CardAGTLaTeX.dvi


@
Applied General Topology

c© Universidad Politécnica de Valencia

Volume 5, No. 2, 2004

pp. 173-190

On the cardinality of indifference classes

Gerhard Herden and Andreas Pallack

Abstract. Let “ - ” be a continuous total preorder on some
topological space (X, t). Then the cardinality or at least lower and
upper bounds of the cardinality of the indifference (equivalence) classes
of “ - ” will be computed. In addition, the relevance of these bounds
in mathematical utility theory and the theory of orderable topological
spaces will be discussed.

2000 AMS Classification: 54F05, 91B16, 06A05.

Keywords: Indifference potency, calculable set, degree of connectedness,
path rank, separation rank.

1. Introduction

Let (X, t) be a topological space that is endowed with a total preorder “ - ”.
The reader may recall that a total preorder “ - ” on X is a reflexive and
transitive relation on X that satisfies the additional property that for all pairs
(x, y) ∈ X2 at least one of the relations x - y or y - x holds. The relation
x ∼ y ⇐⇒ x - y ∧ y - x defines an equivalence relation on X. The correspon-
ding eqivalence classes are the indifference classes of “ - ”. The order topology
t- on X that is induced by “ - ” is generated by the sets L(x) := {y ∈ X|y ≺ x}
and K(x) := {z ∈ X|x ≺ z} where x runs through X.

In mathematical utility theory usually continuous total preorders “ - ” on
X are considered, which means that the order topology t- is coarser than t.

Ideally, the indifference classes of a continuous total preorder “ - ” on (X, t)
only consist of single points. This means from the viewpoint of mathematical
utility theory that different alternatives are perfectly distinguishable by pre-
ferences (utilities) and from the viewpoint of pure mathematics that the given
topology t on X is orderable. The reader may notice that in our terminology
orderability only means that t- ⊂ t. If also t ⊂ t- then we speak of a strictly
orderable topology t on X (cf. Kok [12]). Of course, in general, indifference



174 G. Gerden and A. Pallack

classes are thick. This means that, in general, one cannot escape from the fact
that one has to deal with a preorder.

In general, a topology t on X is not orderable. Indeed, let n > 1 be any
natural number. Then the natural topology tnat on R

n is not orderable (cf.
Section 2 and the nice results in Beardon [2], Candeal and Induráin [6] and
Candeal, Induráin and Mehta [7]).

Clearly, a topology t on X is not orderable if and only if every continuous
total preorder “ - ” on X has at least one indifference class that contains
more than one point. In this case we may select the greatest of all these big
indifference classes. The smallest of these greatest classes then represents in
some sense the degree up to which t is orderable or not. In this way the concept
of an orderable topology t on X seems to be generalizable in a natural way. In
addition, by strengthening this idea we are able to measure in a precise sense
up to which degree alternatives can or cannot be distinguished by preferences
(utilities). Indeed, the fundamental concepts that are introduced and discussed
in Section 3 are based upon this idea.

In order to compute or to measure how big an indifference class [x] of “ - ”
can be two possibilities seem to be natural.

Indeed, let A be an appropriate σ-field in X that contains t and let µ be an
appropriate measure on A. Then for every indifference class [x] of “ - ” the
measure µ ([x]) might be computed. The reader may recall that the continuity
of “ - ” implies that every indifference class [x] of “ - ” is a closed subset of
X. Since t ⊂ A, therefore, µ ([x]) is defined.

In the Arrow-Hahn or Euclidean distance approach to mathematical utility
theory an indifference class [x] of “ - ” is considered as being thick if it contains
a non-empty open subset of X, which means that its Lebesgue measure is non-
zero. In order to rule out thick indifference classes “ - ”, thus, is required
to be locally non-satiated, which means that for every point x ∈ X and every
neighborhood U of x there exists some point y ∈ U such that x ≺ y. Besides
the original Arrow-Hahn approach [1] the reader may consult, in particular,
the book of Bridges and Mehta [5].

On the other hand, one simply could compute the cardinality |[x]| of every
indifference class [x] of “ - ”.

Let n ≥ 1 be a natural number. Usually on Rn the Lebesgue measure λ
is considered. One verifies immediately that there exist total preorders “ - ”
on Rn that are continuous with respect to tnat and only have indifference
classes [x] the Lebesgue measure λ ([x]) of which is zero. For any two points
x = (x1, ..., xn), y = (y1, ..., yn) ∈ R

n one may set, for instance, x - y ⇐⇒
x1 ≤ y1. With the exception of n = 1 in this case the indifference classes are
hyperplanes of Rn which still look very big. In addition, although for n > 1
the natural topology tnat on R

n is not orderable (cf. Section 2) this example
guarantees, however, the existence of continuous total preorders “ - ” on
Rn which, following the spirit of the first possibility, only have thin or small
indifference classes. This means that the first possibility is not appropriate
if one is interested in measuring or computing the degree of orderability of a



Cardinality of indifference classes 175

topology t or its potency of distinguishing between different alternatives by
preferences (utilities). Hence, in the remainder of this paper we mainly study
the second possibility and, thus, compute or at least estimate the cardinality
of indifference classes.

After introducing in Section 3 the basic concepts of this paper in Section
4 we, therefore, concentrate on the computation of the cardinality or at least
lower and upper bounds of the cardinality of indifference classes.

Meanwhile Hilbert spaces or Banach spaces or even general convex spaces are
commonly encountered in mathematical utility theory (cf. [3], [11], [13], [14]).
Thus, in particular, the degree of orderability or the potency of distinguishing
between different alternatives by preferences (utilities) in these spaces will be
computed (Proposition 4.6, Corollary 4.7 and Proposition 4.9). In addition, in
Remark 4.10 the (possible) relevance of these results in mathematical utility
theory will be discussed.

Finally, also a generalization of well known results on orderable connected
topologies will be proved (Proposition 4.11).

2. A first approach

A well known result of Candeal and Induráin [6, Theorem 4] states that for
n ≥ 1 a closed and convex subset C of Rn can be endowed with a continuous
total order if and only if C is homeomorphic and isotonic to a connected subset
of R. An immediate and often quoted consequence of this theorem of Candeal
and Induráin says that Rn for n > 1 cannot be endowed with a continuous
total order. Related but more general results also can be found in Beardon [2]
and Candeal, Induráin and Mehta [7].

