Brian Julsgaard

Entanglement and Quantum
Interactions with
Macroscopic Gas Samples
Brian Julsgaard
Danish National Research Foundation

Center for Quantum Optics - QUANTOP
Department of Physics and Astronomy
University of Aarhus, Denmark
PhD thesis
October 2003
ii
This thesis is submitted to the Faculty of Science at the University of

Aarhus, Denmark, in order to fulfil the requirements for obtaining the
PhD degree in Physics. The studies have been carried out under the
supervision of Prof. Eugene Polzik in the Quantum Optics Laboratory
at the Department of Physics and Astronomy, Aarhus University from
August 1999 to January 2003, and at the Quantum Optics Laboratory,
Niels Bohr Institute, Copenhagen University from February 2003 to Oc-
tober 2003.
iii
Contents
List of Publications vii
Preface ix
1 Outline of the Thesis 1
2 Atoms and Light 3

2.1 The Atomic Ground State Spin . . . . . . . . . . . . . . . . . . . . 3
2.2 Polarization States of Light . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Atom/Light Interactions . . . . . . . . . . . . . . . . . . . . . . . . 8
3 Motivation 13
3.1 Why Quantum State Engineering . . . . . . . . . . . . . . . . . . . 13
3.2 Technological Implementations . . . . . . . . . . . . . . . . . . . . 14
3.3 Continuous Versus Discrete Systems . . . . . . . . . . . . . . . . . 16
4 Experimental Methods 19
4.1 Laser Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.2 Atomic Vapour Cells . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.3 The Rotating Frame and Magnetic Fields . . . . . . . . . . . . . . 24
4.4 Spin Life Times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.5 Optical Pumping . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5 The Effective Interaction Hamiltonian 33

5.1 Electric Dipole Interactions . . . . . . . . . . . . . . . . . . . . . . 34
5.2 The Off-resonant Limit . . . . . . . . . . . . . . . . . . . . . . . . . 35
6 Propagation Equations 41
6.1 General Propagation Equations . . . . . . . . . . . . . . . . . . . . 41
6.2 Probing a Macroscopic Ensemble of Oriented Spins . . . . . . . . . 42
6.3 Probing Transverse Spin Components . . . . . . . . . . . . . . . . 43
6.4 Inclusion of Higher Order Terms . . . . . . . . . . . . . . . . . . . 45
iv CONTENTS
7 Atomic State Characterization 49

7.1 The Magneto-Optical Resonance Method . . . . . . . . . . . . . . 50
7.2 Spin State Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . 54
7.3 Experimental Test of the Modeling . . . . . . . . . . . . . . . . . . 55
7.4 Unresolved Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
7.5 Pulsed Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . 60
7.6 Stark Shifts by the Probe . . . . . . . . . . . . . . . . . . . . . . . 62
8 Recording Quantum Fluctuations 67

8.1 Theoretical Approach . . . . . . . . . . . . . . . . . . . . . . . . . 68
8.2 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . 72
8.3 Experimental Investigation of the Model . . . . . . . . . . . . . . . 73
8.4 Broadband Atomic Noise . . . . . . . . . . . . . . . . . . . . . . . 75
8.5 Discussion of the Results . . . . . . . . . . . . . . . . . . . . . . . . 77
9 Entanglement, Theoretical Approach 79

9.1 Definition of Entanglement . . . . . . . . . . . . . . . . . . . . . . 80
9.2 Entanglement Generation . . . . . . . . . . . . . . . . . . . . . . . 81
9.3 Wave Function Modeling . . . . . . . . . . . . . . . . . . . . . . . . 84
9.4 Rotating Frame and Entanglement . . . . . . . . . . . . . . . . . . 87
9.5 Entanglement Estimation . . . . . . . . . . . . . . . . . . . . . . . 89
9.6 Entanglement Generation and Losses . . . . . . . . . . . . . . . . . 91
9.7 The Atomic Projection Noise Level . . . . . . . . . . . . . . . . . . 93
10 Experimental Generation of Entangled States 95

10.1 Entanglement Demonstration . . . . . . . . . . . . . . . . . . . . . 95
10.2 Changing the Experimental Setup . . . . . . . . . . . . . . . . . . 100
10.3 Magnetic Field Noise . . . . . . . . . . . . . . . . . . . . . . . . . . 101
10.4 Weighted Entanglement Estimation . . . . . . . . . . . . . . . . . . 104
10.5 Discussion of the Results . . . . . . . . . . . . . . . . . . . . . . . . 107
11 Quantum Information Protocols 109

11.1 Teleportation of an Unknown Spin State . . . . . . . . . . . . . . . 110
11.2 Entanglement Swapping . . . . . . . . . . . . . . . . . . . . . . . . 112
11.3 Quantum Memory for Light . . . . . . . . . . . . . . . . . . . . . . 115
12 Summary and Outlook 117
A Polarization States of the Light Field 119

A.1 The Quantized Radiation Field . . . . . . . . . . . . . . . . . . . . 119
A.2 Stokes Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
A.3 Strong, Linearly Polarized Light . . . . . . . . . . . . . . . . . . . 121
A.4 Stokes Operators and Losses . . . . . . . . . . . . . . . . . . . . . . 121
CONTENTS v
B Spins and Density Operators 123

B.1 Quantization Along the z-axis . . . . . . . . . . . . . . . . . . . . . 123
B.2 Quantization Along the x-axis . . . . . . . . . . . . . . . . . . . . . 124
B.3 Commutators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
C Continuous Description of Light and Matter 127

C.1 Continuous Description of the Electromagnetic Field . . . . . . . . 127
C.2 Spatial Description of the Electric field . . . . . . . . . . . . . . . . 128
C.3 Continuous Matter Operators . . . . . . . . . . . . . . . . . . . . . 130
D Dipole Matrix Elements 133

D.1 Calculating Matrix Elements . . . . . . . . . . . . . . . . . . . . . 133
D.2 Dipole Coupling Constants . . . . . . . . . . . . . . . . . . . . . . 135
E Photo Detection Theory 139

E.1 Stokes Vector Detection . . . . . . . . . . . . . . . . . . . . . . . . 139
E.2 Our Experimental Case and Detection . . . . . . . . . . . . . . . . 141
F The Quadratic Zeeman Effect 145
G Spin Decay and Langevin Forces 147
Bibliography 149
vi CONTENTS
vii
List of Publications
[I] B. Julsgaard, A. Kozhekin, and E. S. Polzik, Experimental long-lived en-

tanglement of two macroscopic objects, Nature 413, 400 (2001).
[II] C. Schori, B. Julsgaard, J. L. Sørensen, and E. S. Polzik, Recording Quan-

tum Properties of Light in a Long-Lived Atomic Spin State: Towards Quan-
tum Memory, Phys. Rev. Lett. 89, 057903 (2002).
[III] J. L. Sørensen, B. Julsgaard, C. Schori, J. Hald, and E. S. Polzik, Quantum

noise limited laser probing of atomic spin states, Laser Physics 13, 359
(2003).
[IV] E. S. Polzik, B. Julsgaard, J. Sherson, and J. L. Sørensen, Entanglement and

quantum teleportation with multi-atom ensembles, Phil. Trans. Roy. Soc. A
361, 1391 (2003).
[V] J. L. Sørensen, B. Julsgaard, C. Schori, and E. S. Polzik, Quantum limits

encountered in atomic spin measurements, Spectrochim. Acta B 58, 999
(2003).
[VI] E. S. Polzik, B. Julsgaard, C. Schori, and J. L. Sørensen, in The Expand-

ing Frontier of Atomic Physics, Invited talks of the XVIII International
Conference on Atomic Physics, eds. H. R. Sadeghpour, E. J. Heller, and
D. Pritchard, World Scientific (2003).
[VII] B. Julsgaard, C. Schori, J. L. Sørensen, and E. S. Polzik, Atomic Spins as a

Storage Medium for Quantum Fluctuations of light, Quantum Information
and Computation 3, Special issue, 518 (2003).
[VIII] B. Julsgaard, J. Sherson, J. L. Sørensen, and E. S. Polzik, Characterizing
the spin state of an atomic ensemble using the magneto-optical resonance
method, J. Opt. B: Quantum Semiclass. Opt. 6, 5 (2004).
viii List of Publications
ix
Preface
The work presented in this thesis is the result of a little more than four years
of experimental work under supervision by Prof. Eugene Polzik. During these
years I have experienced many different aspects of research which I have found
very interesting. In addition to the physical insight gained by our research I
learned laser physics, electronics, craftsmanship in the workshop, how to move
entire laboratories to new locations, etc. I also had several opportunities to give
oral presentations of our work at various international conferences.
In August 1999 I joined the Quantum Optics Laboratory at the Department
of Physics and Astronomy at the University of Aarhus. At that time I had the
pleasure to work with Jan Hald, Jens Lykke Sørensen, and Christian Schori who
essentially taught me to adjust mirrors and tame crazy lasers. We co-operated
on very interesting scientific work and we also relaxed together during numerous
coffee breaks. During my time in Aarhus I also had the opportunity to work
with Peter Møller Nielsen, Anton Verchovski, Wolfgang Tittel, Jens Kristian
Mikkelsen, Daniel Oblak, and others.
Time went on, and Jacob Sherson joined my experiment in Aarhus. Since then
we have had an excellent co-operation and obtained many interesting results. He
also joined me in the relocation from the University of Aarhus to the Niels Bohr
Institute at Copenhagen University. During that time I also had the pleasure to
work with Carlos Alzar, Plamen Petrov, Jörg Helge Müller, and others.
From the theoretical side I worked with Alexander Kozhekin who came up
with the idea that two oppositely oriented spins are interesting to study. I am also
very grateful to Anders Sørensen and Klaus Mølmer who always took the time
to give theoretical advice to an experimentalist. I also acknowledge interesting
and useful discussions with Dmitri Kupriyanov. The technical and administrative
staff at the University of Aarhus has always been very helpful, with special thanks
to the workshop, to the electronics department, and to Grete Flarup.
I am very grateful to have experienced all this time together with my super-
visor Eugene Polzik. We have had many enthusiastic discussions and Eugene has
always been a safe guide in the entire process starting from an empty laser table
to the final presentation of our research activities. Without Eugene’s enthusiasm
x Preface
and physical intuition we would never have reached our fruitful results.
Finally, I would like to thank my girlfriend Merete for being understanding
during stressful times or late working hours. In particular I am grateful that she
joined me in the relocation from Aarhus to Copenhagen.
October 2003, Brian Julsgaard.

1
CHAPTER 1
Outline of the Thesis
The subject of this thesis is mainly the generation of entangled states between
macroscopic objects and the quantum mechanical understanding of the interac-
tion between polarized laser light and atomic ground state spins. The work is
primarily experimental but we also present theoretical calculations.
Different readers should read this thesis differently. Some chapters are meant
as pedagogical introductions to the field, other chapters are very technical and
may in some occasions be even clumsy! A general and hopefully understandable
description of the work behind this thesis is of course the most important point.
But we acknowledge the fact that younger students or co-workers following up on
the work may find it useful to read about the detailed calculations and general
considerations which have required a lot of effort for the author of this thesis.
Chaps. 2 and 3 give the general introduction to the field. In Chap. 2 the
physical system of polarized laser light interacting with atomic ground state spins
is introduced. Chap. 3 is a broader review of the field of quantum optics and
puts our work into a general context. These chapters should be relevant for all
readers. Chap. 4 introduces many general aspects and techniques related to the
experiments. This chapter is of course mainly for experimentalists but it also
gives a general characterization of our physical system which is relevant for all
readers. Chap. 5 derives an effective Hamiltonian describing the light/matter
interactions and Chap. 6 states the equations of motion following this Hamil-
tonian. These two chapters are purely theoretical, the hard core experimentalist
may read the main results in Secs. 6.2 and 6.3. In Chap. 7 we start to con-
sider experimental results and techniques in detail. This chapter is devoted to
the magneto-optical resonance method for characterizing atomic ground state
2 Chapter 1 - Outline of the Thesis
spins. These purely classical results are very useful and extensively used for
all other experiments of this thesis. Chap. 8 describes theoretical and experi-
mental investigation of the interaction between light and atoms at the quantum
level. The results demonstrate that our theoretical understanding is correct to
a high degree and that our physical systems are feasible for implementation of
quantum information protocols. The purely theoretical Chap. 9 and the purely
experimental Chap. 10 deal with the generation of entangled states between
macroscopic cesium gas samples. The results presented in Sec. 10.1 demonstrate
on demand creation of entangled states, this is the most important result of this
thesis. Chaps. 7, 8, and 10 present experimental results of increasing importance
and are relevant for most readers. Chap. 9 is not only for the theoretically in-
terested reader, it also covers many practical considerations necessary for the
understanding of the entanglement generation. Chap. 11 gives an impression of
possible future technological implementations of our research for quantum com-
munication purposes. In Chap. 12 we summarize the thesis work and give an
outlook for the future.
The appendices are in general technical. Apps. A and B deal with the
description of polarized light and atomic spins, respectively. App. C is more or
less “text book material” on the quantum mechanical description of light/matter
interactions. The emphasis is put on continuous description of light and matter.
The reason for writing this down is the fact that the author (who is an experi-
mentalist) found it difficult to find relevant references to text books and spent a
huge amount of time on what may be really trivial for a theorist. It is the hope
that other readers find this appendix useful. App. D singles out results of spher-
ical tensor algebra directly applicable for the derivations in Chap. 5 and is only
relevant for the theoretically interested reader. App. E reviews a few results of
photo detection theory and connects these to our experiments (which deal with
polarization states of light). The main problem here is time ordering of light
creation and annihilation operators and turns out to be inelegant and clumsy.
This appendix is reserved for readers who really want to understand every detail
of the theoretical derivations in Chap. 8. App. F reviews the quadratic Zeeman
effect and is included in this thesis for completeness, it is very useful for the un-
derstanding of magneto-optical resonance spectra presented in Chap. 7. App. G
presents a few technical aspects of Langevin forces required for the theoretical
modeling of decaying spins in Chap. 8.
3
CHAPTER 2
Atoms and Light as Interesting

Quantum Systems
In this section we aim at explaining some general properties of atoms and light
relevant for our experiments. Both atoms and light can be described in the lan-
guage of quantum mechanics, but from our perspective this becomes interesting
when we can couple the atomic and light degrees of freedom to each other in
order to exchange quantum information between the two. Atoms are massive
particles and therefore slow in the sense that we experimentally can keep them
at a well controlled location for a long time. This means atoms are good for
storage of quantum information. On the contrary, light is fast and is well suited
for transporting quantum information between atomic systems or to various de-
tection systems. In the following sections we describe in general terms how atoms
and light experimentally can play these roles and we comment on the nature of
the interaction between light and atoms.
2.1 The Atomic Ground State Spin

The experiments in this thesis are carried out using cesium atoms. One reason
for this is the fact that lasers are available for relevant transitions, another is the
fact that the definition of one second relies on the cesium atom. This atom was
the preferred one when I joined the Quantum Optics Laboratory four years ago
and motivating this choice is the availability of a tunable source of squeezed or
entangled light at atomic transitions in cesium (we use this source in Chap. 8).
4 Chapter 2 - Atoms and Light
(a) (c) (d)
mF
mF
mF
mF
Jˆz P̂A
=
=
=
=
4
2
3
1
F=4
Jˆy
F=3 X̂A
Macroscopic spin
Many single spins
(b) (e)
(atoms in F=4)
Interesting Jx
P̂A
F=4
F=3
X̂A
Irrelevant
Figure 2.1: (a) The 6S1/2 ground state of cesium is split into two hyperfine states
with F = 3 and F = 4. If atoms are left alone the spin state will populate all
mF -sublevels evenly as depicted by small circles. (b) In the experimental work
in this thesis we are only interested in the atoms in the F = 4 state and it is
possible to put (almost) all atoms into the outermost state with mF = 4 (with
the x-axis as direction of quantization). (c) The individual spins sum up to a
macroscopic spin along the x-axis, the transverse spin components Jˆy and Jˆz
have quantum uncertainties depicted by a gray disk, see Eq. (2.3). (d) The spin
is often shown from above since only the transverse components have interesting
quantum properties. The new variables X̂A and P̂A are defined in Eq. (2.4). It
is in principle possible to create non-classical states, e.g. a squeezed stated as
shown in (e).
The ground states of cesium are characterized by the outermost electron which
is in the 6S1/2 state, i.e. the orbital angular momentum L is zero. The electron
spin S and thus the total electronic angular momentum J has quantum number
S = J = 1/2. The nuclear spin I of cesium-133 has I = 7/2, and the coupling
between the nucleus and the electron gives rise to the total angular momentum
F = I + J with quantum numbers F = 3 and F = 4, see Fig. 2.1(a).
It is indeed the total angular momentum F which interests us in this thesis
since F and the magnetic quantum numbers mF define the energy levels of the
ground states. Furthermore, we will often restrict ourselves to one hyperfine
level, F = 4, which is possible experimentally since the hyperfine splitting ν hfs =
9.1926GHz is large compared to typical resolutions of our laser systems. And
now a bit of confusion, we choose to denote the total angular momentum of a
single atom by j and for a collection of atoms (in the F = 4 state) we denote the
collective total angular momentum by J, i.e.
N
X
J= j(i) , (2.1)
i=1
where N is the number of atoms in the F = 4 state and j(i) is the total angular
2.1. The Atomic Ground State Spin 5
momentum of the i’th atom. The reason for using J and not F is that J is
generally more used in the literature for spins, and indeed, we wish to think
about our spins more abstractly than just the properties of some atoms. Many
results in this thesis should be applicable in a broader sense than to a collection
of cesium atoms.
In our experiments the number of atoms N will be of the order 1012 and we
will almost always aim at having all atoms polarized along one direction which
we denote as the x-axis, see Fig. 2.1(b). With the x-axis as quantization axis we
thus have mF = 4 for all atoms to a high degree of accuracy, and the collective
spin Jˆx will really be a macroscopic entity. With this experimental choice, we
may treat the x-component of the collective spin as a classical c-number, i.e. we
replace the operator Jˆx by the number Jx . The transverse spin components Jˆy
and Jˆz maintain their quantum nature. They will typically have zero or a small
mean value. The quantum fluctuations are governed by the commutation relation
and the Heisenberg uncertainty relation (with h̄ = 1)
h i
Jˆy , Jˆz = iJx (2.2)
J2
⇒ Var(Jˆy ) · Var(Jˆz ) ≥ x . (2.3)
4
Pictorially, we add many small spins to a macroscopic spin, see Fig. 2.1(c), and
the direction of this macroscopic spin has no precise meaning but can only be
defined within the quantum uncertainty (depicted by the gray disk) stated quan-
titatively by Eq. (2.3). With 1012 atoms the angular quantum uncertainty of the
collective spin direction is of the order 10−6 .
To connect our spin system to quantum mechanics more generally, we note
that the classical Jx enable us to define new quantum variables X̂A and P̂A by
Jˆy Jˆz h i
X̂A = √ , P̂A = √ , ⇒ X̂A , P̂A = i, (2.4)
Jx Jx
where the subscript A refers to “atoms”. The above is exactly on the form seen
in many text books on introductory quantum mechanics. Even though we know
that e.g. Jˆy has a discrete spectrum of eigenvalues, which must be inherited by
X̂A , we effectively have continuous quantum operators as ordinary position x̂ and
momentum p̂ when N is large. For this reason we can depict the transverse spin
variables Jˆy and Jˆz as seen from above as in Fig. 2.1(d,e). The disk or ellipse
symbolizes the Heisenberg uncertainty relation as is often seen in the literature
on e.g. description of the electromagnetic field. As we shall see in Chap. 11,
protocols for quantum information processing can be very well described in the
x, p-language with no specific reference to the cesium atom.
We finalize this section with a very important definition. If all atoms in the
F = 4 state have mF = 4 (or mF = −4) the collective spin is said to be in the
coherent spin state 1 (CSS). In this state all atoms are independent of each other
1 Coherent spin states are discussed more generally in [1]
(a) x z (b) (c)

Ŝz P̂L
y
x-pol y-pol Ŝy

X̂L
Macroscopic Stokes vector

o
45
Sx
Many photons
(x-polarized)
+45◦ -pol −45◦ -pol (d)
P̂L
X̂L
σ+ -pol σ− -pol
Figure 2.2: (a) Some different basis choices for the polarization of a photon. (b)
For many x-polarized photons the Stokes vector has practically a classical S x -
component. The transverse components Ŝy and Ŝz must satisfy the Heisenberg
uncertainty relation (2.8) symbolized by the gray disk. (c) We may restrict
ourselves to a top view of (b) since only the transverse components X̂L and P̂L
play a role concerning the quantum state. (d) Non-classical states of the Stokes
vector are possible, here squeezing of one quadrature.
in the sense that the total wave function is the product function of each atom in
mF = 4. This state also fulfils the minimum uncertainty relation
J2
Var(Jˆy ) · Var(Jˆz ) = x (coherent spin state) (2.5)
4
and we will see in Chaps. 9 and 10 that this equality serves as a reference point
for manifestly quantum states, i.e. states that have no classical analogue.
2.2 Polarization States of Light

All experiments in this thesis involve laser light interacting with atomic spin
states, and it turns out that the polarization of the light is the relevant quantum
variable to describe. Below we introduce the description of polarization states of
light in general terms, for a more rigorous definition see App. A.
Now, consider a pulse of light, or a collection of photons, propagating in the
z-direction. The polarization state is well described by the Stokes operators
1
Ŝx = (n̂ph (x) − n̂ph (y)) ,
2
1
Ŝy = (n̂ph (+45◦ ) − n̂ph (−45◦ )) , (2.6)
2
1
Ŝz = (n̂ph (σ+ ) − n̂ph (σ− )) ,
2
2.2. Polarization States of Light 7
where n̂ph (x) is the number of photons in the pulse with x-polarization, and so
on (the different ways of describing polarization are depicted in Fig. 2.2(a)). The
Stokes operators are dimensionless, they count photons. In other chapters we
will also consider Stokes operators measuring photon fluxes.
We shall make our first approximation right away. We assume that almost all
photons are linearly polarized along the x-direction (we could have chosen any
direction). For a pulse containing many photons this means that we can treat
Ŝx → Sx as a c-number. Note, this is very similar to the approximation of a well
polarized sample of spins in the previous section.
It can be shown that the Stokes vector satisfies angular commutation relations
(see Eq. A.13), i.e.
h i
Ŝy , Ŝz = iSx (2.7)
Sx2
⇒ Var(Ŝy ) · Var(Ŝz ) ≥ (pulse of light). (2.8)
4
The operators Ŝy and Ŝz are our interesting quantum variables, they usually have
zero mean value since a collection of x-polarized photons have polarization ±45 ◦
or σ± with equal probability being one half. But we remark, that if the linear
x-polarized light is rotated by an angle θ around the z-axis the balance between
the ±45◦ -components is changed and Ŝy gets a non-zero mean value. In fact
D E
Ŝy = 2Sx · θ (for θ ¿ 1). (2.9)
We see that Ŝy is a measure of the polarization rotation, and that the quantum
fluctuations of Ŝy can be interpreted as quantum fluctuations in the direction
of polarization. This kind of rotation will prove to be very important when we
consider the interaction of polarized light with atomic spins.
The classical Sx component and the quantum uncertainty of the Ŝy and Ŝz can
be depicted as in Fig. 2.2(b) (just like the case for atomic spins in Fig. 2.1(c)), the
gray disk symbolizes the Heisenberg uncertainty relation (2.8). We may also see
this disk from above as in Fig. 2.2(c,d), where we have defined the new quantum
variables (subscript L refers to “light”)
Ŝy Ŝz h i
X̂L = √ , P̂L = √ , ⇒ X̂L , P̂L = i. (2.10)
Sx Sx
We see, as in the case of atomic spins, the polarization quantum state of light
is similar to the standard position/momentum operators. The mathematical
equivalence of the spin and light operators motivate the search for possible im-
plementations of quantum information protocols that exchange quantum states
between light and atoms.
If we go into a little detail with the operators X̂L and P̂L it actually follows
from Eq. (A.14) that
ây + â†y ây − â†y
X̂L = √ and P̂L = √ , (2.11)
2 i 2
where â†y and ây are the creation and annihilation operators for photons with
y-polarization. In our approximation with strong linear polarization along the
x-axis the x-polarized part serves as a “reservoir” of photons controlling the
strength of the pulse. It also serves as a phase reference for the y-polarized part
which carries the interesting quantum fluctuations.
As a final remark of this section we consider an x-polarized pulse of light
emerging from a laser. Then the y-polarized mode is in the vacuum state and we
easily find
Sx2
Var(Ŝy ) · Var(Ŝz ) = (light pulse, coherent state), (2.12)
4
which is actually valid for the y-polarized mode being in any coherent state. As
for atomic spins, this equality sets the benchmark point for non-classical states
of the light polarization. If Var(Ŝy ) = Var(Ŝz ) = Sx /2 we say that the noise of
Ŝy or Ŝz is at the so-called shot noise level.
2.3 Interaction Between Atomic Spins and

Polarized Light
Let us now introduce the interaction of polarized light with the atomic spin state
of a sample of atoms. A detailed calculation will be given in Chaps. 5 and 6.
First, consider Fig. 2.3(a) which shows the level scheme of the 6S1/2 and 6P3/2
states of cesium. We tune a laser, which we call the probe laser, to the dipole
transition between these two levels and we may choose a detuning ∆ measured
from the F = 4 → F 0 = 5 transition with red detuning being positive.
The polarization state of this laser light may change in different ways if it
interacts with an atomic sample. First of all, the absorption of one polarization
component may be different than the absorption of another polarization compo-
nent. This clearly changes the polarization state, but we will not consider this
situation at all in this thesis. The absorption profile of the transition will look
like the solid line graph in Fig. 2.3(b), in the wings it will fall off as 1/∆2 . Going
to a sufficiently far detuning we can make absorption effects negligible compared
to dispersion effects, the latter fall off as 1/∆. A typical dispersion profile (for
a single F → F 0 transition) is shown with the dashed line graph in Fig. 2.3(b).
Dispersion effects will change the polarization state of light if the index of refrac-
tion is different for two orthogonal polarization components, i.e. if the sample is
birefringent.
Fig. 2.3(c) shows a pulse of light propagating in the z-direction through an
atomic sample which is polarized along the x-direction. Classically, it is clear from
simple symmetry reasoning that we may have nx 6= ny (the sample has linear
birefringence), while e.g. n+45◦ = n−45◦ and nσ+ = nσ− . Thus the x-direction is
an optical axis and x- or y-polarized light will pass on without change, i.e. Ŝx will
be unchanged by the interaction. If we let light with linear polarization along
2.3. Atom/Light Interactions 9
(a) (c) (d)

−∆ nx 6= ny nσ+ 6= nσ−
F=5
6P3/2 x z x z
F=2
Probe 852nm Atomic Atomic

sample sample
F=4 u lse y lse y
6S1/2 ht p h t pu
F=3 Lig Lig
(b) Linear birefringence Circular birefringence
1
∝ ∆12 ∝∆ Ŝy ↔ Ŝz Ŝx ↔ Ŝy
Ŝx constant Ŝz constant
Figure 2.3: (a) The probe laser interacting with cesium atoms and the relevant
level structure. The detuning ∆ is defined relative to the F = 4 → F 0 = 5-
transition and is positive for red detuning. (b) The absorption (solid line) and
refractive index (dashed line) in general for an optical transition. For sufficiently
far detuning absorption effects are small since they fall off faster (as 1/∆ 2 ) than
dispersion effects (fall off as 1/∆). (c) A light pulse on an atomic sample which
is aligned along the x-axis will change its polarization, the sample has linear
birefringence with different indices of refraction nx 6= ny . A ±45◦ -polarized
pulse will become elliptical and vice versa, i.e. Ŝy and Ŝz will change. With the
x-axis as optical axis the Ŝx component is unchanged. (d) For spins oriented
along the propagation direction, there is no preferred direction in the xy-plane,
but the sample has circular birefringence with nσ+ 6= nσ− . Linear polarization
will rotate, i.e. Ŝx and Ŝy change, but the Ŝz -component is unchanged.
the +45◦ -direction interact with the atoms, the light polarization has both an
x- and y-part which are subject to different phase shifts. This will change the
ellipticity of the light, or in other words Ŝy and Ŝz begin to mix up. A quantitative
discussion of this effect is given in Sec. 6.4, see Eq. (6.16) and (6.17).
Another example is shown in Fig. 2.3(d), where the atoms now are oriented
along the z-direction. Classically we have in this case nx = ny = n+45◦ = n−45◦
since there is no preferred direction in the xy-plane. But spins pointing in one
direction along the z-axis is, from a classical view point, the same as a charged
particle rotating in one direction around the z-axis. This suggests that we may
have nσ+ 6= nσ− , the sample has circular birefringence. The different phase
shift experienced by the σ+ - and σ− -part of the light will rotate the polarization
around the z-axis, hence Ŝx and Ŝy begin to mix up. The number of σ+ - and
σ− -photons cannot be changed by this phase shift, thus Ŝz is unchanged in the
process. A quantitative discussion of circular birefringence is given in Sec. 6.2.
For our experiments the linear birefringence turns out to play a minor role
compared to the circular birefringence. Linear birefringence is non-existing for
spin-1/2 particles2 , and if the detuning ∆ is much greater than the excited states
hyperfine splitting ωhfs,e the probe laser only experiences the spin-1/2 properties
of the electron. We have approximately that linear birefringence is proportional
to ωhfs,e /∆2 . Circular birefringence is possible for a spin-1/2 particle and survives
large detunings, i.e. is proportional to 1/∆. Another way of understanding this is
that the dispersive profile shown in Fig. 2.3(b) for each transition F = 4 → F 0 =
3, 4, 5 will interfere destructively for the linear birefringence and constructively
for circular birefringence. A quantitative measure of the above is partly expressed
by the parameters defined in Eq. (5.16).
In the following we assume that linear birefringence is zero, and that light
and atoms are polarized along the x-direction. We are now ready to give a
qualitative derivation of the equations of interaction between polarized light and
atomic spins. These equations are the basis of the work in this thesis.
First, when linearly polarized light passes the atomic spin in the z-direction,
the circular birefringence caused by the spin component ˆ
D E Jz will rotate the polar-
ˆ
ization of the light. Even though the mean value Jz may be zero the quantum
fluctuations of Jˆz cause polarization rotation. This is expressed quantitatively as
Ŝyout (t) = Ŝyin (t) + aSx Jˆz (t), (6.11)
where “in” and “out” refers to the light before and after the interaction. The
Stokes vector with strong x-component Sx is rotated by an angle aJˆz around
the z-axis where a is a constant describing the strength of the interaction. The
above equation is calculated to first order with aJˆz ¿ 1. This is also known as
the Faraday effect. We know that circular birefringence does not affect Ŝz and
we neglected linear birefringence, hence
Ŝzout (t) = Ŝzin (t). (6.12)
The light leaving the atomic sample carries information about the spin component
Jˆz . We see that a measurement of Ŝyout will give information about Jˆz , and if the
interaction constant a is large so that the second term of (6.11) dominates the
first we can get really detailed information about Jˆz . The Heisenberg uncertainty
relation (2.3) then requires the interaction also to affect Jˆy . A calculation shows
that the time evolution of Jˆy is
∂ ˆ
Jy (t) = aJx Ŝz (t). (6.13)
∂t
The physical process involved is the Stark shift of the magnetic sublevels de-
pending on the helicity of the light given by Ŝz . The splitting caused by the light
will, just like a constant magnetic field along the z-direction, cause rotation of
the spin around the z-axis. The strong Jx component will then contribute in the
2 Linear birefringence is caused by alignment terms, e.g. ̂ 2 − ̂2 . For a spin-1/2 we have
x y
̂2x = ̂2y = 1/4, hence no alignment.
2.3. Atom/Light Interactions 11
y-direction, the angular rate of spin rotation being aŜz (t) where we normalized
Ŝz (t) to measure photons per second. Finally, the Jˆz component is unaffected by
the interaction,
∂ ˆ
Jz (t) = 0. (6.14)
∂t
This follows from the fact that Ŝz is constant and the conservation of angular
momentum along the z-direction. Spin flips along z must be accompanied by
changes of σ+ -photons into a σ− -photons or vice versa.
The strong components Sx and Jx are effectively unchanged by the interac-
tion. In our experiments the typical rotation of light polarization or spin polar-
ization amounts to approximately 10−6 radians, and we may clearly assume Sx
and Jx to be constant.
The above equations are linear, they couple polarization states of light to the
spin states of atoms. Note, that quantum properties of the spin state can be read
out on light by (6.11) and that quantum properties of light can be fed into atoms
by (6.13). We study the flow of quantum fluctuations of this kind in detail in
Chap. 8.
Note also, that (6.11) enables us to measure Jˆz (if the first term is small),
and at the same time (6.14) ensures that the state of Jˆz will not be destroyed in
the measurement process. Thus, we are able to perform quantum non-demolition
(QND) measurements of the atomic spin. This will prove very useful for genera-
tion of entangled states as we will describe in Chap. 9.
13
CHAPTER 3
Motivation - Quantum
Information Processing and
Communication
In this chapter we try to give the reader an overview of the field of quantum
information processing and quantum communication. This is a part of physical
science which has grown very rapidly since the beginning of the 1990’ies. The
physics involved is very interesting in itself but technological advancement is also
a motivation for studies in this field. We will briefly touch the areas of quantum
cryptography, quantum computing, quantum communication, and precision mea-
surements. We aim to describe for the reader where our work should be placed
in this context.
3.1 Why Quantum State Engineering

