221A Lecture Notes

Supplemental Material on Harmonic Oscillator

1

Number-Phase Uncertainty

To discuss the harmonic oscillator with the Hamiltonian

H=

p2
1
+ m 2 x2 ,
2m 2

(1)

we have dened the annihilation operator

ip
m
x+
,
2
h
m

a=

(2)

the creation operator a , and the number operator N = a a.

In some discussions, it is useful to dene the phase operator by

a = ei N ,
a = N ei .
(3)
Obviously the phase is ill-dened when N = 0, but apart from that, it is
a useful notion. It is particularly useful when we discuss the classical limit
N
1.
One can dene the phase eigenstate

ein |n .

| =

(4)

n=1

By acting the phase operator ei = a 1N ,

ei |

1
= a
N

ein |n 1

ein |n =
n=1

n=1

ei(m+1) |m = |0 + ei | .

(5)

m=0

It is almost an eigenstate of the phase operator, the failure due to the obvious
problem with n = 0 state as anticipated from its denition. We can also
calculate the inner products

m|eim ein |n =

| =
n=1 m=1

ein( ) .
n=1

(6)

This is almost the delta function

1
2

( ) =

ein( ) .

(7)

n=

The number eigenstate is expressed correspondingly as

2

|n =

dein | ,

(8)

which works for all n except for n = 0.

Again ignoring the subtlety with the n = 0 state, we can derive the
number-phase uncertainty principle. Study the commutator
1
1
1
[N, ei ] = [N, a ] = [N, a] = a = ei .
N
N
N

(9)

Therefore, roughly speaking,

N =i

(10)

Indeed, this makes sense on the phase eigenstate,

ein n|n = i

|N =
n=1

|.

(11)

Therefore, it leads to the canonical commutation relation

[N, ] = i,

(12)

leading to the uncertainty principle

1
N .
2

(13)

Coherent State of Harmonic Oscillator

Ive expanded discussions on the coherent state beyond Sakurai. Here is my

lecture note on this subject.
We saw that the uncertainty of the state |k is actually larger than the
minimum uncertainty
h

xp = (2k + 1).
(14)
2
2

It appears odd that states with larger k, which we expect to behave more
classicaly, are more uncertain. Moreover, expectation values of x and p vanish
for energy eigenstates
k|x|k = 0,

k|p|k = 0.

(15)

Therefore even for large k, the energy eigenstates do not share characteristics
we expect for classical oscillators.
But how do we make a classical oscillator actually oscillate? Lets say we
are talking about a pendulum. To make it oscillate, what we do is to exert a
force on it, pull the pendulum up, make sure the pendulum is settled in your
hand, and release it. Namely, pull, hold, and release. Why not do the same
in quantum mechanics?
To pull a pendulum, we have to add an additional term to the potential
1
V = m 2 x2 F x,
2

(16)

where F is the force we exert on the pendulum. Because the added term is
linear in x, we can complete the square
1
1
V = m 2 (x x0 )2 m 2 x2 ,
0
2
2

(17)

so that the pendulum settles to the position x0 = 0. The force for this
purpose is given by F = m 2 x0 . Because the pulled pendulum still has a
quadratic potential, it is a modied harmonic oscillator. It settles to a ground
state |0 , which is annihilated by the modied annihilation operator
a =

m
ip
(x x0 ) +
2
h
m

=a

m
x0 .
2
h

(18)

Therefore, the new ground state satises the equation

0 = a |0 = a

m
x0 |0 .
2
h

(19)

In other words,
m
x0 |0 .
2
h
This is an eigenequation for the annihilation operator a.
a|0 =

(20)

In general, the eigenstates for the annilation operator can be found as follows. Note that the annihilation operator is not Hermitian, and its eigenvalue
does not have to be real. Dene
f a

fn
fn n
|n ,
(a ) |0 =
|0 =
n!
n=0
n=0 n!

