Fourier series and periodicity

Donal F. Connon

8 December 2014

Abstract

A large number of the classical texts dealing with Fourier series more or less state that
the hypothesis of periodicity is required for pointwise convergence. In this paper, we
highlight the fact that this condition is not necessary.

1. Whither periodicity?

When I first studied Fourier series some 40 years ago, as part of an applied
mathematics course, I was told that it was necessary for the function to be periodic,
i.e. f ( x) = f ( x + p) where (usually) p = 2π . If f ( x) was defined on an interval, say,
[−π , π ] , we were told, as in [13, p.244] and [12, p.299], to extend f ( x) periodically
(and also to graphically display the extended function so as to aid visualisation).

One would then obtain a Fourier series

∞
1
f ( x) = a0 + ∑ (an cos nx + bn sin nx)
2 n =1

and, in view of the fact that both cos nx and sin nx were 2π periodic, it is therefore
clear that we do indeed end up with a representation of f ( x) that is periodic.

However, as I mentioned in [7], I was never able to truly fathom why such periodicity
should, ab initio, be a requirement in the Dirichlet conditions used in the rigorous
treatment of Fourier series (see, for example, Apostol [1, p.319], Bressoud [3, p.223],
Titchmarsh [14, p.401] and Whittaker and Watson [16, p.164]). Why indeed should
the nature of a real function outside of its defined finite domain have any effect on the
pointwise convergence of the corresponding Fourier series?

Apostol [1, p.323] stated that “The hypothesis of periodicity, which appears in all the
convergence theorems dealing with Fourier series, is not as serious a restriction as it
may appear to be at first sight. If a function is initially defined on a finite interval, say
[a, b] , we can always extend the definition of f outside [a, b] by imposing some sort
of periodicity condition. For example, if f (a) = f (b) , we can define f everywhere
on (−∞, +∞) by requiring the equation f ( x) = f ( x + p) to hold for every x where
p = b − a . (The condition f (a) = f (b) can always be brought about by changing the
value of f at one of the endpoints if necessary. This does not affect the existence or
the values of the integrals which are used to compute the Fourier coefficients of f .)”

What Apostol says is of course perfectly correct, but we are still in the dark as to why
it is considered that periodicity is required ab initio. We do not really fare any better
by examining other classical texts such as Zygmund [17, p.8] where it is simply stated
that “…, it is convenient to assume (as we shall always do in what follows) that the
functions whose Fourier series we consider are defined not only in an interval of
length 2π but for all values of x , by the condition of periodicity f ( x + 2π ) = f ( x) .”
[emphasis added]. Hobson [10, p.490] makes a similar comment regarding periodicity
and convenience.

Tolstov [15] starts his book with a section on periodic functions and appears to
require the periodicity condition throughout his text. This is also the approach taken,
inter alia, by Bachman, Narici and Beckenstein [2, p.184], Churchill [6, p.90], Knopp
[11, p.358] and Whittaker and Watson [16, p.164].

2. Demonstration that periodicity is not required

The most fundamental formula in the theory of Fourier series is Dirichlet’s integral
which may be stated [5, p.227] for 0 < a < π

sin μ x π
a
(1) lim ∫ f ( x) dx = f (0+ )
μ →∞ sin x 2
0

where μ is not necessarily an integer.

Let us first examine how Apostol [1, p.317] approaches the theory of Fourier series
where he considers the series expansion on the interval [0, 2π ] . We have the partial
sum

n
1
sn ( x ) = a0 + ∑ (ak cos kx + bk sin kx)
2 k =1

where the Fourier coefficients are

2π 2π
1 1
ak =
π ∫0
f (t ) cos ktdt bk =
π ∫
0
f (t ) sin kt dt

Hence we have
2π
1 ⎡1 n ⎤
sn ( x ) = ∫ f (t ) ⎢ + ∑ (cos kt cos kx + sin kt sin kx) ⎥ dt
π 0 ⎣ 2 k =1 ⎦

2π
1 ⎡1 n ⎤
= ∫ f (t ) ⎢ + ∑ cos k (t − x) ⎥ dt
π 0 ⎣ 2 k =1 ⎦

1
2π
sin ⎡⎣( 2n + 1) (t − x) / 2 ⎤⎦
=
π ∫
0
f (t )
2sin [ (t − x) / 2]
dt

so that

2
 2π
1
(4) sn ( x ) =
π ∫
0
f (t ) Dn (t − x) dt

where the function Dn is called Dirichlet’s kernel.