The quoted consequence of the theorem of Candeal and Induráin also fol-
lows from the remarkable fact that for n > 1 each indifference class [x] of a
continuous total preorder “ - ” on Rn such that x is neither a first nor a
last element of “ - ”contains exactly 2ℵ0 elements. This result is interesting.
Indeed, let n > 1 and let S be a countable subset of Rn. Then a well known
result of Dugundji [8, Chapter 5, Theorem 2.2] states that Rn\S is connected.
As we shall soon see this result of Dugundji also implies the consequence of the
Candeal-Induráin theorem. On the other hand, if ℵ1 < 2

ℵ0 it does not imply
that any continuous total preorder on Rn contains at least one indifference class
that consists of 2ℵ0 elements.

In order to be more precise we are going to outline a proof of the fact that
for n > 1 each indifference class [x] of a continuous total preorder “ - ” on
Rn such that x neither is a first nor a last element of “ - ”contains exactly
2ℵ0 elements that relies on standard arguments from topology. Then it follows,
in particular, that the result of Dugundji does not necessarily imply that any
continuous total preorder on Rn contains at least one indifference class that
consists of 2ℵ0 elements.

Let n > 1 and a, b two different points of Rn. Then a family {fi}i∈I of
paths fi : [0, 1] −→ R

n that connect a and b is said to be separated if for any
pair i, j of different indexes of I the meets of the images Imfi and Imfj of fi



176 G. Gerden and A. Pallack

and fj respectively only consist of the points a and b. Now a straightforward
consideration (cf. the general construction in Proposition 4.6) allows us to
construct a separated family {fi}i∈I of paths fi : [0, 1] −→ R

n that connect

a and b and the cardinality of which is 2ℵ0. This means that the result of
Dugundji can be improved. Indeed, let n > 1 and let S be a subset of Rn such
that |S| < 2ℵ0. Then for every pair of different points a, b ∈ Rn\S there exists
at least one index i ∈ I such that Imfi ∩ S = ∅, and we may conclude that
Rn\S is path connected. In case that x is neither a first nor a last element
of “ - ” it follows that Rn\ [x] = {y ∈ Rn|y ≺ x} ∪ {z ∈ Rn|z ≻ x} is not
connected. Hence, we may conclude that |[x]| = 2ℵ0. Obviously, the original
result of Dugundji only implies that ℵ0 < |[x]| ≤ 2

ℵ0. In Section 4 the afore-
presented observations will be generalized to arbitrary real or complex convex
spaces (cf. Proposition 4.6, Corollary 4.7 and Corollary 4.8).

Let S be a non-empty subset of Rn. Then we choose an arbitrary vector
v ∈ S and postulate for the moment the dimension of S to be the dimension
of the linear subspace of Rn that is generated by S − {v} := {w − v|w ∈ S}.
The reader may verify that the definition of the dimension of S is independent
of any particular chosen vector v ∈ S. Let tnat be the natural topology on R

n.
With the help of our definition of the dimension of S the indifference dimension
of a continuous total preorder “ - ” on (Rn, tnat) or shortly R

n is defined as
the maximum of all dimensions of all indifference classes [(x)] of “ - ”. Then
the indifference dimension of Rn is defined as the minimum of all dimensions
of relations “ - ” where “ - ” runs through all continuous total preorders on
Rn. Of course, these concepts can be generalized to arbitrary real or complex
convex spaces (cf. Remark 4.10). In Proposition 4.9 these types of indiffer-
ence dimensions on Rn (at least implicitly) will be computed. In Remark 4.10
(possible) consequences in mathematical utility theory will be discussed. On
the other hand, these concepts of measuring the size or (thickness) of indif-
ference classes in Rn hardly seem to be generalizable to arbitrary topological
spaces. Hence, in general topology we restrict on computing the cardinality of
indifference classes.

3. Fundamental concepts and inequalities

Let (X, t) be some arbitrarily but fixed chosen non-trivial topological space.
Non-trivial means that X contains at least two (different) elements.

According to our considerations in the introduction for every continuous
total preorder “ - ” on X the weak indifference potency of “ - ” is defined by

winpot(-) :=















0
if “ - ” is an
order on X

sup{| [x] | | [x] is an indifference class of “ - ”, otherwise

.

Then the weak indifference potency of X is defined by



Cardinality of indifference classes 177

winpot(X) := min{winpot(-)|“ - ” is a continuous total preorder on (X, t)}.

In Section 2 the cardinality of an indifference class has been computed with
the help of its potency to separate a (connected) component of X. Let, there-
fore, C be the set of all (connected) components of X. We consider some con-
tinuous total preorder “ - ” on X and choose a component C ∈ C. Now an
indifference class [x] of “ - ” is said to be calculable with respect to “ -|C ”
if C ⊂ [x] or there exist points y, z ∈ C such that y ≺ v ≺ z for every point
v ∈ C ∩ [x].This notation allows us to define the indifference potency inpot(-)
of “ - ” by “0” if “ - ” is an order on X and by

supC∈Cmin{|[x]| [x] is a calculable set with respect to “ -|C ”}, otherwise.

Then the indifference potency of X is defined by

inpot(X) := min{inpot(-)|“ - ” is a continuous total preorder on (X, t)}.

Finally, a subset S of X is said to be calculable if there exists a component
C of X and some continuous total preorder “ - ” on X such that S is an
indifference class of “ - ” and C ⊂ S or there exist points y, z ∈ C such that
y ≺ x ≺ z for every point x ∈ C ∩ S. Let now for every component C ∈ C the
set ca(C) consist of all subsets S of X that are indifference classes of continuous
total preorders “ - ” on X such that C ⊂ S or there exist points y, z ∈ C such
that y ≺ x ≺ z for every point x ∈ C ∩ S. Then we still may define the strong
indifference potency of X by

stinpot(X) :=

{

0, if t is orderable
supC∈C min {|S| |S ∈ ca(C)} otherwise

.

The above definitions imply that the assertions winpot(X) = inpot(X) =
stinpot(X) = 0 and t is orderable are equivalent.

The following lemma justifies the concept of a calculable set.

Lemma 3.1. Let S be a subset of X. Then the following assertions are equiv-
alent:

(i) S is calculable.
(ii) There exists a continuous total preorder “ - ” on X and some compo-

nent C of X such that S is an indifference class of “ - ” and C ⊂ S
or C\S is not a connected subset of X with respect to t-.

