Fourier Series: An Example of Characteristic Behaviors

Concepts of primary interest:

Independent behavior
Completeness
Inner product, a gauge of common behavior
Orthogonality relation
Projection - projecting out a behavior coefficient
Fundamental frequency - relation to the period
Convergence
Rapid changes  greater amplitude for high frequencies
Parseval's identity
Relating series with different periods
Series expansions are good – why choose Fourier?
FS and the Vector Space of periodic functions
Application examples:
Anti-symmetric square wave
Anti-symmetric triangular wave
Oscillator with a periodic driving force
Summing series
Tools of the trade:
The Unit Circle for evaluating trig functions
Converting sums to integrals
Discovery:
Summing function to match a specified function
Give f(x) = the sum of a few terms with integer …

????? Change convergence of coefficients to NO FASTER THAN rather than

as fast as

Contact: [email protected]
See problem 41 on page 79+ ***************************
The value co as defined in the section is equivalent to ½ ao as defined
everywhere else in the known universe. Note that this 'distinct' notation also
leads to a change in the form of Parseval's equality. The rule will be to replace
co by ½ ao throughout to compare with other developments.

Introduction to Fourier Series: More haphazard history  It's just a story with
no basis in fact used to introduce a topic.

Fourier studied the problem of the vibrating string fixed at both ends
discovering that it has a set of characteristic vibration patterns in which the string
oscillates in the shape of any multiple of half cycles of a sinusoid with an
amplitude that oscillates sinusoidally in time at frequencies that are proportional to
that multiple of half cycles (m = m o). The result was so appealing that Fourier
declared that these were the only shapes in which one could expect to find the
string. A musician objected immediately claiming that his violin string had a
triangular shape when he plucked it. Stung by the prospect of being proven wrong,
Fourier boldly proposed that all shapes assumed by the violin string could be
represented as a sum of his sinusoids. Indeed any function can be represented as a
sum of sinusoids with just a few restrictions. The allowed sine and cosine
functions are periodic, and each has zero average value. Restrict the issue to
periodic functions or functions defined over a finite interval which can be assumed
to be extended periodically beyond that interval. Toss in an additive constant to
handle the average value, and it works. It works as long as the correct definition for
working is adopted.

Functions are also restricted in that they must be finite and continuous except for perhaps a finite
number of finite discontinuities. Some more exotic behaviors can be tolerated, but here the

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-2

 behavior assumed is finite, piecewise continuous and possessing only a finite number of
discontinuities in a period. See the Dirichlet conditions presented later for a more precise
statement of the requirements.

In the time-frequency realm, Fourier's conjecture takes the form: Any function
f(t) which is periodic with period T can be represented as a sum of a constant plus
sinusoidal terms with frequencies m  m  2 T  . The sinusoids are the sine and

cosine. The set of frequencies m ensure that each sinusoid completes an integer m
cycles in the time T, the period of f(t). The base frequency for the expansion is
0  1   2 T  which is called the fundamental frequency. This frequency is

determined by the period of the function to be expanded alone. The frequency

of every term in the expansion is an integer multiple (harmonic) of this
fundamental frequency. To be periodic with period T, each sinusoid must complete
an integer number of cycles in the time T. This condition sets the allowed
frequencies to the integer multiples of 2/T independent of any features of the
function other than its period.
 
f (t )  c0 1   am cos nt  
n 1
b
m 1
m sin mt  where m  m  2 T  . [FS.1]

Each term in the expansion represents a distinct behavior of a function with period
T. The first term c0 (1), a fancy way to write c0 , is the average value of the
function with the (1) indicating that it represents the constant behavior of the
function. The terms am cos mt  represent the independent or distinct ways in which
an even function of period T can wiggle, and the terms bm sin mt  represent the
independent or distinct ways in which an odd function of period T can wiggle. It is
crucial to recognize than each term represents an independent behavior. (The
constant c0 is replaced by ½ a0 in almost every other treatment of Fourier series.)
Recall the Cartesian representation of a particle's displacement:

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-3

 
r  x iˆ  y ˆj  z kˆ

Each term in the representation of r is an independent contribution to the
displacement. No amount of x iˆ can change the y or z coordinate of the particle.
The statement of this independence or orthogonality is in the relations:
iˆ  ˆj  0; kˆ  iˆ  0; ˆj  kˆ  0; iˆ  iˆ  1; ˆj  ˆj  1; kˆ  kˆ  1

The first three equations are orthogonality relations, and the last three
normalization relations. A displacement in the x direction is exactly like itself and
nothing like either of the other two possibilities. These (dot) inner product

relations provide the method to find the components of r .
 r  x  iˆ  r  iˆ   x iˆ  y 
  ˆj  z kˆ  x

 
Take the inner product of the x-direction and r to find the x-component of r .
This process is called projection. The orthogonality of the coordinate directions
insures that the projected components are measures of independent (displacement)
behaviors. The choice of unity normalization ( iˆ  iˆ  1; ˆj  ˆj  1; kˆ  kˆ  1 ) for the three
coordinate directions leads to the familiar result that:
2
r   x    y    z 
2 2 2

There is a relation between the norm of the object represented and the sum of the
squares of the expansion coefficients (components). This paragraph may seem a
little silly at this point. Review it after completing this chapter.
Fourier representations work the same way. Orthogonality is the statement that
each term is an independent wiggling of the function. The inner product takes the
form of an integral.
T 2
1
T T 2
g (t ) f (t )  g (t ) f (t ) dt (to be slightly modified later !)

The strange notation g (t ) f (t ) is chosen to represent a general inner product. The

symbolic representation is chosen to be unfamiliar forcing you to regard it as a new

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-4

process. The integral is sometimes designated as the overlap of g(t) and f(t) over
the interval. Each of the contributions: { c0 1;{ am cos mt  };{ bp sin  pt  }} is an

independent wiggling that has zero overlap with any other member of the set.
Therefore, the coefficient in the expansion for a member of the set
{1;{ cos mt  };{ sin  pt  }} can be projected out by multiplying the sum by the
member (behavior) associated with that coefficient and integrating over the range

 T 2 ,T 2  .

Fuzzyspeak; not a definition: An inner product gauges degree to which one entity embodies the
behavior of another entity of the same class. To what degree do two vectors point in the same
direction? To what degree do g(t) and f(t) wiggle together ? More about this choice for the inner
product for wiggling functions is to be presented in a meandering mind section below.

!!! PROJECTION --- HAS COSMIC SIGNIFICANCE !!!

PROJECTION: A process to isolate a coefficient or behavior in an entity. Multiply the entity and the
sum that represents it by a behavior and execute the inner product operation. Use orthogonality
relations. To find the coefficient am for cos mt  , multiply by cos mt  and compute the inner

product with the function f(t) and with its Fourier expansion.

Begin with 1. Find out how much constant behavior is in the function.
T 2 T 2  T 2  T 2
1 1 1 1
1 c0 dt   am 1 cos mt  dt   bm 1 sin mt  dt
T T 2 T T 2 T T 2 T T 2
1 f (t ) dt 
m1 m 1

The integrals of the sine and cosines are over an integral number of cycles; hence
they vanish. Wiggling like a sine or cosine for an integral number of cycles is
nothing like being constant. So:
T 2 T 2
1 1
T T 2 T T 2
c0  1 f (t ) dt  f (t ) dt  f (t ) [FS.2]

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-5

The constant part of the function is just its average value over the interval

 T 2 ,T 2  .

To identify the distinct wiggle contributions, the coefficients for each distinct
wiggle behavior are projected out of the sum by invoking the following
orthogonality relations:
T 2
1
sin  p t  cos mt  dt  0
T T 2  
T 2
1
sin  p t  sin mt  dt  ½  pm
T T 2  
[FS.3]
T 2
1
cos  p t  cos mt  dt  ½  pm
T T 2

The first relation is trivial as the product of sine and cosine is an odd function. The
later two relations are proven below, and a second set of proofs is a homework
problem. That is: proof by exhaustion follows.

The coefficients for the independent wiggle modes become:

T 2 T 2
2 2
T T 2 T T 2  
ap  cos  pt  f (t ) dt ; bp  sin  pt  f (t ) dt for integers p  1 [FS.4]

The factor of two arises because the average values of cos2  pt  and of sin 2  pt 

are one half (while the average value of 1 is 1).

Do not focus on the equations to compute the expansion coefficients. Focus on

the orthogonality relations and the projection process to isolate the
coefficients. That is: use the orthogonality relations to essentially repeat the
derivation of the equations above each time you use them. It sounds extreme,
but the reward is that you command a powerful technique that applies far

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-6

beyond the sine and cosine series expansions.

Exercise: Find the coefficient bp by multiplying the series representation by

 
sin  pt  and integrating over the range  T , T . Use the orthogonality relations.
2 2

The proofs of the orthogonality relations above follow from the identities for
products of sines and of cosines. Recalling that m  m  2 T   m0 :

2
sin( x)sin( y )  1 cos( x  y )  cos( x  y )  

sin  pt  sin mt   1  cos  ( p  m)0t   cos  ( p  m)0t 

2

* Treat the cases p = m and p  m separately where m = m o

T 2 T 2
1 1
p  m:  sin  pt  sin mt  dt  1 cos  ( p  m) t   cos  ( p  m) t  dt
T T 2 2 
0 0 
T T 2

1  sin  ( p  m)0t  sin  ( p  m)0t  

T 2

     0 if pm
2T  ( p  m)0 ( p  m)0  T 2

T 2 T 2
1 1
 sin mt  sin mt  dt  1 cos  0  cos  (2m) t  dt
T T 2 2 
p=m: 0 
T T 2

 sin  (2m)0t  
T 2
1 1
 t    if pm
2T  (2m)0  T 2 2

  
    
where sin m T 2   sin  m  2 T  T 2   sin  m   0 for all integers m is used.

Hence:
T 2

 T   sin  t  sin  t  dt  1 2 

1
T 2
p m pm

1  sin  ( p  m)0t  sin  ( p  m)0t  

T 2

Exercise: Examine the line:     0 . What is

2T  ( p  m)0 ( p  m)0  T 2

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-7

the problem that arises for the p = m case? What is cos[(p - m)t) in the p = m
case?

Exercise: Use the change of variable u = 0t and the trigonometric identity:

2
cos( x) cos( y )  1 cos( x  y )  cos( x  y ) 

cos  p t  cos mt   1 cos  ( p  m)0t   cos  ( p  m)0t 

2
T 2

to show that:  T   cos  t  cos  t  dt  ½ 

1
T 2
p m pm

T 2

Exercise: Repeat the proof that:  T   sin  t  sin  t  dt  ½ 

1
T 2
p m pm using the

change of variable u = 0t.

The Orthogonality Relations for the Fourier Expansion Set

Orthogonality relations are crucial to the application of Fourier series methods.

They are used in 70% of all Fourier problems and applications. Highlight your every
application of these relations. Try to prepare a prose summary of the manner in which the relations
were used to solve each problem. The Fourier expansion set has members
{1;{…, cos mt  ,…};{…, sin  pt  ,…}}.
This set of functions is mutually orthogonal [with a typical member Fi(t)] which
means that the inner product of any two distinct members of the set vanishes:
1
T 2
 0 for i  j
   
*

T T 2 
 F t 
 F t dt   General Orthogonality Relations
 N j for i  j
i j

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-8

The inner product is zero unless the two functions are the same in which case the
result is a non-zero value. This equation is shorthand for the following relations:
[FS.5]
T 2 T 2 T 2
1 1 1
 1 sin mt  dt  0;   
sin  p t cos mt  dt  0;  1 cos  t  dt  0
m
T T 2
T T 2
T T 2

T 2 T 2 T 2
1 1 1
 1 1 dt  1; T   
sin  p t sin mt  dt  1  pm ;   
cos  p t cos mt  dt  1  pm
T T 2 T 2
2
T T 2
2

The relations state that a member of the Fourier expansion set wiggles like itself
and not at all like any other member of the set over the domain [- ½ T, ½ T].
The Kronecker delta ij is a shorthand for 0 if i  j and 1 if i = j.

Inner Products and the Meandering Mind: (This section is not strictly required
to understand the topic of expansions and othogonality, but the ideas may provide
the key to a more global appreciation of the topic.)
The goal is to motivate the definition of the inner product as:
T 2
1
T T 2
g (t ) f (t )  g (t ) f (t ) dt [FS.6]

 
Backtracking to 3D vectors, A  B  Ax Bx  Ay By  Az Bz which becomes more positive
for each component pair that points in the same direction (has the same sign). As
vectors point, this inner (dot) product grows for common pointing (behavior). The
inner product gauges the common behavior of two entities of the same class. How
can this concept be generalized to the inner product of two functions?

Start with a simple case and work up to the real problem. A function assumes a
value at each point in the interval. To begin, represent the function by its values at
N evenly spaced values of its argument. Let tm   T 2   T 2 N  m T N for m =    

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-9

{1,2,…,N} which selects the midpoints of the N equal width bins. The functions
g(t) and f(t) are each represented by N values.
g (t )   g (t1 ), g (t2 ),..., g (t N ) ; f (t )   f (t1 ), f (t2 ),..., f (t N )

Following the expression for the dot product of two displacement vectors, the inner
product of the two functions proposed is:
g (t ) f (t )  g (t1 ) f (t1 )  g (t2 ) f (t2 )  ...  g (t N ) f (t N )   g (tm ) f (tm )
m

The sum of the products of the corresponding values at positions behaviors. Not
a bad start, but to get a better picture, the limit N   is desired which requires
division by N to keep the expression finite.

 N  g (t ) f (t )  g (t ) f (t )  ...  g (t
g (t ) f (t )  1 1 1 2 2 N
m
 N
) f (t N )    g (tm ) f (tm ) 1

The integral of a function is the limit of the sum of terms of that function times an
infinitesimal step of the integration variable. A step from one bin to the next is
t  T so multiplication and division by T is in order. (See Converting Sums to
N
Integrals in the Tools of the Trade section.)

   g (t ) f (t ) T N    1T   g (t ) f (t )  t    1T   g (t ) f (t )dt 
g (t ) f (t )  1
T m
m m
m
m m

This prescription is almost great! The inner product of something with itself
should always be positive. Something should positively behave like itself. To
ensure this feature, a final tweak is made to the definition.

T /2
g (t ) f (t )  1T     g (t ) 
T / 2
*
f (t )dt [FS.7]

The * represents complex conjugation. This definition is correct for real valued
functions and for complex valued ones. Correct is a slippery concept. Many
definitions of the inner product can be correct as long as each properly gauges the
common behavior of the functions. For example, the scale factor is arbitrary. As

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-10

long as the rule is consistently applied, the definition could be modified to be

 1 2T  T / 2  g (t ) * f (t )dt or even just T / 2  g (t ) * f (t )dt .

T / 2 T / 2
Be alert to changes in

notation and normalizations as you review different references.

The goal has been to motivate the choice of the overlap integral as the definition of
the inner product. If you have been paying attention, you should not be buying it!
The inner product is about gauging common behavior. The behaviors under
discussion are remaining constant and wiggling which are represented by the
Fourier coefficients c0 ,..., am ,... ,..., bp ,... which would naturally lead to the inner

product definition (which is similar to the sum of the products of the corresponding
components):

g f   1T   g (t )* f (t )dt  c*g 0c f 0  1 2  a*gm a fm  1 2  bgp

*
b fp [FS.8]
m p

The factors of one-half are an annoying penalty for choosing the expansion set
{1;{ cos mt  };{ sin  pt  }} with the uneven mean square value (normalization).

 cos  t   sin  t 
2 2
12  1; 1 ; 1
m 2 m 2

Other than that, life is wonderful! The two definitions agree. It takes a little
perspiration to establish the equality, but they do agree. The name of some long-
dead mathematician should be tied to this equality. Failing to find a clearly
culpable individual, Parseval is to be given credit for the result. The integral
definition is in the coordinate representation in which the behavior of a function
is to take on a value at a point. The equivalent expression just above is in the
Fourier or frequency-space representation in which the set of certain wiggle
mode behaviors provides a basis for representing functions. As long as the
functions f(t) and g(t) are periodic with period T, these modes provide a complete
set of the possible behaviors for well-behaved functions with period T. In either of

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-11

the representations, the inner product gauges the degree to which the functions
share behavior. The coordinate and frequency representations both provide
complete descriptions of the problem. An engineer studying string clearances in
the design of a violin might choose the coordinate space representation for the
violin string. A musician playing the instrument might be more directly interested
in the tone, the frequency representation.

