A Pedagogical History of Compactness

Authors(s): Manya Raman-Sundström

Source: The American Mathematical Monthly, Vol. 122, No. 7 (August–September 2015),
pp. 619-635
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/10.4169/amer.math.monthly.122.7.619
Accessed: 30-03-2016 12:00 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact [email protected].

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access
to The American Mathematical Monthly

http://www.jstor.org

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
 A Pedagogical History of Compactness
Manya Raman-Sundström∗

Abstract. This paper traces the history of compactness from the original motivating questions
through the development of the definition to a characterization of compactness in terms of nets
and filters. The goal of the article is to clarify the central concepts of open-cover and sequential
compactness, including details that a standard textbook treatment tends to leave out.

Modern mathematics tends to obliterate history:

each new school rewrites the foundations of its
subject in its own language, which makes for
fine logic but poor pedagogy.
R. Hartshorne

1. WHY STUDY THE HISTORY OF COMPACTNESS? Compactness has come

to be one of the most useful notions in advanced mathematics. It can also be seen as
a kind of gate-keeper topic. If an undergraduate mathematics student does not under-
stand compactness it is unlikely that he or she will be able to do higher level mathe-
matics. However, for whatever reasons, when we teach compactness and other topics
of similar importance, we tend to provide little historical motivation. Students are left
more or less on their own to figure out, if they ever do, how definitions and theorems
relate to each other and why they take the specific forms they do.
This paper is an attempt to fill in some of the information that the standard textbook
treatment of compactness leaves out. It is not a historical article, per se, but a synthe-
sis of historical documents with an eye towards clarifying the main ideas related to
compactness. In particular, the paper discusses the origins and development of both
open-cover and sequential compactness, how and why open-cover compactness came
to be favored, and some modern developments involving nets and filters. We will move
fairly slowly and carefully through the early part of the history and a bit more quickly
through the more modern developments.
A list of terms related to compactness is given in the Appendix. Since the terms
have changed names at various points in history, the list can be useful for keeping the
concepts straight. In the main text we will use a combination of historical and modern
formulations of the main definitions, lemmas, and theorems, favoring the best ones for
readability with originals in footnotes.

2. POSSIBLE MOTIVATIONS FOR COMPACTNESS. Compactness grew out of

one of the most productive periods of mathematical activity. In mid to late nineteenth
century Europe, advanced mathematics began to take the form we know today. In the
background was Cantor’s work establishing the beginning of a systematic study of set
theory and point-set topology.1 Also, many mathematicians—including Weierstrass,
Hausdorff, and Dedekind—were worried about the foundations of mathematics and
http://dx.doi.org/10.4169/amer.math.monthly.122.7.619
MSC: Primary 01-01
∗ This paper is based upon my masters thesis [38] at UC Berkeley.
1 Though Cantor himself turned his interests to transfinite sets, the significance of Cantor’s work for topol-

ogy should be credited to Poincaré. Thanks to Jeremy Gray for this comment.

August–September 2015] A PEDAGOGICAL HISTORY 619

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
began to make rigorous many of the ideas that had for centuries been taken for granted.
While some of the nineteenth century work can be traced to mathematical concerns of
the early Greeks, the level of rigor and abstraction reflects a revolution in mathematical
thought.
It is in this context that we will discuss some specific problems that appear to have
motivated the concept of compactness. In particular, we will discuss the influence of
the study of properties of closed, bounded intervals of real numbers (which will be
denoted [a, b]), spaces of continuous functions, and solutions to differential equations.

Properties of [a, b]. In the mid to late nineteenth century, mathematicians began to
really understand and specify essential properties of the real line. This work led to two
different characterizations of the notion which came to be known as compactness. One
characterization, developed by Bolzano and Weierstrass among others, grew out of the
study of functions defined on sequences of real numbers. The other characterization,
which grew out of work by Heine, Borel, and Lebesgue, was based on topological
features, such as the covering of sets by open neighborhoods. We will examine both of
these characterizations in more detail.
The origin of sequential compactness is often traced to a theorem, proved rigorously
by Weierstrass in 1877, which concerns the behavior of continuous functions defined
on closed, bounded intervals of the real line.2 The following statement of the theorem
comes from Fréchet, who referred to this theorem as a result of Weierstrass.

Theorem 2.1 (Weierstrass). Each function continuous in a limited [equivalent to

modern-day “closed and bounded”] interval attains there at least once its maximum.3

a b
Figure 1. A continuous function on [a, b]

Fréchet, who defined sequential compactness in his 1906 thesis, said that his defi-
nition came from his desire to generalize this theorem to abstract spaces [43, p. 244].
Weierstrass’s theorem owes its essential ideas to Bolzano, who in 1817 working in
relative isolation both politically and mathematically in Bohemia, remarkably stated
and proved the following:

Lemma 2.2 (Bolzano). If a property M does not apply to all values of a variable
quantity x, but to all those that are smaller than a certain u, there is always a quantity
2 This date refers to one of the earliest publications of the theorem, see [46]. However, it is likely that

Weierstrass actually proved it and disseminated it orally, via lectures, perhaps ten years earlier.
3 From the original French: Weierstrass a en effete démontré que toute fonction continué dans un intervalle

limité y atteint au moins une fois son maximum [17, p. 848].

620 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 122

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
U which is the greatest for those of which it can be asserted that all smaller x possess
the property M.4

This lemma, today called the least upper bound property for real numbers, was
somewhat of a breakthrough in the conceptualization of real numbers. The proof of
this lemma provided the first real account of the limiting process, and was used to
prove what we now call the intermediate value theorem: if f is continuous on [a, b]
with f (a) < 0 and f (b) > 0, then for some x between a and b, f (x) will be exactly 0.
The idea behind Bolzano’s proof of the lemma was to use interval bisection, that
is to narrow in on the least upper bound by throwing away points of the set that were
below it. This iterative process was essentially the same process used in Weierstrass’s
proof of the maximum value theorem (see [27, p. 953]). In particular, Bolzano’s lemma
allowed Weierstrass to prove that every bounded infinite set of real numbers has a limit
point. It is this property that Fréchet used when he generalized Weierstrass’s theorem
to abstract spaces. We now know this property as the Bolzano–Weierstrass property,
or limit-point compactness.
While Bolzano and Weierstrass were trying to characterize properties of the real
line in terms of sequences, other mathematicians, such as Borel and Lebesgue, were
trying to characterize it in terms of open covers. Borel proved the following lemma in
his 1894 thesis.

