Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

Physics 303/573

Notes on the Harmonic Oscillator Part II


October 5, 2015

The forced harmonic oscillator

The key to understanding the forced harmonic oscillator is to use the linearity of the
equation. Suppose we have two solutions x1 (t) and x2 (t) that satisfy
x
1 + 2 x 1 + 02 x1 =
x
2 + 2 x 2 + 02 x2 =

1
F1 (t)
m
1
F2 (t)
m

(1.1)

for external driving forces F1 (t) and F2 (t), respectively. Then if we consider a linear
combination X(t) = ax1 (t) + bx2 (t), we find it satisfies
1
2

X+2
X+
x1 +b
x2 +2a x 1 +2b x 2 +02 [ax1 (t)+bx2 (t)] = (aF1 (t)+bF2 (t)) (1.2)
0 X = a
m
Also, because of the linear structure, the most general solution with a driving force F (t) is
a linear combination of any solution with F (t) and the general homogeneous solution, that
is the solution with no driving force.
This allows us to solve the harmonic oscillator with an arbitrary external force and initial
conditions.

1.1

Harmonic driving force

Because of the linear structure, we can work with complex driving force F (t) and then just
take the real part at the end; the real part of the solution xF (t) with the complex driving
force F (t) is the solution with a driving force that is the real part of F (t):
x
F + 2 x F + 02 xF =

1
F (t)
m

d2
d
1
Re(xF ) + 2 Re(xF ) + 02 Re(xF ) = Re(F (t))
2
dt
dt
m
(1.3)
1

So we start with

1
F eiF t
(1.4)
m
where F is a complex constant. Since the derivative of an exponential gives back an
exponential, we know that a solution exists that is proportional to eiF t ; we put it into the
equation with an arbitrary amplitude A and find:
x
+ 2 x + 02 x =

x = AF eiF t AF (iF )2 eiF t + 2AF (iF )eiF t + AF 02 eiF t =

1
F eiF t (1.5)
m

and hence
AF [(F )2 + 2iF + 02 ] =

1
F
m

AF =

F
1
2
2
m 0 F + 2iF

(1.6)

We can multiply the numerator and denominator by the complex conjugate of the denominator to find:
1 F (02 F2 2iF )
AF =
(1.7)
m (02 F2 )2 + 4 2 F2
so that we finally have the complex solution
x(t) = AF eiF t =

1 F (02 F2 2iF ) iF t
e
m (02 F2 )2 + 4 2 F2

(1.8)

The real solution is just the real part of (1.8).


1.1.1

Undamped limit

When there is no damping, = 0 and we have simply


AF =

1
F
2
m 0 F2

(1.9)

Notice that the sign of AF depends on whether the driving frequency F is greater or less
than the natural frequency 0 :
If F < 0 , then AF > 0 and the motion is in phase with the driving force, that is
the mass on the spring just tracks the driving force.
If F > 0 , then AF < 0 and the motion is out of phase with the driving force, that
is the mass on the spring moves against the driving force.
Of course, in the undamped case, if F 0 , then AF and the motion blows
up.

1.1.2

The damped case

In the general damped case, we consider (1.7). The magnitude of the amplitude A (for
brevity, in the rest of this section, will just write A rather than AF ) is
|A| =

AA =

1
|F |
q
m ( 2 2 )2 + 4 2 2
0

(1.10)

This reaches a maximum as a function of the driving frequency F when



d
d
|A| = 0 0 =
(02 F2 )2 + 4 2 F2 = 2(02 F2 ) + 4 2
2
dF
dF

(1.11)

and hence
F2 = 02 2 2 = 2 2

(1.12)

This is called the resonant frequency, and is a bit lower than the frequency at which the
spring oscillates when it is not driven. For > 12 0 , there is no resonant frequency, and
the amplitude simply decreases the higher the driving frequency.
In addition to the magnitude of the amplitude, we are interested in the phase shift between
it and the driving force. Defining
A
|A| i
e
=
F
|F |

