Theoretical Quantum Optics

16. Mrz 2011

(lecture notes after Dr. Hartmann, TUM, lectured WS2010/11 and the references: Walls & Milburn, Scully &
Zubairy, Yamamoto & Imamoglu, Mandel & Wolf )

Inhaltsverzeichnis
1

Quantised Electromagnetic Field

1.1

Field Quantisation

1.2

Fock- or Number States

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.3

Coherent states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.4

Squeezed States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

10

1.5

Variance in the Electric Field

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

13

1.6

Multimode Squeezed State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

14

1.7

Quadrature correlations and the Einstein-Podolsky-Rosen (EPR-)Paradox . . . . . . . . . . . . . . .

15

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Coherence properties of the EM eld

17

2.1

Field correlation functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

17

2.2

Properties of the correlation function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

17

2.3

Optical Coherence and Correlation Functions for the double slit . . . . . . . . . . . . . . . . . . . . .

18

2.4

First-order optical Coherence

19

2.5

Second order Optical Coherence: Photon correlation

. . . . . . . . . . . . . . . . . . . . . . . . . . .

19

2.6

Phase dependant correlations - the measurement of squeezing (Homodyne detection) . . . . . . . . .

20

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Representations of the EM eld

22

3.1

Expansion in number states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22

3.2

Expansion in coherent states: P-Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22

3.3

The Wigner-Function

24

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Open Quantum Systems

27

4.1

The (Quantum-) Master Equation

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

27

4.2

Application: Damped Harmonic Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

31

4.3

Representations of the Quantum-Master-Equation

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

34

4.3.1

Representations in number states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

34

4.3.2

P-Representations and Fokker-Planck equations . . . . . . . . . . . . . . . . . . . . . . . . . .

35

4.3.3

Steady state solutions of the Fokker-Planck-Equation

37

. . . . . . . . . . . . . . . . . . . . . .

Interactions Between Atoms and Radiation

38

5.1

Local gauge invariance and minimal coupling

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5.2

Long wavelength approximation & Electron Dipole Representation

. . . . . . . . . . . . . . . . . . .

38

5.3

Two-level-atom and Jaynes-Cumming-Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

40

5.4

Dressed States and Rabi Oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

40

5.5

Spontanous emission of a Two Level Atom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

41

5.6

Resonance Fluorescence

43

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

38

5.6.1

Interlude: Power Spectrum of the emitted light

. . . . . . . . . . . . . . . . . . . . . . . . . .

46

5.6.2

Relation between emitted eld and atomic coherence . . . . . . . . . . . . . . . . . . . . . . .

47

Raman Transitions and Electromagnetically Induced Transparency (EIT)

52

6.1

Raman-Transitions [Marzlin]

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

52

6.2

EIT

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

53

Quantised Electromagnetic Field

1.1 Field Quantisation

In order to do quantum optics, we rst have to

quantize

the EM-eld. Then we can discuss actual quantum states

as number states, the experimentally more common coherent states and exotic states as the squeezed states.

From

E, B to vectorpotential A:

Start with Maxwells equations in Vakuum

B =

(1)

E = t B
E =

(2)

(3)

B = 0 0 t B
Here,

c=

(4)

0 0 1 . Now we want to equivalently express the Maxwell Eqns by a vector potential A and its governing

equation, the wave equation. By (1), we put

B =
with the Coulomb-Gauge condition, to determine

(5)

uniquely:

A=0
and see what further conditions on

(6)

has to be satised in order to reproduce the Maxwells Equations. Eqn (1) is

automatically fullled

B = ( A) = 0
which holds for any vector eld, and using (2) we nd

E = t ( A) {E + t A} = 0
Thus, by virtue of

E = t A.

() = 0

Applying

for any scalar eld

we have that we can express

E + t A = .

Thus

and using (3) and (6) we nd:

4 = 0
and further with Greens Identity:

(7)




d3 r 4 () = ds
n

V
Assuming

=0

at innity for the rhs and plugging in (7) in the left we are left with

d3 r () = 0
V
which implies

= 0

or

= const.

everywhere, with

=0

at innity we have

E = t A
and (1)-(3) are satised. The dynamics of the

eld thus is determined by (4):

( A) +

1 2
A=0
c2 t

vanishing everywhere. Thus

(8)

so that

4A =
determine the Vectorpotential in Vacuum and the

Mode Expansion of

(+)

(r, t) + A

()

(r, t)

A:

1 2
c 2 t A,

E, B

0=A

elds, the physical quantities can be derived from (8), (5).

In order to solve for the real quantity

A,

make the following Ansatz: Expand

A =

in complex conjugate elds which are decomposed in discrete Fourier-Components for nite

volume problems with

A(+) (r, t) =

n
o
A() (r, t) = A(+) (r, t)

X
ak uk (r)eik t
k

where

k > 0

is supposed. Thus, the wave equation for

is transformed in an eigenvalue eqn for

which depend

on the boundary condition of the problem and also have to obtain the Gauge condition:

4uk (r) =

2
k
c2 uk (r),

Since as in single particle simple quantum mechanics, the

uk (r) = 0

4- operator is hermitian, its Eigenvalues are real and the

Eigenstates are an ONB:

d3 r uk (r)up (r) = k,p

(9)

Thus, as a conclusion, we have

A(r, t)

E(r, t)

r
X

}
ak uk (r)eik t + ak uk (r)eik t
20 k
k
r
X }k


i
ak uk (r)eik t ak uk (r)eik t
20
k

where the squareroot-term has been chosen for convenience, such that

Example: Chose periodic boundary conditions in box

with

V =L

a vector of integer values. From the gauge condition,

ak

is dimensionless.

. Here we have

k e = 0

uk (r) =

1
e eikr with
L2/3

c2 k2 = k2 .

We know that the energy of the EM-eld is given by

1
H=
2

d r

1 2
0 E +
B
0
2

This result should be reproduced with the Lagrange formalism in terms of

0 2
1
2
L(A i , k Aj ) = A

( A) =
2
20
The requirement of stationary

X
j,k,l,k0 ,l

(10)

with the Ansatz:

0 2
1
Aj
jkl (l Ak ) jk0 l0 (l0 Ak0 ) L =
2
20
0

t2
3

d rL S =

brings about the Euler-Lagrange Equations,

X
L
L
L
= t
+
m

Ai
(
Ai
m Ai )
m

i = 1, 2, 3

which yield

0 = 0 Ai

2
L n

such that we have a two dimensional space

to chose two distinct polarisation vectors from. The disperison follows from the Eigenvalue eqn

Quantisation Procedure:

k=

X 1X
X 1X
m {jim jk0 l0 (l0 Ak0 )} = 0 Ai
m {ik0 ml0 mk0 il0 } (l0 Ak0 )
0 m
0 m
0 0
0 0

j,k ,l

k ,l

c2 Ai =

{m m Ai m i Am } = 4Ai + i ( A)

m
4

dt L
t1

By the use of the Gauge condition the latter contribution vanishes, so we nd the wave-equation, thus the Lagrangian
density is the right one. From

H=

we obtain

X
d3 r j (r)A j (r) L

with j (r) =

1
H=
2

d r

1 2
1
2
+
( A)
0
0

L
= 0 A j (r)
A j

which is actually the same as (10).

For the quantisation, we need to introduce position and momentum operators,

r
qk (t)

pk (t)

= i

}
ak eik t + ak eik t
2
r k


}k
ak eik t ak eik t
2

It can be checked that the following relations hold

p, q

We now rewrite the Hamiltonian in terms of

1
2

d3 r

qk

pk

pk

k2 qk

with the use of (9), which leads after some algebra to


 X 2

1
pk
2
2
2
0 (t A) +
( A) =
+ k qk2
0
2
2
pk

and

qk

are conjugate, meaning

qk = H/pk

and

pk =

But this is the case as can be checked from (11) and (12). Thus the EM-Field Hamiltonian is a sum

over Harmonic Oscillator Hamiltonians with positions

qk ,

momenta

pk

and frequency

[
qk , pk0 ] i}k,k0 .

This

k .

ak (the
quantum number k also includes the polarisation )
h
i
implies a
k , a
k0 k,k0 and all other a
k -commutators zero. We

move from expansion coecients

that

(12)

Now this resembles a sum over harmonic oscillators if

H/qk .

(11)

For quantisation, we
to operators

a
k

such

nd it convenient to

summarize the result as follows:

Summary
Picture, the
operators

The

QM-eld-operator

is an operator which is mostly used in Heisenberg-Picture (otherwise, in the S-

a are constant and the states carry the time dependence) where it is composed from two time-dependent

a
k (t), a
k (t)

and given by

t) =
A(r,


} 
a
k (t)uk (r) + a
k (t)uk (r)
20 k

In H-Picture, it obeys the Heisenberg Equation of motion:

t) =
t A(r,
where

=
H

X
k

i h i
A, H
}



1
}k a
k a
k +
2

is the same in the S- or H-Picture. Thus we nd

t a
k (t) =

i
ih
= i (}k a)
a
k , H
}
}

a
k (t) = a
k (0) eik t

(13)

States can be labeled in second quantisation as HO occupation number states (which are not time dependent since

n
k (t) = a
k (t)
ak (t) = a
k (0)
ak (0),

see below), and more generally are described by density operators. The actual

elds are given by the expectation values , e.g.

D
E
 n
o
t) = T r t A(r,
t) .
E(r, t) = t A(r,

1.2 Fock- or Number States

From now on we will omit the operator heads. The quantised electromagnetic eld reads

t) =
A(r,

k
where

ak

and

ak


} 
ak (t)uk (r) + ak (t)uk (r)
20 k
k.

annhiliate and create photons in mode

First, consider a single mode

k.

The vacuum state is

dened by

an |0i = 0 n
The

Fock- or Numberstates

are Eigenstates of the number operator

ak ak |nk i = nk |nk i
where

nk

count the number of exications in a mode

which are called photons. The raising and lowering operators

create and destroy photons

ak |nk i =
ak |nk i =
The state

|nk i

nk |nk 1i
nk + 1 |nk + 1i

can be produced from vacuum as follows:

n
ak
|0i
|nk i =
nk !


Since the single mode states

|nk i are Eigenstates of the hermitian operator an an

hnk |mk i = nm

&

they are orthogonal and complete:

|nk i hnk | = 1

k
For various modes, consider a Fock state:

|ni = |nk1 i |nk2 i ... = |nk1 , nk2 , ...i

this means there are

nk1

photons present with quantum numbers

k1, nk2

Photons with quantum number

k2

and

so on... Orthogonality and completeness are expressed as follows:

hn|mi = n,m

&

|nk1 , nk2 , ...i hnk1 , nk2 , ...| = 1

k1,k2,...
The Fock-space is useful for calculations but it is dicult to produce light states which are pure Fock-States, e.g.
have denite Photon numbers in its various modes. It is possible to produce the single photon state in the lab, it
might be valuable in quantum information processing.

1.3 Coherent states

These states minimize the uncertainty relation with respect to amplitude and phase (see section Variance in the
electric eld), thus they are closest to classical light. They are no eigenstates of the photon number operator. In

the following only consider single mode states, thus

Displacement Operator

is xed and the sum over the modes is neglected.

Coherent states are generated by the abstract unitary displacement operator:



D() exp a a

with

an arbitrary complex number. Noting that

(14)

[a , [a , a]] = [a , 1] = 0,

we can use the abbreviated BCH-

Identity
1

eA+B = eA eB e 2 [A,B]

if

[A, [A, B]] = 0

(15)

and write the displacement operator in an alternative fashion:

D() = ea e a e||

/2

(16)

The displacement operator has the properties that it is unitary

D () = D1 () = D()
(from
with





D() = exp a a = exp a a

A = B = a a,

we have from (15) that

A B

e e

=e

D () a D()

D()

which equals

AB

=e

and

B A




exp a a = D1 ()

). Furthermore

= a+

D () a D()

since

(17)

= a +

which can be proved by using

eY X eY = X + [Y, X] +
with

Y = (a a)

and

1
1
[Y, [Y, X]] + [Y, [Y, [Y, X]]] ...
2
6

(18)

X = a:





D () a D() = a a a, a + [a a, a a, a ] + ... = a +
|
|
{z
}
{z
}

Coherent state denition and properties:

A coherent state is created by acting with the displacement ope-

rator on the vacuum (the HO ground state):

|i = D() |0i
Since

D()

is unitary, we can hope to nd a physical process which yields a time evolution operator

A coherent state is an Eigenstate of

D()!

a:
a |i = |i

Proof:

a |i = D()D () a D() |0i = D()(a + ) |0i = D() |0i = |i.

|
{z
} | {z }
1

Eigenvalue

Since

is not hermitian, its

|i

is a complex number.

Expansion in Fock-States:
2

|i = e||
Proof: Consider

/2

X n
|ni
n!
n0

hn|a|i and act with a to the right and to the left:

iterative prescription:

hn|i =

hn 1|i

series expansion of the exponential we have

from which we get

n + 1 hn + 1|i = hn|i, thus we nd an

h0|i
n!
a ||2 /2

hn|i =

h0|D()|0i = h0| ea e
7

(19)

h0|D()|0i. Using the

n!
D
E
||2 /2
a a

|0i = e

0|e

|0 =

e||

/2

D
E

2
0|ea 1|0 = e|| /2

where the last step follows from orthogonality. Finally, having

hn|i =

e|| /2 and the completeness of the number states results in (19).

n!

The coherent state is normalized

h|i = 1, Proof: With (19) we get h|i = e||

P
n0

n n
n!

= e||

P
n0

||2n
n!

e|| e|| = 1.

Coherent states contain an indenite number of photons, e.g photon number measurements on an ensemble
of equally prepared coherent states will yield various values. (As can be seen from the expansion in number

states.) The probability to measure a photon number of

in a coherent state

|i

is poisson distributed. From

the proof of the expansion property, conclude


P (n) = | hn|i |2 =

||

n
2

e||

n!

and as characteristics of the Poisson distribution we have

n = (n) = ||

P(n)
|| increases

Don't confuse the coherent state with a statistical mixture of number states with poisson weights, their density
operator would have only diagonal terms

mix =


n
2
X ||
n!

e|| |ni hn|

while the coherent state density operator has o diagonal terms,

coh =

Coherent states are not orthogonal:

2

e||

||2 +

XX e||2 n m

|ni hm|
n! m!
n m

|h|i| = e

||2

, Proof:

2



 ||2 /2||2 /2 P ()n 
|h|i| = e
n!  =

n0
2

, coherent states become approximately orthogonal for

It holds
2

h|ni = e||

coh = |i h|:

/2

| |

large.

X n

n!
n0

(20)

Overcompleteness. There are more states than necessary to expand every other state in:

d2 |i h| =
where an integration over the complex plane is denoted by

d2 .

