BITOPOLOGICAL SPACES

By J. C. KELLY
[Received 15 February 1962]

1. Introduction
IT is well known that, when the classical conditions on a metric d( , ) for
a set X are relaxed by omitting the requirement
d(x, y) = 0 only if z = y,
there is no difficulty in generalizing the standard theorems of metric
spaces, in particular those concerning metrization of topological spaces.
On the other hand, if one attempts to omit the requirement of symmetry,
the appropriate generalizations are not obvious. Such unsymmetric
distance functions have been studied before by Wilson (9), who used
the term quasi-metrics, Ribeiro (6), and others. To my knowledge, there
has been no previous attempt to systematize the study of these
quasi-metrics.
Any distance function d( , ) on a set X (i.e. any non-negative
real-valued function d( , ) denned on the product X x X and satisfying
the triangle inequality) has the property that the open d-spheres (i.e. sets
of the form {y : d(x,y)<t}) form a base for a topology 3) for X. We say
that Sd is the topology determined by d( , ). The purpose of this note is
to observe that, given on a set X any such (unsymmetric) quasi-metric
p( , ) whatsoever, then associated with p(x,y) is the function
q{x,y) =p(y,x),
and that if one studies the two topologies determined for X by p{ , ) and
q( , ), one regains in such 'bitopological spaces' some of the symmetry of
the classical metric situation, and in consequence one can obtain systematic
generalizations of standard results such as Urysohn's lemma, Urysohn's
metrization theorem, Tietze's extension theorem, and the Baire category
theorem. The results show in general that the existence of quasi-metrics
is related to the existence of real-valued functions which are semi-
continuous relative to two topologies in much the same way as the exis-
tence of metrics is related to the existence of real-valued functions which
are continuous relative to the original topology of the space in question.
(The notion of a bitopological space used in relation to semi-continuous
functions restores sufficient symmetry to enable one to use some of the
existing techniques of continuous functions.)
Proc. London Math. Soc. (3) 13 (1963) 71-89
72 J . C. K E L L Y
Definitions and precise statements of the main results are given in § 2
which follows. Section 3 contains the proofs of the results which apply
to bitopological spaces in general. In §4, we consider some elementary
properties of quasi-metrics, and also the concept of completeness in
quasi-metric spaces. Section 5 contains examples of quasi-metrics; in
particular it is observed that such quasi-metrics arise in a natural manner
when one considers monotonic sequences and monotonic functions in
partially ordered pseudo-metric spaces. We also give counter-examples to
some conjectures; other examples of this sort can be found in Balanzat (1).
The starting-point of our work was the use of a quasi-metric to describe a
semi-space topology and give an alternative proof of a theorem of Bonsall
on monotonic sequences (see ((3) 70) and Example 5.3 below).
Theorem 2.8 would seem to be only a first step in applying the idea of
conjugacy to quasi-metrization, just as Urysohn's theorem is the simplest
metrization theorem. As far as I know, this idea has not been used in
any other work on unsymmetric structures (see e.g. (10)).
Most of the results in the present paper were contained in my thesis
'Quasi-metrics and topologies in partially ordered spaces' (Newcastle,
1960). This was written while I was a research student of Professor
F. F. Bonsall, to whom I am sincerely grateful for his guidance and
inspiring example throughout that time. I am grateful also to Professor
W. H. Cockcroft for his help in the preparation of this paper.
2. Statement of main results
The following terminology will be used:f
2.1. DEFINITION. A quasi-pseudo-metric on a set I is a non-negative
real-valued function p( , ) on the product X x X such that
(i) p(x,x) = 0 (xeX),
(ii) p(x,z)^p(x,y)+p(y,z) (x,y,zeX).
If p( , ) satisfies the condition
(i)' p(x,y) = 0 if and only if x = y {x,yeX),
then p( , ) is a quasi-metric. Of course, if also p(x,y) = p{y,x) (x,yeX),
then p( , ) is a metric, in the classical sense.
