The University of Maine

DigitalCommons@UMaine

Electronic Theses and Dissertations Fogler Library

Spring 5-8-2020

Calculating Infinite Series Using Parseval's Identity

James R. Poulin
University of Maine, [email protected]

Follow this and additional works at: https://digitalcommons.library.umaine.edu/etd

Part of the Analysis Commons, and the Number Theory Commons

Recommended Citation
Poulin, James R., "Calculating Infinite Series Using Parseval's Identity" (2020). Electronic Theses and
Dissertations. 3196.
https://digitalcommons.library.umaine.edu/etd/3196

This Open-Access Thesis is brought to you for free and open access by DigitalCommons@UMaine. It has been
accepted for inclusion in Electronic Theses and Dissertations by an authorized administrator of
DigitalCommons@UMaine. For more information, please contact [email protected].
 CALCULATING INFINITE SERIES USING PARSEVAL’S IDENTITY

By

James Russell Poulin

B.A. University of Maine, 2018

A THESIS

Submitted in Partial Fulfillment of the

Requirements for the Degree of

Master of Arts

(in Mathematics)

The Graduate School

The University of Maine

May 2020

Advisory Committee:

Jack Buttcane, Assistant Professor of Mathematics, Co-advisor

Julian Rosen, Assistant Professor of Mathematics, Co-advisor

Andrew Knightly, Professor of Mathematics

CALCULATING INFINITE SERIES USING PARSEVAL’S IDENTITY

By James Russell Poulin

Thesis Co-advisors: Dr. Jack Buttcane & Dr. Julian Rosen

An Abstract of the Thesis Presented

in Partial Fulfillment of the Requirements for the
Degree of Master of Arts
(in Mathematics)
May 2020

Abstract: Parseval’s identity is an equality from Fourier analysis that relates

an infinite series over the integers to an integral over an interval, which can be
used to evaluate the exact value of some classes of infinite series. We compute
the exact value of the Riemann zeta function at the positive even integers
using the identity, and then we use it to compute the exact value of an infinite
series whose summand is a rational function summable over the integers.
 ii
ACKNOWLEDGMENTS

I would like to thank Julian Rosen for assisting me in researching and writing this paper, as
well as Jack Buttcane and Andrew Knightly. Julian’s Raspberry Pi to run Sage code was
extremely helpful in the research.
 iii

CONTENTS

ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
1. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2. PARSEVAL’S IDENTITY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.1. The Space of Functions L2 ([0, 1]) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2. Orthonormal Basis for L2 ([0, 1]). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3. FINDING THE POSITIVE EVEN ZETA VALUES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
4. NON-NEGATIVE RATIONAL FUNCTIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4.1. Infinite Series for Non-negative Rational Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2
4.2. Example: Sum of n2 +1
Over Z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
289
4.3. Example: Sum of (n2 +4)(n−1/2)2
Over Z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1
4.4. Example: Sum of (n2 +1)(n2 +4)2
Over Z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.5. More Interesting Identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5. GENERAL RATIONAL FUNCTIONS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.1. Infinite Series for General Rational Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.2. Application: Dirichlet L-Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
6. POSITIVE EVEN ZETA VALUES REVISITED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
6.1. The Alternate Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
6.2. How We Found f (x) for ζ(2). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
6.3. Example: ζ(4) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
6.4. Example: ζ(10) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
7. EXTENSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
7.1. Exact Sums Over N. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
7.2. Exact Sums Using Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
BIOGRAPHY OF THE AUTHOR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
 1

1. INTRODUCTION
P∞ 1 π2
It is a well-known fact that the sum of the infinite series n=1 n2 is equal to 6
. The Basel
problem, which is to find the value of this series, was initially solved by Euler in 1735, but
his proof contained a logical gap. However, Euler’s result was later rigorously proven by
Weierstrass, as well as others [3], [10]. Of the numerous ways to solve the Basel problem, one
is to use Parseval’s identity, which states that for a particularly “nice” function f , it holds
that
X 2
Z 1
fˆ(n) = |f (x)|2 dx,
n∈Z 0
R1
where fˆ(n) = 0 f (x)e2πinx dx for every integer n is its Fourier transform. Consider f (x) =
−2πi x − 12 , whose Fourier transform is fˆ(n) = 1/n for n 6= 0 and fˆ(0) = 0. Observe that

1/n2 is an even function of n, so

∞
X 1 1X 1 1X ˆ 2
2
= 2
= f (n) .
n=1
n 2 n6=0 n 2 n6=0

By Parseval’s identity,
2 X 2
Z 1
fˆ(0) + fˆ(n) = |f (x)|2 dx,
n6=0 0

so
∞
1 1 1
π2
Z Z
X 1 2 1 2 1
dx − fˆ(0) 4π 2 x2 − 4π 2 x + π 2 dx − 0 = .


2
= f (x) =
n=1
n 2 0 2 2 0 6

This thesis explores the extent to which Parseval’s identity can be used to evaluate other
infinite series. We will show that Parseval’s identity can be used to compute the exact value
of ∞ 1
P P
n=1 n2k for any k ∈ N, as well as series of form n∈Z g(n), where g ∈ R(x) is summable

over the integers. For instance, we can use Parseval’s identity to find the exact value of series
like n∈Z n21+1 , n∈Z (0.5−n)
1 1
P P P
2 , and n∈Z (3n+1)3 .
 2

2. PARSEVAL’S IDENTITY

There is plenty of theory behind how the sum of a series over Z can be evaluated using
R1
an integral over [0, 1]. After all, such a claim that n∈Z |fˆ(n)|2 = 0 |f (x)|2 dx for certain
P

functions f is not obvious. We shall prove this identity, but we start by discussing which
functions are considered “nice” for the identity to work. These functions arise from the space
we call L2 ([0, 1]).

2.1. The Space of Functions L2 ([0, 1]).

Definition 2.1. We define L2 ([0, 1]) to be the space of all C-valued measurable functions f
R1
on [0, 1] such that 0 |f (x)|2 dx is finite. We identify two functions in this space if they agree
almost everywhere.

The space L2 ([0, 1]) has an inner product defined to be

Z 1
hf, gi = f (x)g(x) dx
0

for f, g ∈ L2 ([0, 1]). Moreover, L2 ([0, 1]) is a normed space whose norm is given by
Z 1
2
kf k = hf, f i = |f (x)|2 dx. [9], p. 139
0

With this inner product and norm, L2 ([0, 1]) becomes a complete inner product space, also
known as a Hilbert space, over C ([6], p. 321, Theorem 5.59). Despite the elements of L2 ([0, 1])
being equivalence classes, we will still refer to a function f as an element of this space because
we are not concerned with two functions that differ by a set of Lebesgue measure zero. Since
kf k2 = kgk2 if any two functions f, g ∈ L2 ([0, 1]) differ by a set of measure zero (that is, they
agree almost everywhere), we will focus only on functions that are continuous on [0, 1].
 3

2.2. Orthonormal Basis for L2 ([0, 1]).

Definition 2.2. A nonempty subset S of a Hilbert space H is orthonormal if the inner

product of any two elements x, y ∈ S is given by


0, x 6= y

hx, yi =

1, x = y,


and such a set S is an orthonormal basis if its span is dense in H.

For our purposes of establishing the theory, we will show that the set {e2πinx | n ∈ Z} is an
orthonormal basis of the Hilbert space L2 ([0, 1]). Let us call this set B. We show that B is
orthonormal using the Fundamental Theorem of Calculus, but we use the Stone-Weierstrass
Theorem to show that it is an orthonormal basis.

Theorem 2.3 (Stone-Weierstrass, [4], p. 823). Suppose the set E is a vector-subspace of the
space of continuous functions C(S, C), where S is a compact, Hausdorff space. If:

(i) for each point x ∈ S there exists an element f ∈ E such that f (x) 6= 0,
(ii) for every pair of distinct points x, y ∈ S there exists an element f ∈ E such that
f (x) 6= f (y), and
(iii) E is closed under multiplication and complex conjugation,

then E is dense in C(S, C) under the max norm.