(21)

Proof:

1 X |mi hn|
1 X |mi hn|
2
n m ||2
r 2 n+m


d |i h| =
d ( ) e
=
r dr e r
dei(nm)

n!
m!
n!
m!
n,m0
n,m0

Here,

= r ei
1

2n,m

has been used. The last integral provides

X |ni hn|

d |i h| = 2

n!

n0

dr er r2n+1 = 2

and thus

X |ni hn| 1

n0

n!

(n + 1) =

|ni hn| = 1

n0

Coherent states are the quantum states which are closest to classical states in the following sense: Consider a
QM harmonic oscillator
particle at position

H = }(a a + 21 ).

Consider a coherent state

|i,

what is the probability to nd the

x?


2

n

D
E2 
it
X

e

1


2
it/2 || /2
i(a a+ 2 )t


e
|  = e
hx|ni
p(x) = |hx|(t)i| =  x|e
n!


n0
Use

Exp[ 21 (x/x0 ) ]
p
Hn (x/x0 )

2n n! nx0

hx|ni =
with

Hn

x0 =

an Hermite polynomial and

p
}/m.

Put

= || ei

and nd:


2

n
X

i(t)


|| e
/2
2
2
1 


p(x) = e|| e(x/x0 )

H
(x/x
)
n
0 

x0 
n!

n0
where the last

|.|

term evaluates to

i
h

2
2i(t)
+ 2 || xx0 ei(t)
Exp ||
2 e
2

p(x) = e|| e(x/x0 )

. We nd

x
2
2
1
e2|| cos (t) e2 2|| x0 cos(t)
x0

Now, introduce the classically expected position

x(t) =

2||Cos(t )

and

x = x0 / 2

and obtain a

Gaussian probability distribution



1
(x x(t))2
p(x) =
Exp
2x2
2x
centered around the classical trajectory with a standard deviation

the momentum distribution by similar means, it can be shown that

satisfy the uncertainty relation as an

equality.

that is constant in time. By calculating

x p = }/2,

so that the coherent states

Here we understand the name coherent state as a state that

sticks togehter instead of changing its shape.

p(x)

Generation of coherent states (YI):

Remember, that a coherent state is created by acting with the displace-

ment operator on the vacuum:

|i = D() |0i
How could we experimentally describe such a translation from the HO ground state? (Pick up the centered wave
ground state wavefunction and shift it away to the point

)

We need a Hamiltonian where a classical external

force couples linearly to the generalized coordinates of the HO. Consider an oscillating dipole (q
operator!), in the S-Picture

VI (t) = q cos (t) E e



it

1
H = } a a +
2
The solution of the SE in the S-Picture
of

+e

it

a+a

cos (t)

not an

, which, after a RWA is of the form:



+ } f (t)a + f (t)a

i}t |(t)i = H |(t)i for |(0)i = |0i is - due to the explicit time dependance

- given by



|(t)i = exp A(t) + C(t)a |0i
where

t
0
C(t) = i dt0 ei(t t) f (t0 )

00

00

00

A(t) = dt f (t ) dt f (t )e
0

i (t0 t00 )

Assume, that the classical driving force is resonant with the harmonic oscillator, that is

f (t) = f0 eit
we nd

1
2
2
A(t) = (f0 t) = |(t)| /2
2

C(t) = i f0 t eit (t)

Thus we have for the generated state

"

#
2

|(t)|

a |(t)|2 /2
|(t)i = exp
+ (t)a |0i = e(t)a |e(t)
|0i = D((t)) |0i = |(t)i
{z
}e
2
acts as

Im()
coh. state

Re()
t increases

1.4 Squeezed States

Minimum uncertainty states:
hermitian operator

Squeezed states are a general class of minimum-uncertainty states. Consider a


2
2
A with variance (A) = A2 hAi , For any two hermitian operators, the uncertainty relation

holds:

1
|h[A, B]i|
2
p
p
q = }/2(a + a ), p = i }/2(a a ).

A B
For example, consider a HO with unit mass, we have
coherent state and have, as before

(q)coh = }/2

and

(p)coh = }/2.

Thus

qp =

}
2

1
2

(22)
We assume a

|h[q, p]i|

and the

coherent state fullll (22) with an equal sign.

In analogy to the relation between

a, a

&

p, q

and in order to gain general insight in the HO problem, dene the

10

hermitian

eld quadrature operators (which, in general both occur in the denitions of EM-Observables E,B)

X1

a + a

X2

i(a a )

[X1 , X2 ]
with the back transformation

a = (X1 + iX2 ) /2

= 2i

leading to a simple statement of the uncertainty relation

X1 X2 1
Consider the following denitions:
Minimum uncertainty state (MUS)
Coherent state: MUS with

Ideal

For a coherent state

X1 = 1 = X2

Squeezed state: MUS with

(Squeezed state: state with

X1

or

X1 X2 = 1

X1

or

X2 < 1,

X2 < 1

see shaded region)

|i, we nd the complex number 2 as a center of the error cycle

in the complex

X1 X2 plane:


|a + a | = + = 2Re()


i |a a | = i( ) = i 2i Im() = 2Im()

h|X1 |i =
h|X2 |i =
and

2
2
2
(X1 ) = |X12 | h|X1 |i = |a2 + aa + a a + a2 | |a + a |
acting to the right with

and to the left with

a ,

we nd:

2
(X1 ) = 2 + 1 + 2||2 + 2 ( + )2 = 2 + 1 + 2||2 + 2 (2 + 2||2 + 2 ) = 1
The free time evolution of a coherent state under

H = }a a is given from above (e.g. in the H-Picture a(t) = a eit )

as a rotation of the error circle or ellipse around the origin.

X2 coh. state
Radius: 1

min. uncertainty states

Exp[r]
squeezed state

coherent state

squeezed states
1

X1

Description of squeezed states:

physical states

forbidden

X2
Exp[-r]

(23)

X1

In analogy to the coherent states, dene the

S() exp

2
2 a

2 a2

where

= r e2i
The squeeze operator obeys the relation

S () = S 1 () = S()
11

squeeze operator,

and is thus unitary. With (18) we can prove

where

S () a S()

a cosh(r) a e2i sinh(r)

S () a S()

a cosh(r) a e2i sinh(r)

S () (Y1 + iY2 ) S()

er Y1 + iY2 er

Y1 + iY2 (X1 + iX2 ) ei

is a rotated complex amplitude. Thus the squeeze operator attenuates one

component of the (rotated) complex amplitude while ampliying the other one. The degree of scaling is given by

r = ||

squeezed state, rst squeezing the vacuum

(squeeze factor). This will be elucidated by the denition of a

(transform the error cycle at zero into an ellipse with the principal axis
to

Y1

and

Y2 )

and then displacing it (move it

2):
|, i = D()S() |0i

A squeezed state has the following expectation values and variances:


hX1 + iX2 i = h2ai = 2 0|S ()D () a D()S()|0 = 2 0|S () (a + ) S()|0 = 2
where (17) was used. Thus the expectation value of

|, i

state in the

X1 iX2

plane is the same as for

|i

state.

Similary, we have

Y1 = er

Y2 = er

thus the state is squeezed in one direction and expanded in the other. The photon number mean and variance is

hN i = || + sinh2 (r)

2
2
(N ) =  cosh(r) e2i sinh(r) + 2cosh2 (r) sinh2 (r)
Example:
shown that

Consider a HO

hq(t)i =

H = }(a a + 21 ),

2}
|| cos(t) and

means, while the mean oscillates with frequency

}
2

2 (a + a ) it can be


cosh2 r + sinh2 r 2cosh(r)sinh(r)cos(2t 2) . That

initially in a squeezed state. For

(q(t)) =

(24)

the variance oscillates with

2 ,

q =

the probability distribution thus

is breathing. This can be seen from following the ellipse in the picture around the circle of free evolution and
monitoring its widht in, say
while the

X2

X1

direction. In the picture below, the

X1

evolution of the squeezed state is shown,

(or momentum) evolution is hidden.

V
breathing

squeezed, displaced vacuum

ground state
q

Generation
h of squeezed
i states (Y&I)

S() exp

2
2 a

2
2a

For squeezing, as we have seen we need to create a unitary evolution

which acts on a vacuum state. This can be obtained by an Interaction Hamiltonian of

the form

HI =


}  2
a + a2
2

which describes simultanous two-photon generation or absorption processes, as can be realized by second order
nonlinear processes where a photon of energy

2}

generates two photons, each of enery

12

~ .

1.5 Variance in the Electric Field

Now apply the general ideas about squeezed and coherent states worked out above and add some meaning to the

X1,2 .

quantities

The electric eld for a single mode in the box can be written as (replace

r
E(r, t) = i
e

a = (X1 + iX2 ) /2)

r

} 
}
i(tkr)

i(tkr)
= e
a(0)e
a (0)e
(X1 sin [t k r] X2 cos [t k r])
3
20 L
20 L3
(25)

For the expectation value of the electric eld, we have

r
hE(r, t)i = e

}
(hX1 i sin [t k r] hX2 i cos [t k r])
20 L3

For the uncertainty in the electric eld, we obtain


2
V (E(r, t)) = E(r, t)2 hE(r, t)i =
where

1
2

V (X1 , X2 ) =

(hX1 X2 i + hX2 X1 i 2 hX1 i hX2 i).

For a minimum uncertainty state

Xj hXj i

for

} 
V (X1 ) sin2 [t k r] + V (X2 ) cos2 [t k r] V (X1 , X2 ) sin [2t 2k r]
3
20 L

j = 1, 2,

X1 X2 = 1,

we have

V (X1 , X2 ) = 0,

j
the proof is as follows: DeneX

then

D E

2 = X 2 hXj i2 = (Xj )2
X
j
j
With Cauchy-Schwartz, we nd

1
1

D
D E D E X herm. D
ED
E
E2 D
E2
j
2 X
1X
2
1 0|X
10 X
2 0|X
2 0 = ||X
1 0||2 ||X
2 0||2  X
1 0|X
2 0  =  X
2 
X
=
X
1
2
2
 Dn
oE 1 Dh
iE2  Dn
oE

1
1, X
2
1, X
2  =  1
1, X
2

X
X
X
+
+
i
= 



2
2
2

Here, from above,

i
1, X
2 = 2i
X

Dn
oE
n
o
1, X
2
1, X
2
X
X
is real, since
Dn
oE
1, X
2
0 = 21
X
= V (X1 , X2 ) = 0.

has been used. We know that

We thus have necessarily in the above equation

is hermitian.

We can now analyze a coherent state and a squeezed state's electric eld distribution:

In a coherent state
(23),

V (X1,2 ) = 1,

|i,

we have

V (X1 , X2 ) = 0

since it is a minimum uncertainty state and we have from

the uncertainty (both in phase and amplitude) is constant (Plot (a))

}
(2Re() sin [t k r] 2Im() cos [t k r])
20 L3

}  2
}
sin [t k r] + cos2 [t k r] =
20 L3
20 L3

hE(r, t)i = e
V (E(r, t))

(shifted sine wave)

In an ideal squeezed state we have to return to the derivation of electric eld mean and variance and write

Xj = Yj ei .

There are two interesting special cases.

1. Plot (b):

=0

and

r>0

corresponds to reduced

X1 = Y1 = er .

values (25) with gaussian distributed random numbers

X1,2

Plotting a variety of electric eld

from inside the error

ellipse, we see that the

phase of the wave (shift in time direction) bears a higher uncertainty than in amplitude.
2. Plot (c):

=0

but

r < 0,

now

Y2 = er

is reduced which corresponds to a larger error in magnitude

than in phase.

13

1.6 Multimode Squeezed State

In this section the concept of a single-mode squeezed state considerd above should be generalized to the case, when
the two photons created simultanously are of dierent energy. Consider a nondegenerate parametric amplier, e.g
the generation of two photons of frequencies

= 1 + 2

from a oscillator with frequency

The interaction

Hamiltonian reads



HI = i} a1 a2 a1 a2

(26)

The corresponding multi-mode-squeezed state is created by rst squeezing the vacuum and then displacing by

1,2 :

|1 , 2 i = D1 (1 ) D2 (2 ) S(G) |0i
where we used the generalized displacement operator

h
i
Di () = exp ai ai

and the

two mode squeezing

operator:

S(G) eG

a1 a2 Ga1 a2

with

G = rei

with the properties

S (G) = S 1 (G) = S(G)

S (G)a1,2 S(G) = a1,2 cosh(r) a1,2 sinh(r) ei
For

(27)

= 0, that is G purely real, we nd for the expansion of the two mode squeezed vacuum state into Fock-States:
|i = S(G) |0i =

X
1
tanhn (r) |n, ni
cosh(r)

(28)

n0

a1,2 |0i = 0 S(G)a1,2 S (G)S(G) |0i = 0. From (27) we replace the rst three terms and we use the Ansatz
|
{z
}
1
P
|i =
cn1 ,n2 |n1 , n2 i. The resulting system of equations for a1 and a2 yield (28).

Proof:

n1 ,n2
Remark: For a state

|i = S(G) |0i,

if we measure the photon number for mode 1 (with result

14

N ),

we project the

state to a number state

span |N, mi,

we know with certainty the photon number of mode 2 (since the expansion
m
(28) is diagonal), although the variance of the photon number is not zero for |i. A state as in (28) is entangled,

there is no way of writing

|i = |1 i |2 i,

i.e. we cannot write the state as a tensor product of the subsystem

states. This occurs if the expansion coecients for the base state do not factorize:

HA HB |iAB =
If

cij = ci cj

cij 6= ci cj

the state is separable, if

cij |iiA |jiB

we have an entangled state.

1.7 Quadrature correlations and the Einstein-Podolsky-Rosen (EPR-)Paradox

In QM, there are states where a measurement on a System A (Photon going to the right) allows to infer the value
of an observable for a system B (photon going of to the left, far away) with certainty although you have multiple
outcomes possible for both these observables.

