Fourier Series & Fourier Transforms

[email protected] 19th October 2003

Synopsis
Lecture 1 :

Review of trigonometric identities Fourier Series Analysing the square wave

Lecture 2:

The Fourier Transform Transforms of some common functions

Lecture 3:
Applications in chemistry

FTIR Crystallography

Bibliography
1. The Chemistry Maths Book (Chapter 15), Erich Steiner, OUP, 1996.

Introduction
Chemistry often involves the measurement of properties which are the aggregate of many fundamental processes. A variety of techniques have been developed for extracting information about these underlying processes. of the most important and is very widely used - eg: Fourier analysis is one in crystallography, X-ray

adsorbtion spectroscopy, NMR, vibrational spectroscopy (FTIR) etc.. As it involves decomposition of functions into partial waves it also appears in many quantum mechanical calculations.

A Little Trigonometry
You will need to be able to manipulate sin() and cos() in order to understand Fourier analysis - a good understanding of the UK's A-level Pure Maths syllabus is sucient. Here is a brief reminder of some important properties.

Units Cos

angles are typically measured in radians:

0 360o is

equivalent to

0 2

radians and sin curves look like this:

Both sin(x) and cos(x) are periodic on the interval period, ie:

and integrate to 0 over a full

cos (x) dx =

sin (x) dx = 0

Wavelength
It should be clear that sin(2x) repeats on the interval interval

and sin(3x) on the

0 2/3

etc. In general sin(nx) and cos(nx) repeat on the interval

2/n.

The repeat distance is the wavelength

and so in general,

= 2/n. 2 .

The discrete family of functions sin(nx), cos(nx) are all said to be commensurate with the period

2 -

that is, they all have wavelengths which divide exactly into

The function sin(kx) for some real number k has an arbitrary wavelength

= 2/k .

is usually referred to as the wavevector.

Note: A simple Mathermatica notebook,

trig_1.nb,

is provided with the course

and can be used to play with sin and cos functions.

Fourier Series
The idea of a Fourier series is that any (reasonable) function, odic on the interval contributions from

f (x),

that is peri-

2 (ie: f (x + 2n) = f (x) sin(nx) and cos(nx).

for all n) can be decomposed into

The general Fourier series may be written as:

f (x) =

a0 2

+ a1 cos (x) + a2 cos (2x) + a3 cos (3x) + . . . + an cos (nx) + b1 sin (x) + b2 sin (2x) + b3 sin (3x) + . . . + bn sin (nx)
(1)

Note:

1.

cos (nx)

and

sin (nx)

are periodic on the interval

for any integer

n.

2. The

an and bn coecients measure the strength of contribution from each harmonic.

Orthogonality
The functions

cos (nx) and sin (nx) can be used in this way because they satisfy the

following orthogonality conditions:

cos (mx) sin (nx) dx =

0 0 2 0

f or all m, n m=n m=n=0 (m = n) > 0 m=n (m = n) > 0

cos (mx) cos (nx) dx =

= =
+

sin (mx) sin (nx) dx =

= 2 )

Note that the integrals only need to extend from

(or any other period of

as the functions simply repeat outside this range.

These conditions can be proved quite readily but it is relatively easy to see why they are true graphically.

cos(mx) sin(nx) ?

This is obvious (!) if you plot cos(x) and sin(x). It is easier to see why it is true by picking a special

sin(mx) sin(nx) n = m?

case; say the integral of sin(2x) sin(x) and plotting.

The symmetry of the plot makes it clear that an integral of this function over any period of

will yield 0.

Is it obvious that this will be true for all cases when Also for the case cos(mx)cos(nx) ? Can you prove it in the general case ?

n=m

Note: A simple Mathermatica notebook,

trig_1.nb,

is provided with the course

and can be used to play with these products and integrals.

cos(mx) cos(mx) - ie: the case when n = m

cos2 (mx) dx =

1 1 sin (2mx) (1 + 2 cos (mx)) dx = x+ 2 2 2m

Finding the coecients

1, cos (nx)or

sin (nx)

and integrating over a complete period leads to terms which are zero apart

from one which corresponds to the coecient

a0 , an

or

bn

respectively, that is:

a0 an bn f (x)

= = =

1 1 1

f (x) dx
+

f (x) cos (nx) dx

f (x) sin (nx) dx

If

is well behaved we can perform these integrals and obtain the Fourier

decomposition of

f (x).

Note: Well behaved in this context means that the function obeys the Dirichlet
conditions.