Lemma 2.3 (Borel). If on a line one has an infinite number of subintervals, such that
every point of the line is interior to at least one of the intervals, then one can determine
effectively a bounded number of intervals from among the given intervals that have the
same property (every point of the line is interior to at least one of them).5

Here a line means a bounded interval. It turns out that Borel’s approach was similar
to the approach Heine used in 1872 to prove that a continuous function on a closed in-
terval was uniformly continuous [23, p. 188]. This theorem was first proven by Dirich-
let in his lectures of 1852, with a more explicit use of coverings and subcoverings than
in Heine’s theorem [15, p. 91]. However, Dirichlet’s notes were not published until
1904, which might explain why he does not get credit for the generalized version of
the Borel lemma (now referred to as Borel theorem). The reason that Heine’s name is
attached to the theorem is that Schönflies, a student of Weierstrass, noticed the connec-
tion between Heine’s work and Borel’s [40, p. 51]. The generalized theorem, which is
now commonly called the Heine–Borel theorem6 , with modern language and notation,
is as follows:

Theorem 2.4 (Heine-Borel). A subset of R is compact if and only if it is closed and

bounded.

While Heine is credited with a theorem he did not prove, it appears that Cousin was
largely overlooked for a lemma he did prove. In 1895 he generalized the Borel lemma
4 From the original German: Wenn eine Eigenschaft M nicht allen Werthen einer veränderlichen Grösse x,

wohl aber allen, die kleiner sind, als ein gewisser u, zukömmt: so gibt es allemahl eine Grösse U , welche die
grösste derjenigen ist, von denen behauptet werden kann, dass alle kleineren x die Eigenschaft M besitzen [7,
p. 41].
5 From the original French: Si l’on a sur une droite une infinitè d’intervalles partiels, tels que tout point

de la droite soit intérieur à l’un au moins des intervalles, on peut déterminer effectivement un nombre limité
d’intervalles choisis parmi les intervalles donnés et ayant la même propriété (tout point de la droite est intérieur
à au moins l’un d’eux) [8, p. 51].
6 For an accessible history of this theorem, along with discussion of a number of different original statements

of this theorem, see [2].

August–September 2015] A PEDAGOGICAL HISTORY 621

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
to arbitrary covers. In the formulation of the lemma below, the plane YOX is just R2 ,
and the region S would, in today’s language, be described as closed and bounded.

Lemma 2.5 (Cousin). In the plane YOX, let S be a connected area bounded by a
closed contour, simple or complex. Suppose that at each point of S or its perimeter
there is a circle, of nonzero radius, having this point as its center. It is then always
possible to subdivide S into regions, finite in number and sufficiently small for each
one of them to be entirely inside a circle corresponding to a suitably chosen point in S
or on its perimeter.7

In other words, if for every point of a closed, bounded region, there corresponds a
finite neighborhood, then the region can be divided into a finite number of subregions
such that each subregion is contained in a circle having its center in the subregion.8
Cousin’s lemma (sometimes referred to as Cousin’s theorem) is generally attributed to
Lebesgue, who was said to be aware of the result in 1898 and published his proof in
1904 [29].9 The Lebesgue lemma is considered itself to be an important consequence
of compactness.
While there is some debate over who was really responsible for the ideas and proofs,
the idea that any closed, bounded subset of R has the open-cover property (sometimes
called the Borel–Lebesgue property) was known when Fréchet first defined compact-
ness formally.

Spaces of Continuous Functions. A second motivation for the notion of compactness

was the study of abstract spaces such as spaces of continuous functions, C 0 [a, b].10
In C 0 [a, b], points are functions (whereas in [a, b] points are real numbers).11 The
properties of [a, b] alone might not have been seen as important to generalize if it
weren’t the case that these properties seemed to be important in more abstract spaces
as well. However, it turns out that infinite dimensional spaces like C 0 [a, b] are not
as well behaved as finite dimensional spaces like Rn . For instance, closed, bounded
subsets of continuous functions on R do not necessarily have the Bolzano–Weierstrass
or open-cover property. The work in this area was done by Ascoli and Arzelà in the
last decades of the 1800s.
The following example illustrates that a closed, bounded subset of continuous func-
tions on R is not, in our modern language, sequentially compact. Consider B, the set of
continuous functions f defined on [0, 1] with  f  ≤ 1. (This is the closed unit ball in
C 0 [0, 1] and   is the sup norm.) We will show that there is a sequence in B that does
not have a convergent subsequence. Let f n (x) = x n . This sequence lies in B, but we

7 From the original French: Soit, sur le plan YOX, une aire connexe S limitée par un contour fermé simple ou

complexe; on suppose qu’à chaque point de S ou de son périmètre correspond un cercle, de rayon non nul, ayant
ce point pour centre: il est alors toujours possible de subdivider S en régions, en nombre fini et assez petites
pour que chacune d’elles soit complètement intérieure au cercle correspondant à un point convenablement
choisi dans S ou sur son périmètre [12, p. 22].
8 Note that the original definition was formulated without the term “neighborhood,” or “voisinage” in

French. The idea of neighborhood was obviously around during Cousin’s time, but was not used consistently.
Formal definitions of the term can be found in [18] and [22].
9 Cited in [25, p. 29].
10 For more details see [16] and [16].
11 We could just as well take C 0 on any set .

622 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 122

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
cannot find a subsequence that converges uniformly to a function in C 0 [0, 1]. Suppose
to the contrary f is such a function. Then

f (x) = lim f nk (x),

k→∞

which would imply that


0 if x < 1
f (x) =
1 if x = 1.

Since f is a discontinuous function, it is not in C 0 [0, 1]. Hence the sequence f n (x)
has no uniformly convergent subsequence.
The problem in this example comes from how the functions converge. If conver-
gence means pointwise convergence, then we get a behavior different from that of, say,
sequences in closed unit balls of Rn . In order to avoid this problem, Ascoli introduced
the notion of equicontinuity [4, p. 566].12 A set E is equicontinuous if and only if for all
 > 0 there exists a δ > 0 such that |s − t| < δ and f ∈ E imply | f (s) − f (t)| < .
The Arzelà–Ascoli theorem, in modern language, then states the following.

Theorem 2.6 (Arzelà-Ascoli). Any bounded equicontinuous sequence of functions in

C 0 [a, b] has a uniformly convergent subsequence.13

Using modern terminology we can state a consequence of this theorem, analogous

to the Heine–Borel theorem.

Theorem 2.7. A subset of C 0 [a, b] is compact if and only if it is closed, bounded, and
equicontinuous.

