A “Bouquet” of Discontinuous Functions for

Beginners in Mathematical Analysis

Giacomo Drago, Pier Domenico Lamberti, and Paolo Toni

Abstract. We present a selection of a few discontinuous functions and we discuss some peda-
gogical advantages of using such functions in order to illustrate some basic concepts of math-
ematical analysis to beginners.

1. INTRODUCTION. In this paper we present a selection of several discontinuous

real-valued functions of one real variable which we believe could be proposed to any
beginner in mathematical analysis, even to students of secondary school dealing with
the first notions of calculus. Some of them are elementary and well known, others a bit
more sophisticated. Most of them are modeled on the Dirichlet function.
Our aim is to point out some pedagogical advantages of using discontinuous func-
tions rather than classical analytic functions. Usually, young students tend to think
of basic mathematical analysis as a set of rules of calculus and, surely, many under-
graduate students would claim to feel more acquainted with the notion of derivative
rather than with the notion of function. This is not surprising at all, if we consider that
the historical evolution of the notion of function has been very long and troubled;
see Youschkevitch [15] and Kleiner [8, 9]. See also Nicholas [11], Deal [3], and
Thurston [13] for an interesting discussion about some pedagogical issues related to
the definition of function. A modern definition of function can be found, e.g., in the
classic book of Bourbaki [2], published more than two hundred years after the defini-
tion of Johann Bernoulli (1718). As is pointed out in [15, p. 79], a significant step in
this process was the formulation of A. Cournot (1841), which we report here for the
convenience of the reader:

We understand that a quantity may depend on another [quantity], even in case

the nature of this dependence is such that it cannot be expressed in terms of a
combination of algebraic symbols.

This level of generality is commonly attributed to Dirichlet, who in 1829 proposed his
celebrated function D defined on (0, 1) as follows:
(
1, if x ∈ (0, 1) ∩ Q,
D(x) = (1)
0, if x ∈ (0, 1) \ Q.

In fact, this example opened a door to a new world: functions are not just formulas,
or analytic expressions, as was commonly assumed in the 18th century. Functions can
be defined by very general laws.
From a pedagogical point of view, deciding the level of generality of functions to
use with young students is not straightforward. According to Kleiner [9, pp. 187–188],
it is possible to teach an elementary model of analysis by placing emphasis solely on
curves and the equations that represent them, without necessarily talking about func-
tions. Kleiner argues that students would find curves more natural than functions and
http://dx.doi.org/10.4169/amer.math.monthly.118.09.799

November 2011] A “BOUQUET” OF DISCONTINUOUS FUNCTIONS 799

teachers should introduce the notion of function only when there is an evident need for
it. Teaching mathematics should follow the historical evolution of mathematics itself:
new definitions and concepts were introduced when the need arose, and it should be so
in teaching too. Thus, the definition of function should be given gradually, following
its historical path. At first, one should introduce functions as formulas, then as rules,
and only at last, if required, as a set of ordered pairs (after all, as Kleiner [9] points
out, “giving this latter definition and proceeding to discuss only linear and quadratic
functions makes little pedagogical sense”).

Figure 1. Graph of the Dirichlet function.

In [14] another approach has been developed. The introduction of [14] reads: “after
all, analysis has few bases: the concepts of function, infinity, limit, continuity, and
differentiability. To visit these locations in a comprehensive way, one should allow
these concepts the faculty of expressing all of their potential and all their fantastic
and incredible situations. To achieve this high level it is necessary to stick closely to
the definition of function as a correspondence of free sets.” The modern definition
of function clearly allows a deeper comprehension of other concepts, such as those
of limit, continuity, and differentiability. It is important that students understand the
far-reaching features of these notions, because only in this way will they be able to
avoid mistakes due to a limited concept of function. Students should be aware from
the very beginning that formulas provide a very small class of functions: analyticity is
a property enjoyed by very special functions. Being discontinuous is not synonymous
with being pathological because most functions are discontinuous. In the same way,
being continuous is not synonymous with being smooth since continuous functions
can be very rough, as in the case of the celebrated Weierstrass functions.
Following this plan, in [14] the authors propose a collection of problems the solu-
tions of which require the modern definition of function: in some cases, using highly
discontinuous functions is not essential but it simplifies the argument and makes the
situation clear.
We note that there is a vast literature devoted to so-called pathological functions. We
mention the classic book by Gelbaum and Olmsted [6] and the recent extensive mono-
graph by Kharazishvili [7]. We also mention Thim [12], which is basically a treatise
concerning continuous nowhere differentiable functions. However, at a pedagogical
level, not much material seems to be available.
In this paper, we adopt the approach of [14] and we discuss some peculiar examples,
some of which are taken from [14]. The focus is on special discontinuous functions,
the definitions of which are algebraically and technically simple in most cases. We
believe that this approach could be used not only with good students but also with weak