Proposition 2.4. The set B is an orthonormal basis for L2 ([0, 1]); that is,


1, if m = n

2πimx 2πinx
he ,e i =

0, if m 6= n,


and the span of B is dense in L2 ([0, 1]).

 4

Proof. We first show that the set B is orthonormal. If m 6= n, then

Z 1
2πimx 2πinx
he ,e i= e2πimx e2πinx dx
0
Z 1
= e2πi(m−n)x dx
0
1
e2πi(m−n)x
=
2πi(m − n) 0
1 1
= − = 0.
2πi(m − n) 2πi(m − n)
R1 R1
If m = n, then the integral evaluates to 0
e2πi(m−n)x dx = 0
1 dx = 1. This shows that
he2πimx , e2πinx i = 0 for each m 6= n, and e2πinx = 1.
We next show that span(B) is dense in L2 ([0, 1]). Since the space of continuous complex-
valued functions C([0, 1]/{0, 1}, C), where the points 0 and 1 are identified, is dense in
L2 ([0, 1]) ([9], p. 153, Theorem 12), it suffices to show span(B) is dense in C([0, 1]/{0, 1}, C)
by showing that it satisfies the hypotheses of Theorem 2.3. Observe that span(B) is a
vector-subspace of C([0, 1]/{0, 1}, C), as well as the fact [0, 1]/{0, 1} is a compact, Hausdorff
space. Since the constant function e2πi(0)x = 1 is in span(B), criterion 2.3(i) is satisfied. For
2.3(ii), consider the exponential function e2πix ∈ span(B). Let x1 , x2 ∈ [0, 1]/{0, 1} such
that x1 6= x2 . Then e2πix1 6= e2πix2 by comparing real and imaginary parts, satisfying 2.3(ii).
Lastly, expanding the product of two finite linear combinations of exponential functions
from B shows that the product is in span(B), and span(B) is clearly closed under complex
conjugation; these satisfy 2.3(iii). With all criteria of Theorem 2.3 satisfied, span(B) is dense
in C([0, 1]/{0, 1}, C) with respect to the max norm. Since the L2 -norm is no larger than
the max norm ([9], p. 142, Corollary 3), span(B) is dense in C([0, 1]/{0, 1}, C) with respect
to the L2 -norm and is therefore dense in L2 ([0, 1]) because C([0, 1]/{0, 1}, C) is dense in
L2 ([0, 1]). We conclude that B is an orthonormal basis of L2 ([0, 1]). 
 5

We lastly define the Fourier transform of a function f ∈ L2 ([0, 1]) to be

fˆ(n) = hf (x), e2πinx i

Z 1
= f (x)e−2πinx dx
0

for every n ∈ Z. Now all the pieces are in place to prove Parseval’s identity.

Theorem 2.5 (Parseval’s Identity). If f ∈ L2 ([0, 1]), then

X 2
fˆ(n) = kf k2 .
n∈Z

fˆ(n)e2πinx be the orthogonal projection

PN
Proof. Let f ∈ L2 ([0, 1]), and let (PN f )(x) = n=−N

of f (x) onto span{e2πi(−N ) , ..., e2πiN }. Since span(B) is dense in L2 ([0, 1]) by Proposition 2.4,
for every  > 0, there exists g ∈ span(B) such that kg − f k < . With g ∈ span(B), there
exists some N such that g ∈ span{e2πi(−N ) , · · · , e2πiN }. It follows that

Pm (f ) − f ≤ kg − f k < 

for all m ≥ N . This shows that

N
X
f (x) = lim fˆ(n)e2πinx .
N →∞
n=−N

Then by the Pythagorean Theorem ([6], p. 576),

N
2 2 2
X
2
kf k = PN (f ) + f − PN (f ) = |fˆ(n)|2 + f − PN (f ) .
n=−N

Letting N approach infinity, f − PN (f ) approaches 0 by what we have just shown, so

N
X
2
kf k = lim |fˆ(n)|2 .
N →∞
n=−N

Since the series converges absolutely, it therefore follows that kf k2 = |fˆ(n)|2 .

P
n∈Z 
 6

3. FINDING THE POSITIVE EVEN ZETA VALUES

Back in the introduction where we used Parseval’s identity to calculate ζ(2) = ∞ 1

P
n=1 n2 ,

where ζ(s) = ∞ 1
P
n=1 ns for any s ∈ C is the Riemann zeta function, observe that our choice of

function f (x) = −2πi x − 12 contains the expression x − 21 . This expression is also known

as the first Bernoulli polynomial, denoted B1 (x). Given how the Fourier transform of f is
conveniently 1/n for all nonzero n, and f is a constant multiple of B1 (x), we look to use
any given Bernoulli polynomial Bk (x) to construct a polynomial whose Fourier transform
is 1/nk for all nonzero n. We will then use this polynomial to evaluate ζ(2k) = ∞ 1
P
n=1 n2k

(similar approaches were taken in [2] and [5]). We begin with a definition of the Bernoulli
polynomials.

Definition 3.1 (cf. [8], p. 19). We define the Bernoulli polynomial Bk (x) for any integer
k ≥ 0 as the polynomial with rational coefficients resulting from the following recurrence
relation:

(i) B0 (x) = 1,
(ii) Bk0 (x) = kBk−1 (x), for all k ∈ N,
R1
(iii) 0 Bk (x) dx = 0, for all k ∈ N.

We further define Bk := Bk (0) to be the k th Bernoulli number.

From criteria 3.1(ii) and 3.1(iii), we immediately find that Bk (1) = Bk (0) for all integers
k ≥ 2:
Z 1
0= Bk−1 (x) dx
0
Z 1
1 0
= Bk (x) dx
0 k
1 1
= Bk (1) − Bk (0).
k k
1
By criterion 3.1(i), we further have B0 (1) = B0 (0) = 1. However, B1 (1) = 2
= −B1 (0), so
Bk (1) = Bk (0) for all non-negative integers k 6= 1. Observe that criterion 3.1(iii) fails for
R1
B0 (x), since 0 B0 (x) dx = 1 is nonzero.
 7

1 π2
Suppose we attempt to use B1 (x) = x − 2
to prove ζ(2) = 6
. Observe that the Fourier
−1 −1
transform of B1 (x) is 2πin
, so we would have to account for the constant 2πi
from the Fourier
transform, adding a rather inconvenient extra step to our calculation of the series. We would
like to adjust any given Bernoulli polynomial Bk (n) for any k ∈ N so that its Fourier transform
becomes 1/nk for all nonzero n and thus removing any adjustments to our calculation.

(2πi)k
Theorem 3.2. Define fk∗ (x) := −k!
Bk (x) for every k ∈ N and 0 ≤ x ≤ 1. Then fˆk∗ (0) = 0,
and fˆk∗ (n) = 1/nk for every n 6= 0.

Proof. The equality fˆk∗ (0) = 0 follows from Definition 3.1(iii). We prove fˆk∗ (n) = 1/nk for
−k!
every n 6= 0 by showing the Fourier transform of Bk (x) is (2πi)k
· n1k using induction on k. For
k = 1, the Fourier transform of B1 (x) is
1 1  
−1 −(1)! 1
Z Z
−2πinx 1 −2πinx
B1 (x)e dx = x− e dx = = · .
0 0 2 2πin (2πi)1 n
R1 −(k−1)!
Now assume that 0
Bk−1 (x)e−2πinx dx = (2πi)k−1
· 1
nk−1
for any k ≥ 2. Then

1 1 Z 1
−Bk (x) −2πinx
Z
−2πinx k
Bk (x)e dx = e + Bk−1 (x)e−2πinx dx
0 2πin 0 2πin 0
   
−1 −1 k! 1
= Bk (1) − Bk (0) − k
· k
2πin 2πin (2πi) n
−k! 1
= · .
(2πi)k nk
−k! 1
This shows (2πi)k
· nk
is the Fourier transform of Bk (x) for every n 6= 0, and therefore the
(2πi)k
Fourier transform of fk∗ (x) = −k!
Bk (x) is 1/nk for every such n. 