Proof. (i) =⇒ (ii): Let S = [x] and let there exist points y, z ∈ C such that

y ≺ x ≺ z. It suffices to verify that C\ [x] is not connected with respect to t-.
Let, therefore, L(x) := {u ∈ X|u ≺ x} and K(x) := {v ∈ X|x ≺ v}. Then L(x)

and K(x) are two disjoint open subsets of
(

X, t-
)

such that L(x) ∩ C\S 6= ∅



178 G. Gerden and A. Pallack

and K(x) ∩ C\S 6= ∅ and C\S ⊂ L(x) ∪ K(x), which implies that C\S is not
connected with respect to t-.

(ii) =⇒ (i): Let “ - ” be some continuous total preorder on X such that
S = [x] for an indifference class [x] of “ - ”. Then we assume the existence of

a component C of X such that C\ [x] is not connected in the order topology t-

on X that is induced by “ - ”. Because of the definition of a calculable set it is
sufficient to prove that there exist points y, z ∈ C such that y ≺ x ≺ z. Since
t- ⊂ t and C is a connected subset of X with respect to t it follows that C
is also a connected subset of X with respect to t-. In addition, the definition
of t- implies that every indifference class [u] of “ - ” is a connected subset

of X with respect to t-. Hence, I := {[v] C ∩ [v] 6= ∅} is a connected subset
of X with respect to t-, which means that I is an interval of (X, -) that is

connected with respect to t-. Furthermore, the reader may recall that a totally
preordered set (Z, .) is connected with respect to its order topology if and only
if it is order dense and Dedekind complete. Order dense means that for any two
points a < b ∈ Z there exists some point c ∈ Z such that a < c < b. Dedekind
complete means that every bounded subset D of Z has a least upper and a
greatest lower bound. These results are well known for chains. The reader
may consult, for instance, the famous book of Birkhoff [4] on lattice theory. By
considering on the set of indifference classes the induced total ordering these
results easily can be generalized to total preorders. Since C\ [x] is not connected

with respect to t- we may summarize our considerations for concluding that
x neither can be a first nor a last element of (I, -). Therefore, the definition
of I guarantees the existence of points y, z ∈ C such that y ≺ x ≺ z, and the
proof is complete. �

Remark and Example 3.2. Let “ - ” be a continuous and total preorder
on X and let C be a component of X. With the help of the proof of Lemma
3.1 it follows that an indifference class [x] of “ - ” is calculable with respect
to ” -|C ” if and only if C ⊂ [x] or C\ [x] is not a connected subset of X with

respect to t-.
In assertion (ii) of Lemma 3.1 the condition that C\S is not a connected

subset of X with respect to t- cannot be replaced by the condition that C\S is
not a connected subset of X with respect to t. Indeed, let X := [0, 1] be the real
unit interval. We set U := [0, 1)∩Q and V := [0, 1)\Q and consider the coarsest
topology t on X that contains tnat and both sets U and V . Then {1} is an
indifference class of the natural total ordering “ ≤=≤|X ” on [0, 1]. In addition,

the definition of t imples that t≤ = tnat ⊂ t. Furthermore, since 1 neither is
contained in U nor in V it follows that (X, t) is a connected space. Moreover,
we may conclude with the help of the definitions of U and V that X\ {1} is
not connected with respect to t. Since X\{1} is connected with respect to t≤

it follows that {1} cannot be calculable with respect to ” ≤|X ” = ” ≤ ”. Now
the connectedness of t and the particular form of the neighborhoods of 1 also
imply that 1 is either a first or a last element of (X, -) for every continuous
total preorder “ - ” on X. Hence, there cannot exist any other continuous



Cardinality of indifference classes 179

total preorder “ - ” on X such that X\ {1} is not connected with respect to

t-, which means that {1} cannot be a calculable set.

In order to prepare the main result of this section the reader may recall that
(X, t) is said to be hereditarily Lindelöf if for every collection {Oi}i∈I of open
subsets of X there exists a countable subset J of I such that

⋃

i∈I

Oi =
⋃

j∈J

Oj.

In addition, we define for every component C ∈ C the degree of connected-
ness of C by

dcon(C) :=

{

min {|Z| |Z ∈ T (C)} , if T (C) 6= ∅
|C| , otherwise

.

With the help of this definition we may define the degree of connectedness
of X by

dcon(X) :=

{

0, if t is orderable
supC∈Cdcon(C), otherwise

.

The reader may compare the definitions of dcon(C) and dcon(X) with as-
sertion (ii) of Lemma 3.1 and the above given example in order to understand
the particular relevance of these concepts within our approach.

Now we still consider the set S(C) of all components C of X such that
dcon(C) < |C| and prove the following proposition that in combination with
Remark and Example 3.4 will help us to clarify in some degree the interrelations
between the afore-introduced concepts.

Proposition 3.3. The following assertions hold:

(i) winpot(X) ≥ inpot(X) ≥ stinpot(X) ≥ dcon(X).

(ii) Let every component C of X be open and closed and let every compo-
nent C of X be metrizable or regular and hereditarily Lindelöf. Then
inpot(X) = stinpot(X) = dcon(X).

(iii) Let every component C of X be open and closed and let S(C) = ∅.
Then winpot(X) = inpot(X) = stinpot(X) = dcon(X).

Proof. (i): Because of the definitions of winpot(X), inpot(X), stinpot(X) and
dcon(X) it suffices to verify that stinpot(X) ≥ dcon(X). Let, therefore, some
component C of X and a continuous total preorder “ - ” on X be arbitrarily
chosen. Then there exists an indifference class [x] of “ - ” such that C∩[x] 6= ∅.
Now we distinguish between the following two cases:

Case 1: C ⊂ [x]. In this case [x] is a calculable subset of X such that
|[x]| ≥ dcon(C).

Case 2: C\ [x] 6= ∅. Since C is connected we may conclude that
I := {v ∈ X|C ∩ [v] 6= ∅} is a connected interval of (X, -) (cf. the correspond-
ing part in the proof of Lemma 3.1). Hence, we may choose some point z ∈ I
that neither is a first nor a last element of I and the indifference class [z] which



180 G. Gerden and A. Pallack

has minimal cardinality with respect to any other indifference class of an inner
point of I. It follows from the definition of a calculable set with respect to
“ -|C ” that [z] is a calculable subset of X with respect to “ -|C ”. Since

C * [z] the first part of Remark and Example 3.2 implies that C\ [z] is not
connected with respect to t-, which, in particular, means that C\ [z] is not
connected with respect to t. Therefore, [z] is a calculable subset of X such that
|[z]| ≥ dcon(C).