800 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 118

students who may take advantage of examples which illustrate profound concepts but
remain at a low level of complexity.

2. EXAMPLES OF DIRICHLET-TYPE FUNCTIONS. In this section we present

several discontinuous functions modeled on the Dirichlet function (1). We think that
students may find such examples easy and eventually amusing, once they are ac-
quainted with function (1). On the other hand, asking a student of a first-year calculus
course to provide such examples could be very challenging.

1. A function continuous at only one point. If we ask a beginner in mathematical

analysis whether a function from R to itself can be continuous at only one point, most
of the time we will get the answer “no” as a consequence of an over-simplified view
of the concept of continuity. Usually, students have an idea of continuity as a global
property of a function: according to this point of view, continuous functions are those
functions whose graphs can be drawn without lifting the chalk from the blackboard.
This vision is not completely wrong since a function is continuous in the whole of an
interval if and only if its graph is a pathwise-connected set in the plane. However, the
following example clarifies the local nature of continuity.

–3 –2 –1 1 2 3

–1

–2

–3
Figure 2. Graph of function F1 .

Let F1 be the function from R to itself defined by

(
x, if x ∈ Q,
F1 (x) =
−x, if x ∈ R \ Q.

The function F1 is continuous only at x = 0.

As a variant of the function F1 , one can consider the function G 1 from R to itself
defined by
(
sin x, if x ∈ Q,
G 1 (x) =
− sin x, if x ∈ R \ Q.

November 2011] A “BOUQUET” OF DISCONTINUOUS FUNCTIONS 801

 2

–2 – 2
–1

–2
Figure 3. Graph of function G 1 .

The function G 1 is continuous only at a countable set of points, precisely at the points
x = kπ with k ∈ Z.

2. A function differentiable at only one point. Loosely speaking, the derivative of

a function at a point is the slope of the tangent line to the graph of the function at
that point. However, the notion of differentiability has a local nature which goes much
beyond the geometric idea of a tangent line.
Let F2 be the function from R to itself defined by
(
x 2, if x ∈ Q,
F2 (x) =
−x 2 , if x ∈ R \ Q.

The function F2 is differentiable only at x = 0 and is discontinuous at any point

x 6= 0. Proving that F2 is differentiable at x = 0 is an easy exercise. However, it may
be instructive to deduce such a proof from the following elementary lemma.

Lemma 2. Let f, g, and h be functions from R to itself such that f (x) ≤ h(x) ≤ g(x)
for all x ∈ R. Let x0 ∈ R be fixed. If f and g are differentiable at x0 , f 0 (x0 ) = g 0 (x0 ),
and f (x0 ) = g(x0 ), then h is differentiable at x0 and h 0 (x0 ) = f 0 (x0 ) = g 0 (x0 ).

3 3

2 2

1 1

–2 –1 1 2 –2 –1 1 2

–1 –1

–2 –2

–3 –3
Figure 4. Graph of function F2 . Figure 5. A randomly generated function differ-
entiable at x = 0.

802 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 118

 Students may find it entertaining to apply the previous lemma to randomly gener-
ated functions h satisfying the condition −x 2 ≤ h(x) ≤ x 2 for all x in a neighborhood
of zero and find out that such functions are differentiable at zero (see Figure 5).