(1.13)

we find (using Eulers identity ei = cos i sin and comparing to (1.7))


tan =

2F
02 F2

(1.14)

As in the undamped case, this changes sign depending on whether F is less than or greater
than 0 ; for F = 0 , tan and hence
=

(1.15)

that is, the motion is 90 degrees out of phase with the driving force. To find the correct
quadrant of the angle, recall that
sin =

2F
>0
|AF |

(1.16)

1.1.3

Transients and boundary conditions

The most general motion of a driven oscillator is a linear combination of a particular


solution and the general solution when the driving force is turned off; when there is any
damping at all, the terms in the solution that are there without the driving force will
eventually die out, and hence are called transients, but they are important for the boundary
conditions. Lets consider an example of a system driven by an external force that starts
at rest in its equilibrium position. The demonstrations with the rubber band and with the
wok were all of this kind.
For such physical systems the driving force is real, so we need to take the real part of our
solution. We can, for example, take F real, so the the force is Re(F eiF t ) = F cos(F t)
Then we have a combination of the real part of (1.8) and (2.52) of the previous lecture:
x(t) = Re(AeiF t ) + xmax et cos(t + )
= |A| cos(F t ) + xmax et cos(t + )
where recall
q
2F
1
F
= 02 2 , |A| = q
, tan = 2
m ( 2 2 )2 + 4 2 2
0 F2
0
F
F

(1.17)

(1.18)

and xmax and are the constants we want to determine. Imposing x(0) = v(0) = 0, we
find
0 = |A| cos + xmax cos
0 = |A|f sin + xmax ( cos + sin )

(1.19)

which we can solve by


xmax = |A|

cos
cos

with
1

tan = (F tan + ) =

(1.20)


02 + F2
02 F2


(1.21)

An example is shown in Figure 1. When the driving frequency is not near resonance, after
a few cycles the random transients die out; near resonance, the motion builds steadily to
its maximum amplitude. This is just what we found in the demonstration. Notice that
when the driving frequency is low F < , the motion is (nearly) in phase with the driving
force, when it is equal to the natural frequency, F = , it is 2 or 90 out of phase, and
when it is high, F > , it is (nearly) or 180 out of phase. In the figure, the driving
frequency is held constant and the natural frequency is varied, as that is easier to visualize.

Figure 1: Transients and phase shifts in the underdamped forced harmonic oscillator. Here
1
F = 1 and = 10
, which is very underdamped. You can see the phase shifts of the
motion as well as the transients at the beginning. Color code: The driving force is cos(t).
When F > = 12 , the motion is out of phase by (almost exactly) . When F < = 2,
the motion is basically in phase. When F = = 1, the motion is near resonance and is
out of phase by 2 .

1.2

General periodic force: Fourier series

When the driving force is an arbitrary periodic function, we can use the linearity of the
harmonic oscillator and simply take the sum of the results of the previous result. Systematically, any periodic function F (t) with period T
F (t) = F (t + T ) ,

T =

2
2
F =
F
T

(1.22)

can be expanded in a basis of simple pure frequency modes in the same way that a note
on a guitar string can be described in terms of a fundamental frequency and its harmonics
or overtones:
!

X
1 X inF t 1
F (t) =
Fn e
=
F0 +
(Fn + Fn ) cos(nF t) + i(Fn Fn ) sin(nF t)
2 n=
2
n=1
(1.23)
This is real if the Fourier coefficients obey Fn = Fn ; then

X
1
Re(Fn ) cos(nF t) Im(Fn ) sin(nF t)
F (t) = F0 +
2

(1.24)

n=1

It is easy to find the Fourier coefficients Fn for a periodic function; note that1

1
Z T
inT1 (einT 1) = in2
(ein2 1) = 0 n 6= 0
1
int
dte
=
1
T 0
n=0

(1.25)

Using this, it is easy to see that


2
Fn =
T

F (t)einF t =

2
F

F (t)einF t

(1.26)