Since C and “ - ” have been arbitrarily chosen we now may conclude with
the help of the definitions of stinpot(X) and dcon(X) that the desired inequality
stinpot(X) ≥ dcon(X) actually holds.

(ii): Let every component C of X be open and closed and let, in addition,
every component C of X be metrizable or regular and hereditarily Lindelöf.
Because of assertion (i) it remains to verify that dcon(C) ≥ inpot(X). There-
fore, we choose for every component C ∈ S(C) some set Z ∈ T (C) that has
minimal cardinality and construct in three steps a continuous total preorder
“ - ” on X such that dcon (X) ≥ inpot (-). Then the definition of inpot (X)
implies that the desired inequality dcon(X) ≥ inpot(X) actually holds. Let,
therefore, some component C ∈ S (C) and a set Z ∈ T (C) that has minimal

cardinality with respect to any other set Z
′

∈ T (C) be arbitrarily but fixed
chosen.

1. In the first step we want to show that, without loss of generality, Z may
be assumed to be a closed subset of X. Indeed, since C\Z is not connected
and C is an open and closed subset of X, there exist open subsets U and V
of C such that U ∩ V ∩ C\Z = ∅ and U ∩ C\Z 6= ∅ and V ∩ C\Z 6= ∅ and
C\Z ⊂ U ∪ V . Now we distinguish between the following two cases:

Case 1: U ∩ V = ∅. In this case we set Z
′

:= C\(U ∪ V ). The inclusion

Z
′

⊂ Z allows us to conclude that U ∩ C\Z
′

6= ∅ and V ∩ C\Z
′

6= ∅. Since
C\Z

′

= U ∪ V it, thus, follows that C\Z
′

is not connected. Z
′

is a closed
subset of X by construction. Hence, in the first case we may assume that Z is
closed.

Case 2: U ∩V 6= ∅. The relation U ∩V ∩C\Z = ∅ implies that U ∩V ⊂ Z.
Let x ∈ U ∩ V be some arbitrarily chosen point. Since the relativized topology
t|C on C is metrizable or regular and (hereditarily) Lindelöf it follows that
(

C, t|C
)

is a normal and, therefore, in particular a completely regular space.
Hence, there exists some continuous function f from C into the real interval
[0, 1] such that f(x) = 1 and f(C\(U ∩ V )) = {0}. Let Z

′

:= f−1
({

1
2

})

.

The continuity of f allows us to conclude that Z
′

is a closed subset of X.
In addition, it follows that Z

′

⊂ U ∩ V ⊂ Z. Furthermore, the sets O :=
{

y ∈ C|f(y) < 1
2

}

and W :=
{

z ∈ C|f(z) > 1
2

}

are two disjoint non-empty

open subsets of C\Z
′

the union of which is C\Z
′

, which means that C\Z
′

is
not connected. Therefore, also in the second case Z may be assumed to be a
closed subset of X.

2. In the second step we want to construct a continuous total preorder
“ -C ” on C such that Z is an indifference class of “ -C ”. Because of the first



Cardinality of indifference classes 181

step we may assume that Z is a closed subset of X. In addition, we already
know that

(

C, t|C
)

is a metrizable or competely regular hereditarily Lindelöf
space. Therefore, we distinguish between the following two cases:

Case 1:
(

C, t|C
)

is metrizable. In this case we choose some metric d on C

that induces t|C. Then we consider the continuous function g : C −→ R
≥0 that

is defined for all points x ∈ C by setting g(x) := d(Z, x). Since g−1({0}) = Z
we may conclude that Z is an indifference class of the continuous total preorder
” -C ” on C that is defined by y -C z ⇐⇒ g(y) ≤ g(z).

Case 2:
(

C, t|C
)

is a completely regular hereditarily Lindelöf space. Now
there exists for every point x ∈ C\Z a continuous function fx : C −→ [0, 1]
such that fx(x) = 1 and fx(Z) = {0}. Moreover, since

(

C, t|C
)

is a here-
ditarily Lindelöf space countably many points x1, x2, ..., xn, ... ∈ C\Z can
be chosen such that C\Z =

⋃

x∈C\Z

f−1x ((0, 1]) =
⋃

n∈N\{0}

f−1xn ((0, 1]), which

implies that Z =
⋂

n∈N\{0}

f−1xn ({0}). Therefore, we set g :=
∑

n∈N\{0}

1
2n

fxn.

Then g : C −→ [0, 1] is a continuous function. Hence, in the same way as in
the first case, we may consider the continuous total preorder “ -C ” on C that
is defined by y -C z ⇐⇒ g(y) ≤ g(z). Since the definition of g implies that
g−1 ({0}) = Z it follows that Z is an indifference class of “ -C ”.

3. In the third step we define the continuous total preorder “ - ” on X such
that dcon(X) ≥ inpot(-). Let, therefore, x ∈ X be an arbitrarily chosen point.
Then there exists a uniquely determined component C of X that contains x,
and we set D(x) := C. Now we choose an arbitrary total ordering “ � ” on C
and set

y - z ⇐⇒















y -C z,
if there exists some C ∈ S (C)
that contains both point y and z

D(y) � D(z), otherwise

for every pair of points y, z ∈ X. Since every component of X is open and
closed and since every total preorder “ -C ” that has been defined is continuous
we may conclude that “ - ” is a continuous total preorder on X. Because of the
definition of S (C) and the second step it follows with the help of the definitions
of inpot(-) and dcon(X), in addition, that dcon(X) ≥ inpot(-) (indeed, the
equality dcon(X) = inpot(-) holds), which settles assertion (ii).

(iii): Let S(C) = ∅. Then the reader may notice that the definition of “ - ”
in the third step of the proof of assertion (ii) also implies that winpot(X) =
dcon(C). This observation finishes the proof of the proposition. �

Remark and Example 3.4. The proof of assertion (ii) of the above propo-
sition motivates the problem of determining all topologies t on X for which
every non-empty closed subset Z of X is the indifference class of an appro-
priate continuous total preorder “ - ” on X. Meanwhile the first author has
obtained some partial results on the characterization of Hausdorff-topologies t



182 G. Gerden and A. Pallack

on X that have the property that every non-empty closed subset of X is an
indifference class of some continuous total preorder ”-” on (X, t) (cf. [10]).
Clearly, these topologies must be completely regular. If t is countably compact
it follows, in addition, that t must satisfy ccc (countable chain condition). The
authors think that these results somewhat justify the assumptions of assertion
(ii) of the proposition.