Since Dn is periodic with period 2π , and postulating that f is also periodic with
period 2π , we can replace the interval of integration by [ x − π , x + π ] and then make
a translation u = t − x to get

x +π
1
sn ( x ) =
π ∫
x −π
f (t ) Dn (t − x) dt

π
1
=
π ∫
−π
f ( x + u ) Dn (u ) du

Since Dn (u ) = Dn (−u ) we obtain (on the assumption that f is periodic)

π
2 f (x + t) + f (x − t)
π∫
(5) sn ( x ) = Dn (t ) dt
0
2

which we may write as

π
4 2
f ( x + 2v) + f ( x − 2v)
=
π ∫
0
2
Dn (2v) dv

Then applying (1) we obtain for 0 < x < 2π

1
lim sn ( x) = [ f ( x − 0) + f ( x + 0)]
n →∞ 2

Enlightenment of the fact that the periodicity condition is redundant finally came
through reading Carslaw’s excellent treatise, Introduction to the theory of Fourier’s
series and integrals [5, p.230].

Instead of assuming that f is periodic, we start with (4)

2π
1
sn ( x ) =
π ∫
0
f (t ) Dn (t − x) dt

x 2π
1 1
=
π ∫
0
f (t ) Dn (t − x) dt +
π ∫
x
f (t ) Dn (t − x) dt

3
and make the substitution t = x − 2v in the first integral and t = x + 2v in the second
integral to obtain

x x
π−
2 2
2 2
π∫
(6) sn ( x ) = f ( x − 2v) D (2v) dv + ∫ f ( x + 2v) Dn (2v) dv
0
n
π 0

Then using (1) we again obtain for 0 < x < 2π

1
(7) lim sn ( x) = [ f ( x − 0) + f ( x + 0)]
n →∞ 2

It is thus seen that there is no need to assume ab initio that f is periodic.

When x = 0 we have

π
2
sn (0) =
π ∫ f (2v) D (2v) dv
0
n

and from (8) below we find that

1
lim sn (0) = [ f (0+ ) + f (2π −)]
n →∞ 2

Similarly, when x = 2π we have

π
2
sn (2π ) =
π ∫ f (2π − 2v) D (2v) dv
0
n

and (8) tells us that

1
lim sn (0) = [ f (0+ ) + f (2π −)]
n →∞ 2

3. Extension of Dirichlet’s integral

As noted by Bromwich [4, p.492] we may increase the range of Dirichlet’s integral
(1) to π in the case where μ = 2 N + 1 and N is an integer. We have

π
π π
sin(2 N + 1) x 2
sin(2 N + 1) x sin(2 N + 1) x
∫
0
f ( x)
sin x
dx = ∫ 0
f ( x)
sin x
dx + ∫ f ( x)
π sin x
dx
2

and make the substitution x = π − t in the second part. This immediately gives us

4
 π
π
sin(2 N + 1) x 2
sin(2 N + 1) x
∫
0
f ( x)
sin x
dx = ∫ [ f ( x) + f (π − x)]
0
sin x
dx

and employing (1) we obtain

π
sin(2 N + 1) x π
(8) lim ∫ f ( x) dx = [ f (0+ ) + f (π −)]
N →∞ sin x 2
0

In fact, as noted below, we may extend the range of the integration well beyond π .
For example, we consider
2π π 2π
sin(2 N + 1) x sin(2 N + 1) x sin(2 N + 1) x
∫ 0
f ( x)
sin x
dx = ∫ f ( x)
0
sin x
dx + ∫
π
f ( x)
sin x
dx

and we write

3π
2π 2π
sin(2 N + 1) x 2
sin(2 N + 1) x sin(2 N + 1) x
∫
π
f ( x)
sin x
dx = ∫
π
f ( x)
sin x
dx + ∫ f ( x)
3π sin x
dx
2

The substitution x = π + t gives us

3π π
2
sin(2 N + 1) x 2
sin(2 N + 1)t
∫
π
f ( x)
sin x
dx = ∫0
f (π + t )
sin t
dt

and with x = 2π − t we obtain

π
2π
sin(2 N + 1) x 2
sin(2 N + 1)t
∫
π
3
f ( x)
sin x
dx = ∫0
f (2π − t )
sin t
dt
2

Therefore, using Dirichlet’s integral (1) we see that

2π
sin(2 N + 1) x 1
lim
N →∞ ∫
0
f ( x)
sin x
dx = [ f (0+) + f (π −) + f (π +) + f (2π −)]
2

More generally we see that

mπ
sin(2 N + 1) x 1 1 m −1
lim ∫ f ( x) dx = [ f (0+ ) + f (mπ −)] + ∑ [ f (nπ −) + f (nπ +)]
N →∞ sin x 2 2 n =1
0

and, if f ( x) is continuous, this may be written as

5
 sin(2 N + 1)π x m −1
m
1
lim ∫ f ( x) dx = [ f (0) + f (m)] + ∑ f (n)
N →∞
0
sin π x 2 n =1