Consider, as another example, finding the value of ζ(6) = ∞ 1 1

P
n=1 n6 . Since n6 is an even
1 2
function of n, ∞ 1 1 1 1
P P
n=1 n6 = 2 n6=0 n6 . The summand n6 = n3 , so we require the function

f3∗ since its Fourier transform is 1/n3 for every nonzero n. The third Bernoulli polynomial
is B3 (x) = x3 − 32 x2 + 12 x, so f3∗ (x) = 43 π 3 ix3 − 2π 3 ix2 + 23 π 3 ix. Furthermore, fˆ3∗ (0) = 0 by
Theorem 3.2. We now have by Parseval’s identity,
1 1 2
π6
Z Z 
1X 1 1 2 1 4 3 3 2
ζ(6) = 6
= f3∗ (x) dx = π x − 2π 3 x2 + π 3 x dx = .
2 n6=0 n 2 0 2 0 3 3 945
 8

Suppose we wish to find the value of ζ(2k) for every k ∈ N. Then we can use the following
corollary to Theorem 3.2.

Corollary 3.3. For all k ∈ N, the following equality holds:

1 ∗ 2
ζ(2k) = f (x) .
2 k

Proof. Let k ∈ N. Then

∞
X 1
ζ(2k) =
n=1
n2k
2
1X 1
=
2 n6=0 nk

1 1 ∗
Z
2
(1) = f (x) dx,
2 0 k

where line (1) is by Parseval’s identity and Theorem 3.2. 

This shows Parseval’s identity is capable of finding the exact value of ζ(s) for every positive
(2πi)k
even integer s = 2k. Furthermore, the relationship fk∗ (x) = −k!
Bk (x) combined with
Corollary 3.3 both allow us to prove the formula for the positive even zeta values in terms of
the Bernoulli numbers.

Corollary 3.4. For all k ∈ N, one has

(−1)k−1 (2π)2k B2k

ζ(2k) = .
2 · (2k)!

Proof. Let k ∈ N, and let fk∗ (x) be as in Theorem 3.2. Then

Z 1
1 2
ζ(2k) = fk∗ (x) dx
2 0
2
1
(2πi)k
Z
1
= · Bk (x) dx
2 0 −(k)!
Z 1
(2π)2k
= Bk (x)2 dx
2 · (k!)2 0
 9
!
(2π)2k (−1)k−1 (k!)2 B2k
(2) =
2 · (k!)2 (2k)!
(−1)k−1 (2π)2k B2k
= . 
2 · (2k)!
R1 (−1)k−1 (k!)2 B2k
The identity 0
Bk (x)2 dx = (2k)!
on line (2) is proved in [1], p. 11, Proposition 1.
As a side note, one may construct these fk∗ (x) directly by the use of a recurrence relation
modeled after that of the Bernoulli polynomials. That way, knowledge of the Bernoulli
polynomials is not required.

Proposition 3.5. We may construct the functions fk∗ for any k ∈ N via the following
recurrence relation:

(i) f1∗ (x) = −2πi x − 12 ,

R 
∗ x ∗ R1 ∗
(ii) fk+1 (x) = 2πi 0 fk (t) dt + 0 xfk (x) dx .

Proof. Criterion 3.5(i) follows from Theorem 3.2 for k = 1. For 3.5(ii), observe that Definition
∗0
Rx
3.1(ii) shows that fk+1 (x) = 2πifk∗ (x) for any k ∈ N, and hence fk+1
∗
(x) = 2πi 0 fk∗ (t) dt + C
R1
for some constant C. Since 0 fk∗ (x) dx = 0, we may find the value of C:
Z 1
∗
0= fk+1 (x) dx
0
Z 1  Z x 
= 2πi fk∗ (t) dt +C dx
0 0
Z 1 Z x 
= 2πi fk∗ (t) dt dx + C
0 0
 Z x  1 Z 1
= 2πix fk∗ (t) dt −2πi xfk∗ (x) dx + C,
0 0
| {z }0
=0

R1
so C = 2πi 0
xfk∗ (x) dx. This constructs criterion 3.5(ii). 
 10

4. NON-NEGATIVE RATIONAL FUNCTIONS

Finding the exact value of the positive even zeta values utilizes ideally adjusted Bernoulli
polynomials to apply Parseval’s identity to. For this section, we transition from series over N
to series over Z as we find another class of series we are able to calculate using Parseval’s
identity. To do so, we first compute the Fourier transform of a general function different from
fk∗ .

4.1. Infinite Series for Non-negative Rational Functions. We consider a function

defined by the product of a monomial and an exponential: fA,r (x) = xA erx , where A ≥ 0 is
an integer and r ∈ C. We first find the Fourier transform fˆA,r (n), and then we rewrite the
summand of a series over Z as the squared modulus of a linear combination of the transformed
functions. That way, the sum of the series is equal to the integral of the squared modulus of
a linear combination of functions xm erx over [0, 1]. We will use M to denote the span of all
r
functions of form xm erx , where m = 0, 1, ..., A and r ∈ C such that 2πi
is not an integer.

r
Lemma 4.1. Let fA,r (x) = xA erx where A ∈ N ∪ {0} and r ∈ C such that 2πi
is not an
integer. Then
A
(−1)A+1 A! X (−1)m A!
fˆA,r (n) = + er
(r − 2πin)A+1 m=0
(A − m)!(r − 2πin)m+1

for all n ∈ Z.

Proof. We prove Lemma 4.1 by induction. For A = 0, we have

Z 1
fˆ0,r (n) = f0,r (x)e−2πinx dx
0
Z 1
= erx−2πinx dx
0
1  
= er−2πin − 1
r − 2πin
0
(−1)0+1 0! r−2πin
X (−1)m 1!
= + e .
(r − 2πin)0+1 m=0
(0 − m)!(r − 2πin)m+1
 11

For the inductive step, assume

1 A−1
(−1)A (A − 1)! (−1)m (A − 1)!
Z X
A−1 (r−2πin)x r−2πin
x e dx = + e .
0 (r − 2πin)A m=0
(A − m − 1)!(r − 2πin)m+1

Then
Z 1
fˆA,r (n) = xA e(r−2πin)x dx
0
1 1
xA e(r−2πin)x
Z
A
= − xA−1 e(r−2πin)x dx
r − 2πin 0 r − 2πin 0
A−1
er−2πin (−1)A+1 A! r−2πin
X (−1)m+1 A!
= + + e
r − 2πin (r − 2πin)A+1 m=0
(A − m − 1)!(r − 2πin)m+2
A
(−1)A+1 A! er−2πin r−2πin
X (−1)m A!
= + + e
(r − 2πin)A+1 r − 2πin m=1
(A − m)!(r − 2πin)m+1
A
(−1)A+1 A! r−2πin
X (−1)m A!
= + e . 
(r − 2πin)A+1 m=0
(A − m)!(r − 2πin) m+1

The Fourier transform of fA,r is a finite series whose sum includes terms that are rational
functions of n. The structure of these rational functions of n is a fraction with a coefficient
(depending on A and r) in the numerator, and the expression (r−2πin)m+1 in the denominator.
When m = A, we reach the largest degree denominator, so we next look to find a function in
r 1
M that depends on A, t, where t = 2πi
is not an integer, whose Fourier transform is (n−t)A+1
.

Lemma 4.2. For all t ∈ C − Z and A ∈ N ∪ {0}, there exists a function FA,t ∈ M , where
1
r = 2πit, such that F̂A,t (n) = (n−t)A+1
.