3. A function with positive derivative at one point, which is not monotone in any
neighborhood of that point. Let F3 be the function from R to itself defined by
(
x 2, if x ∈ Q,
F3 (x) =
2x − 1, if x ∈ R \ Q.

By Lemma 2, the function F3 is differentiable at the point x = 1 and F30 (1) = 2, but
F3 is not monotone in any neighborhood of 1. Those students who are used to identify-
ing increasing functions with functions with positive derivative may find this example
rather bizarre. However, this example points out a well-known subtle distinction con-
cerning increasing functions. In fact, if a function f from R to itself is differentiable
at a point x0 and f 0 (x0 ) > 0 then f is increasing at the point x0 in the sense that
for all x in a convenient neighborhood of x0 we have: f (x) > f (x0 ) if x > x0 and
f (x) < f (x0 ) if x < x0 . Monotonicity is a stronger notion and, in the case of differ-
entiable functions, it occurs when the derivative does not change sign in the whole of
an interval.

Figure 6. Graph of function F3 .

Variants of the function F3 can be easily defined by considering convex combina-

tions of x 2 and 2x − 1 of the form

θ(x)x 2 + (1 − θ (x))(2x − 1),

where θ is a suitable function taking values in [0, 1]. A “continuous version” of F3 is

given for example by the function G 3 from R to itself defined by1
9 9
    
2 2 2
G 3 (x) = sin · x + 1 − sin · (2x − 1),
x −1 x −1
1 The choice of the number 9 in the definition of G is not essential, and is only aimed at emphasizing the
3
oscillations of the graph in Figure 7.

November 2011] A “BOUQUET” OF DISCONTINUOUS FUNCTIONS 803

for all x ∈ R \ {1}, and G 3 (1) = 1. Again, G 03 (1) = 2 > 0, but G 3 is not monotone in
any neighborhood of the point x = 1.
Clearly the function G 3 is much smoother than F3 ; however, defining F3 and prov-
ing its nonmonotonicity is simpler.

2.5

2.0

1.5

1.0

0.5

–0.5 0.5 1.0 1.5

–0.5

Figure 7. Graph of function G 3 .

3. FURTHER EXAMPLES. In this section we present examples of functions with a

specific type of discontinuity at any point of the domain, such as removable discontinu-
ities or jump discontinuities. Although complicated, the definitions of these functions
are constructive and make use of elementary notions. We note that imposing a specific
type of discontinuity at any point of the domain of a function leads to limitations on
the structure of the domain itself. In fact, the following result in Klippert [10] holds.

Theorem 3. Let f be a function from R to itself. Let A be the set of those points x0 ∈ R
such that f is discontinuous at x0 and at least one of the two limits limx→x + f (x) and
0
limx→x − f (x) exists and is finite. Then A is at most countable.
0

4. A function with removable discontinuities at every point of its domain. The

following function F4 is defined on the dyadic rational numbers in (0, 1), i.e., those
numbers in (0, 1) with a finite binary expansion. A simple formal definition is given
below in (6). However, we prefer to begin by describing the elementary geometric
construction given in [14], which could also be presented to beginners.
We construct the function F4 by means of the following iterative procedure (see
Figure 9).
Step 1. Let V = 0 and L = 1, and consider the square in the Euclidean plane with
edges parallel to the coordinate axes whose lower left vertex is (V, V ) and

804 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 118

 1.0

0.8

0.6

0.4

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0
Figure 8. Graph of function F4 (arrested at the fourth iteration).

whose edges have length L. Let

 n 
X L
BV,L = V+ : n∈N .
j=1
2j

Then we set F4 (x) = V + L for all x ∈ BV,L .