Example for the two mode squeezed state (M&W,22.4):

photon splits up into two, the
with

1 + 2 = 0

signal

and the

idler

In parametric down conversion (one high energy

photon), where a two mode squeezed state occurs, the Hamiltonian

and a strong, classical pump-eld is:



H = }1 a1 a1 + }2 a2 a2 + }g0 a1 a2 ei0 t a1 a2 ei0 t = H0 + H1 (t)
Here,

denotes the nonlinear-interaction and

and thus

n
1 (t) n
2 (t) = n
1 (0) n
2 (0)

is the pump amplitude. We can conrm that

i
=0
n
1 n
2, H

so the signal and idler photon are always generated together. The H-Picture

equations of motion are:

t a1 (t) =

1
[a1 (t), H] = i1 a1 (t) iga2 (t)0 ei0 t
i}

It is convenient to introduce new operators which will be shown to be slowly rotating:

A1,2 (t) a1 (t)ei1,2 t

which obey

t A1 (t) = ig0 A2 (t)

t A2 (t) = ig0 A1 (t)

Consider the generalized eld quadrature for a two mode squeezed state

Xj Aj (t)ei + Aj (t)ei
which is hermitian and such that

h
i
aj , iaj = 2i

+
2

Xj , Xj

j = 1, 2

i
h
i
h
Aj (t)ei + Aj (t)ei , iAj (t)ei iAj (t)ei = aj , iaj +

(thus the two observables are analogous to the proposed position and momentum variables in the

original EPR paper). In the two mode squeezed state, we can quantify the correlations between the two modes via

1
V (, )
2
that means if

V (, ) = 0,
0



X1

then we have with introducing

X1

X2

and

2 

X2

eigenbases of the bipartite Hilbertspace:




2
1X X

= V (, ) =
h|1 , 2 i 1 , 2 | X1 X2 |1 , 2 h1 , 2 |i
2
,
1 ,2

1X
2
p1 ,2 (1 2 )
2
1 ,2

Both terms in the sum are

0, so if there is a nite probablity to measure 1

15

and

2 , the two values are necessarily

V (, ) = 0, the two observables X1 , X2

would be perfectly correlated and one would always

be able to infer each single result of a X1 measurement from a preceding measurement of X2 . In the following, we

are interested in + = 0. We now look at the equations of motion for X1 X2

for real and positive = ig0 :

equal. That means if

t X1 t X2
We obtain

+A2 (t)ei + A2 (t)ei A1 (t)ei A1 (t)ei = X1 X2

t X1 X2 = X1 X2
t V (, )(t) =

which leads to


X1 X2 t X1 X2 = 2V (, )(t)

which eventually yields:

V (, )(t) = e2t V (, )(0)

and the observables

EPR-Paradox

X1

and

X2

If we measure

become perfectly correlated.

X1

at a large time, we know the value of

X2

(which is a result of a measurement

a little later) with certainty. That means, that mode 2 must have been in an

X2

eigenstate even before the

measurement of mode 2. At the time we perform the measurement, the two photons are already far apart and
cannot interact, that means mode 2 was in the eigenstate of

X2

even before the measurement on mode 1.

X1 with 6=

was in an eigenstate of X2 before the rst measurement. But in general

If I had decided to measured another variable on mode 1,

can be repeated. Mode 2

h the argument
i

X2 , X2 6= 0 and thus they cannot have

a common eigenstate. We have a contradiction. The conclusion is that in QM there can be nonlocal correlations
between so called entangled states. However, this is not an incidence when information travels faster than light since
due to the probabilistic nature of the rst measurement, its outcome is random and cannot be used to transmit
actual information into the other laboratory (like sending from Munich to LA the message the weather is sunny).

Experimental signicance:

If we can produce many equally prepared initial states as described above and focus

on mode 2 to measure either the position or momentum equivalent observables

X2

or

+/2

X2

we should nd

from the histograms according to their commutator (see above) and the uncertainty principle for the variances,

V (A)V (B)

1
4

|hA, Bi|

 +/2 
V X2 V X2
1

Alternatively, we can measure



+/2
V X2

on mode 2 as before, but in a second series of measurement on mode

1, we measure X1 whose outcomes are, if perfectly correlated in every single measurement equal to
Thus we are able to infer

Vinf X2

from the histogram of the

commute, in principle, the experimental result for

relation. An experiment by Mandel in 1988 gave

X1

X1

measurements. Since

+/2
Vinf X2 is not restricted
V X2




+/2
V X2
Vinf X2 < 1. This does

X2 .
and

+/2

X2

by the uncertainty
not contradict the

uncertainty principle since this was derived for exactly the case discussed before, a number of measurements on the
same state, which we haven't done in the second case. There we used the additional information
a dierent measurement scheme, so that the uncertainty principle looses its validity.

16

V (, ) = 0

and

Coherence properties of the EM eld

This section answers the question: What properties of the EM eld can/do we measure and what does that tell us
about the eld?

2.1 Field correlation functions

Assume to have an ideal photon detector for the eld component
absorption of a eld quanta via the anhiliation operator

aE

E+ (r, t),

that means the detector relies on the

. Thus the transition rate for absorbing a photon

in the detector is, according to Fermi's Golden Rule with the elds initial and nal states

|ii , |f i:

2
Tif =  f |E+ (r, t)|i 
Since we are interested in the total count rate and not in the specic nal state of the eld, we sum over

|f i

to get

the measured intensity:

I(r, t) =


 f |E+ (r, t)|i 2 =
i|E (r, t)|f f |E+ (r, t)|i = i|E (r, t)E+ (r, t)|i
f

If we have a statistical mixture of initial states, we have

I(r, t) = T r (E (r, t)E+ (r, t))

with the density operator

as known from statistical mechanics.

As an example, consider an initial state, that is a single mode Fock-State

and

E (r, t) = E+ (r, t)

we nd

I(r, t) = hnk |E (r, t)E+ (r, t)|nk i =

q


ik t
k
|nk i. Using E+ (r, t) = i }
20 ak (0)uk (r)e

}k
20

|uk (r)| nk .

Motivated by the specic application of the photon detector, we dene the

n-th order eld correlation function

(equivalent to the count rate of a n-photon detector):

G(n) x1 , ..., xn ; x0n+1 , ..., x02n = T r E (r1 , t1 )...E (rn , tn )E+ (rn+1 , tn+1 )...E+ (r2n , t2n )
and recognize that

I(r, t) = G(1) (x, x).

2.2 Properties of the correlation function

If we consider

A=E

(+)

T r A A 0

(x1 )...E

(+)

(xn )

which holds for any operator (since

A A

is always hermitian and positive), and chose

we nd that all correlation functions are non-negative:

G(n) (x1 , ..., xn , xn , ...x1 ) 0

Assuming

A=

n
P

j E (+) (xj )

j=1

we nd

n
P

j l G(1) (xj , xl ) 0

for any

l,j .

Thus, since the quadratic form

j,l=1

(1 , 2 ) G
is positive semidenite, the matrixelements

G(1) (xj , xl )

!
0

form a positiv semi-denite matrix with the property





det G(1) (xj , xl ) 0
For

n=2

we thus nd a generalized Schwarz-Inequality:


2


G(1) (x1 , x1 ) G(1) (x2 , x2 ) G(1) (x1 , x2 )

17

(29)

2.3 Optical Coherence and Correlation Functions for the double slit
We will show how the double slit experiment turns out to measure rst order correlation functions (or simply the
intensities).

slits
r

s1
r1
r2

s2

incident plane wave

Screen

The +-eldcomponent incident at the screen is

+
E+ (r, t) = E+
1 (r, t) + E2 (r, t)
We have for the spherical mode function in a sphere with Radius L

uk (r) =

1 eikr
ek
4L r

and thus, if the eldvalue at the pinhole is determined by the incident wave (in the denominator, replace

s1,2 ' R:

i } eit
a1 eiks1 + a2 eiks2
E (r, t) =
80 L R
+

where

si = ri r.

The intensity on the screen is as before


D
E D
E D
E

G(1) (x1 , x1 ) + G(1) (x2 , x2 ) + 2Re G(1) (x1 , x2 )


a1 a1 + a2 a2 +2  a1 a2  cos
I(r, t) = T r E (r, t)E+ (r, t)
2
R
with

= k(s1 s2 ).

cos = 1 k (s1 s2 ) = 2n.

Thus we see interference fringes with maximas where

Examples:
1. Consider classical two mode coherent eld,

|ii = |i |i

and the

|i

Eigenstates of

ai

with Eigenvalue

we

nd

D
E D
E
E
D


2
I a1 a1 + a2 a2 + 2  a1 a2  cos = 2(1 + cos) ||
and conclude, that we see fringes.
2. Single photon incidend on slit-plane:

|ii =

1
2


a1 + a2 |0i I

1
2

1
2

+ cos,

one sees fringes even from

single photon!
3. Incident number state

|ii = |n1 , n2 i = a1 a2 |0i.

There are no fringes since

As a conclusion, we nd that the interference is due to the

G(1) (x1 , x2 )

from classical light. We need another experiment to tell them apart.

18

E
a1 a2 = 0.

- function, it cannot distinguish quantum-

2.4 First-order optical Coherence

In the double slit, the highest order of optical coherence was associated with the light state which would produce
the largest visibility of interference fringes (this was actually the

denition of coherent light ):






G(1) (x1 , x1 ) + G(1) (x2 , x2 ) + 2 G(1) (x1 , x2 ) G(1) (x1 , x1 ) + G(1) (x2 , x2 ) 2 G(1) (x1 , x2 )
Imax Imin
=
=
Imax + Imin
2G(1) (x1 , x1 ) + 2G(1) (x2 , x2 )
 (1)


G (x1 , x2 )
2 I1 I2
p
G(1) (x1 , x1 )G(1) (x2 , x2 ) I1 + I2

The rst term, which contains the interference information is used to dene a new quantity

It is bounded by

(1)

(x1 , x2 ) = p

due to (29). The larger

g (1)

 (1)

G (x1 , x2 )
G(1) (x1 , x1 )G(1) (x2 , x2 )

the more coherent the eld.

2.5 Second order Optical Coherence: Photon correlation

Consider an experiment outside the domain of one photon optics, namely the intensity time correlation measurement
by

Hanbury-Brown

at time

and

Twiss

(1959). These measurements determine the joint probability that one photon arrives

and another one at time

t + .

With the measurement framework presented above we nd this probablity

to be


G(2) ( ) = E (r, t)E (r, t + )E+ (r, t + )E+ (r, t) = h: I(t)I (t + ) :i h: n(t)n(t + ) :i
where

: A : means normal ordering, e.g. as in the original expressions all a-anhillators to the right and a

creators to

the left. We have used the assumption to be in a stationary state, so the overall total time for the rst measurement
is of no signicance and only the time dierence is of importance. The normalized second order correlation function
is dened as

(2)


a (0)a ( )a( )a(0)
G(2) ( )
( ) 
2 =
2
ha (0)a(0)i
G(1) (0)

A reasonable assumption for a longtime behaviour is that the photons should not care about the other one in any
case, thus


2
G(2) ( ) G(1) (0)

g (2) (0) =

and


a a aa
2
ha ai

g (2) ( ) = 1



a aa a a a
2
ha ai

(not true for thermal eld). For


2
2
n2 hni + hni hni
2

hni

=1+

The dierent states of the EM eld reveal interesting behaviour categorized as bunching

g (2) (0) < 0.

Antibunching is a signature of non-classical light states.

thermal state

g(2)()

bunching (mixture of coherent fields)

1 coherent field

independent photons

anti-bunching (number states)

19

0,

we have

V (n) hni
2

hni

g (2) (0) > 1 and antibunching

Examples:
1. A

thermal state

from a chaotic light source yields




X
= 1 e} exp }a a/kT = 1 e}
exp [}n/kT ] |ni hn|
n
and thus we have no time dependance in the eld. Thus

g (2) (0)

= g (2) ( ) = 1 e}

1

Tr


P

exp [}n/kT ] |ni hn| n2 n

n

2
P
Tr
exp [}n/kT ] |ni hn| n

g (2) ( )
2. Coherent eld,

1 e}

1

1 e}

1

a |i = |i.

exp [}n] (n n)

2
P
exp [}n] n
n
)



4 (
1 + e} e}
1 + e}
e}

3
2
e2}
(1 + e} )
(1 + e} )

We nd neither bunching nor antibunching but have independence of photons

(2)

( ) =

3. Consider a single mode, e.g. red light. A

p()

in a coherent state for all

That means

n
2

P
=


a (0)a ( )a( )a(0)
2

ha (0)a(0)i

||

||

time-uctuating classical eld

=1

is a classical mixture of coherent states.

p() > 0,

(see below in the representation chapter) has to be a probability density, i.e.

and

this implies we have bunching for any classical mixture of coherent states.
4. Number states:

= |ni hn|. g (2) (0) = 1 +

V (n)n
n2

= 1 1/n < 1

since the variance in photon number is zero

for a number state. Thus a number state is a quantum eld.

5. Squeezed states:

|, i

with

= r e2i , r > 0.

Let

be real (error ellipse displaced along X1 axis, ) and use

(24),

V (n) n
|| {cosh(2r) sinh(2r)cos [2] 1} + sinh2 (r)cosh(2r)
=

2
n2
2
|| + sinh2 (r)

for

2 (phase squeezed state), we have

V (n) n = || (e2r 1) + sinh2 (r)(2cosh2 (r) 1),

exhibits bunching. Here, we have the squeezed direction

for

=0

(amplitude squeezed state): here

sinh2 (r)(2cosh2 (r) 1).

For

X1

this state

and the prolonged directionY2

is the squeezed direction.

||  2 sinh2 (r) cosh2 (r)

Finally, the squeezed vacuum,

Y1 = X2

= X1 .

V (n) n = || (e2r 1) +

this state exhibits antibunching.

= 0 has V (n) n = sinh2 (r)(2cosh2 (r) 1) > 0, thus we have bunching.

Conclusion: In any case where we have anti-bunching, the state cannot be a classical mixture of coherent states.

2.6 Phase dependant correlations - the measurement of squeezing (Homodyne detection)

So far, we have been measureing photon numbers or intensities, now, lets turn to the measurement of
two single mode elds (one signal, one reference also termed
transmittivity

20

Local Oscillator)

X .

Consider

combined on a beam splitter with

Beamsplitter
Detector

E1(r,t)

E2(r,t)=Reference field

The single electric eld operators of the incident elds are given by

E1 (r, t)
E2 (r, t)

}
= i
2 V
r 0
}
= i
20 V

i
a ei(krt) h.c.

i
b ei(krt) h.c.

Both elds are taken to have the same polarisation and are phase locked. The mixed eld at the detector is given
by

r
ET (r, t) = i
where

c=
The local oscillator eld
factor

E2 ,

i
} h i(krt)
ce
h.c.
20 V

p
a + i 1 b

which is reected on the backside of the beam splitter gains an additional phase

from relfection. The photon count rate at the detector is




c c = a a + (1 ) b b i (1 ) ab a b
Assume the reference eld

E2

to be in a coherent state with large amplitude

= ||ei : b |i = |i.

Then the

second term dominates the rst term. For the expectation value in the third term we notice that the Hilbertspace
seperates


ab = hai b

and we write

p
p


2
2
c c
= (1 ) || + (1 ) i||ei hai + i a ||ei = (1 ) || + || (1 ) X+/2
with

X = a ei + a ei
We can make

= 1 1

to get rid of the rst term

(1 ) || = O()

while the second term is

O( )

or

subtract the rst term from a measurement with the reference signal alone. Then we can actually measure the
mean quadrature


X+/2

and by changing the phase of the reference eld, all

2
by measuring nc we can measure

hX i

can be measured. Similary,

and actually see if the signal is a squeezed eld!