Notice that when the force F (t) is real, we can complex conjugate and find that the Fourier
coefficients automatically obey Fn = Fn .
If we now go back to the harmonic oscillator, then we can use the solution (1.8) to (1.4)
separately for each term in (1.23) to find
xn (t) = AnF einF t
where
AnF =
1

(1.27)

Fn 02 (nF )2 2inF
2m [02 (nF )2 ]2 + 4n2 2 F2

Clearly, since eint is periodic, we could just as well integrate from

T
2

(1.28)
to

T
2

Because of the linearity of the equations, as discussed at the beginning of these notes, we
can find the general solution simply by adding these up; note that the F0 term just gives
F0
a shift in the position by x0 = 2m
:

X
Fn 02 (nF )2 2inF
inF t
2 (n )2 ]2 + 4n2 2 2 e
2m
[
F
0
F
n=
n=
n=
(1.29)
To take the real part, it is convenient to define the phases of the components of the driving
force

x(t) =

xn (t) =

AnF einF t =

Fn = |Fn |ein

Fn = |Fn |ein

X
1
F (t) = F0 +
|Fn | cos(nF t n ) (1.30)
2
n=1

as well as the phase shift of AnF (see 1.13):


|AnF | in
AnF
=
e

Fn
|Fn |

tan n =

2nF
02 (nF )2

(1.31)

where we have used (1.14) to get the second expression. Then we can write (1.29) as

x(t) =

|AnF | ei(nF tn n )

(1.32)

n=

where the magnitude of AnF is given by (1.10):


|AnF | =

|Fn |
1
q
2m [ 2 (n )2 ]2 + 4n2 2 2
F
0
F

(1.33)

Finally, we get the real motion:


x(t) =

X
F0
|Fn |
cos(F t n n )
q
+
2m
m
[ 2 (n )2 ]2 + 4n2 2 2
n=1

1.3

(1.34)

General force: Fourier transform

To study non-periodic driving forces, we use the same basic method, but take the limit as
the period T . To do this, we need to make some small redefinitions; we take the
limits of integration from T2 to T2 , we define rescaled Fourier coefficients fn = T2 Fn , so
that, plugging into (1.23,1.26), we find:
fn =

T
2

T2

dt F (t)e

t
i( 2n
T )

X
fn i( 2n )t
and F (t) =
e T
T
n=

(1.35)

Taking the limit T , and defining s = Tn , we have fn /T dsf(s), we find:


Z
Z
i2st

f (s) =
dt F (t)e
and F (t) =
ds f(s)ei2st

(1.36)

or, equivalently, letting = 2s and defining F () = f(s)


Z
Z
1
it

F () =
dt F (t)e
and F (t) =
d F ()eit
2

(1.37)

We can now find x(t) as before; using superposition and the basic solution (1.8), we
have
Z
it

x(t) =
d A()e
(1.38)

where

F () 02 2 2i

A()
=
2m [02 2 ]2 + 4 2 2

(1.39)

As before, we find the real solution by taking the real part of this.

1.4

The electrical analog

We begin with the familiar laws of a circuit with a resister, a capacitor, and an inductor
in series, driven by an external time varying voltage. Ohms law tells us that the voltage
drop V across a resistance R is:
V = IR
(1.40)
where the current I is rate of change of the charge Q:
I = Q

(1.41)

When we put a voltage across a capacitor with capacitance C, the charge stored is
Q = CV

(1.42)

Finally, the voltage induced by a current through an inductor with inductance L is

V = LI = LQ

(1.43)

The voltage drop across all three, when they are in series, is simply the sum of the voltage
drops, so we get:
+ RQ + 1 Q = V (t)
LQ
(1.44)
C
which we recognize as a driven damped harmonic oscillator with the identifications:
8

Electrical

Mechanical

Charge

Position

Current

Velocity

Inductance

Mass

Resistance

Damping

(Capacitance)1

1
C

Spring constant

External voltage

V (t)

External force

F (t)

Table 1: The electricalmechanical analog


Dividing through by the inductance, we get the equation
+ 2 Q + 02 Q = 1 V (t)
Q
L

