Density Matrix for Harmonic Oscillator

Jens Michelsen
May 7, 2010
1 QuantumMechanics and QuantumStatistics
1.1 Classical Mechanics vs QuantumMechanics
In classical mechanics the state of the systemis uniquely dened by a point in phase-
space, e.g. for a one-particle systemX=(x, p). In quantummechanics the state of the
system is dened by a state-vector in Hilbert space:
| H (1)
This abstact object can be put in a more concrete formby projecting it onto some ba-
sis we can measure, for instance if we want to knowthe position of a particle then the
projection onto the position-basis gives us the wave-function of the particle:
x| (x) (2)
(x) is interpreted as the probability amplitude of the particle to be at position x. The
probability density is given by:
(x) |(x)|
2
=(x)

(x) (3)
Using this probability density we can calculate the mean- or expectation value of, for
instance the position of the particle.
x =
_
dx (x)x

(x) =
_
dxx(x) (4)
A more general object is the density "matrix"
(x, x

) =(x)

(x

) (5)
1
More accurately this is the density matrix in position-basis
1
.
With this density matrix we can evaluate the expectation value of the momentum
2
p =i

x
:
p =
_
dx
_
i

x
(x)
_

(x) =
_
dx
_
i

x
_
(x, x

=x
(6)
Note that (x) =(x, x).
From the time-dependant Schrdinger eq. and its complex conjugate:
i
t
(x) =

H(x), i
t

(x) =

(x)

H (7)
we nd the equation of motion for the density matrix
i
t
(x, x

) =
_
i
t
(x)
_

(x

) +(x)
_
i
t

(x

)
_
=

H(x, x

) (x, x

)

H
=[

H, ]
(8)
which is called the quantumLiouville equation.
1.2 QuantumStatistics
So far we have only discussed the intrinsic randomness due to the probabilistic na-
ture of QM. In general we also have randomness due to initial conditions etc. We can
imagine that our system is in one of many states |
n
(or in postion basis
n
(x)).
The statistical average is then to be taken wrt an ensemble of states |
n
weighted by
some probability w
n
:
x =

n
w
n
x

n
=

n
w
n
_
dx
n
(x)x

n
(x) (9)
If we then dene a new statistical density matrix
(x, x

) =

n
w
n

n
(x)

n
(x

) (10)
1
The more general form of the density matrix of a pure state is = || and we can de-
ne averages as A = Tr(

A ) =
_
dx
_
dx

A|xx||x

=
_
dx
_
dx

A(x

, x)(x)

(x

) =
_
dx
_
dx

A(x, x

)(x, x

) which for

A = x with x

| x|x =x(x x

) gives us x =
_
dxx(x, x)
2
In the Dirac Bra-Ket notation we have x

| p|x =(x x

)(i

x
) so that
p =Tr( p ) =
_
dx
_
dx

| p|xx| |x

=
_
dx
_
dx

(xx

)(i

x
(x, x

)) =
_
dx
_
i

x
(x, x

)
_
x

=x
2
the averages can again be written
x =
_
dx x(x, x)
p =
_
dx
_
i

x
_
(x, x

=x
(11)
The problem then is to determine the statistical density matrix (x, x

) for a given
system. Inwhat follows I will derive the density matrix for a Harmonic oscillator using
the derivation due to Feynman:
2 Density Matrix of Harmonic Oscillator a la Feynman
The equilibriumdensity matrix should be one which does not depend on time:
t
=
0. We then have from the quantum Liouville equation:
0 =i
t
=[H, ] (12)
we may note that e
H
satises this equation and we will take this to be our density
matrix. How do we nd the representation in position basis? Feynman noted that
formally you can write

= H
_
from

e
H
= He
H
_
(13)
which gives us a differential equation to be solved with the initial conditions
=0 =e
0 H
=1 (14)
which in position basis corresponds to
(x, x