2.2. DEFINITION. Let p{ , ) be a quasi-pseudo-metric on X, and let
q( , ) be defined by q(x, y) = p(y, x) (x, y e X). Then it is a trivial matter to
verify that q{ , ) is a quasi-pseudo-metric on X. We say that p( , ) and
q( , ) are conjugate, and denote the set X with this structure by (X,p,q).
f The terminologies in previous papers on quasi-metrics have been haphazard and
even conflicting. Ribeiro (6) used the term 'weak metric'.
 BITOPOLOGICAL SPACES 73
If p( , ) is a quasi-pseudo-metric on a set X, then {y : p(x, y) < k}, the
open ^-sphere with centre x and radius k>0, is denoted by P(x, k). Just
as in the classical case, the collection of all open ^-spheres forms a base
for a topology. The topology determined in this way by p( , ) will be
denoted by & and will be called the quasi-pseudo-metric topology of
p( , ). Similarly, q( , ) determines a topology 2 for X. Thus the natural
topological structure associated with a quasi-pseudo-metric on a set X
is that of the set X with two topologies.
2.3. DEFINITION. A space X on which are defined two (arbitrary)
topologies 8P and 2 is called a bitopological space and denoted by (X, &>, 2).
The following definitions extend to a bitopological space (X, 0>,2) the
notions of separation properties of a topological space (X, 8P).
2.4. DEFINITION. In a space (X,^, 2), & is said to be regular with
respect to 2 if, for each point x i n l , there is a ^"-neighbourhood base of
^-closed sets, or, as is easily seen to be equivalent, if, for each point x in X
and each ^-closed set P such that x$P, there are a ^-open set U and a
j2-open set V such that
xeU, PcF, and UnV = 0.
(X, &, 2) is, or 0> and 2 are, pairwise regular if 0* is regular with respect
to 2 and vice versa.
2.5. DEFINITION. A space {X, &, 2) is said to be pairwise Hausdorff if,
for each two distinct points x and y, there are a ^-neighbourhood U of x
and a ^-neighbourhood V of y such that U f) V = 0.
We observe that if (X, £P, 2.) is pairwise Hausdorff, then both & and 2
are ^-topologies. If 8P and 2 are determined by conjugate quasi-pseudo-
metrics, it follows immediately from the definitions that the converse is
true. In fact, a pair of conjugate quasi-pseudo-metrics are quasi-metrics if
and only if the associated topologies are pairwise Hausdorff.
The concepts of regularity of a topology with respect to some other
topology and of two topologies being pairwise Hausdorff may be compared
with those of coupling and consistency of two topologies (cf. Weston (8)).
Then 'pairwise Hausdorff' is exactly the same as 'consistent'. On the other
hand, regularity with respect to some other topology is not the same as
the property of being coupled to that topology, although the definitions
are superficially similar.
It is interesting to note that if 0 is regular with respect to 2 and 2 is
coupled to ^ , then ^<= 2.~\ Thus, if 3P and 2 are pairwise regular and
t My thanks are due to the referee for pointing out both this fact and the applica-
tion mentioned here, and also for correcting numerous details elsewhere in the paper.
74 J. C. KELLY
coupled to each other, 0> = 3 and the resulting single topology is regular.
This observation together with Theorem 3 of Weston (8) yield the follow-
ing application:
Let X be a group with unit element e. Suppose that (X, &) and (X, 2.)
are topological groups satisfying the conditions
(i) 2 is regular with respect to 8P,
(ii) X is of the second category relative to 0>,
(iii) corresponding to each ^-neighbourhood Q of e, there is a sequence
{(f),,} of ^-homeomorphisms of X onto itself such that X is covered
by the sets <f>n[,9-c\ (Q)] (n = 1, 2,...).
Then 2 g &. If, in addition, the 'dual5 conditions (obtained from (i), (ii)
and (iii) by interchanging & and 2) hold, then 2 = 0*. For ways in which
condition (iii) can be satisfied, see the remarks following Theorem 3 in (8).
2.6. DEFINITION. A space (X, ^ , 3) is said to be pairwise normal if,
given a ^-closed set A and a ^-closed set B with A C\B = 0, there exist
a J-open set U and a ,^-open set V such that A g U, B c V, and U n V = 0.
Equivalently, (X, &, £) is pairwise normal if, given a J-closed set C and
a ^-open set D such that C^D, there are a ^-open set G and a J-closed
set F such that

