Irish Math. Soc.

Bulletin
Number 75, Summer 2015, 45–57
ISSN 0791-5578

ON THE CONTINUITY OF THE INVERSES OF

STRICTLY MONOTONIC FUNCTIONS

HEIKO HOFFMANN

Abstract. In this short note we present an elementary, but seem-

ingly not well known result on the continuity of the inverse of a
strictly monotonic function and we discuss the relation of this re-
sult to the question when order and subspace topology are identical,
both on the real line as well as in the abstract framework of con-
nected linearly ordered spaces.

1. Introduction.
It is a fundamental question in analysis under which conditions
the inverse of a continuous bijection, say between two topological
spaces, is itself continuous. There are well-known results like the
invariance of domain theorem or the classical (and easy to prove)
result that the inverse of a continuous bijection from a compact
space onto a Hausdorff space is also continuous; see also [5] for a
complete characterisation of all subsets of R such that every con-
tinuous injection defined on a set of this kind is a homeomorphism
onto its range.
It seems that results like the ones just mentioned have influenced
the presentation of similar results at the level of undergraduate
courses. So it seems that the following statement is most wide-
spread in such courses.
If ∅ 6= I ⊆ R is an interval and if f : I → R is continuous and
injective, then f −1 : f (I) → R is continuous, too.
Usually, the proofs given for this result make use of the continuity
of f in such a way that the continuity assumption appears to be
indispensable at a first cursory glance. However, there is a more
general result (see, e.g., [4, 37.1]), which, unfortunately, seems to be
seldom taught in undergraduate courses.
2010 Mathematics Subject Classification. 26A48.
Key words and phrases. continuous inverse, strictly monotonic function.
Received on 19-3-2015; revised 6-6-2015.
c 2015 Irish Mathematical Society
45
46 H. HOFFMANN

If ∅ =
6 I ⊆ R is an interval and if f : I → R is strictly monotonic,
then f −1 : f (I) → R is continuous, too.
This statement demonstrates that the premise of the continuity of
f is entirely superfluous (of course, injectivity must be replaced by
strict monotonicity) and proofs based on this premise might disguise
the deeper reason for this phenomenon. In fact, from the point of
view of topology, the true reason lies in the following observation
(readers not very well familiar with abstract topology may skip the
subsequent explanation at their first reading): a strictly monotonic
function f : I → f (I) is a homeomorphism if I and f (I) both carry
the order topology induced by the order inherited from R instead
of the usual subspace topology. Since the subspace topology is finer
than the order topology the mapping f −1 : f (I) → I is continuous if
f (I) is endowed with the subspace topology and I carries the order
topology. But since for intervals the order and subspace topology
coincide, we conclude that f −1 : f (I) → I is continous where I and
f (I) now both carry the usual subspace topology.
Clearly, the same argument works for every strictly monotonic
function f : A → R (∅ = 6 A ⊆ R) whenever the order and subspace
topology of A coincide. Unfortunately, the above proof (no matter
how simple it is) is in general out of reach for an undergraduate
course due to the topological conceptual framework. So at this point
three questions arise:
(1) Is there a simple (i.e., ideally so simple that it is easily ac-
cessible to undergraduate students with no knowledge of ab-
stract topology) description of those subsets of R for which
the order and subspace topology of A coincide?
(2) Is there an elementary proof for the above statement about
the continuity of the inverse of a strictly monotonic function
defined on such a set?
(3) Does there exist a subset of R such that each strictly mono-
tonic function defined on this set has a continuous inverse,
but the subspace and order topology on this set are distinct?
In this note we answer the first two questions affirmatively and we
present such an elementary proof, which might be easily incorpo-
rated into an undergraduate course. This proof is given in the next
section, where we choose a formulation that completely avoids men-
tion of the order topology and we get along only with notions easily
accessible to undergraduate students. Furthermore, we shall show
 CONTINUITY OF INVERSES 47

that this result is optimal in the sense that on each non-empty sub-
set of R for which order and subspace topology differ there exists
a strictly monotonic function whose inverse is not continuous, thus
giving a negative answer to the third question.
In the last section we take up once again the abstract topologist’s
position in order to complete our picture and to relate Proposition
2.1 and Proposition 2.4 below to the topological point of view de-
scribed above. This link is provided by Lemma 3.3, which in fact
answers the first of the above questions (see Corollary 3.5).