Parseval’s Equality Notation: in other references expect:

 
g f   1T   ½T g (t )* f (t )dt  1 4 a*g 0 a f 0  1 2  a*gm a fm  1 2  bgp
½T
*
b fp
m 1 p 1

Recall the average value of the function co is to be replaced by ½ ao to agree

with the notation more commonly used.
½T ½T
c0  ½  a0  T1  (1) f (t )dt  T1  f (t )dt  f (t ) [FS.9]
½T ½T

Sample Calculations: Fourier series for square and triangular waves

Example 1: Square waves come in several forms. As an example, an odd square

wave with period T stepping from -1 to +1 is presented.
f(t)

-T -T/2 +T/2 T

-1

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-12

The square wave f(t) has a period of T and is extended beyond the base period to
T
display the discontinuities at each multiple of /2. The function is represented as:
 1 for 0  t  T
 2
f (t )   [FS.10]
T
1 for  2  t  0

Clearly, the function is odd and has zero average value. The Fourier series only
includes terms with the behaviors intrinsic to the function. Only the sine terms
match the zero average value and odd function behaviors. Evaluating the integrals
to find c0 and the {… ap…} yields zero contributions. The expansion has the form:
  
f (t )  c0 1   am cos mt  
m 1
 bm sin mt  
m 1
b
m 1
m sin mt  [FS.11]

Using the symmetry, the expression for bP reduces to:

2
T 2
2  0 T 2
 4T2
bp    
sin  p t f (t ) dt   sin   p
t  ( 1) dt   sin   p
t  ( 1) dt    sin  p  2 T t  f (t ) dt
T T 2
T  T 2 0  T 0

Invoking standard practice, the change of variable u  p  2 T  t is elected which

yields dt = T/(2p) du. [The standard choice is the argument of the most complicated
function in the integrand. In this case, either u  p  2 T  t or v   2 T  t is a

good choice. In all cases, a dimensionless variable should be chosen.]

p
4 T   2  p  2 
bp     sin u  du      cos(u )  0    1  cos( p ) 
 
T  2 p  0  p   p 

Cosine of even multiples of  is one; of odd multiples minus one. Hence:

FOURIER SERIES FOR THE ANTI-SYMMETRIC SQUARE WAVE:

 4 
  p  for p odd 4 
 1   2 t 
bp    or f (t ) 

  2m  1  sin  2m  1 T 
 0 for p even m 0

The inverse p dependence means that the coefficients vanish only slowly in the
high frequency (large p) limit. High frequencies (fast wiggles) are necessary to

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-13

follow the very rapid changes at the discontinuities. The factor (2 m + 1) is a
standard representation for the odd integers.

Exercise: Sketch the sines on a plot of the square wave. Use the sketches to
explain the absence of the even p terms and the 1/p dependence of the coefficients
for the odd p terms.

Plotting the first three terms:

Plot[(4/Pi) * ( Sin[x] + Sin[3x]/3 + Sin[5 x]/5 ),{x,-1.4 Pi,1.4 Pi}]

0.5

-4 -2 2 4

-0.5

-1

The square wave is evident with only the first three terms. The series expansion
does miss in the region around the discontinuity, but that is expected. A
discontinuity is a rapid change that can only be represented by rapidly changing
functions, high-frequency sinusoids. At the discontinuity, the series sums to zero,
the average of the limits of the function as the discontinuity is approached from
smaller values of the argument and from larger values. Examining the first nine
terms of the expansion does not alter these observations. The inclusion of the
higher frequencies reproduces the jump more faithfully. One potentially
disturbing point is noted. While the miss region near the discontinuity is shrinking

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-14

in extent, the magnitude of the overshoot just before it is not. It remains at about
9% of the discontinuity for sums of any finite number of terms. This behavior is
the Gibbs Overshoot Phenomenon (after the American physicist J.W. Gibbs who
studied the problem after Albert Michelson brought it to his attention). The
overshoot is discussed further when convergence properties are covered in a later
section.

Plot with first nine non-zero terms (mmax = 17):

Plot[(4/Pi) * ( Sin[x] + Sin[3x]/3 + Sin[5 x]/5 + Sin[7 x]/7 + Sin[9 x]/9

+
Sin[11 x]/11 +Sin[13 x]/13+Sin[15 x]/15+Sin[17 x]/17), {x,-1.2 Pi,1.2
Pi}]

0.5

-3 -2 -1 1 2 3

-0.5

-1

plot through m = 17

Plot[(4/Pi) * Sum[ Sin[m x]/m, {m,1,41,2}],{x,-1.4 Pi,1.4 Pi}]

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-15

 plot through m = 41
Why were the coefficients of even index integer terms zero? Draw a few pictures.
Draw one period of the square wave; add a sketch of sin  p  2 T   for p = 1.

Sketch the product of the functions. Repeat using successive values of p. For p
even, sin  p  2 T   is odd about the points -T/4 and T/4 while the square wave is

even about those points which leads to zero overlap or inner product. Being even
or odd about a value of the argument is a distinct characteristic behavior.

Functions lacking distinct even or odd behavior can be expressed as a sum of an even function
and an odd function. Choosing even and odd about x = a:

f (a   x  a )  f (a   x  a ) f (a   x  a )  f (a   x  a )
f ( x)    f even  f odd
2 2

Example 2: The second example chosen is a triangular wave. This function is

smoother than the square wave as it has no discontinuities. It does have periodic
discontinuities in its first derivative. Be alert to the differences associated with
expanding a smoother function.

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-16

 f(t) 2

-T -3T/4 -T/2 -T/4 +T/4 +T/2 +3T/4 T

-1

-2
T T
base period: - //2 < t < //2
 2  4t for T  t  T
 T 4 2
 4t
f (t )   for  T  t  T [FS.12]
T 4 4

 2  4 t for  T  t   T
 T 2 4

Once again, the function is odd with average value zero.

 
 2 t 
f (t )   bm sin mt  
m1
b
m1
m sin  m
 T 
where

 
T 2 T 4
2 2
bp   sin  p t  f (t ) dt   sin  p  2 T  t  2  4t T dt 
T T 2
T T 2

   
T 4 T 2
2 2
 sin  p  2 T  t  4t T dt   sin  p  2 T  t  2  4t T dt
T T 4 TT4

The integrations necessary to compute the Fourier coefficients are not

extraordinarily difficult, but the process does become tedious. The tools required
to complete the problem are integration by parts and patience.

Integration by parts is based on the identity for the differential of a

product: d (uv)  u dv  v du which can be applied to
d (u n cos[u ])  u n   sin[u ]  n u n1 cos[u ] or u n sin[u ]  d (u n cos[u ])  n u n1 cos[u ] .

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-17

 This result provides the tool to compute

 
p
T 4 n1 2
 t sin  p  2 T  t  dt  T 2 p  u sin u  du
n n

T 4  p
2

 
p
n 1  2p 
 T    u cos u  p   n u n1 cos u  du 
n 2
2 p  
2  p 
2

For our case of interest:

 
p
T 4 2 2
 t sin  p  2 T  t  dt  T 2 p  u sin u  du
T 4  p
2

 2 p 
p
2  p 2 
 T    u cos u  p   cos u  du 
2

 
2  p
2


  T 2 p    p cos     sin     sin  

2
p p  p
 2 2 2 

  p 1 

 
 2 
 2 ( 
2
1) for p odd
 T 
2 p
 p  1 p / 2 for p even

A miracle occurs, and the answer appears. Some of the steps toward that miracle
T
include recognizing that the triangle wave is symmetric about t = /4. The
m 2 t 
functions sin  share this property if m is odd, but not if m is even. The
 T 
Fourier expansion of the anti-symmetric triangular wave under study contains only
sine terms with frequencies that are odd multiples of the fundamental.
m 2 t
Exercise: Sketch the functions sin  
form m = 1, 2, 3 and 4 on a sketch of the
 T 
T
triangular wave for the region 0 < t < /2. For which m values is the integral of the
T
product of the sine and triangular wave non-zero over 0< t< /2?

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-18

 T 2 T 4
2 2
bp   sin  pt  f (t ) dt   4  0 sin  pt  f (t ) dt for p odd
T T 2 T
T 4 T 4
2 4t 32 p 2
bp   4  0 sin  pt  dt  2 0 t sin  pt  dt where  p 
T T T T
p 2
8 8  p 2 p 2 t
bp  2 2
 p 0 u sin u  du 
 2 p2 
u cos u  sin u 
0 where u 
T
 p 1 
8
bp  2 2  1 2 
 
for p odd ; bp  0 for p even [FS.13]
 p

Exercise: Complete the detailed evaluation of the bp suggested in the four lines
above this exercise.

FOURIER SERIES FOR THE ANTI-SYMMETRIC TRIANGULAR WAVE:

8   2 t 1  2 t 1  2 t 
f (t )  sin 1   32 sin 3 T   52 sin 5 T   ....
 2
  T 
[FS.14]
 1
m
8 
 2 t 
  sin  2m  1
2 m 0  2m  1
2
 T 

Plot[{f1[t]- (2 t - t^2)},{t,0,2}]
Again, only the odd index terms appear. As before, for p even, sin  p  2 T   is

odd about the points -T/4 and T/4 while the triangular wave is even about those
points which leads to zero overlap or inner product. For large m, the expansion
coefficients vanish as 1 which is rather more rapid than the 1 for the square
m2 m
wave expansion coefficients. High frequencies are necessary to follow rapid
variations. The triangular wave is smoother (less jerky) than is the square wave
thus it has less high frequency (fast-wiggle) behavior than does the square wave.

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-19

 Conjecture: If a function f(t) is continuous and has continuous derivatives
through pth order, then its Fourier expansion coefficients will vanish no faster
than 1 for large m. The smoother the function is, the less prominent the
m
p 1

very high frequency contributions are. See problem 41.

Plot[(8/Pi^2) * ( Sin[x] - Sin[3x]/3^2 + Sin[5 x]/5^2 ),{x,-1.4 Pi,1.4 Pi}]

0.75

0.5

0.25

-4 -2 2 4
-0.25

-0.5

-0.75

The first three expansion terms provide a fairly faithful rendition of the triangular
wave. It wobbles a little, but it is not too shabby. There are no discontinuities so
the series appears to smoothly approach the desired function. In particular, no
overshoot phenomenon is evident. As a more complete comparison, the nine-term
representation is displayed below. It is just plain great. One can point at the
corners being rounded, but that is to be expected. A 17 0 wave corners 34 times in
the period of f(t). A corner is therefore expected to make its bend in a fraction 1/34
of the period. A plot with expanded detail and dimensionless period 2 verifies
2
this expectation qualitatively. The bend is expected to require about /34 or one-
4
fourth of the /17 range plotted. The important observation is that, as the number
of terms increases, the regions in which the sum is a poor representation of the

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-20

function shrink and the magnitude of the error in these regions also shrinks.
Contrast this with the Gibbs phenomenon at discontinuities.

Plot[(8/Pi^2) * ( Sin[x] - Sin[3x]/3^2 + Sin[5 x]/5^2 -Sin[7 x]/7^2 + Sin[9
x]/9^2 - Sin[11 x]/11^2 +Sin[13 x]/13^2-Sin[15 x]/15^2+Sin[17
x]/17^2),{x,-1.2 Pi,1.2 Pi}]

Plot through m = 17.

Plot[(8/Pi^2) * ( Sin[x] - Sin[3x]/3^2 + Sin[5 x]/5^2 -Sin[7 x]/7^2 + Sin[9 x]/9^2 -
Sin[11 x]/11^2 +Sin[13 x]/13^2-Sin[15 x]/15^2+Sin[17 x]/17^2), {x,Pi/2-
2*Pi/17,Pi/2 + 2*Pi/17}]

Plot through m = 17
Plot[(8/Pi^2) * Sum[ (-1)^m Sin[( 2 m + 1) x]/(( 2 m + 1)^2), {m,0,100}],
{x,Pi/2-2*Pi/17,Pi/2+2*Pi/17}]

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-21

 Plot through m = 201
Exercise: Give the Fourier coefficients for a triangular wave similar to the one
above, but with amplitude A rather than one.

Exercise: Give the Fourier coefficients for a function that is the sum of the unit
amplitude square wave plus the unit amplitude triangular wave discussed above.
How fast do the Fourier coefficients vanish for large m? Why is this dependence
expected?

The Symmetric Triangle Wave: f(t + T/4)

 1
m
8 
 2 t  
  sin  2m  1   2m  1 
 2
m 0  2m  1
2
 T 2

 1
m
8 
 2 (t  T / 4) 
sin  2m  1
 2 m
f (t  T / 4)  
 2m  1
2
0  T 

8   1
m
 2 t  
 2  sin  2m  1   2m  1 
 m0  2m  1 2
 T 2

8   1
m
 2 t 
 2  cos  2m  1 (1) m

 m0  2m  1 2
 T 

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-22

 8 
1  2 t 
f (t  T / 4)   cos  2m  1
 2
m 0  2m  1
2
 T 

The symmetric triangle wave has the same shape as the anti-symmetric triangle
wave so they have the same magnitudes of each frequency component. The
symmetric triangle wave only contains symmetric basis functions, and the anti-
symmetric wave contains only contains anti-symmetric basis functions. Both
functions have zero average value so no co (ao) term in its Fourier series.

Dirichlet Conditions: Under what conditions can a function be represented be a

Fourier series? The following is a statement of sufficient conditions.

If f(t) is a periodic function of period T, and if between -T/2 and T/2 ,the function is
single-valued, has a finite number of finite discontinuities, has a finite number of
T 2


2
maxima and minima, and if f (t ) dt is finite then the Fourier series as defined
T 2

above converges to f(t) at all points where f(t) is continuous, and, at discontinuities,
the series converges to the average of the left and right limiting values of the
function as it approaches the discontinuity; it converges to the mean value at a
discontinuity.

Convergence and Parseval's Relations

Several convergence properties of Fourier series have already been noted. If

the function has a discontinuity at t = tjump, the series converges to the average of
the limits of the function as t approaches tjump from smaller and from larger values.
To discuss he overall convergence of the series for f(t), the sequence of partial

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-23

sums is considered where sn(t) is the sum of the 2 n + 1 lowest frequency terms of
the series. It includes the constant term and all the sine and cosine term with
frequencies up to no. A proper partial sum includes all the terms in the Fourier
series with frequencies up to a maximum no and no terms with frequency higher
than that value. Never use an improper partial sum. (A later section on the Gibbs
phenomenon demonstrates the importance of using partial sums that consistently
include all terms up to a set maximum frequency.)
n n
sn (t )  c f 0 1   a fm cos mt  
m 1
b
m 1
fm sin mt  [FS.15]

If the series converges uniformly to the function f(t), we would have for any given 
there exists an N such that | f(t) – sN(t)| <  for all t  [-½ T, ½ T ]. That is, if and
only if life is great, the sequence of partial sums converges uniformly to f(t) in the
interval [-½ T, ½ T ]. If a series converges uniformly, then there is always an N
term (finite sum) that represents the infinite series to any accuracy desired.
Working in terms of the finite sum, interchanging the order of operations of
summing series and integrating (or differentiating) is clearly allowed. That would
be great, but uniform convergence just does not happen in all cases. At a
discontinuity, the series converges to the average of the left and right limits of the
function, and, in the Gibbs overshoot region, the series diverges from the value of
the function by 8.95% of the magnitude of the discontinuity for any large, but
finite sum of terms.
Fourier partial sum sequences do converge in the mean. For every  there
T /2

2
exists an N such that f (t )  sN (t ) dt   where the limits are just those for any
T / 2

complete period of f(t). This can be rewritten as:

 T /2

Lim   T / 2 f (t )  sN (t ) dt   0
2
[FS.16]
N 

For periodic functions f(t) with 'not too many' discontinuities and that are square

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-24

 T /2
integrable [ the integral  f (t ) dt is defined and finite ], the corresponding
2
T / 2

Fourier series converges in the mean to f(t).

The convergence of the sequence of partial sums to a discontinuous function
f(t) is weak. Consider the Gibbs 9% overshoot at discontinuities (of the
discontinuity). For the nth partial sum, the region in which the sum differs from the
function by an amount of order 9% of the discontinuity has the approximate extent
of T/2 n. Even for large N, there are tiny regions in which the partial sum
misrepresents the function by a finite amount. A Fourier series expansion
converges weakly to a discontinuous function.