21

Representations of the EM eld

This chapter presents various ways to represent the quantum mechanical state of a light eld by specifying its
density operator in terms of various Hilbert-Space bases. We consider a single mode, since each mode is described
by independent Hilbertspaces, this treatment is general: We can formulate the statistical description of the entire
multi-mode eld as the product of the distribution functions for each mode.

3.1 Expansion in number states

The number states, i.e. the eigenstates of the hermitian operator

can be expanded as

form a complete set, thus any density operator

cn,m |ni hm|

n,m=0
Note, that in general, the expansion contains nonzero-o-diagonal terms, only for states where the photon number
distribution is of interest the number state expansion proves useful: As an example, consider a

thermal state. For

a single mode, we have

=
with

Z = T r(eH ) =

e}

n

1 H
1
e
= e}(a a)
Z
Z

= 1/(1 e} ).

(30)

To determine the average photon number, consider with

n0

= } :



X
X
1
1
1
1
1
ne}n =
n = T r(a a) =
en =

= }

Z
Z
Z
1e
e
1
n0

n0

We can expand (30) in number states by the procedure

P
m,n

|mi hm| Z1 e}(a

a)

|ni hn|

as follows

X
X
1 X }n
e
|ni hn| = (1 e} )
e}n |ni hn| =
pn |ni hn|
Z
n0

n0

n0

with

pn = (1 e

}n

)e

1
=
1+n

n
1+n

n

3.2 Expansion in coherent states: P-Representation

The coherent states

|i

form an overcomplete set, i.e. (21). From (19) we know, that for any hermitian operator

we have

( ) m
hn|O|mi n!m!
n!m!
n,m

e|| h|O|i =
This is a taylor-series expansion for

expansion coecients

hn|O|mi n!m!

e|| h|O|i

number state matrix elements

in powers of

, ,

(instead of

Re ()

P F unction

Im ())

with the

diagonal

h 2
i
1
(n m ) e|| h|O|i |=0
n!m!

coherent-state matrix elements

h|O|i for all , all (even-non-diagonal)

hn|O|mi can be found. That is remarkable: It implies that in an |i basis, the density

operator can always be chosen diagonal which is clearly not the case for number states!
The

and

unambigously determined. Thus we nd

hn|O|mi =
That implies, if that we know the

ist dened as

d2 P () |i h|

22


hOi = T r (O) = d2 P () h|O|i holds for any operator O. The last equality can be worked out as
P
T r(O |i h|) =
hn| O |i h|ni = h|O|i. Whenever the photon distribution nn is narrower than the Poisson
such that

|ni, P ()

distribution, as in the case for a number state

becomes badly behaved (e.g. a derivative of a tempered

distribution instead an ordinary function), but this is the price we pay for forcing quantum physics into a classical
format, e.g. forcing

How to nd

|i

to be diagonal in the P representation.

P () from : [S&Z]

Using the denition of the

we write

P F unction

|i,

and the two coherent states

d2 P () h|i h|i

h||i =

Using a calculation similar to the one when conrming the non-orthogonality of coherent states, we nd
2

h||i = e||
Writing

= x + iy

and

= x + iy

d2 P ()e|| e

h
i
2
2
dx dy P (x , y )e(x +y ) e2i(y x x y )

Thus the rhs is just the two dimensional Fourier-Transform of

2

e(x +y )
2

we have

h||i e|| =

P () =

P ()e||

2
2
e||
dx dy h||i e(x +y ) e2i(y x x y ) = 2

and the inverse yields:

d2 h||i e|| e

(31)

Properties of the P-Function

P ()

1=

is normalized in the sense that

d2 P ()

P () 0,

valued or a probablity distribution with

since

|i

P () 0

But it is not always real

states are non-orthogonal.

If we have a classcial eld, that means a mixture of coherent elds

implies bunching, as will be seen below. Whenever

as seen from tracing over

|i,

the

P () 0

holds indeed. This

does not hold, we deal with quantum elds.

The P-Function is convenient for the calculation of correlation functions and any operators which constitute
of normal orderd products of creators and anhilliators,

n m 
a a
= Tr


n
d2 P () |i h| an am = d2 P () ( ) m

(32)

thus, taking the QM expectation value is similar to a classical averaging in statistical mechanics.

The rst two points together give insight in the relation of quantum light and the second order correlation
function, or, more precisely, the bunching or antibunching properties of light. With the denition

d2 P ()||2 ||


a a =

R+
0

g (2) (0) =


a a aa
ha ai

d2 P ()||4
= 
2 = 1 +
d2 P ()||2

h
D
Ei2
2
d2 P () ||2 ||

2
d2 P ()||2

Because we have seen, that for an arbitrary state, antibunching can occur, this is leading to a clear contradiction
of

P ()

beeing a probability distribution.

Squeezing: The variances of

V (X1 ) =

X1 = a + a

a + a

2 E

and

X2 = i a a

can be written as


2
2



2
hai + a
= a + 1 + 2a a + a2 hai + a

23

Now, with the use of (32):

V (X1 )

d2 P () [( + ) (hi + h i)]

1+

V (X2 )

( + ) hi + h i
d P ()

i
i


1+

2

Re [P ()] < 0 for some . Thus squeezing is also a phenomenon

That shows that squeezing is only possible, if

of quantum light.

Examples of states in P-Representation

Thermal state :

As in section 3.1 we have

= (1 e} )

X
X
e}n |ni hn| =
n

We want to make use of the

h||i =

hni

F T 1 ,(31),

n+1

(1 + hni)

n+1

|ni hn|

and note with (20):

2

(1 + hni)

hni

e|| X ||2
h|ni hn|i =
1 + hni n
n!

n



2
n
e||
hni
||2
exp
n =
1 + hni
1 + 1/ hni
(1 + hni)

so that the Fourier Transform can be completed

2

P () =

e||
2
(1 + hni)

d2 exp

||2
1 ||2 /n
e + =
e
1 + 1/ hni
n

Thus, in the P-Representation, a thermal state is given by a Gaussian distribution, which is always >0 and
therefore the thermal state is a classical eld.

Not surprisingly, the

(2)

coherent state |0 i

P-Representation is a delta-peak in the complex plane:

P () =

( 0 )

3.3 The Wigner-Function

Denitions: The

characteristic function

of a density operator

is given by




() = T r ea a
where

C.

Wigner-Function

Then the

W () =
Intuitive Form:

d2 () e

= + iv

and using instead of

1
2

h
i

d2 T r e(a ) (a)

a, a

the more intuitive position and momentum operators

of the HO. We nd

and thus we can replace the exponent by

A more convenient interpretation can be given by splitting the complex integration variable in a

real and imaginary part,

p, q

1
2

is dened by the FT of the characteristic function:

2
} and

v =

2
} v we have with

q + ip
q + ip
a=
=
2}
2}
q
q


2
(p p) + i 2
a (a ) = i }
} v (q q).

d2 = dd = }/2 d
d
v:

W (q, p) =

2}
(2)

h
i
d
d
v T r eiv(qq)i(pp)

24

Introducing

W (q, p) as a distribution in two real numbers, which take the role of the position and momentum varia


2}
bles. It is in general not a probability distribution, since it is not always positive. W (q, p) =
d
d
v T r eiv(qq)i(pp)
(2)2
Here we have

comes close to phase space quantum distribution (real, normalized, although not positive!!) intermediate between
position and momentum representation, satises

W (q, p)dp = hq||qi

p

marginals give prob. distribution to measure momentum

hOi =

W (q, p)dq = hp||pi,

that means

For expectation values of operator

O(p, q):

dp dq O(p, q)W (q, p).

W as Gaussian convolution of P-Function:

The Wigner-Function is a Gaussian convolution of the P-Function:

2
W () =

Proof: Use the BCH-Identity

the

in state

and

e ( a

) (a)

d2 P () e2||

= e(a

) (a) ||2 /2

. Then we write with the denition of

P F unction

W ()

=
=




h
i
1
1
(a ) (a) ||2 /2
(a ) (a) ||2 /2
2
2
2
d P () |i h| e
d T r e
d Tr
e
= 2
e
e
e
2

i
h
2

1
1
2
||2 /2
2
( ) ()
d
= 2
d

P
()
T
r
e
e
d2 e|| /2 d2 P () e( ) ()
2

Now, we have to perform the

for real

w, v .

-Integration. To this end, dene = x + iy

and obtain

d2 = dxdy . Set = w + iv

Then the exponential simplies

( ) ( ) || /2 =
If we nally use

dx ex

/22ivx

1 2
x + y 2 2ixv + 2iyw
2

2e2v

(33)

we conclude

1
W () = 2

d P ()

x2 /22ivx

dx e

y 2 /2+2iwy

dy e

2
=

2(v 2 +w2 )

d P ()e

2
=

d2 P ()e2||

Examples for Wigner-Functions

1. Coherent state

|0 i: P () = (2) ( 0 ),
2 2|0 |2
e

W () =
If we take

0 = (X1 + iX2 ) /2

and

= (x1 + ix2 ) /2
W () =

as well as

D(0 )S() |0i

we nd

2 x21 /2 x22 /2
e
e

we nd a Gaussian in the complex plane with variances

2. Squeezed state: The squeezed state

x
j = Xj xj ,

12 = 22 = 1.

is a pure state, so we can write the density operator as

= D(0 )S() |0i h0| S ()D (0 )

Assume squeezing in

X1

direction, i.e.

= r > 0 , = 0.

Gaussian as before, but with a squeezed variance of

W (
x1 , x
2 ) =

Then we nd for the Wigner-Function a similar

V (
x1 ) = e2r < 1

and

2 x21 /2e2r x22 /2e2r

e
e

25

V (
x2 ) = e2r > 1

Proof:

W () =

1
2

h
D
i
E

1
d2 T r e(a ) (a) = 2
d2 0|S ()D (0 )e(a ) (a) D(0 )S()|0

we use (17) as well as (27) to nd:

1
W () = 2







d2 0|S () exp a cosh r a sinh r + 0 a cosh r a sinh r + + 0 S()|0

We dene the c-numbers

BCH

A cosh r + sinh r

1
W () = 2

d e

Again we dene the o-center coordinates

B
|A|

and

0|e

B = ( + 0 ) c.c.

Aa A a

1
|0 = 2

0 = (
x1 + i
x2 ) /2

and

= v + iw

to nd

= v 2 e2r + w2 e2r
1
2

dvdw eiwx1 eivx2 ew

exp[2r]/2 v 2 exp[2r]/2

Doing the remaining integrals like in (33) yields the result.

= |ni hn|,

we have without a proof:

W (
x1 , x
2 ) =
with

d2 eB e 2 |A|

(v + iw) (
x1 i
x2 ) /2 (v iw) (
x1 + i
x2 ) /2 = iw
x1 iv
x2

W (
x1 , x
2 ) =

3. In a number state,

and rewrite, with the use of the

2
n
(1) Ln (4r2 )e2r

r2 = x
21 + x
22 .

26

Open Quantum Systems

In this chapter we deal with systems in contact to an environment. The rst crucial step is to identify system and
environment by nding a partition of the Hilbert space,

H = HS H E

(34)

System and environment don't need to be spatially seperated. Next consider the Hamiltonian of the problem which
can be written as

H = HS + HE + HI
where

HS,E

act onto

HS,E

and only

HI

(35)

provides the coupling. It is crucial to assume that the coupling is weak, e.g.

||HI ||  ||HS ||, ||HE ||.

Environment

System
HS

HE

HI

4.1 The (Quantum-) Master Equation

Any state in the Hilbertspace (34) can be written with a basis of

|i =

HS , HE

{|s i , |e i},

as

c |s i |e i

,
Any general whole world density operator can be written

R=

r0 0 |s i |e i he 0 | hs0 |

,,0 , 0
Information on the system alone is fully contained in the reduced density matrix

S = T rE (R) =

because for any pure System operator

he | R |e i =

r0 |s i hs0 |

,,0

which acts as an identity to environment states, it holds

hOi = T r(OR) = T rS T rE (OR) = T rS ({O T rE (R)}) = T rS (OS )

as has been extensively proven in stat. Phys. Now the goal is to nd an evolution equation for

Dynamics of

R:

Use

} = 1,

alone.

The whole world density operator obeys the Liouville-vonNeumann eqn of motion

(S-Picture, the states carry the time dependence):

R = i [H, R]

(36)

The density operator with the entries

r0 0

represented by a possibly innitely long vector

(since

.
R

is hermitian, it can be taken as diagonal as well) can be

Since (36) is in any case linear in

R,

we can rewrite it as

= iLR
~
R
where the matrix

is the

Liouvillian

(37)

(or superoperator) of the system. Analogously, dene for any operator

O:

~
[H, O] LO
Analogously to the Hamiltonian (35), we can write
for

L = LS + LE + LI .

Next, consider the general solution of (36)

not dependant on time which can be rewritten using the BCH-Identity (18):

R(t) = eiHt R(0)eiHt = R it [H, R]

t2
[H, [H, R]] + ...
2

Drawing arrows to transform this expression into vector form, we nd

2
~
~ itLR
~ t L2 R
~ + ... = eitL R(0)
~
R(t)
=R
2

(38)

In the following, we will omit the arrows.

Relevant and irrelevant part of R:

for

R)

Dene a projection operator

P acting on any operator O

(we will use it

as

P (O) = E (t0 ) T rE (O)

It is easy to see, that

P2 = P.

The relevant part of

R,

Rrel P (R)
thus erases the correlations between System and environment, resets the environment state and makes a nonseperating

a separating one. Analogously, we use

Rirr = Q(R) = (1 P ) (R)

which contains the correlations alone. It holds that

Nakajima-Zwanzig-Equation:

P2 = P

and

QP = 0, Q + P = id.

We aim to derive an equation of motion for

(37). Take (37) and apply the projection operator and note that

=
P R = E (t0 ) T rE (R)

E (t0 )

Rrel

from the Liouville equation

is constant in time:

d
{E (t0 ) T rE (R)} = R rel
dt

R rel = P (iLR) = iP L(P + Q)R = iP LRrel iP LRirr

Now, if we have

L = L1 + L2

(39)

a general identity is

t
eiL(tt0 ) = eiL1 (tt0 ) i

ds eiL1 (ts) L2 eiL(st0 )

t0

This can be seen from the fact that the lhs. and the rhs. both fullll the same ODE,
same initial value for

t = t0 .