2. Strictly monotonic functions on subsets of R

In this section we do not want to presuppose that the reader is
familiar with abstract topology in order to make sure that this part
of the note is also readable, e.g., for undergraduate students. For
this reason we first clarify some notions occuring in what follows.
The symbol N denotes the set of strictly positive integers, while
˙
N0 := N∪{0}.
Let A be a subset of the reals R. A set C ⊆ A is called a (connected)
component of A if C is an interval (where we include the degenerate
cases of the empty set and singletons) and if each interval I ⊆ A
containing C already coincides with C. Each set A is the disjoint
union of all its connected components. This is most easily seen
by defining an equivalence relation on A by setting a ∼ a0 :⇐⇒
[min{a, a0 }, max{a, a0 }] ⊆ A for a, a0 ∈ A. Then the equivalence
classes of ∼ are precisely the connected components of A.
We call A an open set if for each a ∈ A there exists ε > 0 such that
(a − ε, a + ε) ⊆ A.
We denote by
∂A := {x ∈ R : ∀ε > 0 : (x − ε, x + ε) ∩ A 6= ∅ =
6 (x − ε, x + ε) \ A}
the boundary of A. The set A is closed if and only if ∂A ⊆ A. Notice
that A is closed if and only if R \ A is open.
A function f : A → R is continuous at a point a ∈ A if for every
sequence (an )n in A converging to A the sequence (f (an ))n converges
to f (a). The function f : A → R is continuous (on A) if it is
continuous at each point of A.
Now we can state and prove the announced result on the continuity
of the inverse of strictly monotonic functions.
48 H. HOFFMANN

Proposition 2.1. Let ∅ = 6 A ⊆ R be a set such that every bounded

component of R \ A is either closed or open. Furthermore, let f :
A → R be a strictly monotonic function on A. Then the function
f −1 : f (A) → R is continuous.
Proof. We suppose that f is strictly increasing (the case that f is
strictly decreasing can be treated in a similar way).
Let y0 ∈ f (A) be arbitrary. We want to show that f −1 : f (A) → R
is continuous at y0 . For this purpose, let (yn )n be an arbitrary
convergent sequence in f (A) with limit y0 . We then have to show
that (xn )n := (f −1 (yn ))n ∈ AN converges to x0 := f −1 (y0 ) ∈ A.
It is easy to verify that there are u, v ∈ f (A) with u ≤ v such
that yn ∈ [u, v] for all n ∈ N0 . We put a := f −1 (u) and b := f −1 (v).
Then we have xn ∈ [a, b] ∩ A for all n ∈ N0 . In particular, the
sequence (xn )n is bounded and it thus suffices to verify that x0 is its
only possible limit point in order to conclude that (xn )n converges
to x0 , which completes the proof. Indeed, suppose to the contrary
that x0 is the only possible limit point of the sequence (xn )n , but
this sequence does not converge to x0 . Then we may pass to a
subsequence (xnk )k such that |xnk − x0 | ≥ r for all k ∈ N and some
r > 0. Since (xnk )k is bounded as well, it has an accumulation
point thanks to the Bolzano-Weierstraß theorem, say x00 , and we
deduce |x00 − x0 | ≥ r, i.e., x00 6= x0 on the one hand. But on the
other hand x00 is also an accumulation point of the sequence (xn )n
itself and therefore x0 = x00 by hypothesis and so we end up with a
contradiction.
Suppose now that (xn )n possesses a limit point ξ different from
x0 and let (xnk )k be a subsequence converging to ξ. We then either
have ξ > x0 or ξ < x0 . We only treat the first case (the second one
is analogous) and we shall show that we obtain a contradiction.
First, assume additionally that ξ does not belong to A and denote
by I that component of R \ A that contains ξ. Observe that we have
ξ ∈ ∂I because of ξ ∈ ∂A.
If ξ is the left endpoint of I and if I is not a singleton, then
there exists a k0 ∈ N with xnk0 ∈ (x0 , ξ) and an index k1 ∈ N with
xnk ∈ (xnk0 , ξ) for all k ≥ k1 . This yields
ynk = f (xnk ) ≥ f (xnk0 ) > f (x0 ) = y0
for all k ≥ k1 . As k → ∞ we obtain the contradiction y0 ≥ f (xnk0 ) >
y0 .
 CONTINUITY OF INVERSES 49

