Electronic Notes in Theoretical Computer Science 202 (2008) 19–25

www.elsevier.com/locate/entcs

Continuity Properties of Preference Relations

Marian Baroni1
Department of Mathematics and Statistics
University of Canterbury
Christchurch, New Zealand

Douglas Bridges2
Department of Mathematics and Statistics
University of Canterbury
Christchurch, New Zealand

Abstract
Various types of continuity for preference relations on a metric space are examined constructively. In
particular, necessary and sufficient conditions are given for an order-dense, strongly extensional preference
relation on a complete metric space to be continuous. It is also shown, in the spirit of constructive reverse
mathematics, that the continuity of sequentially continuous, order-dense preference relations on complete,
separable metric spaces is connected to Ishihara’s principle BD-N, and therefore is not provable within
Bishop-style constructive mathematics alone.

Keywords: constructive reverse mathematics, Heine-Borel property, anti-Specker property, uniform

continuity theorem

1 Preferences and Continuity

The notions of preference and utility play a fundamental role in traditional microeco-
nomic theory . Each consumer is assumed to have a consumption set X—typically,
but not essentially, a compact, convex subset of Rn —whose elements are the con-
sumption bundles. Each component of the consumption bundle x is the amount
of a corresponding good or service that the consumer may wish to purchase. It is
assumed that there is a binary relation  of preference on X, where “x  y” means
that the consumer strictly prefers bundle x to bundle y. The early developments
of the theory went so far as to assume that the preferences were represented by a
utility function u : X → R, whereby x  y if and only if u(x) > u(y). It was later
1 Email:[email protected]
2 Email:[email protected]

1571-0661 © 2008 Elsevier B.V. Open access under CC BY-NC-ND license.

doi:10.1016/j.entcs.2008.03.004
20 M. Baroni, D. Bridges / Electronic Notes in Theoretical Computer Science 202 (2008) 19–25

realised that the existence of a utility function representing a given preference re-
lation required justification; furthermore, one could not automatically assume that
the utility function, if it existed, was continuous. This led to the study of necessary
and sufficient conditions for the existence of continuous utility functions, a topic
which has been explored in depth since the appearance of the pioneering work of
Debreu in the 1950s [11,12].
Early in the constructive analysis of preference relations it became clear that
straightforward constructivisation of the classical proofs of the existence and con-
tinuity of utility functions was not possible. In fact, Debreu’s famous theorem on
this subject is false in recursive constructive mathematics [6]. Since the smoothest
constructive path to an existence theorem for utility functions uses a very strong
continuity condition on the preference relation [4], it would be interesting (maybe
useful?) to have an existence theorem under weaker continuity conditions on prefer-
ences. To that end, it makes sense to examine the constructive connections between
various types of continuity of preferences, analogous to ones for continuity of func-
tions. We begin such an examination in this paper.
Let X be a set that is inhabited in the sense that we can construct an element
of it. A binary relation  on X is called a preference relation if it satisfies these
two axioms:
P1 ∀x,y∈X (x  y ⇒ ¬ (y  x)) ;
P2 ∀x,y∈X (x  y ⇒ ∀z∈X (x  z ∨ z  y)) .
The corresponding preference-indifference relation  is then defined by

∀x,y∈X (x  y ⇔ ¬ (y  x)) .

(Of course, we write y ≺ x and y  x as equivalents of x  y and x  y, re-

spectively.) The corresponding reverse preference relation rev is defined on X
by
∀x,y∈X (x rev y ⇔ y  x) .
With each preference relation and each point a of X we associate
 the upper contour set

[a, →) = {x ∈ X : x  a} ,

 the strict upper contour set

(a, →) = {x ∈ X : x  a} ,

 the lower contour set

(←, a] = {x ∈ X : a  x} ,

 and the strict lower contour set

(←, a) = {x ∈ X : a  x} .
 M. Baroni, D. Bridges / Electronic Notes in Theoretical Computer Science 202 (2008) 19–25 21

In this paper we are particularly interested in the case where (X, ρ) is a metric
space. The standard inequality on X is then defined by

∀x,y∈X (x = y ⇔ ρ(x, y) > 0) .

