Fourier Analysis

Fourier series allow you to expand a function on a finite interval as an infinite series of trigonometric
functions. What if the interval is infinite? That’s the subject of this chapter. Instead of a sum over
frequencies, you will have an integral.
15.1 Fourier Transform
For the finite interval you have to specify the boundary conditions in order to determine the particular
basis that you’re going to use. On the infinite interval you don’t have this large set of choices. After
all, if the boundary is infinitely far away, how can it affect what you’re doing over a finite distance? But
see section 15.6.
In section 5.3 you have several boundary condition listed that you can use on the differential
equation u00 = λu and that will lead to orthogonal functions on your interval. For the purposes here
the easiest approach is to assume periodic boundary conditions on the finite interval and then to take
the limit as the length of the interval approaches infinity. On −L < x < +L, the conditions on the
solutions of u00 = λu are then u(−L) = u(+L) and u0 (−L) = u0 (+L). The solution to this is most
conveniently expressed as a complex exponential, Eq. (5.19)

u(x) = eikx , where u(−L) = e−ikL = u(L) = eikL

This implies e2ikL = 1, or 2kL = 2nπ , for integer n = 0, ±1, ±2, . . .. With these solutions, the other
condition, u0 (−L) = u0 (+L) is already satisfied. The basis functions are then

un (x) = eikn x = enπix/L , for n = 0, ±1, ±2, etc. (15.1)

On this interval you have the Fourier series expansion
∞
X ∞
X
f (x ) = an un (x), and um , f = um , an un = am um , um (15.2)
−∞ −∞

In the basis of Eq. (15.1) this normalization is um , um = 2L.

Insert this into the series for f .
∞ ∞
X un , f 1 X
f (x) = u n (x ) = un , f un (x)
n=−∞
un , un 2L n=−∞

Now I have to express this in terms of the explicit basis functions in order to manipulate it. When you
use the explicit form you have to be careful not to use the same symbol (x) for two different things
in the same expression. Inside the un , f there is no “x” left over — it’s the dummy variable of
integration and it is not the same x that is in the un (x) at the end. Denote kn = πn/L.
∞ Z L ∞ Z L
1 X 1 X 0
f (x ) = dx0 un (x0 )* f (x0 ) un (x) = dx0 e−ikn x f (x0 ) eikn x
2L n=−∞ −L 2L n=−∞ −L

Now for some manipulation: As n changes by 1, kn changes by ∆kn = π/L.

∞
1 X π L
Z
0
f (x ) = dx0 e−ikn x f (x0 ) eikn x
2π n=−∞ L −L
∞ Z L
1 X ikn x 0
= e ∆kn dx0 e−ikn x f (x0 ) (15.3)
2π n=−∞ −L

James Nearing, University of Miami 1

 15—Fourier Analysis 2

For a given value of k , define the integral

Z L
0
g L (k ) = dx0 e−ikx f (x0 )
−L

If the function f vanishes sufficiently fast as x0 → ∞, this integral will have a limit as L → ∞. Call that
limit g (k ). Look back at Eq. (15.3) and you see that for large L the last factor will be approximately
g (kn ), where the approximation becomes exact as L → ∞. Rewrite that expression as
∞
1 X ikn x
f (x ) ≈ e ∆kn g (kn ) (15.4)
2π n=−∞

As L → ∞, you have ∆kn → 0, and that turns Eq. (15.4) into an integral.
Z ∞ Z ∞
dk ikx
f (x) = e g (k ), where g (k ) = dx e−ikx f (x) (15.5)
−∞ 2π −∞

The function g is called* the Fourier transform of f , and f is the inverse Fourier transform of g .
Examples
For an example, take the step function

1 (−a < x < a)
f (x) = then
0 (elsewhere)
Z a
g (k ) = dx e−ikx 1 (15.6)
−a
1 h −ika i 2 sin ka
= e − e+ika =
−ik k
The first observation is of course that the dimensions check: If dx is a length then so is 1/k .
After that, there is only one parameter that you can vary, and that’s a. As a increases, obviously the
width of the function f increases, but now look at g . The first place where g (k ) = 0 is at ka = π .
This value, π/a decreases as a increases. As f gets broader, g gets narrower (and taller). This is a
general property of these Fourier transform pairs.
Can you invert this Fourier transform, evaluating the integral of g to get back to f ? Yes, using
the method of contour integration this is very easy. Without contour integration it would be extremely
difficult, and that is typically the case with these transforms; complex variable methods are essential
to get anywhere with them. The same statement holds with many other transforms (Laplace, Radon,
Mellin, Hilbert, etc. )
The inverse transform is