(1.45)

where

1
R
,
02 =
(1.46)
2L
LC
We can differentiate (1.45) and get an equation for the current in terms of the derivative
of the driving voltage:
1
I + 2 I + 02 I = V (t)
(1.47)
L
p
The natural frequency is = 02 2 :
r
R2
1
=

(1.48)
LC
4L2
=

The quality factor (not to be confused with the charge), is given by


damping is
r
0
1 L
=
2
R C

2 ,

which, for small


(1.49)

Everything in our analysis of the harmonic oscillator applies to the electrical analog. However, there is one thing that is slightly different. When we talk about resonance, we are
interested in the frequency that maximizes the amplitude of the current I rather than the
if the amplitude of Q(t) is Q0 , then the amplitude
amplitude of the charge Q. Since I = Q,
of the current is |I0 | = F Q0 . This means that instead of maximizing |A| as in (1.10), we
want to maximize |A|F :
|I0 | =

V0
F
q
L ( 2 2 )2 + 4 2 2
0

(1.50)

Remarkably, when we maximize this, we find that resonant frequency is the undamped
frequency:
r
1
res = 0 =
(1.51)
LC

The WKB approximation

The WKB approximation is very useful for certain problems in quantum mechanics, and
can be introduced here. It applies to harmonic oscillators where the frequency varies slowly
in time. An example might be a pendulum with a spring instead of a fixed length rope;
as the length of the spring varies, the instantaneous frequency of the pendulum varies as
well. If this variation is sufficiently slow, we can use the WKB approximation to describe
the motion of the mass on the end of this spring pendulum.

2.1

Intuitive approach

We consider an equation of the form


x
+ 2 (t) x = 0

(2.52)

where we assume that 2 (t) varies slowly in time. We could easily include a damping
termthe basic ideas would apply as well; you can try to work it out for yourself.
For slowly varying (t) 6= 0, the solution should be approximately
x C (+) ei

dt (t)

+ C () ei

dt (t)

(2.53)

R
where C () are some constants; notice that for constant , dt (t) = t, so this reduces
to the usual result for strictly constant (t) = 0 . Of course, to get a real solution, we
need to take C () = (C (+) ) .
To do better, we write
x = e(t) ,

(t) = a(t) + i(t)

(2.54)

and assume that a(t) is slowly varying. Plugging this into (2.52) gives (up to an overall
factor of e(t) )

+ ()
2 + 2 = 0
(2.55)
The real part of this equation is
2 + 2 = 0
a
+ (a)
2 ()

10

(2.56)

and the imaginary part is


+ 2a = 0

(2.57)

This second equation implies a = 21 , which we can integrate to give:


a(t) = c

1
ln (t)
2

(2.58)

where c is an integration constant; exponentiating, we have:


C
A(t) = q

(t)

C = ec

(2.59)

Returning to the real part, we use the condition that a varies more slowly than to drop
2 >> a
the a
and (a)
2 terms (that is we assume >> a,
()
). To leading order,
= (t) 0 (t) =

dt0 (t0 )

(2.60)

which combines with our expression for A(t) to give an improved approximation for x(t):
C () i R dt (t)
C (+) i R dt (t)
e
+p
e
x(t) p
(t)
(t)

(2.61)

This is the standard leading term in the WKB approximation.


R
We can get higher corrections as follows: starting with the leading solution 0 = dt (t),
we have a0 (t) = c 12 ln (t), which we use to get the next order corrections to (t) and
keep iterating. Thus, the first correction gives:
a =

,
2

a
=

1
+
2 2

 2

(2.62)

so, keeping these leading corrections to (2.56), we get

3
2 = 2
()
+
2 4

 2

(2.63)

which we can integrate to get an improved approximation for (t), which in turn gives an
improved approximation for a(t), etc.