, =0) =(x x

) (15)
Lets solve Eq. (13) for the Harmonic oscillator which in position basis has the Hamil-
tonian
H =
p
2
2m
+
m
2
2
x
2
=

2
2m

2
x
2
+
m
2
2
x
2
(16)
Using scaled coordinates
X =
_
m

x =

X
=
_

m

x
(17)
3
in which the Hamiltonian takes the form
H =

2
_

2
X
2
+X
2
_
(18)
by dening f =(/2) we get the nicer looking differential equation

f
(X, X

, f ) =
_

2
X
2
+X
2
_
(X, X

, f ) (19)
with the initial condition
(X, X

, f =0) =(x x

) =(
_

m
(X X

)) =
_
m

(X X

) (20)
where I used the identity (g(y)) =|g

(y
0
)|
1
(y y
0
) where y
0
is a root of f (y).
We solve this differential equation similar to how we solved the Fokker-planck equa-
tion : by making an ansatz and then comparing coefcients in powers of X.
Ansatz: =e
(a( f )X
2
+b( f )X+c( f ))
(21)
Note that X

only enters differential equation as a parameter a, b, c may depend

. The left hand side of Eq. (19) can be written

f
=
_
a
f
X
2
+
b
f
X +
c
f
_
(22)
and the right hand side
RHS:
_

2
X
2
+X
2
_
=

X
_

X

_
+X
2

=

X
_
(2aX +b)
_
+X
2

=
_
2a (2aX +b)
2
+X
2
_

=
_
2a 4a
2
X
2
4abX b
2
+X
2
_

(23)
Putting it all together we have
_
a
f
X
2
+
b
f
X +
c
f
_
=
_
2a 4a
2
X
2
4abX b
2
+X
2
_
(24)
Cancelling onbothsides andcomparing powers of X gives us three equations:
X
2
:
a
f
=14a
2
a =
1
2
coth2f
X
1
:
b
f
=4ab b =
A
sinh2f
X
0
:
c
f
=2a b
2
c =
1
2
ln
_
sinh2f
_
+
A
2
2
coth2f ln[B]
(25)
4
The last term can be rewritten into the form
c =
A
2
2
coth2f ln
_
B
_
sinh2f
_
(26)
Inserting this into the Ansatz. Eq. (21):
=e

1
2
_
coth2f X
2
+
2AX
sinh2f
+A
2
coth2f
_
e
ln
_
B

sinh2f
_
=
B
_
sinh2f
e

1
2
_
coth2f X
2
+
2AX
sinh2f
+A
2
coth2f
_
=
B
_
sinh2f
e

1
2sinh2f
[cosh2f (X
2
+A
2
)+2AX]
(27)
Nowwe remember the initial condition( f =0) =

m/(XX

), andlook at small
f:
f 0
_
cosh2f 1
sinh2f 2f
(28)
which means that in this limit we get

B
_
2f
e

1
4f
[X
2
+2AX+A
2
]
=
B
_
2f
e

1
4f
(X+A)
2
(29)
This becomes a delta function as f 0 if we take A = X

with the normalization

m/2. The density matrix then takes the form:

, f ) =
_
m
2sinh2f
e

1
2sinh2f
[cosh2f (X
2
+X
2
)2X X

]
(30)
Inthe original variables x =

/mX and=(2/) f we get the nal solution

, ) =
_
m
2sinh2f
e

m
2sinh()
[cosh2f (x
2
+x
2
)2xx

]
(31)
From this you are to calculate the mean square deviations for x, p.
3 Density matrix in energy-eigenbasis
I will nowcalculate the averages x, x
2
, p and p
2
by he use of ladder operators.
This methodis generally simpler because the density matrix has a muchsimpler form
in energy-eigenbasis than in position basis:

nn
=e
E
n

nn
(32)
5
The average value of some operator

A can in this basis be written
A =
1
Z
Tr[

A] =
1
Z

n
|

A
n
e
E
n
(33)
where |
n
is an energy eigenstate: H|
n
=E
n
|
n
. For example, we have
x =
1
Z
Tr[x] =
1
Z

n
|x
n
e
E
n
p =
1
Z
Tr[p] =
1
Z

n
|p
n
e
E
n
(34)
In order to evaluate the averages we need to knowthe matrix elements
n
|x
n
and

n
|p
n
. This is where the ladder operators come in.
Starting from the Hamiltonian:

2
_

2
X
2
+X
2
_
=

2
_
K
2
+X
2
_
(35)
where we dene K =i

X
. Fromthe commutation relations [x, p] =i we nd
[X, K] =i (36)
The ladder operators are dened as
a =
1

2
(X +i K)
a

=
1

2
(X i K)

X =
1

2
_
a

+a
_
K =
i

2
_
a

a
_
(37)
Inserting this into the Hamiltonian:
H =

2
_

1
2
_
a

a aa

+aa
_
+
1
2
_
a

a +a

a +aa

+aa
_
_
=

2
_
a

a +aa

_
(38)
Here we need to keep track of the ordering because a

and a do not commute since

] =aa

a =1 (39)
Using this in the Hamiltonian we get
H =
_
a

a +
1
2
_
(40)
6
Since n a

a is a Hermitian operator it has a set of complete and orthogonal eigen-

a|n =n|n (41)

a then |n is also eigenstate of H:

called "ladder operators"?

|n (44)
I will claim that this is also an eigenstate of n. To show this we act with n on this
state:
n|
+
=a

a
_
a

|n
_
=a

(aa

)
. .
=a

a+1
|n
=a

(n +1)|n =(n +1)a

|n
=(n +1)|
+

(45)
we thus see that, upto normalization we have
a

|n |n +1 (46)
similarly we can show that
a|n |n 1 (47)
This means that a

raises the eigenvalue n by one integer, while a lowers the eigen-

we can thus climb the "ladder" of eigenstates.

)
n
|0 (48)
3.1 Averages
Since the states |n are orthogonal to each other we have that
n|a

n n|n +1 =0
n|an n|n 1 =0
(49)
7
and it follows that
a

=
1
Z
Tr[a

] =0
a =
1
Z
Tr[a] =0
(50)
Due to this we have a quite simple expression for the average position
X =
1

2
_
a

+a
_
=0
K =
i

2
_
a

a
_
=0
(51)
and similarly for the mean square deviations
X
2
=
1
2
(a

+a)
2
=
1
2
_
a

+a

a +aa

+aa
_
K
2
=
1
2
(a

a)
2
=
1
2
_
a

+a

a +aa

aa
_
(52)
because of n|a

n =0 and n|aan =0 we have only

a +aa

(53)
and it remains to evaluate a

a and aa

. To do this I shall use two things.

2. Cyclic property of the trace Tr[ABC] =Tr[BCA] =Tr[CAB]

a = Z
1
Tr[a

a] = Z
1
Tr[a

ae
H
] = Z
1
Tr[a

e
H
e
H
. .
=I
ae
H
]
= Z
1
Tr[a

e
H
e
H
ae
H
. .
ae

] = Z
1
Tr[a

e
H
a]e

= Z
1
Tr[aa

e
H
]e

=aa

(54)
Now using the commutation relation [a, a

] =1:
1 =[a, a

] =aa

a =a

a(e

1) (55)
or
a

a =1/(e

1) and aa

=1/(e

1) (56)
8
We now have all the tools to calculate the mean variance from Eq. (53):
X
2
=K
2
=
1
2
_
1
e

1
e

1
_
=
1
2
cosh(/2)
sinh(/2)
=
1
2
coth(/2) (57)
Going back to the unscaled coordinates we get
x
2
=

m
X
2
=

2m
coth
_

2
_
p
2
=
2
m

K
2
=
m
2
coth
_

2
_ (58)
9