∞
C1
dk ikx 2 sin ka dk ikx eika − e−ika
Z Z
e = e
−∞ 2π k C1 2π ik
dk 1 ik(x+a)
Z h i
= −i e − eik(x−a) C2
C2 2π k

√
* Another common
√ notation is to define g with an integral dx/ 2π . That will require a cor-
responding dk/ 2π in the inverse relation. It’s more symmetric that way, but I prefer the other
convention.
 15—Fourier Analysis 3

1. If x > +a then both x + a and x − a are positive, which implies that both exponentials vanish rapidly
as k → +i∞. Push the contour C2 toward this direction and the integrand vanishes exponentially,
making the integral zero.
2. If −a < x < +a, then only x + a is positive. The integral of the first term is then zero by exactly
the preceding reasoning, but the other term has an exponential that vanishes as k → −i∞ instead,
implying that you must push the contour down toward −i∞.

C3
dk 1 ik(x−a)
Z Z
=i e =
C3 2π k C4
1 e −a)
ik (x 1 . C4
= +i (−1)2πi Res = −i 2πi = 1
2π k=0 k 2π

The extra (−1) factor comes because the contour is clockwise.

3. In the third domain, x < −a, both exponentials have the form e−ik , requiring you to push the
contour toward −i∞. The integrand now has both exponentials, so it is analytic at zero and there
is zero residue. The integral vanishes and the whole analysis takes you back to the original function,
Eq. (15.6).
Another example of a Fourier transform, one that shows up often in quantum mechanics
Z ∞ Z ∞
−x2 /σ 2 −ikx −x2 /σ 2 2 /σ 2
f (x ) = e , so g (k ) = dx e e = dx e−ikx−x
−∞ −∞

The trick to doing this integral is to complete the square inside the exponent.

−1  −1 
−ikx − x2 /σ 2 = x2 + σ 2 ikx − σ 4 k 2 /4 + σ 4 k 2 /4 = (x + ikσ 2 /2)2 + σ 4 k 2 /4
 
σ2 σ2

The integral of f is now

Z ∞
−σ 2 k 2 /4 02 /σ 2
g (k ) = e dx0 e−x where x0 = x + ikσ/2
−∞

The change of variables makes this a standard integral, Eq. (1.10), and the other factor, with the
exponential of k 2 , comes outside the integral. The result is
√ 2 k 2 /4
g (k ) = σ π e−σ (15.7)

This has the curious result that the Fourier transform of a Gaussian is* a Gaussian.

15.2 Convolution Theorem

What is the Fourier transform of the product of two functions? It is a convolution of the individual

* Another function has this property: the hyperbolic secant. Look up the quantum mechanical
harmonic oscillator solution for an infinite number of others.
 15—Fourier Analysis 4

transforms. What that means will come out of the computation. Take two functions f1 and f2 with
Fourier transforms g1 and g2 .
∞
dk 0
Z Z Z
0
dx f1 (x)f2 (x)e−ikx = dx g1 (k 0 )eik x f2 (x)e−ikx
−∞ 2π
dk 0
Z Z
0
= g1 (k 0 ) dx eik x f2 (x)e−ikx
2π
dk 0
Z Z
0
= g1 (k 0 ) dx f2 (x)e−i(k−k )x
2π
Z ∞
dk 0
= g1 (k 0 )g2 (k − k 0 ) (15.8)
−∞ 2 π

The last expression (except for the 2π ) is called the convolution of g1 and g2 .
Z ∞
1
dx f1 (x)f2 (x)e−ikx = (g1 ∗ g2 )(k ) (15.9)
−∞ 2π

The last line shows a common notation for the convolution of g1 and g2 .
What is the integral of |f |2 over the whole line?
∞
dk
Z Z Z
dx f * (x)f (x) = dx f * (x) g (k )eikx
−∞ 2π
dk
Z Z
= g (k ) dx f * (x)eikx
2π
Z *
dk
Z
−ikx
= g (k ) dx f (x)e
2π
Z ∞
dk
= g (k )g * (k ) (15.10)
−∞ 2π

This is Parseval’s identity for Fourier transforms. There is an extension to it in problem 15.10.