(21)

for a complex number f . If you act the annihilation operator on this state,

a ef a |0

fn
fn
a|n =

n|n1 =
n!
n!
n=0
n=1
n=1

fn
(n 1)!

|n1 . (22)

We used the fact that n = 0 state does not contribute because it cannot be
lowered by the annihilation operator. Changing the dummy index n to n + 1,

f n+1
fn

|n = f
|n = f ef a |0
n!
n!
n=0
n=0

(23)

Therefore, this state has an eigenvalue f for the annihilation operator. We

could have guessed it. The commutation relation [a, a ] = 1 says that roughly
speaking a = /a . Therefore, acting a just pulls out the exponent f .
We have not normalized the state yet. Working out the norm,

ef a |0

=
n,m

f n f m
n| |m =
n! m!

(f f )n

= ef f .
n!

(24)

Therefore, the following state is a normalized eigenstate of the annihilation

operator
2

|f = e|f | /2 ef a |0 ,
a|f = f |f .
(25)
This type of state is called coherent state.
Coming back to our problem, the pendulum just before the release is
therefore given by the coherent state
|

m
x0 .
2
h

(26)

Now the interest is in its time evolution. At t = 0, we release the pendulum.

In other words, we let the state evolve according to the original Hamiltonian
without an additional force. We can address the time evolution in Heisenberg
picture easier than in Schrdinger picture.
o
4

In Heisenberg picture, let us rst study the equation of motion for the
annihilation and creation operators. Because H = h(a a+ 1 ) and [a, a ] = 1,

2
we nd
d
i a = [a, H] = ha.
h

(27)
dt
Solving this equation is trivial,
a(t) = a(0)eit .

(28)

a (t) = a (0)eit .

(29)

Similarly, we nd
Solving the denition of the creation, annihilation operators backwards, we
nd the position and momentum operators
x=

(a + a ),
2m

p = i

m
h
(a a ).
2

(30)

Their time-dependence is then immediately obtained as

x(t) =

(aeit + a eit ),
2m

p(t) = i

m it
h
(ae
a eit ). (31)
2

On a coherent state, they have expectation values

h

(f eit + f eit ),
2m

f |x(t)|f

f |p(t)|f

= i

m
h
(f eit f eit ).
2

(32)
(33)

Note that I used f |a = (a|f ) = (f |f ) = f |f . Specializing to the

released pendulum, we have f = m x0 , and hence
2
h
f |x(t)|f
f |p(t)|f

= x0 cos t,
= mx0 sin t.

This result is the same as the classical pendulum.

(34)
(35)

Another important property of coherent states is that they have the minimum uncertainty. We can work it out easily in the following way.
f |x|f
f |p|f
f |x2 |f
f |p2 |f

f |a + a |f =
2m

(f + f ),
2m

(36)

m
h
m
h
f |a a |f = i
(f f ),
(37)
2
2
h

=
f |(a + a )2 |f =
(f 2 + f 2 + (f f + 1) + f f ), (38)
2m
2m
m
h
m 2
h
=
f |(a a )2 |f =
(f + f 2 (f f + 1) f f ).
2
2
(39)
= i

Therefore, we nd
(x)2 =
(p)2 =

,
2m
m
h
.
f |p2 |f ( f |p|f )2 =
2
f |x2 |f ( f |x|f )2 =

(40)
(41)

Finally, we obtain
xp =

,
2

(42)

indeed the minimum uncertainty state.

To sum it up, the coherent state represents the closest approximation of a
classical oscillator, with the minimum uncertainty and oscillating expectation
value of the position and the momentum.
We can obtain the same result in the Schrdinger picture, which is a litte
o
more technical than in the Heisenberg picture. The time evolution of the
coherent state can be obtained as
h
eiHt/ |f

h
= eiHt/ ef a |0 e|f |

2 /2

h
h
h
= eiHt/ ef a eiHt/ eiHt/ |0 e|f |
iHt/ a eiHt/
h
h

ei

it

2 /2

= ef e

= ef a e |0 e|f |
= |f eit eit/2 .

t/
h
2

|0 e|f |

2 /2

2 /2

eit/2
(43)

Therefore the expectation values of the postion and momentum operators

are
f, t|x|f, t

f eit |x|f eit

(f eit + f e+it )
2m
= x0 cos t,
= f eit |p|f eit

=
f, t|p|f, t

m
h
(f eit f e+it )
2
= mx0 sin t,

(44)

= i

(45)

where we used f = m x0 . The results agree with those in the Heisenberg

2
h
picture Eq. (32,33).
In quantum treatment of electromagnetism, light is described by a collection of photons. For a coherent light such as laser, the electric and magnetic
eld behave exactly like in the classical Maxwell theory. Laser is described
in terms of a coherent state.