1 (−2πi)A+1
Proof. Let us multiply (n−t)A+1
by the quantity (−2πi)A+1
to obtain the equivalent expression
1
(−2πi)A+1 · (2πit−2πin)A+1
. Assign r = 2πit, and we have an expression that resembles a
summand of the finite series in Lemma 4.1. Our goal is to find a function of form
A
X
FA,t (x) = (−2πi)A+1 am xm e2πitx
m=0
A
X
= (−2πi)A+1 am xm erx
m=0
 12

1
such that F̂A,t (n) = (n−t)A+1
. We make use of the fact that the finite series in said lemma
includes the term 1/(r − 2πin)A+1 along with the other 1/(r − 2πin)m terms from m = 1 to
m = A. The following (A + 1) × (A + 2) augmented coefficient matrix represents the linear
system
A
X 1
F̂A,t (n) = (−2πi)A+1 am fˆm,r (n) = ,
m=0
(n − t)A+1

where the fˆm,r (n) are as in Lemma 4.1:

 
r er er er er er
 e −1 ··· 0 
 
0 er − 1 2er ··· (A − 2)er (A − 1)er Aer 0 
 

 
 

 0 0 2(er − 1) ··· (A − 2)(A − 3)er (A − 1)(A − 2)er A(A − 1)er 0 
.. .. ... .. ... ... ... .. 
 
.


 . . . . 

 
A! r
0 0 0 ··· (A − 2)!(er − 1) (A − 1)!er e 0 
 
2

 
 

 0 0 0 ··· 0 (A − 1)!(er − 1) A!er 0 
 
0 0 0 ··· 0 0 (−1)A A!(er − 1) 1

Row-reducing this upper-triangular matrix gives us the column vector of coefficients [am ]A
m=0

necessary to construct FA,t . 

1
In general, a rational function h(n) with complex coefficients is not of form (n−t)A+1
.
Although, by using partial fraction decomposition, h can be written as a linear combination
1
of these (n−t)A+1
.

Lemma 4.3. For all h ∈ C(x) in which limn→∞ h(n) = limn→−∞ h(n) = 0, there exists a
function f ∈ M such that fˆ(n) = h(n).

Proof. We decompose such h(n) into its partial fractions

L
X Cj
h(n) = ,
j=1
(n − tj )Aj +1

where Cj , tj ∈ C, Aj ∈ N ∪ {0}, and L is the number of partial fractions in the decomposition.

1
By Lemma 4.2, there exists some function FA,t ∈ M whose Fourier transform is (n−tj )Aj +1
.
 13

We now use linearity to construct the function

L
X
f (x) = Cj FAj ,tj (x)
j=1

whose Fourier transform is

L
X
fˆ(n) = Cj F̂Aj ,tj (n) = h(n). 
j=1

In order to use Parseval’s identity, we require functions of form |h(n)|2 , which is certainly
non-negative. We therefore show that any non-negative rational function of real coefficients
has this form for some h as in Lemma 4.3.

Lemma 4.4. For all non-negative g ∈ R(x), there exists h ∈ C(x) such that g(n) = |h(n)|2 .

Proof. Since g is non-negative, g does not have real zeros or real poles of odd order. Thus,
by the Fundamental Theorem of Algebra, g(n) factors as

α(n − r1 )(n − r1 )(n − r2 )(n − r2 ) · · · (n − rp )(n − rp ) α|n − r1 |2 |n − r2 |2 · · · |n − rp |2

=
(n − s1 )(n − s1 )(n − s2 )(n − s2 ) · · · (n − sq )(n − sq ) |n − s1 |2 |n − s2 |2 · · · |n − sq |2
√ 2
α(n − r1 )(n − r2 ) · · · (n − rp )
=
(n − s1 )(n − s2 ) · · · (n − sq )

where α ≥ 0 is a constant, and each rj , sk ∈ C with j = 1, 2, ..., p and k = 1, 2, ..., q. Observe

that the real zeros and real poles of even order are listed as rj = rj and sk = sk , respectively.
√
α(n−r1 )(n−r2 )···(n−rp )
Define h(n) = (n−s1 )(n−s2 )···(n−sq )
. 

All the pieces are in place to establish the following result.

Theorem 4.5. Given g ∈ R(x) everywhere non-negative and vanishing at infinity, there
exists a function f ∈ M such that g(n) = |fˆ(n)|2 .

Remark. If g is summable over the integers, then combining Theorem 4.5 with Parseval’s
identity shows that
X Z 1
g(n) = |f (x)|2 dx.
n∈Z 0
 14

Hence, this theorem gives us a way to find the exact sum of the series using an integral over
[0, 1].

Proof. Since g is non-negative, there exists a function h ∈ C(x) such that g(n) = |h(n)|2
by Lemma 4.4. By Lemma 4.3, there exists a function f ∈ M such that fˆ(n) = h(n), so
g(n) = |fˆ(n)|2 . 

P
Suppose we would like to determine the sum of g(n) for a given non-negative rational
n∈Z

g summable over Z. By Theorem 4.5, we can find a function f (x) = Ln=1 Cn xAn ern x such
P

that g(n) = |fˆ(n)|2 . Then the value of the infinite series is given by
Z 1 Z 1
2
|f (x)| dx = f (x)f (x) dx
0 0
L X
L
!
X Z 1
= Cn Cm xAn +Am ex(rn +rm ) dx
n=1 m=1 0

L X
X L
(3) = Cn Cm · fˆAn +Am ,rn +rm (0).
n=1 m=1

The value fˆAn +Am ,rn +rm (0) is computed in Lemma 4.1.
If we are given such a function g, then the latter three lemmas and Theorem 4.5 provide
us with a method to find f such that g(n) = |fˆ(n)|2 :

1) Express g(n) as |h(n)|2 with h ∈ C(x) by factoring g.

2) Decompose h(n) into its partial fractions.
3) Construct FAj ,tj (x) as seen in Lemma 4.2 for the j th partial fraction by row-reducing the
corresponding matrix from said lemma.
4) Further construct f (x) = Lj=1 Cj FAj ,tj (x), where L is the number of partial fractions
P

and Cj are coefficients. This gives us fˆ(n) = h(n).

R1
5) By Parseval’s identity, n∈Z g(n) = 0 |f (x)|2 dx.
P

Due to the length of time it takes to find the required function of x, as well as computing
R1
the integral 0 |f (x)|2 dx, the value of these infinite series are best found using software. As
a result of the research, we have developed a code package for the mathematics software
 15

system SageMath [11]. The package, called “ParsevalSum,” inputs a rational function of n
R1
and outputs the exact sum of the series. The value of 0 |f (x)|2 dx given by the finite series
in equation (3) is the operation used to evaluate the integral. ParsevalSum also computes
the even zeta values, and it determines the function f ∈ M such that g(n) = |fˆ(n)|2 . The
software package can be accessed via GitHub:

https://github.com/JamesRPoulin/ParsevalSum.