15.3 Time-Series Analysis

Fourier analysis isn’t restricted to functions of x, sort of implying position. They’re probably more
often used in analyzing functions of time. If you’re presented with a complicated function of time, how
do you analyze it? What information is present in it? If that function of time is a sound wave you
may choose to analyze it with your ears, and if it is music, the frequency content is just what you will
be listening for. That’s Fourier analysis. The Fourier transform of the signal tells you its frequency
content, and sometimes subtle periodicities will show up in the transformed function even though they
aren’t apparent in the original signal.
A function of time is f (t) and its Fourier transform is
∞ ∞
dω
Z Z
iωt
g (ω ) = dt f (t) e with f (t) = g (ω ) e−iωt
−∞ −∞ 2π

The sign convention in these equations appear backwards from the one in Eq. (15.5), and it is. One
convention is as good as the other, but in the physics literature you’ll find this pairing more common
because of the importance of waves. A function ei(kx−ωt) represents a wave with (phase) velocity ω/k ,
 15—Fourier Analysis 5

and so moving to the right. You form a general wave by taking linear combinations of these waves,
usually an integral.
Example
When you hear a musical note you will perceive it as having a particular frequency. It doesn’t, and
if the note has a very short duration it becomes hard to tell its* pitch. Only if its duration is long
enough do you have a real chance to discern what note you’re hearing. This is a reflection of the facts
of Fourier transforms.
If you hear what you think of as a single note, it will not last forever. It starts and it ends. Say
it lasts from t = −T to t = +T , and in that interval it maintains the frequency ω0 .

f (t) = Ae−iω0 t (−T < t < T ) (15.11)

The frequency analysis comes from the Fourier transform.

Z T
ei(ω−ω0 )T − e−i(ω−ω0 )T sin(ω − ω0 )T
g (ω ) = dt eiωt Ae−iω0 t = A = 2A
−T i(ω − ω0 ) (ω − ω0 )

This is like the function of Eq. (15.6) except that its center is shifted. It has a peak at ω = ω0 instead
of at the origin as in that case. The width of the function g is determined by the time interval T . As T
is large, g is narrow and high, with a sharp localization near ω0 . In the reverse case of a short pulse, the
range of frequencies that constitute the note is spread over a wide range of frequencies, and you will
find it difficult to tell by listening to it just what the main pitch is supposed to be. This figure shows
the frequency spectrum for two notes having the same nominal pitch, but one of them lasts three times
as long as the other before being cut off. It therefore has a narrower spread of frequencies.
←− T larger

T smaller −→

Example
Though you can do these integrals numerically, and when you are dealing with real data you will have
to, it’s nice to have some analytic examples to play with. I’ve already shown, Eq. (15.7), how the
Fourier transform of a Gaussian is simple, so start from there.
2 /σ 2 σ 2 2
If g (ω ) = e−(ω−ω0 ) then f (t) = √ e−iω0 t e−σ t /4
2 π

If there are several frequencies, the result is a sum.

σ
An √n e−iωn t e−σn t /4
2 /σ 2 2 2
X X
g (ω ) = An e−(ω−ωn ) n ⇐⇒ f (t) =
n n
2 π

In a more common circumstance you will have the time series, f (t), and will want to obtain the frequency
decomposition, g (ω ), though for this example I worked backwards. The function of time is real, but

* Think of a hemisemidemiquaver played at tempo prestissimo.

 15—Fourier Analysis 6

the transformed function g is complex. Because f is real, it follows that g satisfies g (−ω ) = g * (ω ).
See problem 15.13.
Real
Imag

f g

This example has four main peaks in the frequency spectrum. The real part of g is an even
function and the imaginary part is odd.
Real

f g Imag

This is another example with four main peaks.

In either case, if you simply look at the function of time on the left it isn’t obvious what sort of
frequencies are present. That’s why there are standard, well-developed computer programs to do the
Fourier analysis.

15.4 Derivatives
There are a few simple, but important relations involving differentiation. What is the Fourier transform
of the derivative of a function? Do some partial integration.

df ∞
Z Z
F(f˙) = dt eiωt = eiωt f (t) − iω dt eiωt f (t) = −iω F(f ) (15.12)
dt −∞

Here I’ve introduced the occasionally useful notation that F(f ) is the Fourier transform of f . The
boundary terms in the partial integration will go to zero if you assume that the function f approaches
zero at infinity.
The nth time derivative simply give you more factors: (−iω )n on the transformed function.