Step 2. Inside the square defined in Step 1, consider the sequence of squares with
edges of length L n = 2Ln and parallel to the coordinate axes, and lower left
vertices with coordinates (Vn , Vn ), where Vn = V + n−1 j=1 2 j for all n ∈ N
L
P
with n ≥ 2, and (V1 , V1 ) = (V, V ).
Repeat Step 1 and Step 2, using each square obtained in Step 2 in place of
the original square, and iterate the process.
The domain of the function F4 defined by this procedure is B = ∪V,L BV,L , which is
clearly the set of the dyadic rational numbers in (0, 1).
Looking closely at the construction described above and at the few first iterations
(see Figure 9), it is evident that

lim F4 (x) = v < F4 (v) (4)

x→v

for all v ∈ B. Thus F4 has a removable discontinuity at each point v ∈ B. Indeed, in

order to remove the discontinuity at a point v, it is enough to redefine the function by
setting F̃4 (v) = v.
It is possible to give a non-iterative definition of the function F4 . Indeed the set B
can be represented as

November 2011] A “BOUQUET” OF DISCONTINUOUS FUNCTIONS 805

1.0 1.0

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0

1.0 1.0

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Figure 9. The first four iterations.

np o
B= : n, p ∈ N, p is odd, p < 2 n
(5)
2n

and F4 can be defined on B directly by the equality

 p p+1
F4 = (6)
2n 2n

for all n, p ∈ N, where p is odd and p < 2n . This alternative definition allows an
easy proof of (4) since the summand 1/2n in the right-hand side of (6) vanishes in the
limiting procedure (see also (8) and (9)).
A well-known function which enjoys a similar property can be found in several
textbooks. It is defined as follows. Let G 4 be the function from (0, 1) ∩ Q to itself
defined by

1
 
p
G4 = (7)
q q

806 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 118

 1
2

1
3
1
4
1
5

1 1 1 2 3
4 3 2 3 4
1

Figure 10. Graph of function G 4 .

for all p, q ∈ N, 0 < p/q < 1, p, q coprime. It is not difficult to prove that

lim G 4 (x) = 0 6 = G 4 (x0 ) (8)

x→x0

for all x0 ∈ (0, 1) ∩ Q. The function G 4 can be extended to the whole of (0, 1) by
setting G 4 (x) = 0 for all x ∈ (0, 1) \ Q and turns out to be continuous at any point
x ∈ (0, 1) \ Q. According to Appell [1, p. 10] some students would call function G 4
“Dirichlet light” because its set of discontinuities is “lighter” than the set of disconti-
nuities of the Dirichlet function.
We note that by (6) and (7) it immediately follows that

F4 (x) = x + G 4 (x) (9)

for all x ∈ B. Equality (9) could be used to define F4 via G 4 and to give a further
proof of (4). In this way, the function F4 , whose original definition is based on a geo-
metric construction, can eventually be represented via a function whose definition has
a completely different origin.

5. A function with jump discontinuities at all points of its domain. Monotone func-
tions with many jump discontinuities are well known in real analysis and probability
theory: these functions can be obtained as cumulative distribution functions associated
with discrete measures (see, e.g., Folland [5, p. 102]). Here we present two functions
defined on the set B of the dyadic rational numbers in (0, 1) with jump discontinuities
at every point of B. These functions are defined directly by using binary notation and
do not require any advanced tool.
We recall that the set B in (5) can be represented also in the form
∞ Xn  
[ x j 
B= x j ∈ {0, 1} for all j ∈ {1, . . . , n}, x n = 1 . (10)
n=1 j=1
2j 

Thus, given x ∈ B there exists n(x) ∈ N such that x = n(x) j=1 x j /2 for suitable values
j
P
of x j ∈ {0, 1}, where xn(x) = 1; using binary notation x can be written as

0. x1 x2 · · · xn(x) .

Let F5 be the function from B to itself defined by

F5 (0. x1 x2 · · · xn(x) ) = 0. y1 y2 · · · ym(x) ,

November 2011] A “BOUQUET” OF DISCONTINUOUS FUNCTIONS 807

where the digits y1 , . . . , ym(x) are obtained from the digits x1 , . . . , xn(x) of x by
simply replacing each 0 with 01. For example, F5 (0.01) = 0.011, F5 (0.101001) =
0.101101011, F5 (0.000011) = 0.0101010111, and F5 (0.1) = 0.1. It is easy to prove
that F5 is a monotone function strictly increasing on B. Thus, F5 admits left and right
limits at any point of B. Moreover, it is not difficult to realize that in fact

F5 (v) = lim F5 (x) < lim F5 (x)

x→v − x→v +

for all v ∈ B. Hence F5 is left-continuous with a jump discontinuity at every point

v ∈ B. In order to get rid of the monotonicity one can consider the function G 5 defined
by

G 5 = F4 ◦ F5 .