In Chap. 2 we introduced our experimental systems, spin polarized atomic sam-
ples and polarization states of light. We explicitly concentrated on quantum
variables and not classical mean values. There are some reasons for this being
an interesting approach.
• Quantum states are rich in the sense that it takes many parameters to
describe the quantum state. For our systems, which approximately are
continuous like position x̂ and momentum p̂ of a particle, the quantum state
is in principle described by a wave function ψ which for each x assigns a
14 Chapter 3 - Motivation
complex number ψ(x), a vast amount of complex numbers. We could also

count e.g. individual spin-1/2 particles which each have a Hilbert space of
dimension two. The joint quantum state of n spin-1/2 particles is described
by 2n complex numbers, an amount rapidly increasing with n. This richness
is exploited in quantum computing protocols, as we will discuss in the next
section.
• Quantum states cannot be cloned [2]. This means that there is some sort of
privacy in quantum states. If I have a secret message encoded in a quantum
state an eavesdropper will have to steal the entire quantum state in order
to obtain this message, and accordingly I would loose the quantum state -
a situation that I would recognize and act upon.
• The above point also presents a very big challenge. With only one copy of
the quantum state we are very vulnerable to loss mechanisms. Information
about the quantum state can “leak” into the surrounding environment and
be inevitably lost. Excluding unwanted coupling to external degrees of
freedom is thus a big challenge.
• The fact that we can reach a situation where quantum mechanics is essential
to describe the evolution of physical systems and especially the need of
quantum mechanics for the understanding of encoding and processing of
messages is satisfactory in itself for a physicist and it may be a very crucial
step for future technological achievements.
3.2 Technological Implementations

In the previous section we gave some rather abstract reasons for quantum states
being interesting. In this section we concentrate a little more on what has actually
been proposed or achieved.
• Quantum cryptography is an area of physics dealing with secret com-

munication. The aim is often to distribute a key between two parties,
Alice and Bob, and this key must be unknown to everyone else. Quantum
cryptography exploits the fact that quantum states cannot be cloned [2].
For instance a single photon can bear information sent from Alice to Bob.
An eavesdropper cannot steal this photon or perform measurements on it
without being recognized. Quantum cryptography has been implemented
using single photons or at least weak pulses which very seldom contain more
than one photon [3]. Using single particles as information carriers is elegant
and intuitive. But pulses containing many photons have also been demon-
strated as an implementation for secret communication [4, 5, 6, 7, 8, 9].
If our physical systems should be implemented for cryptography it would
possibly be along these lines.
3.2. Technological Implementations 15
• Quantum computation is a big motivation for exploiting and controlling

quantum systems. The most famous proposals are Grover’s search algo-
rithm [10] and Shor’s algorithm for prime factorizing integers [11]. Both
algorithms will be much faster than any known algorithms on a classical
computer. The theory of quantum computation utilizing qubits (quantum
superpositions |ψi = α |0i + β |1i of two discrete states |0i and |1i) and a
few operations on qubits (e.g. the phase gate and the CNOT-gate) is well
developed. With an appropriate combination of these basic building blocks
any computation can in principle be performed [12]. Many small-scale ex-
perimental demonstrations of quantum computing have been performed.
To mention a few, ion-traps have been proposed and used [13, 14, 15] and
nuclear magnetic resonance (NMR) of molecules have been used to e.g. fac-
tor the integer N = 15 [16]. Scaling up of quantum algorithms to large
systems have not yet been implemented, since it is a very difficult task.
The process is very vulnerable to decoherence and must rely on quantum
error correction. A review of quantum computing including error correction
is given in [17].
The above implementations all deal with discrete quantum systems. There
are some proposals for quantum computation [18] and quantum error cor-
rection codes [19, 20] over continuous variables but the discrete implementa-
tions seem more promising. In [21] it has been shown how distinct coherent
states can “discretize” continuous variables and thus simulate qubits. This
proposal requires generation of superpositions of distinct coherent states
(Schrödinger cat states) which is a difficult task. A proposal for gener-
ating such states in an ensemble of spin states like ours is given in [22].
Schrödinger cat states have been experimentally demonstrated for an elec-
tromagnetic field [23] and for the motional state of a trapped ion [24]
• Quantum communication is the transport of (unknown) quantum states

from one place to another. This could be among e.g. two atomic samples or
from an atomic sample to a beam of light (these process are in some cases
called teleportation). Again, the no-cloning theorem [2] imposes some rules
to follow. If a quantum state is to be sent from Alice to Bob, the initial
state at Alice’s place must be completely destroyed in order for Bob to re-
create exactly the same state. Also, information about the state must not
leak into the environment or be measured by any observer in the process.
These facts require good isolation from external degrees of freedom and
some cleverness in handling the quantum states. In Chap. 11 we will show
that protocols actually exist for physical systems of our kind. In fact,
continuous systems with many particles have an advantage over discrete
systems in teleportation protocols. Quantum teleportation of quadratures
of the electromagnetic field has been achieved [25]. Teleportation of discrete
states of a photon has partly been performed [26, 27]. In these cases,
however, a complete teleportation protocol that always works is difficult to
implement [28, 29].
• Quantum memory is the process of storing and retrieving a quantum

state from some long lived physical system. This could be achieved with
the help from teleportation, e.g. the polarization state of a pulse of light
could be teleported onto the spin state of an atomic sample. After some
time (less than the atomic decoherence time) we could teleport the atomic
spin state back to another pulse of light. Protocols along these lines are very
relevant for us and will be discussed in Chap. 11. Another approach toward
quantum memory is the use of electromagnetically induced transparency to
map states of the electromagnetic field onto atomic variables [30]. This has
been demonstrated experimentally for classical mean values [31, 32, 33] and
is a promising candidate for a real quantum memory.
• Entanglement generation is a key ingredient for quantum teleportation

protocols but also in itself it provides a very interesting study of quantum
mechanics. Generation of entangled states of the electromagnetic field has
been achieved in many cases, some examples can be found in [34, 35, 36, 37].
One of the really famous studies of such states was the first experimental
violation of Bell’s inequalities [38, 39] where the very basis of quantum
mechanics was tested. We demonstrate the creation of an entangled state
between two atomic samples in Chap. 10. See [40, 41] for other examples
of entangled states between massive particles.
• Precision measurements is also a motivation for the study of quantum

states. Spin squeezing can improve signal to noise ratios of certain mea-
surements [42]. In fact in [43] it was shown that the best cesium frequency
standards today are limited by the projection noise of spin states. Spin
squeezing of atomic states have been demonstrated in [44, 45].
In short, there are many possible implementation of our research. Our particu-
lar physical systems with strong light pulses and macroscopic atomic samples is
well suited for teleportation protocols, quantum memory protocols, entanglement
generation, and precision measurements. There will probably be some applica-
bility in the field of quantum cryptography. The field of quantum computation
seems to be the hardest problem to address with our present knowledge.
3.3 Continuous Versus Discrete Systems

Our physical system with many photons and many atoms lead to continuous vari-
ables as introduced in Chap. 2. In this section we motivate the use of continuous
systems for light/matter interactions and we draw attention to (dis)advantages
of these kinds of systems.
A very important motivation for using many particles is the question of in-
teraction strength. If light and atoms only induce weak changes to the quantum
state of each other we would not be able to let quantum fluctuations flow from
one system to the other with high efficiency. The interaction between light and
3.3. Continuous Versus Discrete Systems 17
matter is (in the dipole approximation) basically governed by the Hamiltonian

Ĥint = −d · E where d is the dipole operator of the matter particle and E is the
electromagnetic field. We need this Hamiltonian to be of sufficient magnitude
which can be reached in a number of different ways.
• The electromagnetic field E can be made really big. One way to do this
is to place a single particle inside a high finesse cavity which may enhance
the interaction by many orders of magnitude [46].
• The dipole moment of a single particle can be made large by using Rydberg
states of atoms. Examples are given in [47].
• Yet another possibility is our approach.

P Using many particles we will have
a large joint dipole operator D = di and using a relatively strong field
E we may approach a strong interaction regime. The figure of merit is
here Eq. (6.11), we need the second term aSx Jˆz to be of the order of the
first term Ŝyin to have strong coupling. With the quantum noise limited
variances for spins (2.5) and polarization states of a light pulse (2.12) we
conclude that Var(aSx Jˆz ) ≥ Var(Ŝyin ) when a2 Sx Jx ≥ 1. The parameter a,
which is given by Eq. (6.15), has for our setup typically a value such that
the strong coupling condition is of the order Sx Jx ≥ 1025 .
Of the three approaches above ours is by far the most simple from a technical per-
spective. We use free propagating light through a sample of atoms in the ground
state. When technical challenges has been dealt with, the other two approaches
reveal a more simple and elegant quantum system than ours. Especially, the
internal atomic state is described by a low dimensional Hilbert space contrary to
our practically infinite Hilbert space.
Another aspect of macroscopic continuous systems is robustness to decoher-
ence. Some people often state mistakenly that a very big quantum system de-
coheres very fast simply because of the number of particles involved. This need
not be the case. We deal with the collective properties of e.g. the spin state of
an atomic ensemble with ≈ 1012 particles. The role of each atom is totally neg-
ligible, if one atom is lost the quantum state of the ensemble will be unchanged.
If on the other hand 10% of the atoms are lost or subject to decoherence we will
see the effect on the quantum state of the collective spin variable. The atoms co-
operate in a fashion where each atom contributes very little but the huge number
of particles together make a difference. So, we must screen our atoms in general
from decoherence, but we need not care about a single atom alone.
If we should conclude this chapter with some general remarks about our ap-
proach with many particles and other approaches with small systems it would
be: Large systems often have technical simplicity from an experimental point
of view. They have a clear advantage in the fields of quantum communication
but lack good ideas in the areas of quantum computation and error correction.
Small systems are often very involved experimentally but that said the concep-
tual understanding is simple and elegant. These systems have an advantage for
purposes of quantum computation, quantum cryptography, and error correction.

For quantum communication protocols they meet difficulties.
19
CHAPTER 4
Experimental Methods
In this chapter we discuss general aspects of our experiments. This includes

details about lasers, glass cells with cesium atoms, magnetic fields and Larmor
precession, spin life times, and optical pumping. We concentrate on properties of
lasers and atomic samples common to many experiments and we discuss typical
values of experimental parameters for the work presented in this thesis. More
specific experimental details will be given in the different chapters connected to
the experimental results.
During the past four years we have moved laboratories twice and made a
major upgrade of the experimental setup. Some reasons for the upgrade are
discussed in Sec. 10.2. The contents of the present chapter describe the newer
setup since this will be more relevant to most readers. Important differences are
mentioned in other chapters when appropriate.
4.1 Laser Systems

The level scheme of cesium is shown in Fig. 4.1 together with some transitions
at which we apply laser beams. All our interesting physics mainly takes place
in the F = 4 hyperfine multiplet, and we need lasers for state preparation and
manipulations. The goal is to create high quality coherent spin states.
One laser called the optical pump laser is tuned to the 6S1/2 , F = 4 →
6P1/2 , F = 4 transition (894nm). Its main purpose is to pump atoms into the
extreme F = 4, mF = 4 ground state magnetic sub-level which is the starting
point for many interesting experiments (see discussion in Sec. 2.1). Note, that
20 Chapter 4 - Experimental Methods
−∆
6P3/2
251MHz
F=5
F=4
F=3
201MHz
6P1/2 152MHz
F=2
F=4
1168MHz
Repump Probe
F=3
laser laser
Optical
pump
laser
6S1/2
F=4
9192MHz
F=3
Figure 4.1: The level scheme of cesium and the lasers used in the experiment.
The nuclear spin I = 7/2 creates hyperfine splitting with F = 3, 4 of the cesium
ground state. All our experiments concentrate on atoms in the F = 4 hyperfine
levels and to measure properties of these atoms the probe laser is coupled off-
resonantly with detuning ∆ to the 6S1/2 , F = 4 → 6P3/2 transitions. The repump
laser and the optical pump laser are tuned into resonance with the 6S1/2 , F =
3 → 6P3/2 , F = 4 and 6S1/2 , F = 4 → 6P1/2 , F = 4 transitions, respectively.
These lasers redistribute atoms among the ground state levels by optical pumping,
for more details see Sec. 4.5.
the F = 4, mF = 4 state is a dark state if this laser has σ+ -polarization.

Another laser called the repump laser is tuned to the 6S1/2 , F = 3 →
6P3/2 , F = 4 transition (852nm). This laser is responsible for taking atoms
out from the F = 3 ground state and into the F = 4 ground state. To some ex-
tent the strength of this laser controls the number of atoms in the for us relevant
F = 4 sub-states.
The repump laser and the optical pump laser are both home built diode lasers,
a picture is shown in Fig. 4.2. During the past four years a number of different
laser designs have been used, the present one being the most successful. The
diodes are anti-reflection coated and can be purchased from Eagleyard Photonics
GmbH in Germany. The diodes work very well but we have had some problems
with the life time. The laser cavity consists of the diode back side together with
a diffraction grating with 1800 lines per mm in the Littrow configuration, see
Fig. 4.2(a). This ensures tunability over a broad range of wavelengths. The
light passes an optical isolator and a small fraction is split off for locking to
the right transition by frequency modulated absorption spectroscopy [48, 49], see
Fig. 4.2(b). The remaining beams are used in the experiment for optical pumping
(see Sec. 4.5) creating macroscopic spin states with high degree of orientation.
The lasers are running cw, at present we create pulses of light by acousto-optical
4.1. Laser Systems 21
(a) (b)
Cesium l/4
vapour To experiment
Grating
Isolator
l/2 AOM
Di
od
e
Detector
Figure 4.2: The diode lasers. (a) A picture of a laser diode hidden inside a
copper block directing its beam to a diffraction grating (the small square). A
piezo-element connects the grating to a steel ball which can be fixed by two
screws seen at the top. This design is cumbersome to adjust but very stable
when working. (b) The work principle of the laser. After the grating the light
passes an isolator and a small fraction is reflected on a PBS for saturated FM
absorption spectroscopy [48, 49]. The remaining beam is shaped into pulses by
an AOM. When lasers are working well, more than 25mW of power is available
for the experiments.
modulators (AOMs). Historically we have also been using a chopper for this
purpose, but the AOMs are much more flexible in terms of timing the laser
pulses.
A third laser called the probe laser is detuned by an amount ∆ from the
6S1/2 , F = 4 → 6P3/2 , F = 5 transition (852nm). Different values of the detuning
have been used around ∆ ≈ −1GHz (negative for blue detuning). This detuning
is much smaller than the ground state hyperfine splitting at ≈ 9GHz (so that
light is sensitive to atoms in F = 4 only) and considerably larger than the
hyperfine splitting of the upper state 6P3/2 (this reduces higher order effects
which will be discussed further in Chap. 5). The probe laser beam is produced
by a Microlase Ti:sapphire laser which can deliver typically around 1W of light at
852nm when being pumped by a Coherent Verdi V8 laser delivering 8W of light at
532nm (doubled Nd:YAG). The setup is shown in Fig. 4.3 where also the locking
mechanism is pictured. The locking is similar to the case of diode lasers apart
from the fact that the part split off is passing through a fiber coupled electro-
optical modulator (EOM) which creates strong sidebands with frequencies up to
more than 1GHz. In this fashion detunings up to more than 2GHz are available
(we can use the second side band for locking also). The probe laser is also a
cw laser, pulses were historically created by a chopper but now we use an EOM
with a polarizing beam splitter (PBS). The Verdi laser has relaxation oscillations
around 500kHz which are inherited in the spectrum of the probe laser. We are
vulnerable to laser noise but we find that the noise spectrum is pretty quiet
around 325kHz (it would also be quiet at high frequencies above 1MHz). In
Sec. 4.3 we explain that our experiments are carried out in a way where the
(a) (b) l/2 l/2 To experiment
EOM Cesium l/4

vapour
l/2
Isolator
1 GHz EOM
Verdi V8
Detector
Ti:Sapphire
Figure 4.3: The probe laser. (a) A picture of the Ti:sapphire laser, the green
532nm pump beam is clearly visible. (b) The schematic view of the laser system,
a Coherent Verdi V8 laser delivers 8W of pump power to the Ti:sapphire laser
which typically delivers 1W of light power at 852nm. Part of this light is sent
through an EOM to create strong sidebands at a chosen frequency of up to more
than 1GHz. The sideband is used for FM absorption spectroscopy similarly to
the diode lasers in Fig. 4.2. The main Ti:sapphire beam is shaped into pulses by
an EOM and a PBS.
frequency components around 325kHz are important. The probe laser is used for
quantum measurement of transverse components Jˆy and Jˆz of the atomic spin
and for measuring the macroscopic size Jx of the spin states.
The AOMs and the EOM creating laser pulses can be programmed to pro-
duce pulses with soft edges (not step functions). This is important, a too steep
pulse would have frequency components at 325kHz creating problems for the ex-
periment. We also examined the possibility to use an AOM for the probe laser
but found that this caused too much excess noise in the laser beam. AOMs are
turned on and off by adjusting the power of an electric RF signal (in our case
125MHz). It is a non-trivial task to turn on and off such a signal, this is typically
done by electronic mixers. We found that for the mixers we used there was a
huge amount of noise added in the regime between on and off, hence a quiet soft
pulse was impossible to create. For the case of an EOM the control is done by
a high voltage at DC. It is easy to exclude 325kHz signals from a DC-signal, or
rather our high voltage supply is not even able to work at frequencies around
325kHz.
4.2 Atomic Vapour Cells

Our atomic samples are very conveniently placed inside a paraffin coated glass
cell, see Fig. 4.4. The glass cell consists of a volume not far from being a cube
with six small cylindrical extension, the internal distance between two windows
is 30mm and the volume inside the cell is roughly 18±1cm3 . Taking the glass cell
to be box-shaped this corresponds to an effective transverse area Aeff ≈ 6.0cm2 .
4.2. Atomic Vapour Cells 23
Finger with cesium
Figure 4.4: A picture of a paraffin coated vapour cell and a ruler showing the
scale, the inside separation of two sides is 30mm. Note, that windows give the
possibility to access the atoms with large laser beams from six directions. The
finger contains solid droplets of cesium, the temperature of these decide the
vapour pressure and hence density of atoms in the entire cell volume.
For the cell shown in the figure there are six windows which enable us to access
the atoms by large laser beams from different directions. In addition, there is
mounted a finger which contains droplets of solid cesium. The amount of cesium
vapour in the cell volume is governed by the temperature of the solid cesium.
The cesium atoms are kept in the vicinity of room temperature which means
that atoms are moving with velocity components of the order
1 2 1 2
mCs vrms,1dim = kB T → vrms,1dim ≈ 137m/s, (4.1)
2 2
where kB is Boltzmann’s constant and mCs is the cesium atomic mass. With
a cell dimension of 30 mm it takes of the order of 200µs between each time an
atom collides with the cell wall. We discuss later (see Sec. 4.4) that the atomic
spin life time is much longer than these time scales, the reason that atoms do not
depolarize at the wall collision is the fact that a thin layer of paraffin has been
placed on the inside of the glass walls. Our glass cells have all been produced by
Michael Balabas, S. I. Vavilov State Optical Institute, St. Petersburg, Russia,
for further information about the physics of paraffin coated cells we refer to
[50, 51, 52].
Atoms moving at room temperature also cause Doppler broadening of the
optical line by an amount (see e.g. [53])
r
ν0 2 ln 2kB T
δνD,HWHM = = 189MHz, (4.2)
c mCs
where ν0 is the optical frequency, c the speed of light, and we chose temperature
T = 300K. In our experiments we are concerned with dispersive effects and not
absorption. This also requires the probe laser beam to have a sufficiently far
detuning ∆ À 189MHz.
The fact that atoms move around gives some advantages. First of all, the
probe laser beam cannot fill the whole cell volume but the motion of atoms
ensures that all atoms will pass the beam at some stage during each measurement.
There will also be some averaging effects of atomic motion, some atoms will be
closer to optical resonance than others, but at a later stage these atoms may
be further from resonance. For sufficiently long measurement times all atoms
will experience the same interaction conditions with the laser beam. And finally,
possible inhomogeneities of magnetic fields cause different Larmor frequencies for
different atoms, but some averaging reduces this effect (this is discussed more in
Sec. 4.4).
4.3 The Rotating Frame and Magnetic Fields

All our experiments are carried out with atoms being placed in a constant homo-
geneous magnetic field B along a direction which we define as the x-axis in this
thesis. This will split the magnetic sub-levels and cause spin precession around
the direction of the magnetic field. In App. F we give a detailed discussion of
the splitting of magnetic sub-levels, here we just state that we choose a magnetic
field of magnitude ≈ 0.9Gauss which corresponds to precession frequency (Lar-
mor frequency) of Ω = 325kHz. There are several advantages and dis-advantages
of this magnetic field.
Let us examine the implications of the static magnetic field on the equations
of motion introduced in Sec. 2.3 and stated quantitatively in Eqs. (6.11-6.14).
The magnetic field contribution to the Hamiltonian is Ĥ = h̄ΩJˆx where Ω is
the Larmor frequency. This changes Eqs. (6.13) and (6.14) into ∂ Jˆy (t)/∂t =
−ΩJˆz (t) + aJx Ŝzin and ∂ Jˆz (t)/∂t = ΩJˆy (t), while Eqs. (6.11) and (6.12) are
unaffected. If we introduce rotating frame coordinates (marked with a prime)
Jˆy0 (t) = +Jˆy (t) cos(Ωt) + Jˆz (t) sin(Ωt),

(4.3)
Jˆ0 (t) = −Jˆy (t) sin(Ωt) + Jˆz (t) cos(Ωt),
z
a little algebra shows that the equations of motion are turned into
³ ´
Ŝyout (t) = Ŝyin (t) + aSx Jˆy0 (t) sin(Ωt) + Jˆz0 (t) cos(Ωt) , (4.4)
Ŝzout (t) = Ŝzin (t), (4.5)
∂ ˆ0
J (t) = aJx Ŝzin (t) cos(Ωt), (4.6)
∂t y
∂ ˆ0
J (t) = aJx Ŝzin (t) sin(Ωt). (4.7)
∂t z
Our first observation is the fact that the dynamics of the rotating spin component
are encoded around the Ω = 325kHz sideband of Ŝyout . Lasers are in general much
more quiet at higher sideband frequencies compared to the carrier and since we
are interested in quantum fluctuations of light and atoms interacting with each
4.3. The Rotating Frame and Magnetic Fields 25
(a) (b) (c)
Glass cell
Laser beam access
Aluminum RF-coil
Bias field coils
Figure 4.5: Pictures of the cell mounts and the magnetic field coils. (a) The
glass cell is here placed in an aluminum block which can be heated or cooled
by water. Close to the cell we see a set of coils which are used for creating a
horizontally polarized RF-magnetic field. (b) The next layer consists of eight
coils with different number of windings to create a homogeneous bias magnetic
field along the vertical direction. (c) At the outside we place two layers of µ-metal
and one layer of iron to protect the atoms from external magnetic fields.
other we cannot have our signals dominated by much stronger technical noise
sources.
Having rotating spins also enable us to measure two orthogonal components
Jˆy0 and Jˆz0 with one laser pulse as long as the measurement time T is much longer
than the Larmor period Ω−1 . This fact will be clarified in Sec. 9.4 when we discuss
entanglement generation. But the presence of the magnetic field will at the same
time force us to measure both transverse spin components with the simultaneous
pile up of noise according to (4.6) and (4.7). This is a strong limitation in some
cases, see e.g. the discussion in Sec. 11.3. For a single spin sample the QND
nature of the measurement has disappeared.
The energy splitting caused by the magnetic field ensures that all magnetic
sub-levels are non-degenerate. This is important since the energy barrier prevents
atoms from doing spin flips at a collision. This will be discussed further in Sec. 4.4
but we can now refer to Fig. 4.6 showing the spin life time T1 versus the magnitude
of the constant magnetic field B.
The magnetic field homogeneity must be of sufficiently good quality. If this
is not the case, different atoms following different paths in space will accumulate
different phases and the joint spin state of atoms will decohere. This will be
discussed in detail below in Sec. 4.4. The experimental setup for creating a
high quality magnetic field with good stability is shown in Fig. 4.5. In part (a)
we see the glass cell mounted inside a plastic cylinder in a block of aluminum.
The aluminum can be heated or cooled by water and hence controls the glass
cell temperature1 . A set of coils placed close to the cell enable us to create a
horizontally polarized RF-magnetic field. The role of this is to modulate the spin
state, see the discussion in Chap. 7. In Fig. 4.5(b) we see the coils creating the
constant high quality magnetic field. Eight coils with equal spacing have different
number of windings optimized for high homogeneity. Three independent current
sources connected to these coils allow further optimization of the homogeneity.
The current source driving these coils has a relative stability better than 10 −5 . In
Fig. 4.5(c) we show the outermost magnetic shielding consisting of two layers of
µ-metal and one iron layer. The top and bottom are also shielded and we exclude
laboratory fields from affecting the atoms. All together we create a magnetic field
of sufficient quality, in next section we discuss quantitatively the requirements
for the magnetic field.
4.4 Spin Life Times

The life time of our macroscopic spin states with a large spin component J x
along the x-axis and transverse spin components Jˆy and Jˆz is well described by
two characteristic times T1 and T2 (which are well known concepts from the lit-
erature). T1 is the decay time of the longitudinal spin Jx following the model
Jx (t) = Jx (0)e−t/T1 and T2 is the same for transverse components. We usually
have T1 ≈ 200-300ms which is much longer than T2 ≤ 30ms. The decay mecha-
nisms of Jx must overcome an energy barrier set by the splitting caused by the
magnetic field. The transverse decay will also be affected by phase fluctuations
and hence T2 can be much faster than T1 . We often characterize the transverse
life time by a line width or decoherence rate Γ[Hz] = (πT2 [s])−1 and in this chap-
ter we often discuss the rate Γcom which is a decoherence rate common to all
magnetic sub-levels. This rate is discussed more carefully in Chap. 7 where we
also discuss in detail the methods for measuring both T1 and T2 . In the following
we discuss several experimental parameters that affect T1 and T2 .
As mentioned above the static magnetic field B giving rise to the Larmor
precession must be of a sufficient quality. In Fig. 4.6(a) we show the T1 life time
measured as a function of the magnitude of B. We see that there is a threshold
of the magnetic field strength of about 0.03Gauss above which we may obtain
long spin life times T1 . Our working point is far above this point and we have
usually T1 values of some hundreds of milliseconds.
In Fig. 4.6(b) we see the transverse decoherence rate Γ as a function of an
applied magnetic field gradient. We see that the rate increases quadratically with
the gradient, we can understand this with help from a simple model discussed
in [54]. First, divide the atomic sample into two parts, 1 and 2, along the
1 At the time of writing we are working on replacing the aluminum by non-metallic compo-
nents. Random currents in the aluminum have proved to create magnetic noise disturbing the
experiments, see Sec. 10.3.
4.4. Spin Life Times 27
Γinh [Hz] = 8.7(3)

(b)100
“ h i”2
(a) 400 +0.0158(2) ∂B ∂x
mG
m
νL ≈ 10kHz
80 B0 = 0.93G
300
Line width [Hz]

T1 [ms]
60
200
Usual
working 40
100 point
20
0
0
0.01 0.1 1 -60 -40 -20 0 20 40 60 80
B [Gauss] Field gradient [mG/meter]
Figure 4.6: (a) T1 versus magnetic field strength. We see that a too low field
(below ≈ 0.03Gauss) prevents long spin life time. Our usual working point is
around B = 0.93Gauss for which the spin life time yields some hundred millisec-
onds, here 380ms. (b) The magnetic field also needs a sufficient homogeneity to
prevent dephasing of the spins. We plot the decoherence rate given by the line
width Γcom versus an applied magnetic field gradient and observe the increase
of decoherence rate with increasing gradient. The measurement methods are
explained in detail in Chap. 7.
bias magnetic field direction. If the sample length is L and the bias field B has
strength B0 , the field strength in the two parts will be of order B0 ± ∂Bx /∂x · L
and the difference in Larmor frequency will be (gF µB /h̄)∂Bx /∂x · L according
to (F.4). We follow an atom during the time T it takes for it to decohere. If v is a
typical speed of the atomic motion, the number of visits n1 in part 1 or n2 in part
2 will be of order T v/L, since each visit has duration L/v. The difference
p n 1 − n2
has mean zero and standard deviation of the order std(n1 − n2 ) = T v/L. Thus
the uncertainty δφ in the accumulated phase during Larmor precession is
gF µB ∂Bx L
δφ ≈ L · · std(n1 − n2 ) ≈ 1
h̄ ∂x v
1 ³ g µ ´2 L3 µ ∂B ¶2
F B x
⇒ Γinh ≈ ≈ . (4.8)
T h̄ v ∂x
In the first line we set δφ equal to unity since this is the situation after the time
of decoherence T . We see that the broadening Γinh by inhomogeneities scales
quadraticallyp with the field gradient. If we take gF ≈ 1/4 (see eq. (F.5)), L =
0.030m, v = kB T /mCs = 137m/s at T = 300K, we get gF µB /h̄ = 350Hz/mG
and expect the broadening to be Γinh = 0.024Hz · m2 /mG2 · (∂Bx /∂x)2 .
The experimental investigation can be seen in Fig. 4.6(b) and we definitely
confirm the scaling law predicted above. The numbers match within a factor of
two which puts some confidence to our simple model but this is probably also
partly luck since we were very crude in the model with respect to factors of 2
and π.
(a) 30 Γcom [Hz] = 6.1 + 6.0 · P [mW] (b)

70 T = 45o
25
60
Linw width [Hz]

Linw width [Hz]
20
50
15 T = 20o
40
10 30
5 20
0 1 2 3 4 0 1 2 3 4
Probe power [mW] Density [1010 cm-3]
Figure 4.7: (a) The role of probe power broadening on spin decoherence rate
Γcom . We observe the expected linear dependence from Eq. (4.9), the numbers
are commented on in the text. Note, that we may also deduce the decoherence
rate in the dark by this method, here we get Γdark = 6.1Hz which corresponds to
a spin coherence time T2 = 52ms. (b) The spin decoherence rate versus atomic
density. These data give a typical picture, the decoherence rate increases in a
non-linear manner with increasing density. The density is controlled by raising
the temperature of the vapour cell and measured by polarization rotation, see
Eq. (6.9).
In Chap. 7 we will learn that having Γinh < νQZ is important, where νQZ is
the quadratic Zeeman splitting of some resonance spectra we utilize for spin state
characterization (see the example in Fig. 7.2(b)). Comparing our experimental
result with the splitting due to the quadratic Zeeman effect (F.7) we find for our
particular setup that in order to have Γinh < νQZ we must have 1/B · ∂B/∂x · L <
1.2 · 10−3 .
Another process leading to spin decoherence is the scattering of photons from
the probe laser when atoms undergo real transitions. Even though the probe
laser is detuned by ≈ 1GHz and the populations of excited state levels are very
small they are not zero. The scattering rate Γph can be estimated by the two-level
atom result (see e.g. [55])
γ s 3Iλ3 γ 2
Γph = ≈ , (4.9)
21+s 16π 2 h̄c∆2
I/Isat
where s = 1+(2∆/γ) 2 is the saturation parameter. I is the beam intensity with
Isat = 2π h̄cγ/3λ3 being the saturation intensity. λ is the optical wave length, γ
2
is the natural line width of the optical transition, and ∆ is the detuning (assumed
much greater than γ in the last step of the equation). If we insert typical exper-
2
imental conditions I = 1mW/cm , ∆ = 875MHz, λ = 852nm, and γ = 5.21MHz
we get Γph = 132Hz. This can be compared to experimental results shown
in Fig. 4.7(a). First, however, we need to observe that the scattering rate de-
4.5. Optical Pumping 29
pends only weakly on beam size for a fixed power. This effect is caused by the
atomic motion in and out of the beam. Even though the probe intensity can be
strong, the atoms will only spend little time inside the beam if the size is small.
The vapour cell geometry suggests an effective cross sectional area of the cell of
Aeff ≈ 6cm2 . Then the experimental observation Γcom [Hz] = 6.1 + 6.0 · P [mW]
2
corresponds to Γph [Hz] ≈ 36 · I[mW/cm ]. The experimental spin decoherence
rate is here a factor of almost 4 smaller than the simple two-level atom estimate.
We will not here try to do more correct quantitative estimates but just mention
that the order of magnitude is correct.
The last decoherence mechanism we will comment on is the dependence on
atomic density, see Fig. 4.7(b). These data are typical, the decoherence rate
increases in a non-linear fashion with the atomic density. If the decoherence is
caused by inter-atomic collisions we would expect the rate to increase quadrati-
cally with density, there is probably a quadratic contribution in the figure. But
the situation is more complicated than this. The change in temperature may af-
fect the properties of the paraffin, we have experimentally observed many strange
kinds of dynamics without understanding them very well. Also, the temperatures
are close to the cesium melting point at 28.4◦ C degrees which may also give rise
to complicated effects. We do not wish to understand all these processes in detail
as long as we can measure them. The main point is that at our working values
of atomic density (up to 1011 cm2 ) we have a reasonable decoherence rate but we
cannot go much further up without sacrificing the coherence time.
4.5 Optical Pumping
A very crucial part of our experiments is optical pumping. In Sec. 2.1 we already
mentioned the fact that we are interested in atoms in the hyperfine state F = 4
only and we commented on the importance of atoms being in the extreme m F = 4
state (or close to this state) during the experiments. This is the so-called coherent
spin state and we will see in Chap. 9 that the ability to create this state is crucial
for entanglement generation.
We will not here comment in detail about the theory of optical pumping, it is
a whole scientific field in itself and a review is given in [56]. The basic idea behind
optical pumping is simple though and can bee seen in Fig. 4.8. The repump laser
and optical pump laser which were described in Sec. 4.1 are applied to the atoms
with circular polarization. After some time the extreme F = 4, mF = 4 state
is reached by many atoms. Depending on the strength of the repump laser a
fraction of atoms will be in some of the F = 3 states and hence not contributing
to the collective spin state measured by the probe laser. The ability to adjust the
magnitude Jx of the spin state will prove to be important in Chaps. 8, 9, and 10.
A very important property of the atomic spins in the F = 4 states is the
m=1 2 3 4 m=1 2 3 4 5
6P1/2 6P3/2
σ+ σ+
etc etc
F =4 F =4
F =3 F =3
Figure 4.8: The principle of optical pumping, we wish to put all atoms into the
F = 4, mF = 4 magnetic sub-state. To this end we apply two σ+ -polarized lasers,
the optical pump laser shown on the left part and the repump laser shown on the
right part of the figure. The figure on the left shows an example where an atom
is in the state F = 4, mF = 2. The optical pump lifts this atom into the excited
F = 4, mF = 3 state from where it may decay into the ground states F = 3 or
F = 4 with mF = 2, 3, 4. The mF value will never decrease in this process and
will on average increase. If mF = 4 is reached the atom is in a dark state and
will not move further. The occasional decay to the F = 3 states is counteracted
by the repump laser shown on the right. The number of atoms in the relevant
F = 4 states can be adjusted by the power of the repump laser.
orientation defined by
F
1 X
p= m · hσ̂m,m i , (4.10)
F
−F
where F = 4 in our case and σ̂m,m is the density operator describing the popu-
lation of atoms in the magnetic sub-states |F = 4, mF = mi. If all atoms are in
m = 4 the orientation is equal to unity. For a completely unpolarized sample we
have p = 0.
An experimental example of obtained orientations is shown in Fig. 4.9 where
the orientation of a sample of cesium is studied while the power of the σ + -
polarized optical pumping laser is increased. The pump light is on resonance and
contributes to the decoherence of the spin state with a rate Γpump (this will be
carefully defined in Sec. 7.2). This rate is plotted on the abscissa in the figure
and is a direct measure of how many optical pump photons each atom scatters
on average per second. We see that a few photons per second are sufficient
for obtaining a high degree of orientation, in this example 97%. This should
be compared to a typical decay time T1 of some hundreds of milliseconds, see
Sec. 4.4. The repump laser is on at all times here and with its σ+ -polarization
it helps creating an oriented sample. The repump laser alone can here be seen
to generate a 82% oriented sample. The methods for measuring orientation is
described in detail in Chap. 7.
We have obtained polarizations of up to more than 98% for higher densities.
The optical pumping laser at the 894nm D1-line is essential to this achievement.
We have tried to optically pump on the D2-line with somewhat lower orientation
as a result (a little above p = 0.9). A possible explanation is that the re-scattered
light on the F = 4, mF = 4 → F = 5, mF = 5 transition from one atom
4.5. Optical Pumping 31
1.00
0.95
Orientation
0.90
0.85
0.80
0 1 2 3 4 5 6
Γpump [Hz]
Figure 4.9: An example of orientation p measured for increasing power of the

optical pump laser. On the abscissa we plot the power broadening Γpump caused
by the optical pump laser which in turn is a measure of the strength of the
pumping process. For zero optical pump power the orientation of 82% is created
solely by the σ+ -polarized repump laser. Note, that a high degree of orientation
(here ≈ 97%) is reached with a moderate amount of pumping. The atomic density
is of the order 1010 cm−3 .
affects the state of other atoms. Indeed, according to [57], even with a dark
state when using 894nm pumping light one would expect problems with densities
higher than a critical density ρC = (σR)−1 because radiation will be trapped
inside the sample. Here σ is the cross section for light absorption and R is the
extent of the gas sample. Our atomic sample is Doppler broadened with the
width δνD = 378MHz. With a natural line width of the 894nm D1-transition of
γ = 4.6MHz and a sample extent of R = 3cm we estimate the critical density ρ C
to be roughly ρC ≈ [λ2 /2π · γ/δνD · R]−1 = 2 · 1011 cm−3 . This is only a little more
than our typical values. However, the experiments tell us that the limitations
are still small.
Optical pumping into the coherent spin state with all atoms in the state
|F
D E 4, mDF =
= E 4i can also be seen from another perspective. This state fulfils
ˆ ˆ
Jy = Jz = 0 and any deviation from this state is counteracted by optical
pumping. In Chap. 8 we model the optical pumping by equations ∂ Jˆy (t)/∂t =
−ΓJˆy (t) + F̂y (t) and ∂ Jˆz (t)/∂t = −ΓJˆz (t) + F̂z (t) where Γ describes the rate
of optical pumping and the operators F̂y,z are Langevin forces ensuring correct
quantum statistics. Optical pumping is in this sense a clean up of whatever state
has been created by other processes. If e.g. two spin samples are in an entangled
state the optical pumping will drive the spin state back to being two independent
coherent spin states.
33
CHAPTER 5
The Effective Interaction