Ascoli proved the sufficiency of this condition in 1884 [4, p. 567] and Arzelà the
necessity in 1889 [3, p. 345] (with a clearer proof presented in 1894 [5, p. 226]). This
generalization of Bolzano–Weierstrass’s theorem (although not stated in terms of com-
pactness) was apparently well known after 1880. Moreover, Hilbert seems to have dis-
covered this compactness property independently and published it in 1900 [13, p. 82].
It is unclear whether Arzelà and Ascoli themselves were aware of how their work was
connected with compactness, but Fréchet’s work was influenced by theirs [43, p. 255].

Solutions to Differential Equations. A third motivation for the notion of com-

pactness came from the desire to find solutions to differential equations. Peano, a
contemporary of Arzelà and Ascoli as well as a fellow Italian, realized that the
Arzelà–Ascoli theorem might be useful for demonstrating the existence of such
solutions. He searched for solutions by making a sequence of approximations. He

12 See also [9, p. 27].

13 From the original Italian: La condizione necessaria e sufficiente affinché una successione data di funzioni
u 1 (x), u 2 (x), . . . , u n (x), . . . abbia una funzione limite, nel senso detto sopra, é che, preso un numero positivo
σ piccolo a piacere, si possa sempre determinare un numero intero corrispondente m σ tale che per ogni x,
nell’intervallo a · · · b, si abbia qualunque sia p intero positivo, | u m σ (x) − u m σ + p (x) |< σ . Translated to
English: The necessary and sufficient conditions that a given sequence of functions
u 1 (x), u 2 (x), . . . , u n (x), . . . defined on an interval a · · · b may converge to a limiting function is that,
given an arbitrarily small positive number σ there can always be determined a corresponding integer m σ such
that for all values of x in the interval a · · · b, and for all positive integers p, | u m σ (x) − u m σ + p (x) |< σ [5, p.
226].

August–September 2015] A PEDAGOGICAL HISTORY 623

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
then used what we now call compactness to show that there will be a subsequence that
converges uniformly to a limit (which will be the solution to the differential equation).
To this end, Peano proved the following theorem in 1890.

Theorem 2.8 (Peano). Suppose we are given a system of differential equations in

normal form:

d x1
= ϕ1 (t, x1 , . . . , xn ),
dt
......
d xn
= ϕn (t, x1 , . . . , xn ),
dt

where the functions ϕ1 , . . . , ϕn are continuous in a neighborhood of (b, a1 , . . . , an ).

Then there exists an interval (b, b ), and n functions of t from this interval, x1 , . . . , xn ,
satisfying our system of equations and evaluating to a1 , . . . , an , respectively, at
t = b.14

While it is not clear if Fréchet was aware of this application, applications for the
notion of compactness were apparently known before the term was formally defined.

3. DEVELOPING THE DEFINITION. We will trace below the development of the

two central notions of compactness discussed above, those stemming from sequences
and open covers of real numbers. Again, it is useful to know something about the
climate of the mathematics community at the time of these historical developments.
We will focus on the contributions of only a few central people, but there was actually
a large community of mathematicians who were developing ideas that are now the
foundations for analysis and topology. Many of these mathematicians were in very
close contact with each other, so it is difficult to separate their contributions. Among
them, in France, were Hadamard, Lebesgue, and Fréchet; in Russia, Alexandroff15 and
Urysohn16 ; in Germany, Hausdorff, Hilbert, Schönflies, and Cantor; in Hungary, F.
Riesz; in the Netherlands, Brouwer; in Austria, Vietoris; and in the U.S., Chittenden,
Hedrick, and Moore.
We will start with the work of Fréchet, who coined the term “compact” and gave
definitions for what we now know as countable and sequential compactness. We will
then briefly discuss contributions by Alexandroff and Urysohn, who developed and
stated what we now call open-cover compactness, or simply compactness. We will
show why open-cover and sequential compactness are not equivalent in abstract topo-
logical spaces, providing motivation for a formulation of compactness in terms of nets
and filters which is analogous to sequential compactness.
14 From the original French: Soit donné le système d’équations différentielles, ramené à la forme

normale: ddtx1 = ϕ1 (t, x1 , . . . , , xn ), . . . ddtxn = ϕn (t, x1 , . . . , xn ), où les ϕ1 , . . . , ϕn sont des fonctions contin-
ues aux environs de t = b, x1 = a1 , . . . , xn = an . [Alors il existe] un intervalle (b, b ), et, dans cet inter-
valle, n fonctions x1 · · · xn de t, qui satisfont aux équations données, et qui, pour t = b, prennent les values
a1 · · · an [35, p.182].
15 I will use the spelling “Alexandroff,” rather than the sometimes used “Alexandrov,” since that was the

spelling he preferred [11, p. 634].

16 Urysohn died, tragically, at the age of 26 in a swimming accident off the coast of France. Much of his work

was published posthumously by Alexandroff, who kept up his correspondence with Fréchet after Urysohn’s
untimely death [44, p. 340].

624 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 122

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
Fréchet: Countable and Limit-Point Compactness. While Fréchet was influenced
by many contemporaries and predecessors, it seems he deserves credit as the father of
compactness. It was Fréchet who gave the concept a name, in a paper [17] leading to
his 1906 doctoral thesis. Fréchet also defined metric spaces for the first time, though
not using that term,17 and made inroads into functional analysis, thus providing a con-
text for which the importance of compactness became clear.
Fréchet was a mathematician of big ideas. He preferred definitions that had an in-
tuitive feel rather than analytic power. This preference can be seen in [17, p. 849] in
which he defined a notion of compactness, introducing first a definition of what we
now call countable compactness, using nested intersections, before introducing a char-
acterization using limit points.
In Fréchet’s thesis, he considered three kinds of spaces, which he called L-class,
V-class, and E-class. L-classes were the most general, in which a notion of sequential
compactness was defined. E-classes, which we now call metric spaces, and V-classes,18
a metric space with a weak version of the triangle inequality, were less general, but eas-
ier to work with. The goal was to define compactness for L-classes, but this turned out
unsuccessful because sequential compactness did not have all the properties needed
to generalize to abstract topological spaces (more on this in section 3.2). Fréchet fo-
cused instead on the V- and E-classes, in which notions of modern-day compactness
and sequential or limit-point compactness were equivalent. The following definition
was given for E-classes.

Definition 3.1. A set E is called compact if, whenever E n is a sequence of nonempty,

closed subsets of E such that E n+1 is a subset of E n for each n, then there is at least
one element that belongs to all of the E n ’s.19

The exact nature of Fréchet’s intuition for this definition is unclear, but there might
be two features of compact sets that he wanted to capture. The first is a sense of bound-
edness. The nested intersection property allows us to easily rule out sets that have tails
running to ∞. For instance, we can show that R is not compact. Let E n = [n, ∞).
Each E n is closed because it contains the point n, and clearly E n+1 ⊂ E n . However,
the infinite intersection of these intervals is empty, so R is not compact.

n n+1 n+2 n+3 . .