Replace the function with its Fourier series: Parseval and beyond

Once you have the Fourier series for a function, that series can be used in the
T /2
place of the function. Use the orthogonality relations to consider  T / 2 f (t ) dt .
2

The summation indices are dummy labels. They may be relabeled at will. At
minimum, rename the labels in a product to ensure that the labels are distinct.
Represent f(t)* and f(t) using distinct summation index labels throughout.

T T *
2
 
2 
   

 f (t ) dt    c f 0   a fm cos mt    b fq sin qt    c f 0   a fp cos  pt    b fr sin r t   dt 
2

T T  m 1 q 1   p 1 r 1 
2 2

T
 2   

   f 0 f 0 f 0 
c * c  c * a fp cos  p   b fq sin qt  a * fm cos mt   ...more horrible stuff
 t    dt
T
2
p 1 m 1 q 1 

T
2
  
2
f (t ) dt (after envoking orthogonality )  T  c f 0  1  a fm  1  b fm 
2 2

2
2 2
T
2
 m 1 m 1 

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-25

where the factors of one half are an annoying consequence of the choice of
expansion set functions (or basis functions): {1;{ cos mt  };{ sin  pt  }} which

have uneven square integrals. This relation is Parseval's Equality:

T
2
  
2
 
2 2

2
f (t ) dt  T  c f 0  1 a fm  1 b fm 
2 2
T
2
 m 1 m 1 

Equally as cool and proven using the same techniques, the inner product relation
can be re-expressed:
T /2

 
1
T 
T / 2
 g (t ) 
*
f (t )dt  c*g 0c f 0  1 2  a*gm a fm  1 2  bgp
m
*
b fp
p
[FS.17]

The coordinate and the Fourier representations for the inner product agree at least
as long as our functions are well-behaved. Compare these relations with the
2    
expressions A  A  A  Ax Ax  Ay Ay  Az Az and A  B  Ax Bx  Ay By  Az Bz .

Notation: in other references expect:

g f  1 4 a*g 0a f 0  1 2  a*gma fm  1 2  bgp

*
b fp
m p

Recall the average value of the function c0 is to be replaced by 1

2
a0

to agree with notations commonly used.

T 2 T 2

c0  1  2 a0 
1
T T 2
1 f (t ) dt 
1
T T 2
f (t ) dt  f (t )

Fourier series represent the function:

Operations can be executed on the Fourier series for a function as well as on the
function itself. Examples: differentiation and integration. Properly, the

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-26

techniques are guaranteed valid only if the series converge uniformly. This
'requirement' is to be ignored. Start with the anti-symmetric triangular wave:
 1
m
8 
 2 t 
f (t ) 
 2   2m  1
m 0
2
sin  2m  1
 T 

Compute the series for the derivative of f(t) by differentiating the series term by
term. Sketch the derivative. The series represents a symmetric square wave.
Compare the frequency content with that of the anti-symmetric square wave.

 1
m
df 8 
 2   2 t  
 2
dt 
  2m  1
m 0

 T
 cos  2m  1
  T  

Exercise: Compare the frequency content of the symmetric square wave that
results from the term-by-term differentiation of the anti-symmetric triangular wave
with the frequency content of the anti-symmetric square wave.

Next, f(t) is to be integrated. Sketch the integral of triangular wave f(t).

 1
m
2 t ' 
t  t
8 
g (t )   f (t ') dt ' 
 2   2m  1  sin  2m  1
m 0
2
T 
dt '
0 0

 1
m

T 2  1  cos  2m  1 2T t  


8

 2   2m  1
m 0
3

Exercise: Sketch the integral of the triangular wave. Note that the coefficients of
its Fourier series vanish as m-3 for large m. This dependence suggests that the
function has a continuous first derivative. Does your sketch support this
conclusion?

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-27

Relating Fourier series derived for different periods

Given a Fourier series for a periodic function f(t) with period T1, the Fourier series
for g (t )  f ( TT12 t ) , a function with the same 'shape' that has been argument stretched

to have a period of T2 can been generated by scaling the independent variable in the
original series. The independent variable in the original series can be scaled in the
same manner to yield the Fourier series for g(t). [1 = 2/T1]
 
f (t )  c0   am cos m1t  
m 1
 bm sin m1t 
m 1

 
g (t )  c0   am cos m1
m 1 
 
T1
T2 t   
m 1
bm sin  m1  T  t 

T1
2 

The function g(t) has the a period of T2 , and the same amplitude as f(t).

Many references assume a period of 2. The result is a series of the form:
 
f ( x)  c0   am cos m x  
m 1
 bm sin m x 
m 1
period 2

T1
The series for the same shape and a period of T is generated by replacing x by T2 t

or 2 t . Mathematica assumes a range -½ to ½ or an initial period of T1 = 1. In a

T

T1
Mathematica generated series, replace the independent variable by T2 t  t
T .

Exercise: Given a periodic function f(t) with period T1, show that g (t )  f ( TT12 t ) is a

periodic function with a period of T2.

Physical Application: The Damped Driven Oscillator

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-28

Mechanical systems usually have an inertial element that 'resists acceleration'
with the associated second time derivative ensuring a 1
2 dependence of the


amplitude response to an excitation in the high frequency limit. This dependence

means that the series for the response converges more rapidly than that for the
excitation and that high frequency weirdness is usually not a practical issue.

The steady-state response of the oscillator to a sinusoidal driving force is assumed

to be known. See, for example, the differential equations portion of this handout
set. A condensed derivation follows. The damped driven mechanical oscillator
d 2x dx
obeys the differential equation m 2
b  k x  F (t ) which is rewritten as
dt dt
d 2x dx b k F(t)
2
 2
2
 02 x  f (t ) where = /2m , 0 = /m and f(t) = /m. 
dt dt

MEGACEPT: A linear system responds at the frequency at which it is driven.

If the system is driven at a single (pure) frequency, its steady-state response is
at that same pure frequency with at most a phase shift. The relative amplitude
of the response can depend on frequency. The single pure frequency functions
are sin(t ),cos((t ), e  it . Linear systems obey the superposition principle so the
response to any periodic function can be found by decomposing the driving
function into its Fourier (pure frequency) components, computing the
response to each component and then summing the responses to find the
response to the full periodic driving function. NOTE: The transient
behavior of a linear system has a frequency character determined by the
parameters of the system, not those of the driving force. The steady-state
solution is the long-time behavior after the transients have decayed. The
transient behavior satisfies the homogeneous equation.

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-29

Consider a single frequency driving function: f (t )  f 0 cos(t ) . The response is at
the same frequency, but with a phase shift so x (t )  R( ) f 0 cos(t   ) . The sign in
expression: (t is chosen to be negative as the response lags behind the drive

or cause. The manipulations are more compact if the clock for t is reset by + /.
This yields f (t )  f 0 cos(t   ) and x (t )  R( ) f 0 cos(t ) which states that the drive
leads the response by  which is the same as saying that the response lags the drive
by . Substituting these expressions into the differential equation:
USE: cos(t   )  cos( )cos(t )  sin( )sin(t )
R ( ) f 0   2 cos(t )  2 sin(t )  02 cos(t )   f 0  cos( ) cos(t )  sin( )sin(t ) 

The functions sin(t ) and cos((t ) are independent functions so the coefficients of
each must match on the two sides of the equation.
cos(t ) : R ( )  2  02   cos( ) and sin(t ) : R ( )  2   sin( )

After applying the two most powerful trigonometric identities [ sin 2 ( )  cos 2 ( )  1

and tan( )  sin( ) cos( ) , it follows that:

 2 
 
1 2
 02   2 
2
 R( )   D( )  2
and   tan 1  
   
2 2
0

The steady-state response to a driving function f (t )  f 0 cos(t ) is:

f0   2  
x (t )  R( ) f 0 cos(t   )  cos  t  tan 1  2 
 0    
 2  02    2 
2 2
2 

Note that the result above supports the conjecture that a linear system driven by a
pure frequency responds at that same frequency. The strength of the response does
depend on that frequency. Transients associated with 'turn-on' or start-up' require
that additional frequencies be added to the representation of f(t). Invoking

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-30

superposition for linear systems, the response for an arbitrary periodic driving
 
force, f (t )  c0 1   am cos mt    bm sin mt  ,
m 1 m 1

-1
is: [Recall that: f(t) = m F(t).]

c0 
am cos mt   m  
am sin mt   m 
x(t )   

  02    2m    02    2m 
2 2 2
2 2 2 2
0 m 1 m 1
m m
[FS.18]
 2m 
 m  tan 1  2
 0  m 
2

it looks hopeless. The result is nonetheless very useful, particularly when the
system has a sharp resonance that selects from all the possible m only a few near
0 with the response to frequencies further from resonance being negligible by
comparison.

Why expand in a Fourier series rather than in (say) a Taylor’s series? Series
expansions come in many varieties, and one must choose the type that best suits
the problem. The method of Frobenius revealed the awesome power of Taylor’s
series expansions. The individual expansion base functions (.. xn ..) are simpler
than sines and cosines. Why would you ever bother ? For many important
problems that you study, there exists a set of characteristic solutions. Functions
that just fit-right for that problem. You can recognize their appearance once you
know that the German for characteristic is eigenschaft and that the functions are
eigen-functions. For linear response problems such as the oscillator studied above,
the system responds at the frequency at which it is driven. A pure frequency drive
leads to a response at that same pure frequency. Obviously one wants a series that
represents the driving input in terms of its pure frequency components. Each pure
frequency component can be analyzed independently with no mixing with

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-31

components at other frequencies. The pure single frequency functions are sin(t),
cos(t), eit and e-it. These functions are the bases for Fourier expansions.
Fourier expansions are a leading choice for most linear problems for this reason.
A second strong indicator is that the function is periodic (or can be extended as a
periodic) function. The functions sin(t), cos(t), eit and e-it are periodic and
hence share a strong characteristic behavior with the function of interest.
Expansions in terms of basis functions that share behaviors with the function to be
expanded are more natural.

German: Eigenshaft => nature, virtue, feature, quality, property, attribute

character, characteristic, characteristic quality



Other Fourier Series: The Sine, Cosine and Complex Exponential series

Consider any function defined from 0 to S. That function can be extended as an

even or odd function to the interval -S to S and beyond by assuming that the
function is periodic with period 2 S. If the anti-symmetric extension is adopted,
the Fourier series will only have sine terms.
 S
4
f (t )   bm sin mt  where bm
m 1

2 S 0
sin  p  2 2 S  t  f (t ) dt

If the even extension is adopted, the function may be represented as

 S S
1 4
f (t )  c0   am cos mt  where c0   f (t ) dt and am  cos  m  2 2 S  t  f (t ) dt
m1 S0 2S 0

The penalty for using the odd extension is that the extended function will have
additional discontinuities at t = 0 and at t = -S and + S unless f(t) vanishes at these

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-32

points. The even extension does not introduce additional discontinuities. Recall
that discontinuities ensure that the coefficients vanish no faster than 1 m for large

m.
f(t)
even extension

-S S t

odd extension

Exercise: Consider the function f(t) that consists of two slope-matched parabolic
sections. Consider expanding the function as defined over the interval 0 < t < 2 as
an even or odd function to complete the definition over the period –2 < t< 2.
Would a sine of cosine series converge more rapidly? Assuming that the
coefficients would decrease as m- n for large m predict the value of n for a sine
series expansion and for a cosine series expansion.
1

0.8

0.6

[FS.19] 0.4

0.2

 1  (t  1) 2  2 t  t 2 for 0t 2 -2 -1.5 -1 -0.5 0.5 1 1.5 2

f (t )   -0.2

  1  (t  1)  2 t  t 2t  0
2 2
for -0.4

-0.6

-0.8

-1

Euler's identity and tons of perspiration can be used to develop the complex
Fourier series with the result that one just uses the same set of frequencies for the

 

 T , T  interval. f (t )  d exp i m 2 t  A typical coefficient c p is
 2 2 m
m
 T 

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-33

projected out of the sum by first multiplying both sides by exp   i p 2 T t 
   
     
T /2  T /2

 exp   i p 2 t  f (t ) dt  d  exp   i p 2 t  exp i m 2 t  dt  T d p

T / 2
 T  m
m
T / 2
 T   T 

   
T /2
or d p  1T exp   i p 2 t  f (t ) dt .
T / 2
 T 

NOTE: To project out the coefficient of a complex-valued function eit , the series is
pre-multiplied by the complex conjugate of that function,  eit   eit . The same
*

procedure was followed for the real functions. It just did not matter. The
procedure follows from the form of the inner product. The complex conjugate of
g(t) is required.

 T    g (t ) 
T /2
g (t ) f (t )  1
*
f (t )dt
T / 2

Exercise: Show that:

 T     
T /2
1 exp   i p 2 t  exp i m 2 t  dt   pm [FS.20]
T / 2
 T   T 

 T           
T /2 *
1 exp  i p 2 t  exp i m 2 t  dt   pm  exp ip 2 t  exp im 2 t 
T / 2
 T   T   T   T 

Complex representations are not necessary to describe classical physics. Certain

computational economies result from adopting complex notation, but care must be
taken to avoid interpreting or including the imaginary parts of the results. Few
problems arise as long as only linear operations are involved. Non-linear
operations however mix the real and imaginary (the physical and not-physical)
parts. Evil lurks in the shadows with the imaginary part. [Quantum Mechanics
requires complex values to represent the physics.]

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-34

Offbeat Applications: Summing Series

Sum 1: The square wave example yielded the representation:

 T
4 
 1   2 t   1 for 0  t  2
f (t ) 

   
m 1  2 m  1 
sin

2 m  1  
T  1 for  T  t  0
 2
4 
 1    4 1 1 1 
Examining f (T 4 )  1 

  2m  1  sin  2m  1 2    1  3  5  7  ...
m 1
provides the

sum of the inverses of the odd integers with alternating signs.

 1
m



m 0 2m  1

4
[FS.21]

Sum 2: The triangular wave example

 2  4x for T  t  T
 T 4 2
 1
m
8 
 2 t   4 x
f (t ) 
 2   2m  1
m 0
2
sin  2m  1


T   T
for  T  t  T
4 4
 2  4 x T for  T 2  t   T 4

The trick is to find a value of the argument that allows the sines to be easily
evaluated. Choose t = T/4.
 1  1
m
8 
   8 
f (T )  1  2
4 
  2m  1
m 0
2
sin  2m  1
 2    2   2m  1
m 0
2

2
   1  2
   
 2m  1  8
[FS.22]
m 0
 

Additional sums can be extracted from completed Fourier series by invoking

Parseval's Equality and the identity for inner products of functions.

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-35

 T
2
  
2
 
2 2

2
f (t ) dt  T c
 f0  1 a fm  1 b fm 
2 2
T
2
 m 1 m 1 

T /2

 1
T   g (t )
T / 2
*
f (t )dt  c*g 0c f 0  1 2  a*gma fm  1 2  a*gp a fp
m p

Sum 3: The square wave and Parseval's Equality.

 T
4 
 1   2 t   1 for 0  t  2
f (t ) 

   sin  2m  1
m 1  2 m  1  

T  1 for  T  t  0
 2
T 2
 2
 
2   
1  16   1 
  
2 2

2
f (t ) dt  T  T c
 f0  1 a fm  1 b fm   T    
2 2 2 2 
T  m1 m1  m1  2m  1 
2

2
   1  2
with the result that   
 2m  1 

8
m 0
 
Before you cry foul, think about it. It would have been much worse if the two
approaches were in conflict!

Exercise: Relating sums of odd and even index terms for inverse powers.
2
   1  2 1 1 
1 1  1
   
 2m  1 
  2
8 n odd n
and ….  2 allm  2m 2 22 allm m2
n even n
 
m 0
 
1 1 1 1 1 1 5 1
n
all n
2
 
n even n
2
  2 2  2  2  2
n even n 2 n odd n n odn n 4 n odd n

1 1
Give the values of n
all n
2
and 
n even n
2
where all n means all positive integers.

Exercise: Use the Fourier series for the unit amplitude triangular wave and
Parseval's Equality to develop the sum of an infinite series.