Corresponding to various types of continuity of functions between metric spaces

there are types of continuity for a preference relation  on X. We say that  is
 pointwise continuous at a if both the sets (←, a) and (a, →) are open in X;
 sequentially continuous at a if for each x ∈ X and each sequence (xn )n1 of
points of X converging to x,

x  a ⇒ ∃N ∀nN (xn  a)

and
a  x ⇒ ∃N ∀nN (a  xn ) ;
 nearly continuous at a if for each S ⊂ X and each x in the closure S of S,

x  a ⇒ ∃s∈S (s  a)

and
a  x ⇒ ∃s∈S (a  s) ;
 nondiscontinuous at a if both the sets (←, a] and [a, →) are closed in X.
We say that  is, for example, sequentially continuous on X if it is sequentially
continuous at each point of X. It is straightforward to show that if  is represented
by a pointwise, sequential, or nearly continuous 3 utility function, then  itself has
the corresponding continuity property on X.
Our aim is to investigate, within the framework of Bishop-style constructive
mathematics (BISH), 4 connections between these notions of continuity.
Proposition 1.1 For preference relations on a metric space, pointwise continuity
at a point implies sequential continuity, which implies near continuity, which implies
nondiscontinuity.

Proof. Let  be a preference relation on the metric space X, and let a ∈ X.

Suppose that  is pointwise continuous at a. If x  a, then there exists r > 0
such that y  a whenever ρ(x, y) < r. It follows that for every sequence (xn )n1 of
elements of X converging to x, we have xn  a for all sufficiently large n. The case
a  x is similarly handled. Hence  is sequentially continuous at a.
Next suppose that  is sequentially continuous at a. Let S be a subset of X, and
let x be a point of S with x  a. Then there exists a sequence (xn )n1 of elements
of S converging to x. Since x  a, it follows from sequential continuity that xn  a
3 More on near continuity for functions is found in [7].
4 Mathematics that uses only intuitionistic logic and an appropriate set-theory such as that presented in
[1]. For more on constructive analysis see [3,8]. For background in the constructive theory of preference
and utility, see [2,4,5].
22 M. Baroni, D. Bridges / Electronic Notes in Theoretical Computer Science 202 (2008) 19–25

for all sufficiently large n. The case a  x is similarly handled. Thus  is nearly
continuous at a.
To prove that near continuity implies nondiscontinuity, suppose that  is nearly
continuous at a, and consider a sequence (xn )n1 in X that converges to x and
satisfies xn  a for all n. Assume that a  x. Clearly, x belongs to the closure of
[a, →). Since  is nearly continuous at a, it follows that a  s for some s ∈ [a, →), a
contradiction. Consequently, ¬(a  x) and therefore x  a—that is, x ∈ [a, →). We
prove similarly that the lower contour set at a is closed; so the preference relation
 is nondiscontinuous at a. 2

An irreflexive binary relation R on a set X with an inequality = is said to be

strongly extensional if
∀x,y∈X (xRy ⇒ x = y) .

Proposition 1.2 A nearly continuous preference relation on a metric space is

strongly extensional.

Proof. Omitted. 2

Our next task is to produce necessary and sufficient conditions for the sequential
continuity of a certain type of preference relation on a complete metric space (Propo-
sition 1.5 below). We require the following two lemmas, which are reminiscent of
Ishihara’s tricks [13,9].

Lemma 1.3 Let  be a strongly extensional preference relation on a complete met-

ric space X. Let a, b, x ∈ X satisfy x  a  b, and let (xn )n1 be a sequence in X
that converges to x. Then either xn  b for all n or else there exists n such that
a  xn .

Proof. Construct an increasing binary sequence (λn )n1 such that

λn = 0 ⇒ ∀kn (xk  b) ,
λn = 1 − λn−1 ⇒ a  xn .

We may assume that λ1 = 0. If λn = 0, set yn = x; if λn = 1 − λn−1 , set yk = xn for

all k  n. Then (yn )n1 is a Cauchy sequence in X and so converges to a limit y ∈ X.
Either y  a or x  y. In the first case, if there exists n such that λn = 1 − λn−1 ,
then y = xn ≺ a, a contradiction; hence λn = 0, and therefore xn  b, for all n. In
the case x  y, the strong extensionality of  yields x = y; so there exists N such
that x = yN . Then λn = 1 − λn−1 , and therefore x  a  xn , for some n  N. 2

Lemma 1.4 Under the hypotheses of Lemma 1.3, either xn  b for all sufficiently
large n, or else a  xn for infinitely many n.

