear transformations”, Trans. Amer. Math.
Soc. 68 (1950), 28–46.
14. R. Pol, “A function space C(X) which is weakly Lindelöf but not weakly compactly generated”, Studia
Math. 64 (1979), 279–285.
15. H. P. Rosenthal, “On relatively disjoint families of measures, with some applications to Banach space
theory”, Studia Math. 37 (1970), 13–36.
16. J. van Mill, “A locally connected continuum without convergent sequences”, Topology Appl. 126 (2002),
no. 1-2, 273–280.
17. M. L. Wage, Applications of the set theory to analysis and topology, Thesis, University of Wisconsin–
Madison, 1976.

Faculty of Electrical Engineering, Mathematics and Computer Science, TU Delft,

Postbus 5031, 2600 GA Delft, the Netherlands
E-mail address: [email protected]

Department of Mathematics and Statistics, Fylde College, Lancaster University, Lan-

caster LA1 4YF, United Kingdom
E-mail address: [email protected]

Institute of Mathematics, University of Silesia, Bankowa 14, 40-007 Katowice, Poland

E-mail address: [email protected]