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-36

Recall the term-by-term integration of series. The example was the triangular
wave f(t). Sketch f(t) and its integral.
 1
m
2 t ' 
t  t
8 
g (t )   f (t ') dt ' 
 2   2m  1  sin  2m  1
m 0
2
T 
dt '
0 0

 1
m

T 2  1  cos  2m  1 2T t  


8

 2   2m  1
m 0
3

Note that the average value of g(t) is:

 1
 m
4
g (t ) 
3
  2m  1
m 0
3

Spatial Fourier Series

Most of the material has been directed at temporal Fourier series. The relations are
restated for spatial series. In the space-spatial frequency realm, Fourier's
conjecture takes the form: Any function f(x) which is periodic with period L can be
represented as a sum of a constant plus sinusoidal terms with spatial frequencies
m  m  2 L  . The sinusoids are the sine and cosine. The set of frequencies m

ensure that each sinusoid completes an integer m cycles in the distance L, the
period of f(x). The base frequency for the expansion is 0  1   2 L  which is
called the fundamental spatial (angular) frequency. This frequency is
determined by the period of the function to be expanded alone. The frequency
of every term in the expansion is an integer multiple (harmonic) of this
fundamental frequency.
 
f ( x)  c0 1   am cos m x 
m 1
b m 1
m sin m x  where m  m  2 L  .

Each term in the expansion represents a distinct behavior of a function with period

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-37

L. The periodicity restricts the function to this set of behaviors. The first term c0
(1), a fancy way to write c0 , is the average value of the function with the (1)
indicating that it represents the constant behavior of the function. The terms
am cos m x  represent the independent or distinct ways in which an even function of

period L can wiggle, and the terms bm sin m x  represent the independent or distinct
ways in which an odd function of period L can wiggle. It is crucial to recognize
than each term represents an independent behavior.

Begin with 1. Find out how much constant behavior is in the function.
L2 L2  L2  L2
1 1 1 1
 am  1 cos m x  dx   bm  1 sin m x  dx
L  L 2 L  L 2
1 f ( x) dx  1 c0 dx 
m 1 L L 2 m 1 L L 2

The integrals of the sine and cosines are over an integral number of cycles - hence
they vanish. Wiggling like a sine or cosine for an integral number of cycles is
nothing like being constant. So:
L2 L2
1 1
c0  
L L 2
1 f ( x) dx   f ( x) dx  f ( x)
L L 2
[FS.23]

The constant part of the function is just its average value over the interval

 L 2 , L 2 . Periodic implies all single intervals of length L give the same result.

The coefficients for the independent wiggle modes become:

L2 L2
2 2
a p   cos  p x  f ( x) dx and bp   sin  p x  f ( x) dx for p  1 [FS.24]
L L 2 L L 2

The factor of two arises because the average value of cos 2  p x  and of sin 2  p x 

is just one half (while the average value of 1 is 1 !).

Spatial Fourier series have the same Parseval, inner product and convergence
properties as do the temporal series.

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-38

An ALTERNATIVE NOTATION replaces m by km when the Fourier series is for
function of a spatial variable. The terms spatial frequency and wave number are
used for km. It is common to expand functions dependent on several spatial
dimensions in which cases the collection of spatial frequencies forms a vector.

Hence the final term for kmn is wave vector where the number of subscripts
matches the dimensions of the function's argument..

Beyond periodic functions: THE FOURIER TRANSFORM

The generalization of Fourier series to forms appropriate for more general

functions defined from - to + is not as painful as it first appears. The process is
presented here even though it lies beyond the goals for this section. The process
illustrates the transition from a sum to an integral, a good thing to understand. The
functions to be expanded are restricted to piecewise continuous square integrable
functions. (Less restrictive developments may be found elsewhere.)
The complex form of the Fourier series is the starting point.

 

f ( x)  d exp i m 2 x  [FS.25]
m 
m
 L 

A typical coefficient d p is projected out of the sum by first multiplying both sides

 
by exp   i p 2 L x 
 

     
L/2  L/2

 exp   i p 2 x  f ( x) dx  d  exp   i p 2 x  exp i m 2 x  dx  L d p

L/ 2
 L  m 
m
L/ 2
 L   L 

 L   
L/2
or d p  1 exp   i p 2 x  f ( x) dx .
L/ 2
 L 

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-39

  

The trick is to define k  2 m / L in f ( x)  d exp i m 2 x  to yield
m 
m
 L 

       
 

 exp i m 2 x  2
L m
f ( x)  dm L  dm L exp i kx  k
m 
2  L  
2

where k = 2/L, the change in k as the index m increments by one from term to
term in the sum. In the limit L   , k becomes and infinitesimal, k becomes a
'continuous' rather than a discrete variable [ d m  d (k ) ], and the sum of a great
many small contributions becomes an integral.
 
f ( x)   1
2   

d (k ) L  eikx dk   1    g (k )  eikx dk
2


where from the equation for the dp

L/2
d (k ) L  
L/2
e ikx f ( x) dx  g (k )

The function g(k) is the Fourier transform of f(x) which is the amplitude to find
wiggling at the spatial frequency k in the function f(x). The Fourier transform of
f(x) is to be represented as f (k ) . Sadly, there is no universal treaty covering, the
Fourier transform, and factors of 2 are shuttled from place to place in different
treatments. Some hold that the balanced definition is the only true definition:
 
f ( x)   1   f (k )  eikx dk
  f (k )   1  f ( x) e ikx dx
  [FS.26]
 2    2  

where the twiddle applied to the function symbol denotes the Fourier transform of
that function. In truth the inverse of 2 must appear, but it can be split up in any
fashion.
1 S  S 
f ( x)   1
2  

 f (k )  eikx dk
  f (k )   1
2  
  f ( x)  e
 ikx
dx

1
The common choices are S = 0, S = 1 and S = /2.

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-40

The Fourier transform has a rich list of properties. Suffice it to say that analogs of
the convergence properties, inner products and Parseval relations found for the
Fourier series exist and much more. The current goal is only to show that the
restriction to periodic functions can be relaxed.

Fourier methods are extremely mysterious on first encounter. Pay the price. The
rewards for mastering Fourier methods are enormous and cool. In the time
domain, the Fourier transform identifies the frequency content of a function of
time. Modern SONAR and passive acoustic monitoring systems depend on
examining the received signal transformed into frequency space. Many systems
are identified by their tonals, distinct frequency combinations in their acoustic
emissions. In quantum mechanics, the spatial Fourier transform of the wave
function reveals its plane wave or momentum content. In optics, the spatial
Fourier transform of an aperture predicts the Fraunhofer diffraction pattern of that
aperture. Scaling up to radar wavelengths, the spatial Fourier transform of an
antenna predicts the far-field radiation pattern of that antenna. A result that also
applies to hydrophone arrays in acoustics. There are problems that defy solution in
the time domain that yield results freely when transformed to the (Fourier)
frequency domain.

Fourier Series and the Vector Space of functions with period T

Vector quantities play a central role in the physics. At the lowest level, a vector is
a quantity that has both magnitude and direction. A quick review of the 'algebra'
satisfied by the collection of all possible displacements in three-dimensional space
is presented below. The collection of functions of a single argument that have
period T obeys an analogous 'algebra', and the Fourier basis set { 1, {… cos(mt)

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-41

…}, {… sin(mt) …} } is adequate to build sums representing all the periodic
functions. In an analogous manner, the basis set { iˆ, ˆj, kˆ } supports representations
for all possible displacements.

r  c1 iˆ  c2 ˆj  c3 kˆ
Further, the inner product and all vector operations can be defined in terms of
operations on the expansion coefficients (components) {c1, c2 , c3} .


REVIEW: Displacement Model for a Vector Space
As an introduction, we will review the properties of the collection of all possible

displacements ri of a particle in our familiar model, a flat (Euclidean) infinite three-
dimensional universe.
Rules for the Addition of Displacements
A1. The sum of two displacements is another allowed displacement.
  
r1  r2  r3
A2. Additions can be regrouped without changing the resultant displacement.
     
r1   r2  r3    r1  r2   r3
A3. The order of additions can be changed without changing the sum.
   
r1  r2  r2  r1
A4. There is a displacement that does not change the position of the particle.
   
ri  0  ri for all ri .
 
A5. For any displacement ri , there is another ri that has the opposite action on a particle's
position.
   
for all ri , ri   ri   0 .
Rules for the Multiplication of Displacements by Scalars
 
M1. If ri is an allowed displacement, then c ri is also an allowed displacement.
M2. The scalar multiple of a sum of displacements is the same as the sum of the same multiple of
the individual displacements.
   
c  r1  r2   cr1  c r2 , an allowed displacement
M3. A sum of scalars times a displacement is the sum of the individual scalars times that
displacement.
  
 c  d  ri  c ri  d ri , an allowed displacement
M4. The scalar multiplication of displacements is associative.
  
c  d ri  ri   c d  ri .
M5. One times any displacement is that same displacement.
  
1 ri  ri  ri .

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-42

 Displacements have other features such as a scalar inner (dot) product that yields a measure of the
component of one displacement along the direction of another. The inner product of a
displacement with itself is always positive. Thus the inner product can be used to assign lengths
  
to all the displacements in the space (to set a metric for the space). [ ri  ri ri |] Further,
  
 
we can find a set of three displacements b1 , b2 , b3 such that all the allowed displacements can
be represented as a sum of multiples of the members of this set, a linear combination. That is: all
   
ri  c1 b1  c2 b2  c3 b3 for some scalars c1, c2 , c3 . This set of displacements
  
b , b , b  is a basis for our three dimensional universe. An enormous simplification follows if
1 2 3
  
the basis set b , b , b  is chosen to be orthogonal and unity normalized. For example: the
1 2 3

   
familiar set iˆ, ˆj , kˆ is orthogonal iˆ  ˆj  ˆj  kˆ  kˆ  iˆ  0 in the sense that each represents a
unit displacement that is completely independent of the other two. The set is also unity
normalized as: iˆ  iˆ  ˆj  ˆj  kˆ  kˆ  1 . The components of a vector can be found by projection
using the inner product.
     
cx   r  x  iˆ r ; cy   r  y  ˆj r ; cz   r  z  kˆ r
The inner product of two displacements can be expressed in terms of their components.
 
r1 r2  c1x c2 x  c1 y c2 y  c1z c2 z
The Fourier Series basis set is orthogonal, but it is not unity normalized.

The Fourier vector space of periodic functions

Every well-behaved function of period T can be expanded in a series of the form

 
f (t )  c0 1   am cos mt  
m 1
b
m 1
m sin mt  where m  m  2 T  . [FS.27]

The sum of two functions of period T is another function of period T is also a

function of period T is also a function of period T. Any scalar multiple of a
function of period T is also a function of period T. That is: the addition closure
condition A1 and the multiplication closure condition M1 for vector spaces are
met. Subject to the Dirichlet conditions, the Fourier expansion set
{1;{ cos mt  };{ sin  pt  }} where m  m  2 T  [FS.28]

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-43

provides a complete basis for the expansion. Complete means that basis includes
all behaviors necessary to accurately represent (converge in the mean to) all
reasonable functions with period T. Orthogonal means that each member of the
basis set has a vanishing inner product with any other member of the set. Each
member of the set represents an independent behavior that is not shared with any
other member of the set. The expansion coefficients (components with respect to
the chosen basis) can then be found by using the inner product to project out a

behavior. Choose the cos  pt  behavior. Multiply the series by  cos  pt   and
*

use the orthogonality relations. ( ! Omit the complex conjugation for real
functions.)
T /2 T /2


T / 2
cos  pt  f (t ) dt  
T / 2
cos  pt  c0 dt

 T /2  T /2
  am  cos  pt  cos mt  dt   bm  cos  pt  sin mt  dt
m 1 T / 2 m 1 T / 2

Each function in the basis behaves like itself and not at all like any other so only
the cos-cos integral with m = p survives.
T /2  T /2 
T T
 cos  pt  f (t ) dt  a  m cos  pt  cos mt  dt   am  m  a p
T / 2 m 1 T / 2 m 1 2 2

The familiar result follows:

T /2
2
T T/ 2
ap  cos  pt  f (t ) dt

All operations on the function can be defined in terms of actions on the Fourier
components (coefficients) of the functions. The series for the sum of two functions
(with period T) is the series with components that are the sum of the corresponding
components in the series for the individual functions. The scalar multiple of a
function has Fourier components that are that same scalar multiple of the Fourier
components of the original function. The inner product of two functions can be

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-44

expressed in terms of their Fourier components using Parseval's equality. Given:
 
f (t )  c f 0 1   a fm cos mt  
m 1
b
m 1
fm sin mt 

 
g (t )  cg 0 1   agm cos mt  
m 1
b
m 1
gm sin mt  [FS.29]

T /2
g f  1
T  T / 2
 g (t )  f (t )dt  c*g 0c f 0  1 2  a*gm a fm  1 2  bgp
*

m p
*
b fp

The annoying factors of 1/2 appear because the Fourier basis set is not unity
normalized. Recall that the average value of cos 2  pt  or of sin 2  pt  is one half

(while the average value of 1 is 1 !). Forgive Fourier; he did not realize that he
was defining a basis for a vector space.

Dimension is another issue. The dimension of a vector space is the smallest

number of vectors that span the space - can be formed in a sum to represent every
vector in the space. For the Fourier series space that number is infinite. It is not
very infinite however; it is a mere countable infinity. An infinite set is countable, if
every element can be placed into a one-to-one correspondence with the counting
numbers, the positive integers. One such correspondence is:
c f 0 1  1; a fm cos mt   2 m; b fm sin mt   2 m  1

In a similar fashion, the number of bound states for a hydrogen atom is countable.
Don't sweat it! Several mathematicians assure me that a countable infinity is
hardly more than a finite number in the sense that it causes no problems beyond the
headache currently throbbing between your temples.

STOP ! The lines below are just ideas for additional topics.
Proceed to: Tools of the Trade.
Review this section after you have studied vector spaces. The space of functions

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-45

spanned by the Fourier set is a vector space. Everything that you learn about
vector spaces applies to Fourier series. Parallel developments in Quantum
Mechanics develop basis sets of functions (Hydrogen: 1S, 2S, 2Px, 2Py, 2Pz, 3S,
….) that represent the independent characteristic behaviors for each problem
studied. Each set provides a basis for expansion and another vector space
embodiment. AND, in each case, you are to use the orthogonality relations to
project out the coefficient or component for each characteristic behavior.

Tools of the Trade:

Evaluating Trig Functions using the Unit Circle:

The evaluation of Fourier coefficients often involves specifying values for

expressions of the form: sin[mo ] where  o   p . The answers can literally be read

off the unit circle. The unit circle is a circle of radius one centered on the origin of
the x-y plane. The coordinates of each point on the circle are therefore  cos  ,sin  

where  is the angle measured CCW from the x-axis.

y y
(0,1)
(cos,sin) (-.707,.7 07) (.707,.70 7)

 45º (1,0)
1 (-1,0)
x x

(-.707,-.7 07) (.707,-.7 07)

(0,-1)

Unit Circle for SIN and COS; .707 is a shorthand for 1

2

 0 for m even; m  2 p  1 for m even; m  2 p

cos[m ]   1  
m
sin[m ]  
0 for m odd ; m  2 p  1 1 for m odd ; m  2 p  1

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-46

  0 for m  4 p  0, 4,8,12,... 1 for m  4 p  0, 4,8,12,...
 1 for m  4 p  1  1,5,9,...  for m  4 p  1  1,5,9,...
      0
sin[ m  ]   cos[ m  ]  
 2   0 for m  4 p  2  2, 6,10,...  2  1 for m  4 p  2  2, 6,10,...
 1 for m  4 p  3  3, 7,11,...  0 for m  4 p  3  3, 7,11,...

 0 for m  8 p  1 for m  8 p
 
 1 2 for m  8 p  1  1 2 for m  8 p  1
 1 for m  8 p  2  0 for m  8 p  2
 
    1 2 for m  8 p  3     1 2 for m  8 p  3
sin[m  ]   cos[m  ]  
 4   0 for m  8 p  4  4   1 for m  8 p  4
 1 2 for m  8 p  5  1 2 for m  8 p  5
 
 1 for m  8 p  6  0 for m  8 p  6
 
 1 2 for m  8 p  7  1 2 for m  8 p  7

Exercise: Use the methods above to evaluate 1 m sin(m 2 ) and  

 1 m 1  cos(m 2)  where m is an integer. See problem 16.

Dimensions in integrals and of functions: Integrals represent dimensioned values

in science and engineering while functions and integral are considered
dimensionless (pure numbers) in mathematics. Consider the function bm sin mt  .
The sine function returns a pure number so the coefficient bm must have the
dimensions of the result. Similarly, the arguments of mathematical functions must
be dimensionless. For example mt is the product of  that has the units rad/s and t
which has the units s. The overall units of radians are not a problem as a radian is
the ratio of an arc length measured along a circle to its radius. That is a radian is
itself dimensionless (=m /m).