The function G 5 is particularly interesting since it admits a geometric description in the

spirit of the iterative construction of F4 : in this case one has to consider rectangles with
no common vertices rather than squares (see Figures 11 and 12). The function G 5 is
discussed in detail in Drago [4], where it is also proved by using binary representations
that for all v ∈ B the one-sided limits of G 5 at v exist and satisfy

lim G 5 (x) < lim G 5 (x) < G 5 (v).

x→v − x→v +

1.0

0.8

0.6

0.4

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0
Figure 11. Graph of function G 5 (arrested at the third iteration).

6. A function approaching infinity at any point of its domain. The following ex-
ample is a simple variant of the function G 4 . Let F6 be the function from (0, 1) ∩ Q to
itself defined by
 
p
F6 =q
q

808 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 118

 1.0

0.8

0.6

0.4

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0
Figure 12. Construction of function G 5 .

50

40

30

20

10

0
1 1 1 2 3 1
4 3 2 3 4

Figure 13. Graph of function F6 .

November 2011] A “BOUQUET” OF DISCONTINUOUS FUNCTIONS 809

for all p, q ∈ N, 0 < p/q < 1, p, q coprime. It is not difficult to prove that

lim F6 (x) = +∞
x→x0

for all x0 ∈ [0, 1].

ACKNOWLEDGMENTS. The authors are thankful to Prof. Maurizio Emaldi for pointing out reference [9]
and useful discussions, and also to Dr. Matteo Dalla Riva for pointing out reference [10]. The first two authors
are deeply indebted to the third author, P. Toni, who, in different periods, was their teacher of mathematics
and physics in secondary school, where he has developed and experienced the main pedagogical ideas that
are the basis of the present paper for about three decades. Those stimulating lessons led the second author,
P. D. Lamberti, to define the function F4 . The functions F5 and G 5 were defined by the first author, G. Drago,
in his dissertation [4], written under the supervision of the second author.
The authors are very thankful to a referee for suggesting formula (6), which simplifies a binary representa-
tion given in the first version of this paper, and for pointing out formula (9).

REFERENCES

1. J. Appell, Analysis in Beispielen und Gegenbeispielen, Springer, Berlin, 2009.

2. N. Bourbaki, Eléments de Mathématique, Théorie des Ensémbles, Hermann, Paris, 1970.
3. R. B. Deal, Some remarks on “A dilemma in definition,” Amer. Math. Monthly 74 (1967) 1258–1259.
http://dx.doi.org/10.2307/2315696
4. G. Drago, Funzioni e Funzioni Strane: Alcuni Aspetti Storici e Didattici, B.A. thesis, University of
Padova, Padova, 2009.
5. G. B. Folland, Real Analysis: Modern Techniques and Their Applications, John Wiley, New York, 1984.
6. B. R. Gelbaum and J. M. H. Olmsted, Counterexamples in Analysis, Holden-Day, San Francisco, 1964.
7. A. B. Kharazishvili, Strange Functions in Real Analysis, 2nd ed., Pure and Applied Mathematics, vol.
272, Chapman & Hall/CRC, Boca Raton, FL, 2006.
8. I. Kleiner, Evolution of the function concept: A brief survey, College Math. J. 20 (1989) 282–300. http:
//dx.doi.org/10.2307/2686848
9. , Functions: Historical and pedagogical aspects, Science & Education 2 (1993) 183–209. http:
//dx.doi.org/10.1007/BF00592206
10. J. Klippert, Advanced advanced calculus: Counting the discontinuities of a real-valued function with
interval domain, Math. Mag. 62 (1989) 43–48. http://dx.doi.org/10.2307/2689945
11. C. P. Nicholas, A dilemma in definition, Amer. Math. Monthly 73 (1966) 762–768. http://dx.doi.
org/10.2307/2313993
12. J. Thim, Continuous Nowhere Differentiable Functions, M.A. thesis, Luleå University of Technology,
Luleå, Sweden, 2003.
13. H. A. Thurston, A reply to “A dilemma in definition,” Amer. Math. Monthly 74 (1967) 1259–1261. http:
//dx.doi.org/10.2307/2315697
14. P. Toni and P. D. Lamberti, Esplorando l’Analisi Matematica, Società Editrice Internazionale, Torino,
1996.
15. A. P. Youschkevitch, The concept of function up to the middle of the 19th century, Arch. Hist. Exact Sci.
16 (1976/77) 37–85.