We apply this for

iL(tt0 )

=e

L = L id = LQ + LP

iLQ(tt0 )

and obtain

t
i ds eiLQ(ts) LP eiL(st0 )
t0
28

t X = iLX

and have the

We apply this to

R(t0 )

and act with

iL(tt0 )

Qe

from the left:

iLQ(tt0 )

R(t0 ) = Qe

t
R(t0 ) i ds QeiLQ(ts) LP eiL(st0 ) R(t0 )
t0

Use (38) to nd

iLQ(tt0 )

Rirr (t) = QR(t) = Qe

t
R(t0 ) i ds QeiLQ(ts) LP R(s)
t0

This expression can be inserted in the last term in (39):

t
R rel = iP LRrel iP LQeiLQ(tt0 ) R(t0 )
|
{z
}
source term

This is an exact integro-dierential equation for

Rrel (t)

ds P LQeiLQ(ts) LRrel (s)

(40)

t0

with a source term that has to be removed. We make three

assumptions:

1. The environment state is invariant under the environment dynnamics, e.g. a thermal state with
(E can still evolve under

HI !)

2. The initial state factorizes:

3. The interaction fullls

[HE , E ] = 0.

R(t0 ) = S (t0 ) E (t0 )

P LI P = 0

(MOST AUTHORS (Y&I, Breuer: Open Quantum Systems) ALSO RE-

QUIRE THAT NO BACKACTION OF SYSTEM TO ENVIRONMENT TAKES PLACE IN THE APPROXIMATION OF M-EQN (or the decay really fast, way faster than the coarse grained M-EQN descripton):

R(t) = S (t) E (t0 ) t

From assumption 2 we nd

P R(t0 ) = E (t0 ) T rE (S (t0 ) E (t0 )) = R(t0 ),

term vanishes. Via the integral, the time derivative of

Rrel

at time

thus

QR(t0 ) = 0

depends on times

s < t,

and the source

thus the equation is

non-Markovian.

Proof of

P LE = LE P = 0 and [P, LS ] = 0:

We further employ assumption 1 to proof for any Operator

1.

LE P O = [HE , E (t0 ) T rE (O)] = [HE , E ] = 0

P LE O = P [HE , O] = E (t0 ) T rE ([HE , O]) = E (t0 ) T rE (HE O OHE ) = 0
since the trace is cyclic invariant. Thus we nd:

P LE = LE P = 0

(41)

Next consider

P LS O

P [HS , O] = E (t0 ) T rE (HS O OHS ) = E (t0 ) (HS OS OS HS )

LS P O

[HS , P O] = [HS , E (t0 ) T rE (O)] = E (t0 ) (HS OS OS HS )

Further Simplications:

With these ndings and assumptions, we simplify the Kernel of (40):

P LQeiLQ(ts) LP R(s)

Q2 =Q

P LQeiLQ(ts) QLP R(s)

(42)

Here, we observe

P LQ = P (LS + LE + LI ) Q = LS P Q + 0Q + P LI Q = P LI Q

29

(43)

and

QLP = Q (LS + LE + LI ) P = QP LS + Q0 + QLI P = QLI P

Finally one word about

P 
k bk bk + 12 , E =

1 HE
and
Ze

k
environmental eld. Let us calculate

P LI P O

P LI P = 0

assumption 3:

is true for an often encountered example: Consider


P 
HI = XS gk bk + bk .

P LI P O

(44)

k
for any operator

= E (t0 ) T rE ([HI , E (t0 ) OS ]) = E (t0 )

HE =

This interaction exchanges energy with the

O:

n



o
gk T rE XS OS bk + bk E (t0 ) OS XS E (t0 ) bk + bk
=

= E (t0 ) [XS , OS ]

n

o
gk T rE
bk + bk E (t0 ) = 0

k
The last identity is because in the trace, the operators always contain one creator more or less than anhiliators and
the expectation value thus is zero. Now we use all gathered information on (40):

t
R rel

iP LRrel

ds P LQe

iLQ(ts)

ds P LI QeiLQ(ts) QLI Rrel (s)

LRrel (s) = iP LS P R
t0

t0

t
R rel = iLS Rrel ds P LI QeiLQ(ts) LI Rrel (s)
t0
This last equation is the

Nakajima-Zwanzig generalized Master-Eqn.

Approximations for Markovian Equation:

Weak coupling approximation,

obtain with

only keep leading order of

[LS + LE , Q] = [LS + LE , 1 P ] = 0

and

Q2 = Q

LI ,

this is justied for

||HI || (t t0 )  1.

We

for the exponential:

t
R rel = iLS Rrel

ds P LI Qei(LS +LE )(ts) LI Rrel (s)

t0

We employ

P LI Q = P LI (1 P ) = P LI

and switch back to commutators using (38):

t
E (t0 ) S = i [HS , E (t0 ) S ]

ds E (t0 )T rE

h

HI , ei(HS +HE )(ts) [HI , E (t0 ) S (s)] ei(HS +HE )(ts)

i

t0

Employ

Markov-Approximation, the system doesn't change much by

ei(HS +HE )(ts) S (s) ei(HS +HE )(ts) ,

this neglects

HI

HI

for the evolution of

in short time range:

S (t) '

which will be valid for short

time scales, since it will turn out that the integral kernel will decay fastly. Trace the whole equation with

T rE

to focus on the system:

t
S

i [HS , S ]

ds T rE

h
i
HI , ei(HS +HE )(ts) {HI E (t0 ) S (s) E (t0 ) S (s)HI } ei(HS +HE )(ts)

t0

exp [i (HS + HE ) (t s)] in the commutator and using E constant in time, we nd
I (s t) = exp [i (HS + HE ) (t s)] HI exp [+i (HS + HE ) (t s)] (HI transformed
denition H
Inserting

Picture wrt

HS + HE ):
h
h
ii
t
I (s t), E (t0 ) S (t)
S = i [HS , S ] ds T rE HI , H
t0

30

with the
in the I-

This is the quantum Master-Equation (QME). The environmental eect on the system dynamics is taken care of by
the integral which now is Markovian, e.g. does not involve the systems state in earlier times and thus is a normal
ODE.

4.2 Application: Damped Harmonic Oscillator

Consider the following assignments:



1
HS = 0 a a +
2

HE =

X
k

with the environment in a canonical state



1
k bk bk +
2

HI =

gk a + a



bk + bk

E = exp [HE /kT ] /Z .

For the use of the QME, we need to calculate

I (t):
H
I (t) = exp [i (HS + HE ) t] HI exp [i (HS + HE ) t] =
H

gk a ei0 t + a ei0 t



bk eik t + bk eik t

(45)

Rotating Wave Approximation (RWA):

Assume a weak coupling strenght,

|gk |  0 , k . There are terms in

(45) that would correspond to the creation or anhilliation of two excitations, one in the environment and one in the
system. These terms are not energy-conserving and strongly surpressed, alternatively they are fast rotating in the
interaction picture Hamiltonian

ei(0 +k )t

so that they can be assumed to average to zero when an integration

according to the interaction picture equations of motion is performed. Another argument goes as follows:
Consider an initial state of a system
Thus

d(hH n i)/dt = 0.

|i i.

[H, H] = 0

We have that

and also, for all integers

n,

that

[H n , H] = 0.

We then have

const =

X (is)n
n!

 X
hH n i = eisH =
p(E, t)eisE
E

We nd, that the QM conservation law for the energy tells us, that the probability distribution for the energy of the
system is conserved in time. That means if we obtain an energy distribution for 100 equally prepared systems right
at

t=0

we could also nd this distribution if we let them evolve for a time

t>0

an then take the measurement.

For example, if we have our quantumoptical system in the initial state of atomic ground at energy
excited mode eld at energy

we would nd

p(E) = |hE|i i|

which is

0 /2

widened by the (weak) interaction

the initial state is no eigenstate of the interacting Hamiltonian) and peaked around

E = 0 /2 + .

and

(since

It is unlikely

that we measure the system at a state where both the system and the mode are in excited states, since the conserved
probability

p(0 /2 + )

is quite low.

Continuation for Damped Harmonic Oscillator:

(t) =

We dene

gk bk eik t

k
and obtain

I,RW A (t) = a ei0 t (t) + a ei0 t (t)

H
For the QME integral kernel we calculate

T rE

h

h
ii
I (s t), E S (t)
HI , H





I (s t)E S (t) T rE H
I (s t)E S (t)HI
= T rE HI H




I (s t) + T rE E S (t)H
I (s t)HI (46)
T rE HI E S (t)H

As an example, we calculate the rst trace:



n
 o


I (s t)E S (t) = T rE a + a a ei0 (st) (s t) + a ei0 (st) (s t) E S (t)
T rE HI H

Dening

hOi = T rE (E O)

we can continue


{ (s t) a2 ei0 (st) + (s t) aa ei0 (st) +

 i (st)
+ (s t) a a e 0
+ h(s t)i a a ei0 (st) }S (t)
With the thermal state

(47)

we have that the annormal terms I and IV vanish due to orthogonality. For the integral

from the second term we nd


ds (s t) ei0 (st) =

I2 =
t0

0
n
o
X
X

i(0 p )(st)
ds gk gp T rE bk bp E e
=
d
|gk |2 n
k ei(0 k )
|
{z
}
k,p
k
t0
t0 t
p,k n
k

Since the environment is large, introduce a density of states per frequency,

0
d

I2 =

t0 t
We employ the assumption:

I2 =

d
()|g()|2 n
()ei(0 )
2

with

t0 t

dc e

=c

The integral can be done by employing a damping

0
d

0+1 ei0

t0 t

dx x eix0

0+1 ei0
0

in order to ensure convergence.

( + 1)
+1

(i0 )

0
d ei0

'c

( + 1)
+1

(i )

Since the bath correlations decay rapidly, the kernel of the integral decays fastly for
the lower bound by

Re c

d ei0

0  1. Thus we could replace

Now we turn to the real part, while the imaginary part will be taken care of later:

x = /0

then with substitution

t0 t

I2 = c

> 0,

0
i(0 )

(48)

1
2
()
2 ()|g()| n

():

( + 1)
(i )

+1

c
2

0
d ei0

( + 1)
+1

(i )

c
2
c
2

0
i0

d e

d ei0

( + 1)
+1

(i )

( + 1)
+1

(i )

Summary: The system density matrix

c
2

0
d ei0

c
+
2

ds ei0 s
0

( + 1)
(i )

+1

( + 1)
+1

(is)

= c0

changes due to coupling to the environment on a time scale

|gk |

, the

1
correlations of the environment decay on a time scale 0
as we have seen. In Quantum-Optics we usually have
1
|gk | 0  1 so that the bath correlations decay is the coupling limiting process and the integral boundary in (48)
can be replaced by

Quicker Route:

In (48) we write

right away:

I2 =

d
()|g()|2 n
()
2

0
d ei(0 )

32

and use the identity

0
d ei(0 ) = (0 ) i P V

to arrive at

1
I2 = (0 )|g(0 )|2 n
(0 ) + i P V
2

1
0

()
d ()|g()|2 n
2
0

0
With the denitions

(0 )|g(0 )|2

d ()|g()|2 n
()
PV
2
0
0

PV

d ()|g()|2 {
n() + 1}
2
0

0
we obtain in total for the two non-vanishing terms of the rst trace:

Final Result:

I3

n
(0 ) + i
2

[
n(0 ) + 1] + i
2

Our purpose was to solve the QME where we only have explicitly calculated very few terms:

t
S

I2

i [HS , S ]

ds T rE

h
h
ii
I (s t), E S (t)
HI , H

t0

t
S



i 0 a a, S

ds {T r1 + T r2 + T r3 + T r4 }
t0

t




 

i 0 a a, S
n
(0 ) + i aa S
[
n(0 ) + 1] + i a aS ds {T r2 + T r3 + T r4 }
2
2
t0

The other terms from the remaining traces are similar but with other order of

a, a , S . The complex terms inlcuding

can be absorbed in the leading commutator where they cause a redenition of the frequency





S = i 0,0 a a, S + 2 n
(0 ) 2a S a aa S S aa +
Interpretation:

Consider a system Operator

with

0 0,0 :



[
n(0 ) + 1] 2aS a a aS S a a

hOi = T r (OS ).

We are interested in the amplitude

evolution (S-Picture!!). Use that the trace is cyclic:

d

hai = T r (a S ) = n
(0 ) T r 2aa S a aaa S aS aa + [
n(0 ) + 1] T r 2aaS a aa aS aS a a
dt
2
2
d
hai =
dt
d
hai =
dt
d
hai =
dt
d
hai (t)
dt

n
(0 ) T r aa S a aS aa + [
n(0 ) + 1] T r aaS a aa aS
2
2

n
(0 ) T r S a + a aS a aS aa + [
n(0 ) + 1] T r aaS a a aaS aS
2
2

n
(0 ) T r (S a) [
n(0 ) + 1] T r (aS )
2
2

= hai (t)
2
33

We nd an exponential decay

hai (t) = hai (0) exp [t/2] of the amplitude

(with (25) we nd

hEi 0 as expected)

and we further obtain in a similar way



a a = a a (0) et + n
(0 ) 1 et = n
(0 )
t

Thus while the electric (and magnetic) eld

energy

expectation

values approach zero, the photon number and also the

approach the thermal equilibrium values for the bath temperature! In QM, this is not a contradiction,

since the energy of the eld is given by terms

E2

which are not zero, even if

hEi = 0.

One can also show that the

whole density matrix will approach a canoncial state with bath temperature.

4.3 Representations of the Quantum-Master-Equation

Various representations are useful to solve the damped oscillator QME.

4.3.1

Representations in number states

Consider the QME of the last section example and form matrix elements by applying

n,m
n,m
n,m

hn|...|mi:

X


hn| a a |ji hj| S |mi hn| S |ji hj| a a |mi + n
(0 ) 2 hn| a |ji hj| S |ki hk| a |mi ...
2
j
k,j
X
X

(0 ) 2 nj,n1 hj| S |ki mm,k+1 ...