If ξ is the right endpoint of I (which includes the case that I is a

singleton), then I must be bounded due to x0 < ξ. By assumption
I is either closed or open, but due to ξ ∈ ∂I ∩ (R \ A), the set I
must be closed. Therefore we then have I = [α, ξ] with an α ≤ ξ
such that α ∈ / A.
We may now choose an element z ∈ (x0 , α) ∩ A. (Note that this
is indeed possible: If α < ξ, this follows from α ∈ ∂A and x0 <
ξ, which yields x0 < α. If however α = ξ, then (x0 , ξ) ∩ A is
nonvoid since otherwise we would obtain (x0 , ξ] ⊆ I = {ξ}, which is
impossible.) There exists a k0 ∈ N such that xnk > z for all k ≥ k0 .
This implies
ynk = f (xnk ) ≥ f (z) > f (x0 ) = y0
for all k ≥ k0 and we arrive at the contradition y0 ≥ f (z) > y0 .
Summarizing, we infer that ξ must be an element of A. Here we
distinguish between two cases: (x0 , ξ) ∩ A 6= ∅ or (x0 , ξ) ∩ A = ∅.
In the first case we choose z ∈ (x0 , ξ) ∩ A and proceed as in the
above case where ξ was a right endpoint of the above I to arrive at
a contradiction.
So let us assume that (x0 , ξ) ∩ A = ∅. Then there exists a k0 ∈ N
such that xnk ≥ ξ for every k ≥ k0 . This yields ynk ∈ [f (ξ), ∞) for
each k ≥ k0 , which leads to the contradiction y0 ≥ f (ξ) > f (x0 ) =
y0 .
Altogether we arrive at the conclusion that ξ > x0 is not possible.

Proposition 2.1 gives rise to the following characterisation of the
continuity of a strictly monotonic function.
Corollary 2.2. Let ∅ = 6 A ⊆ R such that every bounded component
of R \ A is closed or open. Then for a strictly monotonic function
f : A → R the following assertions are equivalent.
(a) The function f : A → R is continuous.
(b) Each bounded component of R \ f (A) is closed or open.
If either assertion holds, then the sets A and f (A) are homeomor-
phic. In particular, f is continuous if its range f (A) is closed, open
or an interval. Moreover, the implication “(b) =⇒ (a)” is still true
if we drop the assumption imposed on A.
Proof. Applying Proposition 2.1 to the function f −1 : f (A) → R
gives us the implication “(b) =⇒ (a)”; even without the assumption
50 H. HOFFMANN

imposed on A.
Now assume that f is continuous as well as, without loss of general-
ity, that f strictly increases. Furthermore, suppose to the contrary
that R \ f (A) possesses a bounded component that is neither closed
nor open, thus having the form (u, v] or [u, v). We only treat the
first case.
Then u ∈ f (A), v ∈ / f (A) and there is a strictly decreasing se-
quence (yn )n in f (A) with limit v. We set xn := f −1 (yn ) for n ∈ N
and x := f −1 (u). The sequence (xn )n is strictly decreasing and
bounded from below by x, thus it converges to ξ := inf n∈N xn in R.
The number ξ does not belong to A since otherwise the continuity
of f would imply
v = lim yn = lim f (xn ) = f (ξ) ∈ f (A),
n→∞ n→∞
which is impossible because of v ∈/ f (A). Now consider an arbitrary
z ∈ A with z > x. We then have f (A) 3 f (z) > f (x) = u and
thus f (z) > v. Consequently, there exists an index n ∈ N with
v < yn = f (xn ) < f (z), which implies ξ < xn < z. We conclude that
(x, ξ] is a component of R \ A (because x ∈ A and A 3 xn → ξ ∈ /A
as n → ∞), which contradicts the assumption on A.
The first part of addendum is clear by Proposition 2.1. 
Remark 2.3. The characterisation of the continuity of a strictly
monotonic function obtained in the preceding corollary fails if the
adverb “strictly” is dropped. Indeed, just consider the function
f : { n1 ; n ∈ N} ∪ {0} → R given by f (0) := 0 and f ( n1 ) = 1 (n ∈ N).
As announced we now demonstrate that Proposition 2.1 is in some
sense optimal.
Proposition 2.4. Let ∅ = 6 A ⊆ R be a set such that R \ A possesses
a bounded component that is neither closed nor open. Then there
exists a strictly monotonic, continuous function f : A → R such
that the function f −1 : f (A) → R is discontinuous.
Proof. By assumption R \ A possesses a bounded component having
the form (a, b] or [a, b) (with a < b). We only consider the first case
since the second case is analogous.
Clearly, b is a cluster point of (b, ∞) ∩ A. Therefore we can choose
a strictly decreasing sequence (xn )n in A converging to b. Moreover,
we choose a strictly decreasing sequence (yn )n in R with limit a.
 CONTINUITY OF INVERSES 51