GIACOMO DRAGO is 22 years old. He got his bachelor’s degree with honors from the University of Padova
in 2009. He distinguished himself in the national finals of three consecutive editions of the Kangourou Mathe-
matics Competitions in Italy (2004–2006). He is currently working on his master’s degree in mathematics, and
he dreams of becoming a teacher of mathematics.
Dipartimento di Matematica Pura ed Applicata, Università degli Studi di Padova, Via Trieste 63, 35121
Padova, Italy
[email protected]

PIER DOMENICO LAMBERTI received his Ph.D. in mathematics from the University of Padova in 2003.
He currently has a position as researcher in mathematical analysis at the same university. His research in-

810 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 118

terests are mainly in spectral stability problems for elliptic partial differential equations. He likes to think of
mathematics as a form of art.
Dipartimento di Matematica Pura ed Applicata, Università degli Studi di Padova, Via Trieste 63, 35121
Padova, Italy
[email protected]

PAOLO TONI received his master’s degree in mathematics from the University of Firenze in 1972. He has
taught mathematics and physics in high school for almost 40 years. He was a pioneer in the organization of
mathematics competitions in Italy. His interests are mainly in mathematical creativity and teacher education.
He retired in 2010. Now, he enjoys staying with his nephews and playing the organ. He believes in students’
creativity.
Liceo Scientifico Statale E. Fermi, Via Vittorio Emanuele II 50, 35100 Padova, Italy
[email protected]

 n+ 12
1 
A Ramanujan-Type Formula for 1+ e
n

n+ 12
Motivated by Sanjay Khattri’s proofs [1] that e < 1 + n1 , I offer the follow-
ing formula, which I think Ramanujan would have liked, for the ratio of the two
sides of Khattri’s inequality:
 1 Z 1
1 n+ 2  u2
 
1+ e = exp du
0 (2n + 1) − u
n 2 2

Z 1
u2

= exp du (where m = n(n + 1)/2)
0 8m + 1 − u
2

Z 1
αu 2

= exp du (where α = 1/(8m))
0 1 + α(1 − u )
2

Z 1 
= exp αu 2 − α 2 u 2 (1 − u 2 ) + α 3 u 2 (1 − u 2 )2 − α 4 u 2 (1 − u 2 )3 + · · · du
0

1 1 2 1 4 2 1 6 4 2 1 8 6 4 2 5
 
= exp α − · α2 + · · α3 − · · · α4 + · · · · α − ···
3 5 3 7 5 3 9 7 5 3 11 9 7 5 3
α 2α 4α 6α 8α
     
= exp 1− 1− 1− 1− 1 − ···


3 5 7 9 11
1 1 2 3 4
     
= exp 1− 1− 1− 1− 1 − ··· .


24m 5 · 4m 7 · 4m 9 · 4m 11 · 4m

REFERENCES
 n+0.5
1. S. K. Khattri, Three proofs of the inequality e < 1 + n1 , Amer. Math. Monthly 117 (2010)
273–277. http://dx.doi.org/10.4169/000298910X480126

—Submitted by Michael D. Hirschhorn,

University of New South Wales, Sydney, Australia

November 2011] A “BOUQUET” OF DISCONTINUOUS FUNCTIONS 811