In the preceding definitions, when considering a space and one of its

topologies, we have employed the standard terminology of Kelley (5),
using prefixes 0P-, J2-,... in the obvious way to indicate the topology being
used. We shall continue to do this in what follows. On the other hand,
we shall sometimes say that a bitopological space {X, 0>, 2.) has a particular
topological property, without referring specifically to & or 2, and we shall
then mean that both 8P and 2L have the property; for instance, {X, 0>, 2.) is
said to satisfy the second axiom of countability if both 0> and 2. do so.
The first result we obtain is the following generalization of Urysohn's
lemma:
2.7. THEOREM. / / (X,^,^) is pairwise normal, then given a 2-closed
set F and a ^-closed set H with F CiH = 0, there exists a real-valued function
g on X such that
(i) g(x) = 0 (x eF), g(x) = 1 (xeH), and 0 ^ g(x) ^ 1 {xeX),
(ii) g is @-upper semi-continuous and 2-lower semi-continuous.
On the basis of this result, generalizations to bitopological spaces of
Urysohn's metrization theorem and Tietze's extension theorem are
obtained. We say that a bitopological space (X, SP, 3) is quasi-metrizable
if there is a pair p( , ) and q( , ) of conjugate quasi-metrics such that
 B I T O P O L O G I C A L SPACES 75
0 and 2. are determined by p( , ) and q( , ) respectively. Quasi-pseudo-
metrizability is defined similarly. The quasi-metrization theorem is the
following one:
2.8. THEOREM. Let (X,^,^) be a pairwise regular bitopological space
satisfying the second axiom of countability. Then (X, &>, J?) is quasi-
pseudo-metrizable. If in addition (X,&,£>) is pairwise Hausdorff, it is
quasi -metrizable.
The generalization of Tietze's extension theorem is as follows:
2.9. THEOREM. Let (X ,£?,£) be a pairwise normal bitopological space.
Let i g l be 0-closed and 2,-closed. Let f be a real-valued function defined
on A which is 8P-upper semi-continuous and Q-lower semi-continuous.
Then there exists an extension F off to the whole of X such that F is &-upper
semi-continuous and .2 -lower semi-continuous. If f is bounded, then the
extension F can be chosen so that
} and
:teX} = sup{/(0 : teA}.
We consider in a later section some of the topological properties of
quasi-pseudo-metric spaces. For example, if (X,p,q) is such a space, the
topologies 0 and J2 of this space have the properties of pairwise regularity
and pairwise normality. The distance functions p( , ) and q( , ) have
semi-continuity properties relative to 0 and 2.. The most interesting
result found is one which concerns completeness in a space (X,p,q), and
which generalizes the Baire category theorem for pseudo-metric spaces.
We make the following definitions:
2.10. DEFINITIONS. A sequence {xn} in a quasi-pseudo-metric space
(X,p,q) is a p-Cauchy sequence if, given e>0, there exists an integer m
such that
p(xr,xs)<e {r>s>m).
A subset A of X is p-complete if every ^-Cauchy sequence in A has a
p-limit in A.
2.11. THEOREM. Let (X,p,q) be a quasi-pseudo-metric space. Then X
is of the second category in itself with respect to the topology @ if X is
q-complete.
Naturally, any result stated in terms of £P and 2. has a 'dual', in terms
of J2 and 0>. The definitions of separation properties of two topologies
8P and 2, such as pairwise regularity, of course reduce to the usual
separation properties of one topology 0>, such as regularity, when we
76 J . C. KELLY
take 0* = .2, and the theorems quoted above then yield as corollaries the
classical results of which they are generalizations.
3. Proofs of topological theorems
3.1. Proof of Theorem 2.7. Let (X,^,2.) be a pairwise normal
bitopological space. Let G and H be subsets of X such that G is ^-closed,
H is ^-closed, and G n H = 0.
Let GQ = G and let Kx = X\H. Then GQ is ^-closed, Kx is ^-open, and
GO^KV Since {X,^,M) is pairwise normal, there exist a ^-open set K\
and a J2-closed set G± such that ^ £ ^ 9 ^ 9 ^ , Applying our hypothesis
on (X, 0>, J2) to each pair of sets Go, K^ and G±, Kv we obtain ^-open sets
Kiy Ki and J-closed sets Gif G$ such that

Continuing this process, we obtain two families {Gs} and {Ks}, where
s = p/29 (p = 1, 2,..., 29 — 1; q = 1, 2,...); if s is any other dyadic rational,
let Ks = 0 (s^O), if, = X (s> 1), and G, = 0 (s<0), Gs = X (a&l). Then
Kf^K^G^Gt {r^s^t), and C?a^Kt{s<t). Let gr be the function defined
on X to [0,1] by
g(x) = inf {t:xeKt} {xeX).
Then g(x) = inf{t :xeGt} (xeX).
Clearly, 0^g(x)^l (XGX), g(x) = 0 (xeG),
and gr(a;) = 1 (a; e [X\KX] = ,&).
Just as in the proof of the classical Urysohn lemma, we can show, using
the sets Ks, that g is ^-upper semi-continuous, and, using the sets Gs,
that g is -2-lower semi-continuous. This completes the proof.
Before proving Theorem 2.8, we need the following elementary lemma.
3.2. LEMMA. A pairwise regular bitopological space (X, &>, 21) satisfying
the second axiom of countability is pairwise normal.
Proof. Let {Pn} and {Qn} (n = 1,2,...) be countable bases for 0> and 2.
respectively. Let A be a ^-closed set and B a ^-closed set with

Since (X, &, &) is pairwise regular, for each x in A there is a .2-open set Q,
Q 6 {Qn}, such that

Similarly, for each y in B, there is a ^-open set P, Pe{Pn}, such that

yEP^J2-c\(P)^[X\A]. Proceeding as in the analogous classical lemma
(see e.g. ((5) 113)), we obtain a countable J-open cover of A and a
 BITOPOLOGICAL SPACES 77
countable ^-open cover of B which can be used to construct the required
disjoint sets U and V such that U contains A and is .2-open, whilst V
contains B and is
3.3. Proof of Theorem 2.8. Let (X, &,£) be a pairwise regular bitopo-
logical space satisfying the second axiom of countability. Let {P,,} and
{Qn} in — 1)2,...) be countable bases for 0> and 2. respectively. By the
preceding lemma, (X,£?,J2) is pairwise normal.
Since {X, 0>, 2,) is pairwise regular, for each Pn of the base {Pn} for &
and for each point x of P n , there is a Pm in {Pn} such that

Let the pairs (Pm,Pn) of members of {Pn} such that J2-C1 {Pin)^Pn be
enumerated as {Pmk,Pnk}, and set Sk = Pmk, Tk = Pnk (k = 1, 2, ...)• Then
by Theorem 2.7, for each positive integer k there is a real-valued function
gk on X such that
gk(x) = 0 (xe£-oL[8k])t gk(x) = 1 (ze[X\Tk])t
0^gk(x)^l {xeX),
and gk is ^-upper semi-continuous and .2-lower semi-continuous.
Let
Pi(x* y) = i (i)*&fc(y) - ^(^)} + (
where
{9k(v) - 9k(x)}+ = m a x [0, gk{y) - gk(x)] {k = 1, 2,...).
Then, for fixed x, p-i{x,y) is a ^-upper semi-continuous and .2-lower
semi-continuous function of y. I t is almost obvious that pt( , ) is a quasi-
pseudo-metric. Let ^ denote the topology of jp1( , ). Just as in the proof
of the classical Urysohn metrization theorem, we can show that 0* = 3PV
Now let