One may verify that, in general, winpot(X) > inpot(X) ≥ stinpot(X) >
dcon(X). Indeed, let [0, 1] be the real unit interval. Then we substitute 0 by a
copy R(0) of the reals and every real r ∈ (0, 1] by a copy Q(r) of the rationals
in order to set X := R(0) ∪

⋃

r∈(0,1]

Q(r). Now it follows for every x ∈ X that

x ∈ R(0) or x ∈ Q(r) for some uniquely determined real r ∈ (0, 1]. In this way a
canonical function g : X −→ [0, 1] is given, and we may define a total preorder
“ . ” on X by setting x . y ⇐⇒ g(x) ≤ g(y). We proceed by considering two
arbitrarily but fixed chosen subsets U and V of X such that U ∩ R(0) 6= ∅ and
V ∩ R(0) 6= ∅ and U ∩ Q(r) 6= ∅ and V ∩ Q(r) 6= ∅ for every real r ∈ (0, 1]
and U ∪ V = X and U ∩ V is a non-empty finite set. Then we consider the
topology t on X that is generated by the order topology t. and the sets U and
V . Because of the definition of t we may conclude that t is connected and that
for every continuous total preorder “ - ” on X either the inclusion . ⊂ - or
& ⊂ - holds. It, thus follows that winpot(X) = winpot(.) = winpot(&) = 2ℵ0

and that inpot(X) = stinpot(X) = inpot(.) = inpot(&) = ℵ0. In addition,
since U ∪ V = X and U ∩ V is a non-empty finite set we may conclude that
dcon(X) = |U ∩ V | < ℵ0, which settles the example.

On the other hand, it seems to be difficult to construct a topological space
(X, t) such that inpot(X) > stinpot(X). The authors conjecture, instead,
that inpot(X) = stinpot(X) for every topological space (X, t). Indeed, the
set CT P(X) of all continuous total preorders “ - ” on (X, t) is naturally
ordered by set inclusion. An application of the Lemma of Zorn now implies
that (CT P(X), ⊂) contains minimal elements “ . ”. Then with the help of
some uniqueness argument for ” .|C ” on every component C of X that is well
known for orders (cf., for instance, Eilenberg [9] or any other book or paper
on connected orderable spaces) it should be possible to prove that for any
minimal element “ . ” of (CT P(X), ⊂) the equalities inpot(.) = inpot(X) =
stinpot(X) hold. But at present the authors are not sure about some crucial
technical details of the argument.

In order to also generalize the approach that has been sketched in Section 2
we still introduce the following concepts.

Let x, y ∈ X be different points. Then we denote by P(x, y) the set of all
paths f : [0, 1] −→ X that connect x and y and define the path rank of the pair
(x, y) by



Cardinality of indifference classes 183

prank(x, y) :=















min{|S| |S ⊂ X\ {x, y} and Imf ∩ S 6= ∅
for every path f ∈ P(x, y)},

if P(x, y) 6= ∅

0, otherwise

.

This definition allows us to define for an arbitrarily chosen set D ⊂ X the
path rank of D by prank(D) := min{prank(x, y)|x, y ∈ D, x 6= y}.

Moreover, we define the separation rank of the pair (x, y) by

srank(x, y) :=























min{|F| |F is with respect to set
inclusion a maximal separated
subset of P(x, y)},

if P(x, y) 6= ∅

0, otherwise

,

in order to finally denote for every set D ⊂ X the separation rank of D by
srank(D) := min{srank(x, y)|x, y ∈ D, x 6= y}.

4. Lower and upper bounds and a topological characterization
of real intervals

As in the previous section we consider some fixed chosen non-trivial topo-
logical space (X, t) .

With the help of the proof of assertion (i) of Proposition 3.3 the following
proposition follows immediately.

Proposition 4.1. Let “ - ” be some continuous total preorder on X. Then
for every component C of X there exists at least one indifference class [x] of
“ - ” such that |[x]| ≥ dcon(X).

Let now “ - ” be an arbitrarily chosen continuous total preorder on X.
Then we consider an indifference class [x] of “ - ” for which the set C ([x]) of
all components C of X such that [x] is a calculable set with respect to ” -|C ”
is non-empty. Now the following proposition holds.

Proposition 4.2.

|[x]| ≥
∑

C∈C([x])

dcon(C).

Proof. Let some component C ∈ C ([x]) be arbitrarily chosen. Then we may
conclude with the help of the first part of Remark and Example 3.2 that C ⊂ [x]
or C ∩ [x] ∈ T (C). Hence, it follows for every component C ∈ C ([x]) that
|C ∩ [x]| ≥ dcon(C). Since the components C ∈ C ([x]) are pairwise disjoint
we, thus, may conclude that the desired inequality holds. �

Now we want to generalize the situation that has been considered in Section
2. Let, therefore, two different points x, y ∈ X be arbitrarily chosen. Then the
following proposition holds.



184 G. Gerden and A. Pallack

Proposition 4.3.

srank(x, y) ≤ prank(x, y) ≤ 2ℵ0 · srank(x, y).

Proof. The validity of the inequality srank(x, y) ≤ prank(x, y) is a straight-
forward consequence of the definitions of srank(x, y) and prank(x, y) respec-
tively. Hence, only the inequality prank(x, y) ≤ 2ℵ0 · srank(x, y) has to be
verified. The definitions of srank(x, y) and prank(x, y) respectively allow us
to assume without loss of generality that P(x, y) 6= ∅. Let, therefore, F be
some separated subset of P(x, y) that is maximal with respect to set inclusion.
Then we set P(F) :=

⋃

f∈F

(Imf\{x, y}). Now the maximality of F implies that

P(F)∩Imf 6= ∅ for every path f ∈ P(x, y). Since P(F) ⊂X\{x, y}, thus, the
inequality prank(x, y) ≤ |P(F)| holds. In addition, it follows from the defini-
tion of P(F) that |P(F)| ≤ 2ℵ0 ·|F|. Summarizing these considerations we may
conclude that prank(x, y) ≤ 2ℵ0 · srank(x, y), which still was to be shown. �

Let us now assume, that every component C of X is path connected or
locally path connected which means that the components of X with respect to
path connectedness coincide with the components C of X. Then the following
lemma is an immediate consequence of the definitions of dcon(C) and prank(C)
respectively.

