You are on page 1of 62

Chapter5

FourierAnalysis
Reading:
Kreyszig,AdvancedEngineeringMathematics,10thEd.,2011
Selectionfromchapter11
Prerequisites:
Kreyszig,AdvancedEngineeringMathematics,10thEd.,2011
Complexnumbers:Sections13.1,13.2and13.5

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 1
Contents
1. Fourier analysis. Motivation: Analysis of complex periodic and nonsmooth
functions.
2. Periodic functions. Basic trigonometric function. Trigonometric sum and series.
3. Orthogonal system of functions. Trigonometric system of functions.
4. Fourier and generalized Fourier series.
5. Fourier expansions for functions satisfying the Dirichlet conditions.
6. Complex Fourier series.
7. Fourier series of even and odd periodic functions .
8. Fourier series of nonperiodic function given at finite interval. Halfrange
series.
9. Parseval's identity.
10. Application of the Fourier series: Solving ODEs, forced mechanical oscillations.
11. Application of the generalized Fourier series: The SturmLiouville problem.
Solving the heat conduction equation by the separation of variables.
12. Application of the Fourier series: Frequency spectrum analysis.
13. Fourier transform.
14. Various forms of the Fourier transform.
15. Applications of the Fourier transform.
16. Discrete Fourier transform (DFT). Fast Fourier transform (FFT).

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 2
5.1.Fourieranalysis.Motivation:Analysisofcomplexperiodicandnonsmoothfunctions

Analysis of complex functions is often based on their representation in the form a series
infinite sum of simple functions.
Example: Taylor expansion Representation of a function in the form of the power series

1! 2!
Let's consider a periodic and nonsmooth function

What if we will try to use the Taylor expansion?


In the point 0 we obtain the Taylor series in the form

This is great (accurate results for our ) inside a period, but becomes meaningless outside
the period since the function is discontinuous. The Taylor expansion can be applied only inside
intervals where the function is continuous and has all continuous derivatives.
The major motivation of the Fourier analysis is to develop an approach for series representation
of (almost arbitrary) discontinuous periodic and nonperiodic functions.
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 3
5.2.Periodicfunctions.Basictrigonometricfunction.Trigonometricsumandseries
Many phenomena in science and engineering are
periodic and described in terms of periodic functions.
Examples:
1. Mechanical oscillations (massspring systems,
pendulums, strings, membranes).
2. Oscillations in electrical circuits.
3. Periodic motion of planets.
4. Wave motion (acoustic waves, electromagnetic
waves, radio, etc.).
5. Oscillations of individual atoms in crystalline solids.
Periodic function is a function which satisfies the
following condition
for all :
where parameter is called the period.
Notes:
1. Obviously, if is a period, then is also the period for function . By default, we will
use the term period in order to denote the minimum period of function . The minimum
period is also called the fundamental period.
2. It is sufficient to study any periodic function only at any interval , .
3. Any function given at finite interval , can be periodically extended for any with
the period .
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 4
5.2.Periodicfunctions.Basictrigonometricfunction.Trigonometricsumandseries
Let's consider the functions for fixed parameter
cos sin
is the basic trigonometric function for 0,1, and and are constant coefficients.
Properties of :
1. Common period for all is 2 / , is the fundamental angular frequency.
Proof: cos 2 sin 2
2. Let's introduce the magnitude and phase arctan / . Then
cos sin
Proof: cos cos cos sin sin .
3. If 2 , is the halfperiod, then 2 / / and
cos sin
terms trigonometric sum is

cos sin
2 2
isthefundamental
Trigonometric series is periodfor and

lim cos sin


2 2
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 5
5.2.Periodicfunctions.Basictrigonometricfunction.Trigonometricsumandseries
Twobasicapplicationsofthetrigonometricsumandseries
1. In many mathematical problems, trigonometric sum or series represents an accurate solution
of the problem. In this case, and coefficients and are defined from equation to be
solved and initial/boundary conditions.
Example: SturmLiouville problem (will be considered later)
2. Any periodic function with period can be represented in the form of trigonometric series.

cos sin
2 2

This is similar to the representation of a function in the form of a power series

where coefficients should coincide with the coefficients given by the Taylor expansion

1! 2!
In order to introduce the representation of a function in the form of a trigonometric series, we
need to know how to define coefficients and ( is not arbitrary, it is given by the period:
2 /
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 6
5.3.Orthogonalsystemoffunctions.Trigonometricsystemoffunctions
Let's consider a system of functions , , , , 0,1,2, given at
interval , . The system of functions is called orthogonal in , with respect to the
weight 0 if

0,
,
,

The trigonometric system of functions at fixed is the system


1, cos , sin , cos 2 , sin 2 , , cos , sin ,.
Theorem:
For a given , the trigonometric system of functions is orthogonal at any interval , ,
where 2 / with respect to weight 1 and, moreover,

2 , at 0. (5.3.1)
2
Proof:

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 7
5.3.Orthogonalsystemoffunctions.Trigonometricsystemoffunctions
In order to calculate other integrals, we need to use a number of trigonometric equations that
all follow from
cos cos cos sin sin (5.3.2)
sin sin cos cos sin
Lets first consider the case when . Eq. (5.3.2) results in
1 cos 2 1 cos 2 sin 2
cos , sin , sin cos
2 2 2
Then