Proof. The proof uses inductive application of Lemma 1.3 and is omitted. 2
 M. Baroni, D. Bridges / Electronic Notes in Theoretical Computer Science 202 (2008) 19–25 23

A preference relation  on a set X is said to be order dense if for all x, z ∈ X

with x  z, there exists y ∈ X such that x  y  z.

Proposition 1.5 An order-dense, strongly extensional preference relation  on a

complete metric space X is sequentially continuous if and only if it is both nondis-
continuous and strongly extensional.

Proof. If  is sequentially continuous, then by Propositions 1.1 and 1.2, it is both

nondiscontinuous and strongly extensional.
Suppose, conversely, that  has both these last two properties. Let the sequence
(xn )n1 converge to x in X, and, to begin with, let x  b. Pick a ∈ X such that
x  a  b. By Lemma 1.4, either xn  b for all sufficiently large n, or else, as
we suppose in order to obtain a contradiction, there exists a strictly increasing
sequence (nk )k1 of positive integers such that a  xnk for each k. Since (this is
our nondiscontinuity assumption) the set (←, a] is closed, it follows that x ∈ (←, a];
that is, a  x, a contradiction. Hence, in fact, xn  b for all sufficiently large n.
Finally supposing that a  x, we apply the foregoing argument to the reverse
preference relation rev , to show that xn rev a, and therefore a  xn , for all
sufficiently large n. This completes the proof that  is sequentially continuous. 2

In the case where the preference relation is represented by a utility function,

Proposition 1.5 reduces to Theorem 1 of [13].

2 Sequential and Pointwise Continuity

With classical logic we can prove that, for a preference relation  on the metric
space X, sequential continuity at each point of X implies, and hence is equivalent
to, pointwise continuity throughout X. To see this, suppose that  is sequentially
continuous at each point of X but is not pointwise continuous at x ∈ X. Then either
(←, x) or (x, →) is not open in X. Assuming, for example, that the latter set is not
open, we see that there exist a point y ∈ X and a sequence (yn )n1 converging to y
in X, such that y  x  yn for each n. Since  is sequentially continuous at y, we
must have yn  x for all sufficiently large n, a contradiction.
Clearly, this proof provides no indication of conditions that might ensure that
sequential continuity constructively implies pointwise continuity. In seeking such
conditions, and bearing in mind those preference relations that are represented by
utility functions [4], we are guided by Ishihara’s work [14] relating sequential and
pointwise continuity of functions on a complete metric space.
A set S of positive integers is called pseudobounded if limn→∞ n−1 sn = 0
for each sequence (sn )n1 in S. The following principle, introduced by Ishihara,
holds classically and in both the intuitionistic and recursive models of constructive
mathematics, is unprovable in a natural formalisation of BISH [17], and has proved
extremely significant in constructive reverse mathematics:
BD-N Every countable pseudobounded set of positive integers is bounded.
24 M. Baroni, D. Bridges / Electronic Notes in Theoretical Computer Science 202 (2008) 19–25

In particular, as Ishihara showed in [14] (Theorem 4), BD-N is equivalent to the

proposition
Every sequentially continuous mapping of a complete, separable metric space into
a metric space is pointwise continuous.
For more on the role of BD-N in constructive reverse mathematics, see [10,15,16].

Theorem 2.1 If BD-N holds, then every sequentially continuous, order-dense pref-
erence relation on a separable metric space is pointwise continuous.

Proof. Omitted. 2

We end with a partial converse to Theorem 2.1.

Theorem 2.2 Suppose that every sequentially continuous preference relation on a

complete, separable metric space is pointwise continuous. Then BD-N holds.