Coherent State Wave Functions

Coherent state of course can be studied using the conventional wave functions. It takes a few tricks to workt it out, however.
We use the BakerCampbellHausdor formula. This is a formula important in the study of Lie algebras and Lie groups. The point is that the
product of two exponentials eX eY can be written in terms of many commutators,
eZ = eX eY

(46)

1
1
Z = X + Y + [X, Y ] + ([X, [X, Y ]] [Y, [X, Y ]])
2
12
1
([Y, [X, [X, Y ]]] + [X, [Y, [X, Y ]]]) +
48

(47)

See, for example, http://en.wikipedia.org/wiki/Baker-Campbell-Hausdorff

for more details.

We use this formula for ef a . We take

m
ip
X=f
x,
Y = f
.
2
h
2 m
h

(48)

For this purpose, we will not need any terms more than two commutators
because [X, [X, Y ]] = [Y, [X, Y ]] = 0 and Eq. (47) simplies drastically to
1

eX eY = eX+Y + 2 [X,Y ] .
Then we nd
f a

X+Y

=e

X Y 1 [X,Y ]
2

=e e e

=e

(49)

m
2
h

ip
x f 2 m f 2 /4
h

(50)

Now we are in position to work out the wave function for the coherent
state.
x|f

=
=
=
=
=
=

x|ef a |0 e|f | /2
m
2
2
f ip
h
h
x|ef 2 x e 2 m ef /4 |0 e|f | /2
m

f h
f 2 /4 |f |2 /2
h
2 m x x|0 e
h
e
ef 2 x e
m
1/4

m
2 h
2
2
f h
h
2 m x
h
ef 2 x e
emx /2 ef /4 e|f | /2

2
f 2 |f |2
m
h

m 1/4
m
+
exp f
x
xf

h
2
h
2
h
4
2
2 m
h
m

1/4

exp

m
xf
2
h

1
+ (f 2 |f |2 )
2

(51)

The explicit form of the wave function allows us to calculate the shape of
the probability distribution in real time. For the pulled, held, and released
oscilator, the time-dependent wave function is obtained for f = m x0 eit .
2
h
Therefore,
x|f, t =

1/4

exp

m
x x0 eit
2
h

1 m 2 2it
x (e
1) eit/2 .
2 2 0
h
(52)

We then nd the probability distribution

|(x, t)|2
m
m 2
m 2
=
exp
(x 2x0 x cos t + x2 cos 2t) +
x (cos 2t 1)
0

h
h

2 0
h
m
m
=
exp
(x x0 cos t)2 .
(53)

h
h

Therefore, it is always a Gaussian around x0 cos t which oscillators around

the origin with the amplitude x0 .

Coherent State Representation

One important caveat about the coherent states is that they form an overcomplete set of states. It is easy to calculate

g|f =
n=0 m=0

g n f m
(g f )n

= eg f .
n| |m =
n!
n! m!
n=0

(54)

Even when g = f , it does not vanish.

This is not a paradox. When we proved in class that the eigenstates of an
operator with dierent eigenvalues are orthogonal to each other, we assumed
that the operator was hermitian. The coherent states are eigenstates of the
annihilation operator, which is not hermitian. Therefore, the coherent states
do not form an orthonormal set.
Nonetheless, one can come up with the coherent state representation,
taking the coherent states as the basis kets. This is because of the following
completeness condition,
d2 f

|f ef f f |.