Figure 2. Nested tails

A second feature the nested intersection definition allows us to quickly see is that
sets that have “holes” are not compact. For instance, we can see that X = [a, b] is
compact and Y = [a, b) ∪ (b, c] is not.20 In the latter case, consider E i = [ai , b) ∪
(b, ci ], where ai+1 > ai and ai → b, ci+1 < ci , and ci → b. These sets are clearly
nested and are closed in Y , but the infinite intersection of those intervals is empty.
Hence Y is not compact.
17 The term “metric space” is credited to Hausdorff who used the term in [22, p. 290-1].
18 The letter V comes from the French voisinage meaning neighborhood, L from limite, or limit, and E from
écart, or “gap,” referring to the nonzero distance between two points in a metric space, which according to
[20] derived from Jordan’s use of the term “distance.”
19 From the original French: Si on considère une suite d’ensembles E , E , . . . , E , . . . formés d’éléments
1 2 n
d’un même ensemble compact E, chacun fermé, contenu dans le précédent et possédent au moins un élément,
il y a nécessairement un élément commun à tous ces ensembles [18, p. 7].
20 One could construct a similar example with a hole that is compact such as Z = [a, b] ∪ [c, d], so the

example above only illustrates the intuition.

August–September 2015] A PEDAGOGICAL HISTORY 625

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
 a b a b c

Figure 3. Nested intersections

While Fréchet preferred his intuitive definition involving nested intersections [36,
p. 430], he realized the importance of also providing a more useful, if less intu-
itive, definition. Below is another definition from Fréchet, which uses the Bolzano–
Weierstrass property. This definition applies to V- and E-classes where limit-point,
countable, and sequential compactness are equivalent. Note that for Fréchet, a com-
pact set need not be closed. So for his subsequent definitions and theorems, he often
needed to require that a set be both compact and closed.

Definition 3.2. We will say that a set is [relatively limit-point] compact if it contains
only a finite number of points or if every one of its infinite subsets gives rise to at least
one limit point.21

Since Fréchet did not require that a compact set be closed, he defined the notion of
an extremal set, which is closer to our modern day notion of compact.

Definition 3.3. We will call a set that is both closed and compact an extremal set.22

Fréchet also includes a note which reflects the intuition of this definition, “Within
abstract set theory, extremal sets play a role akin to that of intervals in the theory of
subsets of the real line.”23 We might not know exactly why Fréchet chose the word
“compact,” but his choice of the term was not popular with all mathematicians. For
instance, Schönflies suggested that what Fréchet called compact be called something
like “lückenlos” (without gaps, closer to the modern notion of completeness) or “ab-
schliessbar” (closable), [43, p. 266], suggesting that the intuition behind the term was,
at this point, not fully shared. Despite all of Fréchet’s early concern with intuitive def-
initions and choice of terminology, at the end of his life, he could not remember why
he chose the term:

Doubtless I wanted to avoid a solid dense core with a single thread going off to infinity being
called compact. This is a hypothesis because I have completely forgotten the reasons for my
choice!24

So even in the lifetime of the mathematician who named the concept, the orig-
inal intuition behind the concept was somewhat lost, and Fréchet’s intuitive nested
intersection definition was supplanted by less intuitive but more powerful notions of
limit-point, sequential, and open-cover compactness.

Hausdorff: Compactness on Metric Spaces. One of the obstacles to defining com-

pactness, as we know it today, was to define it in a way that would work for abstract
21 From the original French: Nous dirons qu’un ensemble est compact lorsqu’il ne comprend qu’un nombre

fini d’éléments ou lorsque toute infinité de ses éléments donne lieu à au moins un élément limite [18, p. 6].
22 From the original French: Lorsqu’un ensemble est à la fois compact et fermé nous l’appellerons ensemble

extrémal [18, p. 6–7].

23 From the original French: Le rôle de l’ensemble extrémal dans la théorie des ensembles linéaires.

[18, p. 7].
24 From the original French: ... j’ai voulu sans doute éviter qu’on puisse appeler compact un noyau solide

dense qui n’est agrémenté que d’un fil allant jusqu’à l’infini. C’est une supposition car j’ai complètement
oublié les raisons de mon choix! [36, p. 440].

626 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 122

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
topological spaces. This was a problem for Fréchet, and in the end he had to restrict
his definition to V- and E-classes, leaving open the question of defining compactness
for L-classes, the predecessor of what we now call abstract topological spaces. In the
early 1900s, Hausdorff’s work revolutionized the area of topology, providing defini-
tions that are now standard in the field.25 For instance, in 1914 he introduced what
we now call Hausdorff spaces, in which distinct points have disjoint neighborhoods.
In [22] he defined a set E to be compact if every infinite subset of E has a limit point
in E, where limit point in this context means that every neighborhood of the point
contains infinitely many elements of E. Hausdorff’s notion of compactness, which
we would call limit-point compactness and is equivalent to countable compactness for
Hausdorff spaces, remained the standard notion of compactness throughout the rapid
development of point-set topology in the 1920s.26

Alexandroff and Urysohn: Open-Cover Compactness. While Fréchet was the

first to formally define compactness, his contemporaries in Russia, Alexandroff and
Urysohn, appear to be the first to state it in its most general form, in the context of
abstract topological spaces. It is perhaps for this reason that the two Russians are often
credited with defining the notion (e.g., [30, p. 425S]).
In a paper in 1923, Alexandroff and Urysohn listed open-cover compactness, the
property that every open cover has a finite subcover, as one of three equivalent proper-
ties a set could have to be called compact (in their language “bicompact”). The other
two properties were that all infinite sets have a complete accumulation point and that
nested intersections are nonempty.27 Alexandroff and Urysohn note that these three
properties were already known, although the concept had not been named.28 Alexan-
droff claimed the accumulation point characterization was most important initially, due
to the dominance of the Bolzano–Weierstrass property, but after some years it became
clear that the open-cover property was more fruitful [11, p. 633]. Today it is com-
mon to give the open-cover property as the definition and show the equivalence of one
or both of the other two properties as theorems.29 While more abstract, and perhaps
less intuitive than the other characterizations, the open-cover property brings out more
clearly than the other ones the analogy between compactness and finiteness.30
Alexandroff and Urysohn were actually in close contact with Fréchet [44, pp. 319-
357] during the time they developed their work on compact topological spaces. Al-
though Alexandroff and Urysohn usually get credit for defining open-cover compact-
ness, Fréchet was not unaware of the possibility of using neighborhoods to characterize

25 See [37] for more details and a biographical account of the life of this intellectual giant, including a parallel

career under the pseudonym Paul Mongré, and his eventual suicide in the internment camp Bonn–Endenich in
1942.
26 Thanks to an anonymous reviewer for this information and formulation. See [28] for more detail about

Hausdorff’s role in the development of topology and related fields.