Lemma 4.4.

dcon(C) = prank(C) for every component C of X.

Because of Proposition 4.3 and Lemma 4.4 also the following proposition
holds.

Proposition 4.5.

supC∈Csrank(C) ≤ supC∈Cprank(C) = dcon(X) ≤ 2
ℵ0 · supC∈Csrank(C).

Let (V, t) be some real or complex convex space the dimension dim V of
which is greater than 1. Then our considerations also imply the following
proposition and corollaries (cf. Section 2).

Proposition 4.6.

winpot(V ) = inpot(V ) = stinpot(V ) = dcon(V ) = |V | .

Proof. Let x, y ∈ V be arbitrarily chosen different points. Because of Proposi-
tion 4.3 and Lemma 4.4 and the inequalities dcon(V ) ≤ stinpot(V ) ≤ inpot(V ) ≤
winpot(V ) ≤ |V | it suffices to prove that |V | ≤ srank(x, y). Therefore, we con-
sider some (closed) hyperplane H of V that separates the points x and y. Then
we define for every point z ∈ H some path fz ∈ P(x, y) by setting

fz(λ) :=

{

x + 2λ(z − x), if λ ∈ [0, 1
2
)

z + (2λ − 1)(y − z), if λ ∈ [1
2
, 1]

for every real λ ∈ [0, 1]. A routine argument that uses for different points z,



Cardinality of indifference classes 185

z
′

∈ H the linear independency of the vectors z−x and z
′

−x, respectively y−z
and y − z

′

, implies that the collection F := {fz}z∈H is a separated family of
paths fz ∈ P(x, y). Since |H| = |V | the definition of srank(x, y), thus, implies
that |V | = |F| ≤ srank(x, y), and the desired inequality follows. �

Corollary 4.7. |[x]| = |V | for every continuous total preorder “ - ” on V and
every indifference class [x] of “ - ” such that x neither is a first nor a last
element of “ - ”.

Proof. If x neither is a first nor a last element of “ - ” then L(x) and K(x) are
two non-empty disjoint open subsets of V the union of which is V \ [x]. Hence,
|V | ≥ |[x]| ≥ dcon(V ) and the desired conclusion follows from Proposition
4.6. �

The following corollary strengthens and generalizes Dugundji’s result that
has been quoted in Section 2 to arbitrary real or complex convex spaces.

Corollary 4.8. Let S be a subset of V such that |S| < |V |. Then V \S is path
connected.

Proof. In the proof of Proposition 4.6 it has been shown that srank(x, y) = |V |
for every pair of different points x, y ∈ V . If |S| < |V | this means, in particular,
that for every pair of points x, y ∈ V \S there exists some path f : [0, 1] −→
V \S that connects x and y (cf. the corresponding argument in Section 2). �

Let now n > 1 be a natural number. According to the last paragraph of
Section 2 we define for every non-empty subset S of Rn the dimension dimS
of S by choosing an arbitrary vector v ∈ S in order to then identify dimS with
the dimension of the linear subspace of Rn that is generated by S − {v}. The
reader may recall from Section 2 that the definition of dimS is independent of
any particular chosen vector v ∈ S. Then the following proposition holds.

Proposition 4.9. Let “ - ” be a continuous total preorder on Rn. Then the
following assertions are valid:

(i) dim[x] ≥ n − 1 for every indifference class [x] of “ - ” such that x
neither is a first nor a last element of “ - ”.

(ii) [x] is a hyperplane for every indifference class [x] of “ - ” such that
dim[x] = n − 1 and x neither is a first nor a last element of “ - ”.

Proof. (i): Let [x] be an indifference class of “ - ” such that x neither is a first
nor a last element of “ - ”. By considering instead of “ - ” the continuous
total preorder “ -x ” on R

n that is defined by u − x -x v − x ⇐⇒ u - v we
may assume without loss of generality that [x] contains the zero vector “0”.
Therefore, we may consider the linear subspace V[x] of R

n that is generated
by [x]. Then we assume, in contrast, that dim V[x] ≤ n − 2. Because of this
assumption there exists a linear subspace W of Rn that is generated by two
linearly independent vectors y and z such that V[x] ∩W = {0}. For every vector
c ∈ Rn we now set Lcy := {c + λ(y − c)|λ ∈ R} = {y + γ(c − y)|γ ∈ R} and



186 G. Gerden and A. Pallack

Lcz := {c+λ(z−c)|λ ∈ R} = {z+γ(c−z)|γ ∈ R}. Then we choose two arbitrary
vectors a, b ∈ Rn\V[x]. Since V[x] ∩ W = {0} it follows by direct computation
that is based upon contraposition that Lay ∩ V[x] = ∅ or L

a
z ∩ V[x] = ∅ and

Lby ∩ V[x] = ∅ or L
b
z ∩ V[x] = ∅. We may assume without loss of generality that

Lay ∩ V[x] = ∅ and L
b
z ∩ V[x] = ∅. The remaining possibilities can be settled by

analogous arguments. The equations Lay ∩ V[x] = ∅ and V[x] ∩ W = {0} and
Lbz ∩ V[x] = ∅ allow us to define a path f : [0, 1] −→ R

n\V[x] by setting

f(λ) :=







a + 3λ(y − a), if λ ∈ [0, 1
3
)

y + (3λ − 1)(z − y), if λ ∈ [1
3
, 2
3
)

z + (3λ − 2)(b − z), if λ ∈ [2
3
, 1]

for every real λ ∈ [0, 1]. Since the vectors a, b ∈ Rn\V[x] have been arbitrarily
chosen we may conclude that Rn\V[x] is path connected and, therefore, con-
nected. Now the connectedness of Rn\V[x] implies with the help of a straight-
forward indirect argument that is based upon the fact that every non-empty
open subset of Rn contains a base of linearly independent vectors of Rn that
also Rn\ [x] is connected, which contradicts the proof of Corollary 4.7 and,
thus, finishes the proof of assertion (i).