Then pt( , ) and qx{ , ) are conjugate quasi-pseudo-metrics. For fixed x,

qx(x, y) is a ^-upper semi-continuous and ^-lower semi-continuous function
of y. In particular, {y : qx{x,y) < t] is ^-open, for each t > 0. Thus, if Qx
is the topology of qx{ , ), 2LX c ^ .
By the same method applied to the base {Qn} for .2, we can construct a
family {fk} (k = 1,2,...) of ^-lower semi-continuous and J-upper semi-
continuous functions such that, if
78 J. C. KELLY

an
the topologies J 2 d ^2 °f <?2( > ) a n d Pz( > ) a r e sucn
that J 2 = 2. and

For the set X, let

= 2 (*)%&)-&(*)}++ S
and &(&, y) = qx(x, y) + q2(x, y)

Then the topologies determined by the conjugate quasi-pseudo-metrics

j»3( , ) and <73( , ) are precisely 0> and SL.
It is easily verifiable that if (X, 0>,2L) is pairwise Hausdorff it is quasi -
metrizable. The proof is complete.
We consider next Theorem 2.9, on the extension of a function defined
on a subset of a bitopological space. Unlike the results already proved,
this theorem cannot be obtained by adapting the proof of the correspond-
ing classical theorem (see e.g. ((5) 242)). The usual method of proving
the classical result uses differences of continuous functions, which are
themselves continuous. The functions considered here are semi-continuous,
and taking differences does not preserve semi-continuity. We adopt the
approach used by Bonsall in considering monotonic functions; the proof
of Theorem 2.9 is a trivial modification of Bonsall's method of extending
a bounded continuous monotonic function defined on a subset of a partially
ordered topological space.
Two preliminary results are needed. The first of these is elementary
and its proof is omitted.
3.4. LEMMA. The class UL(X) of real-valued functions which are
&>-upper semi-continuous and Q-lower semi-continuous on a space (X, 0>, 2.)
is complete with respect to the uniform norm on X.
3.5. LEMMA. Let U(A) be the class of real-valued functions which are
defined and 3~-upper semi-continuous on a subset A of a topological space
(X,,T). Let EU(A) be the subclass of U(A) consisting of those functions
which have a 2F-upper semi-continuous extension to X. Then EU(A) is
complete with respect to the uniform norm on A.
Proof. Let {/„} be a Cauchy sequence in EU(A) with respect to the
uniform norm on A. There is a function / defined on A which is the
uniform limit on A of {/„} and such that feU(A). We assert t h a t / has
an extension F to X which is ^"-upper semi-continuous.
 BITOPOLOGICAL SPACES 79
Given e > 0, there is a subsequence {/nJ of {/„}, depending on e, such
that
l/«t(*)-/n,(*) I <«/2* (j>k,k=l,2,...;xeA).
For each positive integer k, there exists a ^-upper semi-continuous
extension of fnk to X, say gk. Let
Fx{x) = gi(x),

F2(x) = [gz(x) u {Fx(x) - e/2}] n {Fx(x) + e/2},...,

Fk+1(x) = [gk+1(x) U {Fk(x) - e/2*}] n {Fk(x) + c/2*},
for all x in X, where Fk. is defined by induction on k. The symbols U and n
have their usual meaning of maximum and minimum. Since U and n pre-
serve semi-continuity, {Fk} is a sequence of 5~-upper semi-continuous
functions on X. We have
F1(x) = g1(x)=fni(x) (xeA),
WW-FiW^cft (xeX),
and
F2(x) = g2{x) = fnt(x) (xeA).
In general,
l^^)-^-,-!^)^^ (xeX),
and
Fk(x) = g,c(x)=fnk(x) (xeA).
Then
Fk[x) - Fk+p(x) | = | Fk(x) - FM(x) + Fk+1(x) -...+ F^Jx) - Fk+V(x) \

f/2M (xeX).

Thus {Fk} is a Cauchy sequence with respect to the uniform norm on the
whole space X. Hence {Fk} converges to a ^"-upper semi-continuous
function F defined on X. It is clear that/(a;) = F(x) (xeA).
3.6. THEOREM. Let A be a subset of a bitopological space (X, 0>, &). Let
UL(A) be the class of all real-valued functions on A which are £P-upper
semi-continuous and 2,-lower semi-continuous. Let EUL(A) be the subclass
of UL(A) consisting of those functions which have a &-upper semi-continuous
and 2,-lower semi-continuous extension to X. Then EUL(A) is complete
with respect to the uniform norm on A.
Proof. The same as that of the preceding lemma, using Lemma 3.4
and functions which are ^ - u p p e r semi-continuous and J2-lower semi-
continuous instead of ^"-upper semi-continuous.
80 J . C. K E L L Y
3.7. Proof of Theorem 2.9. Let (X,^,^,) be a pairwise normal
bitopological space, and A a ^-closed and j2-closed subset of X.
Suppose/is a real-valued function on A which is ^ - u p p e r semi-continuous
and J2-lower semi-continuous. We show that / has an extension F to X
with the required properties.
Let n be a positive integer. For each integer k, positive, negative, or
zero,let
:f(x)>-\ and
n
Then for every integer k, Ukn and Lkn are respectively ^-closed and
^-closed subsets of A, and so of X. Also, Ukn n Lkn = 0.
By Theorem 2.7, there is, for each k = 0,1,2,..., a function uk defined
on X which is ^-upper semi-continuous and .2-lower semi-continuous and
such that
uk(x) = 0 {xeLkn), uk(x) = - (xeUkn),
and
O^uJx)^- (xeX).
Also, there is, for each k = 0, — 1, — 2,..., a function vk defined on X
which is ^-upper semi-continuous and .S-lower semi-continuous and such
that
vk(x) = (xeL,.n), v,.(x) = 0 (xeUkn),
and
Z±<vk(x)<0 (XGX).

