Beyond Mere Convergence

James A. Sellers
Department of Mathematics
The Pennsylvania State University
107 Whitmore Laboratory
University Park, PA 16802
[email protected]
February 5, 2002 REVISED
Abstract
In this article, I suggest that calculus instruction should include a
wider variety of examples of convergent and divergent series than is
usually demonstrated. In particular, a number of convergent series,
such as

k1
k
3
2
k
, are considered, and their exact values are found in a
straightforward manner. We explore and utilize a number of math-
ematical topics, including manipulation of certain power series and
recurrences.
During my most recent spring break, I read William Dunhams book
Euler: The Master of Us All [3]. I was thoroughly intrigued by the material
presented and am certainly glad I selected it as part of the weeks reading.
Of special interest were Dunhams comments on series manipulations and
the power series identities developed by Euler and his contemporaries, for I
had just completed teaching convergence and divergence of innite series in
my calculus class. In particular, Dunham [3, p. 47-48] presents Eulers proof
of the Basel Problem, a challenge from Jakob Bernoulli to determine the
1
exact value of the sum

k1
1
k
2
. Euler was the rst to solve this problem by
proving that the sum equals

2
6
.
I was reminded of my students interest in this result when I shared it
with them just weeks before. I had already mentioned to them that exact
values for relatively few families of convergent series could be determined.
The obvious examples are geometric series

k0
r
k
(with |r| < 1) and telescop-
ing series. I also remembered their disappointment when I observed that the
exact numerical value of most convergent series cannot be determined in a
straightforward way. I tried to excite them with the notion that the con-
vergence or divergence of a given series could be determined via the Integral
Test, Limit Comparison Test, Ratio or Root Test, but this was received with
little enthusiasm.
But now I return to Dunhams book. In [3, p. 41], Dunham notes that
Jakob Bernoulli [2, p. 248-249] proved

k1
k
2
2
k
= 6 (1)
and

k1
k
3
2
k
= 26. (2)
Many teachers of calculus will recognize at least two things about (1) and (2).
First, these series are made-to-order examples to demonstrate convergence
with the Ratio Test. Such examples, where the summands are dened by
the ratio of a polynomial and an exponential function, can be found in a
number of calculus texts, such as [4] and [5]. Second - a much more negative
admission - is that we rarely teach students how to prove equalities like
(1) and (2). We usually stop at demonstrating that such series converge,
and move on to other matters. This is the case with the two calculus texts
mentioned above, and it is an unfortunate situation to say the least.
I contend that students of rst-year calculus would be better served if we
provided a few more tools to them for nding exact values of convergent
innite series. Oddly enough, the series in (1) and (2) are ideal for such a
task.
2
My goal in this note is to present two approaches to nding the exact
value of
a(m, n) :=

k1
k
n
m
k
with |m| > 1 and n N{0} (of which Bernoullis examples (1) and (2) are
special cases).
We begin by noting that, for each |m| > 1,

1
m

< 1, so that a(m, 0) is a

k1
1
m
k
=
1
m
+

k2
_
1
m
_
k
=
1
m
+
1
m

k1
_
1
m
_
k
=
1
m
+
1
m
a(m, 0).
Solving for a(m, 0), we see that it equals
1
m1
. Of course, this result easily
follows from the usual formula for the sum of a convergent geometric series.
Next, we obtain a recurrence for a(m, n), n 1, in terms of a(m, j) for
j < n. Note that
a(m, n) =

k1
k
n
m
k
=
1
m
+

k2
k
n
m
k
=
1
m
+
1
m

k1
(k + 1)
n
m
k
=
1
m
_
1 +

k1
(k + 1)
n
m
k
_
.
3
The argument up to this point is exactly that used in nding the formula for
a(m, 0) above. We now employ the binomial theorem, a tool that should be
in the repertoire of rst-year calculus students.
a(m, n) =
1
m
_
_
1 +