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-47

Recall the triangular wave example:

f(t) 2

-T -3T/4 -T/2 -T/4 +T/4 +T/2 +3T/4 T

-1

-2

The function f(t) has a maximum value of 1. That is f(t) is dimensionless. A

coefficient bp has the representation.

 
T 4
2
 sin  p  2 T  t  4t T dt
T T 4

Pull all the dimensioned quantities out of the integral. Choose u  p  2 T  t . Here T
is the period and t is the time so u is dimensionless.
The recommended choice for the change of variable is the argument of the most
complicated function in the integrand. Here the argument of the sine function was
  ( x x0 ) 
2

chosen. There are some variants of this rule as a form like e  a2 
requires
.
2
u 2  ( x x0 ) a2

  2 p  t 
 
T 4
2 T4
 T   p  2 T  T 2 p  dt
2
 sin  p  2 T  t  4t T dt   sin  p  2 T  t   4 
T T 4 T T 4   2 p  

 4  
 
2 p 2
T 4
 u   T 2 p  du
2
 sin  p  2 T  t  4t T dt   sin u   
T T 4 T  p 2   2 p  

 
 2   p 2 
T 4
2
 sin  p  2 T  t  4t T dt   2 2    sin u  u du 
T T 4   p    p 2 

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-48

In the final form the quantity inside the first set of braces is to include all the
dimensioned quantities and the integral in the second set of braces is to be
 2 
dimensionless, a pure number. In this example  2 2
happens to be
 p 
dimensionless; the first brace in other problems will carry the dimensions
appropriate for the problem.

Converting Sums to Integrals

It is said that an integral is a sum of little pieces, but some precision is required
before the statement becomes useful. Beginning with a function f(t) and a
i N
sequence of values for t = {t1,t2,t3, ….,tN}, the sum  f (t )
i 1
i does not represent the

t
integral 
t
f (t ) dt even if a great many closely spaced values of t are used. Nothing

i N
has been included in the sum to represent dt. One requires  f (t ) t
i 1
i i where

 2  t
ti  1 i 1  ti1  is the average interval between sequential values of t values at ti.

i N
For well-behaved cases, the expression  f (t ) t
i 1
i i approaches the Riemann sum

definition of an integral as the t-axis is chopped up more and more finely. As

i N
illustrated below, in the limit t goes to zero, the sum  f (t ) t
i 1
i i approaches the

t
area under the curve between t< and t>. That is: it represents  t
f (t ) dt provided the

sequence of sums converges, and life is good. The theory of integration is not the
topic of this passage. The goal is simply to remind you that the t must be
factored out of each term that is being summed in order to identify the integrand.

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-49

 f(t) f(tN)

f(t i)
f(t1) f(t k )

t t

t1 t2 ti tk tN
area = f(tk ) t
t< t>

For the discussion of the inner product, the function h(t) = g(t) f(t) was considered
m N
at N equally spaced points between –T/2 and +T/2. This leads to the sum  h(t
m 1
m )

where the points tm   T 2    T 2 N   m  T N  have equal spacing t  T N . As the

number of terms gets large, the sum must be divided by N to keep the result finite
m N
leading to  h(t )  1 N  .
m 1
m The rule for converting sums to integral requires that

t  T be explicitly factored from each term in the sum. Thus

N
m N m N m N

 h(tm ) 1
m 1
 N  
m 1
[ 1T  h(tm )]  N   [ 1T  h(t
T 
m 1
m )] t which becomes

      g (t ) f (t ) dt as N gets large and t small.

T / 2 T / 2
 1 h(t )  dt  1
T / 2  T  T T / 2

 1
m


Converting alternating term sums: 
m 0 2m  1

4
= 0.7853982

 1
7 m

Summing through m = 7: 
m 0 2m  1
 0.7542680

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-50

  1
 m

The goal is to estimate 

m 8 2m  1
by converting the sum to an integral.

Converting a sum to an integral is sure fail unless the change in the value of terms
is small compared to the absolute value of the terms from one term to the next.
The conversion may fail anyway, but unless this condition is met, it’s hopeless.
The series above is an alternating (sign) term series. The magnitude of the change
in value of the sum from term to term is twice the magnitude of the terms. A trick
is needed. Combine positive and negative terms in pairs. It is a trick; think about
the sums below to verify that they are equal.
 1
 m 
 1 1    2 

m 8 2m  1
 4
 4  1 4  3   

 4
16 2  16  3 

The last sum has the advantage that it is a sum of positive terms that vanish more
rapidly than do the alternating terms in the original series. It has better convergence
properties.
To convert to an integral,  must be factored from each term, but  takes on
every integer greater than 4, the starting point, so   1 .
 
 2    2  2 1 19 

 4
16 2  16  3   
4
16 2  16  3     16 2  16  3 d   4 ln 17   .0278064
4

 1
7 m 
2

m 0 2m  1

16  
2
16   3
d   0.7821 With patience one can do better.
4

  4 N 3
2  1 1 
Exercise: Show that:  2
16  16  3
d     
4  1 4  3 
d    u 1 du 4 . Use the  
N N 4 N 1

result to evaluate the integral above.

Bounding the sum: Suitable choices of limits can provide both upper and lower
bounds for the sum. A careful sketch shows that for a monotone decreasing

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-51

   
function F(n), N
F ( x) dx   F ( n)  
m N
N 1
F ( x) dx .

The area of the shaded region is F(N),

and F(x) as drawn is a monotone
F(X)
decreasing function. It follows that:
F(N)
N 1 N
 N
F ( x) dx  F ( N )  
N 1
F ( x) dx

Exercise: Argue that for a monotone

decreasing function,
  
  F ( n)  
N N+1
F ( x) dx  F ( x) dx
N N 1
m N

 1  1  1

7 m   m 7 m 
2 2

m 0 2m  1

16 2  16  3
d   
m0 2m  1

m 0 2m  1

16  16  3
2
d
4 3

 1 < 0.7900. These results are consistent with

 m

or equivalently 0.7821 < 

2m  1 m 0

a sum of /4 ≈ 0.7854. A current estimate might be chosen to be:

 1
m 
7
2  1   17 

m 0 2m  1
  16 2
 16  3
d   0.7542680    ln    0.7855587
 4   15 
3.5

It seems that splitting the difference is reasonable so long as one remembers the
bounds on the proven range.

Summing Series and Special Tricks:

Sine and cosine series can be rich sources of Fourier coefficients to use in
evaluating sums either by evaluating the series for a particular argument or by
employing a Parseval relation for the square of a function of the product of a pair
of functions. Use the convergence properties as a guide. Consider a function like:
f(x) = x ( - x) for the interval [0, ]. Evaluated as a cosine series, it has a a
discontinuity in the first derivative when extended so the coefficients vanish no

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-52

faster than m indicating the powers in the sums that can be directly evaluated and
that m-4 sums can be found using a Parseval relation. If the function is evaluated as
a sine series, its anti-symmetric extension has a continuous derivative so the
coefficients might vanish like m-3 so that m-3 and m-6 sums can be evaluated.


The Riemann zeta is particularly important (p) = n n 1
p
. As many Fourier series

consist of only even or odd index coefficients, it is helpful to note that:

   


n  even  0
n  p  2 p  n  p
n 1

n  odd  0
n  p  (1  2 p ) n  p 
n 1

Translating a Function: Fourier series for a function with an offset argument

How can a representation for a symmetric square s(t) wave be built using a
representation of an anti-symmetric square wave a(t)? A reliable procedure for
translating the argument is required. To begin, prepare a figure.

a(t) s(t’)

t
T /4 t’

t’ t

t = t' +T/4 and a(t)  s(t') s(t') = a(t' + T/4)



4 
 1   2 t
a ( t) =

  2m  1  sin  2m  1
m0 T  anti-symmetric square wave

4  1  

2  t ' T 4  
s(t') = a(t' + T/4) =  sin
 m 0  2m  1  
  2 m  1
T
 symmetric square wave


11/21/2015 Physics Handout Series.Tank: Fourier Series FS-53

 4 
 1   2 t ' 
 sin  2m  1  m    
 m
s(t') = 
 0  2m  1   T 

The sine changes sign for each shift of , and sin[ + /2] = cos[].
 2 t '   2 t ' 
 m      =  1 cos  2m  1
m
 sin  2m  1
 T   T 

4   1m 

 2 t ' 
s(t') =    cos  2m  1 or simply
 m 0  2m  1   T 

4    1m   2 t 
s ( t) =    cos  2m  1 .
 
 m  0  2m  1   T 

The Fourier series for the symmetric square wave has been deduced from the series
for the anti-symmetric square wave by carefully applying an argument translation.

Vector Analog of Inner Product Concepts:

 
The inner product of two 3D vectors is to be defined as: A  B  A B cos AB .
It is assumed that the following properties are known:
The inner product of two vectors returns a scalar value.
   
i.) A   c B   c  A  B  where c is a scalar multiplier.
      
ii.) A   B  C   A  B  A  C
     
iii.) A  B  B  A for all A and B .
    
iv.) A  A  0; A  A  0 iffi A  0

FS1. Establish the orthogonality of the basis set eˆ1 , eˆ2 , eˆ3  iˆ, ˆj , kˆ .

iˆ  iˆ  11 cos(0)  1 iˆ  ˆj  11 cos( 2)  0 iˆ  kˆ  11 cos( 2)  0

and so on until we have eˆi  eˆ j  11 cos(ij )   ij .

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-54

FS2. The claim is that all of our vectors can be adequately represented as linear
combinations of the basis expansion set eˆ1 , eˆ2 , eˆ3  iˆ, ˆj , kˆ .
 3  3
A   Ai eˆi and B   B j eˆ j
i 1 j 1

If the expansions are robust, they can be used in place of the vectors.
Note the use of distinct dummy summation indices.
   3   3  3 3
A  B    Ai eˆi     B j eˆ j   
 i1   j 1
 A B  eˆ  eˆ 
i j i j
 i1 j 1

  3 3 3
A B    
 Ai B j  ij   Ai Bi
i 1 j 1 i 1

Summing over j, the Kronecker delta picks out the one term with j = i.
Hence we have developed a second representation of the inner product that is
expressed in terms of the expansion coefficients.
  3
A  B  A B cos  AB   Ai Bi  Ax Bx  Ay By  Az Bz
i 1

FS3. Suppose that a vector has two expansions in terms of the set.
 3 3
A   Ai eˆi   Ai/ eˆi
i 1 i 1

Show that the coefficients must be equal term by term. Take the inner product on
both sides with an independent expansion set member.
3 3 3 3
eˆ j   Ai eˆi   Ai eˆ j  eˆi  eˆ j   Am/ eˆm   Am/ eˆ j  eˆm
i 1 i 1 m 1 m 1

3 3

 A
i 1
i ji   Am/  jm  Aj  A/j
m 1

In the sum over i, the Kronecker delta picks out the one term with i = j. The
analogous collapse occurs for the sum over m. The conclusion is that the expansion
coefficients must be equal term by term so that the expansion in terms of an

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-55

orthogonal expansion set is unique.

  
FS4. For a vector H  c f F  cg G , one finds the expansion coefficients using the
canonical prescription.
     
Hj 
eˆ j  H

eˆ j H


eˆ j  c f F  cg G

 
c f eˆ j  F

 
cg eˆ j  G 
 c f F j  cg G j
eˆ j  eˆ j eˆ j eˆ j eˆ j  eˆ j eˆ j  eˆ j eˆ j  eˆ j

The expansion coefficients are the same linear combination as are the vectors
coefficient by coefficient. H j  c f Fj  cg G j for j = 1, 2, 3.
 
eˆ j  H eˆ j H
Compare H j   with:
eˆ j  eˆ j eˆ j eˆ j
T 2
1
T T 2
cos  pt  f (t ) dt
cos  pt  f (t ) 2
T 2

T T 2
af p   T 2
 cos  pt  f (t ) dt
cos  pt  cos  pt  1
T T 2
cos  pt  cos  pt  dt

Discovery Exercises:
ALERT! Mathematica and Maple templates are included just after the discovery exercises.

D1.) The goal is to represent 4 sin3 (x) as a sum of sine functions.

f(x) = 4 sin3 (x) = a1 sinx + a2 sin(2x) + a3 sin(3x) +a4 sin(4x)
where the coefficients ai are integers in the range –5 to 5.

Method 1: Plot 4 sin3 (x) for the range - to . Set the ai = {1,0,0,0} and plot
g(x) = 4 sin3 (x) - [a1 sinx + a2 sin(2x) + a3 sin(3x) +a4 sin(4x)]
Adjust the ai until g(x) = 0.

Method 2: Plot | g (x)|2 for the range - to . Set the ai = {1,0,0,0} and then adjust
the a1 to minimize the area under the curve for | g (x)|2. After finding the best value

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-56

for a1, move on to a2. Is method 2 more or less efficient than method 1?

Method 3: The functions sin(n x ) represent the independent ways that an anti-
symmetric function of period 2 can wiggle. Plot the functions pairs f(x) and sin(n
x) over the interval - to . Estimate the degree to which the functions in each pair

vary together over the interval. Compute the integrals  1   sin(n x) f ( x) dx for n =

1, 2, 3 and 4. These integrals are proportional the inner product of sin(n x) and f(x)
and gauge the amount of sin(n x) behavior in f(x).

D2.) The goal is to represent 8 cos4 (x) – 8 cos3 (x) - 3 as a sum of sine functions.
f(x) = 8 cos4 (x) – 8 cos3 (x) - 3 =a0 + a1 cos(x) + a2 cos (2x) + a3 cos (3x) +a4 cos
(4x)
where the coefficients ai are integers in the range –9 to 9.

Method 1: Plot f(x) for the range - to . Set the ai = {1,0,0,0,0} and plot
g(x) =8 cos4 (x) – 8 cos3 (x) - 3 – [a0 + a1 cos(x) + a2 cos (2x) + a3 cos (3x) +a4 cos
(4x) ]
Adjust the ai until g(x) = 0.

Method 2: Plot | g (x)|2 for the range - to . Set the ai = {1,0,0,0,0} and then
adjust the a1 to minimize the area under the curve for | g (x)|2. After finding the best
value for a0, move on to a1. Is method 2 more or less efficient than method 1?

Method 3: The functions cos(n x ) represent the independent ways that a symmetric
function of period 2 can wiggle. Plot the functions pairs f(x) and cos(n x) over the

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-57

interval - to . Estimate the degree to which the functions in each pair vary

together over the interval. Compute the integrals  1   cos(n x) f ( x) dx for n = 1, 2,

3 and 4. These integrals are proportional the inner product of cos(n x) and f(x) and
gauge the amount of cos(n x) behavior in f(x). The correct method to compute an


is: an    cos(n x)

f ( x) dx
. For which n might this equation yield a different result


cos(n x) cos(n x) dx


than  1   cos(n x) f ( x) dx ?

Mathematica Templates for D1 and D2: Mathematica is case sensitive.