Hamiltonian
In this section we will consider the real cesium atom with its hyperfine split
ground and excited states. We couple these atoms off-resonantly to the 6S 1/2 →
6P3/2 dipole transition and we aim at the derivation of an effective Hamiltonian
to describe the physics of this interaction at the quantum level. We neglect
absorption effects and spontaneous emission which is justified if the detuning
from the optical transition is large enough. We are left with dispersive effects
which essentially arise from the shift of atomic energy levels in presence of light
fields. This is known as the Stark effect. As we already briefly mentioned in
Sec. 2.3, the interaction enables us to measure the spin state of atoms, and
properties of the polarization state of light will at the same time be fed into the
spin state of atoms.
This kind of interaction has been studied for some time now, for a historical
review see [56] and references therein. The idea of using the interaction for
QND measurements was given in [58, 59] and the calculations in the present
chapter is closely related to these references. We will concentrate more on the
continuous description of light and matter since this is convenient for describing
the time dynamics that we actually measure. We also put attention to the fact
that cesium is not a spin-1/2 system. This gives rise to higher order terms
of the interaction. We end up with an effective Hamiltonian (5.18) which is a
very convenient starting point for further calculations. The derivation is rather
technical, we put many details in Apps. A-D.
34 Chapter 5 - The Effective Interaction Hamiltonian
5.1 Electric Dipole Interactions

We assume that the light interacting with the atoms has a cross sectional area
A À λ2 where λ is the wave length of the light. Then a one dimensional theory is
sufficient and we only need to care about two polarization modes. With the prop-
agation direction (z-axis) as quantization axis the electric field description (C.9)
will be generalized to
r
h̄ω0 ³ ´
E= â+ (z, t)e+1 + â†+ (z, t)e∗+1 + â− (z, t)e−1 + â†− (z, t)e∗−1 , (5.1)
2²0 A
where the unit vectors e±1 and field operators â± (z, t) and â†± (z, t) are discussed
in Eqs.( A.6) and (A.7). The dipole operator d = −er of a single atom can
conveniently be expressed in tensor components. We write the vector r as
r = e∗+1 r+1 + e∗0 r0 + e∗−1 r−1 , (5.2)
where the tensor components of r are given by

x + iy x − iy
r+1 = − √ , r0 = z, r−1 = √ . (5.3)
2 2
With this definition the dipole operator can be expressed
X ³ ´
0 +
d= d− ∗ ∗ ∗
F,m;F 0 ,m0 e+1 + dF,m;F 0 ,m0 e0 + dF,m;F 0 ,m0 e−1 σ̂F,m;F 0 ,m0 + h.c.,
F,m;F 0 ,m0
(5.4)
where we let F and m sum over ground state levels while the primed letters F 0
and m0 sum over excited states. The dipole moments are defined as
d− 0 0
F,m;F 0 ,m0 = −e hF, m| r+1 |F , m i ,
d0F,m;F 0 ,m0 = −e hF, m| r0 |F 0 , m0 i , (5.5)

d+
F,m;F 0 ,m0
0
= −e hF, m| r−1 |F , m i .0
P
The interaction Hamiltonian
p Ĥint = − dj · E(Rj ) will contain the above dipole
moments and the factor h̄ω0 /2²0 A. We absorb these into a single coupling con-
±
stant gF,m;F 0 ,m0 = ω0 /2h̄²0 Ad±
p
F,m;F 0 ,m0 and the generalization of the Hamilto-
nian (C.15) will turn into (in the rotating wave approximation)
X Z L³
+
Ĥint = h̄ [gF,m;F 0 ,m0 â+ (z, t)
F,m;F 0 ,m0 0 (5.6)

´
−
+ gF,m;F 0 ,m0 â− (z, t)]σ̂F 0 ,m0 ;F,m (z, t) + h.c. ρAdz.
Here, the first term contains the annihilation operator â+ (z, t) for a photon at po-
sition z with polarization σ+ . This operator is accompanied by the density opera-
tor σ̂F 0 ,m0 ;F,m (z, t) which will take an atom from the ground state |F, mi into the
5.2. The Off-resonant Limit 35
excited state |F 0 , m0 i thereby absorbing the photon at position z. The strength

+
of this particular transition is governed by the coupling constant gF,m;F 0 ,m0 which
0 − 0
is non-zero only if m = m+1, while gF,m;F 0 ,m0 is non-zero for m = m−1. These
selection rules and the actual values of the coupling constants will be calculated
carefully in App. D. The values will turn out to be real, and we have omitted
the complex conjugation of these in (5.6).
We also need to state the atomic Hamiltonian to have all fundamental equa-
tions at hand. For the moment we assume that the energy levels of the ground
states are degenerate (there is e.g. no static magnetic field present) and we special-
ize to one of the hyperfine ground states, i.e. F = 4. A possible non-degeneracy
can be accounted for later. We get for the atomic Hamiltonian
5 XZ
X L
Ĥatom = h̄(ω0 + ∆F 0 )σ̂F 0 ,m0 ;F 0 ,m0 (z, t)ρAdz. (5.7)
F 0 =3 m0 0
Here ω0 is the laser frequency which is detuned ∆F 0 from the upper state with
total angular momentum F 0 . The density operator σ̂F 0 ,m0 ;F 0 ,m0 (z, t) measures
the probability for an atom at position z of being in the excited state |F 0 , m0 i
and h̄(ω0 + ∆F 0 ) assigns the appropriate energy in this case.
5.2 The Off-resonant Limit

We will now change the interaction Hamiltonian (5.6) into an effective Hamilto-
nian which depends on the light amplitudes and the ground state spin operators.
We can do this if we assume the optical laser field to be sufficiently far detuned
from atomic resonance. In this case the population of the excited states is neg-
ligible and the coherences σ̂F,m;F 0 ,m0 between the ground states |F, mi and the
excited states |F 0 , m0 i will follow the ground state and the light field adiabati-
cally. We carefully work out the adiabatic elimination and solve for the coherence
σ̂F,m−1;F 0 ,m (z, t) in the following.
The time evolution of σ̂F,m−1;F 0 ,m (z, t) is governed by the Heisenberg equation
∂ σ̂F,m−1;F 0 ,m (z, t) 1 h i
= σ̂F,m−1;F 0 ,m (z, t), Ĥ
∂t ih̄
= −i(ω0 + ∆F 0 )σ̂F,m−1;F 0 ,m (z, t) (5.8)
+
− igF,m−1;F 0 ,m â+ (z, t)σ̂F,m−1;F,m−1 (z, t)
−
− igF,m+1;F 0 ,m â− (z, t)σ̂F,m−1;F,m+1 (z, t),
where the commutation relation (C.17) has been used on the Hamiltonian (5.6)
and two terms proportional to σ̂F 0 ,m;F 0 ,m (z, t) and σ̂F 0 ,m−2;F 0 ,m (z, t) have been
neglected (no population in the excited states). The next step is to acknowledge
that the light amplitudes â+ (z, t), â− (z, t) and the coherence σ̂F,m−1;F 0 ,m (z, t)
are oscillating fast with frequencies ω0 and ω0 + ∆F 0 respectively. We go into the
rotating frame of the light field by introducing slowly varying operators
σ̂F,m−1;F 0 ,m (z, t) = σ̃F,m−1;F 0 ,m (z, t)e−iω0 t

and â± (z, t) = ã± (z, t)e−iω0 t ,
(5.9)
i.e. a tilde marks the operator to be slow. Now the slow version of (5.8) becomes
∂ σ̃F,m−1;F 0 ,m (z, t)
= −i∆F 0 σ̃F,m−1;F 0 ,m (z, t)
∂t
+
− igF,m−1;F 0 ,m ã+ (z, t)σ̂F,m−1;F,m−1 (z, t)
(5.10)
−
− igF,m+1;F 0 ,m ã− (z, t)σ̂F,m−1;F,m+1 (z, t).
On the right hand side we now have a fast term oscillating at ∆F 0 >> T −1 where
T is a typical time scale for the variation of the last two terms. Then it is justified
to put the time derivative equal to zero1 and we get
−1 h +
σ̃F,m−1;F 0 ,m (z, t) = g ã+ (z, t)σ̂F,m−1;F,m−1 (z, t)
∆F 0 F,m−1;F ,m
0
i
−
+ gF,m+1;F 0 ,m ã− (z, t)σ̂F,m−1;F,m+1 (z, t) ,
(5.11)
−1 h −
σ̃F,m+1;F 0 ,m (z, t) = g ã− (z, t)σ̂F,m+1;F,m+1 (z, t)
∆F 0 F,m+1;F ,m
0
i
+
+ gF,m−1;F 0 ,m ã+ (z, t)σ̂F,m+1;F,m−1 (z, t) ,
where we have also stated the result for the coherence σ̃F,m+1;F 0 ,m (z, t). The
physical interpretation of the above equation is quite simple. In the first line we
seek the coherence σ̃F,m−1;F 0 ,m , i.e. we want to know to which extent our atomic
state is in a superposition between the ground state |F, m − 1i and the excited
state |F 0 , mi. Such a superposition can be created in two ways.
In the first term on the right hand side, the population in the ground state
|F, m − 1i parametrized by the density operator σ̂F,m−1;F,m−1 is driven coher-
+
ently toward the excited state |F 0 , mi by the field â+ with strength gF,m−1;F 0 ,m .
The second term describes another possibility, the atomic state could already
be in a superposition between the ground states |F, m − 1i and |F, m + 1i. This
is parametrized by the density operator σ̂F,m−1;F,m+1 . The fraction of the atomic
wave function in the state |F, m + 1i can then be driven into the excited state
−
|F 0 , mi by the field ã− with strength gF,m+1;F 0 ,m .
The solutions (5.11) can now be substituted into the interaction Hamilto-
1 This is the adiabatic elimination, it can be shown to be equivalent to neglecting terms of
magnitude ∆F 0 T times smaller than the retained terms.

nian (5.6) to obtain an effective Hamiltonian;

X Z L nh
eff
Ĥint = c+ (∆, m)â†+ (z, t)â+ (z, t)
m 0
i
+ c− (∆, m)â†− (z, t)â− (z, t) σ̂F,m:F,m (z, t)
(5.12)
h
+ b(∆, m) â†− (z, t)â+ (z, t)σ̂F,m+1;F,m−1 (z, t)
io
+ â†+ (z, t)â− (z, t)σ̂F,m−1;F,m+1 (z, t) ρAdz.
This Hamiltonian only works on time scales long compared to ∆−1 F 0 since this is
the approximation in the adiabatic elimination. We have introduced a number of
coefficients for brevity. The terms containing c± (∆, m) describe the Stark shift
of the ground state |F, mi caused by the coupling to the excited state |F 0 , m ± 1i.
The coefficient is given by
±
X (gF,m;F 2
0 ,m±1 )
c± (∆, m) = −2h̄ . (5.13)
∆F 0
F0
The terms containing b(∆, m) describe the possibility to change ground state
from |F, m − 1i to |F, m + 1i through the excited state |F 0 , mi by absorption of
a σ+ photon and emission of a σ− photon (or vice versa). The coefficient is
+
X (gF,m−1;F −
0 ,m ) · (gF,m+1;F 0 ,m )
b(∆, m) = −2h̄ . (5.14)
∆F 0
F0
± 2 + −
The coefficients (gF,m;F 0 ,m±1 ) and (gF,m−1;F 0 ,m ) · (gF,m+1;F 0 ,m ) are calculated
in App. D, see Eqs. (D.12-D.14). Note, the denominator ∆F 0 in the above two
equations exclude us from applying the sum rules (D.15-D.17) and the entire
description becomes a little more complicated. We notice however that we can
group terms containing 1, m, m2 from (D.12) and the square root from (D.14)
such that (still for the special case of F = 4)
h̄cγ λ2 1
c± (m, ∆) = − (a0 ± a1 m + a2 m2 ),
4A∆ 2π 2 (5.15)
h̄cγ λ2 p
b(m, ∆) = − a2 (4 + m)(5 + m)(4 − m)(5 − m),
4A∆ 2π
where the coefficients a0 , a1 , and a2 are given by
µ ¶
1 1 7
a0 = + + 8 → 4, (F = 4)
4 1 − ∆35 /∆ 1 − ∆45 /∆
µ ¶
1 35 21
a1 = − − + 176 → 1, (5.16)
120 1 − ∆35 /∆ 1 − ∆45 /∆
µ ¶
1 5 21
a2 = − + 16 → 0,
240 1 − ∆35 /∆ 1 − ∆45 /∆
10
a0 → 4
1
a1 → 1
a0, a1, and a2

0.1
0.01 a2 → 0
0.001
0 500 1000 1500 2000 2500 3000
Blue detuning (-∆) [MHz]
Figure 5.1: The parameters a0 , a1 , and a2 defined in Eq. (5.16) for F = 4. These
parametrize the strength of the 0th, 1st, and 2nd order terms in the Hamilto-
nian (5.18), respectively. On the vertical axis is the blue detuning (−∆), and
the arrows indicate the limit for −∆ → ∞. We remind ourselves that the cal-
culations are only valid for dispersive effects. Because of Doppler broadening we
should be careful at low detunings.
where we have chosen to denote the detunings ∆F 0 as ∆50 = ∆, ∆40 = ∆ − ∆45 ,

and ∆30 = ∆ − ∆35 . Red detuning corresponds to positive values of ∆ and the
arrows indicate the limit ∆ → ±∞. The values of a0 , a1 , and a2 are depicted in
Fig. 5.1. If we insert the expressions (5.15) into (5.12) we end up with
h̄cγ λ2 X L ³ h †
Z i
eff
Ĥint =− a0 â+ â+ + â†− â− σ̂m,m
4A∆ 2π m 0
a1 h † i
+ â+ â+ − â†− â− mσ̂m,m
2h i (5.17)
+ a2 â†+ â+ + â†− â− m2 σ̂m,m
p
+ a2 (4 + m)(5 + m)(4 − m)(5 − m)×
h i´
â†− â+ σ̂m+1,m−1 + â†+ â− σ̂m−1,m+1 ρAdz,
where the (z, t) is left out for brevity. The density matrix operators and the
light operators â± , â†± are grouped in a nice way here, the light field operators
can be written in terms of Stokes operators, see Eqs. (A.8-A.12), and the terms
containing density operators can be expressed as spin operators with help from
Eqs. (B.3) and (B.6-B.8). With these substitutions we may finally write the
effective Hamiltonian as
h̄cγ λ2 L ³
Z
eff
Ĥint =− a0 · φ̂(z, t) + a1 · Ŝz (z, t)̂z (z, t)
4A∆ 2π 0 (5.18)
h i´
+ a2 φ̂(z, t)̂2z (z, t) − Ŝ− (z, t)̂2+ (z, t) − Ŝ+ (z, t)̂2− (z, t) ρAdz.
Let us comment on the different terms. The first term containing a0 will just give
a Stark shift to all atoms independent on the internal state but proportional to
ˆ
the photon density phi(z, t). The second term containing a1 rotates the Stokes
vector S and the spin vector J around the z-axis, known as Faraday rotation.
The last terms proportional to a2 are higher order couplings between the light
and the atoms. All these terms conserve individually the z-projection of the
total angular momentum of light and atoms, e.g. the Ŝ− ̂2+ term can change a σ+
photon into a σ− photon (changing the light angular momentum along z by −2h̄
2
while the atoms receive 2h̄ mediated by the atomic raising operator j+ . The total
angular momentum must have its z-projection invariant since the physical system
is axially symmetric around the direction of light propagation (the z-axis). We
remember that the parameters a0 , a1 , and a2 depend on the detuning ∆ and they
are given in Eq. (5.16) for the case of F = 4. In general, the term proportional to
a1 is useful for us and the higher order terms proportional to a2 create different
problems. This will be discussed further in Chap. 6 where the calculations also
will compare the magnitude of the a1 and a2 terms more quantitatively.
We could have performed all the steps in this section for F = 3 and ended up
with the same result, only the a-parameters would be a little different, they are
stated below for completeness.
µ ¶
1 25 63
a0 = + + 24 → 4, (F = 3)
28 1 + ∆24 /∆ 1 + ∆23 /∆
µ ¶
1 45 21
a1 = − + + 80 → 1, (5.19)
56 1 + ∆24 /∆ 1 + ∆23 /∆
µ ¶
1 5 21
a2 = − + 16 → 0,
112 1 + ∆24 /∆ 1 + ∆23 /∆
where we just have chosen to denote the detunings ∆F 0 as ∆2 = ∆, ∆3 = ∆+∆23 ,

and ∆4 = ∆ + ∆24 . Note, red detuning still corresponds to positive ∆, and the
limits for ∆ → ±∞ are the same as in (5.16) for F = 4.
Concluding this chapter we remind ourselves of the approximations of the
Hamiltonian (5.18). We assumed off-resonant interactions, i.e. there are no ab-
sorption effects in our description. This led us to the adiabatic elimination which
is valid if optical beams are far from saturating the optical transition. We made
no specific assumptions about the spin state of atoms or the polarization state of
light, and therefore the Hamiltonian is in general a good starting point for many
calculations involving what is essentially the Stark effect.
41
CHAPTER 6
Propagation Equations
In Chap. 5 we derived an effective Hamiltonian for the off-resonant interaction

of polarized laser light with an atomic spin ensemble. In this chapter we will
take these calculations one step further to derive actual equations of motion for
our interesting spin state operators Jˆy , Jˆz and Stokes operators Ŝy , Ŝz . We start
out by deriving propagation equations in general and we will learn that these are
in fact quite complicated. To couple collective spin operators to collective light
operators we need to perform the approximation that the higher order terms
proportional to a2 of the Hamiltonian (5.18) can be neglected. Doing this we
arrive at the most important equations of this chapter, Eqs. (6.11-6.15). We will
then estimate the role of the higher order terms for our experimental purposes.
Like the previous chapter, the derivations are somewhat technical. To understand
the experiments of this thesis the results of Secs. 6.2 and 6.3 are important.
6.1 General Propagation Equations

We shall be concerned with the spin operators ̂x (z, t), ̂y (z, t), and ̂z (z, t) and
the Stokes operators Ŝx (z, t), Ŝy (z, t), and Ŝz (z, t) where we continue the nota-
tion of Chap. 5. For hthe spin operators
i we state the Heisenberg equations (e.g.
∂̂z (z, t)/∂t = 1/ih̄ · ̂z (z, t), Ĥ ) where we for a start take Ĥ to be the inter-
action Hamiltonian (5.18). Possible magnetic fields acting on the spin operators
can be added later. With help from the commutation rules of App. B.3 and
42 Chapter 6 - Propagation Equations
Eq. (C.18) we get

∂ cγ λ2 n
̂x (z, t) = a1 Ŝz ̂y (6.1)
∂t 4A∆ 2π³ ó
+ a2 2Ŝy [̂x ̂z + ̂z ̂x ] − (2Ŝx − φ̂)[̂z ̂y + ̂y ̂z ] ,
∂ cγ λ2 n
̂y (z, t) = −a1 Ŝz ̂x (6.2)
∂t 4A∆ 2π³ ó
+ a2 −(2Ŝx + φ̂)[̂x ̂z + ̂z ̂x ] − 2Ŝy [̂z ̂y + ̂y ̂z ] ,
∂ cγ λ2 n o
̂z (z, t) = a2 4Ŝx [̂x ̂y + ̂y ̂x ] − 4Ŝy [̂2x − ̂2y ] , (6.3)
∂t 4A∆ 2π
where we for brevity have left out the (z, t)-notation on all operators on the right
hand side. For the Stokes variables we can easily generalize Eq. (C.8) to be
directly applicable to Stoke operators
µ ¶
∂ ∂ 1 h i
+c Ŝ(z, t) = Ŝ(z, t), Ĥint . (6.4)
∂t ∂z ih̄
Furthermore, we may in the following neglect effects of retardation, that is we
assume the speed of light c is infinite. Since we in Chap. 5 already restricted
ourselves to describing dynamics on a long time scale by deriving an effective
Hamiltonian this does not impose strong restrictions. Neglecting retardation is
mathematically equivalent to leave out the ∂/∂t term above. Then we get
∂ γρ λ2 n o
Ŝx (z, t) = a1 Ŝy ̂z + a2 · 2Ŝz [̂x ̂y + ̂y ̂x ] , (6.5)
∂z 4∆ 2π
∂ γρ λ2 n o
Ŝy (z, t) = −a1 Ŝx ̂z − a2 · 2Ŝz [̂2x − ̂2y ] , (6.6)
∂z 4∆ 2π
∂ γρ λ2 n o
Ŝz (z, t) = a2 2Ŝy [̂2x − ̂2y ] − 2Ŝx [̂x ̂y + ̂y ̂x ] , (6.7)
∂z 4∆ 2π
where we again leave out the (z, t)-notation on the right hand side. Equa-
tions (6.1-6.3) and (6.5-6.7) are coupled to each other, and they are not even
closed. On the right hand side we see spin operators like e.g. ̂x ̂y + ̂y ̂x which
again from Heisenberg’s equations will get its own time evolution, and so on.
From here we can go into many different directions depending on the actual
physical system under consideration. For the rest of this chapter we special-
ize into different relevant cases, and we will also in these connections give some
physical interpretations to the equations above.
6.2 Probing a Macroscopic Ensemble of Oriented

Spins
A simple and useful tool for characterizing an ensemble of oriented atomic spins
is the Faraday rotation of a linearly polarized laser beam propagating along
6.3. Probing Transverse Spin Components 43
the direction of atomic spin orientation. We assume that a sample of spins are
oriented along the direction of light propagation, i.e. the z-axis, and we will
neglect quantum fluctuations for the moment. In this case the only spin operator
with non-zero mean value in the equations of Sec. 6.1 is ̂z in the propagation
equations for Ŝx and Ŝy . If a beam of light traverses a sample of atoms of length
L we may show that the classical values of the Stokes operators evolve as
Sxout = Sxin cos(2θF ) − Syin sin(2θF ),

(6.8)
Syout = Sxin sin(2θF ) + Syin cos(2θF ),
where the angle θF is given by (in radians)
a1 γλ2 ρL
θF = − · h̂z i . (6.9)
16π∆
If a linearly polarized beam of light is rotated by the angle θ, the Stokes vector
will be rotated 2θ. Thus, in the above, θF is the polarization rotation caused
by the spin orientation along the direction of light propagation. The equations
are valid for both F = 3 and F = 4 in the cesium ground state, where a1 ≈ 1
defined in Eq. (5.19) or (5.16) is depending on F . γ is the FWHM line width of
the excited 6P3/2 state, λ is the optical wave length of the transition, ρ is the
atomic density, and ∆ is the optical detuning (red being positive), L is the sample
length, and h̂z i is the expectation value of the total angular momentum along
the direction of light propagation of a single atom in the sample. We will see in
Chap. 7 that this polarization rotation is a very useful tool for characterizing the
spin state of an atomic ensemble.
6.3 Probing Transverse Spin Components

The most essential physical setting in this thesis is the situation of a macroscopic,
oriented sample along the x-direction with an off-resonant probe propagating in
the z-direction. In this case the probe measures a transverse spin component
which is interesting to us at the level of quantum fluctuations, as mentioned in
the introductory Chap. 2.
In this section we neglect the higher order effects of the atom/light interaction,
i.e. we assume a2 = 0 in the equations of Sec. 6.1. We have chosen the x-axis to
coincide with the direction of spin orientation, and we shall also assume a linearly
polarized probe along the x-direction (which actually is not strictly important,
we could have chosen any direction). The interaction will practically be such that
the state of light and the state of atoms do not deviate much from this situation,
and we may describe the x-components of the spin j and the Stokes operator S
by constant c-numbers, i.e. Ŝx → Sx and ̂x → jx . Under these assumptions we
have a zero on the right hand side of Eqs. (6.1, 6.3, 6.5, 6.7). Equations (6.2)
and (6.6) yield
∂ cγ λ2
̂y (z, t) = − · a1 Ŝz (z, t)jx ,
∂t 4A∆ 2π (6.10)
∂ γρ λ2
Ŝy (z, t) = − a1 Sx ̂z (z, t).
∂z 4∆ 2π
Since we here have jx and Sx constant in the whole sample of atoms we can easily
integrate over z to get equations for the collective properties of the sample. We
RL
define the collective spin variable (with capital letter J) Jˆx = 0 ̂x (z, t)ρAdz
and so on for y, z-components. We also note that with ∂ Ŝz (z, t)/∂z = 0 we may
write c · Ŝz (z, t) = Ŝzin (t) where the latter is the Stokes vector Ŝz at the beginning
of the sample normalized to photons per second. Summarizing the above we get
equations

Ŝzout (t) = Ŝzin (t), (6.12)
∂ ˆ
Jy (t) = aJx Ŝzin (t), (6.13)
∂t
∂ ˆ
Jz (t) = 0, (6.14)
∂t
2
γ λ
a = − a1 . (6.15)
4A∆ 2π
The Stokes operator Sx is classical, and Ŝy , Ŝz are the quantum variables of the
light polarization state normalized to photons per second. For atoms with J x
being classical we have quantum variables Jˆy and Jˆz normalized such that they
describe the total spin of all atoms in the sample, i.e. Jx is of the order of the
number of atoms in the sample. The interaction parameter a depends on the
FWHM line width γ of the excited state 6P3/2 , the optical wave length λ, the
detuning ∆ (red being positive), the beam cross section A, and the parameter
a1 ≈ 1 defined in Eq. (5.19) or (5.16). Physically, a is the rotation angle of
the macroscopic spin J around the z-axis per circularly polarized photon, or the
rotation angle in Stokes vector space of S around the z-axis per unit of angular
momentum along the z-axis. The above equations are the cornerstone for all
experiments in this thesis. We remark that they arise from the first order terms
proportional to a1 in the equations of Sec. 6.1. The term aSx Jˆz (t) in Eq. (6.11)
enables us to read out properties of the spin state to the light. At the same time
the term aJx Ŝzin (t) of Eq. (6.13) feeds noise back to the spins. We often call the
latter term for the back action term.
6.4. Inclusion of Higher Order Terms 45
6.4 Inclusion of Higher Order Terms

Now, let us turn to the higher order terms proportional to a2 of the equation
in Sec. 6.1. For simplicity, we will assume that the atoms are oriented almost
perfectly along the x-axis. This is in general the case for our experiments. We
start out by an analysis of the classical mean values. In this case the only non-zero
spin operators® on the right hand side of the equations in Sec. 6.1 are h̂x i ≈ ±F
and ̂2x − ̂2y ≈ F (F − 21 ), see Eq. (B.15). The effect of the ̂2x − ̂2y terms can be
understood by considering the propagation√of a beam of light which is initially
linearly polarized along e+45 = (ex + ey )/ 2. Then the most important terms
of Eqs. (6.6) and (6.7) can be written
∂ ∂
Ŝy (z, t) = +k Ŝz (z, t) and Ŝz (z, t) = −k Ŝy (z, t)
∂z ∂z
(6.16)
a2 γρ λ2
with k = − F (2F − 1),
4∆ 2π
where Eq. (B.15) was used to evaluate ̂2x − ̂2y . The solution of these equations
is simply
Ŝy (z, t) = +Ŝy (0, t) cos(kz) + Ŝz (0, t) sin(kz),