= i0,0
(nn,j hj| S |mi hn| S |ji mm,j ) + n
2
j
k,j

= i0,0 (n m) n,m + n
(0 ) n hn 1| S |m 1i m ...
= i0,0

Including the other terms which have not been calculated explicitly

n,m

= i0,0 (n m) n,m +




p

n+m+2
n+m
nmn1,m1
(n + 1) (m + 1)n+1,m+1
+ n
(0 )
n,m + [
n,m
n(0 ) + 1]
2
2

We are now interested in the populations

pn n,n

(the diagonal elements) only:

= n
(0 ) {npn1 (n + 1) pn } + [
n(0 ) + 1] {(n + 1) pn+1 npn }

pn

The steady state condition

pn = 0

leads to

n
(0 ) {npn1 (n + 1) pn }

[
n(0 ) + 1] {(n + 1) pn+1 npn }

The solution is given by

pn
pn1

}0
n
(0 )
= e kT
n
(0 ) + 1

{npn [
n(0 ) + 1] n
(0 ) (n + 1) pn }
{npn n
(0 )npn }

= {(n + 1) pn n
(0 ) npn [
n(0 ) + 1]}
= {npn n
(0 ) npn }

Thus we end up with a thermal state at the reservoir's temperature:

pSteadyState
=
n

34

0
1 }
kT n
Ze

...true

4.3.2

P-Representations and Fokker-Planck equations

As a prerequisite, we aim to calculate

1
2

+ i y

(|i h|)

for

= x + iy

with (19) and with

1
2

i y
;

=
=

(|i h|)

(|i h|)

n
X ( a)m
X m
2 X a
||2 X n

e||

|ni
hm| =
e
|0i h0|

n!
m!
n!
m!
n0
n0
m0
m0





2
2
1

i
e(x +y ) Exp (x + iy) a |0i h0| Exp [(x iy) a]
2 x
y






1 (x2 +y2 )
e
a Exp (x + iy) a |0i h0| Exp [(x iy) a] + Exp (x + iy) a |0i h0| a Exp [(x iy) a] +
2






2
2
1
+ e(x +y ) a Exp (x + iy) a |0i h0| Exp [(x iy) a] Exp (x + iy) a |0i h0| a Exp [(x iy) a]
2




2
2
1
+ (2x + 2iy) e(x +y ) Exp (x + iy) a |0i h0| Exp [(x iy) a]
2


a |i h|

With these preperations let us write as in the density operator in p-representation:

S = =
Consider the expression

d2 P () |i h|

in the QME and use the previously computed identity:

a S =

d2 P () a (|i h|) =

d2 P () ([ + ] |i h|)

We use a partial integration where the boundary terms vanish due to

P () 0

for

||

(normalization).

a S

d2 {( + ) P ()} |i h|

Thus we can establish the mapping

a S ( ) P ()
and its analogs

aS

P ()

S a

P ()

S a

a aS

S a a

( ) P ()
( ) P ()
( ) P ()

We use that to transform the damped oscillator QME to a Fokker-Planck-Type Equation in

the commutator term by going into the interaction picture, w.r.t.

aI (0)ei0,0 t

and

S,I = eiH0,0 t S eiH0,0 t .

The time derivative of

H0,0 = 0,0 a a,

S,I

thus

P,

where we get rid of

aI (t) = eiH0,0 t a eiH0,0 t =

yields

S,I = iH0,0 S,I + eiH0,0 t S eiH0,0 t + iS,I H0,0 = eiH0,0 t S eiH0,0 t + i [H0,0 , S,I ]
We apply

eiH0,0 t ...eiH0,0 t

to the QME:

S = i [H0,0 , S ] +





n
(0 ) 2a S a aa S S aa + [
n(0 ) + 1] 2aS a a aS S a a
2
2

35

eiH0,0 t S eiH0,0 t + i [H0,0 , S,I ]

S,I

n
(0 ) 2eiH0,0 t a eiH0,0 t S,I eiH0,0 t a eiH0,0 t ... + ...
2


n
(0 ) 2aI (0) ei0,0 t S,I aI (0) e+i0,0 t ... + ...
2

One nds with swapping the exponential time dependences of the creator and annhilliator

S,I

=
+




n
(0 ) 2aI (t)S,I aI (t) aI (t)aI (t)S,I S,I aI (t)aI (t) +
2



(
n(0 ) + 1) 2aI (t)S,I aI (t) aI (t)aI (t)S,I S,I aI (t)aI (t)
2

Now, we employ the replacement rules from above to map the QME to a FPE for
skipped)

t PI (, t) =
This can be rewritten by introducing
Driftvector

h
2

in the interaction picture (steps

i
2
( + ) + n
(0 ),
PI (, t)

= x1 + ix2

and using a vector notation

x = (x1 , x2 )

= (x1 , x2 )

A = 2 x (moving PI (x, t) to origin), Diusionmatrix D = n

(0 )diag(1, 1) (shaping PI (x, t) to circular

shape):

2
X
1
t PI (x, t) = A +
xj xl Dj,l PI (x, t)
2
j,l=1

Application of the FPE:

Consider the expectation values dened as

hxj i

dx1 dx2 xj P (x1 , x2 )

Employing the FPE we nd by the use of integration by parts (diusion plays no role for expectation value evolution)

d
hxj i =
dt
d
hxj i =
dt
d
hxj i =
dt

2
X
1

xj xl Dj,l P (x, t) = dx1 dx2 xj (P 1 a1 + a1 1 P + P 2 a2 + a2 2 P )

dx1 dx2 xj A +
2
j,l=1

dx1 dx2 (xj P 1 a1 P 1 (xj a1 ) + xj P 2 a2 P 2 (a2 xj ))

dx1 dx2 (a1 P j,1 a2 P j,2 ) = hAj i

Similary, we have

d
1
hxk xl i = hxk Al i + hxl Ak i + hDkl + Dlk i
dt
2
This can be applied to our specic case with

A, D

given as above:

hxj i = hxj i
dt
2

hxj i (t) = hxj i (0) e 2 t

as well as

d
2
x
dt 1


x21 + n
(0 )


d
d
2
2
V (x1 ) = x21 + n
(0 )
hx1 i = x21 + n
(0 ) + hxj i = V (x1 ) + n
(0 )
dt
dt

The solution is given by

d
V (x1 , t) = V (x1 , 0)et + n
(0 ) 1 et n
(0 )
dt
If we start, for example, in a squeezed state, we will end up in a thermal state that has lost all information about
the initial state:

36

X2

(InteractionPicture:
non rotating)

X1

4.3.3

Steady state solutions of the Fokker-Planck-Equation

The steady state condition for the generalized FPE with some given

xj

1X
Aj +
xl Djl
2

Aj , Djl

reads

!
PI (x, t) = 0

l=1

Try the Ansatz for

fullling the steady-state condition:

Aj P =
Divide by P (assume

l=1

l=1

l=1

X
X
1X
xl (Djl P ) 7 2Aj P P
xl Djl =
Djl xl P
2

P 6= 0)
2
2
X
X
Djl xl lnP = 2Aj
xl Djl
l=1

Assume

P (x) = e

(x)

/N

and

l=1

invertible. Then we nd

"
#
2
2
X

1X
1
xl Djl Fk (x)

= 2 Dkj Aj
xk
2
j=1
l=1

where

Fk (x)

is known from the problem. In short,

F = .

If for

and

holds

2
Fl
Fk
2
=
=
=
xk xl
xk
xl
xl xk

we say that the potential conditions are satised. Then we can obtain

by integration where the path is arbitrary:

x
F dv

(x) =
As an example consider again the damped oscillator,

(x) = (/ n
(0 ))

A = 2 x, D1 = diag(1, 1)/ n
(0 ): xl Djl = 0

x
1

x1
d
x1 x
1 +

d
x1 x
1
0

the steady state distribution is a Gaussian with zero mean and variance
thermal state:

P (x) = e

x2 +x2
2
0)

1
2n
(

37

/N

x21 + x22
2
n(0 )

n
(0 )

which is the p-representation of a

Interactions Between Atoms and Radiation

We will consider atoms (mostly alkali atoms) interacting with the radiation eld. The elds couple to the charges
of electrons and nuclei via the minimal coupling Hamiltonian. We use the convention of
charge is

q = e.

H=

rj = H/pj

and the electrons

1
2
(p + eA(r, t)) eV (r) + Hrad
2m

Consistency check: Lorentz force for classical particle

use

e>0

Minimal coupling form of Hamiltonian:

and

pj = H/rj .

(49)

Consider the classical minimal coupling Hamiltonian,

We have

rj =
pj = ej V

1
(pj + eAj )
m

X
eX
(pl + eAl ) j Al = ej V e rl j Al
m
l

X
X
m
rj = pj + e rl l Aj = ej V + e
(rl l Aj rl j Al ) mr = eE e (r B)
l

5.1 Local gauge invariance and minimal coupling

Consider a free electron SE

i}t = }2 /2m 2

under a local gauge transformation:

e (r, t) = ei(r,t) (r, t).

The

SE should be invariant under this phase transformation, so add terms which can be adjusted after the transformation:

"

1
i}t =
2m

}
+ eA
i

2

eV + Hrad

Now, do the transformation:

"

1
i}t =
2m
The SE is now invariant, if

and

#

2


}
}
}
+ e A + (r, t)
e V + Hrad
i
e
e

transform as:

}
Ve = V
e

e = A + }
A
e
This gauge transformation must leave the observable elds

E, B

invariant which can be approved. The minimal

coupling Hamiltonian (49) is not the only Hamiltonian with can be adjusted, so that the SE is invariant under local
transformations, but it is the one of lowest order.

5.2 Long wavelength approximation & Electron Dipole Representation

Consider a system of charged particles and a radiation eld where a possible magnetic eld eect is ignored.

H=



X 1
X
1 X qj ql
1
2
(pj qj A(rj )) +
+
}k,v ak,v ak,v +
2mj
80
|rj rl |
2
j
j,l

We consider neutral atoms, so that

origin and assume that

qj = 0

and

k,v

aBohr / '= 104  1.

Let the atom core be centered at the

A(rj ), which varies at a spatial scale of , has a spatially constant value in the region where

the wavefunctions are essentially nonzero. Then

H=



X 1
X
1
2
(pj qj A(0)) + Vcoul +
}k,v ak,v ak,v +
2mj
2
j
k,v

38

(50)

Dene the atomic dipole moment and see how it enters the expression:

P
qj rj . Consider a unitary transformation
j

(|

i = T |i

and

H 0 = T HT

describes the same physics with the same SE as in the unprimed case)



X
X
i
i
a

T = exp d A(0) = exp
qj rj A(0) = exp
k,v k,v k,v a
k,v
}
} j
k,v

with

A(0) =

Pq
k,v

}
2k 0

uk,v (0)ak,v + uk,v (0)ak,v

k,v =

In the exponential of

are position operators, so

and thus

uk,v (0) d
2k }0

acts trivally on

(and thus on

Vcoul

as well as on

A(0))

but

not on the momentum operator where we need (18):

X
X
i
i
p0j = T pj T = exp
qj rj A(0) pj exp
qj rj A(0) = pj + qj A(0)
} j
} j
Similary we get

k,v
a
0k,v = T a
k,v T = a
k,v +

and

,
Ta
k,v T = a
k,v +
k,v

0
0
0
H 0 = T HT = Hatom
+ Hrad
+ Hint
The parts of the new primed Hamiltonian, described by the old unprimed operators are in detail:

0
Hatom

0
Hrad

0
Hint

X 1
X 1
2
p2j + Vcoul +
|uk,v (0) d|
2m
2
j
0
j
k,v


X
1

}k,v ak,v ak,v +

2
k,v


X

a
k,v a

}k,v

k,v
k,v
k,v = d E(r = 0)
k,v

The atomic Hamiltonian contains an additional term related to the possible light modes and the dipole moment.
This term contains the operator

and is thus a non-constant perturbation to the potential. It is due to the fact,

that the atom, as treated in the QMI lecture does not exist, it inevitably is aected by the electromagnetic eld
around.

Important:

Note, that in the primed system, we have to use

H0

to generate the H-eqn. of motion for the operator

actually is the dielectric displacement operator

appearing in H 0 . It turns out (AckehaltMilonni1984) that E
E
int
=D
0 = E
0 + 4P0 in the primed picture. Whenever we want to do caculations in states we have specied in
E
the unprimed system, we have to remember to transform them according to

| 0 i = T |i

before we act with the

primed Hamiltonian! The interaction in dipole approximation is manifestly gauge invariant since the dipole moment
couples on the observable dielectric displacement directly.
We now use a set of energy Eigenstates (dened in the primed picture!) for

0
Hatom
,|i.

This is most suitable for

alkali atoms, where the valence electron is weakly bound and can be treated with an eective core potential. Then
the dipole operator can be written

d=

|i h| d |i h| =

X
d |i h|
,

39

If we chose plane waves,

uk,v = ev Exp[ikr]/ V

0
Hint

we have (in old unprimed operators!)

r


XX
}k
e
e
E(0)

v a
= d
=
d |i h| i
k,v v a
k,v
20
V
V
, k,v

If we focus on a case with a specied (real) polarisation

r
g,,k = i

and dene

k
ev d
}20 V

(51)

we nd

0
Hint
=}

X 

g,,k ak + g,,k
ak |i h|

(52)

,,k

5.3 Two-level-atom and Jaynes-Cumming-Model

We want to estimate the magnitude of

g,,k

to a volume smaller than approximately

the

D2

line connecting states

|i , |i

with

that we cannot conne photons

(box with boundary conditions). Using

= 2c/ we get for Rb87 with

= 780nm

r
}|g,,k |
Therefore the weak coupling condition
(this will reduce the sum over

and observe as a natural limit on

and

|d | = 2.56 10

29

cm.

}2c
|d | 6.9 1024 J 6.1 1010 Hz
20 4

|g,,k |  |E E | = 2.56 1019 J

is fullled and the RWA will be applied

in 52). Assume that only one transition couples appreciably to radiation and use

spin notation for the two level system:

= |i h| |i h|

= |i h|

= |i h|

The Hamiltonian can be written as

H=




X
X
}0
1
z +
}k ak ak +
+}
gk ak + + gk ak
2
2
k

(53)

If one restricts oneself to only one radiative mode, we get the celebrated

Jaynes-Cummings-Model (JCM). It is

a good approximation for atoms in a cavity, since there the radiation indeed has a discrete spectrum with only one
relevant mode.

HJCM =

}0
2 z

+ }r a a + } ga+ + ga

5.4 Dressed States and Rabi Oscillations

Consider a two level atom in JCM-approximation with ground- and excited state

|gi , |ei.

We see that the number

of excitations is conserved since the relevant operator commutes with the Hamiltonian


 

a a + |ei he| , HJCM = a a + |ei he| , ga |ei hg| + ga |gi he| =
ga aa |ei hg| + ga |ei hg| + ga aa |gi he| gaa a |ei hg| ga a a |gi he| ga |gi he| = 0
With this nding, we can diagonalize the Hamiltonian in independent subblocks for each excitation number. In
the n-excitation subspace consider the two possible states

40

|n, gi =(0,
1)T

and

|n 1, ei =(1,
0)T .

Then forming the

matrix elements of

HJCM ,

the new, simplied Hamiltonian reads with detuning

g n

H=}

!
g n
0

0 
+ } r n
2


= 0 r

Diagonalizing, the eigensystem is

En, = }r n
For


p
}
}0
2 + 4ng 2

2
2

we have:

En, = }r n
with Eigenstates (which are also called

}0
}g n
2

dressed states) :

|n 1, ei |n, gi 1

|ni =
=

2
2

!)