1=

Here, d2 f = df1 df2 for f = f1 + if2 and f1,2

Let us prove the completeness relation.

R.

n,m

d2 f f f f n
f m
|n m|
e

n!
m!

n,m

|f |d|f |d f f f n+m ei(nm)

e
|n m|

n!m!

n,m

d2 f

|f ef f f | =

|f |d|f | f f f n+m 2n,m

e
|n m|

n!m!

=
=

2|f |d|f |

=
n

dt
n

|f |2n |f |2
e
|n n|
n!

tn t
e |n n|
n!
9

(55)

=
n

1
(n + 1)|n n|
n!
|n n| = 1.

(56)

In the third last line, we changed the variable to t = |f |2 . From the completeness relation for the energy eigenstates |n , the last expression is indeed
the unit operator.
Therefore, any state can be expressed as a linear combination of coherent
states. In particular, the energy eigenstates are
|n =

d2 f

|f ef f f |n =

n
d2 f
f
|f ef f .

n!

(57)

The coherent state representation is quite interesting because the twodimensional integral on f can be regarded as a phase space integral. Recall
the denition of the annihilation operator Eq. (2) and setting f = a in this
representation, we nd
f=

m
ip
x+
.
2
h
m

(58)

Therefore, we can identify

1
d2 f
=

dx dp
m
m 1
dx
dp =
,
2
h
2 m
h
2
h

(59)

indeed the normal phase space volume.

Let us see how one can calculate expectation values of operators using the
coherent states. Note that any operator made up of x and p can be rewritten
in terms of a and a . Furthermore an operator can be brought to the form
that all annihilation operators are moved to the left, and creation operators
to the right using their commutation relations. Therefore we can cast any
operators to the form O = an am without a loss of generality. Then its
expectation value can be calculated as
|O|

=
=

|an am |
d2 f

|an |f ef f f |am |

The operators of the form am an are said to be normal ordered. Maybe I should
call those in the order we use here abnormally ordered.

10

d2 f n m

f f
|f ef f f |

d2 f n m

f f | f | |2 ef f .

=
=

(60)

Therefore, the combination | f | |2 ef f can be viewed as the probability

density on the phase space, where the operator an am is simply brought to
the numbers f n f m .
This observation allows us to view a state as a probability density on
the phase space. First of all, the classical motion is along a zero-thickness
circle on the phase space. It is always at a point at a given moment, and
the point moves along the circle as time evolves. This is shown as the rst
picture in Fig. 1. Note that the time corresponds to the phase, while the
energy to the number.
On the other hand, the quantum mechanical energy eigenstates have the
phase space density
2

| f |n |2 ef

|f |2n |f |2
f n

ef f =
e
.
=
n!
n!

(61)

The main support for this distribution is shown in the middle picture of Fig. 1.
It basically forms a ring in the phase space with the constant energy, smeared
a little bit so that the energy varies roughly from n to (n + 1) . The
h
h
area is given by its uncertainty xp = (2n + 1) /2, while the higher energy
h
states appear as successively outward rings. The fact that it is spread out
over the entire ring is a reection of the energy-time uncertainty principle.
Because we have specied energy, we dont know anything about time, and
we cant say at what phase it is.
The coherent state is very close to a point on the phase space resembling
the classical mechanics. The phase space density for the normalized state

|g eg g/2 is
| f |g eg

g/2

|2 ef

= |ef g eg

g/2

|2 ef

= e|f g| .

(62)

This is a two-dimensional Gaussian centered at f = g, and its main support

is depicted in the right picture of Fig. 1. It has the minimum uncertainty
and its area is much smaller than the energy eigenstate. The patch moves
along the circle clockwise just like the classical oscillator.

11

Figure 1: Classical oscillator is a point on the phase space (x, p space) moving
along an elliptic orbit. The quantum mechanical energy eigenstate is spread
out along the ellipse with no notion of motion. The uncertainty xp is
larger for higher levels because of a constant width around the oribit. The
coherent state is a patch of the minimum uncertainty, and the whole patch
moves along the classical orbit.

12