(6.17)
Ŝz (z, t) = −Ŝy (0, t) sin(kz) + Ŝz (0, t) cos(kz).
This is nothing more than the birefringence of the atomic sample which is ori-
ented along the x-axis (with the x- and y-axes as major axes). The difference
in phase shift experienced by x- and y-photons turns linear polarization into cir-
cular polarization and vice versa. This is a complication to the simple physical
setup described in Sec. 2.3 where a strong linearly polarized beam of light passes
through the atomic sample with constant Ŝz -component and the Ŝy -component
only reading out the spin component Jˆz . But there is more to this, the term
Ŝy (̂2x − ̂2y ) in Eq. (6.3) will change the mean value of ̂z . This is just another
way of stating, that if Ŝz is subject to changes these will also affect ̂z since
the projection of the total angular momentum along z must be conserved. Also,
in Eq. (6.17) if Ŝz in the propagation builds up a considerable non-zero mean
value the terms proportional to Ŝz in Eqs. (6.1) and (6.2) will start rotating the
macroscopic spin around the z-axis. This again affects ̂2x − ̂2y which started it all.
We see the complicated structure of the interaction now and we really wish to
minimize these effects. To characterize the strength of these effects the relevant
parameter is kL where k is given in Eq. (6.16) and L = 3.0cm is the length of
our atomic samples. For F = 4 we can conveniently relate kL to θF defined in
Eq. (6.9) by
7π a2
kL = θF [deg]. (6.18)
90 a1
For a typical detuning ∆ = −1GHz and a corresponding typical large value of
θF ≈ 30deg we get kL ≈ 7% which we have to keep in mind. kL will decrease with
detuning and with atomic density, it is an atomic property and is independent

on the power of the laser.
The term ̂2x − ̂2y describes alignment in the xy-basis. In the equations of
Sec. 6.1 there are also spin terms ̂x ̂y + ̂y ̂x which in a completely analogous
way describe the alignment along the directions rotated by ±45◦ in the xy-plane.
These will have non-zero mean if we choose to orient atoms along either of the
±45◦ -directions instead of the x-direction.
To sum up so far, for mean values we understand all terms on the right hand
side of Eqs. (6.5-6.7), terms including a1 give rise to the Faraday effect caused
by circular birefringence. The terms including a2 are responsible for the linear
birefringence.
But equations are also valid for fluctuations, quantum or classical. For our
experiments we should keep in mind that with our constant bias magnetic field
and the rotating frame, see Sec. 4.3, we should concentrate on frequency compo-
nents around the Larmor frequency Ω for what regards ̂y , ̂z , Ŝy , and Ŝz . For
instance, since ̂2x − ̂2y ≈ F (F − 12 ) primarily has a DC component, it is the AC
components of Ŝy and Ŝz which couple to each other in Eqs. (6.6) and (6.7).
Similarly, from the fact that Eq. (B.13) consists of first off-diagonal elements,
we know that ̂x ̂y + ̂y ̂x primarily has frequency components at Ω. Then it is
the DC component of Ŝx and AC component of Ŝy that contribute to the time
evolution of ̂z in Eq. (6.3).
Taking these considerations into account we conclude that DC terms of the
time evolution of ̂x are small. Thus for a sample oriented along the x-direction we
have a pretty stable system. What regards Eq. (6.2) the spin operator ̂y ̂z + ̂z ̂y
is small and has frequency components at 2Ω. This must couple to frequency
components at Ω or 3Ω of Ŝy for the product to contribute at Ω in the time
evolution of ̂y . AC components of Ŝy are small independent of the direction of
probe polarization, so we neglect the last term of Eq. (6.2).
Now, turn to the last term of Eq. (6.3). Since ̂2x − ̂2y is a DC term, the
AC components of Ŝy at Ω will feed into ̂z . This term can have a considerable
magnitude, we wish to compare it to our favorite term which is the a1 Ŝz ̂x back
action term of Eq. (6.2). When we measure noise properties we have to square
the fluctuations, and the correct comparison between the unwanted pile up in ̂ z
and the wanted back action noise (BAN) in ̂y is
µ ¶2
Bad pile up a2 Noise(Ŝy )
= 4(2F − 1)2 . (6.19)
BAN a1 Noise(Ŝz )
For our typical values of detuning we have a2 /a1 ≈ 10−2 and the above tells us
that (Bad pile up)/BAN ≈ 0.02 · Noise(Ŝy )/Noise(Ŝz ) for F = 4. If our laser
beam is polarized along the x- or y-axis with a clean linear polarization, the
noise of Ŝy and Ŝz at the frequency Ω will most likely both be limited by shot
noise, i.e. by quantum noise (amplitude noise of the laser does not feed into Ŝy
and Ŝz in the case of clean linear polarization). In this case the unwanted noise
6.4. Inclusion of Higher Order Terms 47
only contributes ≈ 2% of the total noise pile up. But if we choose arbitrary
polarization directions in the xy-plane, the Ŝy -component will have non-zero
mean value, and the fluctuations at Ω will essentially be the amplitude noise of
the laser at Ω. In this case, to keep the last term of Eq. (6.3) from piling up
extra noise requires the laser intensity to be shot noise limited at Ω (which is a
more difficult condition to meet than clean linear polarization). We thus have
one motivation for choosing the laser to be polarized along the x- or y-direction
and not in between.
Now, let us turn to the ̂x ̂z + ̂z ̂x term of Eq. (6.2) and the ̂x ̂y + ̂y ̂x term
of Eq. (6.3). These spin operators have frequency components at Ω and we must
then consider the DC components of (2Ŝx + φ̂) and 4Ŝx , respectively. In the
following we show, that the effect of these terms change the Larmor frequency Ω
by a small amount, and we wish to calculate this for different directions of the
linearly polarized laser.
To this end, assume that the laser has photon flux φ(t) and is polarized along
e1 = cos(α)ex + sin(α)ey . Then we have mean values
D E D E
c(2Ŝx (z, t) + φ̂(z, t)) = (cos(2α)+1)φ(t) and 4cŜx (z, t) = 2 cos(2α)φ(t).
(6.20)
Furthermore, since we assume a strong orientation along the x-axis we may re-
late the spin operators ̂x ̂y + ̂y ̂x and ̂x ̂z + ̂z ̂x to ̂y and ̂z by Eqs. (B.13)
and (B.14). After some algebra we deduce that the time evolution of ̂y and ̂z
become
µ ¶
∂ cos(2α) + 1
̂y (z, t) = − Ω + ΩS ̂z (z, t) + . . . , (6.21)
∂t 2
µ ¶
∂ 2 cos(2α)
̂z (z, t) = + Ω + ΩS ̂y (z, t) + . . . , (6.22)
∂t 2
where the dots remind us that there are more terms in Eqs. (6.2) and (6.3) which
we leave out for brevity. Above Ω is the magnetic field contribution and Ω S is a
Stark induced contribution normalized such that ΩS is the extra contribution for
α = 0, i.e. for light polarization parallel to the spin orientation along the x-axis.
We have ΩS given by (in Hertz)
γλ2 a2
ΩS [Hz] = · φ(t) · 2(2F − 1)σ jx . (6.23)
16π 2 A∆
Now, if α 6= 0 the parentheses in Eqs. (6.21) and (6.22) are unequal, but for
Ω À ΩS we may easily show that the effective Larmor frequency becomes the
average of the two parentheses. This amounts to
γλ2 a2 (2F − 1)σ jx
Ωeff [Hz] = Ω[Hz] + · φ(t) · (1 + 3 cos(2α)) · , (6.24)
16π 2 A∆ 2
where σ jx = ±1 for h̂x i = ±F . We see the Stark contribution to the Larmor
frequency acts in opposite directions for oppositely oriented spin samples. In-
serting typical values for the detuning ∆ = 1GHz and for the laser intensity of
1mW/cm2 we obtain a shift in Larmor frequency of 160Hz for α = 0. For two

oppositely oriented spin samples the difference in Larmor frequency under these
conditions is 320Hz. This can be a problem, but the shift is reduced to zero
if α = 54.7◦ . Thus we have a motivation for choosing polarization directions
different from the x- or y-directions. But this opposes our desire to be polarized
exactly along the x- or y-direction where pile up of laser noise is small according
to Eq. (6.19). We shall study this Stark shift experimentally in Sec. 7.6 where
we also present an alternative calculation of the change in Larmor frequency.
To conclude this section, we cannot easily solve analytically the equations of
motion with the higher order terms proportional to a2 included. The collective
variables loose their meaning in this case. The most important effects include the
mixing (6.17) of Ŝy and Ŝz by linear birefringence, the possible pile up of laser
noise discussed around Eq. (6.19), and the shift in Larmor frequency discussed
in Eq. (6.24). The Ŝy , Ŝz mixing has strength parametrized by kL ≈ 7% in the
typical case. The laser noise pile up will probably be a few percent for x- or
y-polarized probe beam. The effects of the Stark shifted Larmor frequency will
be present for x- or y-polarization and absent for α = 54.7◦ .
We also repeat the fact that Eqs. (6.1-6.3) and (6.5-6.7) are not closed, but for
a well oriented sample with h̂x i ≈ ±F and with the approximations in Eqs. (B.12-
B.14) we do have a closed set of linearly coupled equations. Even in this case an
analytical solution will be very cumbersome. We shall not pursue any solution
in this thesis.
49
CHAPTER 7
Atomic State Characterization
In this chapter we develop methods for the characterization of the atomic spin
states that occur in different contexts in our experiments. We have several mo-
tivations for this; we would like to know the number of atoms in our sample
(especially in order to check scalings of noise as we will discuss in Sec. 9.5), know
the degree of orientation to tell whether we are in the coherent spin state or close
to it, measure the decoherence time of the state for estimating the life time of
interesting quantum states. We are able to address all the above questions. The
contents of this chapter are published in [VIII].
We start out with some notation. We will consider a sample with N atoms in
one hyperfine ground state F of cesium and describe the spin state with density
operators σ̂ij given by
N N
1 X (k) 1 X
σ̂ij = σ̂ij = |iik hj|k , (7.1)
N N
k=1 k=1
where i, j = −F, −F +1, . . . , F , the sum is done over all individual atoms and |ji k
refer to the magnetic sub-level with mF = j of the k’th atom. With the x-axis
as quantization axis we may express the total macroscopic angular momentum J
50 Chapter 7 - Atomic State Characterization
(a) (b)
Bbias 20
BRF
λ/2 - 10
x
E/h [GHz]
z
0
r Atomic
Probe lase sample
y -10
D E D E
i(t) ∝ Ŝyout (t) ∝ Jˆz (t) -20
0.0 0.2 0.4 0.6 0.8 1.0
B [Tesla]
Figure 7.1: (a) The setup for magneto-optical resonance. A constant magnetic
field Bbias is applied parallel to the atomic spin orientation along the x-axis. An
RF-magnetic field is applied along the y-axis, and the Jˆz component of the spin
is measured by a probe laser propagating along z. Magnetic resonance effects
are read out optically by the probe in the photo current i(t). (b) The energy
levels of the magnetic sub-levels of the F = 3 and F = 4 ground states in cesium
according to Eq. (F.2). We operate at Bbias ≈ 1Gauss which is far into the lower
linear regime where quadratic effects are small.
of the atoms in the hyperfine state F as
Jˆx = N
X
mσ̂m,m , (7.2a)
m
F −1
C(F, m)
Jˆy = N
X
{σ̂m+1,m + σ̂m,m+1 } , (7.2b)
2
m=−F
F −1
C(F, m)
Jˆz = N
X
{σ̂m+1,m − σ̂m,m+1 } , (7.2c)
2i
m=−F
p
where C(F, m) = F (F + 1) − m(m + 1), see Eqs. (B.9-B.11). In addition to
the number of atoms N we see that the macroscopic spin Jˆx is described by the
diagonal terms σ̂m,m and the quantum variables Jˆy and Jˆz are described by the
first off-diagonal terms σ̂m,m+1 . We need to characterize each of the relevant
density operators in detail in terms of magnitude and decoherence time.
7.1 The Magneto-Optical Resonance Method

We will use the so-called magneto-optical resonance to investigate the spin state
of atoms, the basic setup is shown in Fig. 7.1(a). Atoms are placed in an ex-
ternal, constant magnetic field Bbias as discussed in Sec. 4.3. Applying another
external radio frequency (RF) magnetic field we may induce transitions between
the magnetic sub-levels if the RF is in resonance with the level splitting. More
7.1. The Magneto-Optical Resonance Method 51
exactly, as we will derive below, the RF-magnetic field couples to and drives the
coherences σ̂m,m+1 and σ̂m+1,m . According to Eq. (7.2c) this will lead to a mod-
ulation of the spin component Jˆz . This modulation can be read out optically
by a probe laser, i.e. the photo current i(t) of the detector setup
D Eof Fig. 7.1(a),
which measures Ŝy , will be proportional to the mean value Jˆz according to
Eq. (6.11).
The energy of the magnetic sub-levels of an atom in an external magnetic
field B can be calculated and is given without approximation in Eq. (F.2). These
levels are shown in Fig. 7.1(b). We see that for low magnetic fields the energy
dependence on the magnetic field strength is linear with small corrections caused
by the presence of the hyperfine splitting. In our experiments the magnetic field
strength Bbias is around 1 Gauss which means that non-linear corrections are very
small. We shall still include them to second order, which gives the Hamiltonian
for a single atom with total angular momentum j
Ĥ = gF µB j · B + O(B 2 ), (7.3)
where µB is the Bohr magneton and the gF -factor is defined in Eqs. (F.4)
and (F.5). The bias magnetic field along the x-axis contributes gF µB Jx Bbias +
2
O(Bbias ) to the Hamiltonian while the RF-magnetic field directed along the y-axis
contributes gF µB Jˆy |BRF | cos(ωt + φ) where the RF-magnetic field is character-
ized by the amplitude |BRF |, frequency ω, and phase φ. Retaining only the linear
term is sufficient here. The Hamiltonian may now be written
F
X
Ĥ = h̄ωm · σ̂mm
m=−F (7.4)
g F µB ¡
̂+ BRF e−iωt + ̂− BRF
∗
eiωt ,
¢
+
4
where ̂± = ̂y ± i̂z and BRF = |BRF |e−iφ is the complex amplitude. The
first term is the bias magnetic field contribution where we take the second order
corrections into account by explicitly stating the energy levels h̄ωm of the m’th
sub-level. The second order correction is calculated in Eq. (F.7) and will be
discussed in more detail in the next section. The second term is the RF-magnetic
field contribution in the rotating wave approximation. We may wish to write the
Hamiltonian entirely in terms of the density operators σ̂ij :
F
X
Ĥ = h̄ωm · σ̂mm
m=−F
(7.5)
F −1
g F µB X ¡
C(F, m)σ̂m+1,m BRF eiωt + h.c.
¢
+
4
m=−F
which follows directly from Eqs. (7.2b) and (7.2c). The equations of motion are
now determined by
∂ σ̂ij 1 h i
= σ̂ij , Ĥ + decay terms, (7.6)
∂t ih̄
where the first term is the coherent evolution of the system, and the interaction
with the environment will be put in by hand as decay terms.
We will now solve Eqs. (7.5) and (7.6), and to illuminate the method for solv-
ing these equations, we will pick out a single example and work it out thoroughly.
The time derivative of e.g. σ̂12 is
∂ σ̂12 1 h i
= σ̂12 , Ĥ − Γ/2 · σ̂12
∂t ih̄
= −i(ω2 − ω1 )σ̂12 − Γ/2 · σ̂12
(7.7)
igF µB ©
+ C(F, 1)[σ̂22 − σ̂11 ]BRF e−iωt
4h̄
∗
eiωt ,
ª
+ [C(F, 0)σ̂02 − C(F, 2)σ̂13 ]BRF
where we have just inserted the Hamiltonian (7.5) into (7.6) and added the decay
term, −Γ/2 · σ̂12 . We will restrict ourselves to a description of spins in the case
where Jˆy , Jˆz ¿ Jx , i.e. the angle θ that the spins deviate from being oriented
along the z-axis is much less than unity. From Eqs. (7.2a-7.2c) the order of
magnitude can roughly be written as O(σ̂m+1,m ) ≈ θ ·O(σ̂m,m ), and following the
same lines O(σ̂m+2,m ) ≈ θ 2 ·O(σ̂m,m ). It is then justified to neglect the coherences
σ̂02 and σ̂13 in the above equation. For brevity we will define ω21 = ω2 −ω1 , which
is the frequency corresponding to the transition from mF = 2 to mF = 1. This
frequency is the Larmor frequency which typically has values around 325kHz.
This is fast compared to the inverse time scale for dynamical evolution of the
spin state which never exceeds 1kHz. Since the RF-magnetic field frequency ω
will be in the vicinity of ω21 it will be convenient to define the slowly varying
operators
σ̂ij = σ̃ij e−iωt . (7.8)
Using this definition, Eq. (7.7) will turn into
∂ σ̃12
= (i[ω − ω21 ] − Γ/2)σ̃12
∂t (7.9)
igF µB
+ C(F, 1)BRF [σ̂22 − σ̂11 ].
4h̄
The constant Γ will describe the decay of the transverse spin components. With
the small angle assumption θ ¿ 1 the population difference (σ̂22 − σ̂11 ) will not be
affected by the RF-magnetic field. In addition, the typical life time of populations
is T1 ≈ 200 − 300ms and we can safely assume (σ̂22 − σ̂11 ) to be constant. Any
decays of populations can even be compensated by external pumping processes.
With an external pumping process we may let transients decay (takes time 1/Γ)
7.1. The Magneto-Optical Resonance Method 53
and we are left with the steady state solution for the operator σ̃12 . By setting
∂ σ̃12 /∂t = 0 in the above equation we get
igF µB BRF C(F, 1)e−iωt

σ̂12 = − [σ̂22 − σ̂11 ]. (7.10)
4h̄ · (i[ω − ω21 ] − Γ/2)
This method applies to all density operators σ̂m,m+1 , and substituting into
Eqs. (7.2b) and (7.2c) we get
F
( −1
)
iωt
ig F µ B B RF N [F (F + 1) − m(m + 1)] · e
Jˆy = Re
X
[σ̂m+1,m+1 − σ̂m,m ] ,
4h̄ i(ωm+1,m − ω) − Γm+1,m /2
m=−F
(7.11a)
F
( −1
)
igF µB BRF N [F (F + 1) − m(m + 1)] · eiωt
Jˆz = Im
X
[σ̂m+1,m+1 − σ̂m,m ] .
4h̄ i(ωm+1,m − ω) − Γm+1,m /2
m=−F
(7.11b)
These equations can be interpreted as a number (2F ) of two-level systems that

all interact with the driving RF-magnetic field. Two adjacent magnetic sub-
levels m and m + 1 act as one two-level atom with the usual Lorentzian response
(resonance frequency ωm+1,m and line width Γm+1,m FWHM). Each two-level
system does not respond with exactly the same weight which is reflected in the
factor F (F + 1) − m(m + 1). All the resonances add up coherently to give the
overall response of the spin state to the RF-magnetic field. Note, that Jˆy and Jˆz
oscillate at the driving frequency ω and not the “natural” frequencies ωm+1,m .
This is the steady state behavior with damped transients. In Sec. 7.5 we will
comment on non-steady state behavior of the spin system.
We conclude this section with some comments which are relevant for our
particular experimental setup. We may D write
E the photo current of the detector
ˆ
setup shown in Fig. 7.1(a) as i(t) = α · Jz = α · Im{A(t)} where α is a constant
depending on experimental parameters, and A(t) reflects the mean value of the
curly bracket of Eq. (7.11b). We know from this equation that A(t) ≡ A(ω)eiωt
will posses only a single frequency component, namely the local oscillator fre-
quency ω driving the transverse spins Jˆy and Jˆz away from zero. The amplitude
of this frequency component is experimentally measured by inserting the photo
current i(t) into a lock-in amplifier and decomposing the signal into sine and
cosine components:
i(t) = α · Im{A(ω)eiωt }
= α · (Re{A(ω)} sin(ωt) + Im{A(ω)} cos(ωt)) . (7.12)
We set the lock-amplifier to give the sum of the squared amplitudes of the sine
and cosine components which in our case will be exactly α2 |A(ω)|2 . We shall call
this signal our magneto-optical resonance signal at frequency ω (MORS(ω) in
short). Combining the above with Eq. (7.11b) and ignoring irrelevant constants
we find
¯ F −1 ¯2
¯ X [F (F + 1) − m(m + 1)] ¯
MORS(ω) = const · ¯N hσ̂m+1,m+1 − σ̂m,m i¯ .
¯ ¯
¯ i(ωm+1,m − ω) − Γm+1,m /2 ¯
m=−F
(7.13)
We see that the signal depends on the number of atoms N in the relevant hyper-
fine state F , the resonance frequencies ωm+1,m , the transverse spin decay rates
Γm+1,m , and the populations σ̂m,m . All these parameters are of importance for
us.
7.2 Spin State Modeling

In the previous section we derived how the spin J responds to an external RF-
magnetic field, our motivation is to use this knowledge to characterize the spin
state, i.e. we wish to deduce parameters like σ̂m,m , Γm+1,m and so on. Now, for
cesium in the e.g. F = 4 hyperfine ground state there are nine populations σ̂ m,m
and eight line widths Γm+1,m together with the resonance frequencies. To fit an
experimentally measured spectrum (see e.g. Fig. 7.2(b)) to all these parameters
will be very hard and in the following we will develop a model to significantly
reduce the number of free parameters. We will just tailor a model and the
justification for this model will be an experimental test.
Let us consider a case where we wish to orient all atomic spins along the
x-direction, i.e. attempt to put many atoms into the m = F sub-state. This can
be done experimentally by illuminating the atoms with circularly polarized light,
as was described in Sec. 4.5. It is then convenient to define the orientation p as
an order of merit
F
1 X Jx
p= m · hσ̂m,m i = . (7.14)
F NF
−F
Note, that with this definition p = 1 if all atoms are in the extreme m = F sub-
level, and p = 0 for a completely unpolarized sample with σ̂m,m = 1/(2F + 1)
for all m. We try to let the orientation p be the only parameter describing
P the
relationship between the nine populations σ̂m,m . With the condition σ̂m,m = 1
we have thus reduced eight free parameters to a single one.
Now, we describe ensembles of atoms and given p we will assume that the
spin state maximizes the entropy Ŝ = −Tr(σ̂ ln σ̂). To find the individual σ̂m,m
we use the method of Lagrange multipliers. We must solve
∂ ³ X X ´
S−α hσ̂m,m i − β m · hσ̂m,m i = 0
∂ hσ̂m,m i
⇒ hσ̂m,m i = e−1−α · e−β·m . (7.15)
7.3. Experimental Test of the Modeling 55
We now just need to adjust α and β in order that Tr(σ̂) = 1 and p is as desired.
Doing this is more or less a computational problem and in principle not difficult.
For the physical understanding we just need to remember that we can write
hσ̂m,m i = hσ̂F,F i ²F −m where ² = eβ is a parameter which is a function of p. This
can go directly into Eq. (7.13).
For the eight line widths Γm+1,m in the case of cesium we will make a model
with two free parameters. First, a common line width Γcom is assigned to all
transitions independent of m. The physical cause for this type of decay could
be magnetic field inhomogeneities, collisions, and loss mechanisms common to
all atoms. In addition, if we wish to create a well oriented sample with m ap-
proaching F we will need to illuminate the atoms with resonant circularly po-
larized light. In our case this is the 894nm 6S1/2 , F = 4 to 6P1/2 , F 0 = 4 line.
This light causes excitations from the atomic ground sub-level m with a rate
γm ∝ | h4, m, 1, 1|4, m + 1i |2 = (4 − m)(5 + m)/40, where the second term is the
square of Clebsch-Gordan coefficients. For a magnetic transition between ground
sub-levels m and m + 1 the resonant pumping light will contribute to the line
broadening proportional to γm + γm+1 . Thus we define the width Γpump caused
by the optical pumping process such that
19 − 2m − m2
Γm+1,m = Γcom + Γpump , (7.16)
4
where the normalization is such that for the m = 3 → m = 4 transition we have
Γ4,3 = Γcom + Γpump .
Finally, we must have the resonant frequencies as parameters in our model.
We will write this as a central frequency ωcenter and a splitting ωsplit such that
µ ¶
1
ωm+1,m = ωcenter + ωsplit m + . (7.17)
2
From the quadratic Zeeman effect we should have ωsplit = 2π · νQZ (see Eq. (F.7))
but we keep it as a free parameter since in practice this splitting will also depend
on the Stark shifts by the probe laser, see Sec. 7.6.
To sum up, a possible description of the ground spin state involves the total
spin Jx and the orientation p together with the line widths Γcom and Γpump , and
the frequencies ωcenter and ωsplit . An equivalent but computationally easier way
to represent Jx and p is to use the number of atoms NF = N hσ̂F,F i of atoms in
m = F as one parameter and the parameter ² such that the population Nm can
be expressed as Nm = N hσ̂m,m i = NF ²(F −m) .
7.3 Experimental Test of the Modeling

Let us make an experimental test of the magneto-optical resonance method and
the models discussed in Sec. 7.2. To this end we setup our lasers as shown in
Fig. 7.2(a). In the following all lasers run cw. The optical pump and repump
(a) D2 (b)
0.0020 Γcom = 9.4Hz
D1
Γpump = 0.0Hz
x Atomic 0.0015 Jx = 0.122[a.u.]
MORS(ω) [a.u.]
λ/2 sample
p = 0.346
y z 0.0010
BRF
λ/2 0.0005
Bias
field 0.0000
Optical pump and 325000 325100 325200 325300 325400

Probe laser σ+ repump lasers i(t) Frequency [Hz]
Figure 7.2: (a) The experimental setup for testing the magneto-optical resonance
method and the spin state modeling. In addition to the situation discussed in
Fig. 7.1(a) another linearly polarized laser (the DC-Faraday probe) is propagating
parallel to the macroscopic spin polarization created by optical pump lasers along
the x-axis. The MORS signal is recorded in the photo current i(t) while the
detectors D1 and D2 can tell about the polarization rotation θDC of the DC-
Faraday probe. (b) An experimental spectrum (dots) of magnetic transitions
among the nine sub-levels of the F = 4 ground state in cesium. The solid
line is a fit to Eq. (7.13) using the model described in Sec. 7.2. The many
peaks tells us that atoms are distributed among all nine levels resulting in a low
orientation p = 0.346. The line width 9.4Hz is a direct measure of the decay
rate of spin coherence. According to Eq. (F.5) the corresponding F = 3 signal is
approximately 1kHz away and does not interfere here.
lasers are tuned as described in Sec. 4.1 and their polarization can be adjusted
at will. The probe laser is split into two parts, one measuring the transverse spin
component Jˆz along the z-axis and the other one (called the DC-Faraday probe)
measuring the longitudinal, macroscopic spin component Jx along the x-axis.
The detuning of the probe laser is ∆ = −1GHz.
The rotation θDC of the DC-Faraday probe is described by Eq. (6.9) and is
directly proportional to the macroscopic spin (7.2a). The probe measuring Jˆz will
lead to the MORS signal defined in Eq. (7.13). An example of a spectrum is shown
in Fig. 7.2(b). In this case the optical pump laser is off and the repump laser
has elliptical polarization which is relatively far from being circular. Here we see
that there are eight peaks corresponding to the eight possible transitions between
adjacent levels among the nine magnetic sub-levels in the F = 4 hyperfine state.
The dots are experimental points and the solid line is a fit to the model (7.13)
with Jx , p, Γcom , ωcenter , and ωsplit as free parameters. Γpump is set to zero (since
the optical pump laser is off). We see that the solid line matches the experimental
points very well. Note, that p = 0.346 is the only parameter really describing the
relative strength of the individual peaks, while the other parameters are common
to all peaks. This gives strong support to the model described by Eq. (7.15) with
hσ̂m,m i = hσ̂F,F i ²F −m . With ω = 2πν (experimentally we prefer Hertz and not
7.3. Experimental Test of the Modeling 57
(a) 0.040 (b) 0.10

Γcom = 9.4Hz Γcom = 9.4Hz
0.035
Γpump = 0.0Hz 0.08 Γpump = 5.5Hz
0.030
Mors(ω) [a.u.] Jx = 0.248[a.u.] Jx = 0.336[a.u.]
Mors(ω) [a.u.]
0.025 p = 0.823 p = 0.967
0.06
0.020
0.015 0.04
0.010
0.02
0.005
0.000 0.00
-0.005
325000 325100 325200 325300 325400 325000 325100 325200 325300 325400
Frequency [Hz] Frequency [Hz]
Figure 7.3: Two examples of experimental (dots) and fitted (solid line) traces.
The left graph was obtained with pure σ-polarized repump laser and no optical
pumping. On the right graph a small amount of optical pumping light is added
giving rise to a non-zero Γpump . One can, with a careful look, see that the lines
now have different widths. Note also, that the height has grown by a factor of
three compared to the graph on the left.
radians per second) the center frequency is νcenter ≈ 325250Hz which predicts
a quadratic Zeeman splitting of 23Hz according to Eq. (F.7). The observed
splitting νsplit = 22Hz is very close, the small deviation is due to Stark shifts
from the laser beams. For the line width we find Γcom = 9.4Hz (FWHM). This
corresponds to a life time of the spin coherence of T2 = 34ms, and we see that
the experimental spectrum supports the model that all eight lines have the same
width. Finally, the fit to the model (7.13) also gives a value of Jx = 0.122 which
is in arbitrary units because we do not know the constant in front of Eq. (7.13).
Let us look at two other examples. First, with the settings as discussed above
(and the optical pump laser still off) we adjust the polarization of the repump
laser to be as circularly polarized as experimentally possible. This gives the
spectrum shown in Fig. 7.3(a). Now the spectrum is much more asymmetric and
the fit gives p = 0.823. This single parameter still seems to describe the shape
with good accuracy. The third example we will show is seen in Fig. 7.3(b). Here
the situation is as before but now with a weak optical pump present with pure
circular polarization. We observe an additional broadening of the left most peak
by Γpump = 5.5Hz and we also note that the second peak seems much broader
(should have an additional broadening by 15.1Hz according to Eq. (7.16)). Since
the fit and the experimental points follow each other very well, we get support
for the modeling of Γpump . The orientation p = 0.967 shows that we are very
close to having all atoms in mF = 4 with only a moderate amount of optical
pumping light.
In the three examples described above and more spectra of the same kind
we get a fitted value for the macroscopic spin Jx . Now, with the DC-Faraday
rotation signal θDC giving an independent measure of Jx we may compare the
fitted Jx with θDC to get another consistency check of the model. This is shown
in Fig. 7.4(a) where we plot the fitted Jx (in arbitrary units) as a function of
θDC . The lowest points are taken with the repump laser only and varying repump
polarization. The upper eight points are taken with purely circular optical pump
of increasing intensity in addition to a purely circular repump laser. We see a
very nice agreement between the fitted and the directly measured values giving
strong support to both the derivations leading to Eq. (7.13) and the modeling of
the spin state described in Sec. 7.2.
Let us conclude this section by noting that we developed a very simple model
describing the spin state of atoms in the presence of optical pump and repump
lasers. The parameter ² = eβ defined in the discussion around Eq. (7.15) is relying
on “equilibrium physics” (the derivation includes maximizing the entropy). It is
pure luck that this simple model is sufficient, our experience tell us that a long
T1 is required to observe this. We indeed have seen bad vapour cells for which
this model does not hold, but this is no problem as long as we can choose a good
vapour cell and test it. In the case of well oriented samples like Fig. 7.3(b) we can
of course not say whether the model is good for the extreme magnetic sub-levels
around mF = −4 where the population must be exponentially small according
to the model. But in this case the spin state is clearly defined by only a few
magnetic sub-levels around mF = 4 and everything is fine anyway. With these
techniques as a starting point we may turn off the pumping lasers and we are
left with a long lived and well characterized spin state for further experimental
investigation.
7.4 Unresolved Lines

The spectra shown in the previous section have been more or less well resolved
which enabled us to directly determine the orientation p. Now, how much infor-
mation can we extract if the line widths are much broader than the quadratic
Zeeman splitting ωsplit ? In this case it will be hard to observe asymmetry in the
spectrum like e.g. Figs. 7.3.
First, assume that all atoms are subject to decoherence with the same rate
described by Γcom À ωsplit and decay from pumping light is a small contribution.
In this approximation we set Γpump = ωsplit = 0 and Eq. (7.13) reduces to
¯ P ¯2
¯ mσ̂m,m ¯ ∝ |Jx |2 ,
¯
MORS(ω) = const · ¯¯2N (7.18)
i(ω0 − ω) + Γcom ¯
where ω0 = ωm+1,m for all m. We see that in this case the spectrum will be
a single Lorentzian the size of which is only depending on Jx . In this case the
independent measure from the DC-Faraday probe will only contribute exactly
the same information and we will not be able to deduce the orientation p, e.g. we
cannot tell the difference between having N = 1012 , p = 1 and N = 2 · 1012 ,
p = 0.5.
7.4. Unresolved Lines 59
0.35 (a) 1.25 (b) 1.02

J zx 1.00
0.30
1.20 Gpump
0.98
0.25
0.96
Gpump / Gpump,0
Jzx from fit [a.u.]
1.15
Jzx / Jxz,0
0.20 0.94
1.10 0.92
0.15
0.90
0.10 1.05 0.88
Model fits poorly
0.05 Model fits well 0.86
1.00
0.84
0.00
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 0.90 0
.92 0
.94 0
.96 0
.98 1
.00
qDC [deg] Fixed orientation p
Figure 7.4: (a) Consistency check of the models. Fits to different spectra give
an estimate of Jx . This can be compared directly to the independently measured
DC-Faraday rotation signal θDC which is proportional to Jx . We indeed observe a
straight line through the origin. Note, the model description hσ̂m,m i ∝ ²m proved
to be less accurate for the lowest four points, but by coincidence the points still
fit well. (b) The interdependence of p, Jx and Γpump in the limit where Γpump
dominates both the common width Γcom and the quadratic Zeeman splitting
ωsplit . Jx,0 and Γopt,0 reflect the values at p = 1. If we can measure e.g. Jx
independently with an accuracy of 2% the orientation can be defined within 1%
in the example shown.
On the other hand, if Γpump dominates Γcom and ωsplit we will get a signal
that depends on the internal atomic spin state. The reason for this is the fact
that different resonance lines have different line widths according to Eq. (7.16).
To examine this approximation we set Γcom = ωsplit = 0 and try to fit the rest of
the parameters to a spectrum which is a perfect Lorentzian. The correct fitting
parameters of course have p = 1 and Γpump equal to the Lorentzian width but
in practical life other sets of parameters will also fit the spectrum to an extent
which one would find reasonable. We find that orientations in the range p = 0.9
to p = 1 all fit the perfect Lorentzian pretty well. We now fix p to a value in
this range and make a fit of the Lorentzian. The resulting values of Jx and Γopt
are shown in Fig. 7.4(b). We see that if we can estimate one of the parameters
Jx or Γopt independently we should be able to calculate the orientation p. For
instance, a measurement of Jx (by the DC-Faraday probe) to an accuracy of 2%
will fix the orientation p to 1%. Keeping track of the optical pump power could
lead to an estimate of Γopt and this could also help making bonds on p. One only
needs to have one fix point, e.g. if one knows that we have p = 1 perfectly in one
case, or if one can reduce Γpump to the point where the spectral lines become
resolved and a calibration like Fig. 7.4(a) can be performed.
7.5 Pulsed Experiments

All previous derivations and measurements in this chapter have been carried out
in cw settings, i.e. Eqs. (7.11a) and (7.11b) assume constant values of frequency
ω and decay rates Γm+1,m . This is indeed valid if lasers are running cw and if
we scan the frequency ω slowly enough. But some experiments must be carried
out in a setting with pulsed lasers, e.g. one might wish to prepare the spin
state in the maximally oriented state F = 4, mF = 4 by illuminating atoms
by a pulse of resonant, circularly polarized laser light, see e.g. the creation of
entanglement in Chaps. 9 and 10. For the magneto-optical resonance method
to be useful in such experiments it must be utilized in the correct experimental
conditions which now means time varying decay rates Γm+1,m . In this section
we outline the extensions into the pulsed regimes and discuss the applicability of
the magneto-optical resonance method for characterization of spin states under
these circumstances.
We assume for simplicity that atoms are pumped to the F = 4, mF = 4
state to an extent that we only need to consider transitions between mF = 3 and
mF = 4. The extension to all levels should be straightforward (but cumbersome).
For these two levels we may write Eq. (7.9) as
∂ σ̃34
= (i∆ − Γ/2)σ̃34 + iχ[σ̂44 − σ̂33 ], (7.19)
∂t
where ∆ = ω − ω43 and χ = gF µB BRF C(F, 3)/4h̄. We assume as in Sec. 7.1
that the populations σ̂44 and σ̂33 can be treated as constants corresponding to
small angle deviations from the x-axis. Then the solution of the above equation
is straightforward
σ̃34 (t) = σ̃34 (0)e(i∆−Γ/2)t

iχ ³ ´ (7.20)
− [σ̂44 − σ̂33 ] 1 − e(i∆−Γ/2)t .
i∆ − Γ/2
This solution starts out with σ̃34 (0) at t = 0 and makes a damped oscillation
toward the steady state value −iχ[σ̂44 − σ̂33 ]/(i∆ − Γ/2). Note, this steady state
value is exactly the result in (7.10), and it is reached in a time ≈ Γ−1 . With the
solution of σ̃34 we can continue to find the actual spin, e.g. Jˆz given by (7.2c)
and predict the results of a measurement.
Experimentally, we set up pumping lasers and a probe laser measuring the
transverse spin state as in Fig. 7.1(a). The lasers are turned on and off with
acousto- and electro-optical modulators. The decay rate in the absence of lasers
is denoted Γdark which is typically small. When the probe laser is on an additional
broadening of Γprobe is present leading to a total decay rate of Γprobe + Γdark .
During the optical pumping pulse the atoms are typically subject to a high decay
rate given in total by Γpump +Γdark . The probe laser is typically turned on shortly
after the optical pumping has been turned off and is maintained for a time shorter
than the decay time (Γprobe + Γdark )−1 . We are thus in the transient regime of
7.5. Pulsed Experiments 61
(a) 0.3ms Pump

(b) 5 0.3ms Pump
14 Measured
1.0ms 0.5ms Probe 1.5ms 0.5ms Probe
Simulation
MORS(ω) [a.u.] 12 4
MORS(ω) [a.u.]
10 15ms time 15ms time
3
8 Measured
Height Simulation
6 2
Width
4
1
2
0 0
324000 325000 326000 324000 325000 326000
Frequency [Hz] Frequency [Hz]
Figure 7.5: Examples of magneto-optical resonance signals in the pulsed regime.