15.5 Green’s Functions

This technique showed up in the chapter on ordinary differential equations, section 4.6, as a method to
solve the forced harmonic oscillator. In that instance I said that you can look at a force as a succession
of impulses, as if you’re looking at the atomic level and visualizing a force as many tiny collisions by
atoms. Here I’ll get to the same sort of result as an application of transform methods. The basic
technique is to Fourier transform everything in sight.
The damped, forced harmonic oscillator differential equation is

d2 x dx
m 2
+b + kx = F0 (t) (15.13)
dt dt

Multiply by eiωt and integrate over all time. You do the transforms of the derivatives by partial
integration as in Eq. (15.12).
Z ∞ Z ∞
iωt
dt e 2
[Eq. (15.13)] = −mω x̃ − ibω x̃ + k x̃ = F̃0 , where x̃(ω ) = dt eiωt x(t)
−∞ −∞
 15—Fourier Analysis 7

This is an algebraic equation that is easy to solve for the function x̃(ω ).
F̃0 (ω )
x̃(ω ) =
−mω 2 − ibω + k
Now use the inverse transform to recover the function x(t).
∞
dω −iωt dω −iωt F̃0 (ω )
Z Z
x(t) = e x̃(ω ) = e
−∞ 2π 2π −mω 2 − ibω + k
dω e−iωt
Z Z
0
= 2
dt0 F0 (t0 )eiωt
2π −mω − ibω + k
dω e−iωt
Z Z
0
0
= dt F0 (t )0
2
eiωt (15.14)
2π −mω − ibω + k
In the last line I interchanged the order of integration, and in the preceding line I had to be sure to use
another symbol t0 in the second integral, not t. Now do the ω integral.
∞ ∞ 0
dω e−iωt dω e−iω(t−t )
Z Z
0
2
eiωt = 2
(15.15)
−∞ 2π −mω − ibω + k −∞ 2π −mω − ibω + k
To do this, use contour integration. The singularities of the integrand are at the roots of the denomi-
nator, −mω 2 − ibω + k = 0. They are
C1

√
−ib ± −b2 + 4km
ω= = ω±
2m
C2

Both of these poles are in the lower half complex plane. The contour integral C1 is along the real axis,
and now I have to decide where to push the contour in order to use the residue theorem. This will be
0
governed by the exponential, e−iω(t−t ) .
0
First take the case t < t0 , then e−iω(t−t ) is of the form e+iω , so in the complex ω -plane its
behavior in the ±i directions is as a decaying exponential toward +i (∝ e−|ω| ). It is a rising exponential
toward −i (∝ e+|ω| ). This means that pushing the contour C1 up toward C2 and beyond will make
this integral go to zero. I’ve crossed no singularities, so that means that Eq. (15.15) is zero for t < t0 .
0
Next, the case that t > t0 . Now e−iω(t−t ) is of the form e−iω , so its behavior is reversed from
that of the preceding paragraph. It dies off rapidly toward −i∞ and rises in the opposite direction.
That means that I must push the contour in the opposite direction, down to C3 and to C4 . Because
of the decaying exponential, the large arc of the contour that is pushed down to −i∞ gives zero for its
integral; the two lines that parallel the i-axis cancel each other; only the two residues remain.

C3

∞ 0
dω e−iω(t−t )
Z X
2
= −2πi Res (15.16)
−∞ 2π −mω − ibω + k ω ±
C4
 15—Fourier Analysis 8

The denominator in Eq. (15.15) is −m(ω − ω+ )(ω − ω− ). Use this form to compute the residues.
Leave the 1/2π aside for the moment and you have

0 0
e−iω(t−t ) e−iω(t−t )
=
−mω 2 − ibω + k −m(ω − ω+ )(ω − ω− )

The residues of this at ω± are the coefficients of these first order poles.