Now Lkn g Lfa and U^ e Ukn. Thus,

if f{x)^0 then vk(x) = 0 (k = 0, - 1, - 2 , . . . ) ;
if / ( z K O then uk(x) = 0 (k = 1,2,...).
We define a real-valued function fn on X as follows:
00 0

&—1 fc— -oo

/ n is ^-upper semi-continuous and J2-Iower semi-continuous.

Let

Thus

and
= 0,±l,±2,...}.
 BITOPOLOGICAL SPACES 81
l
Consider the value of fn(x) on Ck , for some particular k. There are two
cases, according as k ^ 1 or k ^ 0.
(i) k>\. Let zeCb« = U&
Since

since
1
n
also, : k Thus,
n'
(t-D
Jn\x) ~ n
and, by definition

(ii) k^O. Letting xeCkn, we can show in a similar way that

where A; is a non-positive integer, and so

Combining (i) and (ii), we have

It is clear that the restrictions fn\ A (n = 1,2,...) converge uniformly

t o / o n A, and form a Cauchy sequence with respect to the uniform norm
on A. By Theorem 3.6, / has an extension F to X which is ^-upper
semi-continuous and .2-lower semi-continuous.
Finally, if/ is bounded, let the infimum and supremum of / on A be
a and b respectively. Since the constant functions are continuous, the
function F' defined by
F'(x) = {a u F(x)} n 6 (XEX)
is an extension of/ with the required properties.
4. Quasi-pseudo-metrics
Let (X,p,q) denote an arbitrary quasi-pseudo-metric space. The
following elementary results state some basic facts about quasi-pseudo-
metrics in general.
5388.3.13 G
82 J. C. KELLY
4.1. PROPOSITION. For a fixed point x in X, p{x,y) is a 8P-upper
semi-continuous and 2,-lower semi-continuous function of y; and for a fixed
point y in X, p{x,y) is a &-lower semi-continuous and M-upper semi-
continuous function of x.
The proof depends on showing that, for every x in X, the sets
{y : p{x,y) < k] and {y : p{x,y) ^ k} are respectively ,^-open and ^-closed.
We omit the details.
4.2. PROPOSITION. Let & and 2, be the topologies on X determined by
p( , ) and q( , ) respectively. Then (X, SP, Q) is pairwise regular and
pairwise normal. (X, 0*, 2L) is pairwise Hausdorff if and only if p{ , ) and
q( , ) are quasi-metrics.
Proof, (i) Since, for every x in X and each k > 0, the set {y : p(x, y) ^ k)
is ^-closed, each point x has a ^-neighbourhood base of ^-closed sets.
Hence & is regular with respect to 2.. Similarly, J2 is regular with
respect to &.
(ii) Let A and B be disjoint subsets of X such that A is ^"-closed
and B is ^-closed.
Define, for xm X,
p(x,A) = inf{p(x,a) : aeA}
and q(x,B) = ini{q(x,b) :beB}.
Then A = {y : p(y, A) = 0} and B = {y : q(y, B) = 0}.
Let U= {x:p{x,A)<q(x,B)}
and V = {x : q(x, B) < p{x, A)}.
Then U n V = 0, A S U, B c V, and we can establish pairwise normality
by showing that U and V are respectively i?-open and
We show that V is ^-open. Suppose that XOEV and that
p{xQ,A)-q{x0,B) = k>0.
Let xeP(xo,kl±). Then

= p{xQ,x) + q(x0,B),
and p(xQ,A)^p{x0,x)+p{x,A).
Thus p{x, A) - q(x, B) >p(x0, A) - q{xQ, B) - 2p(x0, x) > fc/2,
and so x e V, that is, P(x0,fc/4)c V.
Clearly, (X, &>, SI) is pairwise Hausdorff if and only if p( , ) and q( ,
are quasi-metrics.
 BITOPOLOGICAL SPACES 83
It may happen that p(x, y) is a continuous function of x or of y relative
to either & or 2.. In showing some consequences of assuming continuity
properties, we use the following elementary result:
4.3. LEMMA. / / @P is a larger topology than 2, then & is pseudo-metrizable.
Proof. Let d(x,y) = max {p{x,y),q(x,y)} (x,yeX). Then the topology
2 of d( , ) is the smallest topology larger than both & and 2. Thus