In a cw experiment we estimate Γpump = 770Hz, Γdark = 18Hz, and Γprobe = 2Hz.
The timing of pulses is shown in the two insets, the only difference between (a)
and (b) is the pump duration being 1.0ms and 1.5ms, respectively. The thick
solid line is a simulation as described in the text, the thin solid line is a measured
spectrum. The only free parameter in the simulation is the height which is
common for both figures (and for all data points in Fig. 7.6 below). We define
the width and height of the traces as shown in part (a), these will be discussed
in Fig. 7.6.
Eq. (7.20) and given the frequency ω of the driving RF-magnetic field we cannot
obtain a simple estimate of the amplitude of the response at that frequency as
in (7.13). Instead we have time varying quadrature components of the measured
photo current i(t) and we simply integrate these over time in the presence of the
probe laser. From the perspective of modeling we need to evolve σ̃34 according
to (7.20) with the relevant decay rates and integrate the result over the time of
the probe laser pulse.
We perform the pulsed experiment and compare to simulations, experimental
examples are shown in Fig 7.5. Given the laser powers we may from cw experi-
ments deduce parameters like Γpump , Γdark , and Γprobe for use in the simulations,
and we choose the timing of pulses as shown in the insets of the figures. Only a
common height to the simulated spectrum is a free parameter. From the figures
we see that the model and the experimental data match to a high degree, only
small details in the experimental spectra are not covered by the model. These
details are partly noise and partly some asymmetry which arises from a possible
non-perfect orientation and from the fact that the optical pump laser contributes
a small Stark shift. The structure consists of a central peak which is dominant at
short pump durations Tpump . We observe small sidebands with separation 67Hz
−1
= Tpump which increase in magnitude for longer pump durations. For very large
Tpump we end up with a single, broad structure. As we show in Fig. 7.5(a) we
define a height and width for the simulated and measured spectra. These are
compared in Fig. 7.6, we see there is a very nice agreement for a longer range of
100 700
(a) (b)
600
Width FWHM [Hz]

500
Height [a.u.]
400
10
Measured 300 Measured
Simulated Simulated
200
100
1 0
0 2 4 6 8 10 0 2 4 6 8 10
Pump duration [ms] Pump duration [ms]
Figure 7.6: Comparison of heights and widths of the simulated and measured
magneto-optical spectra, see Fig. 7.5. We have Γpump ≈ 770Hz which corresponds
to a characteristic time around a millisecond. We see that for Tpump ≥ 2ms there
is essentially no change in the height and width of the spectrum, the steady state
value of Eq. (7.20) has been reached with the strong Γpump .
pump durations than the two shown in Fig. 7.5.

We do in general find good agreement with simulated and measured spectra.
As another observation, we will state that with a good approximation the width of
the pulsed spectrum grows proportionally with Γpump if we just to a reasonable
−1
extent have Γdark + Γprob ¿ Γpump ¿ Tprobe . In addition, we may show by
dimensional arguments that (for fixed orientation) the area A of the spectrum is
connected to the width Γ and to the macroscopic spin Jx by A = const · Jx2 /Γ.
In the experiment discussed in Sec. 10.1 this was partially used to estimate J x .
7.6 Stark Shifts by the Probe

We conclude this chapter with a study of the influence of the Stark effect from the
probe laser on the magneto-optical resonance lines. There are several motivations
for this. Firstly, this is a direct experimental test of the higher order terms
(proportional to a2 ) in the interaction Hamiltonian (5.18), and secondly, we will
get some understanding related to the fact that the laser beam does not cover all
atoms at the same time. Finally, some experimental diagnostics can be applied
from the Stark effect.
Let us calculate the Stark effect from the probe laser on the magnetic sublevels
|F, mi. We let the light be strong and linearly polarized along the vector
e1 = ex cos α + ey sin α, (7.21)
i.e. α is the angle between the macroscopic spin direction (the x-axis) and the
probe polarization direction. The Stark effect on magnetic sub-levels is in our
7.6. Stark Shifts by the Probe 63
case much weaker than the splitting caused by the constant magnetic field and
can be calculated in non-degenerate perturbation theory from the interaction
Hamiltonian (5.18).
D E The a0 term is common to all levels, the a1 term is zero
on average since Ŝz = 0, and we are left with the higher order components
proportional to a2 . For the linearly polarized probe we may show that
D E φ(t) D E φ(t)
Ŝ+ (t) = e2iα and Ŝ− (t) = e−2iα , (7.22)
2 2
where φ(t) is the photon flux and Stokes operators are normalized to photons per
second. Then the higher order terms of the interaction Hamiltonian for a single
atom can then be written
eff h̄γ λ2
a2 · φ(t) · ̂2z − [̂2x − ̂2y ] cos(2α) − [̂x ̂y + ̂y ̂x ] sin(2α) .
¡ ¢
Ĥint =−
4A∆ 2π
(7.23)
Now, in the basis quantized along the x-axis we may easily derive
hm| ̂2x |mi = m2 ,

F (F + 1) − m2
hm| ̂2y |mi = ,
2 (7.24)
F (F + 1) − m2
hm| ̂2z |mi = ,
2
hm| ̂x ̂y + ̂y ̂x |mi = 0.
We can now calculate in perturbation theory the contribution of the Stark shift
to the energy levels from the above Hamiltonian.
h̄γ λ2
· ¸
Stark 1 + 3 cos(2α) 1 + cos(2α)
Em = a2 · φ(t) · · m2 − F (F + 1) . (7.25)
4A∆ 2π 2 2
Comparing to the quadratic Zeeman effect of App. F we easily derive that the
Stark effect causes an additional splitting between two resonance lines of the
magneto-optical resonance signal of
γλ2 a2
νStark [Hz] = − · φ(t) · [1 + 3 cos(2α)]
16π 2 A∆
(7.26)
P [mW] · a2 (∆) · [1 + 3 cos(2α)]
= 1.03 · 106 Hz ,
A[cm2 ] · ∆blue [MHz]
where we inserted γ = 5.21MHz and λ = 852nm, and we related the flux φ(t)
to the probe power P . a2 was defined in Eq. (5.16). We remark that we have
here presented an alternative calculation of Eq. (6.24). The additional factor
(2F − 1)σ jx /2 arises from the fact that in Eq. (6.24) we calculated the shift of
the outermost resonance lines m = F ↔ m = F − 1 or m = −F ↔ m = −F + 1.
20 (a) (b)
Stark splitting/Power [Hz/mW]

15
Stark splitting [Hz]

10
5 10
0 ∆ = 1000MHz
P = 2.21mW
-5 α = 0deg
-10 y = C[1+3cos(2π[α-α0]/180o)]
C = 4.35(2) y = Ca2/∆
-15 α0 = 4.7(2) C = 6.80(4)*105
1
-20
-40 -20 0 20 40 60 80 100 120 500 1000 1500 2000 2500 3000
Polarization angle (α) [deg] Blue detuning (-∆) [MHz]
Figure 7.7: (a) The Stark splitting νStark versus polarization direction α. The
dots are experimental and the solid line a fit. We confirm the 1 + 3 cos(2α)
dependence predicted by Eq. (7.26). The fitted constant C can be compared
quantitatively with the theory, see the text for details. (b) The Stark splitting
per milliwatt of probe power νStark /P versus blue detuning (−∆). Again there is
agreement with theory, we confirm the dependence on a2 /∆. The small deviation
at low detuning can be explained by Doppler broadening.
The Stark splitting can be measured with high precision by magneto-optical

resonance signals like the example shown in Fig. 7.2(b). From the experimentally
measured splitting we subtract the quadratic Zeeman contribution (F.7) ν QZ =
23.0Hz to obtain the Stark contribution νStark .
We set up an experiment to measure MORS for different values of the angle
α, the probe power P , the detuning ∆, and the beam cross section A. First of
all, we find that the Stark splitting is independent on the beam cross section
A. The reason for this is the fact that atoms are moving in and out of the
laser beam. For a constant power P all atoms see the same average number of
photons independent on A. We also confirm experimentally that the splitting is
proportional to the power P .
Next, we examine the dependence on the angle α. The results is shown in
Fig. 7.7(a). We make a fit to Eq. (7.26) with an overall constant C and an
offset angle α0 as only free parameters. We find an offset α0 = 4.7deg which
we take as a mis-calibration of the polarization direction α. The experimental
uncertainty in α is about one degree. We also find that the dependence on
(1 + 3 cos(2α)) is confirmed, and the constant C matches the prediction (7.26)
if we take A = 5.8cm2 . This is very close to the effective transverse area of our
vapour cell Aeff ≈ 6.0cm2 which we mentioned in Sec. 4.2.
We also examine the dependence of the Stark splitting on the blue detuning
(−∆). To obtain a higher precision we take several points with different probe
powers P . We plot the measured splitting versus probe power and make a linear
fit to the data. This results in a slope νStark /P which is plotted in Fig. 7.7(b).
This is fitted to the model y = Ca2 /∆blue and we see that the data and the solid
7.6. Stark Shifts by the Probe 65
line fit match very well. A small disagreement for low detuning is probably a
result of the Doppler broadening. We find agreement with Eq. (7.26) if we take
A = 6.1cm2 . Again there is good agreement with an effective area of 6.0cm2 .
We conclude that our predictions match very well with experiments, both
qualitatively and quantitatively. Thus we have confidence that the Hamilto-
nian (5.18) is correct. We also learn that the atomic motion can be modeled
by an effective transverse area Aeff = 6.0cm2 of the vapour cell (at least for
classical mean values). Finally, we saw that the measurements could indicate a
mis-calibration of the polarization angle α. Also, by comparing the Stark split-
ting for two atomic samples we have the possibility to measure the light loss in
the propagation between the samples (we cannot place power meters inside the
vapour cells).
67
CHAPTER 8
Recording Quantum Fluctuations

of Light in Atoms
In this chapter we examine the interaction between atomic spins and the polar-
ization state of light at the quantum level. Our motivation for this is two fold.
First of all, we would like some evidence that the theory described in Chaps. 5
and 6 leads to correct predictions for experimental results. In the present chapter
we take the basic equations of interaction (6.11-6.14) as our starting point. With
these at hand, we develop an understanding of the dynamical evolution of the
quantum spin state in the presence of a laser beam in a non-classical polarization
state. We will find convincing agreement between theory and experiment. Sec-
ondly, the experimental results demonstrate that our atomic spin states indeed
are sensitive to the quantum fluctuations of a light beam. This is one necessary
step toward the realization of a full scale quantum memory for light based on
back action of quantum measurements. We discuss this more carefully in Secs. 8.5
and 11.3.
Other approaches toward a quantum memory for light exist. With the aid
of electromagnetically induced transparency, the amplitudes of an electric field
can be mapped onto coherences of atomic ground state spins [30]. This has been
demonstrated experimentally in [31, 32, 33] for classical mean values of ampli-
tudes and phase. Theoretically this should also work for quantum fluctuations.
The contents of the present chapter are published in [II,VII]. The work of the
present chapter also sheds light on the ultimate sensitivity of spin measurements.
This is discussed in [III,V].
A schematic view of the experiment is shown in Fig. 8.1. An atomic sample
68 Chapter 8 - Recording Quantum Fluctuations
Bbias
Spectrum
λ/2 - analyzer
x
z
laser
Probe
σ+
Sq. y
Coh.
Optical pump and i(t)
repump lasers
Figure 8.1: Schematic view of the experimental setup. A probe laser is sent
through a sample of oriented atoms. The optical and repump laser maintain the
spin orientation. A detection system measures the fluctuations of the light leaving
the atomic sample. A constant magnetic field Bbias moves atomic fluctuations
to the Larmor frequency Ω.
is placed in a constant magnetic field Bbias and the atomic spins are oriented
parallel to this magnetic field along the x-axis. The spins will rotate with Larmor
frequency Ω = gF µB Bbias /h̄ (see App. F). The spin orientation is maintained by
the optical pump and repump lasers which are running cw. The optical pump and
repump power can be adjusted, especially the resonant optical pump laser will
cause decay of the spin state with a rate Γ depending on laser power. The probe
laser emerges from a source that will be described in Sec. 8.2. The quantum
polarization state of the probe laser can be coherent or squeezed. The probe
laser passes the atoms and exchanges quantum fluctuations with these. The
outgoing
D laser E beam is measured by a detection system, and the photo current
i(t) ∝ Ŝyout (t) (see Eq. (E.1)) is fed into a spectrum analyzer. Our goal is to
predict the spectrum of i(t) and connect this to the quantum variables of the
atomic spins and the polarization state of the probe laser.
8.1 Theoretical Approach

We write up Eqs. (6.11-6.14) with the addition of the Larmor rotation and the
spin decoherence. This reads

Ŝzout (t) = Ŝzin (t), (8.2)
∂ ˆ
Jy (t) = −ΩJˆz (t) − ΓJˆy (t) + F̂y (t) + aJx Ŝz (t), (8.3)
∂t
∂ ˆ
Jz (t) = ΩJˆy (t) − ΓJˆz (t) + F̂z (t). (8.4)
∂t
The magnetic field adds the term h̄ΩJˆx to the Hamiltonian which leads to the
first terms of Eqs. (8.3) and (8.4). The decoherence of the spin variables is put in
8.1. Theoretical Approach 69
by hand with a decay rate Γ. This decay term must be accompanied by Langevin
forces F̂y and F̂z to preserve the correct quantum statistics. The properties of
these Langevin forces are calculated in App. G.
The above equations are conveniently solved inD Fourier space. InE addition,
with the knowledge of the correlation function Ŝyout (ω)Ŝyout (−ω 0 ) we may
directly deduce
√ R the spectrum of fluctuations of the photo current defined by
Φ(ω) = 1/ 2π hi(t)i(t + τ )i eiωτ dτ . We will find mathematically
D E D E
φ(ω) · δ(ω − ω 0 ) ∝ Ŝyout (ω)Ŝyout (−ω 0 ) + Ŝyout (−ω)Ŝyout (ω 0 ) . (8.5)
This is essentially the Wiener-Khintchine theorem [86], forR more details see
App. E. We define the Fourier transform by Ŝyout (ω) = √12π Ŝyout (t)eiωt dt and
with some algebra the above equations turn into
Ŝyout (ω) = Ŝyin (ω) + aSx Jˆz (ω), (8.6)

Ŝzout (ω) = Ŝzin (ω), (8.7)
(Γ − iω)F̂y (ω) − ΩF̂z (ω) + (Γ − iω)aJx Ŝzin (ω)
Jˆy (ω) = , (8.8)
(Ω − ω)(Ω + ω) − 2iΓω + Γ2
ΩF̂y (ω) + (Γ − iω)F̂z (ω) + ΩaJx Ŝzin (ω)
Jˆz (ω) = . (8.9)
(Ω − ω)(Ω + ω) − 2iΓω + Γ2
Inserting (8.9) into (8.6) gives the Fourier components

D of the outgoing EŜy oper-
ator. We need to calculate the correlation function Ŝyout (ω)Ŝyout (−ω 0 ) and to
this end we need the correlation functions of the Langevin forces F̂y , F̂z , and
the Stokes operators Ŝyin , Ŝzin . The Stokes operators are discussed in App. E, the
Fourier transform of Eqs. (E.10) and (E.11) yields
D E Sx
Ŝyin (ω)Ŝyin (−ω 0 ) = ²y δ(ω − ω 0 ),
2 (8.10)
D
in in
E Sx
Ŝz (ω)Ŝz (−ω ) = ²z δ(ω − ω 0 ),
0
2
where we have introduced the squeezing parameters ²y and ²z such that ²y =
²z = 1 for the input light in the coherent state (classical laser light) and ²y < 1
or ²z > 1 for light with squeezed Ŝy -component or vice versa. The Heisenberg
uncertainty relation requires ²y · ²z ≥ 1. The ²-parameters measure the noise
level of the Stokes components relative to the level with coherent state light.
Physically, the delta function δ(ω − ω 0 ) times the constant Sx /2 gives the light
noise a white power spectrum which grows proportionally to the flux of photons.
The Langevin forces are derived in App. G, Fourier transforming Eq. (G.6)
yields D E D E
F̂y (ω)F̂y (−ω 0 ) = F̂z (ω)F̂z (−ω 0 ) = Γ|Jx |δ(ω − ω 0 ), (8.11)
where Γ is the spin decay rate introduced in Eqs. (8.3) and (8.4), and |Jx | is the
magnitude of theDmacroscopic
E spin state along x. There are also non-zero cross
correlations like F̂y F̂z which are all delta-correlated as the direct terms.
Now, from Eqs. (8.6) and (8.9) we note that (apart from c-numbers) Ŝyout (ω)
only depends on the quantum variables Ŝyin (ω), Ŝzin (ω), F̂y (ω), and F̂z (ω). Since
all these have correlation function proportional to δ(ω − ω 0 ) it follows that this
is the case for Ŝyout (ω). From Eq. (8.5) and some lengthy but simple algebra we
derive the spectrum Φ(ω) of the photo current i(t). Leaving out irrelevant front
factors we get
1 2 2 ½ 2 2
4 a Sx
¾
Sx a J x Sx ² z
Φ(ω) = ²y + + 2Γ|J x | + neg. freq. (8.12)
2 (Ω − ω)2 + Γ2 2
We have made the narrow band approximation Γ ¿ Ω and |Ω − ω| ¿ Ω which
clarifies the resonant structure of the power spectrum around Ω with width Γ.
There is also a similar negative frequency component around −Ω. The cross
correlations of the Langevin forces and Stokes operators do not contribute here,
this is a result of the special form of Eq. (8.5) which again is a result of time-
ordered light field operators (see App. E).
The spectrum (8.12) is very intuitive. The first term is the noise of the
incoming light, it has a white spectrum and is proportional to the photon flux.
For ²y = 1 we are at the shot noise level (SNL). The second term has a Lorentzian
profile centered at the Larmor frequency Ω with line width Γ (HWHM). This
corresponds to the slow time dynamics of the spin state evolution being of order
Γ−1 . The first term inside the curly brackets is called back action noise, it is a
result of the quantum fluctuations of light being stored in the spin state. The
last terms in the curly brackets is the projection noise of the spin state.
To find a convenient way to compare theoretical predictions with measurement
we integrate the power spectrum over frequencies. We define the back action noise
area and the projection noise area by
µ ¶3 µ ¶2
πa4 Jx2 ²z Sx Sx
BANA²z = , PNA = 2πa2 |Jx | , (8.13)
Γ 2 2
i.e. the integral over the first and second narrow band term in the curly brackets
in (8.12). We note that the BANA scales with the photon flux to the third power,
with Jx2 , and inversely with Γ. The PNA scales linearly with the spin size Jx , this
is the usual finger print of quantum noise imposed by the Heisenberg uncertainty
principle. Note also the quadratic scaling with photon flux and the independence
of the width Γ, projection noise cannot be washed away by decay processes. The
shot noise level SNL = Sx /2 connects the BANA1 (with coherent state light,
²y = ²z = 1) and the PNA in the following way
p
PNA = 2 πΓ(BANA1 ) × (SNL). (8.14)
We may denote the area of additional noise contributions to the atomic noise
as technical noise area TNA. Technical noise (from e.g. lasers or radio stations)
8.1. Theoretical Approach 71
4.5 e y =1
(a) ey <1 (b)
4.0
0,003 Electronic noise
/Hz]
Spectral density 3.5 TNA
Spectral density [VRMS

3.0
2
2.5 0,002
2 Γ= 191Hz
PNA
2.0
1.5
BANA 0,001
1.0
0.5 Shot noise level
0.0 0,000
-400 -200 0 200 400 -900 -600 -300 0 300 600
Ω -ω Ω - ω [Hz]
Figure 8.2: (a) A pictorial description of different noise contributions. The shot
noise of light has a white spectrum and is represented by the flat dotted line. The
narrow band contribution is divided in to three terms, the back action noise, the
projection noise, and the technical noise (see text for more details). The back
action contribution (shaded with gray scale) can be considered as the “memory
part” of the atoms. (b) An example of a measured spectrum. The solid line
is obtained with the input light in a vacuum state (²y = ²z = 1). When the
input mode is in a squeezed state (dashed line) the Lorentzian part from atoms
increases while the wings decrease. The peak on the right is technical noise. In
the experiment Ω = 325kHz.
can scale in different ways, but an important fact is that external sources will be
common to all atoms in the sample. As a consequence the TNA will scale as J x2 .
The different kinds of noise are illustrated in Fig. 8.2(a). On top of the
flat shot noise level (SNL) are the narrow band contributions PNA, BANA, and
TNA. These three contributions add up to one joint Lorentzian structure, in
Fig. 8.2(b) we show two experimental spectra of this kind. The solid line is taken
with coherent state probe (²y = ²z = 1). The dashed line differs from the solid
line solely by a changed quantum state of the input light, we have squeezed Ŝy
such that ²y < 1 and ²z > 1. We note the reduction in the base line according
to the first term in Eq. (8.12), and we see that the Lorentzian peak has become
larger caused by the extra noise in the anti-squeezed component Ŝz , see Eq. (8.13)
for BANA. We wish to single out the individual contributions PNA, BANA,
and TNA from experimental measurements. To this end we may fit the narrow
structures of Fig. 8.2(b) by a Lorentzian shaped function obtaining the total
narrow band noise area. In the case of coherent probe (solid line in Fig. 8.2(b))
we denote the area ACOH and in the case of a squeezed probe (dashed line in
Fig. 8.2(b)) we get the area ASQ . We must have
ACOH ≡ BANA1 + PNA + TNA, ASQ ≡ BANA1 · ²z + PNA + TNA, (8.15)

- 6P3/2 875MHz
Syout Probe
Fourier transform
450 6P1/2
Optical
Sout pumping Repumping
Jout
F=4
B-field + optical pumping 6S1/2
F=3
y
z S in
x Second harmonic
ây generator
Ax
Optical Parametric
Oscillator
Ti:Sapphire
Figure 8.3: The experimental setup. A Ti:sapphire laser is frequency doubled

in a second harmonic generator and down-converted in an optical parametric
oscillator to give the quantum field ây . This field is mixed with a strong field
Ax derived from the same Ti-sapphire laser to obtain non-classical polarization
states of Ŝy and Ŝz , see Eq. (A.14). This light is sent through an atomic cesium
ensemble. The light polarization and the atomic spin state exchange quantum
fluctuations, and the resulting laser beam is measured at a polarization state
analyzer. The different lasers used in the experiment are depicted in the inset.
and we can solve this to obtain the experimental values
ASQ − ACOH ²z ACOH − ASQ

BANA1 = , PNA + TNA = . (8.16)
²z − 1 ²z − 1
The shot noise level (SNL) and the squeezing parameters ²y , ²z are easy to
access by measuring noise of light without atoms. The PNA can be estimated
from Eq. (8.14) and we have thus separated all different noise contributions.
8.2 Experimental Setup

The experimental setup for the study of quantum fluctuation exchange between
atoms and light is shown in Fig. 8.3. The light incident on atoms described by
quantum variables Ŝyin and Ŝzin is engineered by overlapping a strong x-polarized
beam of light with amplitude âx (t) ≈ Ax and a quantum field linearly polarized
along the y-axis with amplitude ây (t) on a polarizing beam splitter. The latter
is generated by frequency doubling a ti:sapphire laser and subsequent down-
conversion in an optical parametric oscillator (OPO) below threshold. This pro-
cess is not a part of my thesis work, we refer to [37, 60] for details. The important
fact is that the OPO generates squeezed vacuum which is characterized by noise
8.3. Experimental Investigation of the Model 73
reduction (below the standard quantum limit √ of classical light) in either of the√
quadrature amplitudes x̂(t) = (ây (t) + â†y (t))/ 2 or p̂(t) = (ây (t) − â†y (t))/i 2
while the other gets more noisy. This corresponds exactly to the case introduced
in Chap. 2 around Eqs. (2.10) and (2.11), we stress that the quantum fluctua-
tions of the polarization state of light is entirely given by the quantum state of
the y-polarized mode emerging from the OPO.
The ti:sapphire laser is blue detuned by ∆ = −875MHz from the D2 line
and the strong x-polarized beam has power up to 5mW limited by saturation
of our Sy detectors. The noise level in the squeezed vacuum is -4.5dB in the
quiet quadrature and 8.0dB in the noisy quadrature both relative to the classical
noise level. However due to propagation losses, non perfect detection, finite
beam overlap and weakly birefringent optics after the PBS only about -3.0dB
squeezing and 7.0dB excess noise is left at the detection. For squeezing of Ŝy
this corresponds to ²y = 0.5 and ²z = 5.0. The bandwidth of squeezing is
8.0MHz HWHM and consequently we can consider our polarization squeezed
probe broadband relative to the atomic spin noise resonance which has a width of
no more than 1kHz FWHM (this was assumed in App. E leading to the theoretical
Eqs. (8.10)).
The laser beam now passes through the atomic cesium sample which has
macroscopic angular momentum Jx created by the repump and optical pump
lasers, see Sec. 4.1. Both lasers are circularly polarized with the same helicity.
By adjusting the relative power of the lasers we are able to control the number
of atoms in the F = 4 ground state. The decay rate Γ is almost only set by the
optical pumping laser which is resonant with atoms in F = 4. The degree of
spin polarization (better than 95%) and the number of atoms is measured by the
magneto-optical resonance method described in Chap. 7.
The output Stokes parameter Ŝy is measured by a polarizing beam splitter.
The power spectrum of the photo current i(t) is recorded in a frequency window
varying from 1.6kHz to 3.2kHz around Ω. The resulting spectrum is a narrow
Lorentzian centered at Ω = 325kHz with a width Γ in the range of 100Hz to 1kHz
FWHM.
8.3 Experimental Investigation of the Model

Now, let us investigate our theoretical model in detail. First of all, the shape of
the experimental traces in Fig. 8.2(b) resembles the theoretical prediction (8.12),
there is a Lorentzian structure on top of a flat noise level. The flat back ground
level is observed to rise or fall with respect to the shot noise level when Ŝy
is anti-squeezed or squeezed, respectively. We also observe an increase of the
Lorentzian peak when Ŝy is squeezed and Ŝz is anti-squeezed, see Fig. 8.2(b).
The opposite example with squeezed Ŝz has only led to a very marginal decrease
of the Lorentzian peak size. First of all, the effect is ten times less pronounced
(²z = 0.5 instead of ²z = 5.0). Just as important is technical difficulties with
phase locking the strong x-polarized beam and the y-polarized quantum field
-1
(a) (b)
BANA 0,30
BANA = const S2.8(2)
x
-2 0,25
log (Backaction G/N )
]
2
Noise power [VRMS

2
0.20
-3
0.15
-4
0.10
-5 PNA derived from BANA
0.05 PNA=0.041(1) Jx
-6 0.00
0 1 2 3 4 5 6 7 8
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
Jx [a.u.]
log (Probe power [mW])
Figure 8.4: The back action noise area BANA as a function of probe power on
log-log scale (a). The probe power is proportional to Sx . Varying the probe
power will cause slight changes to the number of atoms N and the line width
Γ. This is taken care of by plotting BANA·Γ/N 2 on the ordinate axis. We see
the data is consistent with BANA∝ Sx3 . (b) The projection noise area derived
from (8.14). The data is, in fact, compiled from two series with different Γ’s,
giving strong quantitative support of the model.
in the case of Ŝy -anti-squeezing. With the qualitative confirmation of the spec-
trum (8.12), we now turn to a more quantitative comparison between theoretical
predictions (8.13, 8.14) with experimental results. First, we try to vary the probe
power (i.e. to vary Sx ). This will have a slight influence on the number of atoms,
i.e. on Jx , and some influence on Γ due to power broadening. From (8.13) we
expect BANA · Γ/N 2 to be proportional to Sx3 . This is indeed confirmed exper-
imentally, see Fig. 8.4(a). We also wish to confirm that BANA scales with J x2
and that the predicted projection noise area PNA scales linearly with Jx , given
by Eqs. (8.13). Note, these two scaling relations are equivalent since we predict
the PNA from the BANA and the shot noise level. The experimental results are
shown in Fig. 8.4(b). We again confirm the predicted scaling, and in addition,
the constant of proportionality is consistent for different values of Γ.
Finally, let us examine the BANA as a function of Γ. The line width Γ
is controlled by increasing the power of the optical pump laser. Doing this only
affects the macroscopic spin weakly, and we can directly examine the scaling. This
is done in Fig. 8.5(a), where we clearly see the correct scaling with Γ−1 . We may
also consider the noise not originating from the back action effect. This residual
noise area (abbreviated RSN) given by the projection noise plus the technical
noise (PNA + TNA, see Eq. (8.16)) is plotted together with the predicted PNA
in Fig. 8.5(b). We see that the RSN decreases with increasing Γ and seems to
approach the PNA. The physical interpretation is simple. The technical noise
picked up from some unknown source can be reduced by increasing the power
of the optical pump laser. This will exactly destroy the spin state created by
technical noise and push the spin toward the coherent spin state. The coherent
8.4. Broadband Atomic Noise 75
BANA
-1,02(2)
BANA=35(1)*(2 Γ ) (a) RSN (b)
-1,26
] (2G )
]
Noise power [VRMS
Noise power [VRMS

PNA
2
2
0,1 1
0,1
100 2G [Hz] 1000 100 2G [Hz] 1000
Figure 8.5: (a) The measured back action noise area (BANA) for the vacuum
light input as a function of decay rate Γ on a log-log scale. (b) The measured
residual spin noise (RSN) and the inferred projection noise area (PNA) calculated
from (8.13).
spin state has an inherent quantum noise (the PNA) which can by no means be
reduced. Thus we see in Fig. 8.5(b) that the PNA is independent on Γ, and that
the residual noise will not decrease below the PNA.
In Fig. 8.5(a) we also see, that the back action noise BANA can indeed be
reduced in the same way as technical noise. Back action noise is a pile up of
quantum noise from the probe laser and the optical pump will clear up this
again. The stronger the optical pump, the lesser piling up of back action noise
(and technical noise).
A comment should be added to Fig. 8.5(b). Theoretically, the RSN should
really converge to the PNA in the limit of large Γ if our understanding is correct.
Experimentally, we could not go further than Γ = 1kHz, but even so one may
still suspect that the RSN is converging to a slightly higher value. There could
be several reasons for this, e.g. non-perfect orientation of the spin, higher order
effects discussed in Sec. 6.4, effects of beam geometry, etc. We do not have
experimental data suggesting which of these effects (if any) play an important
role here so we will not dig further into that. But this does not change the overall
impression of the results. There is a very good agreement between the theoretical
predictions and the experimental data, and we have a good understanding of the
various noise sources. All together the Faraday effect has been studied in detail
and found to agree with the theory on the level of quantum fluctuations.
8.4 Broadband Atomic Noise

For completeness, in this section we will briefly discuss some broad band noise
which has been observed in addition to the narrow band structure given by
Eq. (8.12). Consider Fig. 8.6(a) which shows a noise spectrum similar to the one
in Fig. 8.2(b). Here we see the recorded signal in presence of atoms and in the
7
(a) (b)
White atomic noise to shot noise level

6
Noise spectral density [mVrms2/Hz]

5
Peak height
1.0
4
oise
3
level
tomic
Beam diameter
2 4mm
noise
ea
6mm
Whit
1 0.1 8mm
Shot
10mm
0
0 500 1000 1500 1 10 100
Frequency [Hz] Peak height to shot noise level
Figure 8.6: (a) An example of the recorded spectrum with atoms (upper trace)
and without atoms (lower trace). We see that in the case with atoms we have
some white atomic noise on top of the shot noise level. The frequency is relative
to the demodulation frequency at 325kHz. (b) Plotted is the white atomic noise
level as a function of the peak height of the narrow structure (see part (a) of the
figure), note the log-log scale. The noise levels are normalized to shot noise. The
probe power and beam size are varied in this experiment. See text for details.
absence of atoms. The important fact is that the flat background is higher than
the shot noise level in presence of atoms, i.e. the atoms contribute some broad
band noise which we denote white atomic noise (white since it seems to be quite
broad band).
In Fig. 8.6(b) this white atomic noise is plotted versus the peak height of the
narrow band atomic noise (both normalized to the shot noise level) where we
for different beam sizes vary the probe power. The beam size is controlled by
clipping a large beam with an iris. The fact that Fig. 8.6(b) reveals a linear
dependence (slope close to unity on log-log scale) just tells that when the narrow
atomic noise increases the white noise part increases in the same way, the increase
is here caused by variations in the probe power. We have not extensively studied
the white noise dependence on various parameters but we do have indications
that it scales linearly with number of atoms and quadratically with the probe
power. What is also clearly seen from the figure is the fact that the white noise
seems to contribute more for a smaller beam size. The reason for plotting the
white noise as a function of the peak height (a property of the narrow noise) is
that we exclude the growth of the white noise to be caused by a simple overall
growth in noise. We explicitly see that the fraction of white noise increases with
decreasing probe diameter.
To understand the white atomic noise we need to carry out more experiments,
it could be interesting to know the bandwidth of this noise but bandwidth limi-
tations of detectors and electronics exclude us from doing this in an easy fashion.
But we certainly have sufficient information to start speculating on the reason
for the white noise. One should remember that broad band noise corresponds
8.5. Discussion of the Results 77
to fast decoherence times, if the white noise contribution is relatively large there
are relatively many atoms which live for a very short time or which are only
measured for a very short time. Since the white noise contribution is largest for
small beam sizes we propose that it is caused by atoms which have spent little
time inside the volume illuminated by probe light. If an atom traverses a beam
of 5mm diameter with a typical speed of 137m/s (see Sec. 4.2) the time duration
will be of order 37µs corresponding to a typical frequency bandwidth of order
27kHz which is broad band with our typical frequency scales. The narrow peak
would then be caused by atoms which have had several opportunities to enter
and leave the region illuminated by the probe laser. We stress that this expla-
nation would gain more confidence with more experiments. In the experiments
reported in previous sections a Gaussian beam with waist 5.1mm was used. The
white atomic noise was observable but all results were concentrating solely on
the narrow part of the atomic noise.
8.5 Discussion of the Results