1 cos 2
cos
2 2

1 cos 2
sin
2 2

1
sin cos sin 2 0
2

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 8
5.3.Orthogonalsystemoffunctions.Trigonometricsystemoffunctions
Now let's consider cos cos at . Eq. (5.3.2) results in

cos cos cos sin sin


cos cos cos sin sin

Sum of these two equations results in

1
cos cos cos cos
2
Then

1
cos cos cos cos 0
2

Similarly (see Kreyszig, page 479) one can prove that at

cos sin 0, sin sin 0

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 9
5.4.FourierandgeneralizedFourierseries
Let's consider the system of functions , , ,... which are orthogonal at , with
respect to weight and assume that some periodic function with period 2 /
can be represented in the form:

(5.4.1)

The coefficients in Eq. (5.4.1) can be found with the following theorem:
Theorem:
If function can be represented in the form given by Eq. (5.4.1), then coefficients in this
series are unique and can be found with the following Euler formulas:

, 1 (5.4.2)

Proof:
Let's multiply Eq. (5.4.1) by and integrate it from to :

or
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 10
5.4.FourierandgeneralizedFourierseries

, , .

Now let's use the orthogonality: , 0 if . Then


, , .
The representation of the function in the form given by Eq. (5.4.1) where coefficients are
calculated with Eqs. (5.4.2) is called the generalized Fourier series.

Fourierseries
Now lets apply our general theorem to the trigonometric system of functions
, , : 1, cos , sin , cos 2 , sin 2 , , cos , sin ,.
where 2 / 2 is the fundamental period of function .
According to Eq. (5.3.1) 2 , /2 at 0. Then lets rewrite Eq.
(5.4.1) in the form

, , , , ,
/2

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 11
5.4.FourierandgeneralizedFourierseries
In the sum, lets group together cos and sin functions of the same argument. Then
1 ,1 , cos , sin
cos sin
2 /2 /2 /2

Now lets introduce the Fourier coefficients and of function :

,1 2

/2
TheseareFourier
coefficients offunction , cos 2
cos 1,2 (5.4.3)
/2

, sin 2
sin
/2

Then
ThisistheFourier
series offunction cos sin (5.4.4)
2

The representation of function in the form (5.4.4) where coefficients are calculated with
Eqs. (5.4.3) is called the Fourier series of function .
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 12
5.4.FourierandgeneralizedFourierseries
DifferentformsoftheFourierseries 2 / 2
If we use :

cos sin
2
/ / /

, cos , sin

If we use :
2 2
cos sin
2

2 2 2 2 2
, cos , sin

If we use :

cos sin
2

1 1 1
, cos , sin

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 13
5.4.FourierandgeneralizedFourierseries
In terms of magnitude and phase:
, arctan /

cos sin
2 2

Example: Let's consider the periodic function given by the plot

1 2, 1,
1 1 Atperiod 1,1 ,

1
Fourier coefficients according to the Euler formulas (5.3.2)

0, cos 0

1 1
sin cos cos cos

2 2 1
cos
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 14
5.4.FourierandgeneralizedFourierseries

2 1
sin
2 , , 1

We have obtained good approximation of a function with discontinuities valid at any !


What if we will try to use the Taylor expansion?

1! 2!
In the point 0 we have

This is great (accurate results for our ) inside a period, but becomes meaningless outside
the period if the function is discontinuous.
Conclusion: The Fourier expansion provides good approximation ax arbitrary of even
discontinuous (but periodic functions), while the Taylor expansion can be applied only inside
intervals where the function is continuous and has all continuous derivatives.
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 15
5.4.FourierandgeneralizedFourierseries
Jean Baptiste Joseph Fourier (21 March 1768 16 May 1830) was a French
mathematician and physicist best known for initiating the investigation of
Fourier series and their applications to problems of heat transfer and vibrations.
The Fourier transform and Fourier's Law are also named in his honor.

cos sin
2
ButwasJosephFourierthefirstwho"invented"theFourierSeries?
Claudius Ptolemy (c. AD 90 c.168) Epicycle Imperfections in the
was a GrecoEgyptian astronomer, Ptolemaic system were
who invented the Ptolemaic discovered through
geocentric model of universe, where observations accumulated
each planet is moved around Earth over time. More levels of
by a system of two spheres: one epicycles (circles within
called its deferent, the other, its circles) can be added to
epicycle. the model to match more
accurately the observed
Deferent planetary motions.
Cosine and sine functions parametrically define the circular
motion.
Representation of the visible trajectory of a planet in the
cos form the system of deferent + epicycles is equivalent to the
sin expansion of the trajectory into the Fourier series.
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 16
5.5.FourierexpansionsforfunctionssatisfyingtheDirichlet conditions
Our goal is to formulate conditions for a function which guarantee that the Fourier series
for this function exists and converges to the values of the function. Such conditions are known
as the Dirichlet conditions.
Let's consider some periodic (real value) function with period . We say that this function
satisfies the Dirichlet conditions if
1. This function is a piecewise monotonous function in any interval , (This means that
has a finite number of extrema in this interval).
2. This function is a piecewise continuous function in any interval , (This means that
has a finite number of discontinuities in this interval).
3. This function has finite limits in the ends of the interval , and finite left and right
hand limits at any discontinuity (i.e. all discontinuities are jump or step discontinuities).