We remark that it is enough here to assume that & is coupled to 2. For

2 is regular with respect to ^ , and we then have 2 c 0>.
4.4. PROPOSITION, (i) //, for each x in X, the function p(x,y) is
^-continuous in y, then 8P is regular.
(ii) //, for each x in X, the function p(x,y) is 2-continuous in y, then 2
is pseudo-metrizable.
Proof, (i) It is enough to note that ^-continuity of p(x,y) implies that
{y : p(x, y) < k) is a ^"-closed set (k > 0).
(ii) If p(x, y) is ^-continuous then each open ^-sphere P(x, k) is j2-open.
Thus 2 is larger than 3P and the preceding lemma applies.
In the rest of this section, we consider completeness in a space
(X,p,q).
4.5. Proof of Theorem 2.11. Suppose that X is of the first category in
itself relative to 0». Then X = (J {Xn : n = 1,2,...}, where each Xu is
nowhere dense in X, relative to 8P. Throughout this proof, cl ( ) is under-
stood to be ^-cl ( ). Recall that A is nowhere dense in X, relative to 8P,
if cl (A) has an empty interior. Let P'(x, k) = {y : p(x, y) < k) (xeX, k > 0).
Let us choose xx e [X\c\ (Xx)]. Then there is ex with 0 < ex < 1 such that
Xx n P'(xv ex) = 0 . The open set P(x1} et) is not contained in cl (X2), since
X2 is nowhere dense in X. Thus there is x2, with x2eP(x1) ex) and e2, with
0 < e2 < \, such that

Continuing in this way, we construct the sequences {xn} and {en} such
that xn e X, 0 < en < I/n}
P'(Xn+l>€n+l)^P(Xn>en),
and XnnP'(xn,en) =0 (n=l,2,...).
Thus, if m>n, P(x7n,em)^P(xn,€n), and therefore p(xn,x,n)<€n, and so
q(xm,xn) <en (n — 1, 2,...). Hence {xn} is a ^-Cauchy sequence in X, which
by hypothesis has a (/-limit xQ in X. I t follows easily that q(x0, xn) ^ en,
84 J. C. KELLY
that is, p(xn,x0)^en (n= 1,2,...). Then xQeP'(xnien), and so xo$Xn
(n = 1,2,...). This is contradictory, since X = (J {Xn : n = 1,2,...}. The
proof is complete.

The definition of ^-Cauchy sequence in a space (X,p,q), which we have

given in § 2.10, coincides with the usual definition when p( , ) is a pseudo-
metric. Recall that if p( , ) is a pseudo-metric, {xn} is a ^-Cauchy sequence
if, given e>0, there exists an integer m such that p(xr,xs)<€ (r,s^m).
This definition, if carried over to quasi-pseudo-metrics, does not seem
natural; for, if p( , ) and q( , ) are conjugate quasi-pseudo-metrics, and
d{x, y) = max [p(x, y), q(x, y)] (x, yeX),
a p -Cauchy sequence would then be simply a Cauchy sequence relative to
the pseudo-metric d( , ). It seems that no definition is entirely satis-
factory. Definition 2.10 has the disadvantage that a p-convergent
sequence need not be ^-Cauchy (see counter-example 5.8).
There remains the problem of completing a quasi-pseudo-metric space
(X,p, q) with respect to p( , ), or, more precisely, of finding a w-complete
space {X*, u, v) such that (X,p) is homeomorphic with a ^-dense subspace
of [X*, u). I have not been able to find a satisfactory equivalence relation
on the set of #>-Cauchy sequences in (X,p,q).

5. Examples
We consider first quasi-pseudo-metrics on quasi-ordered sets. Let
(X, d, ^ ) denote a set X on which are defined a pseudo-metric d( , ) and
a quasi-order ^ . (For the terminology and notation used for quasi-orders
and partial orders see e.g. (2).) (X,d, ^ ) is said to be a quasi-ordered
pseudo-metric space. We may assume that
d{x,y)>l (x,yeX).
Let A( , ) be the function o n l x l defined by
X(x,y) = 0 if x^y,
= 1 otherwise (x,yeX).
Let /x( , ) be defined by
fi(x, y) = 0 if x > y,
= 1 otherwise (x,yeX).
Then A( , ) and JJL( , ) are conjugate quasi-pseudo-metrics for X. We
use A( , ) and /x( , ) to define two pairs of less trivial quasi-pseudo-
metrics for X.
 BITOPOLOGICAL SPACES 85
We remark that if px{ , ) and p2( , ) are two (arbitrary) quasi-pseudo-
metrics for a set X, so are p3( , ) and pA( , ), defined as follows:
(i) Pzfay) = ™&x[Pi(x,y),p2(x,y)] {x,yeX). The topology determined
> ) is the smallest which is larger than the topologies determined by