Let us summarize and discuss the results of this chapter. We started out in
Sec. 8.1 with the basic equations of interaction (6.11-6.14) and developed a model
to describe the particular experimental setup of a polarized laser beam interacting
with a sample of polarized spins. The theoretical predictions were demonstrated
experimentally and we conclude that we understand the light/matter interaction
in detail, this includes the exchange of quantum fluctuations between light and
matter.
The fact that the y-polarized mode could be illuminated with squeezed vac-
uum demonstrated the sensitivity of the spin state to a quantum field of light.
The atoms responded to this light in a way which was not much weaker than
e.g. uninteresting technical noise sources, the clear difference between the two
Lorentzians in Fig. 8.2(b) is solely due to the quantum statistics of the y-polarized
mode emerging from the squeezing source (the OPO). In fact, for our specific val-
ues of the gain of the OPO the y-polarized field contains about one photon per
second per Hz of the bandwidth [60]. The atoms measured in Fig. 8.2(b) are
only sensitive to frequency components of light in a width Γ = 200Hz. Thus the
difference between the solid an dashed line in this case is the absence/presence
of about 200 photons/sec. The atoms have a characteristic memory time of the
order Γ−1 which means that on the time scale of the spin life time the atoms have
been influenced by roughly one photon. This definitely underlines the quantum
sensitivity of the interaction. We note that the measuring and averaging time
necessary to create the two Lorentzians in Fig. 8.2(b) is much more than one spin
life time but the difference of atomic state is the presence of a single photon per
spin decoherence time.
If we direct the discussion toward the outlook for a future implementation
of quantum memory we need to understand what interesting properties of the
quantum field can be stored. First of all, for a real quantum memory we need to
(a) (b) (c)

Sample B1
Cells Sx B2
B3
Sz B4
Cells Laser
Laser beam Time beam
Mirror
Figure 8.7: (a) A possible scenario for storing details about a light pulse. A
rotating mirror will direct a laser beam to different storage cells thereby recording
different time bins of the pulse. This is further illustrated in (b), the interaction
is on when Sx is high (symbolized by the dashed line). The evolution of Ŝz is then
sampled at different times. (c) The same in Fourier space, different frequency
components are recorded in each cell since the magnetic field varies.
store the properties of non-commuting variables. For instance, for a pulse of light
with polarization state described by the variables Ŝy and Ŝz it is desirable to map
Ŝy → Jˆy and Ŝz → Jˆz . This is not what we demonstrate in the present chapter.
We show how the Ŝz -component of light piles up in the atomic spin state. Starting
out with the light field in a squeezed state should lead to atomic spins ending
up in a squeezed state in order to call the exchange of quantum fluctuations a
complete quantum map. A protocol for performing the full quantum map will
be discussed in Chap. 11.
In addition to the above discussion we also note, that the atoms are only
sensitive to fluctuations around the Larmor frequency Ω in the bandwidth Γ.
For storage of the state of light in atoms it is desirable to have a long memory
time and accordingly a small Γ. Hence we can only store a single frequency
component of a light pulse in a single atomic sample, or in other words, if the
Larmor frequency is zero we only store the average value of Ŝz over the light
pulse. To store more detailed time dynamics of a light pulse we could imagine a
setup as depicted in Fig. 8.7(a). Chopping a long pulse into smaller pieces and
storing each piece in a separate atomic sample accomplishes some time resolution.
In a classical picture as in Fig. 8.7(b) this corresponds to sampling the value of
Ŝz at different times. But this must be equivalent to sampling different frequency
components of light, we could choose to shine the light through several atomic
samples each with its own Larmor frequency as depicted in Fig. 8.7(c). This
would be sampling in frequency space. If atoms were stationary a magnetic field
gradient would accomplish the same. A possible implementation for this could
also be a inhomogeneously broadened rare-earth doped solid [33].
In short, the experiments described in this chapter is a first step in the direc-
tion of implementation of a real quantum memory. In Chap. 11 we describe the
remaining steps to reach this goal.
79
CHAPTER 9
Entanglement, Theoretical
Approach
This chapter is devoted to theoretical questions and definitions in connection

to entanglement generation between samples of cesium gases. We consider two
systems 1 and 2 as shown in Fig. 9.1Dwhere E two macroscopic
D E spins are oriented
oppositely along the x-direction with Jˆx1 = Jx = − Jˆx2 . This setting opens
up the possibility to generate an entangled state similar to the EPR-example
in [61] as we will see in the following sections. The trick is to perform a suit-
able measurement which will “collapse” the state of the atomic spins into an
entangled state, see [62]. This is a different approach than other experimental
demonstrations of entanglement between massive particles [40, 41].
In this chapter we define what we mean by entanglement and we describe

how to understand the process of entanglement generation by a measurement.
The latter follows quite naive models which are motivated by the fact that the
polarization state of light and the atomic spin state can be described collectively
in the X̂, P̂-representation as we discussed in Chap. 2. We also discuss general
experimental aspects like the rotating frame, losses for light propagation, non-
perfect detectors, and methods for proving the generation of entangled states.
Actual detailed description of experiments and data analysis is given in Chap. 10.
80 Chapter 9 - Entanglement, Theoretical Approach
Jˆy1
Sy-detector
Light pulse Jˆz1 Jx2 = −Jx
Jx1 = Jx
Jˆy2
Jˆz2
Figure 9.1: The schematic setup of two spin states for entanglement generation.
Two spins are prepared in coherent spin states with opposite directions and the
same magnitude Jx . This setup opens up the possibility to perform joint quantum
non-demolition (QND) measurements on the two states [62].
9.1 Definition of Entanglement

Let us characterize entangled states between samples of macroscopic ensembles
of cesium gases that we use in our experiments. To this end let us start with
the usual definition of entanglement. We consider two systems 1 and 2 which are
described by a joint wave function ψ. If this wave function cannot be written as
a product ψ = ψ1 (x1 ) · ψ2 (x2 ) where ψ1,2 are wave functions of the individual
systems 1 and 2 depending on parameters x1,2 of two systems, then the state of
the two systems is entangled. A well known entangled state is the singlet state
of two spin-1/2 particles |ψi = √12 (|↑↓i − |↓↑i). It is easy to see that this state
cannot be expressed as a product wave function.
Now, the above is a pure state definition, for mixed states described by density
operators there is a similar definition. Consider again two systems 1 and 2 and let
the joint density operator be given by ρ. If ρi1 and ρi2 are sets of density operators
describing the individual systems 1 and 2, and if ρ cannot be decomposed into a
sum of products of these, i.e.
X
ρ 6= pi ρ1i ⊗ ρ2i ⇔ State is entangled, (9.1)
i
where the pi ’s are positive, then the system is in an entangled state. The above
definitions are intuitive, they tell that individual stories ψ1 and ψ2 or for mixed
states ρ1i and ρ2i are not enough to characterize the entire system described by
ψ or ρ, something extra is needed. But while being intuitive the definitions are
not necessarily easy to apply for an experimentalist in the laboratory. However,
from this definition Refs. [63, 64] have derived the following result:
For two continuous variable systems 1 and 2 a sufficient condition for having
entanglement is satisfaction of the inequality
³ ´ ³ ´
Var X̂1 + X̂2 + Var P̂1 − P̂2 < 2, (9.2)
where X̂1 , P̂1 and X̂2 , P̂2 , are continuous variables describing the two sub-systems
9.2. Entanglement Generation 81
h i
satisfying the commutation relations X̂j , P̂j = i for j = 1, 2. If we consider
Gaussian states this is also a necessary condition. This entanglement criterion is
easy to apply for an experimentalist, variances of measured variables are easily
estimated by statistical means.
Now, in Chap. 2 we described how our spin systems could be regarded as con-
tinuous variables described by operators X̂ and P̂ . For our system of oppositely
oriented spins the above inequality turns into
³ ´ ³ ´
Var Jˆy1 + Jˆy2 + Var Jˆz1 + Jˆz2 < 2Jx ⇔ State is entangled. (9.3)
This entanglement criterion is intuitive. If we assume the two systems to be

independent of each other, how low can the left hand side of the inequality
then be? The answer lies in Eq. (2.5). For minimum uncertain states (coherent
spin states) we have Var(Jˆy1 ) = Var(Jˆy2 ) = Var(Jˆz1 ) = Var(Jˆz2 ) = Jx /2. For
independent states we have Var(Jˆy1 + Jˆy2 ) = Var(Jˆy1 ) + Var(Jˆy2 ) and similar
for the z-components. Taking all this together we see that equality of the above
criterion is the best we can obtain classically.
We should also here make a reference to [65] which raises some intriguing
questions about the validity of the use of Eq. (9.2) for spin systems. However, it
is agreed that the criterion (9.3) is valid.
9.2 Entanglement Generation
Let us now discuss the methods we apply in order to generate entanglement.

First of all, consider the commutator
Dh iE D E
Jˆy1 + Jˆy2 , Jˆz1 + Jˆz2 = i Jˆx1 + Jˆx2 = i(Jx − Jx ) = 0. (9.4)
The fact that this commutator has zero mean for two oppositely oriented spins en-
sures the existence of entangled states, the variances Var(Jˆy1 + Jˆy2 ) and Var(Jˆz1 +
Jˆz2 ) can simultaneously be arbitrarily small and break the inequality of the en-
tanglement criterion (9.3).
We will generate the entangled state by a quantum non-demolition (QND)
measurement of the two operators Jˆy1 + Jˆy2 and Jˆz1 + Jˆz2 as discussed in [62]. To
describe this process we will for simplicity start out with two oppositely oriented
spins J1 and J2 in zero magnetic field, we comment on the rotating frame with
a non-zero magnetic field below. Placing the spins as depicted in Fig. 9.1 and
applying a probe laser beam through both samples, the equations of interaction
(6.11-6.14) will turn into
Ŝyout (t) = Ŝyin (t) + aSx (Jˆz1 (t) + Jˆz2 (t)), (9.5)
Ŝzout (t) = Ŝzin (t), (9.6)
∂ Jˆy1 (t) + ∂ Jˆy2 (t)
= aJx1 Ŝzin (t) + aJx2 Ŝzin (t) = 0, (9.7)
∂t
∂ Jˆz1 (t) + Jˆz2 (t)
= 0, (9.8)
∂t
where the Stokes operators here are normalized to photons per second. We see
that a measurement of Ŝyout will provide information about Jˆz1 + Jˆz2 , and if the
second term of Eq. (9.5) is large compared to the first one we will make effective
measurements of the spins. At the same time, the vanishing time derivatives (9.7)
and (9.8) ensures the measurement to be non-destructive. For the time derivative
of Jˆy1 + Jˆy2 to be zero, it is important that the spins are opposite with same
magnitude Jx and the absence of light losses between the samples is also required
here. If losses are present the Ŝz -component seen by the two spin samples will
be different, see Sec. 9.6.
We will let the probe laser be on for a time duration T while measuring the
Ŝy -component of the light. We measure the photo current i(t) and for the rest
of this chapter we assume that it is (in appropriate units) equal to Ŝyout . This
corresponds to a perfect detector efficiency, see detection theory in App. E and
.
Eq. (E.1). We will write i(t) = Ŝyout where the dot symbolizes that a measurement
RT
is performed. Defining the integrated value of the outcome as A = 0 i(t)dt, we
find that A is a the outcome of a measurement of the operator
Z T Z T
.
A= Ŝyout (t)dt = Ŝyin (t)dt + aSx T (Jˆz1 (0) + Jˆz2 (0)). (9.9)
0 0
After this process we may apply another laser beam along the y-direction, or
alternatively rotate the two spins by 90 degrees with help from a magnetic field
and apply the same laser once again. Then a non-destructive measurement of
Jˆy1 + Jˆy2 is performed and the z-components are unaffected.
Now, we will quantify the magnitude of the second term versus first term in
Eq. (9.9). If we calculate the variance we obtain
ÃZ ! Z Z
T T T D E
out
Var Ŝy (t)dt = Ŝyin (t)Ŝyin (t0 ) dtdt0
0 0 0
³ ´
+ a2 (Sx T )2 Var Jˆz1 (0) + Jˆz2 (0)
(9.10)
Sx T Sx T ¡
+ a2 (Sx T )2 Jx = 1 + 2a2 Sx T Jx
¢
=
2 2
Sx T ¡ 2
¢
≡ 1 + 2κ ,
2
9.2. Entanglement Generation 83
D E
where we used the fact that Ŝyin (t)Ŝyin (t0 ) = Sx /2 · δ(t − t0 ), see Eq. (E.10). We
defined κ2 = a2 Sx T Jx as a figure of merit for the spin contribution compared
to the light noise contribution. Above we assumed the spins to be prepared in
the coherent spin state with variances equal to Jx /2, see Eq. (2.5). For efficient
entanglement generation we need to have κ2 large compared to unity. We see
that we gain by increasing the magnitude of the spins Jx , by increasing the total
photon number nph = Sx T /2, or by adjusting the interaction parameter a given
by Eq. (6.15).
It is here appropriate to connect the concept of entanglement between spin
states to the concept of spin squeezing. If we in the above considerations only
had one spin sample (sample 1 for instance) the measurement of Ŝyout for a laser
beam propagating along the z-direction would obtain information about Jˆz1 ,
RT
and the variance of 0 Ŝyout (t)dt would have been Sx T /2 · (1 + κ2 ). At the same
time the effect of the probe laser on the Jˆy1 -component would be governed by
∂ Jˆy1 (t)/∂t = aJx Ŝzin (t) which leads to
Z T
Jˆy1 (T ) = Jˆy1 (0) + aJx Ŝzin (t)dt ⇒
0
³ ´ ³ ´ Z T Z T D E
Var Jˆy1 (T ) = Var Jˆy1 (0) + a2 Jx2 dt dt0 Sz in (t)Ŝzin (t0 ) (9.11)
0 0
Jx a2 Jx2 Sx T Jx
= + = (1 + κ2 ),
2 2 2
where the initial spin state is assumed to be the coherent spin state with variance
Jx /2. We see that when we for higher κ2 obtain better information about the
Jˆz1 -component of the spin we pile up more noise in the Jˆy1 -component which
is required in order not to violate Heisenberg’s uncertainty principleh (2.3). This
i
uncertainty relation is a consequence of the non-commuting property Jˆy1 , Jˆz1 =
iJx . Now, for our two samples we have commutators
h i h i
Jˆy1 + Jˆy2 , Jˆz1 − Jˆz2 = 2iJx and Jˆz1 + Jˆz2 , Jˆy1 − Jˆy2 = 2iJx , (9.12)
while all the other combinations vanish, i.e.

h i h i
Jˆy1 + Jˆy2 , Jˆy1 − Jˆy2 = Jˆz1 + Jˆz2 , Jˆz1 − Jˆz2 =
h i h i (9.13)
Jˆy1 + Jˆy2 , Jˆz1 + Jˆz2 = Jˆy1 − Jˆy2 , Jˆz1 − Jˆz2 = 0.
This motivates the interpretation of entanglement to be squeezing of two inde-

pendent modes Jˆy1 + Jˆy2 and Jˆz1 + Jˆz2 at the expense of anti-squeezing the
conjugate variables Jˆz1 − Jˆz2 and Jˆy1 − Jˆy2 , respectively (such representation has
been discussed in [66]). We have the two corresponding Heisenberg uncertainty
relations
³ ´ ³ ´
Var Jˆy1 + Jˆy2 Var Jˆz1 − Jˆz2 ≥ Jx2 and
³ ´ ³ ´ (9.14)
Var Jˆz1 + Jˆz2 Var Jˆy1 − Jˆy2 ≥ Jx2 .
In the case of two spin samples a calculation similar to Eq. (9.11) will show how
much noise is piled up in the anti-squeezed variables Jˆy1 − Jˆy2 and Jˆz1 − Jˆz2 ,
³ ´ ³ ´
Var Jˆy1 (T ) − Jˆy2 (T ) = Var Jˆz1 (T ) − Jˆz2 (T ) = Jx (1 + 2κ2 ). (9.15)
All equations in this section are coherent evolution of the spin states. But we also
perform measurements which change the spin states in a non-coherent way and
hopefully create the entangled states. In next section we try to model this mea-
surement and the corresponding spin state evolution. The above considerations
will play an important role in the understanding of the results.
9.3 Wave Function Modeling

The fact that the collective properties of light and atoms can be described by
position and momentum like operators (see Chap. 2) motivates a simple model
where the entire spin state is described by a wave function ψ(x) like it is well
known for a single particle in elementary quantum mechanics. For simplicity we
will in the following describe a single spin ensemble and the effect of a measure-
ment as we already discussed around Eq. (9.11). As in Eqs. (2.10) and (2.4) we
define RT RT
0
Ŝy (t)dt Ŝz (t)dt
X̂L = √ and P̂L = 0 √ (9.16)
Sx T Sx T
for light pulses (indexed by L) of duration T and for the atomic spin state (indexed
by A) we define
Jˆy Jˆz
X̂A = √ and P̂A = √ . (9.17)
Jx Jx
h i
These definitions fulfil X̂, P̂ = i. We now assume the state of light and atoms
to be described by a Gaussian wave function on the form
(x − x0 )2
µ ¶
ψ(x) = N exp − + ip0 x , (9.18)
4σx2
where N is a suitable normalization constant. A Gaussian distribution is mo-

tivated by the fact that our physical system is composed of a huge number of
particles each having their own statistical properties. The central limit theorem
ensures that the collective properties will be Gaussian if the individual particles
are not too far from being independent of each other. The above state fulfils
9.3. Wave Function Modeling 85
D E D E
X̂ = x0 , P̂ = p0 , Var(X̂) = σx2 , and Var(P̂) = σp2 = 1/(4σx2 ). We always
have Var(X̂)Var(P̂) = 1/4, i.e. we are in the minimum uncertainty state. For
the coherent state of light or atoms we have σx2 = σp2 = 1/2. The operators are
described by the usual substitution X̂ → x and P̂ → −i∂/∂x. We may change
from x to p representation by a Fourier transform
1
Z
ψ(x) = √ eipx ψ(p)dp,
2π
(9.19)
1
Z
ψ(p) = √ e−ipx ψ(x)dp,
2π
which for Eq. (9.18) would turn into
(p − p0 )2
µ ¶
ψ(p) = N exp − − ix 0 p . (9.20)
4σp2
Now, prepare the atomic sample to be in the coherent spin state described by
ψ(pA ) = N exp(−p2A /2) and the light also in the coherent state described by
ψ(pL ) = N exp(−p2L /2). Now, when light and atoms interact, the evolution is
governed by the Hamiltonian (this is essentially Eq. (5.18) with neglected higher
order terms)
Ĥ = h̄aŜz (t)Jˆz (t) ⇒

Ã ! Ã !
i T T
Z Z
exp − Ĥdt = exp −ia Ŝz (t)Jˆz (t)dt (9.21)
h̄ 0 0
³ p ´ ³ ´
= exp −ia Sx T Jx P̂L P̂A = exp −iκP̂L P̂A ,
where we used the fact that Jˆz (t) is constant and κ is as defined in the previous
section. With the joint state of atoms and light given by ψ(pL , pA ) = exp(−[p2L +
p2A ]/2) we calculate the evolution to be
Ã Z !
T
[p2 + p2A ]
µ ¶
ψ(pL , pA ) → exp −i Ĥdt ψ(pL , pA ) = N exp − L − iκpL pA .
0 2
(9.22)
If we Fourier transform this into the x-basis of light corresponding to Ŝy we
obtain (compare Eqs. (9.18) and (9.20) to see this)
[xL − κpA ]2 p2A

µ ¶
ψ(xL , pA ) = N exp − − . (9.23)
2 2
So far we described coherent evolution of the state under the influence of the
interaction Hamiltonian. Now we perform a measurement of X̂L which we model
.
by letting xmeas
L = X̂L , i.e. we assume the light part of the state to collapse to a
definite measured number. The remaining atomic state is then described by the
above equation with the variable xL replaced by the constant xmeas
L ,
Ã κ
!
meas 2
[pA − 1+κ 2 xL ]
ψ(pA ) → N exp − 1 . (9.24)
4 · 2(1+κ 2)
D E
κ meas 1
This state reveals the statistics P̂A = 1+κ 2 xL and Var(P̂A ) = 2(1+κ 2 ) . We
D E 2
also have X̂A = 0 and Var(X̂A ) = 1+κ 2 . Let us convert this back to the spin
variables Jˆy and Jˆz . We get
³ ´ J ³ ´ J 1
x x
Var Jˆy = · (1 + κ2 ) and Var Jˆz = · . (9.25)
2 2 1 + κ2
For the Jˆy -component we obtain exactly the variance given by Eq. (9.11). The
back action noise from the light Ŝz -component supplies the extra noise added to
Jˆy . The Heisenberg uncertainty then allows the variance of the Jˆz -component to
be reduced up to a factor 1 + κ2 . In our simple pure state model we stay in the
minimum uncertainty state and obtain exactly the reduction of Var(Jˆz ) by the
factor (1 + κ2 ).
For the mean values of the spin componentsR it is convenient to use the
R units
. R
of the detected signal, i.e. since we measure i(t)dt = Ŝyout (t)dt = Ŝy dt +
aSx T Jˆz we multiply the spins by aSx T and calculate
Z T
D E D E κ2
aSx T Jˆy = 0 and aSx T Jˆz = i(t)dt. (9.26)
1 + κ2 0
We remember from the previous section that κ2 is the ratio of atomic noise to light
noise in a measurement with one spin sample. The correction factor κ2 /(1 + κ2 )
is then intuitively understandable. This factor is the ratio of atomic noise to the
total noise and it is the atomic portion that bears the spin state information. In
the limit κ2 → 0 we obtain no information about the spin state, and there is no
back action. Then the best bet for the mean value is the initial coherent spin
state value of zero. On the other hand, if κ2 À 1 the light noise is negligible and
the correction factor should be unity.
Now, we may apply the above considerations to the case of two spin samples.
If we measure Jˆz1 + Jˆz2 by integrating the equation Ŝyout (t) = Ŝyin (t) + aSx (Jˆz1 +
Jˆz2 ) we know from Eq. (9.10) that the ratio of atomic to light noise is 2κ2 .
Integrating the measured Ŝyout (t) we find that the mean value of the state after
the measurement is
Z T
D E 2κ2
aSx Jˆz1 + Jˆz2 = i(t)dt. (9.27)
1 + 2κ2 0
From Eq. (9.15) we remember that the pile up of noise in the conjugate variable
Jˆy1 − Jˆy2 is Jx (1 + 2κ2 ). With the preservation of the minimum uncertainty
9.4. Rotating Frame and Entanglement 87
relation in our simple model we find the variance of the created state to be
³ ´ Jx
Var Jˆz1 + Jˆz2 = (9.28)
1 + 2κ2
which is the coherent spin state variance reduced by a factor 1 + 2κ2 . The same
considerations apply for the y-components of the spins. We have checked that
the above results for two spin samples (in zero magnetic field) match the result
of a wave function analysis with Gaussian wave packets.
Our models are quite simple and rely on the assumption that the collection of
spins or photons can be described by a wave function depending on the collective
variables. To get more insight into the microscopic picture of the measurement
process we refer to [67] in the case of measurements on a single spin ensemble
and to [68] for the opposite spin setup with entanglement generation.
9.4 Rotating Frame and Entanglement

As we already discussed in Sec. 4.3 we place the atomic samples in magnetic
fields. The spin precession forces us to consider the above calculations in the
rotating frame and we introduce the rotating frame coordinates Jˆy10
, Jˆz1
0
, Jˆy2
0
,
ˆ0
and Jz2 in analogy with Eq. (4.3). After a little algebra Eqs. (9.5-9.8) can be
written
Ŝyout (t) = Ŝyin (t)

³ ´
+aSx cos(Ωt)[Jˆz1 0
(t) + Jˆz2
0
(t)] + sin(Ωt)[Jˆy1
0
(t) + Jˆy2
0
(t)] , (9.29)
Ŝzout (t) = Ŝzin (t), (9.30)
∂[Jˆy1
0
(t) + Jˆy2
0
(t)] ∂[Jˆz1 0
(t) + Jˆz2
0
(t)]
= = 0. (9.31)
∂t ∂t
We see that the rotating spin coordinates are constants of motion (regarding the
coherent evolution) and that these can be measured simultaneously by measuring
Ŝyout . The sine and cosine components can be separated electronically from the
measured photo current i(t) if the time of measurement T fulfils ΩT À 1. If we
define Z T Z T
A= i(t) sin(Ωt)dt and B= i(t) cos(Ωt)dt, (9.32)
0 0
we get for ΩT À 1
T
aSx T h ˆ0
Z
.
i
A= Ŝyin (t) sin(Ωt)dt + Jy1 (0) + Jˆy2
0
(0) ,
0 2
T
(9.33)
aSx T h ˆ0
Z
.
i
B= Ŝyin (t) cos(Ωt)dt + Jz1 (0) + Jˆz2
0
(0) .
0 2
These operators are measured by multiplying the photo current by cos(Ωt) and
sin(Ωt) respectively and integrating over time T . The ratio of measured atomic
noise to light noise can be evaluated and we find that A and B are outcomes of
stochastic variables with variances
Sx T
Var (A) = Var (B) = (1 + κ2 ), (9.34)
4
where κ2 = a2 Sx T Jx is as defined in previous sections and we assumed the spin
state at t = 0 to be the coherent spin state. Compared to the previous section
the atomic to light noise ratio is κ2 and not 2κ2 . The reason is the fact that
during the time T we now spent time measuring both y- and z-components of
the spins and not just a single component.
We may also consider the conjugate variables Jˆy1
0
(t)−Jˆy2
0
(t) and Jˆz1
0
(t)−Jˆz2
0
(t)
as we did in previous section. The evolution of these can be shown to be
∂[Jˆy1
0
(t) − Jˆy2
0
(t)]
= 2aJx Ŝzin cos(Ωt), (9.35)
∂t
∂[Jˆz1
0
(t) − Jˆz2
0
(t)]
= −2aJx Ŝzin sin(Ωt), (9.36)
∂t
and we can integrate this from t = 0 to t = T to evaluate the amount of noise
piled up in these variables. The result turns out to be
³ ´ ³ ´
Var Jˆy1
0
(t) − Jˆy2
0
(t) = Var Jˆz1
0
(t) − Jˆz2
0
(t) = Jx (1 + κ2 ). (9.37)
Now we may compare the results of this section to the previous one and extrapo-
late the reasoning with minimum uncertainty states to the rotating frame. After
having performed the measurement our best estimate for the mean values and
variances are
Z T
aSx T D ˆ0 ˆ0
E κ2
Jy1 (T ) + Jy2 (T ) = i(t) sin(Ωt)dt,
2 1 + κ2 0
T
aSx T D ˆ0 κ2
E Z
Jz1 (T ) + Jˆz2
0
(T ) = i(t) cos(Ωt)dt, (9.38)
2 1 + κ2 0
³ ´ ³ ´ Jx
Var Jˆy1
0
(T ) + Jˆy2
0
(T ) = Var Jˆy1
0
(T ) + Jˆy2
0
(T ) = .
1 + κ2
Again, let us comment on the results. If κ2 → ∞ we simply have zero variance of
RT
Jˆy1
0
+ Jˆy2
0
and Jˆz1
0
+ Jˆz2
0
while 0 i(t) sin(Ωt)dt = aSx T /2 · [Jˆy1
0
+ Jˆy2
0
] holds (and
ˆ0 ˆ0
similarly for Jz1 + Jz2 ). This is just a perfect measurement of the spin components
and is consistent with the total neglect of the term Ŝyin (t) in Eq. (9.29). On the
other hand, if κ2 → 0 the measurement is effectively non-existing and the spin
state is unaltered. Hence we must have variances equal to the initial coherent
state variance Jx and the mean value should be zero independent on the measured
photo current i(t), the latter is just light noise bearing no information about the
spin state.
9.5. Entanglement Estimation 89
This is not
Var(A) + Var(B)
experimental
data.
2
κ
Jx
Figure 9.2: The theoretically expected noise in a measurement of transverse spin

components Jˆy1 + Jˆy2 and Jˆz1 + Jˆz2 if the spins are prepared in oppositely oriented
coherent spin states, see Eq. (9.39). There is a constant noise contribution from
light (dashed line) and the atomic contribution will grow linearly with the spin
magnitude Jx following the solid line.
9.5 Entanglement Estimation

As experimentalists we wish to create an entangled state between two cesium gas
samples, in this section we describe some strategies for proving that an entangled
state has been created. To this end we must keep in mind the entanglement
criterion (9.3) and be able to use it. One of the most important tasks is to
calibrate this inequality in the sense that we must know which values of the
variances correspond to equality. To this end remember that if the two samples 1
and 2 are independent of each other then Var(Jˆy1 +Jˆy2 ) = Var(Jˆy1 )+Var(Jˆy2 ) and
Var(Jˆz1 + Jˆz2 ) = Var(Jˆz1 ) + Var(Jˆz2 ). Furthermore, for coherent spin states the
spins will be in the minimum uncertainty state where these individual variances
amount to Jx /2, see Eq. (2.5). We conclude that for two oppositely oriented
coherent spin states we have the equality Var(Jˆy1 + Jˆy2 ) + Var(Jˆz1 + Jˆz2 ) = 2Jx .
If we for a given magnitude Jx can create coherent spin states and measure these
variances we have calibrated the inequality.
Experimentally we will be in the rotating frame and we will create the coher-
ent spin states by optical pumping processes, see Sec. 4.5. Then we will measure
the values A and B as in Eq. (9.33) and we will do this several times. By statis-
tical means we can then estimate the mean and variance of A and B and study
the statistics as a function of the spin magnitude Jx .
For Jx = 0 we only see the light noise (and maybe some electronics noise
from the detectors). It is easy to judge whether this light noise is limited by
quantum noise (also called shot noise) by varying the photon number n ph of the
laser pulse, the noise variance should grow linearly with nph , see Eq. (2.12). We
will very often normalize our units of noise to the shot noise of light since the
light noise is easy to measure.
Now, we would start to increase Jx to non-zero values and measure the noise
variances once again. If the additional atomic noise is limited by quantum pro-
jection noise, the atomic contribution should increase proportionally with J x , see
Eq. (2.5). This fact is also reflected by the equality of the criterion (9.3). In
Fig. 9.2 we show what a plot of the measured variances Var(A) + Var(B) ver-
sus Jx should look like in the case of the atomic noise being characterized by
the quantum projection noise of the coherent spin states. The solid line would
then be a fit to the measured points and serve as calibration for the entangle-
ment criterion (9.3). The dashed line symbolizes the shot noise of light. Note
from Eq. (9.33), that the sum Var(A) + Var(B) actually measures the variances
Var(Jˆy1
0
+ Jˆy2
0
) + Var(Jˆz1
0
+ Jˆz2
0
) + light noise, i.e.
µ ¶2 ³
Sx T aSx T ³ ´ ³ ´´
Var(A) + Var(B) = + Var Jˆy1
0
+ Jˆy2
0
+ Var Jˆz1
0
+ Jˆz2
0
2 2
 ³ ´ ³ ´
Sx T  Var Jˆy1
0
+ Jˆy2
0
+ Var Jˆz10
+ Jˆz2
0
= 1 + κ2 
2 2Jx
(9.39)
such that the contribution above the dashed line is exactly the left hand side
of (9.3). To connect this calibration to the considerations in the previous section
we note that the atomic contribution is equal to κ2 in the units where shot noise
of light has unity variance.
We will now turn our attention toward the evidence of entanglement genera-
tion. The possibly entangled state is created by the first measurement of A and
B which we denote by A1 and B1 (indices 1 for first pulse). We need to perform
a second measurement A2 and B2 in order to characterize the state created by
the first one.
Let us discuss what could be observed if the pure state model of Sec. 9.3
holds exactly. In this case the first pulse measurement A1 and B1 both have zero
mean and variance 1 + κ2 (which we already showed in Fig. 9.2). The second
measurement variables A2 and B2 will theoretically fulfil
κ2 A 1 κ2 B 1
hA2 i = and hB 2 i = ,
1 + κ2 1 + κ2
(9.40)
κ2 A 1 κ2 B 1 κ2 1 + 2κ2
µ ¶ µ ¶
Var A2 − 2
= Var B2 − 2
=1+ 2
= .
1+κ 1+κ 1+κ 1 + κ2
For the variances of A2 (or B2 ) minus the mean value the term 1 is the light noise
contribution of the second pulse and the term κ2 /(1 + κ2 ) is the variance of the
created entangled state. If by measurements and statistical calculations we could
confirm these results we would have proved experimentally that the entangled
state was created. Let us also here discuss the expected variance of the second
measurement alone. We would obtain
¶2
κ2 κ2
µ
2
Var(A2 ) = Var(B2 ) = 1 + + (1 + κ ) = 1 + κ2 . (9.41)
1 + κ2 1 + κ2
9.6. Entanglement Generation and Losses 91
The term 1 is the shot noise of the second pulse, the term κ2 /(1 + κ2 ) is the
variance of the created entangled state, and the last term arises from the fact
that the mean value of e.g. A2 is random, A1 has variance 1+κ2 and the correction
factor κ2 /(1 + κ2 ) must be squared. The result is 1 + κ2 which is the same as the
variance of the first pulse. This fact reflects the QND nature of the measurement.
If we perform a measurement and if a ”secret” observer had already performed
the same measurement on the spins without telling us the outcome we would not
be able to notice.
Now, the pure state model and its implications discussed above need not
hold. The model could simply be wrong, or several physical effects could change
the picture. For instance if decoherence is strong between the first and second
measurement pulse entanglement could be hard to observe. In order to apply
the above considerations in a more realistic setting we assume more generally
that the first measurement (A1 , B1 ) creates a state with mean value (αA1 , αB1 )
and variance Var(Jˆy1 + Jˆy2 ) + Var(Jˆz1 + Jˆz2 ) = 2Jx · σ 2 . In this case we can
estimate α and σ 2 by performing a large number N of measurements each giving
results (A1 [i], B1 [i], A2 [i], B2 [i]), where i is indexing the different measurements.
The variables A2 [i] − αA1 [i] and B2 [i] − αB1 [i] will have zero mean and variance
1 + σ 2 in units of shot noise, the term 1 arises from the shot noise of the second
measurement. We calculate
ÃN N
!
2 1 X
2
X
2
1+σ = (A2 [i] − αA1 [i]) + (B2 [i] − αB1 [i]) , (9.42)
N −1 i i
where α must be chosen such that the right hand side is minimal. The minimiza-
tion procedure is equivalent to a linear fit through (0, 0) if we plot the second
pulse results A2 and B2 versus the first pulse results A1 and B1 . If σ 2 is below
the level of the coherent spin state κ2 (symbolized by the atomic part above
the dashed line in Fig. 9.2) we have created an entangled state. In Sec. 10.4
we discuss experiments connected to this method. In Sec. 10.1 we discuss other
experiments where we for technical reasons did not have the ability to utilize the
above method but we were able to estimate the variance of A2 − A1 and B2 − B1 ,
i.e. we were forced to put alpha to unity. In this case we have
ÃN N
!
2 1 X
2
X
2
1+σ ≤ (A2 [i] − A1 [i]) + (B2 [i] − B1 [i]) , (9.43)
N −1 i i
i.e. our estimation of the entangled state variance is not optimal but put an upper
bound on σ 2 which is sufficient for entanglement demonstration. Note, our pure
state model predicts α = κ2 /(1 + κ2 ). The simple method with α = 1 should
work best for a large atomic to shot ratio, κ2 À 1.
9.6 Entanglement Generation and Losses