0 0
e−iω+ (t−t ) e−iω− (t−t )
at ω+ : and at ω− :
−m(ω+ − ω− ) −m(ω− − ω+ )

The explicit values of ω± are

√ √
−ib + −b2 + 4km −ib − −b2 + 4km
ω+ = and ω− =
2m √ 2m
−b2 + 4km b
Let ω0 = and γ=
2m 2m

The difference that appears in the preceding equation is then

ω+ − ω− = (ω 0 − iγ ) − (−ω 0 − iγ ) = 2ω 0

Eq. (15.16) is then

0
" 0 0 0 0
#
∞
dω . e−iω(t−t ) e−i(ω −iγ )(t−t ) e−i(−ω −iγ )(t−t )
Z
= −i +
2
−∞ 2π −mω − ibω + k −2mω 0 +2mω 0
−i 0 0 0 0 0
e−γ (t−t ) − e−iω (t−t ) + e+iω (t−t )
 
= 0
2mω
1 −γ (t−t0 )
sin ω 0 (t − t0 )


= 0
e
mω

Put this back into Eq. (15.14) and you have

Z t
1 0
dt0 F0 (t0 )G(t − t0 ), G(t − t0 ) = e−γ (t−t ) sin ω 0 (t − t0 )


x(t) = where 0
(15.17)
−∞ mω

If you eliminate the damping term, setting b = 0, this is exactly the same as Eq. (4.34). The integral
stops at t0 = t because the Green’s function vanishes beyond there. The motion at time t is determined
by the force that was applied in the past, not the future.

Example
Apply a constant external force to a damped harmonic oscillator, starting it at time t = 0 and keeping
it on. What is the resulting motion?

0 (t < 0)
F0 (t) =
F1 (t > 0)
 15—Fourier Analysis 9

where F1 is a constant. The equation (15.17) says that the solution is (t > 0)
Z t Z t
0 0 0
x(t) = dt F0 (t )G(t − t ) = dt0 F1 G(t − t0 )
−∞ 0
Z t
1 0
dt0 e−γ (t−t ) sin ω 0 (t − t0 )


= F1 0
0 mω
Z t
F1 0  0 0 0 0 
= 0
dt0 e−γ (t−t ) eiω (t−t ) − e−iω (t−t )
2imω 0
 t0 =t
F1 1 (−γ +iω 0 )(t−t0 ) 1 (−γ −iω 0 )(t−t0 )
= e − e
2imω 0 γ − iω 0 γ + iω 0 0
 0 − γt t =0
F1 2iω e
2iγ sin ω 0 t + 2iω 0 cos ω 0 t
 
= −
2imω 0 γ 2 + ω 02 γ 2 + ω 02
F1 h
−γt
 0 γ 0
i
= 1 − e cos ω t + sin ω t
m(γ 2 + ω 02 ) ω0
Check the answer. If t = 0 it is correct; x
(0) = 0 as it should.
If t → ∞, x(t) goes to F1 / m(γ 2 + ω 02 ) ; is this correct? Check it out! And maybe simplify the result
in the process. Is the small time behavior correct?
15.6 Sine and Cosine Transforms
Return to the first section of this chapter and look again at the derivation of the Fourier transform. It
started with the Fourier series on the interval −L < x < L and used periodic boundary conditions to
define which series to use. Then the limit as L → ∞ led to the transform.
What if you know the function only for positive values of its argument? If you want to write
f (x) as a series when you know it only for 0 < x < L, it doesn’t make much sense to start the way
I did in section 15.1. Instead, pick the boundary condition at x = 0 carefully because this time the
boundary won’t go away in the limit that L → ∞. The two common choices to define the basis are

u(0) = 0 = u(L), and u0 (0) = 0 = u0 (L) (15.18)

Start with the first, then un (x) = sin(nπx/L) for positive n. The equation (15.2) is unchanged, save
for the limits.
∞
X ∞
X
f (x ) = an u n ( x ) , and um , f = um , an un = am um , um
1 n=1

In this basis, um , um = L/2, so

∞ ∞
X un , f 2 X
f (x ) = un (x) = u , f u n (x )
n=1
un , un L n=1 n

Now explicitly use the sine functions to finish the manipulation, and as in the work leading up to
Eq. (15.3), denote kn = πn/L, and the difference ∆kn = π/L.
∞ Z
2X L 0 nπx0 nπx
f (x ) = dx f (x0 ) sin sin
L 1 0 L L
∞ Z L
2 X nπx
= sin ∆kn dx0 f (x0 ) sin nπx0 /L (15.19)
π 1
L 0
 15—Fourier Analysis 10

For a given value of k , define the integral

Z L
gL (k ) = dx0 sin(kx0 )f (x0 )
0

If the function f vanishes sufficiently fast as x0 → ∞, this integral will have a limit as L → ∞. Call that
limit g (k ). Look back at Eq. (15.19) and you see that for large L the last factor will be approximately
g (kn ), where the approximation becomes exact as L → ∞. Rewrite that expression as
∞
2X
f (x ) ≈ sin(kn x)∆kn g (kn ) (15.20)
π 1