(ii) Let r(x,y) = mm['pl(xty)i'pi{xiy)] (x,yeX). r( , ) does not

necessarily satisfy the triangle inequality. Let
n
2h(x> V) = inf Ii rfa, xi+1),
t= O

the infimum being taken over all finite sets xo,xv ...,xn+1 with x = xQ and
y = s»+i-
Consider again the space (X,d, ^ ).
5.1. DEFINITIONS. Let
p(x,y) = m&x[d(x,y),X(x,y)]
= d{x,y) if x^y,
= 1 otherwise,
and let q(x, y) = max [d(x, y), p(x, y)]
= p(y,x) (x,yeX).
Then p( , ) and q( , ) are conjugate quasi-pseudo-metrics for X. I call
the topologies & and J2 of p{ , ) and q( , ) the positive and negative cone
topologies respectively.
5.2. DEFINITIONS. Let
r(x,y) = min[d(x,y),X(x,y)]
and s{x, y) = min [d(x, y), n(x, y)] (x, yeX).
n
Let u(x, y) = inf £ r(xi} xi+1),
the infimum being taken over all finite sets xo,xv ...,xn+1 with x = x0
and y = xn+1. v( , ) is defined similarly in terms of s( , ). Then u( , )
and v( , ) are conjugate quasi-pseudo-metrics for X. Obviously, if x^y
then u(x, y) = v(y, x) = 0. I call the topologies % and Y" of u{ , ) and
•y( , ) the upper and Zower quasi-pseudo-metric topologies for X respectively.
These topologies are related to the two topologies determined by the
families of open upper sets and open lower sets respectively (see ((4) 136)).
5.3. EXAMPLE. Let (X,d, ^ ) be a partially ordered vector space such
that d( , ) is a translation-invariant metric. Then the positive cone
topology 3P determined by p( , ) on the semi-space of positive elements
86 J . C. KELLY
of X is precisely the semi-space topology. (For the definition of semi-
space topology, see ((3) 70-71).)
A sequence {xn} converges to xQ relative to 0> if and only if {x.n} converges
to x0 relative to the metric topology and xn^x0 {n = 1,2,...). {o;n} is a
j9-Cauchy sequence if and only if it is a d-Cauchy sequence and xn ^ xn+1
(TI= 1,2,...), that is, {xn} is a cZ-Cauchy monotonic decreasing sequence.
Similarly, a g-Cauchy sequence is a rf-Cauchy monotonic increasing
sequence. Let X+ = {x : xeX,x^ 0}. Then Theorem 2.11 asserts that X+
is of the second category relative to 0> if every monotonic increasing
cZ-Cauchy sequence in X+ has a limit in X+. This is Lemma 9.1 of Bonsall
((3) 70), stated for a semi-space.
5.4. EXAMPLE. The simplest example of a quasi-ordered pseudo-metric
space is the linearly ordered metric space formed by the set R of real
numbers with the usual order and metric. Let d( , ) be the metric for R
defined by
d(x,y) = mm[l,\x-y\] (x,yeR).
With the notation of Definitions 5.1 and 5.2 we define the quasi-pseudo-
metrics p{ , ), q{ , ) and u( , ), v( , ). Then
p{x,y) = d{x,y) if %^y,
=1 if x>y,
and q(x, y) = d(x, y)if x>y,
= 1 if
if x<y
x<y (x,yGR).
Also, u>{x,y) = 0 if x^y,
= d(x,y) if x>y,
and v(x, y) — 0 if x^y,
= d{x,y) ifif x<y
x<y {x,yeR).
The topology 0* determined by p( , ) is precisely the topology for R
which has as a base the family of all half-open intervals of the form
[a,b) = {x : a^x<b}. Similarly, SI has as a base the family of all half-
open intervals of the form (a, 6] = {x : a < x ^ 6}.
The topology tfl determined by u{ , ) is precisely the topology for R
which has as a base the family of all sets of the form {x : x> a}. Similarly,
"V has as a base the family of all sets of the form {x : x < a}. (The it-spheres
U(x, l/n) forming a base for °tt are the sets {y : y >x— 1/n}; the u-spheres
V(x, \jn) forming a base for 1^ are the sets {y : y<x+ l/n}.)
5.5. MONOTONIC FUNCTIONS. Let (X,^,^) be a pairwise regular
bitopological space. Clearly, the relation ^ , defined by 'x ^ y if and only
if xe£?-c\(y)' is a quasi-order. We show that XG^-C\{Z) if and only if
 BITOPOLOGICAL SPACES 87
zeJ£-6l(x). Then y^x if and only if yeJ2-c\(x). Let @(x) denote the
set of all ^-neighbourhoods of x in X. Let
P(x) = {y:xe 0>-d (y)} = (){G:Ge 3P{x)}
= fl {G : Ge0>{x) and 0 is ^-closed},
since 0> is regular with respect to SL. Thus P(x) is J-closed. Suppose
z ^ - c l ( z ) , i.e. z$P(x). Since $L is regular with respect to @, there exist
a j2-open set Q and a ^-open H such that ZGG, P(X)Cjy, and Gf)H = 0.
Thus 2 £ J2-C1 (X). Similarly, z <£ J-cl (a;) implies x $ 0>-c\ (z).
Let 5T be the smallest topology larger than both SP and J . If ^ and Q
are pairwise regular, it is straightforward to verify that &~ = (t? + J2) is
regular. (The notation ^ + J2 for y is due to Weston (8).) L e t / be a
real-valued function on X which is ^-lower semi-continuous and j2-upper
semi-continuous. Denoting by °U and "V the topologies for the real line
determined by the functions u( , ) and v( , ) respectively in § 5.4, this
means that / i s &->-'% and Sl^-'V continuous. Then the inverse image
of any open w-sphere U{f(x),e), with centre f(x) and radius e, contains a
^-neighbourhood of x, say P ; thus f[P] £ U(f{x), e). Then