Now we put g(xn ) := yn for n ∈ N and g(a) := a and we extend

g on (xn+1 , xn ) linearly. This gives us a strictly increasing, contin-
uous function g : {a} ∪ (b, x1 ] → R, which we extend to a strictly
increasing, continuous function g : (−∞, a] ∪ (b, ∞) → R in any
way. Then the function f := g|A (note that A ⊆ (−∞, a] ∪ (b, ∞))
is strictly increasing and continuous, but its inverse is discontinuous
at a. In fact, we calculate limn→∞ f (xn ) = limn→∞ yn = a = f (a),
while limn→∞ f −1 (yn ) = limn→∞ xn = b 6= a = f −1 (a). 
Remark 2.5. (a) Proposition 2.1 and Proposition 2.4 together
characterise those nonvoid subsets A of R such that each
(continuous) strictly monotonic function f : A → R possesses
a continuous inverse. These are precisely those non-empty
sets A such that all bounded connected components of R \ A
are closed or open.
(b) By Proposition 2.1, the function g|(−∞,a)∪(b,∞) (where g is
as in the proof of Proposition 2.4) has a continuous inverse.
Therefore the point a is the only discontinuity of the above
f −1 .
(c) Combined with the order topological considerations in the
introduction, Proposition 2.4 furnishes a proof that the order
and subspace topology of A do not coincide whenever R \ A
possesses a bounded component that is neither closed nor
open. The converse is also true, see Corollary 3.5 below in
the next section.

3. Strictly monotonic functions on subsets of

connected linearly ordered spaces
In this section we want to go beyond the scope of real functions
and embed the results of the preceding section into a more general
framework in order to supply the topological background underlying
these results. In particular, we want to explore whether there exists
a reasonable generalisation of Proposition 2.1. As we shall see, it
turns out that there is indeed a perfect analogue in the setting of
connected linearly ordered spaces (see Proposition 3.7 below).
We presuppose from now on that the reader is acquainted with the
most basic notions of abstract topology. Nevertheless we start by
reviewing some important notions.
In what follows (X, ≤X ) and (Y, ≤Y ) are linearly ordered sets, both
52 H. HOFFMANN

endowed with their respective order topology τ(X,≤X ) and τ(Y,≤Y ) . If

no confusion is to be expected, we drop the indices and simply write
≤.
Let ∅ =
6 A ⊆ X. We may endow A with two reasonable topologies:
the subspace topology, denoted by τA , generated by all sets of the
form (−∞, x) ∩ A or (x, ∞) ∩ A where x ∈ X, and the topology
induced by the order on A inherited from X, denoted by τ(A,≤) ,
generated by all sets of the form (−∞, a) ∩ A or (a, ∞) ∩ A where
a ∈ A. We always have τ(A,≤) ⊆ τA , but this inclusion can be strict.
Notice that τ(A,≤) = τA if, for instance, A is compact with respect
to the subspace topology or an interval (see, e.g., [2, 4A2R (m)]).
We adopt the usual convention to write (−∞, x) resp. (x, ∞) resp.
(−∞, x] resp. [x, ∞) instead of {x0 ∈ X : x0 < x} resp. {x0 ∈ X :
x0 > x} resp. {x0 ∈ X : x0 ≤ x} resp. {x0 ∈ X : x0 ≥ x}. We write
sup A = ∞ if A is not bounded from above and inf A = −∞ if A
is not bounded from below and by convention −∞ < x < ∞ for all
x ∈ X (even if (X, ≤) has a minimum or maximum; in particular
±∞ ∈ / X).
Recall that (X, ≤) is called Dedekind complete if every non-empty
subset A of X with an upper bound has a least upper bound denoted
by sup A. If (X, ≤) is Dedekind complete and A is a nonvoid subset
of X with a lower bound, then A possesses a greatest lower bound
denoted by inf A (see, e.g., [1, 314B (b)]).
One says that (X, ≤) is dense provided that for any two elements
x, x0 ∈ X with x < x0 there exists an x00 ∈ X such that x < x00 < x0 .
A subset A of X is called (order-) convex if for any two elements
a, a0 ∈ A the interval [min(a, a0 ), max(a, a0 )] is contained in A.
The next lemma collects some basic facts concerning order topolo-
gies, which are probably folklore. For this reason we omit the easy
proofs (see also Exercise 26G in [7] for assertion (a))