(ii): Let x ∈ Rn be any point that neither is a first nor a last element of
“ - ” such that dim [x] = n − 1. In order to prove that [x] is a hyperplane
of Rn we may assume, as in the proof of assertion (i), that 0 ∈ [x]. Using the
notation of the proof of assertion (i) it suffices to verify that V[x] ⊂ [x]. We
assume, in contrast, that there exists some non-empty subset S of V[x] that is
not contained in [x]. We may assume without loss of generality that |S| = 1.
Indeed, the last argument in the proof of assertion (i) shows in which way the
case that |S| > 1 can be reduced to the case that |S| = 1. Let, therefore,
S = {v} for some vector v ∈ V[x]. Then we set V

−
[x]

:= V[x]\{v} and prove that

Rn\V −
[x]

is connected. As in the proof of assertion (i) it, thus, follows that also

Rn\ [x] is connected, which contradicts the proof of Corollary 4.7. Hence, if
we are able to verify that Rn\V −

[x]
is connected nothing remains to be shown.

In order to prove that Rn\V −
[x]

is connected we choose two arbitrary points a,

b ∈ Rn\V −
[x]

and distinguish between the cases that a = v or b = v and that

neither a = v nor b = v.
Case 1: a = v or b = v. In this case, we may assume without loss of

generality, that a = v and that b 6= v. Since a = v and b /∈ V[x] it follows that

{a + λ(b − a)| λ ∈ R} ∩ V −
[x]

= Lab ∩ V
−
[x]

= ∅. This means that f : [0, 1] −→

Rn\V −
[x]

defined by f (λ) := a + λ(b − a) for every real λ ∈ [0, 1] is a path that

connects a and b.
Case 2: a 6= v and b 6= v. Now it follows that {a + λ(v − a)|λ ∈ R} ∩ V −

[x]
=

Lav ∩ V
−
[x]

= ∅ and that {v + γ(b − v)|γ ∈ R} ∩ V −
[x]

= Lv
b
∩ V −

[x]
= ∅. Hence,

f : [0, 1] −→ Rn\V −
[x]

defined by



Cardinality of indifference classes 187

f (λ) :=

{

a + 2λ(v − a), if λ ∈ [0, 1
2
)

v + (2λ − 1)(b − v), if λ ∈ [1
2
, 1]

for every real λ ∈ [0, 1] is a path that connects a and b (cf. the definition of
fz in the proof of Proposition 4.6). Since a, b ∈ R

n\V −
[x]

have been arbitrarily

chosen we may summarize our considerations for concluding that Rn\V −
[x]

is

path connected and, therefore, also connected. Thus, the proof of assertion (ii)
is complete. �

Remark 4.10. An analysis of the proof of Proposition 4.9 implies that Propo-
sition 4.9 can be generalized to arbitrary real or complex convex spaces V
the dimension dim V of which is greater than 1. But then the conclusion
dim [x] ≥ n−1 has to be replaced by the condition that in case that 0 /∈ [x] the
linear subspace of V that is generated by [x] coincides with V and that in case
that 0 ∈ [x] the linear subspace of V that is generated by [x] coincides with V
or is a maximal linear subspace of V . In combination with Proposition 4.6 and
Corollary 4.7 this additional result, thus, provides a rather complete survey on
the size of indifference classes of a continuous total preorder “ - ” on a real or
complex convex space. In this sense Proposition 4.9 completes Proposition 4.6
and Corollary 4.7. Indeed, in mathematical utility theory a particular chosen
set of coordinates or a particular chosen base of linearly independent vectors of
a real or complex convex space V can be interpreted as the collection of those
(latent) factors or dimensions that influence preferences between alternatives.

Let, for the moment, an indifference class of a continuous total preference
relation “ - ” on V for which x neither is a first nor a last element of “ - ” said
to be an inner indifference class of “ - ”. Then we learn from Proposition 4.6,
Corollary 4.7 and Proposition 4.9 that inner indifference classes not only have
maximal cardinality but that, in addition, preferences between alternatives y
and z, the corresponding indifference classes of which are inner, can be assumed
to be determined by different expressions of exactly one coordinate or different
expressions of the same coordinates or different expressions of nearly the same
coordinates, which means that one of the given alternatives y or z can be
influenced by one more coordinate than the other alternative. Indeed, if [y]
and [z] are hyperplanes then the disjointness of [y] and [z] implies that the
preference between [y] and [z] can be assumed to be influenced by exactly one
coordinate (cf. the corresponding example in the introduction). Otherwise, at
least one of the indifference classes [y] or [z] is influenced by a maximal set of
coordinates, which means that [y] or [z] can be assumed to be influenced by at
most one more coordinate than any other inner indifference class of “ - ”.

In order to complete these considerations we still consider a continuous total
preference relation “ - ” on a real or complex convex space V that is locally
non-satiated (cf. the introduction) and has the additional property that all
its inner indifference classes are path connected. In this case the authors con-
jecture that there exists a homeomorphism h : V −→ V such that every inner



188 G. Gerden and A. Pallack

indifference class of the continuous total preorder ” -h ” on X that is defined
by h(x) -h h(y) ⇐⇒ x - y is a hyperplane of V .

If, however, an arbitrary abstract topological space is given an appropriate
generalization of Proposition 4.9 hardly seems to be possible, and the cardinal-
ity approach gains additional importance.

We conclude by proving the following proposition that is a slight general-
ization of well known results in general topology (cf. Eilenberg [9] or Willard
[15, Section 28] or Kok [12, Chapter II]). It provides a new proof of these clas-
sical results and includes for every natural number n ≥ 1 the family of all
connected and convex subspaces of the Euclidean space (Rn, tnat). Therefore,
it also generalizes Theorem 4 of Candeal and Induráin [6] and is closely re-
lated to generalizations of this theorem by Beardon [2] and Candeal, Induráin
and Mehta [7] (cf., in particular, [7, Theorem 4]), where necessary and suffi-
cient conditions for the agreement of the Euclidean and the order topology on
totally ordered subsets of Rn are discussed.

Proposition 4.11. Let (X, t) be a separable, connected and locally connected
Hausdorff space. Then the following assertions are equivalent:

(i) Every non-trivial connected subspace
(

Y, t|Y
)

of (X, t) contains some
point y such that Y \{y} splits into two components.

(ii) There exists some continuous total order “ � ” on X.

(iii) There exists some real interval I such that (X, t) is homeomorphic to
(I, tnat).