27 From the original French: En appelant point d’accumulation complète d’un ensemble E situé dans l’espace

topologique R tout point ξ tel que la puissance de la partie de E contenue dans un voisinage quelconque du
point ξ est égale à celle de tout l’ensemble E—les trois propriétés suivantes sont équivalentes: (A) Tout en-
semble infini situé dans R possède un au moins point d’accumulation complète, (B) Toute suite bien ordonnée
descendante d’ensembles fermés situés dans R possède au moins un point commun à tous les ensembles de la
suite, (C) De toute infinité de domaines recouvrant l’espace R on peut extraire un nombre fini jouissant de la
même propriété [1, p. 6].
28 Sans prétendre de donner une notion nouvelle, nous allons appeler bicompacts les espaces R vérifiant une

quelconque (et, par suite, toute les trois) des propriétés énoncées [1, p. 6].
29 An early example of this is [26, p. 135–137].
30 Reminiscent of the joke, attributed to Herman Weyl [24, p. 499], ”What is a compact city? It is one that

can be guarded by finitely many near-sighted policemen.”

August–September 2015] A PEDAGOGICAL HISTORY 627

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
compactness, an idea suggested by his advisor Hadamard in 1905.31 The first definition
that Fréchet gave, in terms of nested intersections, is the dual of, and hence logically
equivalent to, countable open-cover compactness.

Open-Cover vs Limit-Point Compactness. Although Fréchet may have been moti-

vated originally to define compactness for abstract topological spaces, he in fact re-
stricted himself to metric spaces. His approach of looking at sequences and limits was
not as general as the approach of using open covers, which resulted in what we now
take to be the correct definition of compactness.32 Here we look at examples which
illustrate why sequential compactness and open-cover compactness are not equivalent.
We will use the concept of the least upper bound property, namely that any nonempty
set containing an upper bound necessarily has a least upper bound.

Sequentially Compact Does Not Imply Compact. Consider S = {α | α is an ordinal

number and α < } with the order topology, where  is the first uncountable ordinal
number.33 See Figure 4 below. The first infinite ordinal, ω, is the first ordinal after
“exhausting” the natural numbers. The first uncountable ordinal, , is the ordinal after
“exhausting” the countable ordinals.
We know that all closed subsets of compact sets are compact (and all compact sets

1 2 ω ω + 1 2ω ω2 ωω Ω
SΩ

Figure 4. Representation of S

are closed). So S is not compact since it is not closed in the compact set S ∪ {}.
However, it turns out that S is limit-point compact. To see why this is true, we will
use the fact that any countable subset of S has an upper bound in S . If we take any
infinite subset of S , it has a countably infinite subset, which we will call X . Since
X is countable, it has an upper bound, let’s call it b, in S . But the interval [1, b] is
compact since S has the l.u.b. property. So there must be a point in [1, b] which is
a limit point (of both X and any set containing it). Thus, S is limit-point compact.
Essentially the same argument shows that any sequence in S must have a convergent
subsequence in S , so S is sequentially compact.

Compact Does Not Imply Sequentially Compact. Just as we can have a space that is
compact but not sequentially compact, we can also have a space that is sequentially
compact but not compact.34 Consider the set of all functions from the interval [0, 1]
to itself with the topology of pointwise convergence. This can be thought of as the
infinite product [0, 1][0,1] with the product topology, which is compact by Tychonoff’s
theorem.35 However, if f n (x) is the nth digit in the base-2 decimal expansion of x
31 See [28, p. 212]. The undated letter from Hadamard to Fréchet and can be found in full in [43, p. 246].
32 For more detail on how the open-cover definition became standard, see [11, p. 633–635]. An interesting
anecdote attributed to Melvin Henrikson is included in this paper concerning the emotions behind the estab-
lishment of the now standard topological definitions. An attempt was made in 1925 in Ann Arbor to nail down
the definitions, but after an intense hour long debate, the meeting broke up—mostly over disagreement on how
to define compactness!
33 Refer to [42, p. 68] for further discussion of this example.
34 For more detail see [42, p. 125–126].
35 The product of compact topological spaces is compact, or in German, Das Produkt von bikompakten

Räumen ist wieder bikompakt, originally proved in [45, p. 772], though the theorem is sometimes credited to
Čech [16].

628 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 122

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
(using the expansion that terminates in 0’s if x is a dyadic rational), the sequence f n ,
which is a sequence in the set of all functions from [0, 1] to itself, has no pointwise
convergent subsequence. It does have convergent subnets,36 a concept that will be de-
fined in the next section, but not proper convergent sequences.

1 Ω
b
Figure 5. Illustration of b in S .

4. NETS AND FILTERS. In the previous section we saw that the two important
properties of compactness, those stemming from the Bolzano–Weierstrass property
(sequential compactness) and the Borel–Lebesgue property (open-cover compactness),
are not equivalent in abstract topological spaces. Open-cover compactness is more
general and applicable and for these reasons is now the concept to which the term
“compactness” refers. However, it is possible to define open-cover compactness in a
way that is analogous to sequential compactness, using notions of nets and filters.37
These two concepts are very different on the surface, but they give rise to the same
notion of convergence in abstract topological spaces.

Moore and Smith: Nets. The theory of nets was developed by E. H. Moore and his
student H. L. Smith, and published in 1922 [32].38 It is unclear whether Moore and
Smith knew how nets could be used to define compactness. This connection is usually
credited to Birkhoff [26, p. 64], who applied Moore–Smith theory to general topolog-
ical spaces. However, in the paper in which Moore and Smith introduce the concept
of nets, they also generalize some of Fréchet’s compactness results [32, p. 118]. Our
goal here is to express compactness in terms of nets, and we will use the S example
to motivate and illustrate net compactness.
The problem in the S example is that while  is a limit point of S (any neigh-
borhood of  contains points of S ), no sequence in S converges to  [26, p. 76].
If we are limited to taking a countable number of elements in the sequence, we will
never reach . Nets provide one way of getting around this problem by allowing us
to have something like uncountable sequences. In our discussion of both nets and fil-
ters, we will consider only topological spaces, on which the notion of neighborhood is
defined.39
To see how nets are a generalization of sequences, it is useful to think of sequences
as functions on the natural numbers.