An Example
Consider the square wave:

f (x) = 1 = 0 f (x) = f (x + 2)

0x< x < 0

This appears to be a dicult case - the rather angular square wave does not look as if it will be readily expanded in terms of sine and cosine functions. The coecients in the expansion can be determined from the formulae given above.

a0

is determined by:

a0 =

f (x)dx =

1 dx =
0

1 =1
is simply because

where the restriction of the integral to the region zero in the region Similarly:

f (x)

is

0.

an =
just draw

1 cos (nx) dx = 0
0

cos (x)to

see that its integral from

is zero, so

an = 0 f or all n
For the b coecients we have,

bn

= = =

1 + 1 sin (nx) dx 0 1 cos(nx) n 0 1 (1 cos (n)) n

but,

cos (n) = +1 = 1
so,

n even n odd

bn =

0
2 n

n even n odd f (x)

having determined all of the coecients we can write the series for

f (x) =

1 2 + 2

sin (x) +

sin (3x) sin (5x) + + ... 3 5

The sum continues to an innite number of terms. We can see how it converges to the square wave by plotting the truncated sum containing a nite number of terms - lets call the sum containing n-trigonometric terms average value of the square wave.

fn (x) then f0 (x) =

1 2 is just the

f1 (x) =

1 2

sin (x)

is plotted below

which is the best approximation that can be made using just a constant and a sine wave - not great.

f2 (x) =

1 2

sin (x) +

2 3 sin(3x) looks like this:

which is beginning to look more like a square well. The weight in each contribution is falling and with each additional term the ne detail of the square wave is being rened. Including 50 terms,

f50 (x),

we get

which is a pretty decent approximation to the original square wave. Note that the little spikes at the edge of the square wave are present even after including many hundreds of terms (although they become ner and ner) they are a consequence of trying to describe a discontinuous step function with smooth sine waves  this was noticed and studied by the mathematician JW Gibbs in the late 1890's.

A more compact notation

ei = cos () + i sin ()
We can write;

cos () =

1 i e + ei 2

sin ()

1 i e ei 2i

Using these relations we can rewrite the Fourier series, equation 1, in the more compact exponential notation;

f (x)

a0 1 1 + ai einx + einx + bi einx + einx 2 2 n=1 2i n=1

which can be rearranged as;

f (x) =

cn einx
n=

In the exponential notation the orthogonality conditions are;

eimx einx dx = 2

and so the coecients given by

if m = n

cn =

1 2

einx f (x)dx

The original and compact notations are equivalent and the fore directly related to

cn

coecients are there-

an

and

bn

c0 cn cn

= = =

1 a0 2 1 (an bn ) 2 1 (an + bn ) 2

This more compact notation is used in almost all applications.

Fourier Transforms
2
(ie:

Functions of arbitrary periodicity

f (x + 2n) = f (x)

for all n). This can be generalised to functions periodic

on any interval. Functions with a periodicity of 2L (ie: posed into contributions from from the period 2L. The Fourier series may then be written as:

f (x + 2Ln) = f (x) for all n) can be decomsin n x and cos n x which are periodic on L L

f (x) =

a0 2

+ a1 cos

x x x x + a2 cos 2 + a3 cos 3 + . . . + an cos n L L L L x x x x + b1 sin + b2 sin 2 + b3 sin 3 + . . . + bn sin n L L L L

or

f (x) =

a0 x x + an cos n + bn sin n 2 L L n=1

or, in the exponential notation,

f (x) =

cn ein L
n=

cn =

1 2L

+L L

ein L f (x)dx

Note: The limits of integration cover a single period of the function which is not
2L rather than 2 . This allows a function of arbitrary period to be analysed.

Nonperiodic functions
Fourier series are applicable only to periodic functions but non-periodic functions can also be decomposed into Fourier components - this process is called a Fourier

Transform.
Imagine a function that is of a nite extent that is much less than the periodicity, 2L, as pictured below,

If L becomes very large (tends to innity) then we will have an isolated, aperiodic, function. We will use this limiting process to develop the equations for the Fourier Transform from the Fourier Series. Consider the Fourier Series for this function;

f (x) =

cn ein L
n=

Consider the limit in which L becomes very large. If we dene;

kn =
then

n L

f (x) =

cn eikn x
n=

and it is clear that for very large L the sum contains a very large number of waves with wavevector

kn and

that each succesive wave diers from the last by a tiny

change in wavevector (or if you prefer, wavelength),

k = kn+1 kn =

10

As was shown in the lecture, in the limit of large L able, the discrete coecients,

becomes a continuous vari-

cn ,

become a continuous function of

k , c (k)and

the

summation can be replaced by an integral and,

f (x) c(k)

= =

1 2

c(k)eikx dk f (x)eikx dx

These pair of equations are very often rescaled by substituting obtain;

c(k) =

2c(k)

to

f (x) c(k)

= =

1 2 1 2

c(k)eikx dk
+

f (x)eikx dx

The functions f and c are called a Fourier transform pair - c is the Fourier transform of f and f is the (inverse) transform of c.

11