As L → ∞, you have ∆kn → 0, and that turns Eq. (15.20) into an integral.
2 ∞
Z Z ∞
f (x ) = dk sin kx g (k ), where g (k ) = dx sin kx f (x) (15.21)
π 0 0

This is the Fourier Sine transform. For a parallel calculation leading to the Cosine transform, see
problem 15.22, where you will find that the equations are the same except for changing sine to cosine.
2 ∞
Z Z ∞
f (x ) = dk cos kx g (k ), where g (k ) = dx cos kx f (x) (15.22)
π 0 0

What is the sine transform of a derivative? Integrate by parts, remembering that f has to
approach zero at infinity for any of this to make sense.
Z ∞ ∞ Z ∞ Z ∞
0
dx sin kx f (x) = sin kxf (x) − k dx cos kx f (x) = −k dx cos kx f (x)
0 0 0 0

For the second derivative, repeat the process.

Z ∞ Z ∞
00 2
dx sin kx f (x) = kf (0) − k dx sin kx f (x) (15.23)
0 0

15.7 Wiener-Khinchine Theorem

If a function of time represents the pressure amplitude of a sound wave or the electric field of an
electromagnetic wave the power received is proportional to the amplitude squared. By Parseval’s identity,
the absolute square of the Fourier transform has an integral proportional to the integral of this power.
This leads to the interpretation of the transform squared as some sort of power density in frequency.
|g (ω )|2 dω is then a power received in this frequency interval. When this energy interpretation isn’t
appropriate, |g (ω )|2 is called the “spectral density.” A useful result appears by looking at the Fourier
transform of this function.
dω dω *
Z Z Z
0
|g (ω )|2 e−iωt = g (ω )e−iωt dt0 f (t0 )eiωt
2π 2π
dω *
Z Z
0
0 0
= dt f (t ) g (ω )eiωt e−iωt
2π
Z *
dω
Z
0 0 −iω (t0 −t)
= dt f (t ) g (ω )e
2π
Z
= dt0 f (t0 )f (t0 − t)* (15.24)
 15—Fourier Analysis 11

When you’re dealing with a real f , this last integral is called the autocorrelation function. It tells you
in some average way how closely related a signal is to the same signal at some other time. If the signal
that you are examining is just noise then what happens now will be unrelated to what happened a few
milliseconds ago and this autocorrelation function will be close to zero. If there is structure in the signal
then this function gives a lot of information about it.
The left side of this whole equation involves two Fourier transforms ( f → g , then |g |2 to it’s
transform). The right side of this theorem seems to be easier and more direct to compute than the left,
so why is this relation useful? It is because of the existence of the FFT, the “Fast Fourier Transform,”
an algorithm that makes the process of Fourier transforming a set of data far more efficient than doing
it by straight-forward numerical integration methods — faster by factors that reach into the thousands
for large data sets.
 15—Fourier Analysis 12

Problems
15.1 Invert the Fourier transform, g , in Eq. (15.7).
2 2
15.2 What is the Fourier transform of eik0 x−x /σ ? Ans: A translation of the k0 = 0 case
2 2
15.3 What is the Fourier transform of xe−x /σ ?
15.4 What is the square of the Fourier transform operator? That is, what is the Fourier transform of
the Fourier transform?
15.5 A function is defined to be

1 (−a < x < a)
f (x ) =
0 (elsewhere)

What is the convolution of f with itself? (f ∗ f )(x) And graph it of course. Start by graphing both
f (x0 ) and the other factor that goes into the convolution integral.
Ans: (2a − |x|) for (−2a < x < +2a), and zero elsewhere.
15.6 Two functions are
 
1 (a < x < b ) 1 (A < x < B )
f 1 (x ) = and f 2 (x ) =
0 (elsewhere) 0 (elsewhere)

What is the convolution of f1 with f2 ? And graph it.

15.7 Derive these properties of the convolution:
(a) f ∗ g = g ∗ f (b) f ∗(g ∗ h) = (f ∗ g )∗ h (c) δ (f ∗ g ) = f ∗ δg + g ∗ δf where δf (t) = tf (t),
δg (t) = tg (t), etc. (d) What are δ 2 (f ∗ g ) and δ 3 (f ∗ g )?
15.8 Show that you can rewrite Eq. (15.9) as

F(f ∗ g ) = F(f ) . F(g )

where the shorthand notation F(f ) is the Fourier transform of f .