Lemma 3.1. Let (X, ≤) be a linearly ordered set endowed with the
order topology and ∅ =
6 A ⊆ X.

(a) The space X is connected if and only if (X, ≤) is Dedekind

complete and dense.
(b) If A is connected w.r.t. τA , then it is convex.
(c) If A is an interval, then A is convex. The converse is true
provided that (X, ≤) is Dedekind complete.
 CONTINUITY OF INVERSES 53

(d) If (X, ≤) is connected and A is an interval, then A is con-

nected w.r.t. τA .
(e) If (X, ≤) is connected, then the set of subsets of X that are
connected (w.r.t. the subspace topology) coincides with the
set of intervals.
Example 3.2. Subintervals of R, the extended real line R ∪ {±∞},
the long line (see, e.g., counterexample 46 in [6]), the extended long
line (see, e.g., counterexample 46 in [6]), the unit square with the
lexicographical order (see, e.g., counterexample 48 in [6]) or lexi-
cographic cubes (see section 2 of [3]) are examples for connected
linearly ordered spaces (X, ≤).
For a connected linearly ordered space (X, ≤) the subsequent
lemma provides a catchy characterisation of all subsets of X for
which the subspace and order topology coincide. Moreover, the fol-
lowing Lemma 3.3 (resp. Corollary 3.5) links Proposition 2.1 and
Proposition 2.4 to the topological consideration from the introduc-
tion and completes our picture. In fact, using Lemma 3.3 we obtain
a perfect generalisation of Proposition 2.1 as we shall see later on.
Lemma 3.3. Let (X, ≤) be connected and ∅ = 6 A ⊆ X. Then the
order and subspace topology of A coincide if and only if every com-
ponent of X \ A w.r.t. the subspace topology τA is closed or open.
Proof. We first suppose that X \ A possesses a component which is
neither closed nor open. Thanks to part (e) of Lemma 3.1, there are
a, b ∈ X with a < b such that either [a, b) or (a, b] is a component
of X \ A. (Notice that all other kinds of intervals are surely closed
or open.) We only treat the first case since the second one can be
handled analogously.
We first observe that a cannot be the least element of (X, ≤)
(provided there exist any at all) because otherwise [a, b) would be
open. As a consequence, the set (−∞, a) is nonvoid. In addition,
(t, a) ∩ A 6= ∅ for each t ∈ (−∞, a). The latter assertion results from
the fact that for a point t ∈ (−∞, a) with (t, a) ∩ A = ∅ one would
obtain (t, b) ⊆ X \ A. Since (t, b) is connected by Lemma 3.1 (d)
and [a, b) is a connected component of X \ A with [a, b) ∩ (t, b) 6= ∅,
we infer (t, b) ⊆ [a, b). But as (X, ≤) is dense, the set (t, a) is non-
empty. This yields (−∞, a) ∩ [a, b) 6= ∅, which is absurd. We now
put A := {t ∈ X : t < a}, we let  denote the partial order ≤ on
A and we choose xt ∈ (t, a) ∩ A for each t ∈ A. Then (A, ) is an
54 H. HOFFMANN

upwards directed (nonvoid) set and (xt )t∈A is a net in A ∩ (−∞, a)