The dynamics of an arbitrary state

|n (t)i = cn1,e (t) |n 1, ei + cn,g (t) |n, gi

can be determined by solving the SE (S-Picture)

cn1,e (t)

"
= i

cn,g (t)

!
g n

g n

For example, if we start with an atomic excited state

atomic state:

pe (t) =
with Rabi-Frequency



1
+ 0 n
2

cn1,e (0) = 1,

!#

cn1,e (t)

cn,g (t)

we obtain for the probability of an excited


1
1 + cos 2 ngt
2

R = 2 ng

doubly degenerate with g=0

Pe(t)

1
0
t

2excitations,
2 21/2g Splitting
|e>|1> or |g>|2>
|e>|0> or |g>|1>

2g Splitting

|g>|0>

5.5 Spontanous emission of a Two Level Atom

Any atom couples to its electromagnetic environment according to the Hamiltonian (53) where the polarizations
have been added

H=


X
X
}0

z +
}k ak ak +
gk,v ak,v + + gk,v
ak,v
2
k

k,v

41

Since the coupling is weak,

|gk,v |  0 }

we can use the M-Eqn approach and treat the modes

15

0 10 Hz

environment for the atom. For typical values of

and

T = 300K

ak,v

as a quantum

or lower, the light modes close to

resonance are practically empty:

n
(0 ) =

1
' 1013
exp [}0 ] 1

Here we use a trick: Take the situation from section 4.2 where we treated not a damped atom but a damped
harmonic oscillator. In the derivation, we never used the commutation relations and we essentially think of our two
level system (atom) as the two lowest level of the oscillator ladder. With the replacements

and

a +

we rewrite the Master Equation:

at = i
The question is what is

i
z , at + {2 at + + at at + }
2
2
0,0

(54)

in this context. If we go back to (47), we note that only the third term is not zero by the

I3

emtpy modes approximation. The

integral is given similary to the

I2

integral with 1 instead of

n
k

by

0
d

I3 =

k,v
easily nd as in the previous chapter (with

d2
I3 =
6 2 }0 c3

L 3
2

|gk,v |2 ei(0 k )

k,v

tt0

If we use 2 polarisations and replace

d3 k

as well as replacing the dot product in (51) by

= 3-Kernel

0
tt0

we

in this realistic situation)

3 i(0 )

d cos

d e
0

d2
= 2
}0 c3

0
d

ei0
4

The real part can be calculated as

d2
1
Re[I3 ] = 2
}0 c3 2

ei0
d2 03

=
=
4

6}0 c3
2

while the imaginary part formally diverges. This is due to the use of the long wavelength approximation, which
should only be used for a nite range of
redenition of

in

0,0 .

Thus, also the imaginary part has a nite value, which is absorbed in a

Thus, the measureable transistion frequencies of atoms are changed by the surrounding

vacuum.
If we are interested in the fate of an excited state, we put

pe = he|at |ei

in (54):

"
pe

= i0,0

#
X

he|z |i h| at |ei he|at |i h| z |ei +

=e,g

X
{2 he| |i h| at |i h| + |ei
2 ,=e,g
he| + |i h| |i h| at |ei he| at |i h| + |i h| |ei}

The rst and the second (due to

= |gi he|) line vanish, and the third only contributes for = g

as expected, the excited state decays exponentially with damping rate

pe =

{2pe }
2

and

pe (t) = pe (0)et

A similar calculation gives the time dependance of the coherence,

he| at |gi = T r(at ) = h i

h i (t) = h i (0) exp [i0,0 + /2] t

42

as

= e. Thus,

which decays at a rate

/2

and oscillates.

5.6 Resonance Fluorescence

Consider a two level atom that is continously driven by a coherent laser mode and emits photons in the surrounding
vacuum.

|e>

Laser

spontanous
emission in
electromagnetic
vacuum
(measure spectrum)

|g>

The corresponding Hamiltonian with the laser mode

H=

kL

is given by the same expression as above,




X
X 
}0

z +
}k ak ak + }
gk ak + + gk ak + }L aL aL + } gL aL + + gL
aL
2
k6=kL

k6=kL

It diers by the previous case by the initial state which is

|(0)i = |L i |0k6=kL i |atom i

We will apply a M-Eqn treatment with the assumption that on the coarse grained temporal description it provides,
the environment state is not altered from the initial state, thus we can model the continous laser beam as above,
although to be strict, in the model given, the energy would be redistributed from the laser mode to the atom and
other modes and a steady state with reduced laser intensity would be reached.

Transformation to vacuum initial state and classical driving eld expression:

Instead of naively using

the SE to compute the time evolution of the state, we consider a unitary transformation

 E

1
(L (t)) |i
 = DL

with (see below)

L (t) = L eiL t

Then,


E

(0) = |0L i |0k6=kL i |atom i
1
DL
(L (t)) = DL (L (t)) would shift the
 laser
E state for all times t back to the vacuum state, if we had a free

photon hamiltonian (this is not the case here, (t)
evolves away from the initial vacuum state!). The dynamics of
h
i

the free laser photons with HL = }L aL aL in the S-Picture is given by |L (t)i = exp iL aL aL t |L (0)i. Thus
Here,

we have

h
i
n
X L (t)n
2 X L (0)
!

|ni = exp iL aL aL t e|L (0)|

|ni
n!
n!
n0
n0
h
i

of the number states, exp iL aL aL t |ni = exp [iL nt] |ni.

2

|L (t)i = e|L (t)|

With the H-Eigenvector properties

|L (t)i = e

|L (0)|2

X L (0)eiL t

n!
n0
43

n
|ni

Thus

1
L (t) = L eiL t . Then the claim follows by acting with DL
(L (t)) on DL (L (t)) |0i = |L (t)i.
 E



1

interested in the equation of motion for  . With D

(L (t)) = D(L (t)) = exp L (t)a + L
(t)a

and we can infer:

We are

we compute

d  E
i} 
dt


d 1
d
= i}
D (L (t)) |i + D1 (L (t)) i} |i
dt
dt



 E
d

L (t)a + L (t)a  + D1 (L (t)) HD (L (t)) D1 (L (t)) |i

= i}
|
{z
}
dt
1

We go on by using

(
D

(L (t))

aL
aL

)
D (L (t)) =

aL + L (t)

aL + L (t)

in the second part on the rhs and nally obtain:

i}

d  E

dt

 E

 E


 E


= }L L (t)a + L
(t)a  + H  + } L |L (t)|2 + L L (t)a + L (t)a  +
 E


+ (}gL L (t)+ + }gL
L (t) ) 

The rst and the fourth block cancel, the third is constant in time and thus produces just an additional phase.
Thus, for the transformed state

we have a corresponding Hamiltonian:

= H + }L (t)+ + }L (t)
H

with a c-number

We have transformed the problem to an equivalent problem, where

all

L (t) = gL L eiL t

modes are in vacuum initially but a

(55)

classical

driving eld shows up in the Hamiltonian.

Transformed problem: M-Eqn treatment and Matrix-Equation:

We can now perform the analysis along

the lines of the spontanous emission problem with an M-Eqn treatment where we assume all radiation belongs to
the thermal (n (0 )

at =

' 0)

environment. Only the operator in the commutator is changed.

i
i h 0,0
}
z + }L eiL t + + }L eiL t , at + {2 at + + at at + }
}
2
2
eiL z t/2 which leads with
AI (t) = eiL z t/2 A(t)
+ +,I (t) = + eiL t and ,I (t) = eiL t where + is the

We transform from this S-Picture to an interaction picture with:

(18) and

[z , + ] = 2+

to the replacement

S-Picture operator (z doesn't change). The implicit time dependance of the density operator changes accordingly

I (t) = eiL z t/2 (t) eiL z t/2 ,


at = i

but we will skip any index in the following and use

= 0,0 L :

z + L + + L , at + {2 at + + at at + }
2
2

We note, that we have no explicit time dependance any more. We assume

L to be real, note hj i (t) = T r {j,I (t)I (t)}

and consider for example

d
h+ i (t)
dt

= iL T r {+,I (t)I (t)} + T r {+,I (t) I (t)}

= iL h+ i (t) + Z

44

where

is given by the use of

+ + = 0

and similar equalities as well as the cyclic invariance of the trace




Z = e
T r + i
z + L , I (t) + + {I (t)+ }
2
2
 
 



z + 1
1 z
1 z
iL t
= e
T r i + I (t) + L
I (t) + I (t) +
I (t)L

I (t)+
2
2
2
2
2 2
n
o

= eiL t T r i (+ I (t) + L z I (t)) + I (t)

2


h+ i (t) iL eiL t hz i (t)
=
i
2

d
iL t
Thus,
hz i (t). The equation for h i (t) is just the complex conjugate
dt h+ i (t) = i0 2 h+ i (t) iL e
iL t

and for

hz i (t),

we obtain:

d
hz i (t) = T r {z I (t)}
dt
d
hz i (t)
dt

= T r {iL z [+ + + ] (z + 1) }
= T r {2iL [+ ]} hz i (t) = 2iL (h+ i (t) h i (t)) hz i (t)

In total, we have to consider:

h+ i (t)

v0 = h i (t)
hz i (t)
and we nd:

d 0
v = A0 v 0 + b
dt

with

i0

A0 =

i0

2iL
These equations are termed

iL eiL t

(56)

iL eiL t

2iL

Optical Bloch Equations.

b= 0

There is a trick to simplify the problem and replace

A0

by a

matrix constant in time. To this end we dene

h
+ i (t)

h+ i (t) eiL t

v = h
i (t) h i (t) e+iL t
h
z i (t)
hz i (t)
Then we nd

d
v = Av + b
dt

(57)

with

i /2

0
2iL

i /2
2iL

A=

Steady state solution:

Consider

t f (t) = af (t) + b,

iL

b= 0

iL

thus

f (t) = b/a + c eat

and for

Re(a) < 0, f (t ) =

b/a. It can be shown, that (57) has a nite steady state, as the real parts of the eigenvalues are negative. Then the
steady state solution is given by

vss = A1 b

h
+ iss = h+ iss eiL t =
From these solutions we see, that

(as the nonzero Eigenvalues also show, that

2i ( + i) L
2 + 42 + 82L

hz iss < 0,

h iss = h+ iss

hz iss =

A1

exists) , in detail

2 + 42
2 + 42 + 82L

meaning a simple laser drive can never generate population inversion

(that means, the ground state occupation probability is always higher than the excited state population probability).

45

We further see, that for vanishing laser power, i.e.

General solution:

|L | 0

and thus

General solutions of (57) for zero detuning

L 0,

the ground state is reached.

and initial ground state can be obtained from

the equivalent form

t (v vss ) = A(v vss )

(58)

and read:

with

(|L |

hz i (t)

h+ i (t)

2 /4 162L /2.

< /8,

We have two regimes, depending if

is imaginary (|L |

> /8,

strong driving) or real

<z>

strong driving

+1

weak driving

t
-1

-1

5.6.1

weak driving).

<z>
+1




82L
3
3t/4
1e
sinh t 1
cosh t +
2 + 82L
4






3
3t/4
2iL 2
1e
cosh t +
+
sinh t eiL t
+ 82L

16

Interlude: Power Spectrum of the emitted light

Consider

s
D
E

 X }2 k k 0

+
ik t ik0 (t+ )
0e
0
E (r, t)E (r, t + ) =
u
u
e
a
a
k
k
k
k
420
0
k,k

For a stationary eld, this quantitiy should be independent of

implies

ak ak0

= kk0 nk

X }k
20

k
+

Integral has been extended to

energy in the eld,

|uk |2 eik nk =

1
20 c3

since

though) and thus

k = k0 .

This

nk

1
20

d3 k
}k eik nk =
(2)3

d 2
}ei nk
(2)3

is zero for

k < 0.

(59)

Compare this expression to the overall

X
1
d 2
H=
}k nk + const =
d 3
}nk + const.
c
(2)3
k

Thus we see that

(it depends on

so that


E (r, t)E+ (r, t + ) ss

where the

hE (r, t)E+ (r, t + )iss

can be regarded as the Fourier-Transform of the energy per frequency

interval, which leads after an inverse transformation:

S(r, ) = lim

d ei hE (r, t)E+ (r, t + )i

46

5.6.2

Relation between emitted eld and atomic coherence

In the previous subsection we have seen that in order to compute the power spectrum, we need to know the emitted
electric eld. Therefore, we consider the Hamiltonian (55) (where the driving laser was represented classically) which
reads after a slight redenition of

g:

X
X
= }0 z +
H
}k ak ak i
(gk ak + h.c.) + }L (t)+ + }L (t)
2
k

(60)

The Heisenberg equations of motion read:

gk
i
a k = [H, ak ] = ik ak
}
}
and thus

ak (t) = e

ik t

gk
ak (0)
}

t
ds (s) eik (ts)

(61)

t
r
X }k
X k d eikr
ikr
ds (s) eik (ts)
E (r, t) = i
e ak (t) =
20 V
20 V
+

where the rst part of (61) rhs has been dropped, since it will eventually cancel when

hE (r, t)E+ (r, t + )i is formed

with the (vacuum-) ground state. We replace the k-Sum by an integral and note in a second step

d
E (r, t) =
16 3 0

t
3

d k k e

ikr

ds (s) e

ik (ts)

d
=
8 2 0 c2 r

We obtain

(s) = i0 + z
0  |gk |/}, |L (t)|

|L |.

(
X  g
k

The

(s)

d e

 t

ds (s) e

is

ak

(t) = ei0 t (t)

[z , ] = 2 .

is slowly varying on the

and compute the integrals

d 2 ei(r/c+st)

ds
(s) e
0

and

+ L (t)

i0 s

ds (s) ei(ts)

with subsequent

and we continue with taking the

out of the remaining integral at its peak value.

I =

ds
(s) e
0

We are interested in the high

t + r/c

E (r, t)

peaks at

ir/c

integral is calculated by introducing a cuto,

slowly varying

t
2 i(r/ct)

ir/c

we know that the quantity

We return to the calculation of

I =

(s) from (60) using [+ , ] = z

scales of |gk |/},

To continue, we also need the H-picture time evolution of

By the notion that

k r = /c cos:

i0 s

2i
3

(r/c + s t)

t
3
= 2i
(t r/c) ds ei0 s / (r/c + s t)
0

range. The integral with

1/ (s (t + r/c))

can be taken to zero, since the kernel

outside the integration range. The limits of the remaining integral can be taken to innity, because

its kernel is essentially nonzero only between 0 and t. Thus we get

I = 2i
(t r/c)

ds ei0 s / (r/c + s t) = 2
(t r/c)02 ei0 (tr/c) = 202 (t r/c)

and nally, still in the H-Picture:

E+ (r, t) =

02 d
(t r/c)
40 c2 r

(62)

This result makes sence, since the electric eld is dependant on the coherence of the atom (a dynamic quantity

47

giving a measure of the probability to change between the two levels at an earlier time) Here, we have ignored the
polarization of the eld. A general treatment reveals

02 d 
r r
d
(t r/c)
2
40 c r
r
r

E+ (r, t) =

(63)

The power spectrum is given by

S(r, )

lim

d e

 I0 (r)
E (r, t)E (r, t + ) =
2

I0 (r)

d ei G( )

G( )

lim h+ (t) (t + )i h+ ( )iss


2
 02 d 
r  r 

 40 c2 r d r r 
t

Thus we are led to the task to compute the quantity

Quantum Regression Theorem (QRT)

h+ ( )iss .