Lefthandlimit:

0 lim
,

Righthandlimit:

0 lim
,

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 17
5.5.FourierexpansionsforfunctionssatisfyingtheDirichlet conditions
Example 1: A function satisfying the Dirichlet conditions

Example 2: Function sin 1/ does not satisfy the Dirichlet conditions, since it is not a
piecewise monotonous function in any interval containing 0.

sin 1/

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 18
5.5.FourierexpansionsforfunctionssatisfyingtheDirichlet conditions
Theorem: Dirichlet theorem
If is a periodic function with the fundamental period 2 / and satisfies the Dirichlet
conditions, then
1. The Fourier series

cos sin
2
where the Fourier coefficients are calculated with the Euler formulas (2.3.4) converge at any .
2. is also a periodic function with the fundamental period .
3. If is continuous in the point , then

4. If is discontinuous in the point , then values of is the halfsum of left and right
hand limits of :
1
0 0
2
5. The Fourier coefficients approach 0 when : , 0 if .
6. The "speed" of convergence of and to zero with increasing depends on the degree of
smoothness of : If has discontinuous derivatives of order , then
1

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 19
5.5.FourierexpansionsforfunctionssatisfyingtheDirichlet conditions
Example:

1 2, 1,
1 1 Atperiod 1,1 ,

2 1
sin
2 , , 1

We see that:
1 1
0 0 ,
2
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 20
5.6.ComplexFourierseries
Lets assume that some periodic function can be represented in the form of the Fourier
series

cos sin (5.6.1)


2

2 2 2
, cos , sin , 1,2

Lets show that every cos sin can be represented in a form containing
complex numbers using the Euler formula for the complex exponent:

cos sin (5.6.2)


Then

cos , sin
2 2

cos sin
2 2

(5.6.3)
2 2
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 21
5.6.ComplexFourierseries
Now lets introduce the complex Fourier amplitudes ( ):
1 1
cos sin
2

1 1
cos sin
2

2
Now lets substitute (5.6.3) into Eq. (5.6.1). Then we obtain:

1 (5.6.4)
,

Eq. (5.6.4) is the complex Fourier series of the periodic realvalued function .
If the complex Fourier amplitudes are found, then :
2 2

2 2

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 22
5.7.Fourierseriesofevenandoddperiodicfunctions
Function is called the odd function if

Function is called the even function if


.
Examples:
Oddfunctions:sin ,Evenfunctions:cos ,

1 1
1 1
1 1
1
Simple properties:
1. If and are even functions and and are odd functions, then
and are even functions, is the odd function.
2. If is an odd function, then integral over any interval symmetric with respect to 0 is
equal to zero

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 23
5.7.Fourierseriesofevenandoddperiodicfunctions
Fourier expansion of an even periodic function :
Lets take /2 :

1 1 1
, cos , sin 0,

cos
2
For an even function, phase arctan / 0, in the complex Fourier series .
Fourier expansion of an odd periodic function :
Lets take /2 :

1 1 1
, 0 cos 0 , sin ,

sin

For an odd function, phase arctan / 0, in the complex Fourier series .


ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 24
5.7.Fourierseriesofevenandoddperiodicfunctions
Even/Odd decomposition of periodic function :
Any function can be uniquely decomposed into a sum of even, , and odd, , functions:

where
1 1
,
2 2
It allows one to introduce another form of the Fourier series:

cos
2

sin

cos sin
2
where
1 1 1
, cos , sin

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 25
5.8.Fourierseriesofnonperiodicfunctiongivenatfiniteinterval.Halfrangeseries
Lets consider a nonperiodic function given at the finite interval , (| | )
and assume that satisfies the Dirichlet conditions.
Example: Nonperiodic Periodicextension

Although is nonperiodic, it can be expanded into the Fourier series. For this purpose we
need to periodically extend for arbitrary . For instance, we can introduce a new periodic
function with the period which is defined as follows:

(5.8.1)

where is the integer part of the real number .


The periodic extension satisfies the Dirichlet conditions and, thus, can be represented in
the form of the Fourier series. In the interval , the Fourier expansion of will
coincides with in all points except the discontinuities of .
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 26
5.8.Fourierseriesofnonperiodicfunctiongivenatfiniteinterval.Halfrangeseries
For any nonperiodic given in a finite interval, there is infinitely large number of different
periodic extensions with different fundamental periods .
Example:

1
1 1

0.5 1.5 0.5 0.5


In particular, for any given at , one can introduce an extension which can be
either even or odd periodic function of the period 2 . The Fourier series for
obtained with the help of even or odd periodic functions are called halfrange Fourier
expansions.
Let's first consider the case 0. Then we can introduce the even or odd periodic extension in
two steps:
1. First, we introduce auxiliary function given at , .
0
a. For the odd extension:
0
0
b. For the even extension:
0
2. Second, we introduce a periodic expansion for given by Eq. (5.8.1).
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 27
5.8.Fourierseriesofnonperiodicfunctiongivenatfiniteinterval.Halfrangeseries
Example:
Odd Even
1 1
1
1
0 1 1 1 0 1
0
1
If 0, then the odd or even extension can be introduced by two ways:
1. If 0 or 0 we can introduce an extended function in a way illustrated in the figure in
the previous slide.
2. In the general case, we can introduce a shifted function , which is defined in
0 and then apply our twostep algorithm from the previous slide to .
The halfrange Fourier expansions are convenient to use, since only half of all Fourier
coefficients should be determined: 0 for even periodic extensions, 0 for odd periodic
extensions.
Question: Which extension is better?
In general, the better results are obtained with an extension which removes the discontinuities
in . Then, according to the Dirichlet theorem, the convergence of and to zero with
increasing is faster and we can retain smaller number of terms in the Fourier sum in practical
calculations.
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 28
5.8.Fourierseriesofnonperiodicfunctiongivenatfiniteinterval.Halfrangeseries
Example: Odd Even
1 1
1
1
0 1 1 1 0 1
0
Odd extension: 1
Seesect.3.3:
1
1 1 2 1
sin

Even extension: 1 2 1 1
cos
1 2

1 0 1
2 1

2 2 1 1
2 cos sin sin

In this case the even extension is better, since it removes discontinuities and provides ~1/ .
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 29
5.9.Parsevals identity
Many applications of the Fourier series use the following theorem:
Theorem: Parsevals theorem
Lets consider a periodic function that satisfies the Dirichlet conditions and, thus can be
represented in the form of the Fourier series

cos sin (5.9.1)


2
Then Parsevals identity holds:

1 1 (5.9.2)
2 2
Proof:
Lets rewrite Eq. (5.9.1) using the following notation for the functions of the trigonometric
system: 1, cos , sin , etc., and /2, , , etc.

And then let's calculate the product

(5.9.3)

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 30
5.9.Parsevals identity
Now we can integrate Eq. (5.9.3) over a period:

But the trigonometric system of functions is orthogonal with respect to the weight 1, i.e.
0
0
/2 0

Consequence: If we use the complex Fourier series

Then Parsevals identity takes the form

1
(5.9.4)

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 31
5.9.Parsevals identity
In order to prove it, lets use the definition of :

, ,
2 2 2
Then /4

1
2
2 2 2
,

Energyspectrum
Many applications of Parseval's identity are based on the interpretation of /2 as
energy (or power) associated with a particular term/oscillation in the Fourier series.
Example 1: Mechanical oscillations.
Assume that describes displacement of an oscillating mass in the massspring system.
Let's first consider a harmonic oscillation of an undamped system with the equation of motion
0
The solution is cos sin (here ) and the energy "stored" in
such a harmonic oscillation is equal to ( / )
1 1

2 2 2 2
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 32
5.9.Parsevals identity
Thus, if we have a nonharmonic oscillation that can be represented as a superposition of
infinite number of harmonic oscillations (i.e. in the form of the Fourier series), the averaged
over a period energy stored in the oscillation is a sum of energies stored in the individual
harmonic oscillations (in every harmonic oscillation and does not depend on time):
1 1
2 2

Distribution of averaged energy over different oscillation frequencies is called the energy
(power) spectrum of oscillation . In this regard, we say that any periodic function
(oscillation) has a discrete or point spectrum, since the energy of such oscillation is stored in
countably many isolated frequencies .

3
Example 2: Joule heat in electrical circuits. Let's consider a part of an electrical circuit with
resistance and current . Then the Joule heat dissipated during time is equal to
. If we would calculate the Fourier coefficients for , then value /2 is
proportional to the contribution of oscillation of frequency to the total electric power .
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 33
5.10.ApplicationoftheFourierseries:SolvingODEs,forcedmechanicaloscillations
Mechanicalmassspringsystem Newtonssecondlawofmotion:

1.Elasticrestoringforce(Hookslaw):

isthespringconstant (springstiffness)

2.Damping(friction)force:

: Displacement
3.Input(driving)externalforce

(5.10.1)

Undamped oscillation: 0
Dampedoscillation: 0
Freeoscillation: 0
Forcedoscillation: 0

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 34
5.10.ApplicationoftheFourierseries:SolvingODEs,forcedmechanicaloscillations
Solutionforharmonicdrivingforce(SeeSection2.8)
We considered only the harmonic driving force, cos , where is the input angular
frequency
cos

In the case of damped oscillation ( 0), 0 when , so that (see Section 2.8)

cos sin (5.10.2)

, ,

Using the Fourier series, we can generalize the solution for an arbitrary periodic driving force
. Let's assume that
is the periodic function which satisfies the Dirichlet conditions, and thus, can be
expanded into the Fourier series.
Mean value of is zero: 0.
is the even function (only for the sake of simplicity, the general case can be considered).
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 35
5.10.ApplicationoftheFourierseries:SolvingODEs,forcedmechanicaloscillations
Then ( 2 ):
1
cos , cos

From Eq. (5.10.1):

cos (5.10.3)

Let's consider an equation for a single Fourier term ( / ):


cos (5.10.4)
The particular solution of the nonhomogeneous ODE (5.10.4) is given by Eq. (5.10.2):
cos sin

Since Eq. (5.10.3) is linear, the particular solution of this equation takes the form:

cos sin (5.10.5)