11

2.2

Systematic approach

Actually, it is conventional to organize this expansion in a more systematic way. Let us


introduce a characteristic large frequency 1 and define:
=

(t)

1X
n (t) n

(2.64)

n=0

Plugging this into

+ ()
2 + 2 = 0

(2.65)

and multiplying through by 2 , we get

n (t) n+1 +

n=0

n (t) m (t) n+m + 2 (t) = 0

(2.66)

m (t) nm (t) n + 2 (t) = 0

(2.67)

n=0 m=0

We can rewrite the double sum to get:

X
n=0

n (t) n+1 +

X
n
X
n=0 m=0

Now we expand in powers of ; the leading term is


( 0 (t))2 + 2 = 0

(2.68)

which implies that 0 gives leading contribution to 0 (t) that we found above:
Z
i
0 = i dt (t) 0 = 0

(2.69)

The next term is

0 (t) + 2 0 1 = 0

(2.70)

which imples that 1 just gives the leading contribution to a(t) that we found above (the
dependence on can be absorbed into the constant c:
1 = c0

1
1
ln (t) = a0 (t) = c ln (t)
2
2

c = c0

1
ln
2

(2.71)

The next equation is

1 + ( 1 )2 + 2 0 2 = 0

(2.72)

which can be solved for 2 . The general equation for n is


2 0 n +
n1 +

n1
X
m=1

12

m nm = 0

(2.73)

2.3

A related exactly soluble problem

Recall that in the previous set of notes, we considered the equation


= 0 where D = d
(D + A)(D + B)x = x
+ (A + B)x + (AB + B)x
dt

(2.74)

and said that the most general solution to this equation can be expressed as a solution to
the system of first-order equations
x + B x = xA (t) ,

x A + A xA = 0

(2.75)

We can actually give the solution to this explicitly in terms of integrals, even when A and
B are functions of time. The equation x A + A xA can be integrated to give:
Rt

xA (t) = e A
 Rt 
To solve the equation for x, we substitute x = e B y; then

(2.76)

 Rt 
x = e B (y B y)

(2.77)

 Rt  Rt 
Rt
y = e B e A = e (BA)

(2.78)

and hence (2.75) reduces to

which we can immediately integrate to


Z
y=

t R

(BA)

(2.79)

and hence we find the total solution is


 R 0

 Rt 0 0 Z t
t
00
00
00
x = e dt B(t )
dt0 e dt (B(t )A(t ))

(2.80)

where in the final expression (though not in the intermediate expressions (2.76)-(2.79)) I
have been careful to explicitly indicate the dummy integration variables.
If we want to have no damping, then from (2.74), we need A + B = 0; for a mechanical
system, we also require that the 2 (t) is real, so A2 + A must be real. However, we also
> 0.. If we write A = a + ib,
want a stable system, which means that 2 (t) = (A2 + A)
then

2 = b2 a2 a i(2ab + b)
(2.81)

13

the condition that the frequency is real implies


a=

b
2b

(2.82)

Plugging this back into (2.81), we find


3
2 = b2
4

b
b

!2
+

b
2b

(2.83)

Note the striking resemblance of (2.82),(2.83) to (2.62),(2.63). If we plug




b
1d
B = A =
ib =
ln(b) ib
2b
2 dt

(2.84)

into our general solution (2.80), we find


x(t) =
=

Z
Z
1  i R t b
1  i R t b  t  2i R t b 
e
b e
= e
b
b


1  i R t b  (+) i  2i R t b 
e
C
+
e
2
b
C (+)  i R t b  C ()  i R t b 

+
e
e
b
b

i d  2i R t b 
e
2 dt

(2.85)

where C () are two integration constants. Notice that except for the correction (2.83) to
, this is precisely the form (2.61). As for the WKB approximation, to make x(t) real, we
need C () = (C (+) ) .