If a; ^ y , f(y) e f) {U(f(x), e) :.e> 0}, i.e. f(x) ^f(y). Hence / is monotonic

increasing. Almost obviously, / is continuous relative to y .
If SP and 21 have countable bases, then Lemma 3.2 and Theorems 2.7 and
2.8 show that there are enough ^-lower semi-continuous and j2-upper
semi-continuous functions on X to determine the topologies SP and J2 by
conjugate quasi-pseudo-metrics p( , ) and q( , ) (see Proof 3.3). ST is
determined by the pseudo-metric d{ , ), where
d(x, y) = p(x, y) + q(x, y) (x, yeX).
The family 8F of ^"-lower semi-continuous and .S-upper semi-continuous
functions, determining Q, are monotonic increasing and continuous rela-
tive to 3T. The family ^ of ^-upper semi-continuous and ,,2-lower semi-
continuous functions, determining £P, are monotonic decreasing and
continuous relative to ST. Then ST is determined by ^U.^". Moreover,
the quasi-order ^ is determined by J5", in the sense that x ^ y if and only
if/(x) 4:f(y) (fe^). (Almost obviously, x^y if and only if p(x,y) = 0.)
In the rest of this section, we consider counter-examples to some
conjectures about quasi-pseudo-metric spaces. The topology 3P of a
quasi-pseudo-metric p( , ) on a set X need not be regular, as can be seen
88 J . C. KELLY
by considering the topologies °ll and "V on the real line. The following
example shows that.a quasi-metric need not have a normal topology.
5.6. COUNTER-EXAMPLE. Let (X,p,q) be the real line R with the quasi-
metrics p( , ) and q( , ) defined in § 5.4, whose topologies 0* and 2. are
the topologies with bases consisting of half-open intervals. Let RxR
have the product topology SP x SP determined by the quasi-metric A( , )
where

for each x,y in RxR, and x = (xvx2), y = (2/1,2/2)- Sorgenfrey (7) has
shown that & x & is not normal.
5.7. COUNTER-EXAMPLE. A quasi-metric determines a ^-topology, but
this need not be a Hausdorff topology (in contrast to the case of a metric).
Let R denote the set of real numbers, and let an element x of the product
space RxR be (xvx2). Let d( , ) be the pseudo-metric on RxR defined

d{x,y) = min[\x2-y2\,l] {x,yeRxR).

Let <; be the quasi-order on RxR defined by x^y if and only if either
(i) x = y, or (ii) x2<y2. Let p( , ) and q( , ) be the conjugate quasi-
pseudo-metrics for J? x R defined as in § 5.1. Then p( , ) has a ^-topology
& which is not Hausdorff.
5.8. COUNTER-EXAMPLE. In a quasi-pseudo-metric space {X,p,q), a
^-convergent sequence need not be a £>-Cauchy sequence.
Let X be the set of all sequences of real numbers having at most
one non-zero term. Let the zero sequence be denoted by 0. Let the
member of X having non-zero nth term x be denoted by x \n (n = 1,2,...).
For each n = 1,2,..., define
p(x\n,y\n) = min[\x — y\,l] if x^y,
= 1 if x>y.
n m
If m # n , define p(x \ , y \ ) — 1.
Then p( , ) is a quasi-metric on X. The sequence {xn} of elements of X,
defined by
1n
Xn = - (71=1,2,...),

converges to 0 relative to p( , ). Yet {xn} is not a #>-Cauchy sequence,

since

Finally, a p-Cauchy sequence need not be a g-Cauchy sequence, as may

be seen by examining the topologies 0P and SI of § 5.4 on the real line.
 B I T O P O L O G I C A L SPACES 89

REFERENCES
1. M. BALANZAT, 'On the metrisation of quasi-metric spaces', Gaz. Mat. Lisboa 12
no. 50 (1951) 91-4.
2. G. BIRKHOFF, Lattice theory, revised edition, American Math. Soc. Colloquium
Publications XXV (New York, 1948).
3. F. F. BONSALL, 'Linear operators in complete positive cones', Proc. London Math.
Soc. (3) 8 (1958) 53-75.
4. 'Semi-algebras of continuous functions', ibid. (3) 10 (1960) 122-40.
5. J. L. KELLEY, General topology (New York, 1955).
6. H. RIBEIRO, 'Sur les espaces a metrique faible', Portvgaliae Math. 4 (1943)
21-40 and 65-8.
7. R. H. SORGENFREY, 'The topological product of paracompact spaces', Bull.
American Math Soc. 53 (1947) 631-2.
8. J. D. WESTON, 'On the comparison of topologies', J. London Math. Soc. 32
(1957) 342-54.
9. W. A. WILSON, 'On quasi-metric spaces', American J. Math. 53 (1931) 675-84.
10. A. CSASZAR, Fondements de la topologie ginirale (Akade"miai Kiad6, Budapest,
1960).

The University
Hull