Methods of Experimental Physics Semester 1 2005

Lecture 5: Fourier Transforms and Dierential Equations

5.1 Mass-Spring System
y
x
D
k
m
F(t)
Figure 1: A general mass spring system
Lets examine a mass-spring system and consider its response to an external force applied to the mass. A
general diagram of such a system is given in Figure 1 above. The rst job is to write down the governing
equation and then to solve them using our knowledge. We will rst do this the way that you are probably
relatively familiar with, and then I am going to solve it using a Fourier transform approach. I hope that
you will see that the Fourier transform approach makes life much easier.
We will assume the the force from the spring can be expressed as F
SPRING
= k(x y) where x and y
are measured from the equilibrium points given in the diagram above. The damping can be expressed
as F
DAMPING
= D( x y) i.e. it is proportional to the velocities at the end of the springs, and opposed
to the velocity that we attempt to impose upon it. To make life simple lets assume that y = 0. So using
Newtons laws we write:
m x = F
total
= F kx D x (1)
Conventionally one can also rewrite the equations using the substitutions:
2f
0
=

k
m
(2)
Q =
2f
0
m
D
(3)
D/m =
2f
0
Q
(4)
where f
0
is called the natural frequency or resonant frequency of the system, Q is called the Q-factor of
the system and tells us how long the energy persists in the system (i.e. it is actually 2 times the number
of cycles before the energy has fallen to 1/e of some initial value. We will use these substitutions below.
5-1
Lecture 5: Fourier Transforms and Dierential Equations 5-2
5.1.1 Solving the mass-spring system in the time domain
This is a bit of a nightmare so what I am going to do is assume that there is no external force i.e.
F(t) = 0 and then we will solve this. It is horrible enough and you will get a feel for it. Later we will
solve it using the Fourier transform method and we can do it in the full glory.
Lets assume that the solution can be written as x(t) = Ce
t
- this allows for both an oscillatory solution
as well as a damped exponential solution. If we substitute this and its derivatives into Eq. 1 then we
obtain:
2
+ + (2f
0
)
2
= 0 after cancelling a factor of Ce
t
. This polynomial has a two roots:
=

2
4
(2f
0
)
2
(5)
We should distinguish three dierent cases: low damping ( < 4f
0
), heavy damping ( > 4f
0
), and
critical damping ( = 4f
0
)
5.1.1.1 Low Damping
In this case we see that the argument of the square root function is negative and thus is complex
(i.e. since x(t) = Ce
t
this means x(t) will be oscillatory). Let us write =

2
2f
d
where
f
d
=

f
0
2

2
16
2
. There are thus two solutions and all linear combinations of these two linear solutions
are also solutions i.e.:
x(t) = exp

t
2

[C
1
exp(2f
d
) +C
2
exp(2f
d
)] (6)
If one requires that the solution is real (as we probably should for physical reasons) then C
2
= C

1
. If
one thinks about it a little you can see that this linear combination of oppositely rotating exponential
solutions can be rewritten as:
Ae
it
+A

e
it
= 2(A) cos(t) +2(A) sin(t) = 2|A| cos (t +) where = arctan [(A)/(A)] (7)
so the net result of this is an exponentially damped oscillatory solution with an arbitrary phase.
2 4 6 8 10
-0.75
-0.5
-0.25
0.25
0.5
0.75
1
Figure 2: A lightly damped sinwave
5.1.1.2 Heavy Damping
This is the case when both of the solutions for are real and negative. Doing the same maths as before
we get to a solution that looks like:
x(t) = C
1
e
t(
1

2
)
+C
2
e
t(
1
+
2
)
(8)
Lecture 5: Fourier Transforms and Dierential Equations 5-3
0.2 0.4 0.6 0.8 1
0.2
0.4
0.6
0.8
1
Figure 3: Heavily damped mass-spring system: time in units of the oscillator period
where
1
= /2 and
2
=

2
4
4f
0
2

2
By imposing boundary conditions we obtain the specic
values for C
1
and C
2
. If we impose x(0) = 1 and = 8f
0
we obtain Figure 3. We note that if the
damping is really high then the system is essentially exponentially damped.
5.1.1.3 Critical Damping
By choosing = 4f
0
we obtain only a single solution to the problem, yielding x(t) = C exp(

2
t). This
only gives us one arbitrary constant where we need two (as this is a second order d.e.) in order to satisfy
the full range of boundary conditions. The correct solution is:
x(t) = (C
1
+ 2C
2
f
0
t)e
2f
0
t
= (C
1
+C
2
t/2)e
t/2
(9)
Figure 4 shows a much quicker damping than the heavily damped oscillator shown above in Figure 3!
0.2 0.4 0.6 0.8 1
0.2
0.4
0.6
0.8
1
Figure 4: Critically damped: time in units of the period of the oscillator
5.1.2 Forced Vibration
Now lets turn to the possibility of forced vibration of the mass. After we start to drive the device there
will be some transient behaviour while the device comes into a steady-state with the driving force. We
will ignore this and only be interested in the steady-state solution.
Lecture 5: Fourier Transforms and Dierential Equations 5-4
Lets use a complex notation for the both the applied force,