k1
_

n
j=0
_
n
j
_
k
j
_
m
k
_
_
=
1
m
_
1 +
n

j=0
_
n
j
_

k1
k
j
m
k
_
=
1
m
_
1 +
n1

j=0
_
n
j
_

k1
k
j
m
k
+

k1
k
n
m
k
_
=
1
m
_
1 +
n1

j=0
_
n
j
_
a(m, j) + a(m, n)
_
=
1
m
a(m, n) +
1
m
_
1 +
n1

j=0
_
n
j
_
a(m, j)
_
Solving for a(m, n) yields
_
1
1
m
_
a(m, n) =
1
m
_
1 +
n1

j=0
_
n
j
_
a(m, j)
_
or
a(m, n) =
_
1
m1
_
_
1 +
n1

j=0
_
n
j
_
a(m, j)
_
. (3)
As a sidenote, it is interesting to see from (3) that, for rational values of
m, the numerical value of a(m, n) must be rational for all n 0. This can
be proven via induction on n. We noted above that a(m, 0) =
1
m1
which
is rational as long as m is rational. Then, assuming a(m, j) is rational for
0 j n 1, (3) implies a(m, n) is also rational. Hence, no values such as

2
6
will arise as values for a(m, n) whenever m is rational.
The recurrence in (3) can be used to calculate with relative ease the exact
value of
a(m, n) =

k1
k
n
m
k
4
for all |m| > 1 and n N {0}. For example, since
a(2, 0) =

k1
1
2
k
= 1,
we have
a(2, 1) =

k1
k
2
k
=
_
1
2 1
__
1 +
_
1
0
_
a(2, 0)
_
= 1 + 1 = 2,
and
a(2, 2) =

k1
k
2
2
k
= 1 +
_
2
0
_
a(2, 0) +
_
2
1
_
a(2, 1)
= 1 + 1 + 2 2 = 6,
which is the result labeled (1). Finally,
a(2, 3) =

k1
k
3
2
k
= 1 +
_
3
0
_
a(2, 0) +
_
3
1
_
a(2, 1) +
_
3
2
_
a(2, 2)
= 1 + 1 + 3 2 + 3 6 = 26,
which is (2).
Of course, recurrence (3) could be used to calculate a(m, n) for larger
values of m and n. However, this might prove tedious for extremely large
values of n. With this in mind, we now approach the calculation of a(m, n)
from a second point of view.
We begin with the familiar power series representation for the function
1
1 x
:
1
1 x
= 1 + x + x
2
+ x
3
+ x
4
+ . . . , where | x| < 1 (4)
5
Andrews [1] recently extolled the virtues of (4) in the study of calculus. Our
goal in this section is to manipulate (4) via dierentiation and multiplication
to obtain a new power series of the form
f
n
(x) := x + 2
n
x
2
+ 3
n
x
3
+ 4
n
x
4
+ . . . =

k1
k
n
x
k
for a xed positive integer n. This is done by applying the x
d
dx
operator
to
1
1x
n times. Then a(m, n) equals f
n
_
1
m
_
, which is easily computed
once f
n
(x) is written as a rational function. (Note that we dene f
0
(x) by
f
0
(x) := x
_
1
1x
_
=

k1
x
k
.)
As an example, we apply the x
d
dx
operator to
1
1x
and get
x
d
dx
_
1
1 x
_
= x
d
dx
(1 + x + x
2
+ x
3
+ x
4
+ . . .)
or
f
1
(x) =
x
(1 x)
2
= x + 2x
2
+ 3x
3
+ 4x
4
+ . . . =

k1
kx
k
.
Hence,

k1
k
2
k
= f
1
_
1
2
_
=
1
2
_
1
1
2
_
2
= 2.
We can apply the x
d
dx
operator to
1
1x
twice to obtain f
2
(x) :
f
2
(x) = x
d
dx
_
x
d
dx
_
1
1 x
__
= x
d
dx
_
x
(1 x)
2
_
=
x
2
+ x
(1 x)
3
.
Thus,
f
2
(x) =
x
2
+ x
(1 x)
3
= x + 2
2
x
2
+ 3
2
x
3
+ 4
2
x
4
+ . . . =

k1
k
2
x
k
.
6
Hence,

k1
k
2
2
k
= f
2
_
1
2
_
=
1
2
+
_
1
2
_
2
_
1
1
2
_
3
= 6
upon simplication. This, as we have already seen, is (1).
Additional applications of the x
d
dx
operator can be performed to yield
f
1
(x) =
x
(1 x)
2
=