15.9 Derive Eq. (15.10) from Eq. (15.9).

15.10 What is the analog of Eq. (15.10) for two different functions? That is, relate the scalar product
of two functions, Z ∞
f1 , f2 = f1* (x)f2 (x)
−∞
R
to their Fourier transforms. Ans: g1* (k )g2 (k ) dk/2π
15.11 In the derivation of the harmonic oscillator Green’s function, and starting with Eq. (15.15), I
assumed that the oscillator is underdamped: that b2 < 4km. Now assume the reverse, the overdamped
case, and repeat the calculation.
15.12 Repeat the preceding problem, but now do the critically damped case, for which b2 = 4km.
Compare your result to the result that you get by taking the limit of critical damping in the preceding
problem and in Eq. (15.17).
 15—Fourier Analysis 13

15.13 Show that if f (t) is real then the Fourier transform satisfies g (−ω ) = g * (ω ).
What are the properties of g if f is respectively even or odd?

15.14 Evaluate the Fourier transform of



A a − | x| (−a < x < a)
f (x ) =
0 (otherwise)

How do the properties

 of the transform vary as the parameter a varies?
Ans: 2A(1 − cos ka) k 2

15.15 Evaluate the Fourier transform of Ae−α|x| . Invert the transform to verify that it takes you back
to the original function. Ans: 2α/(α2 + k 2 )

15.16 Given that the Fourier transform of f (x) is g (k ), what is the Fourier transform of the the
function translated a distance a to the right, f1 (x) = f (x − a)?

15.17 Schroedinger’s equation is

∂ψ h̄2 ∂ 2 ψ
−ih̄ =− + V (x )ψ
∂t 2m ∂x2

Fourier transform the whole equation with respect to x, and find the equation for Φ(k, t), the Fourier
transform of ψ (x, t). The result will not be a differential equation.
Ans: −ih̄∂ Φ(k, t)/∂t = (h̄2 k 2 /2m)Φ + (v ∗ Φ)/2π

15.18 Take the Green’s function solution to Eq. (15.13) as found in Eq. (15.17) and take the limit
as both k and b go to zero. Verify that the resulting single integral satisfies the original second order
differential equation.

15.19 (a) In problem 15.18 you have the result that a double integral (undoing two derivatives) can
be written as a single integral. Now solve the equation

d3 x C
= F (t)
dt3

directly, using the same method as for Eq. (15.13). You will get a pole at the origin and how do
you handle this, where the contour of integration goes straight through the origin? Answer: Push the
contour up as in the figure. Why? This is what’s called the “retarded solution” for which the value
of x(t) depends on only those values of F (t0 ) in the past. If you try any other contour to define the
integral you will not get this property. (And sometimes there’s a reason to make another choice.)
(b) Pick a fairly simple F and verify that this gives the right answer.
Rt
Ans: 12 −∞ dt0 F (t0 )(t − t0 )2

15.20 Repeat the preceding problem for the fourth derivative. Would you care to conjecture what 3 1/2
integrals might be? Then perhaps an arbitrary non-integer order?
Rt
Ans: 16 −∞ dt0 F (t0 )(t − t0 )3

15.21 What is the Fourier transform of xf (x)? Ans: ig 0 (k )

 15—Fourier Analysis 14

15.22 Repeat the calculations leading to Eq. (15.21), but for the boundary conditions u0 (0) = 0 =
u0 (L), leading to the Fourier cosine transform.
15.23 For both the sine and cosine transforms, the original function f (x) was defined for positive x
only. Each of these transforms define an extension of f to negative x. This happens because you
compute g (k ) and from it get an inverse transform. Nothing stops you from putting a negative value
of x into the answer. What are the results?

15.24 What are the sine and cosine transforms of e−αx . In each case evaluate the inverse transform.

15.25 What is the sine transform of f (x) = 1 for 0 < x < L and f (x) = 0 otherwise. Evaluate the
inverse transform.

15.26 Repeat the preceding calculation for the cosine transform. Graph the two transforms and com-
pare them, including their dependence on L.

15.27 Choose any different way around the pole in problem 15.19, and compute the difference between
the result with your new contour and the result with the old one. Note: Plan ahead before you start
computing.