Let us discuss the role of losses for the generation of entanglement. Consider
Fig. 9.3 where we place two spin samples 1 and 2 next to each other and shine
D E
i(t) ∝ Ŝydet (t)
Jx,1 Jx,2
η1 η2
Figure 9.3: Light is propagating through two cells with macroscopic spins J x1
and Jx2 oriented along the x-axis. We parametrize the loss between the two
cells by η1 and the losses after the second cell by η2 (also including the detector
efficiency). The detector measures the Ŝy -component of light.
a pulse of light through them as usual. We assume losses to be present between

samples 1 and 2 modeled by the transmission efficiency η1 and between sample
2 and the detector modeled by η2 . In this case the Stokes operators of light will
transform according to Eqs. (A.15) and (A.16). If the light before sample 1 is
described by Sxin ≡ Sx , Ŝyin , and Ŝzin , we may show that
Sxdet = η1 η2 Sx , (9.44)
³ ´
Ŝydet = η1 η2 Ŝyin + aη1 η2 Sx Jˆz1 (t) + Jˆz2 (t)
r r
η1 (1 − η1 )Sx η1 η2 (1 − η2 )Sx
+ η2 V̂y1 + V̂y2 (9.45)
2 2
for the light reaching the detectors. The strong Sx -component is attenuated by
the overall efficiency η1 η2 . The Ŝy component consists of the attenuated input
field Ŝyin , a readout of Jˆy1 + Jˆy2 with some attenuation, and finally some added
vacuum noise described by operators V̂y1 and V̂y2 (see Sec. A.4). Even in the
case of losses between samples 1 and 2 (η1 < 1) we readout Jˆz1 and Jˆz2 with
equal weight. Atoms will see the back action from the Ŝz -components which at
sample 1 and 2 amount to
Ŝzcell 1 = Ŝzin , (9.46)
r
η1 (1 − η1 )Sx
Ŝzcell 2 = η1 Ŝzin + V̂z1 . (9.47)
2
Since these two are different, there will not be a perfect back action cancellation
in the case of opposite spins (Jx1 = −Jx2 = Jx ). The evolution of spins can
be found by time integration of Eqs. like (4.4-4.7) which is straightforward but
cumbersome. We will not carry out the calculations here but just state the
results.
• If we prepare two coherent spin states with x-components Jx1 and Jx2 like
in Fig. 9.3 and if we perform measurements like those described in connec-
tion to Eq. (9.39) we do not expect the measured noise to depend linearly
9.7. The Atomic Projection Noise Level 93
on the spin magnitude, a quadratic component is added. This quadratic

contribution arises from the non-perfect cancellation of back action noise.
• The optimal balancing of spins which reduces the quadratic contribution

most is having Jx1 = −Jx2 . This is independent on η1 .
• The shot noise measured on the detectors depend only on the number of
photons received by the detector.
• In the case of balanced spins Jx1 = −Jx2 = Jx the measured noise (in
units of shot noise) follows Eq. (9.48) below. Here we define PN/SN as the
projection noise (PN) to shot noise (SN) ratio. This increases linearly with
Jx . The result PN/SN = κ2 = a2 Sx T Jx in case of no losses is modified to
PN/SN = η1 η2 κ2 .
µ ¶2
PN 1 − η1 PN
Measured noise = 1 + + . (9.48)
SN 12η1 η2 SN
The quadratic term vanishes if η1 = 1, then back action cancellation is perfect.
The front factor 1/12η1 η2 is valid for a quantum noise limited Ŝz -component of
light. If there is additional classical noise from e.g. the laser the detected back
action is larger.
For entanglement generation, the non-canceled back action noise leads to the
fact that we cannot measure Jˆy1 + Jˆy2 without disturbing Jˆz1 + Jˆz2 to some ex-
tent. Since back action noise increases quadratically with Jx this is an increasing
problem for increasing PN/SN ratios. At the same time, a large PN/SN ratio is
required in order to generate strong entanglement as we saw in the previous sec-
tions. Hence, avoiding losses between samples 1 and 2 and a good spin balancing
is important for efficient entanglement generation. On the other hand, for weak
entanglement with small κ2 we should not worry too much about the back action
cancellation.
9.7 The Atomic Projection Noise Level

In previous sections we discussed the entanglement criterion (9.3) and its cali-
bration by linearity of the spin projection noise level like depicted in Fig. 9.2 and
discussed in Sec. 9.5. The linearity of the observed noise is a strong indication
of the quantum projection noise. For instance, all technical noise sources from
external fields acting on the spin state would contribute to all atoms and lead to
extra noise which scales quadratically with the atomic spin Jx . However, having
support from theoretical estimations on the ratio of atomic to shot noise κ 2 will
always be welcome. This section is devoted to such an estimate.
We wish to utilize the DC-Faraday effect described in Eq. (6.9). If a beam with
linear polarization is propagating along the x-direction (along the macroscopic
spin) the direction of polarization will rotate by an amount
a1 γλ2 Jx
θDC [rad] = − . (9.49)
16πAeff ∆
This is Eq. (6.9) multiplied by the effective transverse area Aeff of the vapour cell
in the numerator and denominator and we exploit the fact that ρAeff L h̂x i = Jx .
From Sec. 9.4 we know that in the rotating frame we expect atomic to shot noise
ratio
aJx
κ2 = a2 Sx T Jx = 2aSx T = 2aSx T θDC [rad]. (9.50)
2
In the last step above we assumed that the a = γλ2 a1 /8πAeff ∆ really is the
correct a to insert in the equation for κ2 . The probe beam does not have cross
section A = Aeff but somewhat smaller (typically around 50% of that area). An
argument in favour of using Aeff anyway is the fact that atoms move in and out
of the beam leading to an effective smaller interaction strength since atoms are
inside the beam on average during time T · A/Aeff . If this model is valid we arrive
at following estimate for κ2
18.6 · P [mW] · T [ms] · a1 (∆) · θDC [deg]

κ2 = . (9.51)
∆blue [MHz]
We converted θDC into degrees instead of radians, P is the power of the probe, T
the time duration of the pulse, ∆blue = −∆ the blue detuning, and the parameter
a1 was defined in Eq. (5.16). We connected Sx = (photon flux)/2 = P/2h̄ω where
ω = 2πν = 2πc/λ is the optical frequency of the laser beam and λ = 852nm is
the wave length. We also inserted L = 3.0cm and Aeff = 6.0cm2 . All parameters
on the right hand side are easy to access experimentally and serve as a good
estimate for the ratio κ2 . In addition to the above equation we should remember
that κ2 is reduced by losses and detector inefficiencies, see Sec. 9.6.
An alternative estimation method is to perform absorption measurements to
find the density of atoms in the vapour cell. This leads to an estimate for J x
which again estimates κ2 = a2 Sx T Jx .
95
CHAPTER 10
Experimental Generation of
Entangled States
In this chapter we demonstrate the generation of entanglement. This is the most

important result of this thesis, the experimental work was done in 2001 and
we devote Sec. 10.1 to the discussion hereof. The results are published in [I].
The experimental setup of this experiment deviates a little from the discussion
of Chap. 4 and we comment this when appropriate. After the 2001 entangle-
ment experiment we decided to rebuild the setup for reasons that we discuss in
Sec. 10.2. On top of this we moved our laboratories twice, once internally at the
Department of Physics and Astronomy at the University of Aarhus, and then
from Aarhus to the Niels Bohr Institute at Copenhagen University.
This new experimental setup has at the time of writing not yet demonstrated
entangled states for various reasons but the problems encountered on the way
have in some cases been very interesting. We describe some of these problems in
Sec. 10.3. In Sec. 10.4 we discuss the present state of the entanglement experi-
ment. We conclude this chapter in Sec. 10.5 with a summary of our results.
10.1 Entanglement Demonstration

Let us now turn to the experimental demonstration of entanglement generation.
The experimental setup is shown in Fig. 10.1, this is a little different than the
setup mentioned in Chap. 4. Here two cells are situated next to each other, they
are not placed in their own magnetic shield as in Fig. 4.5. Also, the timing of laser
96 Chapter 10 - Experimental Generation of Entangled States
532nm 852nm
Sy-detection Verdi V8 Ti:sapph
- Noise monitor
Chopper
Diode lasers
894nm
852nm
Figure 10.1: Experimental setup for entanglement generation. In this experiment

the vapour cells are situated next to each other. The cells can be surrounded
by a solenoid and a magnetic shield (not shown here). The laser settings are as
described in Sec. 4.1 but the timing is controlled by a chopper. A noise detec-
tor monitors the classical noise of the probe laser. After the probe laser from
the ti:sapphire laser has passed the atomic samples its Ŝy -component is mea-
sured. The coil generating the RF-magnetic field for spin state characterization
(Chap. 7) is partly visible above the cells.
pulses is controlled by a chopper. The probe laser is detuned by ∆ = −700MHz,

the duration is Tprobe = 0.45ms, and the power is P = 5.0mW. The optical pump
laser is tuned to the 6S1/2 , F = 4 → 6P1/2 , F = 4 transition and the repump laser
is tuned to the 6S1/2 , F = 3 → 6P3/2 , F = 4 transition as described in Sec. 4.1.
The duration of pumping is Tpump = 0.45ms. The total experiment cycle is 2ms,
it is shown in Fig. 10.2(b). The polarization of the probe laser is 54◦ with respect
to the direction of spin polarization. According to Eq. (6.24) this is the condition
for having no Stark shifted Larmor frequencies. From Eq. (6.19) we also know
that this setting is vulnerable to excess laser noise. For this reason we use a
photo detector for monitoring the classical noise of the probe laser as shown in
Fig. 10.1. If the classical noise increases too much we simply wait for it to settle
down again before continuing experiments.
The magnitude of the spin components Jx1 and Jx2 are measured (on a rel-
ative scale) by the magneto-optical resonance method as described in Chap. 7,
especially by some of the pulsed methods mentioned in Sec. 7.5. The spin co-
herence time is also measured by MORS, in absence of lasers we typically find
T2 = 15 − 30ms. At the time of this experiment we had not developed the DC-
Faraday measurement for direct measurement of Jx1 and Jx2 . The absolute value
of the spins Jx1 = −Jx2 ≡ Jx is estimated by measuring the optical depth α of the
atomic sample with a weak, resonant, non-saturating probe beam. The optical
depth relates input intensity I0 to the output intensity I by I = I0 exp(−α). The
density ρ of atoms can then be found by α = ρσL where L is the sample length
and the relevant cross section σ for our Doppler broadened atoms is estimated
by σ = λ2 /2π · γ/δνD . Here γ is the optical line width, δνD is the Doppler width,
and λ is the optical wave length. With Doppler broadening present, roughly a
10.1. Entanglement Demonstration 97
(a) (b)
Gate on-off
- Signal Oscilloscope HF output
gate (averaging)
Gate on-on
LF-lockin
HF-lockin amplifier LF local oscillator
amplifier
LF-lockin
Pump
Probe
amplifier Pulse sequence
0.45ms 2.0ms
Figure 10.2: (a) The signal road from balanced detectors to noise variance esti-
mation. The photo current passes a signal gate which may include or exclude any
of the two pulses. The HF lockin amplifier demodulates the signal at the Larmor
frequency Ω. The sine and cosine components at Ω from the HF lockin amplifier
are fed into two LF lockin amplifiers demodulating the signal at the cycle fre-
quency 500Hz. The outputs are averaged and squared in an oscilloscope. (b) At
the bottom is shown the laser pulse sequence. The pump and probe lasers both
have pulse durations 0.45ms. The total cycle is 2ms. Above this is shown the
corresponding local oscillator field for the LF lockin amplifiers. Further above is
shown an example of HF lockin output. Mixing these with the LF local oscillator
will lead to estimation of A1 and B1 for a one pulse experiment and of A1 − A2
and B1 − B2 for a two pulse experiment.
fraction γ/δνD of the atoms are on resonance. This method is a bit crude and
hence we should not trust the magnitude in absolute units to better than a factor
of 2-3. But on the relative scale we still estimate to have 5% precision.
The spins must be balanced with opposite orientation, Jx1 = −Jx2 . This
is done by simultaneously measuring MORS for the two samples. The MORS
should vanish when the spins are balanced if the pumping conditions are sim-
ilar and if the Larmor frequencies coincide. We estimate this can be done to
a precision of 5%. Rough adjustments of spin magnitudes are controlled by the
temperature of the vapour cells. Fine tuning is performed by adjusting the power
of the repump laser. The two Larmor frequencies are adjusted to coincide by a
small extra coil next to one of the cells (a black wire is visible on the left cell
in Fig. 10.1). The orientation is also estimated by measuring MORS, for this
we increase the magnetic field by roughly a factor of two to better resolve the
quadratic Zeeman effect.
When the probe light has passed the two atomic samples its Ŝy -component
is measured by a set of balanced noise detectors. The differential photo current
is then handled as shown in Fig. 10.2(a). First, an electronic gate turns on or
off the observation of the two pulses, for CSS noise calibration we only need
the first pulse, and for entanglement estimation we need both, see discussions in
Sec. 9.5. Then the gated photo current is fed into a high frequency (HF) lockin
amplifier. The local oscillator of this is adjusted to the Larmor frequency of the
atomic samples. The two outputs of the HF-lockin amplifier deliver the sine and
cosine components of the photo current, or in other words, a signal dependent on
Jˆy1
0
+ Jˆy2
0
and Jˆz1
0
+ Jˆz2
0
, respectively, see discussion in Sec. 9.4. These signals are
taken to two low frequency (LF) lockin amplifiers the local oscillator of which
is a sine wave running at the chopping frequency 500Hz corresponding to the
experimental cycle time of 2ms. The demodulation at this frequency ensures that
in a two pulse experiment, the value of the two measurements are subtracted from
each other. This is also illustrated in Fig. 10.2(b). Now, by setting the lockin
amplifier in “r”-mode (deliveringPthe magnitude of the P frequency component at
500Hz) we obtain after squaring i (A1 [i] − A2 [i])2 and i (B1 [i] − B2 [i])2 in the
language of Secs. 9.4 and 9.5. This is exactly what we need for entanglement
estimation according to Eq. (9.43). To be complete, for the LF-lockin amplifier
to work properly, we need to integrate over a time duration longer than 2ms
(otherwise a 500Hz frequency component cannot be singled out). This means
that we really average over some pulses (maybe 5-10) before squaring in the
above sums. This is perfectly OK for independent measurements.
Now let us concentrate on the noise results. We start out by one pulse mea-
surements and move the Larmor frequency Ω far away from the local oscillator
frequency at 325kHz. Then we only measure the noise of light, and we check that
the Ŝy -measurement is limited by shot noise by observing the linear dependence
of the noise on the probe power.
Next, we move the Larmor frequencies of the two atomic samples back to
coincide with the local oscillator at 325kHz. The spins are balanced with opposite
orientation and with magnitude Jx . The noise properties of the CSS is now
measured, the result is shown in Fig. 10.3(a). On the abscissa is the magnitude
Jx and on the ordinate we plot the variance of A1 plus the variance of B1 , and
we normalize this to the result for shot noise of light. We see that for low J x
the measured points depend linearly on Jx which is the fingerprint of the atomic
spin projection noise according to the discussion around Eq. (9.39) and Fig. 9.2.
We make a fit to the linear part of the graph which is now our calibration for the
CSS noise level. The non-linear part for higher Jx may arise from non-canceled
back action, pile up of technical noise, etc. We do not know in detail the source
of this extra noise but this is also irrelevant. The important point is the fact that
we reached the linear dependence for our calibration. We note, that the slope of
the linear fit is 0.81 when Jx is measured in units of 1012 . We compare this to
the theoretical value κ2 = a2 Sx T Jx , where a = γλ2 a1 /8πAeff ∆. Inserting the
correct detuning and Aeff ≈ 6.0cm2 we obtain κ2 = 0.83 · Jx [1012 ] which is close
to the measured value. But this excludes the effect of losses and we remember
that our estimate of Jx in absolute units is quite crude. We only conclude that
we find agreement within the right order of magnitude for κ2 .
Next, we turn to the two pulses experiment for demonstrating the entangle-
ment generation. After each pumping pulse we shine two pulses of light through
the samples to measure the spin components Jˆy1 0
+ Jˆy2
0
and Jˆz1
0
+ Jˆz2
0
twice. There
is a 0.5ms delay between these two pulses. The resulting noise of the difference
10.1. Entanglement Demonstration 99
8 (a) (b)
Var(A1-A2) + Var(B1-B2) [to CSS noise]

Var(A1) + Var(B1) [to shot noise] 7 2.0
6
1.5
5
4
1.0
3
2
0.5
1
0 0.0
0 2 4 6 0 2 4 6
Jx [1012] Jx [1012]
Figure 10.3: (a) The measured noise (squares) in a single pulse experiment, the
noise variance is normalized to shot noise. On the abscissa we plot J x , and
for Jx = 0 we measure the shot noise of light. We see that the noise variance
increases linearly with Jx which is the fingerprint of the quantum projection noise
of the atomic spins, see the discussion connected to Eq. (9.39) and Fig. 9.2. The
straight line is a fit to the linear part of the data points, the slope is 0.81. (b)
The measured noise (stars) of the difference between two pulses, here normalized
to the CSS noise (i.e. divided by the straight line of part (a)). The straight line
fit of part (a) has uncertainty in the slope which we recalculate into the error bars
in part (b). The data points should fall below horizontal solid line in order to
demonstrate entanglement, we see this is indeed the case for the points at higher
Jx . The dotted line is the shot noise of a single light pulse, the dash-dotted line is
twice the shot noise. The data points cannot fall below twice the shot noise, the
reasons for them being higher than this level are decoherence, losses, etc. The
lowest data point is (36 ± 7)% below unity.
is plotted in Fig. 10.3(b). Here the abscissa is as in part (a) of the figure but the
ordinate is now normalized to the CSS noise level, i.e. every data point is divided
by the value of the straight line in Fig. 10.3(a). In these units the horizontal
solid line is the boundary we have to be below in order to fulfil the entanglement
criterion (9.3). This corresponds to σ 2 < κ2 as we discussed in Sec. 9.5. We see
that this is indeed the case for the points at higher Jx and we have proved the
generation of entangled states. Since we normalize the data points to the CSS
noise level the uncertainty of the slope in Fig. 10.3(a) contributes to the error
bars of Fig. 10.3(b). The lowest data point is positioned (36 ± 7)% below the
CSS noise level.
The dotted line is the shot noise level of light, and the dash-dotted line is
twice this level. In our two pulse experiment we can never obtain data points
below this level since the noise of Ŝyin for each pulse is uncorrelated to all other
noise sources. In the ideal case with no decoherence we should on the other
hand reach down to this level. The exact reason that we are not at this level is
unknown but losses, decoherence, and technical noise will play some role. The
crucial point is the fact that we are below the horizontal line and demonstrate
the generation of entanglement between the two atomic samples. Since the delay
between the two laser pulses is 0.5ms we know that the life time of our entangled
state is at least that long.
Some comments are in place for the obtained results. As we discussed above,
the fact that the averaging procedure involves several pulses before squaring
is OK if the atoms are independent of each other. From the magneto-optical
resonance methods we can estimate the refreshing rate Γpump of the optical pump
laser. We know that the transverse spins under influence of pumping decay as
Jy (t) = Jy (0) exp(−Γpump t/2). We estimate that Γpump Tpump /2 ≈ 1 so that we
do refresh atoms to a high degree. One may ask whether this level is enough.
The straight line observation of Fig. 10.3(a) is a strong indication that we do
refresh the spin sample sufficiently, we know that noise piling up in atoms scale
quadratically with Jx .
Another issue is the orientation of atoms. We know that for the lower linear
part of Fig. 10.3(a) it is better than 95%. This is measured by MORS in a setting
where the optical pump power is reduced to resolve the quadratic Zeeman effect
(in addition to increasing the magnetic field). We assume that a higher optical
pump power will do a better job. A non-perfect orientation compared to the
100% orientated coherent spin state can be regarded as a thermal excitation and
we expect this state to be more noise than the CSS. We try to deliberately reduce
the orientation and observe the effect on the noise. We do not find any increase
in the noise level, however.
10.2 Changing the Experimental Setup

The experiments of the previous section are performed in the old experimental
setup. We decide to change the experimental setup for several reasons which we
will comment on now.
For the mounting and magnetic shielding of vapour cells we refer to Fig. 10.4
where the old and new setups are shown. In the old setup two vapour cells are
placed adjacent to each other inside the same magnetic shielding making laser
access very cumbersome. In the new setup the shielding and vapour cells are
designed such that laser access is possible from six directions. This setup is
suitable for experiments involving three or four vapour cells (we will discuss such
experiments in Chap. 11).
On the laser side, in the old setup we use a chopper to control the timing
of pulses as we discuss in Sec. 10.1. While being very simple, this setup is not
flexible. We wish to vary the time duration of pumping and probing lasers.
We would also like to vary the delay between various pulses in order to study
decoherence effects. In addition, the new setup with AOMs and EOMs controlling
10.3. Magnetic Field Noise 101
(a) (b)
Laser access
Laser access
Figure 10.4: (a) The old experimental setup. Two vapour cells are placed inside
the magnetic shielding cylinder shown. We only have laser access from one end
and need to place mirrors inside the cylinder (see picture in Fig. 10.1). (b) The
new experimental setup. Here each vapour cell is situated in its own magnetic
shielding with laser access from six different directions (shown with red arrows).
This setup is more suitable for three or four sample experiments.
laser pulses opens up the possibility to shape the temporal profile of the laser
pulse.
What regards the data acquisition we would like to implement the weighted
entanglement estimation methods of Eq. (9.42) rather than the more simple
method of Eq. (9.43). The data acquisition system of the old and new ex-
periments are discussed around Figs. 10.2 and 10.6(a), respectively. The new
acquisition methods also enable us to do more detailed statical analysis on our
data.
We need to control the temperature of vapour cells. In the old setup an
electric heater is placed inside the magnetic shielding. This can in some cases
create magnetic fields disturbing the experiments and heating has to be turned
on and off during measurements. The new setup utilizes water heating/cooling
in an aluminum block as depicted in Fig. 4.5. As we shall see in the next section
this aluminum also creates magnetic noise and we are presently working on a
non-metallic temperature control using air flow.
Moving from Aarhus to Copenhagen also meant a change in environment.
The laboratory magnetic fields in Copenhagen are more noisy than in Aarhus.
10.3 Magnetic Field Noise

For the experiments we need to exclude external fluctuating magnetic fields. If
a static magnetic field is added to our bias magnetic field, the Larmor frequency
Ω = 325kHz will depend on the external field. For entanglement generation we
need to have a stable Larmor frequency for several minutes. In fact, we have
observed stochastic changes of the Larmor frequency at the level of up to 100Hz.

The time scale for these changes is of the order of 1 second or slower. These
variations correspond to magnetic field changes of 0.3 milligauss at the atoms.
The magnetic noise is much more quiet between roughly 1:30am and 4:30am
in the night. This is consistent with the night stand still of the local trains which
are running about 1km away from the laboratory at the Niels Bohr Institute. The
variations has been discovered to be in the vertical component of the magnetic
field which is shielded poorly (only a factor 10-15) by the construction shown in
Fig. 4.5(c). The remedy is to rotate the mounts shown in Fig. 4.5 by 90 degrees,
the shields have a much higher efficiency perpendicular to the curved surface. So
the problem is eliminated but this story tells us in very understandable terms
that our measurements are quite sensitive.
Another problem that we have encountered is fluctuating magnetic fields at
the Larmor frequency. As discussed in Chap. 7, by introducing a time-varying
magnetic field transverse to the x-axis of spin polarization we modulate the trans-
verse components of the spin for atomic spin state characterization. This has
proved very useful, but if fluctuating magnetic fields are present beyond our con-
trol we encounter problems. In Fig. 4.5(a) we show how our vapour cells are
placed on an aluminum block. Below we will see how this aluminum block cre-
ate random fields disturbing our experiments and we will try to estimate the
magnitude of these fields.
With atoms polarized along the x-axis we are sensitive to magnetic field fluc-
tuations along the y- and z-axis. With a magnetic field B present, the Hamil-
tonian describing the effect on the entire atomic spin J reads (to first order)
Ĥ = gF µB B · J, see Eq. (F.4). With Bx being our usual static magnetic field
we get Larmor precession and we consider the rotating frame coordinates Jˆy0 and
Jˆz0 . In absence of decay mechanisms we can easily show that
∂ ˆ0 g F µB Jx
Jy (t) = (+ cos(Ωt)Bz (t) − sin(Ωt)By (t)) ,
∂t h̄ (10.1)
∂ ˆ0 g F µB Jx
J (t) = (− sin(Ωt)Bz (t) − cos(Ωt)By (t)) ,
∂t z h̄
where By (t) and Bz (t) are the fluctuating magnetic fields which we assume for
the moment to be common to all atoms. We wish to integrate the above equations
formally and calculate the variance of Jˆy0 (t) and Jˆz0 (t). To this end we assume that
hBi (t)Bj (t0 )i = SB · δ(t − t0 )δij where i, j = y, z and SB is a constant describing
the spectral noise of the magnetic field. This kind of correlation assumes that the
magnetic noise has fast time dynamics compared to our typical spin evolution
time. Then we obtain
D E D E µ g µ J ¶2
ˆ02 ˆ02 F B x
Jy (t) = Jy (0) + SB t,
h̄
D E D E µ g µ J ¶2 (10.2)
F B x
Jˆz (t) = Jˆz (0) +
02 02
SB t.
h̄
10.3. Magnetic Field Noise 103
0.80
0.75
Atomic to shot noise

0.70
0.65
0.60
0.55 -5
y = 0.507 + 6.4 10 x
0.50
0 1000 2000 3000 4000 5000

Pump probe delay [µs]
Figure 10.5: The measured noise of a single pulse experiment with a single cell
prepared by optical pumping. On the horizontal axis is the delay time between
the pump pulse preparing the spin state and the probe pulse measuring the spin.
The noise level is believed to be close to the CSS level at t = 0, and the linear
increase is attributed to the noise generated by random magnetic fields. For this
experiment θF = 13.9◦ and ∆ = −875MHz. Removing the aluminum block also
removes increase in the noise level.
We see that the variance of Jˆy0 and Jˆz0 increases linearly with time. In Fig. 10.5
we show the results of an experiment confirming this behavior. For a single
spin sample we prepare the coherent spin state as good as we can with optical
pumping. Then we wait time t (called the pump-probe delay) before shining a
pulse measuring Jˆy0 and Jˆz0 . The measured variance is shown in Fig. 10.5 as a
function of t. The units on the vertical axis is the shot noise level of light. We
believe that 0.50 is not far from the CSS noise level and we conclude from the
data that the slope of the increasing noise is 128s−1 in units of the CSS noise.
Since the CSS variance of Jˆy0 and Jˆz0 is Jx /2 we get from Eq. (10.2) the theoretical
slope
³ g µ ´2
F B
Slope = 2Jx SB (10.3)
h̄
in CSS noise units. To calculate our experimental estimate of SB we need to
find Jx . With ∆ = −875MHz and a DC-Faraday measurement of θF = 13.9◦ we
find by Eq. (6.9) that Jx = 1.5 · 1012 . Then we derive SB ≈ 9 · 10−32 Tesla2 s.
The noise in a bandwidth δω is SB δω where δω is measured in rad/s. Taking
the square root and converting to Hz we conclude that √ our approach estimates
magnetic field fluctuations of magnitude 7 · 10−15 Tesla/ Hz.
Our assumption that all atoms experience the same magnetic field is probably
wrong, and accordingly the real fluctuating field is probably higher than our
estimate. In [69] the random magnetic field from a metal filling one half plane
(e.g. x < 0 as in our case) is discussed. The origin of these fields is the thermal
random currents inside the metal. At a certain point in space (x, y, z) with x > 0
their Eq. (10) reads
∞
µ20 ω 2 h̄²0 Im²(ω) tij
Z
dτ hBi (t + τ )Bj (t)iT eiωτ =
−∞ 1 − e−h̄ω/kB T 16πx
(10.4)
3µ2 σ0 kB T
→ 0 δij for i, j = y, z,
32πx
where tij = 3/2 for i = j = y, z. The approximation after the arrow assumes ω to
be much slower than the inverse of the electron relaxation time scale τ in order to
connect the relative permittivity ²(ω) to the DC-conductivity σ0 of aluminum by
simple Drude theory Im²(ω) = σ0 /²0 ω. We also assume the typical energy to be
less than thermal fluctuations h̄ω ¿ kB T . These conditions are easily fulfilled for
ω ≈ 325kHz. According to the above our approximation hBi (t)Bj (t0 )i = SB ·δ(t−
t0 )δij is valid. Inserting relevant values, including x ≈
√ L = 3.0cm, we calculate
√ −12
the theoretical estimate of SB = 1.2 · 10 Tesla/ Hz. We see this number
is more than two orders of magnitude higher than our experimental estimate.
This is probably an effect of us neglecting spatial correlations. Furthermore, the
result for a half plane filled with metal must overestimate the field somewhat.
We can probably√ conclude that the field seen by our atoms is somewhere above
10−13 Tesla/ Hz which still is a small field. We do indeed perform quantum
limited measurements of magnetic fields and further study of ultimate sensitivity
could be interesting. For our entanglement generation we learn that metal should
be avoided close to our atoms.
10.4 Weighted Entanglement Estimation

In this section we describe some aspects of our search for entangled states with
our newer experimental setup. We place the atomic vapour cells in mounts as
depicted in Fig. 10.4(b) (but rotated by 90 degrees). The separation between the
two setups is roughly 30cm. We shine the probe laser as usual to measure the
transverse components of the atomic spins, and we may also direct part of the
probe laser along the direction of the macroscopic spins in order to measure J x
directly. Then we have the full capability to characterize the spin states by the
methods of Chap. 7.
After the probe laser has passed the atomic spins the Ŝy -component is mea-
sured giving photo current i(t). The following data handling is depicted in
Fig. 10.6(a). The HF lockin amplifier gives as outputs sin(Ωt)i(t) and cos(Ωt)i(t)
which are integrated in a home built integrator over the probe pulse duration with
results A and B, see Eqs. (9.32) and (9.33). These numbers are stored in a com-
puter and we can process the data afterward. An example of such processing
is the analysis discussed in connection to Eq. (9.42) and shown in Fig. 10.6(b).
Here we see the result of the second pulse measurement A2 plotted versus the first
10.4. Weighted Entanglement Estimation 105
(a)
- HF-lockin
amplifier
Integrator
Computer
analysis
Integrator
Figure 10.6: (a) Schematic view of the signal processing (compare to

Fig. 10.2(a)). The differential photo current from the detectors is demodulated
in the HF lockin amplifier to give the sine and cosine components at the Larmor
frequency Ω as outputs. These are integrated and we obtain two numbers A and
B bearing information about the spin state, see Eqs. (9.32) and (9.33). For each
probe laser pulse these numbers are stored in a computer and can be used for
further analysis. (b) An example of correlations between the first and second
probe laser pulse giving results A1 and A2 . We plot A2 versus A1 (10.000 points)
and perform a linear fit, this is exactly the method described around Eq. (9.42).
We see the slope α = 0.181 is non-zero, the first pulse result can clearly be used
to predict to some extent the second pulse measurement.
pulse result A1 . If atomic noise is much greater than shot noise, and if everything
else is ideal, these points should be on a common straight line through (0, 0) with
unity slope. In other words, the QND measurement would be perfect and the
two measurement results should be identical. But in the case we describe in this
section we have atomic noise less than shot noise, the scattering of data points
in the Fig. 10.6(b) demonstrates this. We perform a linear fit of the data and
get a slope of α = 0.181 along the lines of Eq. (9.42). The non-zero slope clearly
indicates a correlation between A1 and A2 (and a similar plot can be made for B1
and B2 ). The question is, are these correlations strong enough to demonstrate
entanglement following the criterion (9.3)?
The answer to this question follows from a thorough analysis as described
in Sec. 9.5. First, we must calibrate the noise level of the coherent spin state
(CSS). This is done by measuring the variances Var(A1 ) + Var(B1 ) as a function
of the magnitude Jx of the oppositely oriented macroscopic spins. This is plot-
ted in Fig. 10.7(a) with black squares. The shot and electronic noise has been
subtracted, i.e. we only plot the atomic noise (normalized to shot noise). On the
horizontal axis is the DC-Faraday angle θF which we remember is proportional
to Jx . The solid line is a linear fit through (0, 0) of the black points, and it
0.4
0.6 (a) (b)
0.5 0.3
Atomic to shot noise

0.4
0.2
α
0.3
0.2 1st pulse

2nd pulse 0.1
0.1 Weighted
Fit to CSS noise
0.0 0.0
0 2 4 6 8 0 2 4 6 8 10
θF [deg] θF [deg]
Figure 10.7: Data demonstrating the weighted method of entanglement estima-

tion discussed around Eq. (9.42). (a) We vary the magnitude Jx of the macro-
scopic spin and measure this conveniently by the DC-Faraday rotation angle
θF , see Eq. (6.9). Against this is plotted with black squares the atomic noise
variance of the first pulse, with red triangles the second pulse noise, and with
green stars the weighted method estimate σ 2 of the state generated by the first
pulse. The straight line fit (solid line) is a calibration of the CSS noise level, and
with the green stars above this level our data shows no entanglement. The fact
that the red triangles are above the black squares indicate decoherence, technical
noise pile up, or the like. (b) The experimental fitting parameter α (squares) of
Eq. (9.42) versus θF and the theoretical estimate α = κ2 /(1 + κ2 ) (solid line) in
the ideal case with no losses or decoherence. We see that in some sense we are
only effective to the level of 50%.
serves as calibration of the CSS noise level. For the experimental points shown
we used laser detuning ∆ = −875MHz, probe power P = 4.0mW, and probe
duration Tprobe = 650µs. The probe laser is polarized along the x-axis parallel
to the macroscopic spin direction, we wish to examine the setting where classical
laser noise is least likely to play a role according to Eq. (6.19). The setting of
the optical pump and repump lasers are as described in Sec. 4.1.
We now use Eq. (9.51) to estimate the slope theoretically. We find the ex-
perimental slope to be 48% of the theoretical. If loss of light occurs between
the vapour cells and between the cells and detectors the theoretical level should
be decreased correspondingly, see Sec. 9.6. But a 52% overall loss is more than
expected, a level of 20% would be more acceptable. We conclude either that the
simple theoretical estimate of Sec. 9.7 is good only within a factor ≈ 2, or that
there effectively may be more losses than we expect.
Now, in the language of Sec. 9.5 the straight line of Fig. 10.7(a) calibrates κ 2
and the weighted method of Eq. (9.42) determines the slope α (see Fig. 10.6(b))
and the variance σ 2 of the possibly entangled state. The value of σ 2 is plotted
in Fig. 10.7(a) with green stars and α is plotted in Fig. 10.7(b). Since the green
stars are not below the straight line we have no entanglement in the example
10.5. Discussion of the Results 107
shown, i.e. we do not have σ 2 < κ2 .