k1
kx
k
,
f
2
(x) =
x
2
+ x
(1 x)
3
=

k1
k
2
x
k
,
f
3
(x) =
x
3
+ 4x
2
+ x
(1 x)
4
=

k1
k
3
x
k
,
f
4
(x) =
x
4
+ 11x
3
+ 11x
2
+ x
(1 x)
5
=

k1
k
4
x
k
,
f
5
(x) =
x
5
+ 26x
4
+ 66x
3
+ 26x
2
+ x
(1 x)
6
=

k1
k
5
x
k
, and
f
6
(x) =
x
6
+ 57x
5
+ 302x
4
+ 302x
3
+ 57x
2
+ x
(1 x)
7
=

k1
k
6
x
k
.
We see that
f
n
(x) =
g
n
(x)
(1 x)
n+1
for each n 1 where g
n
(x) is a certain polynomial of degree n. Indeed,
the functions g
n
(x) are well-known. Upon searching N.J.A. Sloanes On-Line
Encyclopedia of Integer Sequences [6] for the sequence
1, 1, 1, 1, 4, 1, 1, 11, 11, 1, 1, 26, 66, 26, 1, . . . ,
which is the sequence of coecients of the polynomials g
n
(x), we discover
that these are the Eulerian numbers e(n, j). They are dened, for each
value of j and n satisfying 1 j n, by
e(n, j) = je(n 1, j) + (n j + 1)e(n 1, j 1) with e(1, 1) = 1. (5)
7
With this notation, it appears that, for n 1,
f
n
(x) =
n

j=1
e(n, j)x
j
(1 x)
n+1
.
Using (5), this assertion can be proven in a straightforward manner via in-
duction. Moreover, we know from [6, Sequence A008292] that
e(n, j) =
j

=0
(1)

(j )
n
_
n + 1

_
.
This can be used to write the rational version of f
n
(x) for any n 1 in a
timely way. So, for example, we see that
f
8
(x) =
x
8
+ 247x
7
+ 4293x
6
+ 15619x
5
+ 15619x
4
+ 4293x
3
+ 247x
2
+ x
(1 x)
9
,
which implies

k1
k
8
5
k
= f
8
_
1
5
_
=
1139685
2048
.
We have thus seen two dierent ways to compute the exact value of

k1
k
n
m
k
with |m| > 1 and n N {0}, one with a recurrence and one
with power series. I encourage us all to share at least one of these techniques
with our students the next time we are exploring innite series.
References
[1] G. Andrews, The Geometric Series in Calculus, American Mathematical
Monthly 105, no. 1 (1998), 36-40.
[2] J. Bernoulli, Tractatus de seriebus innitis, 1689.
[3] W. Dunham, Euler: The Master of Us All, The Dolciani Mathematical
Expositions, no. 22, Mathematical Association of America, Washington,
D.C., 1999.
8
[4] C. Edwards and D. Penney, Calculus with Analytic Geometry, Fifth Edi-
tion, Prentice Hall, 1998.
[5] R. Larson, R. Hostetler, and B. Edwards, Calculus: Early Transcendental
Functions, Second Edition, Houghton Miin Company, 1999.
[6] N. J. A. Sloane, The On-Line Encyclopedia of Integer Sequences, pub-
lished electronically at http://www.research.att.com/njas/sequences/.
Keywords: innite series, convergence, divergence, Euler, Bernoulli, ratio test,
recurrence, binomial theorem, Eulerian numbers
Biographical Note: James A. Sellers is currently the Director of Undergraduate
Mathematics at the Pennsylvania State University. Before accepting this position
he served for nine years as a mathematics professor at Cedarville University in
Ohio. As a mathematics professor, James loves to teach mathematics to under-
graduates and perform research with them.
Prior to going to Cedarville he received his Ph.D. in mathematics in 1992 from
Penn State, where he met his wife Mary. James truly enjoys spending time with
Mary and their ve children. He agrees with Euler that mathematics can often be
enjoyed and discovered with a child in his arms or playing round his feet.
9