2
4.2. Example: Sum of n2 +1
Over Z. For a simple example, consider the series

X 2
.
n∈Z
n2 +1

One may take the constant 2 outside the summation, but we incorporate it into the example
to demonstrate that the following works nicely with constants. Observe that
√
X 2 X ( 2)2
=
n∈Z
n2 + 1 n∈Z (n − i)(n + i)
√ 2
X 2
= ,
n∈Z
n+i
√
so we will look for a function f such that fˆ(n) = 2
n+i
. The denominator has form (n − t)A+1
from Lemma 4.2, where t = −i and A = 0. Hence, our r = 2πit = 2π and A = 0, so we
row-reduce the matrix
   
to obtain 1 .
e2π − 1 1 1 e2π −1

Thus, our fˆ(n) is equivalent to

√ !
2
fˆ(n) = (−2πi) fˆ0,2π (n).
e2π − 1

Therefore, by linearity, our required function f is

√ !
−2πi 2
f (x) = e2πx .
e2π − 1
 16

Finally, the exact sum of the infinite series is therefore

Z 1
X 2
2
= |f (x)|2 dx
n∈Z
n +1 0

2π(e2π + 1)
= ≈ 6.3067.
e2π − 1
289
4.3. Example: Sum of (n2 +4)(n−1/2)2
Over Z. For an example that requires partial fractions,
consider
X 289
.
n∈Z
(n2 + 4)(n − 1/2)2
Again, the constant 289 can be taken out, but the partial fraction decomposition is more
convenient with the constant. Observe that

X 289 X 172
=
n∈Z
(n2 + 4)(n − 1/2)2 n∈Z
(n + 2i)(n − 2i)(n − 1/2)2
2
X 17
= ,
n∈Z
(n + 2i)(n − 1/2)

so we will look for a function f such that fˆ(n) = 17

(n+2i)(n−1/2)
. Using partial fraction
decomposition, we get
−2 + 8i 2 − 8i
fˆ(n) = + .
n + 2i n − 1/2
The denominator of each partial fraction has form (n − t)A+1 from Lemma 4.2. The quantity
1
n+2i
shows that the corresponding A = 0 and t = −2i, hence r = 2πit = 4π. Thus, we
row-reduce the matrix
   
to obtain 1 .
e4π − 1 1 1 e4π −1

1
Similarly, the quantity n−1/2
shows that the corresponding A = 0 and t = 1/2, hence
r = 2πit = πi. Thus, we row-reduce the matrix
   
eπi − 1 1 to obtain 1 − 21 .
 17

Thus, our fˆ(n) is equivalent to

 
−2 + 8i
fˆ(n) = (−2πi) fˆ0,4π (n) + (−2πi)(−1 + 4i)fˆ0,πi (n).
e4π − 1

Therefore, by linearity, our required function f is

 
16π + 4πi
f (x) = f 0,4π (x) + (8π + 2πi)f 0,πi (x)
e4π − 1
 
16π + 4πi
= e4πx + (8π + 2πi)eπix .
e4π − 1

Finally, the exact sum of the infinite series is therefore

Z 1
X 289
= |f (x)|2 dx
n∈Z
(n + 4)(n − 1/2)2
2
0

30π(e4π + 1)
= 68π 2 − ≈ 576.8847.
e4π − 1
1
4.4. Example: Sum of (n2 +1)(n2 +4)2
Over Z. For an example that uses a larger matrix,
consider
X 1
.
n∈Z
(n2 + 1)(n2 + 4)2
Observe that

X 1 X 12
=
n∈Z
(n2 + 1)(n2 + 4)2 n∈Z
(n + i)(n − i)(n + 2i)2 (n − 2i)2
2
X 1
= ,
n∈Z
(n + i)(n + 2i)2

so we will look for a function f such that fˆ(n) = 1

(n+i)(n+2i)2
. Using partial fraction decompo-
sition, we get
1 1 i
fˆ(n) = − + + .
n + i n + 2i (n + 2i)2
The denominator of each partial fraction has form (n − t)A+1 from Lemma 4.2. The first
1
quantity n+i
shows that the corresponding A = 0 and t = −i, and hence our r = 2πit = 2π.
 18

Thus, we row-reduce the matrix

   
to obtain 1 .
e2π − 1 1 1 e2π −1

1
The second quantity n+2i
shows that the corresponding A = 0 and t = −2i, hence r = 2πit =
4π. We then row-reduce the matrix
   
to obtain 1 .
e4π − 1 1 1 e4π −1

1
The third quantity (n+2i)2
shows that the corresponding A = 1 and r = 4π. We then
row-reduce the 2 × 3 matrix
   
e4π
4π
 e −1 e4π 0   1 0 (e4π −1)2 
  to obtain  .
4π 1
0 −(e − 1) 1 0 1 − e4π −1

Thus, our fˆ(n) is equivalent to

   
1 1
fˆ(n) = (−2πi) − fˆ0,2π (n) + (−2πi) 4π fˆ0,4π (n)
e2π − 1 e −1
 ! 
4π
 
i ie
+ (−2πi)2  − 4π fˆ1,4π (n) + fˆ0,4π (n) .
e −1 (e − 1)2
4π

Therefore, by linearity, our required function f is

2πie2πx 2πie4πx 4π 2 ixe4πx 4π 2 e4π ie4πx

f (x) = 2π − 4π + 4π − 4π .
e −1 e −1 e −1 (e − 1)2

Finally, the exact sum of the infinite series is therefore

Z 1
X 1
= |f (x)|2 dx
n∈Z
(n + 1)(n2 + 4)2
2
0

5πe8π + 32πe6π − 24π 2 e4π − 32πe2π − 5π

= ≈ 0.1104.
144e8π − 288e4π + 144
 19

Note that there are multiple choices of fˆ in each of these examples, especially this one.
We could have chosen, for instance,

1
fˆ(n) =
(n + i)(n − 2i)(n + 2i)

for this example. With this choice, our corresponding function f is

πie4πx 2πie2πx πie−4πx

f (x) = − + .
2(e4π − 1) 3(e2π − 1) 6(e−4π − 1)
R1
Nonetheless, the integral 0
|f (x)|2 dx has the same value as the previous integral. This
shows that some choices of fˆ may be better than others. In this case, the second choice of fˆ
only requires row-reducing augmented 1 × 2 matrices within Theorem 4.5, and integration by
parts or computing a triple sum in equation (3) are both not required when evaluating the
integral.

4.5. More Interesting Identities. Theorem 4.5 provides us with some interesting identities.

Proposition 4.6. For all t ∈ R − Z, one has

X 1
2
= π 2 csc2 (πt).
n∈Z
(n − t)

2
Proof. Suppose t ∈ R − Z. With 1
(n−t)2
= 1
n−t
, our choice of fˆ(n) is 1
n−t
, which means our
−2πi·e2πitx
A = 0 and r = 2πit. We thus find that f (x) = e2πit −1
. Now we apply Parseval’s identity:

2
1
−2πi · e2πitx
Z
X 1
= dx
n∈Z
(n − t)2 0 e2πit − 1
Z 1 2
2π
= dx
0 |e2πit − 1|
4π 2
=
2 − 2 cos(2πt)

= π 2 csc2 (πt),

with the last equality holding by the half-angle identity of sin2 (πt). 
 20

This identity makes sense since adding 1 (or any integer) to n in the summand does not
change the value of the series. Let us next consider the same series, but for t a complex
number.

Proposition 4.7. More generally, for all t ∈ C such that Im(t) 6= 0, one has

X 1 π − πe−4πIm(t)
= .
n∈Z
|n − t|2 Im(t)(e−4πIm(t) − 2e−2πIm(t) cos(2πRe(t)) + 1)

−2πi·e2πitx
Proof. With f (x) = e2πit −1
from before, now our integral becomes

2
1
−2πi · e2πitx
Z
X 1
= dx
n∈Z
|n − t|2 0 e2πit − 1
2
1
−2πi · e2πix(Re(t)+iIm(t))
Z
= dx
0 e2πi(Re(t)+iIm(t)) − 1
1
4π 2 e−4πIm(t)x
Z
= dx
0 |e−2πIm(t) cos(2πRe(t)) − 1 + ie−2πIm(t) sin(2πRe(t))|2
1
4π 2 e−4πIm(t)x
Z
= dx
0 e−4πIm(t)−2e−2πIm(t) cos(2πRe(t))+1
π − πe−4πIm(t)
= . 
Im(t)(e−4πIm(t) − 2e−2πIm(t) cos(2πRe(t)) + 1)

Proposition 4.8. For all k ∈ R× , one has

X k2
= πk · coth(πk).
n∈Z
n2 + k 2

k2
Proof. Let k be any nonzero real number. Observe that the summand factors as (n+ik)(n−ik)
,
so
2
X k2 X k
= .
n∈Z
n2 + k 2 n∈Z
n + ik

Thus, our choice of fˆ(n) is k

n+ik
. We see that our A = 0 and t = −ik, so r = 2πk. We then
−2πik·e2πkx
find that f (x) = e2πk −1
.
 21

By Parseval’s identity,
2
1
k2 −2πik · e2πkx
X Z
= dx
n∈Z
n2 + k 2 0 e2πk − 1
Z 1
4π 2 k 2
= 2πk 2
e4πkx dx
(e − 1) 0
πk(e4πk − 1)
=
(e2πk − 1)2
πk(e2πk + 1)
=
e2πk − 1
= πk · coth(πk). 

k2
This identity, in particular, is very interesting. The sum of n2 +k2
as n ranges through all
the integers is analogous to the integral
∞ ∞
k2
Z  
x
dx = k · arctan
−∞ x2 + k 2 k −∞

= π|k|

where x ranges through all real numbers.