In order to check that Eq. (5.10.5) is a solution of (5.10.3): Substitute Eq. (5.10.5) into (5.10.3).
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 36
5.11. Application of the generalized Fourier series: The SturmLiouville problem. Solving the heat
conduction equation by the separation of variables
The boundary value problem for the secondorder linear ODE:
0, , , 0 (5.11.1a)
0 (5.11.1b)
0 (5.11.1c)
where , , , and are real numbers, is called the SturmLiouville problem.
Note: The SturmLiouville problem is important for solving PDEs with separating variables
(example will be considered below).
Any SturmLiouville problem has the trivial solution 0 (can be proved by substitution). The
fundamental property of the SturmLiouville problem, however, is a nonuniqueness of solution
at some particular values of : At some , other, nontrivial ( 0) solutions also exist.
If the SturmLiouville problem has a nontrivial solution at some , such a is called the
eigenvalue of the SturmLiouville problem, and is called the eigenfunction corresponding
to the eigenvalue .
To solve the SturmLiouville problem means to find all pairs of eigenvalues and eigenfunctions.
Example: 1, 0, 1, 0, , 1, 0, 1, and 0
0, 0 0, 0 (5.11.2)
If 0, then and only the trivial solution ( 0) satisfies
the b.c. (boundary conditions) in Eq. (5.11.2)
If 0, then cos sin and it can satisfy the b.c. in Eq. (5.11.2) if
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 37
5.11. Application of the generalized Fourier series: The SturmLiouville problem. Solving the heat
conduction equation by the separation of variables
0, , 0, 1, 2,
but 0 gives us again the trivial solution and should be excluded.
Thus, the pairs of eigenvalues and eigenfunctions for the problem (5.11.2) are
, sin , 1,2,3, (5.11.3)
Note: Coefficient in the eigenfunction is an arbitrary nonzero value. Thus, every
eigenfunction for a given is nonunique (This is similar to the nonuniqueness of eigenvectors
for a given eigenvalue in the matrix eigenvalue problem).
The SturmLiouville problem is closely related to the generalized Fourier expansions, since
eigenfunctions, corresponding to different , form orthogonal systems of functions with respect
to the weight as stated by the following theorem:
Theorem:
Let's assume that , , , and in the SturmLiouville problem (5.11.1) are real
valued and continuous and 0 in the interval .
Let and be eigenfunctions that correspond to different eigenvalues and
( ).
Then and are orthogonal on with respect to the weight , i.e.
0,
, (5.11.4)
,
Proof: See Kreyszig, p. 502.
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 38
5.11. Application of the generalized Fourier series: The SturmLiouville problem. Solving the heat
conduction equation by the separation of variables
if 0,thenthefollowingsingular SturmLiouville problemcanbesolved:
0,
(5.11.5)
0
If 0,thenthefollowingsingular SturmLiouville problemcanbesolved:
0,
(5.11.6)
0
if , then the following SturmLiouville problem with periodic boundary conditions
can be solved:
0, (5.11.7)
,
Note: Eigenfunctions of problems (5.11.5)(5.11.7) also form orthogonal systems of functions
(see the proof in Kreyszig, p. 502).
ApplicationoftheFourierexpansionsforsolvingPDEswithseparatingvariables
Example: Let's consider the onedimensional unsteady heat conduction problem:
(5.11.8)
(5.11.9)
0, : Initial conditions
(5.11.10)
/ / 0: Boundary conditions thermally insulated boundaries
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 39
5.11. Application of the generalized Fourier series: The SturmLiouville problem. Solving the heat
conduction equation by the separation of variables

, (5.11.11)

Let's try to represent the solution in the form


, (5.11.12)
and substitute it into Eq. (5.11.11). Then we obtain

or


The LHS depends only on , the RHS depends only on . The identity is possible only if

, (5.11.13)

Let's consider the second equation together with the boundary conditions given by Eq. (5.11.10)
0, 0 0, 0
where / . This is the SturmLiouville problem: If 0, then cos
sin and it can satisfy the b.c. if 0, , 0,1,2, . The eigenfunctions ,
corresponding to eigenvalues , form an orthogonal system:
, cos
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 40
5.11. Application of the generalized Fourier series: The SturmLiouville problem. Solving the heat
conduction equation by the separation of variables

Then a particular solution of the first Eq. in (5.11.13) for a given takes the form:
/ exp
Now any is the solution of Eq. (5.11.11) that satisfies the boundary conditions
(5.11.10). Due to linearity of the original equation (5.11.11) and boundary conditions, the
solution can be represented in the form:

, cos (5.11.14)
2 2
Coefficients should be found based on the initial condition given by Eq. (5.11.9):

0, cos (5.11.15)
2
Since the RHS of Eq. (5.11.15) is the Fourier series of the function , can be determined
as regular Fourier coefficients. Note that is nonperiodic, but can be naturally extended to
an even periodic function with the period 2 . It explains the absence of sinterms in (5.11.15).
Note: Depending on , , and , the eigenfunctions of the SturmLiouville problem
can be not only trigonometric functions, but also many other functions. In particular, solutions
of the boundary value problems for PDEs with separating variables in domains with axial
symmetry result in eigenfunctions in the form of the Legendre polynomials.
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 41
5.12.ApplicationoftheFourierseries:Frequencyspectrumanalysis
Let's assume that we represented some periodic function in the form of the Fourier series:
1
, (5.12.1)

if the independent variable is time, , then we call a signal in the time domain, and
is the angular frequency.
Spectrum (spectral) analysis, also referred to as frequency domain analysis or spectral density
estimation, is the technical process of decomposing a complex signal into simpler parts. Any
process that quantifies the various amounts (e.g. amplitudes, energies, powers, intensities, or
phases), versus frequency can be called spectrum analysis.
The Fourier series gives us a signal as a combination of simpler parts. Every part has the form of
a harmonic oscillation. We can look at as at measure of importance of individual harmonic
oscillation of given frequency in the signal.
Every complex amplitude can be represented in the form including amplitude and phase
:

0: , , tan
2 2
(5.12.2)
0: , , tan
2 2
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 42
5.12.ApplicationoftheFourierseries:Frequencyspectrumanalysis
Spectral analysis implies that we plot the amplitude spectrum, versus , and phase
spectrum, versus .
Amplitude spectrum:

3
Phase spectrum:

Note 1: Amplitude spectrum is the even function, phase spectrum is the odd function.
Note 2: Spectrum Fourier analysis of periodic functions results in discrete spectrums.
Note 3: Amplitude can be thought as a measure of representativeness of oscillations with
given frequency in the signal.
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 43
5.12.ApplicationoftheFourierseries:Frequencyspectrumanalysis
Example: Periodic rectangular wave of period 2 2. What happens if ?
Signal Amplitudespectrum

Thefunctioniseven, 0,
/2, , 0

Wecananalyzetheamplitude
spectrumintermsof
1 2

Nonperiodicfunction
obtainedat
Note 1: The difference between neighbor frequencies depends on the halfperiod :
/ , are proportional to .
Note 2: In the limit of the nonperiodic function, when , 0 and the discrete
spectrum evolves into the continuous one.
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 44
5.13.Fouriertransform
With the Fourier series we can study properties of periodic functions, or periodic extensions of
nonperiodic functions given at a finite interval. The motivation of the Fourier transform is to
extend the developed approach to nonperiodic functions and, in particular, to nonperiodic
functions that can attain nonzero values at arbitrary .
We limit our consideration by absolutely integrable functions , i.e. such functions for which
the following integral exists

(5.13.1)

It means that the area between the plot of | | and axis is finite. It is possible only if
| | 0 when . But this last condition is insufficient. | |

Examples: /2 /2
1. exp | | is absolutely integrable function.
2. 1/| | is not absolutely integrable function.
In order to find equations of the Fourier transform, let's consider some absolutely integrable
function , choose some arbitrary 0, and then
1. Introduce a new function given in a finite interval at /2 /2.
2. Introduce periodic extension of .
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 45
5.13.Fouriertransform
The periodic extension can be expanded into the Fourier series. According to the Dirichlet
theorem, in the interval /2 /2, the Fourier series for coincides with with
exception of points of discontinuities, so we can write
/
1
, ,
(5.13.2)
2 2
/
whereweusethecomplexrepresentationoftheFourierseriesandnotation 2 / .
Nowlet'sseehowEq.(5.13.2)evolveswhen .
The complex amplitude corresponds to the oscillation with angular frequency
2 / . These frequencies form a discrete spectrum. The difference between neighbor
frequencies 2 / 0 when . Thus, the limit of nonperiodic
function (obtained at ) is characterized by a continuous spectrum when the angular
frequency can attain any real value. Then let's rewrite Eq. (5.13.2) as follows
/

,
2
/
Notethat ~ 0 when .Thenextstepistointroduce 2 / :
/
1
(5.13.3)
2
/
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 46
5.13.Fouriertransform
Then
1
(5.13.4)
2
Note that so far Eqs. (5.13.3) and (5.13.4) are equivalent to Eq. (5.13.2), but they allow us to
consider the limit . In this limit, the RHS of Eq. (5.13.4) becomes the Riemann integral
sum, i.e. it transforms to the integral with the integrand . Thus, in the limit
1
(5.13.5)
2

1
(5.13.6)
2
If exists, then it is called the Fourier (integral) transform of , representation of in
the form of Eq. (5.13.6) is called the inverse Fourier transform or Fourier integral of .
Note1:TheFouriertransformisthecomplexvaluedfunction.
Note 2: The Fourier transform can be formulated in many forms that usually differ from each
other by the choice of other variables instead of and different coefficients before the integrals
instead of 1/ 2 . It can be also formulated in purely real form. The different forms of the
Fourier transform will be considered in section 5.14.
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 47
5.13.Fouriertransform
Example:
1/ 0 1/

0 0,
0
Let'scalculatetheFouriertransform:
1 1 1
2 2 2

Let'susetheEulerformula cos sin :