Definition 4.1. A sequence (denoted {xn }n∈N = {x1 , x2 , x3 , . . .}) is a function which
assigns to each element n of the natural numbers, N, a functional value xn in a set X .

36 Every function from [0, 1] to {0, 1} is the limit of some subnet.

37 In this section we are more interested in the concepts and not the historical development, so we will be less

careful than in earlier sections about giving original definitions and theorems, though we provide references
for anyone who wants to track down the original formulations.
38 Little biographical information about Smith is available. He received his Ph.D. from University of Chicago

under Moore and got a job at Louisiana State University, but apparently after his important work on nets and
filters, he dropped into obscurity [19, p. 563].
39 This treatment follows [26, pp. 62–70] and [31, pp. 281–283, 286–289].

August–September 2015] A PEDAGOGICAL HISTORY 629

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
 We would like to replace N with a set that can be uncountable but has an ordering
similar to that of N. In other words, we want to stipulate conditions for an ordering
relation on a generic set that generalizes the way > orders natural numbers. We will
call this relation  to suggest the connection to >, and we will say that this relation
“directs” a given set.

Definition 4.2. A nonempty set D, with the relation  is called directed if and only if
(i) if d1 , d2 , d3 ∈ D such that d1  d2 and d2  d3 , then d1  d3 ;
(ii) if d1 , d2 ∈ D, then there is a d3 ∈ D such that d3  d1 and d3  d2 .

So the definition of a net is simply the definition of a sequence with N replaced

by the notion of a directed set. From now on, D will stand for a directed set with the
relation  as defined above.

Definition 4.3. A net (denoted {xd }d∈D or simply {xd }) is a function which assigns to
each element d of a directed set D a functional value xd in a set X .

Once we know what a net is, we can state what it means for it to converge. Again we
can derive the definition for net convergence and limit point by taking the definitions
involving sequences and simply replacing N and > with D and .

Definition 4.4. A net {xd } converges to a ∈ X (denoted {xd } → a) if and only if for
every neighborhood U of a, there is an index d0 ∈ D such that if d  d0 , then xd ∈ U
(i.e.. if the net is eventually in each neighborhood of a).

Definition 4.5. A point a is a limit point of {xd } if for every neighborhood U of a and
every d0 ∈ D, then there is a d  d0 such that xd ∈ U.

In order to state compactness in terms of nets, we also need the concept of subnet,
the analog of subsequence.40 Part of the definition of subsequence generalizes eas-
ily, but the other part requires us to think about subsequences in a slightly different
way than we are accustomed. The first defining property of subsequence is that each
element of the subsequence can be identified with an element of the sequence. This
property is generalized in (i) below. The second defining property requires that the
subsequence is ordered in a similar way as the sequence. Usually we require the in-
dices of the subsequence, like the indices of the sequence, to be strictly increasing. In
other words, for a subsequence {xnk } of a sequence {xn }, the n k are positive integers
such that n 1 < n 2 < n 3 · · · . But the feature of this condition which turns out to be
important is simply the fact that as k → ∞, so do the n k . This property is generalized
in (ii) below.

Definition 4.6. A subnet of a net {xd }d∈D is a net {yb }b∈B where B is a directed set and
there is a function ϕ : B → D such that:
(i) yb = xϕ(b) and
(ii) ∀d ∈ D, ∃b0 ∈ B such that if b  b0 , then ϕ(b)  d.

We are now ready to characterize compactness in terms of nets.

40 Incidentally, Kelley, who first coined the term “net” had considered using the term “way” so the analog

of subsequence would be “subway.” McShane also proposed the term “stream” for net since he thought it was
intuitive to think of the relation of the directed set as “being downstream from” [31, p. 282].

630 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 122

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
Theorem 4.7. A topological space X is compact if and only if either,
(i) every net of points of X has a limit point in X , or
(ii) every net of points of X has a convergent subnet in X .

Notice that these definitions are precisely the same as limit point and sequential
compactness for metric spaces with the term “net” substituted for “sequence.”
Applying these definitions to the S example, we can show why S is not compact.
If we take a net {xd } of elements of S , it is no longer the case that there will necessarily
be a limit in S . In particular, let D = S and xd = d. Then {xd } converges to , which
is not in S . Thus no subnet of {xd } will converge to a point in S .

Cartan (and Smith): Filters. Nets are not the only way of generalizing sequences.
Another generalization of sequence is a filter, defined by H. Cartan in 1937.41 While
different from a net, both nets and filters give rise to the same notion of convergence on
topological spaces. That is to say, on abstract topological spaces, they are essentially
the same.42 Nonetheless, some mathematicians find nets more intuitively appealing
and useful, while others prefer filters.
The idea behind filters was foreshadowed by F. Riesz in 1907 when he provided ax-
ioms for topology based on limit points instead of a metric [39]. We note that though
his topological axioms are not equivalent to the standard ones we use today, and his
work did not result in a fruitful line of research. Riesz defines a concept called an
“ideal” which is essentially the same as what we now call an ultrafilter. Smith inde-
pendently discovered filters as an attempt to explain what was lacking in the theory of
nets that he and Moore proposed.
Following our treatment of nets, we will now define the notions we need to state
compactness in terms of filters and then apply our compactness result to show S is
not compact. As with nets, we can look at convergence of sequences to motivate the
idea of convergence of filters. However, whereas with nets the focus was on the index
set, with filters the focus is on neighborhoods.43

Definition 4.8. Let X be a set. A set

of subsets of X is called a filter if and only if
(i) ∅ ∈
/
,
(ii) A1 ⊂ A2 ⊂ X and A1 ∈
⇒ A2 ∈
, and
(iii) A1 , A2 ∈
⇒ A1 ∩ A2 ∈
.

As with nets, we should define what it means for a filter to converge.

Definition 4.9. A filter

converges to a ∈ A (denoted
→ a) if and only if each
neighborhood of a is a member of
.

There is a natural way to associate a filter with any sequence. If x1 , x2 , x3 , . . . is a

sequence in X , we can associate with this sequence a filter
on X such that ∀a ∈
X, {xn } → a if and only if
→ a. In particular, let
= {A ⊂ X | ∃k A such that
∀i ≥ k A , xi ∈ A}. So the tails of the sequence are contained in neighborhoods which
41 See also [10, p. 8].
42 To show equivalence on abstract topological spaces, there is for example an exercise in Kelley that estab-
lishes a dictionary mapping between them (i.e., given a net you can find a filter, and vice versa) [26, p. 83].
But there is a subtle distinction for a particular type of limit found in the advanced theory of integration [41, p.
371].
43 This treatment follows [14].