 22

5. GENERAL RATIONAL FUNCTIONS

Theorem 4.5 provides a way to find the exact sum of a non-negative rational function over
Z, but most rational functions are not non-negative. Of course, we can handle non-positive
rational functions by simply factoring out −1 from the summand, which results in a non-
negative summand, but how would we handle rational functions that are neither non-negative
nor non-positive? Such an expression cannot be written in the form |h(n)|2 , but we may
instead consider a difference of such expressions. Such a difference can be determined for
any rational function by using the following lemma, and thus extending Theorem 4.5 even
further.

5.1. Infinite Series for General Rational Functions.

Lemma 5.1. Any rational function with real coefficients is a difference of two non-negative
rational functions.

p(n)
Proof. Let q(n)
be a rational function of n with polynomials p and q of real coefficients.
p(n)q(n)
Multiplying both the numerator and denominator by q(n) yields (q(n))2
, for which the
denominator is a non-negative polynomial. Considering the numerator, it is a polynomial
nk
Pdeg(pq)
of form k=0 ak nk where ak ∈ R is the coefficient of the nk term. For k even, (q(n))2 is a

non-negative rational function. For k odd, observe that

nk nk+1 + nk + nk−1 nk+1 nk−1

= − −
(q(n))2 (q(n))2 (q(n))2 (q(n))2

is a linear combination of non-negative rational functions since k + 1 and k − 1 are even, and
nk+1 + nk + nk−1 = nk−1 (n2 + n + 1) ≥ 0 for all real n. Lastly, add the non-negative rational
functions together, and then add the non-positive rational functions together. Factoring out
−1 from the non-positive quantity, we are left with a difference of two non-negative rational
functions. 

We now have an immediate corollary to Theorem 4.5 that gives us a means to calculate
the exact sum of any summable rational function over the integers using Parseval’s Identity.
 23

Corollary 5.2. Suppose g ∈ R(x). There exist functions u, v ∈ M such that

2 2
g(n) = û(n) − v̂(n) .

Remark. Recall that M is the span of all functions of form xm erx , where m = 0, 1, ..., A with
r
non-negative integer A, and r ∈ C such that 2πi is not an integer. If g is summable over the
P
integers, then the exact value of n∈Z g(n) is that of the difference of integrals
Z 1 Z 1
2
|u(x)| dx − |v(x)|2 dx.
0 0

Proof. By Lemma 5.1, g(n) is a difference of two non-negative rational functions of n, each
of which we know has form |fˆ(n)|2 for some function f ∈ M by Theorem 4.5. Denote this
difference as g(n) = |û(n)|2 − |v̂(n)|2 for functions u, v ∈ M . 

Remark. It should be noted that a more standard method for finding the sum of an infinite
series for a general rational function over Z involves contour integration. By taking a contour
integral of the function g(z) cot(πz), where z is a complex variable and g is a rational function
summable over Z, we can use the residue theorem to calculate the contour integral and write
P
the resulting value in terms of n∈Z g(n). Our method uses Parseval’s identity instead of
contour integration, and thus provides flexibility to how we may find the exact value of
P
n∈Z g(n) for such g.

5.2. Application: Dirichlet L-Functions. A useful application to Corollary 5.2 is finding

certain special values of Dirichlet L-functions.

Definition 5.3. A Dirichlet character χ of modulus k, where k ∈ N, is a group homomorphism

χ : (Z/kZ)× → C× . By defining


χ(n mod k), gcd(n, k) = 1

χ(n) :=

0,
 gcd(n, k) > 1,

the domain of χ is extended to Z. Furthermore, we define the Dirichlet L-function as the

 24

infinite series
∞
X χ(n)
L(s, χ) =
n=1
ns
for any s ∈ C. Lastly, we say that χ is odd if χ(−1) = −1, and we say it is even if χ(−1) = 1.

Remark. Since χ is a group homomorphism, we automatically find that χ(1) = 1.

Consider the sum S(A; q, k) defined to be

X 1 X 1
S(A; q, k) := A
= A
,
n∈Z
n n∈Z
(kn + q)
n≡q mod k

where q, k ≥ 1 and A ≥ 2 are integers. Observe that Theorem 4.5 allows us to compute this
sum for even A, and Corollary 5.2 allows us to compute this sum for odd A. If A has the
same parity as χ, meaning that either both A and χ are even or they are both odd, then we
have the following proposition to relate S with L:

Proposition 5.4. Let k ≥ 3 be an integer, χ a Dirichlet character of modulus k, and A ≥ 2

an integer with the same parity as χ. Then

bX
k/2c

L(A, χ) = χ(q)S(A; q, k).

q=1

Proof. With such k, 2, and χ, the right-hand side of the equation is

 
bX
k/2c b k/2c ∞ −1
X 1 X  X χ(q) X χ(q) 
χ(q) A
= 
A
+ 
A 
q=1 n∈Z
n q=1

n=1
n n=−∞
n
n≡q mod k n≡q mod k n≡q mod k
 
bX
k/2c ∞ ∞
 X χ(n) X χ(−n) 
=  + 
q=1

n=1
nA n=1
(−n)A 
n≡q mod k n≡−q mod k
 
bX
k/2c ∞ ∞
 X χ(n) X χ(−1) χ(n) 
(4) =  + · 
q=1

n=1
nA n=1
(−1)A nA 
n≡q mod k n≡−q mod k
 25
 
bk/2c ∞ ∞
X  X χ(n) X χ(n) 
(5) =  + 

q=1 n=1
nA n=1
nA 
n≡q mod k n≡−q mod k

∞
X χ(n)
=
n=1
nA

= L(A, χ).

χ(−1)
Line (4) uses the fact that A and χ have the same parity, thus (−1)A
= 1. Line (5) requires
that gcd(k, k/2) > 1 to avoid repeated terms q = k/2 whenever k is even, which is true for
k ≥ 3. 

Proposition 5.4 fails for k = 2, but since



1, gcd(n, 2) = 1

χ(n) =

0, gcd(n, 2) > 1


is the only Dirichlet character of modulus 2, and L(s, χ) = (1 − 2−s )ζ(s), nothing is new for
k = 2. We may show that L(s, χ) = (1 − 2−s )ζ(s) since
∞
X χ(n)
L(s, χ) =
n=1
ns
∞
X 1
(6) =
n=0
(2n + 1)s
1 1 1 1 1
= s
+ s + s + s + s + ···
1 3 5 7 9
   
1 1 1 1 1 1 1 1 1 1 1
(7) = + + + + + ··· − s + + + + + ···
1s 2s 3s 4s 5s 2 1s 2s 3s 4s 5s

= ζ(s) − 2−s ζ(s)

= (1 − 2−s )ζ(s).