1
cos sin
2
where
1 1 sin
cos sin
2 2 2

1 1 1 cos 2sin /2
sin cos
2 2 2 2
Notethat istheevenfunctionand istheoddfunction.
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 48
5.13.Fouriertransform
The conditions when can be represented by the Fourier integral are given by the following:
Theorem:Fourierinversetheorem
Let'sconsiderafunction ,whichsatisfiesthefollowingconditions:
1. isabsolutelyintegrable.
2. isapiecewisecontinuousineveryfiniteinterval.
3. hasfinitelefthand, 0 ,andrighthand, 0 ,derivativesineverypoint,i.e.
0 0
0 lim , 0 lim
, ,
Then
1.TheFouriertransformexists
1
2
2.Ineverypoint ,where iscontinuous, canberepresentedbytheFourierintegral,
1
2
3.Atapoint ,where isdiscontinuous,
1 1
0 0
2 2
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 49
5.13.Fouriertransform
Note 1. Existence of lefthand, 0 , and righthand, 0 , derivatives implies
existence of lefthand, 0 , and righthand, 0 , limits in every point.
Note 2. The conditions of the Fourier inverse theorem are similar to the Dirichlet conditions, but
more strict. In the Fourier inverse theorem, it is required additionally
a. Existence of left and righthand side derivatives (only existence of left and righthand
limits is requited by the Dirichlet conditions).
b. The function should be absolutely integrable (Note that any nonzero periodic
functions is not absolutely integrable).

ME501,MechanicalEngineeringAnalysisI,AlexeyVolkov 50
5.14.VariousformsoftheFouriertransform
FouriertransformandFourierintegral
1 1
, (5.14.1)
2 2
are used in many different forms. A few of such forms are considered below. We will mark these
different forms of the Fourier transform by the subscript "*" in order to distinguish them from
our basic form given by Eq. (5.14.1).
1.Formbasedonthe"true"frequency /
Then 2 , 2 and
1 2
,
2 2
Nowlet'sintroduce

2 (5.14.2a)

Then

2 (5.14.2b)

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 51
5.14.VariousformsoftheFouriertransform
2.NonsymmetricformI
Let'sintroduce
1

2 (5.14.3a)
2
Then

(5.14.3b)

3.NonsymmetricformII
Let'sintroduce
2 1
(5.14.4a)

Then
1
(5.14.4b)
2

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 52
5.14.VariousformsoftheFouriertransform
4.Twocomponentform.RealformoftheFourierintegral
Let'susetheEulerformula cos sin .ThenEq.(5.14.4a)reducesto

1
cos sin , (5.14.5a)

where
1 1
cos , sin (5.14.5b)

Note that , , i.e. and are always even and odd


functions, correspondingly. Now let's use the Euler formula in Eq. (5.14.4b):
1
cos sin
2

1
cos cos sin sin
2
Sinceboth cos and sin areoddfunctions,finallywehave
1
cos sin (5.14.5c)
2
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 53
5.14.VariousformsoftheFouriertransform
Butnowweseethattheintegrand, cos sin ,istheevenfunction,sowecan
rewritethelastequationas

cos sin (5.14.5d)

Note: Eqs. (5.14.5b)(5.14.5d) allows us to formulate the Fourier integral in the purely real form
(without complex numbers).
5. Fourier transform for even and odd functions
If is even, then 0 and cos is the even function, so we can write
2
cos , cos (5.14.6)

The Fourier integral in the form of Eq. (5.14.6) is called the Fourier cosine integral.
If is odd, then 0 and sin is the even function, so we can write
2
sin , sin (5.14.7)

The Fourier integral in the form of Eq. (5.14.7) is called the Fourier sine integral.

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 54
5.14.VariousformsoftheFouriertransform
6. Fourier transform for halfrange functions given at
The case of functions given only in the halfrange, e.g., only at 0, is an important particular
case, since signals (function of time, ) are recorded and studied only starting at some initial
time.
If values of are given only at 0, then, in order to introduce its Fourier transform and
integral, we have at least three options:
1. To assume that 0 at 0. It gives us the Fourier integral in the general form of Eqs.
(5.14.5b) and (5.14.5d).
2. To extend for 0 evenly, i.e. to assume that at 0. It gives us the
representation of in the form of the Fourier cosine integral, Eq. (5.14.6).
3. To extend for 0 oddly, i.e. to assume that at 0. It gives us the
representation of in the form of the Fourier sine integral, Eq. (5.14.7).

Note: Different extensions of at 0 produce different Fourier transforms, but the Fourier
integral in all these cases will have the same values at all 0 except points where is
discontinuous. It is guaranteed by the Fourier inverse theorem. In particular, the value of the
Fourier integral at 0 can be different depending on whether the extended function has or
does not have a discontinuity in this point.

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 55
5.15.ApplicationsoftheFouriertransform
Like the Fourier series, the Fourier transform is used in order to
1. Solve initial and boundary value problems for ODEs and PDEs.
2. Perform signal processing (Filtering, etc.).
3. Perform spectrum (spectral) analysis.
Fourier transforms of many "basic" functions are tabulated like derivatives and antiderivatives
(see Kreyszig, pages 534536).
If the independent variable is the time, , then is often called the signal in the time
domain. The Fourier transform is considered as the image of the signal in the frequency
domain.
Example of signal processing:
Findtheimage Edittheimageinthe Restorethecorrectedsignalin
(Fouriertransform) frequencydomain thetimedomain(Fourierintegral)

The Fourier transform can be represented in the form



2
where and are the amplitude and phase

, tan
2
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 56
5.15.ApplicationsoftheFouriertransform
Thus, the general complex Fourier transform corresponds to two real spectra: Amplitude
spectrum and phase spectrum
If is the even function, then /2 is a realvalued function and only the
amplitude spectrum is of interest.
The amplitude can be thought as a measure of representativeness of oscillations with
given frequency in the signal.
Example: See http://en.wikipedia.org/wiki/Fourier_transform