Theorem: Assume a set of operators

Yj

fullls

X
d
hYj (t)i =
Gjl (t) hYl (t)i
dt

(64)

where no assumptions on the initial states have been made (they do not need to be even physical). Them the two
time correlation functions fullll the same equation of motion for

instead of

t:

X
d
hYk (t)Yj (t + )i =
Gjl ( ) hYk (t)Yj (t + )i
d

(65)

Proof: In the H-Picture, we have

hYi (t)i = T r eiHt Yj eiHt 0 = T r Yj eiHt 0 eiHt

(66)

We compare this expression with





hYk (t)Yj (t + )i = T r eiHt Yk eiHt eiH(t+ ) Yj eiH(t+ ) 0 = T r Yk eiH Yj eiH eiHt 0 eiHt

= T r Yk eiH Yj eiH 0 (t) = T r Yj eiH 0 (t)Yk eiH
| {z }

(67)

We see that

hYk (t)Yj (t + )i

the evolution equation for

in (66) equal

has the same

hYj (t)i

in (67). Although

evolution in

as

hYi (t)i

in

but with dierent initial conditions. Since

was assumed to hold for all initial states, we can conclude the proof by letting

is not a density operator, due to the condition above, the theorem holds.

Note, that for stating (64), it does not matter if the expectation value at time

t (a measurable quantity) was derived

in the S-, I- or H-Picture. In constract, for an expression as in (65), it is essential to dene the operators in the
H-Picture.

Computation of the Mollow-Spectrum:

uorescence state vector

v,

When we apply the Quantum regression theorem for resonance

we have to remember, that the dynamics of

did not derive from a unitary time

development but from the M-Equation (with no special initial state specied). We can still apply the theorem since
the M-Equation dynamics can be viewed as approximation of a complicated and unknown unitary dynamics using
the Hamiltonian of system, bath and interaction.

48

Consider the dynamics of the atomic DOF means, which was solved in a previous section

h
+ (t)i h
+ iss

d
(v(t) vss ) = A (v(t) vss )
dt

v(t) vss = h
(t)i h
iss
h
z (t)i h
z iss
The

h
+ iss

are dealt with as time independent c-numbers and are moved out of the expectation values. Then use the

, but now for operators in the H-Picture,

QRT for various correlation functions

values of the form

hA(t)B(t )i

because otherwise the expectation

are ill dened. (We take the detuning as zero):

h
+ (t)
+ (t + )i h
+ (t)i h
+ iss

h
+
+ ( )iss h
+ iss h
+ iss

g ( ) = h
+ (t)
(t + )i h
+ (t)i h
iss gss ( ) gt ( ) = h
+
( )iss h
+ iss h
iss
h
+ (t)
z (t + )i h
+ (t)i h
z iss
h
+
z ( )iss h
+ iss h
z iss
The parameter

is free in the second equation of the QRT, so it has been taken to innity.

d
gss ( ) = Agss ( )
d
with initial conditions taken from the notion that

2
+
= 0, + = (1 + z ) /2, + z = + :

h
+
+ iss h
+ iss h
+ iss

gss ( = 0)

= h
+
iss h
+ iss h
iss =
h
+
z iss h
+ iss h
z iss

2 2

gss ( = 0)

4
( 2 +82 )2
2
4 2 2
1
2 2( 2 +82 ) ( 2 +82 )2
2i 3
22i
+82 + ( 2 +82 )2

lim h
+ (t)
(t + )iss =

X =

h
+ iss h
+ iss
1
2

+ h
z iss /2 h
+ iss h
iss =
h
+ iss h
+ iss h
z iss

The solution for the quantity of interest (the middle entry of

with

gss ( ))

is, after adding

h
+ iss h
iss ,



42
2
e /2
X+ (3/4) X (3/4+)
+

e
+
e
2 + 82 2 + 82
2
2
2

34

4
2
2 +82 4 2 8 . Now we remember

+ (t) = + (t) eiL t and

(t + ) = (t + ) e+iL (t+ )

and nd

G ( )

D
E
lim h+ (t) (t + )i = lim
+ (t) eiL t
(t + ) eiL (t+ )

t
iL

=
ss

lim h
+ (t)
(t + )iss e

That means we have to attach an additional

eiL

to the quantity found in (68)!

We compute the spectrum as follows

I
(r)
I0 (r)
0
d ei G( ) + d ei G( )
d ei G( ) +
d ei G( ) =
S(r, ) =
2
2
0

(68)

49

With the help of

G( )

= h+ (t) (t )iss = h+ (t0 + ) (t0 )iss

X
= T r {+ (t0 + ) (t0 ) 0 } =
hi| + (t0 + ) (t0 ) 0 |ii
i

E X
XD

=
0 (t0 ) +
(t0 + ) i|i =
hi| 0 + (t0 ) (t0 + ) |ii
i

=
we nd
height

T r (0 + (t ) (t + )) = G( )

I0 (r)
i
G( ). We need predominantly the Lorentzian curve in the spectrum, having a
Re d e
0
and a full width at half maximum of 2:

S(r, ) =

1/

Re d ei(0 ) =
0

( 0 ) + 2

L0 , ()

Weak driving regime: ||  /8, The atom absorbs the laser photon and directly emits it into the resonant
mode

S(r, ) ' I0 (r)

4 2
( 0 )
2 + 82

S(r,)

Strong driving regime: ||  /8,

we obtain the

many times with the laser eld before emitting it,

Mollow -Triplett,

where the atom exchanges the photon

is the strength of the Atom-Laser interaction, which gives

rise to sidebands.

I0 (r)
S(r, ) =
2


2

4 2
1
1
( 0 ) + L0 ,/2 () L0 2,3/4 ()
2
2
+ 8
2
4

S(r,)

02

0+2

Antibunching properties of scattered light:

The photon statistics, as described in section 2, is linked to the

quantity

g (2) ( ) =

G(2) ( )
|G(1) (0)|2

50

We calculate

G(2) ( ) = hE (r, t)E (r, t + )E+ (r, t + )E+ (r, t)i

using (62) and making use of the QRT, which

applies here as well. We nd



3
g (2) ( ) = 1 e3 /4 cosh [ ] +
sinh [ ]
4

g(2)()
1

strong driving

antibunching

classical
light,
uncorrelated
photons

weak driving

In all driving regimes, the plot indicates non-classical antibunching. That is clear when one remebers, that the atom
has to make a transition into the ground state to emit a photon, than some waiting time (for excitation) is needed
before the next photon can be emitted. The transition from actual oscillations to the monotonously increasing shape
is close, but not at

= 0 ( = 8). The oscillations in the strong driving regime are reminescent of Rabi oscillations,

where the atom was in the exicted state for sure after

= /

51

and a second emission is therefore more likely.

Raman Transitions and Electromagnetically Induced Transparency

(EIT)

Consider an electronic system which is more complex than a two level atom and consists of at least three levels as
shown in the Figure:

L2
L1

We are interested in the coupling of light to the transition between states

|ji, |j 0 i.

In the long wavelength limit

(see above), the coupling strength is proportional to the matrix element, which might vanish if not consistent with
selection rules

D
E
0 ek
gk,j,j 0 j|d|j
where

D
E
0 = dk,j,j 0
j|d|j

is only nonzero if lj 0

= lj 1

and

mj 0 {mj , mj 1}.

In our three-level system above we

make use of this fact and chose the atomic levels such that no direct transition between level

|1i

and

|2i

is possible

(in fact it might be possible in higher orders beyond long wavelength approximation but highly surpressed). This is
the fact in
in state

87Rb

|2i

for example, where we have for the spontanous emission rates

0 < 21 < 104 31 .

are long lived and can be used to store information. Next we use

} = 1

and

Thus excitations

jl = |ji hl|,

then the

Hamiltonian (transformed as in the resonance uorescence section with two classical driving elds) reads with
setting

to zero:

H = 3 33 + 2 22 + (1 (t)31 + 2 (t)32 + h.c.)

where

1 (t) = 1 eiL1 t .

We use (18) and

We transform in an interaction picture with respect to

H0 = L1 33 + (L1 L2 ) 22 .

[33 , 31 ] = |3i h3| |3i h1| |3i h1| |3i h3| = 31 :

HI

= eiH0 t {3 33 + 2 22 + (1 (t)31 + 2 (t)32 + h.c.) L1 33 (L1 L2 ) 22 } eiH0 t

HI

= eiH0 t {(1 (t)31 + 2 (t)32 + h.c.)} eiH0 t + 1 33 + {2 (L1 L2 )}22

{z
}
|

HI

1 eiL1 t eiL1 t 31 + 2 eiL2 t eiL2 t 32 + h.c. + 1 33 + 22

In the case of two photon resonance,

1 = 2 =

= 0,

(69)

the two dashed lines in the gure coincide and we nd eventually with

a time independent Hamiltonian:

HI = (1 31 + 2 32 + h.c.) + 33

6.1 Raman-Transitions [Marzlin]

If we go to another interaction picture with respect to

33

we nd from (70)

HI (t) = 1 31 eit + 2 32 eit + h.c.

Formally, the equation of motion in the interaction picture reads

d
|(t)i = iHI (t) |(t)i
dt
52

(70)

which can be translated in

c1,2 |1,2 |.

The last two lines indicate

ic3

1 eit c1 + 2 eit c2

ic2

2 eit c3

ic1

1 eit c3

The formal solution for the rst line is with

c3 (t = 0) = 0

t
n
o
0
0
c3 (t) = i dt0 1 eit c1 (t0 ) + 2 eit c2 (t0 )
0
Now we consider the case of large detunings and weak driving:

||  |1,2 |
Then we have

c1,2 |1,2 |  || and the exponentials under the integral vary much faster than the states. Integrate

only over the rapidly variing terms:


c3 (t) '






2 it
1 it
e
1 c1 (t) +
e
1 c2 (t)

This means the excited state is always weakly occupied. Use

1,2 eit

ic1,2 (t) =

c3 (t)

in the evolution equations for

c1,2 (t):






1 it
2 it
e
1 c1 (t) +
e
1 c2 (t)

which is, for the rst component:

ic1 (t) =
In the time evolution of

c1 ,

1
2
1 eit c1 (t) +
1 eit c2 (t)

the quickly variing terms play no role, so we nd

ic1 (t) =
The same holds for the

c3

c2

2
1
c1 (t) +
c2 (t)

state. We see that for the dynamics of the two lower states

is irrelevant (adiabatic elimiation of excited state). The eective Rabi-Frequency

the coupling between the two (not directly coupled) lower states and induces the
lation of the excited state,

|1 + 2 | /2 .

So, if

|c3 (t)|

is of the order

|1 + 2 | /2

and

ef f =

c2 ,

the excited state

1 2 / determines

Raman-Transitions.

The popu-

and the spontanous emission losses scale like

we have the relation

1 2

2
 2 |1 + 2 |

between the wanted transition and the unwanted emission process rates.

6.2 EIT
We again consider the Hamiltonian as in (70) in the two-photon resonance case.

53

c1

strong control

L2

weak probe

L1

H = (1 31 + 2 32 + h.c.) + 33
We note that

has an Eigenstate with

H |0 i = 0
|0 i =

We have

h3|0 i = 0,

|3i

level in this so called

=0

prevents emission of light. If we start in the state

the atom would stay in

|0 i and we can adiabatically transfer


1 2
2
2

z E = 0 t P
c2 t

is the polarisation. It is dened as the dipole moment per volume and can be calculated in our situation as

1X
he rj i = natoms d13 31 eiL1 t + d23 32 eiL2 t + c.c
V j

where we decided to use the density matrix elements

Take the

E1

eld which connects the

and

31

and

32

(72)

in the Interaction picture (all without proof ).

states as the probe, then

E1 = E1 eiL1 t + c.c.

1 =

d31
E1
}

We are interested in its propagation in dependence of the second control eld

E2

that the eld amplitudes are constant and that we have spontanous emission from

2 =

with

3 1, 2

in there for future calculations, but for the moment assume

21
for the steady state case ( 31

= |1i h1| +
order for

from (69) (we leave

1
(213 31 33 33 ) +
2
2

(223 32 33 33 ) + (212 21 22 22 )
2
2

31

(1)

and from

= 0):

We are especially interested in the coherences which appear in (72) (31

to

1,2

= i [HI , ] +
+

d32
} E2 . We assume

with rate

2 with the much smaller rate . Thus have to employ a master equation approach with HI

We set

|0 i

Consider Maxwells equation in 1d in a medium without currents

P (t) =

state

the destructive interference

|1i |2i.

where

dark-state

1 /2

and slowly change the ratio of the laser powers

Optical response:

2 |1i 1 |2i
p
21 + 22

thus the upper level is never excited, even for resonance

from both lasers in the

population between

(71)

= |3i h1| = 31 ),

to this reason we solve


1 
= i +
31 i2 21 i1 (11 33 )
2


= i +
31 + i2 31 + i1 23
2

= 0 = 21 ).

We assume strong control and weak probe,

for a small parameter

(73)

|2 |  |1 |,

thus

|0 i ' |1 i.

and expand the density matrix for the perturbed dark state into

(Thus we assume the system is in the state

:
31 =

= |0 i h0 |

d31
2 (2 i) E1
} (2 i) (2 i) 822
54

in rst order!). We solve (73) in linear

with

= 1 + 2 .

We nd

P (t)

and by comparison to

=
For the case of two-photon-resonance,

P = 0 E , E = E1 + E2

natoms |d13 |2
2 (2 i)
}0
(2 i) (2 i) 822

=0

and

very small, we have

does not feel the medium, we have transparency induced by

E1

we get

E2

= P = 0,

even on resonance

that means the probe eld

= 0,

where an incident eld

would be absorbed.

Slow light:

In the setup described above, we have to leading order of

=
The refractive index is

n=

1+

with

= 0:

natoms |d13 |2
}0
222

and the group velocity for the probe eld is given by




c
natoms |d13 |2
vg =
= ... =
' c/ 1 +
k
222 }0
n L1 n

For

natoms |d13 |2  222 }0

very

slow light, e.g. vg  c is achievable, see Hau et. al., Nature, 397, 594, (1999)

55

E1