Nonlinear Harmonic motion

We began our discussion of the Harmonic oscillator by noting that it arises universally if one
Taylor expand a smooth force around an equilibrium point. Lets now consider the higher
terms in the expansion. Because the system is nonlinear, we can no longer simply consider
the superposition of different solutions, and the initial conditions change the solution in
complicated ways. Let us start by proceeding naively; for a force
F (x) = m(02 x + f (x))

14

(3.86)

where we have defined the normalized nonlinear terms by f (x), we expand both the force
and the solution in a small parameter :
f (x) =
x(t) =

X
n=2

fn n1 (x(t))n
xn (t)n

(3.87)

n=0

Plugging in the expansion for the position x(t) into the expansion for force F (x), we
find
f (x) = f2 x20 + 2 (f3 x30 + 2f2 x0 x1 ) + 3 (f4 x40 + 3f3 x20 x1 + f2 (x21 + 2x0 x2 )) + ... (3.88)
We can now expand the nonlinear harmonic oscillator equation:
x
+ 02 x = f (x)

(3.89)

(for simplicity we consider the undamped case, but not much changes with damping), and
expand it order by order in
O(1) :

x
0 + 02 x0 = 0

O() :

x
1 + 02 x1 = f2 x20

O(2 ) :

x
2 + 02 x2 = f3 x30 + 2f2 x0 x1

O(3 ) :

x
3 + 02 x3 = f4 x40 + 3f3 x20 x1 + f2 (x21 + 2x0 x2 ))

(3.90)

etc. The O(1) equation is the usual linear harmonic oscillator and is easily solved, and all
the subsequent equations get an effective driving force from the lower order terms.
If the boundary conditions are such that we take x0 (t) = c0 eit or x0 (t) = c0 eit , that
is, a complex solution which is a pure exponential (which arises in quantum mechanics
but is unphysical for classical oscillating masses), then (3.90) is the whole story, and each
correction xn (t) has a frequency n induced by the forcing term.
However, if the boundary conditions are such that x0 (t) has terms involving both eit and
eit simultaneously, then the driving terms which arise from powers and products of the
xn s contain admixtures of lower frequencies, and it is more useful to resum the series. The
general discussion is quite complicated, so I will focus on some examples.
Consider a quadratic term x2 ; notice it breaks the symmetry F (x) = F (x). When
this symmetry is not broken, the center of the motion (the time-independent term) stays
at xc = 0; for F (x) 6= F (x), which would be relevant to, e.g., a rubber band, the center
of the motion xc depends on the amplitude.
x
+ 02 x = x2
15

(3.91)

We write
x(t) = xc + A cos(t) + ...

(3.92)

where we fix the amplitude A of the fundamental oscillation and see what we get; to leading
order, we have
(02 2 )A cos(t) + 02 xc = [x2c + 2xc A cos(t) + A2 cos2 (t)]
A2
= [x2c + 2xc A cos(t) +
(cos(2t) + 1))]
2

(3.93)

Collecting the constant terms, we find:


02 xc

x2c

2
1
+ A2 xc = 0
2

p
04 22 A2
A2
+ O(2 )
=
2
202

(3.94)

which shows that the center of the motion is shifted from the equilibrium point as the
amplitude increases by the nonlinearity; this does not happen when F (x) = F (x).
Notice also that when the nonlinearity is too strong or the amplitude is too large, (3.94)
breaks down. This makes sense: a quadratic force will dominate over the linear restoring
force when it becomes strong enough, and the mass will eventually run away. This is best
seen by computing the potential:
F (x)
d V (x)
= 02 x + x2 =
m
dx m

1
1
V (x) = m02 x2 mx3
2
3

(3.95)

which clearly is unbounded for x large, so if the oscillations are large enough, the mass will
2
escape the minimum at x = 0 and go over the maximum at x = 0 .
Returning to (3.93) and looking at the terms linear in cos(t), we find
(02 2 )A = 2xc A 2 = 02 2xc = 02

2 A2
+ O(3 )
02

(3.96)

so the fundamental frequency is shifted from 0 by an amount that depends on the


amplitude A:
2 A2
= 0 [1
+ O(3 )]
(3.97)
204
Finally, we see that there is a term proportional to cos(2t) in (3.93); this looks like a
driving force and gives rise to higher harmonics which may be calculated using the methods
developed above.

16

You might also like