In [8] we used this to obtain a version of the Poisson summation formula

m m ∞m
1
∑ f ( n) = ∫ f ( x) dx + [ f (0) + f (m) ] + 2∑ ∫ f ( x) cos 2nπ x dx
n =0 0
2 n =1 0

Letting m → ∞ , and assuming that lim f (n) = 0 (which is of course required for the
n →∞

convergence of the infinite series), we obtain

∞ ∞ ∞ ∞
1
f (0) + ∑ f (n) = ∫ f ( x) dx + 2∑ ∫ f ( x) cos 2π nx dx
2 n =1 0 n =1 0

We may also note that

sin(2 N + 1) x
a a a
sin 2 Nx cos x
∫
0
f ( x)
sin x
dx = ∫ f ( x)
0
sin x
dx + ∫ f ( x) cos 2 Nx dx
0

and, for suitable functions, the Riemann Lebesgue lemma tells us that
a
lim ∫ f ( x) cos 2 Nx dx = 0
N →∞
0

and thus for 0 < a < π

a
π
lim ∫ f ( x) cot x sin 2 Nx dx = f (0+ )
N →∞ 2
0

4. Open access to our own work

This paper contains references to various other papers and, rather surprisingly, all of
them are currently freely available on the internet. Surely now is the time that all of
our work should be freely accessible by all. The mathematics community should lead
the way on this by publishing everything on arXiv, or in an equivalent open access
repository. We think it, we write it, so why hide it? You know it makes sense.

REFERENCES

[1] T.M. Apostol, Mathematical Analysis, Second Ed., Addison-Wesley Publishing

Company, Menlo Park (California), London and Don Mills (Ontario), 1974.

[2] G. Bachman, L. Narici and E. Beckenstein, Fourier and Wavelet Analysis.

Springer-Verlag, New York, 2000.

6
[3] D. Bressoud, A Radical Approach to Real Analysis.
The Mathematical Association of America, 1994.

[4] T.J.I’a Bromwich, Introduction to the theory of infinite series. Third edition.
AMS Chelsea Publishing, 1991.

[5] H.S. Carslaw, Introduction to the theory of Fourier’s Series and Integrals.
Third Ed. Dover Publications Inc, 1930.
https://archive.org/stream/introductiontoth00carsrich#page/34/mode/2up

[6] R.V. Churchill, Fourier series and boundary value problems.

Second Edition. McGraw-Hill Book Company, Inc. 1963.

[7] D.F. Connon, Some series and integrals involving the Riemann zeta function,
binomial coefficients and the harmonic numbers. Volume V, 2007.
http://arxiv.org/abs/0710.4047

[8] D.F. Connon, Some applications of the Dirichlet integrals to the summation of
series and the evaluation of integrals involving the Riemann zeta function.
http://arxiv.org/abs/1212.0441, 2012.

[9] P.G.L. Dirichlet, Sur la convergence des séries trigonométriques qui servent à
représenter une fonction arbitraire entre des limites données.
Journal für die reine und angewandte Mathematik, 4, (1829), 157-169.
http://www.digizeitschriften.de/zeitschriften/open-access/

[10] E.W. Hobson, The theory of functions of a single variable and the theory of
Fourier’s series, Volume II. Cambridge, 1907.
https://archive.org/stream/theoryoffunction032497mbp#page/n13/mode/1up/search/fourier

[11] K. Knopp, Theory and Application of Infinite Series.

Second English Ed. Dover Publications Inc, New York, 1990.

[12] M.R. Spiegel, Schaum's Outline Series of Theory and Problems of Advanced
Calculus. McGraw Hill Book Company, 1963.

[13] G. Stephenson, Mathematical Methods for Science Students.7th Ed.

Longman Group Limited, London, 1970.

[14] E.C. Titchmarsh, The Theory of Functions.

Second Ed., Oxford University Press, 1932.

[15] G.P. Tolstov, Fourier Series. (Translated from the Russian by R.A.Silverman)
Dover Publications Inc, New York, 1976.

[16] E.T. Whittaker and G.N. Watson, A Course of Modern Analysis: An

Introduction to the General Theory of Infinite Processes and of Analytic
Functions; With an Account of the Principal Transcendental Functions. Fourth
Ed., Cambridge University Press, Cambridge, London and New York, 1963.

7
[17] A. Zygmund, Trigonometric Sums.
Cambridge Mathematical Library, 2002.

Wessex House,
Devizes Road,
Upavon,
Pewsey,
Wiltshire SN9 6DL
dconnon@btopenworld.com