Proof. Let S be a countable pseudobounded subset of N, and without loss of gener-

ality assume that 0 ∈ S. We first invoke Proposition 1 of [14] to produce a complete,
separable subset X of R and a sequentially continuous mapping u : X → {0, 1} such
that
 
0 ∈ X ∧ u(0) = 0 ∧ ∀m m ∈ S ⇒ 2−m ∈ X ∧ u(2−m )) = 1 .
Note that 1 ∈ S. Now,
x  y ⇔ u(x) > u(y)
defines a sequentially continuous preference relation  on X. Suppose that this
preference relation is pointwise continuous on X. Then

(←, 1) = {x ∈ X : u(x) = 0}

is an open subset of X that contains 0. Choose a positive integer N such that if

x ∈ X and |x| < 2−N , then u(x) = 0. If m ∈ S and m > N, then 2−m ∈ X,
so u(2−m ) = 0, by our choice of N ; but u(2−m ) = 1, by definition of u. This
contradiction shows that m  N for all m ∈ S. Hence S is bounded. 2

The preference relation  used in the proof of Theorem 2.2 is not order dense.
We do not know whether BD-N is a consequence of the proposition “Every sequen-
tially continuous, order-dense preference relation on a complete, separable metric
space is uniformly continuous”.
 M. Baroni, D. Bridges / Electronic Notes in Theoretical Computer Science 202 (2008) 19–25 25

Acknowledgement
The authors thank the Department of Mathematics & Statistics at the University
of Canterbury for supporting Marian Baroni as a doctoral student while this paper
was begun. They also thank the referee for pointing out several errors in the original
version, and for pointing out a major simplification in the proof of Theorem 2.1.

References
[1] Aczel, P., and M. Rathjen, “Notes on Constructive Set Theory”, Report No. 40, Institut Mittag–Leffler,
Royal Swedish Academy of Sciences, 2001.

[2] Baroni, M.A., “The Constructive Theory of Riesz Spaces and Applications in Mathematical Economics”,
Ph.D. Thesis, University of Canterbury, Christchurch, New Zealand, 2004.

[3] Bishop, E.A., and D.S. Bridges, “Constructive Analysis”, Grundlehren der Math. Wiss. 279 (1985),
Springer-Verlag, Heidelberg.

[4] Bridges, D.S., The constructive theory of preference relations on a locally compact space, Proc.
Koninklijke Nederlandse Akademie can Wetenschappen, Ser. A, 92 2 (1989), 141–165.

[5] Bridges, D.S., Constructive methods in mathematical economics, Mathematical Utility Theory, J. Econ.
(Zeitschrift für Nationalökonomie), Suppl. 8 (1999), 1–21.

[6] Bridges, D.S., and F. Richman, A recursive counterexample to Debreu’s theorem on the existence of a
utility function, Math. Soc. Sciences 21 (1991), 179–182.

[7] Bridges, D.S., and L.S. Vı̂ţă, Apartness spaces as a framework for constructive topology, Ann. Pure
and Applied Logic 119 (2003), 61–83.

[8] Bridges, D.S., and L.S. Vı̂ţă, “Techniques of Constructive Analysis”, Universitext, Springer-New-York,
2006.

[9] Bridges, D.S., D. van Dalen and H. Ishihara, Ishihara’s proof technique in constructive analysis, Proc.
Koninklijke Nederlandse Akad. Wetenschappen (Indag. Math.) N.S., 14 2 (2003), 163–168.

[10] Bridges,D.S., H. Ishihara, P.M. Schuster and L.S.Vı̂ţă, Strong continuity implies uniform sequential
continuity, Arch. Math. Logic 44 (2006), 887–895.

[11] Debreu, G., Representation of a preference ordering by a numerical function, Decision Processes (R.
Thrall, C. Coombs, and R. Davies, eds), John Wiley, New York, 1954.

[12] Debreu, G., “Theory of Value”, John Wiley, New York, 1959.

[13] Ishihara, H., Continuity and discontinuity in constructive mathematics, J. Symbolic Logic 56 4 (1991),
1349–1354.

[14] Ishihara, H., Continuity properties in constructive mathematics, J. Symbolic Logic 57 2 (1992), 557–
565.

[15] Ishihara, H., Sequential continuity in constructive mathematics, Combinatorics, Computability and
Logic (C.S. Calude, M.J. Dinneen and S. Sburlan, eds), Springer-Verlag, London, 2001, 5–12.

[16] Ishihara, H., and S. Yoshida, A constructive look at the completeness of D(R), J. Symb. Logic 67
(2002), 1511–1519.

[17] Lietz, P., “From Constructive Mathematics to Computable Analysis via the Realizability
Interpretation”, Ph.D. Thesis, Technische Universität, Darmstadt, Germany, 2004.