that converges in (X, ≤) to a, as one easily verifies. In particular,
for every x ∈ A with x < b, which implies x < a, resp. for each
x0 ∈ A with x0 > b, there is a t0 ∈ A with xt ∈ (x, ∞) ∩ A, resp.
with xt ∈ (−∞, x0 ) ∩ A for all t  t0 . Therefore (xt )t∈A converges to
b with respect to the order topology on A.
If the order topology and the subspace topology of A coincided,
then we could infer that (xt )t∈A converges to b with respect to the
subspace topology on A, hence in (X, ≤), which would yield a = b
(the order topology is always Hausdorff, see, e.g., [2, 4A2R (c)]) in
contrast to a < b. As a result, the subspace topology of A is strictly
finer than the order topology of A. This establishes the only-if-part.
Now we conversely assume that each component of X \ A is either
closed or open. In order to show that in this case the order and
subspace topology of A are identical, it suffices to verify that each
set of the form (−∞, ξ) ∩ A or (ξ, ∞) ∩ A, where ξ ∈ X, is open
with respect to the order topology on A. We show this only for
(−∞, ξ) ∩ A because the remaining case can be treated similarly.
In the cases ξ ∈ A, (−∞, ξ) ∩ A = ∅ or (−∞, ξ) ∩ A = A
the assertion is clear. Therefore we may assume that ξ ∈ / A and
(−∞, ξ) ∩ A 6= ∅ and (−∞, ξ) ∩ A 6= A or equivalently that ξ ∈ /A
and (−∞, ξ) ∩ A 6= ∅ and (ξ, ∞) ∩ A 6= ∅. We denote by I that
component of X \ A that contains ξ. Due to (−∞, ξ) ∩ A 6= ∅,
(ξ, ∞) ∩ A 6= ∅ and part (b) of Lemma 3.1, the set I is bounded
from above and from below. Thanks to Lemma 3.1 (e) the set I is
an interval and we thus deduce that there are a, b ∈ X such that
I ∈ {(a, b], [a, b), [a, b], (a, b)}.
We next show that none of the cases I = [a, b) or I = (a, b] can
occur. We establish this claim only for the first case because an
analogous argument works in the second one. The same argument
as utilised above in the proof of the only-if-part gives us a net (xt )t∈A
in A ∩ (−∞, a) that converges in (X, ≤) to a. (For this notice that
(−∞, a) is non-empty because A ∩ (−∞, ξ) 6= ∅ and [a, ξ] ⊆ X \ A.)
By hypothesis, I is closed or open. If I were open, then there would
exist a t0 ∈ A such that xt ∈ I for all t  t0 , which is impossible
because of (−∞, a) ∩ I = ∅. Hence, I is closed. Employing that
(X, ≤) is dense, one easily shows that b is a cluster point of I, which
yields b ∈ I ⊆ X \ A. But as [a, b] is connected by Lemma 3.1 (d)
 CONTINUITY OF INVERSES 55

and a strict superset of I, which is a connected component of X \ A,

the point b belongs to A. Contradiction!
Altogether we therefore either have I = [a, b] with a ≤ ξ ≤ b and
a, b ∈ X \ A or I = (a, b) with a < ξ < b and a, b ∈ A.
In the first case we can choose as before a net (xt )t∈A in A∩(−∞, a)
converging in (X, ≤) to a. We then obtain
[
(−∞, ξ) ∩ A = (−∞, a) ∩ A = ((−∞, xt ) ∩ A) ,
t∈A
so that (−∞, ξ) ∩ A is a union of sets open with respect to the order
topology on A and consequently itself open with respect to the order
topology on A.
In the second case we observe that [b, ∞)∩A is closed with respect
to the order topology on A (because of b ∈ A). For this reason
(−∞, ξ) ∩ A = A \ ([b, ∞) ∩ A)
is open with respect to the order topology on A. 
Example 3.4. (a) Lemma 3.3 applies to all closed subsets of a
connected linearly ordered space (X, ≤). Indeed, let A be a
closed subset of X. Then X \ A is open and can be expressed
as a union of disjoint open intervals (see, e.g., [2, 4A2R (j)]).
Hence, each connected component of X \ A is open.
(b) Furthermore, we may apply Lemma 3.3 to open subsets U
of a connected linearly ordered space (X, ≤). To see this,
note that each component of X \ U is closed in (X \ U, τX\U )
because components of a topological space are always closed
in this space. Hence, each component of X \ U is closed in
(X, ≤) since X \ U is closed in (X, ≤).
(c) Lemma 3.3 also applies to each subset D of a connected lin-
early ordered space (X, ≤) which is dense in X. In fact,
thanks to Lemma 3.1 every component of X \ D is an inter-
val. Since (X, ≤) is dense, each non-empty interval that is
not a singleton has nonvoid interior. Hence, every component
of X \ D is a singleton and consequently closed as (X, ≤) is
Hausdorff.
We record the following simple consequence of Lemma 3.3.
Corollary 3.5. Let ∅ 6= A ⊆ R. Then the order and subspace
topology of A coincide if and only if every bounded component of
R \ A is either closed or open.
56 H. HOFFMANN