August–September 2015] A PEDAGOGICAL HISTORY 631

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
are members of the filter. The condition that each neighborhood of a is in the filter is
then equivalent to the condition that the sequence is eventually in any neighborhood
of a.

a
A
x5

X
x2
x1
x0

Example with kA = 5

Figure 6. Member of filter with k A = 5

In order to define compactness in terms of filters, we need one more notion, that of
an ultrafilter.

Definition 4.10. A filter in X is an ultrafilter if and only if no filter in X properly

contains it.

The notion of ultrafilter is not exactly analogous to subsequence, but in the formu-
lation of compactness, it serves the same purpose.

Theorem 4.11. A topological space is compact if and only if every ultrafilter on X

converges to a point in X .

Now we can return again to our example and get a sense in terms of filters for why
S is not compact.44 We want to show that there is an ultrafilter on S that does not
converge. Consider all the neighborhoods of  in S ∪ . Let
= {A ⊂ S | ∃α ∈ S
such that ∀β ≥ α, β ∈ A}.

α Ω

Figure 7. Member of filter on S

This clearly satisfies the definition of a filter. Let  be any ultrafilter containing

. We claim  does not converge in S . Suppose it did. Say that  → b. Now pick
some α > b with α ∈ S . Then A+ = {β : β ≥ α} ∈  since A+ ∈
⊆ . We also
have A− = {β : β < α} ∈  since A− is an open neighborhood of b (and we claim 
converges to b).
But A+ ∩ A− = ∅, which violates the definition of a filter, so our assumption must
be wrong. Thus,  must not converge, and hence S is not compact.
44 The proof of this claim, in particular when we assert that there is an ultrafilter containing our filter, actually

relies on the axiom of choice.

632 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 122

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
 A– A+

b α

Figure 8. Illustration of A− and A+

Final Comments. Here ends the story, a sort of co-evolution of the two different but
related notions of sequential and open-cover compactness. Today when we use the
term “compact” we mean open-cover compact, but this paper, and the terms listed in
the Appendix, show that this was not always the case. The story of how open-cover
compactness came to be seen as the right one is a story of developing mathematics
without always knowing where it was going, how important terms should be defined,
and how widely they might be applied.
It might be worth noting, in closing, that even this paper, which attempts to char-
acterize the evolution of compactness, is only a sort of snapshot. In the time it took
to write this paper and revise it, textbooks have changed. For instance, the latest ver-
sion of a standard general topology textbook [34] now includes a discussion of nets
and filters, whereas the earlier version [33], available during the writing of the original
version of this paper, did not.
The lesson to be drawn is simply that mathematics evolves and changes as concepts
become clearer and are applied in more general situations. This may be obvious to the
mathematician who is involved in making these conceptual advances, but may be less
clear to the student who sees a textbook as a collection of established facts. Textbooks
are, as perhaps they should be, a distillation of what we currently know. They are also
historical documents in their own right. Being aware of this fact may help students
mature as mathematicians.

ACKNOWLEDGMENT. Many thanks to those who have supported this project in ways both small and
substantial: Hendrik Lenstra, Hans Wallin, Johan de Jong, Lucien Le Cam, Jeremy Gray, Umberto Bottazz-
ini, Gerald Folland, Michael Saks, Klas Markström, Victor Falgas-Ravry, Edouard Servan-Schreiber, Sophie
Laplante, and Lars-Daniel Öhman. Without this help this paper would not have been half as good, and I would
not have learned half as much. I am also indebted to three anonymous reviewers for their careful and meticu-
lous feedback. And I am grateful to several librarians, including Mikael Rågstedt at the Mittag–Leffler Institute,
who helped me track down original sources. This project was funded in part by a stipend from Wenner–Gren
Foundation. All mistakes are, of course, my own.

REFERENCES

1. P. Alexandroff, P. Urysohn, Sur les espaces topologiques compacts, Bull. Int. Acad. Pol. Sci. Lett. Ser. A
(1923) 5–8.
2. N. Andre, S. Engdahl, A. Parker, An analysis of the first proofs of the Heine–Borel theorem - Borel’s
proof, Loci (2013), http://dx.doi.org/10.4169/loci003890.
3. C. Arzelà, Funzioni di linee, Att. Real. Acca. Lin. 4 no. 5 (1889) 342–348.
4. G. Ascoli, Le curve limite di una varietà data di curve, Att. Real. Acca. Lin. 3 no. 18 (1883/4) 521–586.
5. G. Ascoli, C. Arzelà, Sulle funzioni di linee, Mem. R. Acad. Sci. Inst. Bologna 5 (1894) 225–244,
https://archive.org/stream/memoriedellar55189596racc#page/224/mode/2up.
6. G. Birkhoff, E. Kreyszig, The establishment of functional analysis, Historia Math. 11 (1984) 258–321.
7. B. Bolzano, Rein analytischer Beweis des Lehrsatzes dass zwischen je zwey Werthen, die ein entgegenge-
setztes Resultat gewähren, wenigstens eine reelle Wurzel der Gleichung liege, Prague, 1817.
8. E. Borel, Sur Quelques Points de la Théorie des Fonctions. Gauthier-Villars, Paris, 1894.
9. N. Bourbaki, Topologie Générale. Book X, Hermann, Paris, 1949.
10. ———, Topologie Générale. Book XVI, Hermann, Paris, 1953.