Line (6) uses the fact that χ of modulus 2 sends positive even integers to zero and sends
positive odd integers to one. Line (7) adds and subsequently subtracts the reciprocals of
positive even integers raised to the power s.
 26

An example of using Parseval’s identity to calculate a Dirichlet L-Function is finding the

value of L(3, χ), where χ is an odd Dirichlet character of modulus 3. Observe that


1, n ≡ 1 mod 3






χ(n) = −1, n ≡ 2 mod 3




n ≡ 0 mod 3

0,


is the only odd Dirichlet character of modulus 3. Then by Proposition 5.4, we have that
∞
X χ(n) X 1
L(3, χ) = = .
n=1
n3 n∈Z
(3n + 1)3

1 4
√
3π 3 .
P
Using Corollary 5.2 on n∈Z (3n+1)3 , we find that L(3, χ) = 243
 27

6. POSITIVE EVEN ZETA VALUES REVISITED

A similar method to finding the required function f given a summable rational function of
n applies to finding such f given a rational function 1/n2A for any A ∈ N. This method does
not depend on knowing f1∗ from Proposition 3.5 nor the Bernoulli polynomials beforehand
and instead is solely dependent on the summand. Thus, we show an alternative method to
Theorem 3.2 by first finding the Fourier transform of just the monomial fA (x) = xA for any
A ∈ N and then writing 1/n2A as a linear combination of these fˆA (n). That way, we find a
polynomial function f such that fˆ(n) = 1/n2A for all n 6= 0. Extra adjustments are necessary
to account for the fact that 1/n2A is undefined at n = 0, and thus the rational function is
not summable over the integers.

6.1. The Alternate Method.

Lemma 6.1. Let fA (x) = xA for any A ∈ N. Then fˆA (0) = 1

A+1
, and for all n 6= 0, one has

A
X −A!
fˆA (n) = m nm
.
m=1
(A − m + 1)!(2πi)

Proof. If n = 0, then
Z 1 Z 1
−2πi(0)x 1
fA (x)e dx = xA dx = .
0 0 A+1

Suppose n 6= 0 and proceed by induction. For A = 1, we have

1
−(1)!
Z
1
fˆ1 (n) = xe−2πinx dx = − = .
0 2πin (1 − 1 + 1)!(2πi)1 n1
−(A−1)!
For the inductive step, assume fˆA−1 (n) = A−1
P
m=1 (A−m)!(2πi)m nm
. Then
Z 1
fˆA (n) = xA e−2πinx dx
0
1 1
xA e−2πinx
Z
A
=− + xA−1 e−2πinx dx
2πin 0 2πin 0
A−1
1 A X −(A − 1)!
=− +
2πin 2πin m=1 (A − m)!(2πi)m nm
 28

A
1 X −A!
=− +
2πin m=2 (A − m + 1)!(2πi)m nm
A
X −A!
= m nm
. 
m=1
(A − m + 1)!(2πi)

We proceed to the next lemma similarly to how we proceeded from Lemma 4.1 to Lemma
4.2.

Lemma 6.2. For all A ∈ N, there exists a polynomial function f such that fˆ(n) = 1/nA for
all n 6= 0.

am xm such that fˆ(n) = 1/nA for

PA
Proof. We look to find a function of form f (x) = m=1

all n 6= 0. We make use of the fact that the finite series in Lemma 6.1 includes the term
1/nA along with the other 1/nm terms from m = 1 to m = A − 1. The following A × (A + 1)
augmented coefficient matrix represents the linear system
A
X
fˆ(n) = am fˆm (n),
m=1

where the fˆm (n) are as in Lemma 6.1:

 
 1 1 1 1 ··· 1 1 1 0 
 
 0 2 3 4 ··· A−2 A−1 A 0
 

 
 
 0
 0 6 12 ··· (A − 2)(A − 3) (A − 1)(A − 2) A(A − 1) 0 

 
 0
 0 0 24 ··· (A − 2)(A − 3)(A − 4) (A − 1)(A − 2)(A − 3) A(A − 1)(A − 2) 0 

.
 .. .. .. .. .. .. .. ..

.. 
 .
 . . . . . . . . 

 
(A−1)! A!
 0 0 0 0 ··· (A − 2)! 0
 
 2 3 

A!
 
 0
 0 0 0 ··· 0 (A − 1)! 2
0 

 
0 0 0 0 ··· 0 0 A! −(2πi)A

Row-reducing this matrix gives us the column vector of coefficients [am ]A

m=1 necessary to

construct f . 
 29

We now have the following result.

Theorem 6.3. For any A ∈ N, there exists a polynomial function f such that
∞
1 1
Z
X 1 2 1 ˆ 2
2A
= f (x) dx − f (0) .
n=1
n 2 0 2

1 2
P∞ 1
P∞
Proof. Observe that n=1 n2A is equivalent to n=1 nA . By Lemma 6.2, there exists a
polynomial function f such that fˆ(n) = 1/nA for all n 6= 0. We can thus find the exact sum
R1
of the series by instead finding the value of 0 |f (x)|2 dx and fˆ(0). The equality by Parseval’s
identity is
2 2
Z 1
2
X
fˆ(0) + fˆ(n) = f (x) dx.
n6=0 0

1 1
P∞ 1
Since 1/n2A is an even function of n, we have
P
2 n6=0 n2A = n=1 n2A , and therefore,

∞
1 1
Z
X 1 2 1 ˆ 2
= f (x) dx − f (0) . 
n=1
n2A 2 0 2
R1 2
We may evaluate the integral 0
f (x) dx directly to find
Z 1 A X
A
2
X ap aq
(8) f (x) dx = ,
0 p=1 q=1
p+q+1

where this f is as in the statement and proof of Lemma 6.3, and am is the mth coefficient of
f . Furthermore, we know fˆ(0) = 1
A+1
by Lemma 6.1, so we end up with

∞ A A
X 1 1 1 X X ap aq
=− +
n=1
n2A 2(A + 1)2 2 p=1 q=1 p + q + 1

which is no longer in terms of integrals.

This alternate method is how we programmed ParsevalSum to compute the (2A)th zeta
R1
value, as well as find a polynomial function f such that ζ(2A) = 12 0 |f (x)|2 dx − 12 |fˆ(0)|2 .
ParsevalSum uses equation (8) to evaluate the integral.

6.2. How We Found f (x) for ζ(2). This provides some extra insight for how we came
up with f (x) = −2πi x − 12 for our choice of function when calculating the value of ζ(2).


 30

However, this method instead yields f (x) = −2πix because this method constructs the
required function f with a zero constant term. Since we need to find the value of ∞ 1
P
n=1 n2 ,

we first will look for the function f whose Fourier Transform is fˆ(n) = 1/n for all n 6= 0.
Note that we will use the function fˆA (n) as the Fourier transform of fA (x) = xA for such n.
Following Theorem 6.3,

1
= a1 fˆ1 (n)
n
 
−1
= a1
2πin

for all n 6= 0, so a1 = −2πi. Therefore, by linearity, f (x) = −2πix. With this choice of f ,
R1 R1 2 R1 2 2
we find that 12 0 |f (x)|2 dx = 12 0 | − 2πix|2 dx = 2π3 and 12 |fˆ(0)|2 = 12 0 −2πix dx = π2 .
We then get

∞ 2
1 1
Z 1
2π 2 π 2 π2
Z
X 1 2 1
2
= |−2πix| dx − −2πix dx = − = .
n=1
n 2 0 2 0 3 2 6
P∞ 1
6.3. Example: ζ(4). Consider ζ(4). To compute n=1 n4 , we will look for a function f
such that fˆ(n) = 1/n2 for all n 6= 0. We next row-reduce the matrix
   
2
 1 1 0  1 0 −2π 
to obtain .