Mathematica 5.2 Syntax ` is to the left of the 1 key

<<Calculus`FourierTransform` loads the Fourier package
Mathematica 6 Syntax
Fourier transform library is preloaded
<<FourierSeries` New load command needed to load the
Fourier series library
FourierTrigSeries[f(x),x,terms]
generates <terms> terms of the Fourier series for f(x) with period 2.
UnitStep[x] = 0 for x < 0; = 1 for x > 1

Plot[4 (Sin[x])^3, {x,-Pi,Pi}, PlotRange -> {-4,4}, PlotStyle ->RGBColor[1,0,0]]

a0=1; a1=1; a2=0; a3 = 0; a4 = 0 *execute this line whenever the a’s are changed

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-58

 ENTER of SHIFT-RETURN executes a line

Plot[4 (Sin[x])^3 - (a1 * Sin[x] + a2 * Sin[2 x] + a3 Sin[3 x] + a4 Sin[4 x]),

{x,-Pi,Pi}, PlotRange -> {-4,4}, PlotStyle ->RGBColor[1,0,0]]

Plot[(4 (Sin[x])^3 -
(a1 * Sin[x] + a2 * Sin[2 x] + a3 Sin[3 x] + a4 Sin[4 x]))^2,
{x,-Pi,Pi}, PlotRange -> {-4,4}, PlotStyle ->RGBColor[1,0,0]]

Plot[{4 (Sin[x])^3 , Sin[3 x] }, {x,-Pi,Pi}, PlotRange -> {-4,4},

PlotStyle ->{RGBColor[1,0,0],RGBColor[0,0,1]}]
Note that the function to be plotted has been replaced by a list of two functions

(1/Pi) * Integrate[ Sin[ 3 x] *(4 (Sin[x])^3 ),{x,-Pi,Pi}]

Cos[x] is the cosine function

A space or an * represent multiplication
Plot[4 (Sin[x])^3, {x,-Pi,Pi}, PlotRange -> {-4,4}, PlotStyle ->RGBColor[1,0,0]]
[ … ] use square brackets to enclose function arguments
( … ) use parentheses to group terms
{ … } use braces to enclose a list Pi = 
Plot[ function[x], { abscissaVariable, lower, upper}, optional properties]

PlotRange -> {ymin,ymax} set the vertical or ordinate range

PlotStyle -> RGBColor[red, green, blue] sets the plot line color

Integrate[ Sin[ 3 x] *(4 (Sin[x])^3 ),{x,-Pi,Pi}]

Integrate [ function[x], { integrationVariable, lower, upper}]

Maple Sample Code:

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-59

> a0:=1; a1:=0; a2:=0; a3:=0; a4:=0;

> f:=(x)-> 4*(sin(x))^3- (a1*sin(x)+a2*sin(2*x)+a3*sin(3*x)+a4*sin(4*x));

> with(plots):plot(f(x),x=-Pi..Pi);

> with(plots):plot({4*(sin(x))^3, sin(1 * x)},x=-Pi..Pi);

> (1/Pi)* int(sin(x)*f(x),x=-Pi..Pi);

Warm Up Problems:

FSWUP1. PROJECTION (A MOST WONDERFUL PROCESS): Consider a

vector represented as Vi  ai1 B1  ai 2 B2  ...  ain Bn in terms of the basis set { ..

|Bi  .. }. Further, assume that the basis set is mutually orthogonal.  Bj|Bi
Bj|Bi ij. The coefficient aik represents the amplitude of the |Bk  behavior in the
vector |Vi .
a.) Use the inner product to develop a general expression for aik in terms of the
ratio of inner products for vectors expanded in an orthogonal basis.
b.) Use the inner product projection procedure to project out the component Ay in

the expression A  Ax iˆ  Ay ˆj  Az kˆ and where the inner product is are standard dot

product to find an expression for Ay. Compare with an application of the general

B V ˆ
j A i
procedure given an orthogonal basis. aik  k i  Ay 
Bk Bk ˆj ˆj

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-60

  
c.) Consider a function, f (t )  c f 0   a fm cos mt    b fn sin nt  . Use the inner
m 1 n 1

product to project out the coefficient afk in this expansion. Some of the
orthogonality relations for the Fourier trig basis set are:
T 2 T 2 T 2
1 1 1
 sin  pt  cos mt  dt  0;  sin  pt  sin mt  dt  cos  pt  cos mt  dt 
T T 2
1  pm ; 1  pm
T T 2 T T 2 2 2

T 2
1 
and the inner product is f g    f (t )  g (t ) dt . Show that the process develops
T T 2
T 2
2
the standard expression: a p   cos  pt  f (t ) dt
T T 2

DeMoivre’s Theorem and Trig Identities

The Euler relation ei = cos + i sin anchors the development of the theorem. It
follows that 
ei = cos + i sincos(n) + i sin(n) (cos + i sinn
coupling this identity with the Binomial theorem:
(cos + i sinn =
n! n!
(cos  ) n  (cos  ) n 1 (i sin  )1  (cos  ) n  2 (i sin  ) 2  ...  (i sin  ) n
(n  1)!1! (n  2)!2!

Consider the case n = 2: (cos  )2  2 i cos  sin   (sin  )2  cos(2 )  i sin(2 ) .

Equating the real parts and separately the imaginary parts,
(cos  ) 2  (sin  ) 2  cos(2 ); 2 cos  sin   sin(2 ) [Trig.30]

1  cos(2 ) 1  cos(2 )
Exercise: Complete the proofs that sin 2   and cos 2   .
2 2

Problems:

1.) A geometric proof of the identity cos() = cos() cos() + sin() sin()

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-61

was presented in the trigonometry handout. Replace  by - to develop the
corresponding identity for cos(+). Use the identities for cos(-) and cos(+)
to find expressions for cos() cos() and sin() sin() in terms of cos(-) and
cos(+). Use the results to establish the (a) center and (b) rightmost of the

following orthogonality relations. [ m = m (2 /T)]
T 2 T 2 T 2
1 1 1
 sin  pt  cos mt  dt  0;  sin  pt  sin mt  dt  cos  pt  cos mt  dt 
T T 2
1  pm ; 1  pm
T T 2 T T 2 2 2

Pay special attention to the factors p – m and p + m. Which can be zero? Treat the
cases p = m and p m separately.
The leftmost relation is easily established using a symmetry argument. (c) Do so.
(d) Prove one more of the orthogonality relations for the Fourier expansion set:
{1;{…, cos mt  ,…};{…, sin  pt  ,…}}.
It will involve 1 and either a sine or a cosine. If you feel that you have suffered to
complete this problem, you have earned the right to use the orthogonality relations
freely in the future. Note that a minimum of four evaluations is required.
T 2
1
Conclusion: Given the inner product: f g   f (t ) g (t ) dt , the set of
T T 2

functions {1;{…, cos mt  ,…};{…, sin  pt  ,…}} is mutually orthogonal if
2
the frequencies are m = m /T. The constant 1 has a norm squared of 1 while all
others have a norm squared of ½.

2.) Consider two functions of period T represented by their Fourier series:

 
f (t )  c f 0 1  a fm cos mt   b fn sin nt 
m 1 n 1

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-62

  
g (t )  cg 0 1   agk cos k t  
k 1
b
p 1
gp sin  p t 

Show that the two expressions for the inner product of periodic functions are
equivalent. Beware! The sum indices m and n have been used twice. Change to
four distinct index symbols before substituting into the integral. As sine and cosine
 
are real functions,  g (t )  a*gm cos mt    b* sin  t  .
*
 c*g 0 1  gn m
m 1 n 1

T /2
g f  1
T   T / 2
 g (t )  f (t )dt  c*g 0c f 0  1 2  a*gm a fm  1 2  bgp
*

m p
*
b fp


Recall that: A
m 1
m  m  A . Hint: Use the orthogonality relations. SPECIAL NOTE: The

summation indices are dummy labels. They may be relabeled at will. At minimum, rename
the labels in a product to ensure that the labels are distinct. Begin by rewriting the
representations of f(t) and g(t) using distinct summation index labels throughout.

3.) The equation below is to be valid.

   
c f 0   a fm cos mt    b fn sin n t   cg 0   agk cos k t    bg  sin t 
m 1 n 1 k 1  1

Show that this equality requires that:

c f 0  cg 0 ; a fm  agm for all m  1; b fn  bgn for all n  1

Each term in a Fourier series represents an independent behavior. Two such series
can only be equal if they are equal term by term.

4.) Consider the functions f(t) and g(t) with their corresponding Fourier
representations:
 
f (t )  c f 0 1   a fm cos mt  
m 1
b
m 1
fm sin mt 

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-63

  
g (t )  cg 0 1   agk cos k t  
k 1
b
p 1
gp sin  p t 

What is the Fourier representation for the function h(t) = A f(t) + B g(t) where A
and B are constants? Begin with the standard relations for the an’s as an example:
T 2 T 2 T 2
2 2 2
a fp  
T T 2
cos  p t  f (t ) dt ; agp  
T T 2
cos  p t  g (t ) dt ; ahp 
T T 2
cos  p t  h(t ) dt

Substitute h(t) = A f(t) + B g(t). Conclude that Fourier series add component-wise
or coefficient-wise in a fashion similar to vector addition.

5.) Consider the case of a function f(t) with period 2 to be expanded over the
interval - to . What are the allowed values of m? Give the expressions to
calculate co, {…am…} and {…bm..}.

6.) Back to the violin string running from 0 to L. Fourier series can also be used
to represent functions of spatial coordinates. The function f(x) is to be periodic
with period 2 L. (The string shape function is to be extended as a function that is
anti-symmetric about x = 0.) What are the allowed spatial frequencies km for the
Fourier series of f(x)? Give the expressions to calculate co, {…am…} and {…bn…}.
 
Which are certainly zero? f ( x)  c0   am cos  km x    bn sin  kn x 
m 1 n 1

7.) Consider the (INVERTED) symmetric unit amplitude triangular wave.

f(t) 1
t

-T -3T/4 -T/2 -T/4 +T/4 +T/2 +3T/4 T

-1

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-64

  1  4 t for 0t T
 T 2
f (t )  
 1  4 t for T t 0
 T 2

Find the Fourier series for f(t). Sketch the anti-symmetric triangular wave, one of
the example functions treated above, and this symmetric triangular wave on the
same plot. Compare the frequency content of the two triangular waves. Scale the
new wave to have amplitude T/4. That is define g(t) = (T/4) f(t). Give the Fourier
series for g(t). Prepare a sketch of dg/dt. What is the slope of g(t) at t = T/4 ?
Differentiate the Fourier series for g(t) term by term. Comment on the form of the
series after differentiation.

8.) Slope-matched parabolic sections. Consider the function of period 4 defined

over the interval [-2,2] by the equations:
1

0.8

0.6

0.4

 1  (t  1)  2 t  t
2
for2
0t 2
0.2

f (t )   -2 -1.5 -1 -0.5 0.5 1 1.5 2

  1  (t  1)  2 t  t 2t  0
2 2
for -0.2

-0.4

-0.6

-0.8

-1

2
Compute df dt . Is df dt continuous at x = 0 ? Compute d f 2 . Is the second
dt
derivative continuous at x = 0? Compute the Fourier series for f(t). How fast do the
coefficients vanish with respect to m for large m? Note: The function is odd about
zero so only sines appear. The function is even about T/4 so only sine terms that are
odd multiples of the fundamental are to be included.
T 4 1
2
b2 m 1   4   sin 2 m 1t  f (t ) dt  2  sin  2m  1    t   2t  t 2  dt
T 0 0

Sketch the function and a few sines with even and odd multiples of the

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-65

fundamental frequency. Argue that only the odd multiple sine waves will
p
3
contribute. ANSWER: bp  32 for p odd; = 0 for p even.

8 extension 1.) Continue problem 8. Differentiate the series term by term.

Differentiate the series again term by term. Compare the result to series for the
unit amplitude triangle and square waves. What amplitude do you expect for the
first and second derivative of the matched parabola function defined in problem 8?
The derivative of a function of an odd (even) function is an even (odd) function.
How is this fact embodied in the results of differentiating the series expansion term
by term?

8 extension 2.) Continue problem 8. The sum of the terms for m greater than M is

bounded by   32  m  dm   2.0641 2 . Argue that the series converges uniformly
3

M
M

to the function 2 x - x2 on the interval [0,2].

9.) Integrate the series for the unit amplitude anti-symmetric square wave term by
term. Choose the lower limit as ti = -T/4. Each integral term should be of the form:
sin  p 0t ' dt ' . Sketch the expected result. What is the average value of the
t
bp 
T / 4

function represented by the integrated series? Why was the lower limit chosen to
be ti = -T/4? Compare with problem 7. Note that t' is a dummy label for the
integration variable. Comment. (For a detailed discussion of dummy labels in integrals see
the Differential Equations section where the solution of DEs by integration is treated.)

10.) Use the series representations for the anti-symmetric square wave and
triangular wave and the (Parseval) relation for the inner product to compute an
infinite sum involving the cubes of the inverses of the odd integers. [See #2.]

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-66

11.) Develop the Parseval and Parseval Inner Product relations for the Fourier
series using complex exponential basis functions.

12.) This problem illustrates an issue related to convergence. Sketch the anti-
symmetric square wave. Sketch the derivative of the square wave. Differentiate
the series representation for the square wave term by term, Does that series
converge? The series for the square wave does not converge uniformly. Discuss
the specific behavior of the convergence of the series representation of the square
wave that fails to meet the standard for uniform convergence.

13.) The solution to the driven damped oscillator for a single cosine driving term
was found to be
f0  2 
x (t )  R( ) f 0 cos(t   )  cos(t  tan 1  2
)
 0   
 
2 2
 02   2 
2 2

for a driving force F(t) = m fo cos[t]. A standard notation defines the Q or

quality factor of the oscillator as Q  mo b . Show that the equation for x(t) can be
expressed as:
  
x (t ) 
 Fo k  
cos  t  tan 1    o  
  Q 1   o 2   
1    
2
2 1    
    o 
2 2
o
Q

14. Assume that an oscillator has k = 105 k/m ; Q = 100; o = 120  s-1. The driving
force is a 20 Hz triangular wave with amplitude 10 N: (See prob. 13)

80 N  1 1 
F (t )  2 
sin  40 s 1 t   2 sin 120 s 1 t   2 sin  200 s 1 t   ....
  3 5 

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-67

Give the expression for x(t) complete through the 100 Hz term. Make an insightful
comment about the frequency of the dominant term in the response of the
oscillator.

15.) The second example chosen was an anti-symmetric triangular wave. Noting
T
that the function is odd about t = 0 and even about t = /4, only the coefficients of
the odd m sine terms are non-zero. Argue that these may be computed as
T 4
8  4t 
bm 
T  sin  t   T 
0
m dt for m odd. Compute the integrals to find bm for m odd.

Collect everything and present the Fourier series for f(t) defined just below.

f(t) 2

-T -3T/4 -T/2 -T/4 +T/4 +T/2 +3T/4 T

-1

-2

 2  4t for T  t  T
 T 4 2
 4t
f (t )   for  T  t  T
T 4 4

2  4t T for  T 2  t   T 4

Find the Fourier series for f(x), which has a period of 2.
 0 for   x  0


f ( x)   1 for 0 x 
2

 0 for   x 
2

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-68

After you compute the series, compare your result with the partial solution given
below. Use the unit circle tool to evaluate trig functions for arguments that are a
multiple of ½ .
1 1  cos( x) cos(3x) cos(5 x)  1  sin( x) sin(3 x) sin(5 x) 
f ( x)       ...       ... 
4  1 3 5   1 3 5 

2  sin(2 x) sin(6 x) sin(10 x) 
     ... 
 2 6 10 

In the final form, the contents of each set of parentheses above should be replaced
by an infinite sum with an index that assumes integer values. According to the
rules for partial sums, the function might be better represented as:
1 1  cos( x) cos(3 x) cos(5 x) 
f ( x)       ... 
4  1 3 5 

1  sin( x) sin(3x) sin(5 x)  2  sin(2 x) sin(6 x) 
     ...      ... 
 1 3 5   2 6 

What was the change? What requirement or standard is met by the second form
that the first fails to meet by the change?

17.) To what value does the Fourier series found in problem 16 converge for x = 
 1
m


? Show that this leads to the result: 
m 0 2m  1

4
To what value does the


series converge for x = /2 ? Show that this leads to the same sum! What sum can
be computed using Parseval's relation and the series?

18.) A function is said to be periodic with period T if f(t) = f(t + T) for all t.
T T
For a period function f(t) with period T, show that f(- /2) = f(+ /2) and that
f(t) = f(t + 2 T) for all t.

19.) Operations can be executed on the Fourier series for a function as well as on

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-69

the function itself. Examples: differentiation and integration. Properly, the
techniques are guaranteed valid only if the series converge uniformly. This
'requirement' is to be ignored. Start with the anti-symmetric triangular wave:
 1
m
8 
 2 t 
f (t ) 
 2   2m  1
m 0
2
sin  2m  1
 T 

Compute the series for the derivative of f(t) by differentiating the series term by
term. Sketch the derivative. The series represents a symmetric square wave with
amplitude 4/T. Compare the frequency content with that of the anti-symmetric
square wave.