Remark 3.6. If we combine Proposition 2.1, Proposition 2.4 and

Corollary 3.5, we arrive at the following result:
Let ∅ =
6 A ⊆ R. Then the order and subspace topology of A coincide
if and only if every (continuous) strictly monotonic function f :
A → R possesses a continuous inverse f −1 : f (A) → A, where A
and f (A) are endowed with their respective subspace topologies.
As previously promised we now arrive at the announced generali-
sation of Proposition 2.1.
Proposition 3.7. Let (X, ≤X ) and (Y, ≤Y ) be two linearly ordered
spaces, where (X, ≤X ) is connected. Let A be a nonvoid subset of
X such that each component of X \ A (w.r.t. the subspace topology
τA ) is closed or open. Assume that f : A → Y is an injective order-
preserving or injective order-reversing mapping. Then the inverse
mapping f −1 : (f (A), τf (A) ) → (A, τA ) is continuous.
Proof. If f is order-reversing, then we define another total order ≤rY
on Y via
y ≤rY y 0 :⇐⇒ y ≥Y y 0
for y, y 0 ∈ Y . Then f : (A, ≤X ) → (Y, ≤rY ) is order-preserving and
it is not hard to verify that τ(Y,≤Y ) = τ(Y,≤rY ) . For this reason we may
and will assume w.l.o.g. that f is order-preserving. Then
f : (A, τ(A,≤X ) ) → (f (A), τ(f (A),≤Y ) )
is a homeomorphism. Because of τ(f (A),≤Y ) ⊆ τf (A) and τ(A,≤X ) = τA
(the latter assertion results from Lemma 3.3 and the hypothesis),
we conclude that
f −1 : (f (A), τf (A) ) → (A, τA )
is continuous as claimed. 
We close this note with the following Question:
Does also an analogue of Proposition 2.4 hold in all connected lin-
early ordered spaces? Or to put it another way: Is an analogon of
the characterisation obtained in Remark 3.6 valid in every connected
linearly ordered space?

References
[1] D. H. Fremlin, Measure Theory – Volume 3, Torres Fremlin, Colchester, 2004.
[2] D. H. Fremlin, Measure Theory – Volume 4, Torres Fremlin, Colchester, 2006.
 CONTINUITY OF INVERSES 57

[3] R. G. Haydon, J. E. Jayne, I. Namioka, C. A. Rogers, Continuous functions

on totally ordered spaces that are compact in their order topologies, J. Funct.
Anal. 178 (2000), 23–63.
[4] H. Heuser, Lehrbuch der Analysis – Teil 1, 13th ed., B. G. Teubner, Stuttgart,
Leipzig, Wiesbaden, 2000.
[5] M. J. Hoffman, Continuity of inverse functions, Math. Mag. 48 (1975), 66–
73.
[6] L. A. Steen and J. A. Seebach Jr., Counterexamples in Topology, Holt, Rine-
hart and Winston, Inc., New York-Montreal, Que.-London, 1970.
[7] S. Willard, General Topology, Addison-Wesley Publishing Co., Reading,
Mass.-London-Don Mills, Ont., 1970.

Heiko Hoffmann received his Bachelor and Master degree from Saarland Uni-
versity in 2008 and 2010, respectively. He has recently completed his PhD thesis
at the Karlsruhe Institute of Technology (KIT).
Department of Mathematics, Institute for Analysis, Karlsruhe
Institute of Technology (KIT), 76128 Karlsruhe, Germany
E-mail address: [email protected]