F exp(t) where

F = F
0
exp() and the
movement of the object is given by xexp(t). Of course we want a real solution in both cases but we
will apply this at the end. If the guiding equation (Eq. 1) is: m

x =

F k x D

x then rearranging we
get:
x =

F
4m
2
(f
2
0
f
2
+f/(2))
(10)
This equation is termed a response function or a transfer function as it relates the behaviour of the
position of the mass to the driving acceleration (F/m) i.e.
R =
1
4
2
(f
2
0
f
2
+f/(2))
(11)
If we want to examine the form of this equation we could either plot the real and imaginary components
or alternatively the magnitude and phase. The magnitude of the function is:
|R|
2
=
1
16
4
(f
2
0
f
2
)
2
+
2
(2f)
2
(12)
To nd the phase lets assume that the response function can be written as R = exp() where is
obviously equal to |R| given in Eq. 12 while is yet to be found:
x = R

F
m
= F
0
e

m
=
F
0
m
e
(+)
(13)
Lets not forget that the displacement should be real so we will take the real part of this to yield:
(

x(t)e
it
) = x(t) =
F
0
m
cos( + +t) (14)
Now since exp() = R/ = cos() + sin() by denition, we are able to derive the phase from:
tan() =
(R/)
(R/)
=
f
2(f
2
0
f
2
)
(15)
An example for three dierent systems of the magnitude and phase response is given in Figure 5. It
10 20 30 40
-45
-40
-35
-30
-25
response curves.nb 1
(a) magnitude response on log scale
10 20 30 40
-175
-150
-125
-100
-75
-50
-25
response curves.nb 1
(b) phase response (

)
Figure 5: Three systems of varying losses
can be seen that there is a strong amplication of motion at the so-called resonant frequency, in fact the
amplication is equal to Q factor. The phase is seen to always be negative - that is, the motion of the
mass is either in synchrony with the applied force if it is applied below resonance, or it lags the force by
up to 180

for frequencies well above resonance.

Lecture 5: Fourier Transforms and Dierential Equations 5-5
5.1.3 Solving the mass-spring system in the frequency domain
Lets now use the Fourier method to solve the dierential equation. One will see that it is possible to
obtain the transfer function of response function very easily. Regaining the time response from the
response function can be quite dicult but often one is not interested in that. All of the information
is contained within the response function. I note that the notation is going to get awful if we use the
logical variable for the time variation of force (i.e. F(t) for the time variation and F(f) for the frequency
domain). Anyway, to get around this problem, lets use n(t) andN(f) (for Newton) for the Force variable.
Returning to Eq. 1 we can write Newtons law as:
m x +D x +kx = n(t) (16)
Taking the Fourier Transform on both side we get:
m(2f)
2
X(f) + 2fDX(f) +kX(f) = N(f) (17)
where we have dened X(f) x(t) and N(f) n(t) as the Fourier conjugates of the time domain
functions.
(1
m
k
(2f)
2
+ 2f
D
k
)kX(f) = N(f) (18)
Reusing our denitions from above in Eqs 2-4 (2f
0
=

k
m
and Q =
2f
0
m
D
) we get:
(1
f
2
f
2
0
+
f
f
0
Q
)kX(f) = N(f) (19)
and thus:
X(f) =
1
k
1
1
f
2
f
2
0
+
f
f
0
Q
N(f) (20)
which is the equivalent of Eq. 10 obtained earlier. If we rewrite this equation as X(f) = G(f)N(f) then
the relating function, G(f), is termed the transfer function, response function or sometimes the impulse
function.
This nal terminology comes about from considering the situation of applying an impulse of force to a
system i.e. n(t) = (t). The Fourier conjugate of the impulse function is N(f) = 1 and thus trivially
X(f) = G(f) i.e. the systems response to an impulse is indeed its transfer function. The nal bit of
confusing nomenclature is that the time domain version of the transfer function is called the impulse
response of the system. This is really just the obvious conclusion to the earlier remarks - if we hit the
system with a short response then the Fourier transform of the resulting motion is just the transfer
function, and so its motion considered in the time domain is the impulse response of the system.
I note also that using this model we see three regimes of behaviour
1. when f f
0
, |G(f)| 1/k, 0 (spring regime)
2. when f = f
0
, |G(f)| = Q/k , /2 (resonance)
3. when f f
0
, |G(f)| =
1
m

1
2f

2
, (free mass)