 1
m
df 8 
 2   2 t 
result:  2
dt 
  2m  1
m 0

 T
 cos  2m  1
  T  


Factor out the amplitude. Compare this result directly with the anti-symmetric
square wave shifted by T/4.
44 
1    2  T 
    sin  2m  1    t  4  
T  m0  2m  1    T  
44 
1    2 t    
 
T 
  2m  1  sin  2m  1 
 T
  
2   
m0  
To finish use:
  2 t   
sin  2m  1    
  T 2 
 2 t     2 t   
cos  2m  1  sin  2m  1   sin  2m  1  cos  2m  1 
 T   2  T   2

20.) Operations can be executed on the Fourier series for a function as well as on
the function itself. Examples: differentiation and integration. Properly, the
techniques are guaranteed valid only if the series converge uniformly. This
'requirement' is to be ignored. Start with the anti-symmetric triangular wave:

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-70

  1
m
8 
 2 t
f (t ) 
 2   2m  1
m 0
2
sin  2m  1
 T 

Next, f(t) is to be integrated. Sketch the integral of triangular wave f(t).

 1
m
2 t ' 
t  t
8 
g (t )   f (t ') dt ' 
 2   2m  1  sin  2m  1
m 0
2
T 
dt '
0 0

 1
m

T 2  1  cos  2m  1 2T t  


8

 2   2m  1
m 0
3

Note that the coefficients of its Fourier series vanish as m-3 for large m. This
dependence suggests that the function has a continuous first derivative. Does your
sketch support this conclusion? Compare your result with problem 8 for the slope
 1
m

3
matched parabolas. Note that (see problem 10):   2m  1
m 0
3

32
. The series for

the integrated function has a non-zero constant term. What does that tell you? How
does the function for problem 8 differ?

21.) Consider the function f(t) that consists of two slope-matched parabolic
sections. Consider expanding the function as defined over the interval 0 < t < 2 as
an even or odd function to complete the definition over the period –2 < t< 2.
Would a sine of cosine series converge more rapidly? Assuming that the
coefficients would decrease as m- n for large m predict the value of n for a sine
series expansion and for a cosine series expansion.

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-71

 1

0.8

0.6

0.4

 1  (t  1)  2 t  t
2 2
for 0t 2
0.2

f (t )   -2 -1.5 -1 -0.5 0.5 1 1.5 2

  1  (t  1)  2 t  t 2t  0
2 2
for -0.2

-0.4

-0.6

-0.8

-1

22. The Euler identity provides representations of the sine and cosine in terms of
complex exponentials.
eimx  e  imx eiox  e  ipx
sin(mx)  and cos( px) 
2i 2
Use these relations to transform the orthogonality integral for the sines and cosines
to integrals of complex exponentials to establish:
T 2 T 2 T 2
1 1 1
 sin  pt  cos mt  dt  0;  sin  pt  sin mt  dt  cos  pt  cos mt  dt 
2 pm T 
1  ; 1  pm
T T 2 T T 2 T 2
2

It is easiest to treat the p≠m cases and the p=m cases separately. If you feel that
you have suffered to complete this problem, you have earned the right to use the
orthogonality relations freely in the future. Hints: After substituting the complex
T 2
1
exponentials, one needs calculate integrals of the form  e i 0t dt where 0  2 T .
T T 2

Compute the integral for  a non-zero integer. Do not forget the representation of
sine in terms of the complex exponentials as you do this. Also, read the Tools of
the Trade section of this handout. Give expression for  in terms of m and n.
Suppose that  is zero; what is true about m and n? Set  to zero and evaluate the
integrand and then integrate.

23.) The Euler identity provides representations of the sine and cosine in terms of
complex exponentials.

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-72

 ei  e  i ei  e  i
sin( )  and cos( ) 
2i 2

Use these to establish the identity: cos() = cos() cos() + sin() sin()
Give the corresponding identity for cos(+). Solve the identities for cos(-) and
cos(+) to find expressions for cos() cos() and sin() sin() as sums of terms
linear in trigonometric functions.

24.) Find the cosine series that represents f(x) =  x – x2 over the interval [0,].
a.) As the function is expanded in a cosine series, its full period is 2 and its
extension across zero must be even. Sketch f(x) for -2 < x < 2.
b.) At what order derivative does the function have its first discontinuity? The
coefficients are therefore expected to vanish no faster than what inverse power of
the coefficient index?
c.) Give the equations used to compute the expansion coefficients c0 and ak for a
general k. Be sure that you adjusted the standard Fourier series relations to account
for the range of integration and the restriction to a cosine series. Compute them.

d.) Show that f(x) = 2/4 – (x - /2)2. 

d.) Note that f(x) is even about /2. What does this tell you about ak for k odd?
?Answer: f(x) = 2/2 – cos(2x) – 1/4 cos(4x) – 1/9 cos(6x) – 1/16 cos(8x) - …
25.) Slope-matched parabolic sections. Consider the function of period 4 defined
over the interval [-2,2] by the equations:

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-73

 1

0.8

0.6

0.4

 1  (t  1)  2 t  t
2 2
for 0t 2
0.2

f (t )   -2 -1.5 -1 -0.5 0.5 1 1.5 2

  1  (t  1)  2 t  t 2t  0
2 2
for -0.2

-0.4

-0.6

-0.8

-1


32
It has a Fourier expansion f (t )   sin  2 m  1    t  . Consider the case
m  0  (2 m  1)
3 3

t = 1. To what value should the series converge. Use sin  2 m  1     1 and
 
m
 
m

 1
this result to find the sum  (2 m  1)
m0
3
.

26.) Start with the Fourier series for the anti-symmetric square wave.

f(t)

-T -T/2 +T/2 T

-1

The square wave f(t) has a period of T and is extended beyond the base period to
T
display the discontinuities at each multiple of /2. The function is represented as:
 1 for 0  t  T
 2
f (t )   [FS.31]
1 for  T 2  t  0

FOURIER SERIES FOR THE ANTI-SYMMETRIC SQUARE WAVE:

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-74

  4 
  p  for p odd 4 
 1   2 t 
bp    or f (t ) 

  2m  1  sin  2m  1 T 
 0 for p even m0

t T T
a.) Sketch T / 4
f (t ) dt . It should be a triangular wave ranging between - /4 and /4.

4 
 1   2 t 
  2m  1  sin  2m  1
t
b.) Compute  T / 4  m0 T 
dt .

c.) Multiply your result by -4/T. Use part a.) to sketch the result.
d.) Shift the result. Replace t by t - T/4. Sketch the outcome expected.
******* change the wording ! ********
e.) Give the expressions for cos( - m ) for m = 4 n, 4 n +1, 4 n + 2 and 4 n + 3
where n is an integer.
f.) Express the result of part d.) using sines rather than cosines.
g.) Compare the result to the Fourier series for the unit amplitude anti-symmetric
triangular wave.

27.) Slope-matched parabolic sections. Consider the function of period 4 defined

over the interval [-2,2] by the equations:
1

0.8

0.6

0.4

 1  (t  1) 2  2 t  t 2 for 0t 2
0.2

f (t )   -2 -1.5 -1 -0.5 0.5 1 1.5 2

  1  (t  1)  2 t  t 2t  0
2 2
for -0.2

-0.4

-0.6

-0.8

-1


32
It has a Fourier expansion f (t )   sin  2 m  1    t  . Use Parseval's
m0  (2 3
m  1) 3


1
relation to compute the sum 
m0 (2 m  1)6
. Use the extended Parseval's relation

(Equation [FS.17]) and Fourier series calculated in this handout to compute the

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-75

 
1 
(1) m
sums 
m0 (2 m  1) 4
and 
m 0 (2 m  1)5
. Before you use a pre-calculated series for a

particular function, modify the function to make explicit that its period has been set
to T = 4 to match the period of the slope-matched parabolas. Parseval’s equality is
for functions with the same period; that period, of course, can be T for both.
Answers: 6/960; 4/96; 55/1536.

28.) Develop the Fourier series for: f ( x)  x 2    x    assuming that the

function is extended periodically with period 2. Use the result to compute
m1

 1 x 2   3  4  1


cos(m x)
   x    ; 


m
m2
. Answer: 12
m 1 m 1 m2
 
1 1
29.) Consider the Reimann zeta  (n)   n and the sum E ( n )   n . Show
p 1 p p 1  2 p 

 
that  (n)  E (n)  1  21n  (n) . What is (n) - 2 E(n)? Use the result of the problem

Answer:  6

above to compute (2).

30.) Consider the function: f(x) = x ( - x) for the interval [0, ]. Expand it in a
cosine series and in a sine series. Check the extended functions for discontinuities
and predict the fall-off in the magnitudes of the coefficients in the large index
limit.

31.) Use the previous two problems to show that:


1 2 
(1) n 1 2 
(1) n 1 3
 2
  2
  3

n 1 n  n 1  n 12 n 1  2n  1 32


32.) Use the previous three problems to show that:

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-76

 
1 4 
1 6
 4
  6

n 1 n 90 n 1  n 945

Take the time to repeat the derivation of Parseval equalities to ensure that you
use the correct normalization factors.

33.) Explain why one cannot use a combination of the sine and cosine series
for f(x) = x ( - x) found above to examine sum of terms like n-5.

34.) Blackbody Radiation: The following integral rises in the study of blackbody
radiation.
 q 3 dq  1
 0 e 1 0
q
  q 3 dq e  q 1  e  q  
  
  q 3 dq e  q 1  e  q  e 2 q  e 3q  ...   q 3 dq  e  nq
0 0
n 1


Show that this integral is equal to 6! n 4 . This sum can be evaluated using
n 1

Parseval’s equality. Choose the Fourier cosine series for x (x - ) for 0 < x < .

x( x      2 6   cos(22mx)
m 1 m
 
    q 3 dq  
Show that   x( x     dx  36  (½) m 4   m4  90 leading to   .
0
m 1 m 1
0 e q  1 15

35.) Square waves come in several forms. Consider an odd square wave with
period T stepping from 0 to +1. It is a square wave with a 50% duty cycle.

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-77

The square wave f(t) has a period of T and is extended beyond the base period to
T
display the discontinuities at each multiple of /2. The function is represented as:
 1 for 0  t  T
 2
f (t )  
 0 for  T 2  t  0

Next. consider an odd square wave with period T stepping from 0 to +1. It is a
square wave with a 25% duty cycle.

The square wave g(t) has a period of T and is extended beyond the base period to
display the discontinuities. The function is represented as:
 1 for 0  t  T
 4
g (t )  
 0 for  3T 4  t  0

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-78

Compare the fundamental frequency for the expansion of f(t) and the allowed
spectrum of frequencies in the Fourier expansion of f(t) with those for the
expansion of g(t).

36.) Consider the set of all complex-valued functions that are periodic with period
T /2
T and that have a finite square integral.  T / 2
| f (t ) |2 dt finite . The sum of two

functions is:
Sum[f(t), g(t)] = f(t) + g(t)
The scalar multiple of a function is:
ScalarMult[d, f(t)] = d f(t)
For complex valued functions, complex numbers are the appropriate scalar field.
(If the set is restricted to real-valued functions, then the real numbers should be
chosen.) Show that the set defined with the prescribed operations is a vector space.

37.) A definition for the inner product of two functions that are periodic with
period T and finite square integral has been proposed.
T /2
g f  1T   
T / 2
 g (t ) 
*
f (t )dt

a.) What type of value does this operation return?

b.) Verify that the operation satisfies the axioms required of an inner product for
the set of functions prescribed in the previous problem.

38.) Consider the set of all complex-valued functions that can be formed as linear
combinations of the functions { 1, { .. , cos[mot], … }, { … , sin[not], … }}

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-79

where o = 2/T and m = mo subject the restriction * below is a vector space with
the operations on the representative members g(t) and f(t):
 
f (t )  c f 0 1   a fm cos mt  
m 1
b
m 1
fm sin mt 

 
g (t )  cg 0 1   agm cos mt  
m 1
b
m 1
gm sin mt  

 
Sum[f(t), g(t)] =  c f 0  cg 0  1  a fm  a gm  cos m t    b fn  bgn  sin mt 
m 1 n 1

The scalar multiple of a function is:

 
ScalarMult[d, f(t)] = d c f 0 1    d a fm  cos mt  
m 1
  d b  sin  t 
n 1
fn n

 

2 a 2 b
2 2
Retriction *: c f 0 2  1
fm  1
fm is finite .
m 1 m 1

Show that the set defined with the prescribed operations is a vector space. The
scalars can be the real numbers or the complex numbers.

39.) A definition for the inner product of the vectors in the previous problem has
been proposed.
 
 c*g 0 c f 0 1  2a  2b
g (t )| f (t ) 1 *
a fm 1 *
b fn s
gm gn
m 1 n 1

a.) What type of value does this operation return?

b.) Verify that the operation satisfies the axioms required of an inner product for
the set of functions prescribed in the previous problem.

40.) One period of a function is given below along with its Fourier series.

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-80

  0 for  1  t  1
 2

f (t )   1 for 1  t  1
2 2

 0 for 1  t  1
2
1 2  cos( t ) cos(3 t ) cos(5 t ) 
f (t )       ... 
2  1 3 5 

a.) What is the period of the function?

b.) What are the allowed frequencies m in the expansion of a function with
this period?
c.) Sketch the function for – 2 < t< + 2.

d.) To what value does the Fourier series converge at t = - /2?
e.) To what value does the Fourier series converge at t = +1?
f.) The Fourier coefficient am for this function is computed as:

T 2
2 1/ 2

T T 2
ap  cos  p t  f (t ) dt  cos  p t  [1] dt
1/ 2

1/ 2
Start with  1/ 2
cos  p t  [1] dt and show the steps to find ap for all p>0 and

hence generate (and display) the series for f(t) written in a summation form.
h.) What property of the function f(t) is represented by the constant co = ½ ?

41.) Slope-matched parabolic sections and beyond. Consider the function of

period 4 defined over the interval [-2,2] by the equations:
1

0.8

0.6

0.4

 t1 (2  t )1 for 0t 2
0.2

f (t )   1 -2 -1.5 -1 -0.5 0.5 1 1.5 2

 t (2  t ) 2t 0
1
for -0.2

-0.4

-0.6

-0.8

-1

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-81

 2
This function and its first derivative are continuous. However, d f is
dt 2
discontinuous. It has maximum and minimum values of +1 and -1.

Consider the function:

1
0.75
0.5

 t n (2  t ) n for 0  t  2
0.25
g n (t )   n 1 n
 (1) t (2  t ) for  2  t  0
n
-2 -1 1 2
-0.25
-0.5
-0.75
Figure for n = 6.
-1

a.) Show that this function and its first n derivatives are continuous and that
d n 1 g n
is discontinuous. Confirm that gn(t) has maximum and minimum values
dt n 1
of +1 and -1.
b.) It is expected that the Fourier expansion coefficients will vanish no faster than
1
/m N for large m. What is N?
c.) In the standard Fourier expansion series, which coefficients are sure to vanish?
Explain.
d.) Compute the Fourier series for g2(t).
T 4 1
2
b2 m 1   4   sin 2 m 1t  f (t ) dt  2  sin  2m  1    t  t n (2  t ) n  dt
T 0 0

p
3
ANSWER: bp  32 for p odd; = 0 for p even.

termsFS  21; FourierTrigSeries

 x2 1  2  x2  UnitStepx  x2 1  2  x2  x2 1  2  x2 , x, termsFS

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-82

 12  2 Sin 2  x 4  3 2  Sin 6  x
  

12  25 2  Sin 10  x 12  49 2 Sin 14  x 4  27 2  Sin 18  x

2 5 162 5

  

12  121 2  Sin 22  x 12  169 2 Sin 26  x 4  75 2  Sin 30  x
6250  5 33614  5 39366 5

  

12  289 2  Sin 34  x 12  361 2 Sin 38  x 4  147 2  Sin 42  x
322102 5 742586 5 506250 5

 
2839714 5 4952198 5 2722734 5

The coefficients for the terms beginning with 12 in the numerator are of the form:
(12  m 2  2 )
5
. Note that these terms approach zero as m-3 for large m. This is
2m

consistent with the statement that the coefficients vanish no faster than m-[n+1]
where the nth derivative has the first one that is discontinuous.

42.) A standard trigonometric Fourier series for a function f(x) with period L has
 
the form: f ( x)  c0 1   am cos  mk0 x   b m sin  mk0 x  where ko = 2/L.
m 1 m 1

a.) Show that this can be cast in the form:

 
f ( x)  c0 ei (0)   ½(am  ibm ) eimk0 x  ½(am  ibm ) eimk0 x 
m 1

m 
m eimk0 x


This result justifies the form of the complex Fourier series, 
m 
m eimk0 x .

L/2
b.) Show that ( L1 )   L / 2 (eimk x ) eink x dx   mn .
0 0


c.) Pre-multiply f ( x)  
m 
m eimk0 x by (eipk0 x ) and use the orthogonality relation

developed in part b to project out the coefficient p.