Proof. Since the proof of the implication “(iii)=⇒(i)” is straightforward it suf-
fices to verify the validity of the implications “(i)=⇒(ii)” and “(ii)=⇒(iii)”.

(i)=⇒(ii): We construct by transfinite induction a continuous total order
” � ” on X.

α = 0 : We set -0:= X × X.
0 < α is not a limit ordinal: Let Iα−1 := {[x] | [x] is a non − trivial

indifference class of ” -α−1 ”}. In case that Iα−1 = ∅ we set �:= -α−1 in
order to then finish the induction process. Otherwise, we choose an arbitrary
indifference class [x] ∈ Iα−1. Because of the induction hypothesis we may
assume that

(

([x] , t|[x]
)

is a connected (and locally connected) Hausdorff space
and that [x] = X or there exists a single point s ∈ X such that

s = max{u ∈ X|u ≺α−1 x}

or there exists a single point t ∈ X such that

t = min{v ∈ X|x ≺α−1 v}.

We abbreviate these assumptions on [x] by (*). In addition, the induction
hypothesis allows us to assume that the sets dα−1(s) := {u ∈ X|u -α−1 s}
and iα−1(s) := {u ∈ X|s -α−1 u} or dα−1(t) := {v ∈ X|v -α−1 t} and
iα−1(t) := {v ∈ X|t -α−1 v} are closed subsets of X. Let us abbreviate these
last assumptions by (**). Since the afore-presented cases can be settled by



Cardinality of indifference classes 189

analogous arguments we may concentrate without loss of generality on the case
that there exist single points s, t ∈ X such that s =max{u ∈ X|u ≺α−1 x}
and t =min{v ∈ X|x ≺α−1 v}. Assertion (i) implies the existence of some
point z ∈ [x] such that [x] \{z} splits into two (non-empty) connected and
then, obviously, locally connected components Cα1 and C

α
2 . The assumptions

(**) imply that [x] is an open subset of X. Since (X, t) is a locally connected
Hausdorff space we, thus, may conclude that also Cα1 and C

α
2 are open subsets

of X. Furthermore, these assumptions on (X, t) allows us to assume without
loss of generality that the (topological) closure Cα1 of C

α
1 is C

α
1 ∪ {s, z} and

that the (topological) closure Cα2 of C
α
2 is C

α
1 ∪ {z, t}. Hence, we set

-α:=-α−1 \{(u, v) ∈ X × X|u ∈ C
α
2 ∪ {z} and v ∈ C

α
1 or u ∈ C

α
2 and

v ∈ Cα1 ∪ {z}}. Summarizing the afore-considered arguments it follows that
both sets Cα1 and C

α
2 are indifference classes of ” -α ” and that the assumptions

(*) and (**) are also satisfied with respect to ” -α ”.
0 < α is a limit ordinal: In this case we set -α:=

⋂

β<α

-β.

The reader may verify that in this way, actually, a total order ” � ” on X
has been constructed. With the help of the assumptions (**) we may conclude,
in addition, that ” � ” is a continuous total order on X.

(ii)=⇒(iii): Let ” � ” be a continuous total order on X. The general
assumptions of the proposition imply with the help of Eilenberg’s well known
re-presentation theorem [9] that there exists some real interval I such that
(

X, t�
)

is homeomorphic to (I, tnat). It, thus, suffices to show that t
� = t.

The continuity of “ � ” allows us to conclude that t� ⊂ t and it remains to
verify that also t ⊂ t�. Since (X, t) is locally connected we may choose some
connected open subset O of X. Then we must prove that O ∈ t�. Because of
the assumption that “ � ” is a continuous total order on X it follows from the
connectedness of O that O is an interval of (X, �). Since O is an open subset of
X the connectedness of (X, t) and the continuity of “ � ” imply, moreover, that
O must be an open interval of (X, �). Hence, O ∈ t� and nothing remains to
be shown. �

Acknowledgements. The authors are grateful to two anonymous referees
for a very careful reading of an earlier version of this paper and for several
valuable comments and suggestions that have resulted in improvements.

References

[1] K. Arrow and F. Hahn, General competetive analysis, (Oliver and Boyd, Edinburgh,
1971).

[2] A.F. Beardon, Totally ordered subsets of Euclidean space, J. Math. Econom. 23 (1994),
391-393.



190 G. Gerden and A. Pallack

[3] T. Bewley, Existence of equilibria in economics with infinitely many commodities, J. of
Econom. Theory 4 (1972), 514-540.

[4] G. Birkhoff, Lattice Theory, (American Mathematical Society, Rhode Island, 1940 (First
edition), 1967 (Third edition)).

[5] D. Bridges and G.B. Mehta, Representation of preference orderings, (Springer, New
York, 1995).

[6] J.C. Candeal and E. Induráin, Utility functions on chains, J. Math. Econom. 22 (1993),
161-168.

[7] J.C. Candeal, E. Induráin and G.B. Mehta, Further remarks on totally ordered repre-
sentable subsets of Euclidean space, J. Math. Econom. 25 (1996), 381-390.

[8] J. Dugundji, Topology, (Allyn and Bacon, Boston, 1966).
[9] S. Eilenberg, Ordered topological spaces, Amer. J. Math. 24 (1941), 305-309.

[10] G. Herden, On the semicontinuous and continuous analogue of the Szpilrajn theorem,
in preparation, University of Essen, 2003.

[11] L. Jones, A competetive model of product differentiation, Econometrica 52 (1984), 507-
530.

[12] H. Kok, Connected orderable spaces, (Mathematical Centre Tracts, Am-sterdam, 1973).
[13] A. Mas-Colell, The price equilibrium existence problem in topological vector lattices,

Econometrica 54 (1986), 1039-1053.

[14] G.B. Mehta, Infinite dimensional Arrow-Hahn theorem, Preprint, University of Bris-
bane, 1989.

[15] S. Willard, General topology, (Addison-Wesley, Mass.-London-Don Mills, 1970).

Received July 2002

Accepted June 2003

Gerhard Herden (gerhard.herden@uni-essen.de)
Universität Duisburg-Essen, Fachbereich 6 (Mathematik), Universitaets-strasse
2, D-45117 Essen, Germany

Andreas Pallack (andreas.pallack@uni-essen.de)
Universität Duisburg-Essen, Fachbereich 6 (Mathematik), Universitaets-strasse
2, D-45117 Essen, Germany