August–September 2015] A PEDAGOGICAL HISTORY 633

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
11. R. Chandler, G. Faulkner, Hausdorff compactifications: A retrospective, in Handbook of the History of
General Topology. Edited by C. E. Aull, R. Lowen. Springer-Verlag, Berlin, 2001, http://dx.doi.
org/10.1007/978-94-017-0468-7.
12. P. Cousin, Sur les fonctions de n variables complexes, Acta Math. 19 (1895) 1–61.
13. J. Dieudonné, History of Functional Analysis. Elsevier Science, Paris, 1981.
14. J. Dixmier, General Topology. Springer-Verlag, New York, 1985, http://dx.doi.org/10.1007/
978-1-4757-4032-5.
15. P. Dugac, Sur la correspondance de Borel et le theoreme de Dirichlet–Heine–Weierstrass–Borel–
Schoenflies–Lebesgue, Arch. Inter. Hist. Sci. 39 no. 122 (1989) 69–110.
16. G. Folland, A tale of topology, Amer. Math. Monthly 117 no. 8 (2010) 663–672.
17. M. Fréchet, Généralisation d’un théorème de Weierstrass, C. R. Acad. Sci. Paris 139 (1904) 848–850.
18. ———, Sur quelques point du calcul fonctionnel, Rend. Palermo 22 (1906) 1–74.
19. P. R. Halmos, Has progress in mathematics slowed down? Amer. Math. Monthly 97 no. 7 (1990) 561–588.
20. Handbook of the History of General Topology. Ed. by C. E. Aull, R. Lowen. Springer-Verlag, Berlin,
2001, http://dx.doi.org/10.1007/978-94-017-0468-7.
21. R. Hartshorne, Algebraic Geometry. Springer-Verlag, New York, 1977, http://dx.doi.org/10.
1007/978-1-4757-3849-0.
22. F. Hausdorff, Grundzüge der Mengenlehre. Verlag von Veit, Leipzig, 1914.
23. E. Heine, Die Elemente der Funktionenlehre, J. Reine Angew. Math. (1872) 172–188.
24. E. Hewitt, The role of compactness in analysis, Amer. Math. Monthly 67 no. 6 (1960) 499–516.
25. T. H. Hildebrandt, The Borel theorem and its generalizations, in The Chauvenet Papers: A Collection
of Prize-Winning Expository Papers in Mathematics. Ed. by J. C. Abbott. Mathematical Association of
America, Washington, DC, 1925.
26. J. Kelley, General Topology. D. Van Nostrand, Princeton, 1955.
27. M. Kline, Mathematical Thought: From Ancient to Modern Times. Oxford Univ. Press, Oxford, 1972.
28. T. Koetsier, J. Van Mill, By their fruits ye shall know them: Some remarks on the interaction of general
topology with other areas of mathematics, in History of Topology. North-Holland Publishing, Amsterdam,
1999. 199–239.
29. H. Lebesgue, Leçons sur l’Intégration et la Recherche des Fonctions Primitives. Gauthier-Villars, Paris,
1902.
30. Mathematical Society of Japan, Encyclopedic Dictionary of Mathematics. Second edition. MIT Press,
Cambridge, MA, 1987.
31. E. J. McShane, Partial orderings and Moore–Smith limits, in The Chauvenet Papers: A Collection of
Prize-Winning Expository Papers in Mathematics. Ed. by C. Abbott. Mathematical Association of Amer-
ica, Washington, DC, 1950.
32. E. H. Moore, H. L. Smith, A general theory of limits, Amer. J. Math. 44 (1922) 102–121.
33. J. R. Munkres, Topology: A First Course. First Edition. Prentice-Hall, Englewood Cliffs, NJ, 1975.
34. ———, Topology: A First Course. Second Edition. Prentice-Hall, Englewood Cliffs, NJ, 2000.
35. G. Peano, Démonstration de l’intégrabilité de équations différentielles ordinaires, Math. Ann. 37 (1890)
182–228.
36. J. P. Pier, Historique de la notion de compacité, Historia Math. 7 (1980) 425–443.
37. W. Purkert, The double life of Felix Hausdorff/Paul Mongré. Math. Intelligencer, Trans. by H. and D.
Rowe, 30 no. 4 (2008) 36–50.
38. M. Raman, Understanding Compactness: A Historical Perspective, Master of Arts Thesis, Univ. of Cali-
fornia, Berkeley, 1997.
39. F. Riesz, Die genesis des Raumbegriffes, Math. Natwiss. 24 (1907) 309–353.
40. A. Schönflies, Die Entwickelung der Lehre von den Punktmannigfaltigkeiten, in Jahresbericht Deutsh
Mathers Verein. B.G. Teubner, Leipzig, 1900.
41. H. L. Smith, A general theory of limits, Natl. Math. Mag. 7 no. 8 (1938) 371–379.
42. L. Steen, J. A. Seebach, Counterexamples in Topology. Holt, Rinehart and Winston, New York, 1970,
http://dx.doi.org/10.1007/978-1-4612-6290-9.
43. A. Taylor, A study of Maurice Fréchet: I. His early work on point set theory and the theory of functionals,
Arch. Hist. Exact Sci. 27 no. 3 (1982) 233–295.
44. ———, A study of Maurice Fréchet: II. Mainly about his work on general topology 1909–1928, Arch.
Hist. Exact Sci. 34 no. 3 (1985) 279–380.
45. A. Tychonoff, Ein Fixpunktsatz, Math. Ann. 111 no. 1 (1935) 767–776.
46. K. Weierstrass, Einleitung in die Theorie der Analytischen Funktionen. Vorlesung Berlin, written down
by A. Hurwitz, worked out by P. Ullrich, 1878.

MANYA RAMAN-SUNDSTRÖM is an Associate Professor of mathematics education. She works mostly

on the teaching and learning of mathematical proof and on mathematical aesthetics.

634 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 122

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms
Department of Science and Mathematics Education, Umeå University, Sweden, SE-90187
[email protected]

Appendix: Terminology There are many notions related to (but not necessarily
equivalent to) compactness. Table 1 contains a list of some of these notions. For a
more comprehensive list see [42].

Table 1. Flavors of compactness

[open-cover] compact: Every open cover has a finite subcover.
(also called the Borel–Lebesgue property)
sequentially compact: Every sequence has a convergent subsequence.
countably compact: Every countable open cover has a finite subcover.
limit-point compact: Every infinite subset of X has a limit point in X .
(also called Fréchet compact or the Bolzano–Weierstrass property)
relatively compact: The closure is compact.
pseudo-compact: Each continuous real valued function on X is bounded.
finally compact: Every open cover has a countable subcover.
(also called Lindelöf compact)

Many of these concepts are related. For instance, compactness implies countable
compactness implies limit point compactness. Sequential compactness implies count-
able compactness. If we put further restrictions on our spaces we can get implications
in the other direction. In T1 spaces, limit-point compactness implies countable com-
pactness. In first countable spaces, countable compactness implies sequential com-
pactness. In second countable spaces, sequential compactness implies compactness.
In particular, we know that in compact metric spaces, which turn out to be second
countable, the first four notions of compactness in Table 1 are equivalent.
It took some time as compactness was applied to different types of spaces for rela-
tionships like these to be worked out. It also took time for names to stabilize. Table 2
lists different terms used for compactness-related ideas used by some of the most influ-
ential mathematicians in the historical development. In this paper, the modern names
have been used.

Table 2. Names of historical compactness-related terms

Who When Their term Modern term
Fréchet 1906 compact relatively sequentially compact
extremal sequentially compact
Russian School 1920s bicompact compact
(Alexandroff, etc.) compact countably compact
Bourbaki 1930s quasi-compact compact
compact compact and Hausdorff

August–September 2015] A PEDAGOGICAL HISTORY 635

This content downloaded from 128.83.63.20 on Wed, 30 Mar 2016 12:00:05 UTC
All use subject to http://about.jstor.org/terms