2
sin( x) sin( y )  1 cos( x  y )  cos( x  y )

43.) cos( x) cos( y )  1  cos( x  y )  cos( x  y ) 

2
sin( x) cos( y )  1 sin( x  y )  sin( x  y )
2

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-83

Give the Maclaurin series for sin(x) and cos(x) correct to order x2. Use these
expansions to verify the three trig identities for products of sines and cosines to
second order for small arguments.

44.) Show that the series for the anti-symmetric triangle wave converges uniformly
to the triangle wave by showing that the choice of N > 1/4 is adequate to ensure
that | f(t) – sN(t)| <  for all t  [-½ T, ½ T ]. Assume that | f(t) – s(t)|  0 for all t 
[-½ T, ½ T ].

f(t) 2

-T -3T/4 -T/2 -T/4 +T/4 +T/2 +3T/4 T

-1

-2
T T
base period: - //2 < t < //2
 2  4t for T  t  T
 T 4 2
 4t
f (t )   for  T  t  T
T 4 4

 2  4 t for  T  t   T
 T 2 4

 1
m
8 
 2 t 
f (t )   sin  2m  1
 2
m 0  2m  1
2
 T 

45.) Consider the discontinuous sawtooth f(t) = ½t in the base period [-, ].

Show that the series representation is: f (t )   ( 1)n sin(nt ) . What series sum can
n 1

n 1

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-84

be computed by examining f(½)? Identify the series and compute the sum that
follows from the Parseval relation for this series.
Note that  t sin(mt ) dt   m  cos(mt ) dt .

t
46.) Look at the previous problem. Define g(t) =  0
f (t ) dt  over the same base

period. Compute the series directly and by integrating the series from the previous
problem term by term.

47.) Apply de Moivre’s theorem to prove that:

a.) sin 3   3 4 sin   1 4 sin 3 and cos3   3 4 cos   1 4 cos 3 .
b.) sin 4   3 8  1 2 cos 2  18 cos 4 and cos 4   3 8  1 2 cos 2  18 cos 4 .
c.) Explain why each identity above contain all sines or all cosines on the right.

48.) Sine series for a step up/down.

0 for 0  x   4  0 for 0  x   16
 
f ( x)  1 for  4  x  3 4 g ( x)   1 for  16  x  3 16
 0 for 3  x   0 for 3  x  
 4  16 4

a.) Beginning with f ( x)  b
m 1
m sin m x  , identify the allowed values for m.

b.) Multiply the expression by sin[p x] and execute the projection procedure to
develop an expression for bp.
c.) Compute the sine series expansion for f(x).
d.) The function has an angular least increment of /4. How many terms might
the expansion pattern have before the pattern begins to repeat?
e.) Repeat the expansion for g(x).

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-85

 f.) Based on your results what do speculate is true about the expansion
coefficients for f(x) over the interval [0, L] as compared to those for the
linearly stretched function g(x) = f(x/c) over the interval [0, c L]. How does
the set of allowed m change? What c takes one from part a to part e?
2 2 
 sin  (8m  1) x  sin  (8m  3) x  sin  (8m  5) x  sin  (8m  7) x  
Ans: f ( x)  
 m0  8m  1

8m  3

8m  5

8m  7



2 2 
 sin  (8m  1) 4 x  sin  (8m  3) 4 x  sin  (8m  5) 4 x  sin  (8m  7) 4 x  
g ( x)  
 m 0  8m  1

8m  3

8m  5

8m  7



Plot[(2 Sqrt[2]/Pi) (Sin[x] - Sin[3 x]/3 – Plot[(2 Sqrt[2]/Pi) (Sin[x] - Sin[3 x]/3 -Sin[5 x]/5 +
Sin[5 x]/5 + Sin[7 x]/7), {x,0,Pi}] Sin[7 x]/7 + Sin[9 x]/9 - Sin[11 x]/11-Sin[13 x]/13 +
Sin[15 x]/15), {x,0,Pi}]

Lessons Learned

Problem 1. Orthgonality relations are extremely important. The relations are so

important that one should sweat blood for hours proving them. The expressions for
the sine and cosine in terms of complex exponentials are useful. Learn them; love
them.

Problem 2. A Parseval identity is a representation of an inner product as the

sum of the products of the corresponding expansion coefficients of the vectors.
The representation of the inner product is not unique. Inner products return a scalar

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-86

value that gauges the degree to which two entities share common behavior. For
vectors we have two procedures to gauge the
 
A  B  A B cos   Ax Bx  Ay By  Az Bz

degree to which the two are directed in the same direction. One can have two sets
of basis behaviors that completely characterize the members of the set under study.
Consider a 3D Cartesian coordinate system (unprimed) and a second system
(primed) with axes that are rotated relative to those of the first system. Each vector
has a representation in each system.

A  Ax iˆ  Ay ˆj  Az kˆ  Ax/ iˆ /  Ay/ ˆj /  Az/ kˆ /

B  Bx iˆ  By ˆj  Bz kˆ  Bx/ iˆ /  By/ ˆj /  Bz/ kˆ /

The inner product can be computed in either representation

 
A  B  Ax Bx  Ay By  Az Bz  Ax/ Bx/  Ay/ By/  Az/ Bz/

Either set of basis behaviors, iˆ, ˆj, kˆ or iˆ / , ˆj / , kˆ /  is adequate to describe all the 3D

vectors. Such sets are said to be complete. Problem two reveals that there are two
equally valid methods to compute the inner product of two functions that are
periodic with period T. And, in particular, the set { 1, …, cos( m 0 t), …. , …, sin(
2
n 0 t), ….} is a complete set to characterize those same functions give that 0 =

/T and that m and n are the counting integers 1, 2, 3, … .

Problem 3. If the basis is chosen to be orthogonal and complete, every entity in the
set of interest can be represented as a sum of scalar multiples of members of that
basis. To find the scalar multiple for a particular basis member, compute the inner
product of the member with the entity being represented. To find the x component

of a vector A , one computes the inner product of the x character basis member iˆ

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-87

 
with the vector A and normalizes by dividing by the inner product of the basis
 
iˆ  A iˆ A
member with itself. Ax   . Verify that
iˆ  iˆ iˆ iˆ

T 2
1
T T 2
cos  pt  f (t ) dt
2
T 2 cos  pt  f (t )
T T 2
af p  cos  t
 p  f (t ) dt  T 2

1 cos  pt  cos  pt 

T T 2
cos  pt  cos  pt  dt

Two expansions in terms of the same set of orthogonal behaviors can only be equal
 
if the coefficients are equal behavior by behavior. A  B  Ax  Bx , ....

Problem 4. The Fourier series expansion is a linear operation. The various Fourier
coefficients are added and multiplied coefficient by coefficient just as vectors are
added component by component. h(t) = A f(t) + B g(t) leads to:
  
ahp= A afp + B agp. just as C  a A  b B leads to Cx = a Ax + b Bx.
Even though the result seems obvious, one should anchor the result by computing
the coefficients for h(t) in the canonical fashion.
T 2 T 2 T 2
1 2 2
T T 2
ch 0  h(t ) dt ; ahp   cos  p t  h(t ) dt ; bhp   sin  p t  h(t ) dt
T T 2 T T 2

Hence,
T 2
2
cos  p t   A f (t )  B g (t )  dt
T T 2
ahp 

 2 T 2   2 T 2 
 A   cos  p t  f (t ) dt   B   cos  p t  g (t ) dt   A a fp  B agp
 T T 2   T T 2 

Problem Discussion

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-88

Problem 8 with matched slope parabolas:

Plot of the function

Partial Sum Sn = (32/(Pi^3)) Sum[ Sin[ m Pi t/2]/(m^3),{m,1,n,2}]

The sum index begins with m = 1 and sums terms through index n incrementing by
2 to include only the odd terms.
NOTE: As 2 t - t2 is subtracted from the partial sum of the Fourier series, the error
term is plotted.

Plot[(32/(Pi^3)) Sum[ Sin[ m Pi t/2]/(m^3),{m,1,1,2}] -( 2 t - t^2), {t,0,2},

PlotRange -> All, PlotPoints -> 1024]

Plot of the ERROR with mmax = 1

Plot[(32/(Pi^3)) Sum[ Sin[ m Pi t/2]/(m^3),{m,1,11,2}] -( 2 t - t^2), {t,0,2},

PlotRange -> All, PlotPoints -> 1024]

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-89

 Plot of the ERROR with mmax = 11

Plot of the ERROR with mmax = 31

Plot of the ERROR with mmax = 101

What behavior of the function occurs at t = 0 and at t = 2 that causes the

convergence to be slower in the regions of those points.

Gibbs overshoot phenomenon and consistent partial sums:

The Gibbs phenomenon was discovered by Albert Michelson McQuarrie, page 735 while
he was summing a Fourier series using coefficients for a periodic function that he
measured using an optical-mechanical system. Noting the overshoot, he brought it
to the attention of J. Willard Gibbs who computed its magnitude.

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-90

The overshoot phenomenon demonstrates the importance of completing partial
sums consistently to an upper frequency.
The Gibbs phenomenon was
discovered by Albert Michelson
while he was summing a
Fourier series using coefficients
for a periodic function that he
measured using an optical-
mechanical system.

Michelson was a U S Naval

Academy midshipman and
instructor. One of his speed-of - Michelson-Stratton
light measurements was made at harmonic analyzer
Albert Michelson (pre-discovery photo) the academy.

Consider the Fourier series for the function f(x) which has period 2.
 1 for   x  0
 4
 3
f ( x)   4 for 0 x 
2

  1 4 for   x 
2

Examine an incorrect partial sum. Note the first two sums are through a frequency
201 while the last term in the sum reaches a frequency of 402. The result is the
improper appearance of Gibbs (overshoot) ringing at points that alias the positions
of the discontinuities. (A proper partial sum contains all terms of frequency up to
the maximum and no others.)

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-91

Plot[(1/Pi) *
Sum[ ( Sin[( 2 m + 1) x]/( 2 m + 1)+
(-1)^m Cos[( 2 m + 1) x]/( 2 m + 1)
+2 * Sin[(4 m + 2) x]/(4 m + 2)), {m,0,100}],
{x,-4,4}]

A proper partial sum contains all terms of frequency up to the maximum and no
others. The sum below contains all term with frequencies up to 202 and none
higher. Note that the series now appears to converge uniformly in regions in which

the function is continuous. The ringing behaviors at ±  and at - /2 are artifacts of
an improper partial sum.

Plot[(1/Pi) *
(Sum[ ( Sin[( 2 m + 1) x]/( 2 m + 1)+ (-1)^m Cos[( 2 m + 1) x]/( 2 m + 1)
), {m,0,100}] + Sum[2 * Sin[(4 m + 2) x]/(4 m + 2), {m,0,50}]),
{x,-4,4}, PlotPoints -> 512, PlotRange -> All]

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-92

The sum is repeated keeping all terms up to  = 402.

The Gibbs phenomenon occurs only at the discontinuities and, as the frequency
limits was doubled, the domain over which it occurs is halved. Correct partial
sums include all terms with frequency up to the maximum chosen and none
higher.

Linear Transformations and the Various Fourier Series:

Read only if you have studied Vector Spaces and Linear Transformations.

 
Consider the relations between: f (t )  c0 1  a
m 1
m cos  m0t   b
m 1
m sin  m0t 

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-93

  

where 0  2  and: f (t )  d exp i m 2 t  . These represent the vector
m 
m
 T 

f(t), an element in a function space, using two choices of orthogonal basis set
vectors (functions). Following the conventions discovered for partial sums, all
expansions are to be order in increasing frequency order.
Traditional Basis Set:
{ 1, cos(ot), sin(ot), cos(2ot), sin(2ot), cos(3ot), sin(3ot), ... } =
{T1(t), T2(t), T3(t), T4(t), T5(t), T6(t), T7(t), ... }
Complex Basis Set:
{ 1,eiot, e-iot,ei2ot, e-i2ot, ,ei3ot, e-i3ot, ,ei4ot, e-i4ot,... } =
{C1(t), C2(t), C3(t), C4(t), C5(t), C6(t), C7(t), ... }

The function f(t) is represented by column vectors in each scheme.

 c0   d0 
a  d 
 1  1
b1   d 1 
   
a d
f (t )   2  f (t )   2 
 b2   d 2 
   
 a3   d3 
b  d 
 3  3 
  TRADITIONAL    COMPLEX

The transformation matrix from the traditional to the complex basis has matrix
elements {i,j} = Ci|Tj.


References:
1. K. F. Riley, M. P. Hobson and S. J. Bence, Mathematical Methods for Physics
and Engineering, 2nd Ed., Cambridge, Cambridge UK (2002).

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-94

2. J.W. Brown and R. V Churchill, Fourier Series and Boundary Value
Problems, 6th Edition, McGraw-Hill, Boston (2001).
3. H. F. Weinberger, Partial Differential Equations, Blaisdell, Watham, MA
(1965).
4. Discovery exercises are inspired by or directly borrowed from the Physics
Education Group at Oregon State headed by Corinne A. Manogue.
5. Donald A. McQuarrie, Mathematical Methods for Scientists and Engineers,
University Science Books, Sausalito, CA (2003).

Fourier Series Collected:

1.) An odd square wave with period T stepping from -1 to +1 is presented.

f(t)

-T -T/2 +T/2 T

-1

The square wave f(t) has a period of T and is extended beyond the base period to
T
display the discontinuities at each multiple of /2. The function is represented as:
 1 for 0  t  T
 2
f (t )   [FS.32]
1 for  T 2  t  0

FOURIER SERIES FOR THE ANTI-SYMMETRIC SQUARE WAVE:

 4  for p odd
  4 
 1   2 t 
bp    p  or f (t )    2m  1  sin  2m  1
 0  m 0 T 
 for p even

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-95

2.) The second example chosen is a triangular wave. This function is smoother
than the square wave as it has no discontinuities. It does have periodic
discontinuities in its first derivative. Be alert to the differences associated with
expanding a smoother function.

f(t) 2

1
t

-T -3T/4 -T/2 -T/4 +T/4 +T/2 +3T/4 T

-1

-2
T T
base period: - //2 < t < //2
 2  4t for T  t  T
 T 4 2
 4t
f (t )   for  T  t  T [FS.33]
T 4 4

 2  4 t for  T  t   T
 T 2 4

Once again, the function is odd with average value zero.

 
 2 t 
f (t )  b
m 1
m sin mt   b
m 1
m sin  m
 T 
where

FOURIER SERIES FOR THE ANTI-SYMMETRIC TRIANGULAR WAVE:

8   2 t  1  2 t  1  2 t  
f (t )  sin  m   2 sin  m   2 sin  m   ....
 2
  T  3  T  5  T  
[FS.34]
 1
m
8 
 2 t 
  sin  2m  1
2 m 0  2m  1
2
 T 

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-96

3.) Slope-matched parabolic sections. Consider the function of period 4 defined
over the interval [-2,2] by the equations:
1

0.8

0.6

0.4

 1  (t  1)  2 t  t
2 2
for 0t 2
0.2

f (t )   -2 -1.5 -1 -0.5 0.5 1 1.5 2

  1  (t  1)  2 t  t 2t  0
2 2
for -0.2

-0.4

-0.6

-0.8

-1


32
It has a Fourier expansion: f (t )   sin  2 m  1    t  .
m0  3
(2 m  1) 3

4.) Asymmetric 25% square pulse which has a period of 2.

 0 for   x  0


f ( x)   1 for 0 x 
2

 0 for   x  
2
In the final form, the contents of each set of parentheses should be replaced by an
infinite sum with an index that assumes integer values. Use the unit circle tool to
evaluate trig functions for a set of even spaced arguments.
1 1  cos( x) cos(3 x) cos(5 x)  1  sin( x) sin(3 x) sin(5 x) 
f ( x)       ...       ... 
4  1 3 5   1 3 5 

2  sin(2 x) sin(6 x) sin(10 x) 
     ... 
 2 6 10 




11/21/2015 Physics Handout Series.Tank: Fourier Series FS-97

Fourier series for: f ( x)  x 2    x    assuming that the function is extended
m1

 1
periodically with period 2. Use the result to compute 
m 1
m2
. Answer:


cos(m x)
x 2   3  4  1    x    ; 
 m 

m 1 m2 12

11/21/2015 Physics Handout Series.Tank: Fourier Series FS-98