  
2 2
0 2 −(2πi) 0 1 2π

This shows that

1
fˆ(n) = 2
n
= 2π 2 · fˆ2 (n) − 2π 2 · fˆ1 (n),

and so our desired function is f (x) = 2π 2 x2 − 2π 2 x by linearity. The sum of the series is
therefore
∞ 2
1 1 1
π4
Z Z
X 1 2 1
4
= 2π 2 x2 − 2π 2 x dx − 2 2 2
(2π x − 2π x) dx = .
n=1
n 2 0 2 0 90
 31

6.4. Example: ζ(10). For a more complicated example, consider ζ(10). In order to compute
P∞ 1 ˆ 5
6 0. We next
n=1 n10 , we will look for a function f such that f (n) = 1/n for all n =

row-reduce the matrix

   
2 5
 1 1 1 1 1  0  1 0 0 0 0 45
π i 
   
0 2 3 4 5 0  0 1 0 0 0 0
   
  
   
 to obtain  0 4 5 
−9π i  .
  
 0 0 6 12 20 0 0 1 0 0
   
   2 5


 0 0 0 24 60 0 

 0
 0 0 1 0 3
π i 

   
5 4 5
0 0 0 0 120 −(2πi) 0 0 0 0 1 − 15 π i

This shows that

1
fˆ(n) = 5
n
4 5 ˆ 2 4 2
=− π i · f5 (n) + π 5 i · fˆ4 (n) − π 5 i · fˆ3 (n) + π 5 i · fˆ1 (n),
15 3 9 45
4 5 5
and so our desired function is f (x) = − 15 π ix + 23 π 5 ix4 − 49 π 5 ix3 + 45
2 5
π ix by linearity. The
sum of the series is therefore
∞ 2
1 1
Z Z 1
X 1 2 1
= f (x) dx − f (x) dx
n=1
n10 2 0 2 0
!
1 2π 10 1
= − |0|2
2 93555 2
π4
= .
93555
 32

7. EXTENSIONS

Over the course of this paper, we have established many results that allow us to use
Parseval’s identity to calculate the exact value of an infinite series, as well as some useful
applications. There are minor extensions that take the idea of our results to introduce other
methods to finding the exact sum of an infinite series.

7.1. Exact Sums Over N. While Parseval’s identity finds the exact sum of series over Z,
we are often interested in finding the exact sum of series over N instead. Calculus already
includes a couple classes of functions for which we can compute exact sums over N, but
Theorems 3.2 and 4.5 introduce two more classes of functions. We may thus find the exact
sum of a linear combination of the following list of functions over the natural numbers:

• telescoping summands
• geometric summands
• summands of 1/n2k for any k ∈ N
• even rational summands over Z. If g is an even rational function summable over
the integers, then we may find its exact sum over the natural numbers by using
P∞ 1
P 1
P
n=1 g(n) = 2 n∈Z g(n) − 2
g(0) and applying Parseval’s identity to n∈Z g(n)

using either Theorem 4.5 or Corollary 5.2.

For example, one can find the value of ∞ 2n2 +n+2

P
n=1 n4 +n3 +2n2 +2n by decomposing the summand
P∞  1 
and using linearity to obtain ∞ 1 1
P
2
n=1 n +2 + n=1 n − n+1
. The first sum may be found
by instead using Theorem 4.5 to compute 12 n∈Z n21+2 − 14 , and the second sum evaluates
P

to 1 as it is a telescoping series. Likewise, we can find ∞ n4 +3n

P
n=1 n4 ·3n by rewriting the series

as ∞
P 1
P∞ 1
n=1 n4 + n=1 3n . Theorem 3.2 allows us to find the value of the first series, and the

second series is a geometric series starting at n = 1.

7.2. Exact Sums Using Fourier Series. Another way to compute the exact sum of a
series over Z is to use the Fourier series expansion of f . That is, if f is a continuously
differentiable function on (0, 1) with the endpoints identified, and the right-derivative exists at
x = 0 and the left-derivative exists at x = 1, then it is equal pointwise to its Fourier series as
 33

fˆ(n)e2πinx = n∈Z fˆ(n) cos(2πnx) + i n∈Z fˆ(n) sin(2πnx). If we let x = 0 and

P P P
f (x) = n∈Z

x = 1, then f (0)+f (1)

=
P ˆ ˆ
2 n∈Z f (n) provided that f (n) is summable ([7], p. 37, Theorem 2.3.4).

If instead x = 1/2, then we get the sum of the alternating series f (1/2) = n∈Z (−1)n fˆ(n)
P

provided it is summable. The former gives us a way to compute the exact sum of a series over
Z whose summand is a rational function in C. If n∈Z (−1)n fˆ(n) diverges because either
P

the real or imaginary part of (−1)n fˆ(n) is not summable, but the other part is summable,
then we may instead compute only the sum of the summable part. The same thing applies
to n∈Z fˆ(n).
P
P 
1 n
Consider, for example, Im n∈Z n+i . The real part of the summand n2 +1 is not summable

over Z, but its imaginary part − n21+1 is. Take fˆ(n) = n+i 1
, and apply Theorem 4.5 to
P ˆ 2
P 1 ˆ
n∈Z |f (n)| = n∈Z n2 +1 using our initial choice of f (n). We then get the resulting
2πx
   
f (0)+f (1)
f (x) = −2πie 1
P
e2π −1
we need to find that Im n∈Z n+i = Im 2
= −π coth(π), which is
what we would expect from Proposition 4.8. If we instead use x = 1/2, then the alternating
n
series n∈Z (−1)
P
n+i
is summable in both the real and imaginary parts and has exact value
π (−1)n ·n
f (1/2) = − e2πe
2π −1 i. The real part evaluates to 0 because the real part of the summand n2 +1

is a summable odd function of n by the Alternating Series Test.

 34

REFERENCES
[1] Takashi Agoh and Karl Dilcher. Integrals of Products of Bernoulli Polynomials. Journal
of Mathematical Analysis and Applications, 381(1):10–16, 2011.

[2] Krishnaswami Alladi and Colin Defant. Revisiting the Riemann Zeta Function at Positive
Even Integers. International Journal of Number Theory, 14(07):1849–1856, 2018.

[3] Oscar Ciaurri. Euler’s Product Expansion for the Sine: An Elementary Proof. The
American Mathematical Monthly, 122(7):693–695, 2015.
[4] Louis De Branges. The Stone-Weierstrass Theorem. Proceedings of the American
Mathematical Society, 10(5):822–824, 1959.
[5] Asghar Ghorbanpour and Michelle Hatzel. Parseval’s Identity and Values of Zeta
Function at Even Integers. arXiv preprint arXiv:1709.09326, 2017.
[6] Anthony W Knapp. Basic Real Analysis. Springer Science & Business Media, 2005.
[7] J Korevaar. Fourier analysis and related topics. Amsterdam, Spring, 2011.

[8] M Ram Murty. Problems in Analytic Number Theory, volume 206. Springer Science &
Business Media, 2008.

[9] Halsey Lawrence Royden and Patrick Fitzpatrick. Real Analysis, volume 32. Macmillan
New York, 1988.

[10] David Salwinski. Euler’s Sine Product Formula: An Elementary Proof. The College
Mathematics Journal, 49(2):126–135, 2018.

[11] The Sage Developers. SageMath 9.0, the Sage Mathematics Software System (Version
0.5.2), 2020. https://www.sagemath.org.
 35

BIOGRAPHY OF THE AUTHOR

James was born in Waterville, Maine on March 30, 1997. He was raised in South China,
Maine and graduated from Erskine Academy High School in 2015. He attended the University
of Maine and graduated in December 2018 with a Bachelor’s degree in Mathematics, as well
as two Minors: one in Statistics and one in Music. He joined the University of Maine’s
Graduate School immediately after graduating. He spent four years as a member of the Black
Bear Men’s Chorus, and he worked as a Teacher’s Assistant during the latter 1.5 years at
the university. James is a candidate for the Master of Arts degree in Mathematics from The
University of Maine in May 2020.