Signal(oscillationoffrequency Amplitudespectrum(dominatedby
3 withvaryingmagnitude) theoscillationoffrequency 3)

3 5
Integrands

cos 2 3 exp DominantfrequencyUnderrepresentedfrequency


ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 57
5.16.DiscreteFouriertransform(DFT).FastFouriertransform(FFT)
In the majority of applications, the signal function is given in discrete or tabulated form
... +1point
: points
... forperiodicity

This is typical for signals obtained as results of measurements in physical experiments. We


cannot perform signal processing or spectrum analysis of such signal by directly applying the
Fourier transform, because calculation of the image

is possible only if we know for all . The Discrete Fourier transform (DFT) is a special
modification of the Fourier transform that can be applied to discrete (tabulated) signals.
Reductionofadiscretesignaltoastandardform
Let's consider signals with evenly spaced points: , / 1 .
Let's turn the signal into the periodic function, assuming that there is an additional point
, where . Such extended signal can be reduced to a
"standard" periodic discrete signal given in the interval 0 2 :
2
2
For the standard discrete signal, 2 / and .
Standard discrete signal is given by the (column) vector , ,, .
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 58
5.16.DiscreteFouriertransform(DFT).FastFouriertransform(FFT)
Interpolationbytrigonometricpolynomials
Since is given in a finite range 0 2 , we could represent it in the form of complex
Fourier series, Eq. (5.6.4), with 2 / 1:

We cannot define infinitely large number of complex amplitudes if we know values of in


only points. Instead, let's approximate with term complex trigonometric sum:

(5.16.1)
Hereweuse

The function in the RHS of Eq. (5.16.1) is called the complex trigonometric polynomial, since it
is an analog of "regular" polynomials in the form

Our goal is to find in Eq. (5.16.1) based on data in the vector , ,, . We will do
it, assuming that is the interpolation polynomial, i.e.

(5.16.2)

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 59
5.16.DiscreteFouriertransform(DFT).FastFouriertransform(FFT)
Combining Eqs. (5.16.2) and (5.16.3), one can write that

If we introduce , then the last equation can be rewritten as follows:

, 0, , 1 (5.16.3)
Eq. (5.16.3) is a linear system with respect to and can be rewritten in the matrix form as

, where

and square matrix has elements , . The solution this system is



The elements , of the inverse matrix are equal to (See Kreyszig, Sect. 11.9, p. 529)
1
, , where
( and are complex conjugate) and, thus
1 1
, (5.16.4)

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 60
5.16.DiscreteFouriertransform(DFT).FastFouriertransform(FFT)
Eq. (5.16.1) and (5.16.4) defines the trigonometric interpolation polynomial which is used
instead of trigonometric (Fourier) series for tabulated functions.
DiscreteFourierTransform(DFT)
We connected the Fourier transform with as (See slide 46): 2 / . In
the case of a tabulated function / 1/ and it is reasonable to introduce the
discrete Fourier transform (DFT) of as a (column) vector , ,, , where

(5.16.5)

The DFT can be also introduced in the matrix form


:DFT (5.16.6)

where the Fourier matrix , has elements , .


If we know the image of the signal, , and want to find the signal itself, we just need to resolve
the linear system given by Eq. (5.16.6) with respect to :
:InverseDFT (5.16.7)
where 1/ , / or (compare with Eq. (5.16.4)):
1
(5.16.8)

ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 61
5.16.DiscreteFouriertransform(DFT).FastFouriertransform(FFT)
Fourier transform of discrete (tabulated) data is a vector (table) , not function.
In applications, the number of points in the signal is usually large, 1. In this case the
direct application of Eq. (5.16.6) is inefficient, since it requires ~ 1 ~ arithmetic
operations. Direct calculations of DFT at large N is extremely lengthy operation!
FastFourierTransform(FFT)
Fast Fourier Transform (FFT) is a DFT at specific values of , where calculations are organized
in a special manner that allows one to reduce the number of arithmetic operations.
Usually in FFT, 2 . In this case the number of arithmetic operations can be reduced to
log , e. g. , 1000, 1000000, log 6900,more than 100fold acceleration!
The most popular algorithm of FFT is the CooleyTukey algorithm. This method (and the
general idea of an FFT) was popularized by a publication of J. W. Cooley and J. W. Tukey in
1965, but it was later discovered that those two authors had independently reinvented an
algorithm known to Carl Friedrich Gauss around 1805. He developed an FFTtype algorithm
to interpolate the orbit of asteroids Pallas and Juno from sample observations.
See details in
Kreyszig, Sect. 11.9, pp. 528532
https://en.wikipedia.org/wiki/Fast_Fourier_transform
https://en.wikipedia.org/wiki/Cooley%E2%80%93Tukey_FFT_algorithm
The majority of mathematical software has buildin capabilities for FFT. In MATLAB, FFT of
tabulated data X can be performed with function fft ( X ). See
https://www.mathworks.com/help/matlab/ref/fft.html?requestedDomain=www.mathworks.com
ME501,MechanicalEngineeringAnalysis,AlexeyVolkov 62