The reasons for the lack of entanglement is the subject of current investigation
and at the time of writing it would be mostly speculation to point at a specific
problem. However, we know (without going into details here) that we have
not been aware of the full implications of the Stark shifted Larmor frequency
discussed around Eq. (6.24). Also, the value of the fitting constant α plotted
in Fig. 10.7(b) is roughly 50% of the ideal case theoretical value κ2 /(1 + κ2 ),
see Eq. (9.38). This indicates that the information provided by the first pulse
(A1 , B1 ) is only “half as good” as we expect in the ideal case.
Yet another indication that we are further away from ideal measurements
that we would like is the fact that the noise variance of the second pulse alone
(Var(A2 ) + Var(B2 ) minus shot noise) shown with red triangles in Fig. 10.7(a)
is at a higher level than the first pulse variances. This indicates either technical
noise pile up, decoherence, or maybe non-canceled back action. Further study is
required to understand the decoherence mechanisms behind our data.
But we certainly do conclude that we are not far from observing the entan-
glement in the new setup. The measurement apparatus works well and supplies
us with useful information for understanding the decoherence mechanisms. We
also await the possibility to go to higher κ2 when our hot air heating system
is up running (we removed aluminum heating systems for reasons discussed in
Sec. 10.3).
10.5 Discussion of the Results

Let us discuss and summarize the results of this chapter. First of all, the experi-
mental results given in Sec. 10.1 demonstrate the generation of entangled states
between two macroscopic cesium gas samples. This is the most important result
of this thesis. The created states have in the best case a variance fulfilling
Var(Jˆy1 + Jˆy2 ) + Var(Jˆz1 + Jˆz2 ) ≤ (64 ± 7)% · 2Jx (10.5)
which should be compared to the entanglement criterion (9.3). The 64% = (100
- 36)% arise from the fact that the lowest points of Fig. 10.3(b) are 36% below
the CSS noise level. This experimental estimate is an upper bound since we here
did not take into account the fact that the initial state is the CSS (this is the
difference between Eqs. (9.42) and (9.43)).
Our entanglement lives for at least 0.5ms which is a relatively long life time
for atomic systems. With our old experimental setup this delay is controlled by a
chopper and we do not have much flexibility to test other delays between the first
laser pulse creating the entanglement and the second pulse necessary to verify
the generation of an entangled state.
We note the fact that the entangled state is created on demand, we perform a
measurement with a laser pulse and this inevitably drives the spin samples into
the entangled state. The mean values of Jˆy1 + Jˆy2 and Jˆz1 + Jˆz2 are random,
they will be distributed within the CSS variance and can be extracted from
the integrated photo current of the detectors measuring Ŝy . This is directly
implemented in the newer experiments. The mean values are necessary for future
work like teleportation of atomic states
D (see Chap.
E D11). To create
E entangled states
with definite mean values, e.g. Jˆy1 + Jˆy2 = Jˆz1 + Jˆz2 = 0, requires the
rotation of the spin state by e.g. a magnetic field. This is still to be implemented.
We have chosen to rebuild the entanglement experiment to make it more suit-
able for future teleportation protocols, see Chap. 11. This includes the separation
of vapour cells to larger distances, design and build up of new cell mounts and
magnetic shields, more complicated data acquisition and analysis, and installa-
tion of new vapour cells. This process has involved many technical problems and
challenges, a few interesting ones are discussed in Sec. 10.3. With the new setup
we have reached the CSS noise level for weighted entanglement estimation, see
Fig. 10.7(a), or in other words, equality of the criterion (9.3) has been reached
and generation of entangled states should not be far away. This setup will also
enable us to perform detailed study of the decoherence of entangled states since
we will be very flexible in the timing of laser pulses.
109
CHAPTER 11
Quantum Information Protocols
In this chapter we discuss various protocols for quantum communications pur-

poses, this includes the concept of teleportation which is the transport of an
unknown quantum state from one place (which we call Alice’s place) to another
(which we call Bob’s place). According to the no-cloning theorem [2] a quan-
tum state cannot be copied. As a result, in the teleportation process we are not
allowed, even in principle, to obtain any information about the initial quantum
state to be teleported from Alice to Bob. If we obtain some of this information,
the exact same information cannot be sent to Bob. This fact imposes some re-
strictions on the “handling” of the quantum states and we must follow a certain
protocol. A very important resource in such protocols is the use of entangled
states, just like the ones we demonstrate in Chap. 10.
A protocol for teleporting the state of a spin-1/2 system was discovered in
1993 [70] and different experiments along these lines have been carried out [26,
27, 29]. Teleportation of continuous variables was proposed in 1994 [71], and the
experimental demonstration of continuous variable teleportation for quadratures
of the electromagnetic field was demonstrated in [25]. Parts of the contents of
the present chapter has been published in [IV].
Experimental studies of teleportation will in practical life have some limita-
tions, therefore it is convenient to define the fidelity F as a figure of merit for
the teleportation protocol
®¯2
F = Average ¯ ψ in ¯ ψ out ¯ or F = Average ψ in ¯ ρ̂¯ ψ in ,
¯ ¯ ¯ ¯ ®
(11.1)
¯ in ®
where the input state is described by the wave function ¯ψ and the output state
by wave function |ψ out i or in the case of a mixed state by the density operator
110 Chapter 11 - Quantum Information Protocols
(a) (b) 1.0

(A2, B2)
3 0.8
2
Fidelity
Add 0.6
Cell 3
result
0.4
Cell 1 Cell 2
0.2
1
(A1, B1)
Alice Bob 0.0
0 2 4 6 8 10
Atomic to shot noise (κ2)
Figure 11.1: (a) The teleportation protocol. Alice has two spin samples 1 and
3, the quantum state of sample 3 is unknown and we wish to teleport this to
Bob’s sample 2. First (marked with an arrow containing the number 1) entangle
sample 1 and 2 by a laser pulse. Next measure the joint state of sample 1 and
3 by a similar laser pulse. Last we need by classical communication to transfer
numbers A2 and B2 from Alice to Bob. This information is required to complete
the teleportation. (b) The fidelity F of the teleportation protocol as given by
Eq. (11.5). The F = 1/2 and F = 2/3 limit are reached by κ2 = 1.62 and
κ2 = 3.56, respectively.
ρ̂. The average is performed over input states. If F = 1, then all input states
are teleported perfectly from Alice to Bob. To exceed the limit F = 1/2 for any
coherent state we need quantum entanglement, hence this is a boundary between
classical and quantum communication [72]. If in addition the limit F = 2/3 is
reached we can be sure that the teleported state is the best existing “copy” of
the initial state [73].
11.1 Teleportation of an Unknown Spin State

Let us now turn to the protocols for teleportation of spin states. The analysis
of continuous variables teleportation for quadratures of light in realistic experi-
mental conditions is given in [74], here some calculation methods for Gaussian
continuous variables are also outlined. A proposal for spin state teleportation
was given in [62] and in the following we review the basic principle behind.
Consider Fig. 11.1(a) where three vapour cells containing cesium are drawn.
Cell 1 and 3 are placed at Alice’s site while cell 2 is placed at Bob’s site which
in principle can be far away. We prepare the atomic spins in these three cells in
coherent spin states along the x-axis as usual such that Jx1 = Jx , Jx2 = −Jx , and
Jx3 = −Jx . Then two adjacent cells will be polarized along opposite directions.
The interesting quantum variables are as always Jˆy and Jˆz (we work in the
rotating frame but leave out the primes which will be used for other purposes
below) and the aim now is to teleport an unknown state of sample 3 described
11.1. Teleportation of an Unknown Spin State 111
by operators Jˆy3 and Jˆz3 to sample 2.

The first step is to entangle the spins in sample 1 and 2 by our well known
methods from Chaps. 9 and 10. We assume the atomic to shot noise ratio to
be very large (κ2 À 1) such that we essentially perform perfect measurements.
Firing a laser pulse through cells 1 and 2 yields results
Jˆy1 + Jˆy2 = A1 and Jˆz1 + Jˆz2 = B1 (11.2)
which holds as operator equations for ideal measurements, sample 1 and 2 have
changed from independent coherent states into a highly entangled state with
Var(Jˆy1 + Jˆy2 ) = Var(Jˆz1 + Jˆz2 ) → 0.
The next step is to fire a similar laser pulse through sample 1 and 3 to
perform an ideal measurement of the spin components. We let primes denote the
operators at a time after this second measurement and the primeless operators
refer to operators on spin states before the measurement. We have
Jˆy1 + Jˆy3 = Jˆy1

0
+ Jˆy3
0
= A2 and Jˆz1 + Jˆz3 = Jˆz1
0
+ Jˆz3
0
= B2 . (11.3)
Note that back action noise is piled up in both spins 1 and 3 such that these
individual spins are changed dramatically, i.e. Jˆy1 0
= Jˆy1 + BAN and Jˆy3 0
=
ˆ
Jy3 + BAN and similarly for z-components. Here BAN means back action noise.
The sum of primed and unprimed operators are identical, this is the QND nature
of the measurement caused by the back action cancellation. We thus measure
the properties of the initial state of sample 3 plus the entangled state of sample 1
with sample 2. At the same time Jˆy2 0
= Jˆy2 and Jˆz2
0
= Jˆz2 , there is no interaction
going on at sample 2. Now, the final state of sample 2 can be deduced by
Jˆy2
0
= Jˆy2 = A1 − Jˆy1 = Jˆy3 − (A2 − A1 ),
(11.4)
Jˆz2
0
= Jˆz2 = B1 − Jˆz1 = Jˆz3 − (B2 − B1 ).
Here Jˆy3 and Jˆz3 refer to the state of sample 3 before the firing of the second laser
pulse. The second equalities hold as a consequence of the initial entanglement
between sample 1 and 2, see Eq. (11.2). The third equalities follow from the
result of the second QND measurement, see Eq. (11.3). We see that to complete
the teleportation we need to add the numbers A2 − A1 and B2 − B1 to Jˆy2 0
and
ˆ0
Jz2 , respectively, which can be done by suitable magnetic fields. The numbers
A2 and B2 are completely random and without these Bob has no use of the spin
state of sample 2. For Bob to know these numbers we must establish classical
communication between Alice and Bob. The initial state of Alice’s sample 3 has
been destroyed by pile up of back action noise but has also been recreated in
Bob’s sample 2.
Note, the initial entanglement between samples 1 and 2 could have been
prepared while the cells were sitting next to each other. Then one sample could
be moved far away, and the rest of the protocol could proceed. This approach
of course involves some technical problems (e.g. the magnetic field should be
carried together with the atomic sample). But it underlines the profound nature
of teleportation, when the entanglement has been established the rest is local
measurements and classical communication.
The above considerations are worked out for ideal measurements. For a finite
atomic to shot noise ratio κ2 the analysis is covered by [62] which again relies on
the calculations in [74]. We will just state the result, the fidelity F is given by 1
Áµ ¶
1 1
F =1 1+ + 2 . (11.5)
1 + κ2 κ
We plot this quantity in Fig. 11.1(b). To overcome the threshold F = 1/2 we
need κ2 = 1.62. With κ2 = 3.56 we may break the F = 2/3 limit. We also note
that for κ2 À 1, Eq. (11.5) can be approximated by F ≈ 1 − 2/κ2 . Hence, to
obtain F ≥ 95% we need κ2 ≥ 40. We remind ourselves that these calculations
are results of a Gaussian wave function modeling which is similar to our pure state
considerations for entanglement generation in Sec. 9.3. The only non-ideal factor
included in this approach is the fact that atomic to shot noise is finite (κ 2 6= ∞).
Hence, as experimentalists, given κ2 we should take the above equation as a
theoretical upper limit.
In the experimental demonstration of entanglement generation in Chap. 10
we estimated the variance of the entangled state to be 64% of the CSS noise level,
see Eq. (10.5). In this estimation, we neglected the knowledge about the initial
CSS which leads us to believe we created a state with variance (relative to the
CSS noise level) 1/κ2 rather than the more correct 1/(1 + κ2 ) of Eq. (9.38). If
we assume that our estimate corresponds to and effective 1/κ2 = 64% we obtain
κ2 ≈ 1.6, i.e. we now model our operationally obtained entanglement at the real
κ2 ≈ 2.8 (see Fig. (10.3)) by a virtual but perfect experiment working at κ2 ≈ 1.6.
This effective κ2 is close to the teleportation fidelity of F = 1/2.
The teleportation protocol is a very suitable extension to our experiment, in
addition to the magnetic field for adding A2 − A1 and B2 − B1 we essentially
have two laser beams similar to the one we applied for entanglement generation
in Chap. 10. Hence, in principle, the teleportation protocol with three cells is
straightforward to implement. However, as experimentalists we need to demon-
strate that the teleportation is successful with some fidelity F. For this purpose
we would need to perform measurements on sample 2 after the teleportation has
been completed. But with our rotating frame this is not straightforward, we
cannot measure and characterize a single spin component (Jˆy2 or Jˆz2 ) as we re-
marked in Sec. 4.3. This fact motivated us to consider a four cell protocol which
is described in the next section.
11.2 Entanglement Swapping

Let us consider the setup shown in Fig. 11.2(a). At Alice’s site we place cells 1
and 3, and at Bob’s site we place cells 2 and 4. We create coherent spin states in
1 In Eq. (3) of [62] we substituted 2κ2 → κ2 to account for the rotating frame.
11.2. Entanglement Swapping 113
(a) (b) 1.2

(A3, B3) 3
[Var(Jy2 + Jy4) + Var(Jz2 + Jz4)]/2Jx

(A2, B2) 1.0
2 Add result
0.8
Cell 3 Cell 4
1 0.6
Cell 1 Cell 2 0.4

1 0.2
(A1, B1)
Alice Bob 0.0
0 2 4 6 8 10
Atomic to shot noise (κ2)
Figure 11.2: (a) The four cell entanglement swapping protocol. Alice has two
spin samples 1 and 3, and Bob has samples 2 and 4. First we entangle sample 1
with 2 and sample 3 with 4 by two laser pulses. Next measure the joint state of
sample 1 and 3 by a similar laser pulse. Last we need by classical communication
to transfer numbers A3 and B3 from Alice to Bob. If the measurement is ideal,
Bob can now use this information to rotate samples 2 and 4 into an entangled
state with Jˆy2 + Jˆy4 = 0 and Jˆz2 + Jˆz4 = 0. (b) Given the atomic to shot noise
ratio κ2 the solid line shows the best obtainable variance of the entangled state by
the protocol (calculated numerically). We see that κ2 > 1 is required to observe
entanglement between samples 2 and 4. The dashed line is the naive guess 3/κ 2
for the variance. At large values of κ2 the two lines agree.
all four samples oriented such that Jx1 = Jx4 = Jx and Jx2 = Jx3 = −Jx . Then
any two adjacent samples have opposite orientation and we may perform QND
measurements on any of these pairs. Now we wish to generate entanglement
between samples 2 and 4 without direct interaction between these. This is done
in the following manner where we assume ideal measurements (κ2 À 1).
First fire two laser pulses marked with the arrow (1) in Fig. 11.2(a). These
perform measurements
Jˆy1 + Jˆy2 = A1 and Jˆz1 + Jˆz2 = B1 ,

(11.6)
Jˆy3 + Jˆy4 = A2 and Jˆz3 + Jˆz4 = B2 .
After this step sample 1 is entangled with sample 2 and sample 3 with sample
4. The next step is to shine a similar laser pulse (shown with arrow (2) in
Fig. 11.2(a)) through samples 1 and 3 to measure
Jˆy1 + Jˆy3 = Jˆy1

0
+ Jˆy3
0
= A3 and Jˆz1 + Jˆz3 = Jˆz1
0
+ Jˆz3
0
= B3 . (11.7)
In analogy with the previous section we let the primed operators refer to spin
states after the measurement on samples 1 and 3, and we remember that the
individual samples e.g. Jˆ1y and Jˆy3 are destroyd by pile up of back action noise
while the sum of the two is conserved. Now let us consider our knowledge about
samples 2 and 4. We have
Jˆy2
0
+ Jˆy4
0
= Jˆy2 + Jˆy4 = (A1 − Jˆy1 ) + (A2 − Jˆy3 ) = A1 − A2 − A3 ,
(11.8)
Jˆz2
0
+ Jˆz4
0
= Jˆz2 + Jˆz4 = (B1 − Jˆz1 ) + (B2 − Jˆz3 ) = B1 − B2 − B3 .
This measurement does not involve samples 2 and 4 which is the reason for the
first equalities to hold above. The second equalities follow from the entangle-
ment (11.6) created by the first laser pulses. The third equalities are valid as
a consequence of the second step laser pulse (11.7). The numbers A1,2,3 and
B1,2,3 are completely random, but if Alice sends the values A3 and B3 to Bob
by classical communication he is able to, as the final step of the protocol, add
−A1 + A2 + A3 and −B1 + B2 + B3 to e.g. Jˆy4 and Jˆz4 , respectively. We see
that we then arrive at Jˆy2 + Jˆy4 = 0 and Jˆz2 + Jˆz4 = 0 and a definite entangled
state has been created between samples 2 and 4 even though these samples never
interact directly. This is known as entanglement swapping.
Note, if we compare Figs. 11.1(a) and 11.2(a) the protocols are very similar.
We may consider the four cell protocol as a teleportation experiment where sam-
ple 3 is the unknown quantum state to be teleported to sample 2. In Fig. 11.2(a)
this state just happens to be an entangled state together with sample 4. After
the completion of the protocol it is sample 2 which is entangled with sample 4,
an entangled state has been teleported.
Now, the above considerations are valid for perfect measurements, let us dis-
cuss what happens in the case of a finite atomic to shot noise ratio κ2 . The
protocol works fine in the rotating frame, and to measure e.g. Jˆy1 + Jˆy2 we recon-
sider Eq. (9.33). If we have no prior knowledge about the spin state of samples
1 and 2 and given the results A we could guess
Ã !
Jˆy1 + Jˆy2
Z T
2
√ = A− Ŝyin (t) sin(Ωt)dt . (11.9)
Jx κ 0
This is just a rewriting of Eq. (9.33). The value of Jˆy1 + Jˆy2 is here the number
A plus some fluctuating quantity (the integral). If we calculate the variance of
the above we get (the number A has variance zero)
³ ´
Var Jˆy1 + Jˆy2 1 1
= 2 = 2. (11.10)
Jx a Sx T J x κ
Having three of these kinds of measurements in the protocol leads us to a naive
guess for the uncertainty in the final state of Jˆy2 + Jˆy4 or Jˆz2 + Jˆz4 to be 3/κ2
in units of Jx . This is plotted in Fig. 11.2(b) as a dashed line. But we do not
have to be that naive, in the limit κ2 → 0 this estimate approaches infinity which
is wrong. With κ ¿ 1 we hardly touch the spins and we know that we should
obtain the coherent spin state limit of unity on the vertical axis of Fig. 11.2(b).
We extend the wave function modeling of Sec. 9.3 to the four cell protocol. This
is very cumbersome and we get help from numerical methods. The result is
11.3. Quantum Memory for Light 115
shown as the solid line in the figure, we see that it reaches the correct limits for
κ2 → 0 and κ2 → ∞. The interesting part is that with κ2 > 1 we are in principle
able to detect entanglement between samples 2 and 4. The operational estimate
κ2 ≈ 1.6 of the previous section is well above this limit.
11.3 Quantum Memory for Light

In this section we investigate the possibility to teleport the quantum state of a
light pulse onto atomic spins, and later on teleport this state back to light. Read-
ing the introduction in Secs. 2.1 and 2.2 we definitely see that the polarization
state of a light pulse and the spin state of atoms are very similar, below we show
how the off-resonant interaction can connect the two in a teleportation scheme.
To this end we need a source of entanglement between spin states or between
beams of light. The contents of this section is close to the ideas of Ref. [75]
which discusses similar protocols.
Now, let us assume that we are given a light pulse of duration T in an unknown
quantum polarization state. We describe this pulse by Ŝyin (t) and Ŝzin (t) and the
relevant quantum variable is the collective property
R T in R T in
Ŝy (t)dt Ŝ (t)dt
in
X̂L = √ 0
and P̂L = 0 √ z
in
. (11.11)
Sx T Sx T
This is similar to Eq. (2.10). As a resource of entanglement we assume to have
two ensembles 1 and 2 of spins fulfilling Jˆy1 + Jˆy2 = 0 and Jˆz1 + Jˆz2 = 0, i.e. they
are perfectly entangled. Like Eq. (2.4) we redefine the spins as
Jˆy1 Jˆz1
X̂A1 = √ and P̂A1 = √ ,
Jx Jx
(11.12)
Jˆy2 Jˆz2
X̂A2 = √ and P̂A2 = − √ .
Jx Jx
h i
Now X̂, P̂ = i for both light and the two atomic samples. The entanglement
condition (at time t = 0 reads) X̂A1 (0) + X̂A2 (0) = 0 and P̂A1 (0) − P̂A2 (0) = 0.
In the following we assume no static magnetic field, i.e. we are not in the rotating
frame. We let the incoming light propagate along the z-axis and let it interact
with the spin sample 1. The equations of motion (6.11) and (6.13) will with the
above definitions transform into
X̂Lout = X̂Lin + κP̂A1 (0) and X̂A1 (T ) = X̂A1 (0) + κP̂Lin ,
(11.13)
X̂A2 (T ) = X̂A2 (0) and P̂A2 (T ) = P̂A2 (0).
Sample 2 is unchanged, there is no interaction going on. If we put κ = 1 we see

that the outgoing X̂Lout contains information about the light pulse, X̂Lin , and the
initial atomic state of sample 1, P̂A1 (0). We measure this light with outcome A
such that A = X̂Lin + P̂A1 (0). The remaining information (X̂A1 (0) and P̂Lin ) is
stored in the atoms. If we shine a strong light pulse along the y-direction with
κ2 À 1 we may perform a measurement of X̂A1 (T ) with outcome B, i.e. B =
X̂A1 (0) + P̂Lin . We may now consider the state of sample 2,
X̂A2 (T ) = X̂A2 (0) = −X̂A1 (0) = P̂Lin − B,

(11.14)
P̂A2 (T ) = P̂A2 (0) = P̂A1 (0) = −X̂Lin + A.
Now, if we add the numbers B and −A to X̂A2 and P̂A2 , respectively, we have
completed the teleportation P̂Lin → X̂A2 and −X̂Lin → P̂A2 .
Now, we could reverse the process, assume we have two entangled beams of
light and an unknown atomic quantum spin state. Calculations similar to the
above will shown that we also in this case may teleport the spin state to the light
pulse. Hence, with both the possibility of writing a quantum state of light onto
atoms and reading out this state to another light pulse, we have a protocol for
a complete quantum memory. This result should be seen in connection to our
discussion in Sec. 8.5.
We demonstrate in Chap. 10 how to generate an entangled state between
atomic spins. Also, in our laboratory others have demonstrated entanglement
between two beams of light, see [37]. Hence, in principle we have everything at
hand to implement a full quantum memory. However, the above protocols do
not work in the rotating frame. Experimentally we prefer the rotating frame for
several reasons, see Sec. 4.3, and a useful protocol is still to be found.
An alternative to quantum memory is the quantum cloner. An unknown co-
herent polarization state of light can be optimally cloned onto two oppositely ori-
ented spin states by a single passage of the light through the two atomic samples
followed by a measurement [76]. This protocol requires no initial entanglement
between the two spin samples.
117
CHAPTER 12
Summary and Outlook
In this thesis we study the interaction between polarization states of laser light
and ground spin states of atomic cesium samples. With strong laser pulses and
macroscopic gas samples containing ≈ 1012 atoms our quantum mechanical de-
scription of the collective properties of light and matter becomes very similar to
ordinary position X̂ and momentum P̂ of a single particle. This is also known
as continuous variable quantum systems. In this approximation the atoms and
light are very similar from a mathematical point of view, and with an off-resonant,
dispersive interaction we are able to couple these two different physical systems
together.
In the limit of off-resonant coupling we neglect all absorption effects but
calculate all dispersive effects of the interaction between laser light and the real
hyperfine split ground state of cesium. This leads to an effective Hamiltonian
and to light/matter propagation equations. The calculations are then heavily
supported by experiments of both classical and quantum nature. During this
process we learned that our physical systems of light and atoms really are sensitive
to the quantum fluctuations of each other. In particular we followed the quantum
fluctuations of a squeezed beam of light into atomic degrees of freedom and back
onto light. This study brings optimism for possible future implementations of
e.g. quantum memory.
The off-resonant atom/light interaction also allows us to create entangled
states between two separate macroscopic atomic gas samples. We exploit the
fact that we can perform quantum non-demolition measurements on the joint
spin system of two atomic samples. This will, on demand, drive the spin samples
into an entangled state. We perform a simple theoretical analysis of this and
118 Chapter 12 - Summary and Outlook
we demonstrate the generation of entangled states experimentally. The obtained

entanglement corresponds to a noise reduction of (36 ± 7)% below the classical
limit, and the entangled state lives for at least 0.5ms. Experimental entanglement
generation is the most important result of this thesis.
It is feasible to extend our experimental procedures to teleportation protocols
with three or more atomic samples. Our operationally obtained entanglement
could optimistically lead to teleportation of an unknown quantum state with fi-
delity close to the limit F = 1/2 which cannot be broken by classical means.
A four sample entanglement swapping experiment seems feasible, and our op-
erationally obtained entanglement should be sufficient to observe entanglement
created by a swapping protocol. The present work in the laboratory is directed
along these lines.
In addition to the important results relevant for future quantum information
and communication protocols we learned many aspects of atomic physics for the
practical characterization and utilization of atom/light interactions for different
purposes. With our efforts on upgrading the experimental setup we also expect
to be able to study the dynamical evolution of entangled states.
119
APPENDIX A
Polarization States of the Light

Field
In this appendix we define our notation for Stokes operators, i.e. the operators
suitable to describe the polarization state of the light field. We will also discuss
vacuum noise and losses for Stokes operators.
A.1 The Quantized Radiation Field

The quantization of the radiation field is well described in many text books and
we will here quote results from [77]. The electric field operator can be expressed
r
X h̄ωλ ³ ´
E=i âλ eλ eikλ ·r − â†λ e∗λ e−ikλ ·r . (A.1)
2ε0 V
λ
Here λ runs over all modes, i.e. over all directions in k-space and all polarizations.
The vector kλ describes the direction of propagation, the angular frequency is
given by ωλ = c|kλ |, and the complex unit vector eλ describes the direction of
polarization. The vector eλ is perpendicular to kλ for all λ, or in other words,
it is the transverse part of the field which is described above. The creation and
annihilation operators âλ and â†λ fulfil the commutation relation
h i
âλ , â†λ0 = δλλ0 . (A.2)
We will make a few simplifications to (A.1). We assume that the field is propa-
gating in the positive z-direction and that the transverse extent of the beam is
120 Appendix A - Polarization States of the Light Field
much larger than an optical wavelength. In this case a one-dimensional theory is

sufficient. The transverse cross section will be denoted A and the quantization
volume is then V = A · L, where we quantize the beam over a spatial distance L
along the z-axis. We will also make an overall phase change1 to get rid of the i
in front of E . Now the electric field may be written
r
X h̄ωλ ³ ´
E= âλ eλ eikλ z + â†λ e∗λ e−ikλ z . (A.3)
2ε0 AL
λ
The Hamiltonian describing the radiation field is given by

µ ¶
X 1
ĤR = h̄ωλ â†λ âλ + . (A.4)
2
λ
A.2 Stokes Operators

To introduce the notation of Stokes operators it will suffice to consider only a
single longitudinal mode of (A.3) with wave vector k. We explicitly specify the
directions of polarization along x, y or +, − by
r
h̄ω ¡
[âx ex + ây ey ]eikz + [â†x ex + â†y ey ]e−ikz
¢
Ek =
2ε0 AL
r (A.5)
h̄ω ³ ikz † ∗ † ∗ −ikz
´
= [â+ e+1 + â− e−1 ]e + [â+ e+1 + â− e−1 ]e .
2ε0 AL
In the first line ex and ey are unit vectors along the x- and y-directions with
corresponding creation and annihilation operators âx , ây , â†x , and â†y . In the
second line the unit vectors e±1 are defined by
ex + iey ex − iey
e+1 = − √ and e−1 = √ . (A.6)
2 2
This particular definition ensures that â+ is the annihilation operator of a σ+ -
polarized photon and so on. This choice of linear and circular unit vectors fixes
the relation between linear and circular creation and annihilation operators,
âx − iây âx + iây
â+ = − √ and â− = √ . (A.7)
2 2
Now, the Stokes operators are defined by
1¡ † 1³ † ´
âx âx − â†y ây = − â+ â− + â†− â+ ,
¢
Ŝx = (A.8)
2 2
1¡ † 1 ³ † ´
†
â+ â− − â†− â+ ,
¢
Ŝy = âx ây + ây âx = − (A.9)
2 2i
1 ¡ † ¢ 1³ † ´
Ŝz = †
âx ây − ây âx = â+ â+ − â†− â− , (A.10)
2i 2
1 This is just a convention. This particular way of defining the fields is common in the
quantum optics literature.

A.3. Strong, Linearly Polarized Light 121
where we have the operators in both the x, y and the +, − basis. Ŝx is the number
of photons polarized along ex minus the number of photons polarized along ey ,
√
Ŝy is the number of photons polarized along e+45 = (ex + ey )/ 2 minus the
√
number of photons polarized along e−45 = (ex − ey )/ 2, and finally, Ŝz counts
the photon difference between σ+ and σ− polarization. In the circular basis we
may find it useful to express
Ŝ+ = Ŝx + iŜy = −â†+ â− ,

(A.11)
Ŝ− = Ŝx − iŜy = −â†− â+ .
We may also define the total photon number φ̂ by
φ̂ = â†x âx + â†y ây = â†+ â+ + â†− â− . (A.12)
The Stokes operators defined as above satisfy the usual angular momentum com-
mutation relations
h i h i
Ŝj , Ŝk = i²jkl Ŝl and Ŝz , S± = ±Ŝ± (A.13)
which can be derived from the commutation relations for for the field opera-
tors (A.2). All Stokes operators commute with φ̂.
A.3 Strong, Linearly Polarized Light

In our experiments we often use a linearly polarized beam of light with strong
intensity. If the light is polarized along the e.g. x-axis we may treat operators
âx and â†x as numbers, we make an overall phase choice such that âx → Ax and
â†x → Ax with Ax real. Then we write
A2x
Ŝx ≈ Sx = ,
2
Ax
ây + â†y , (A.14)
¡ ¢
Ŝy ≈
2
Ax
ây − â†y .
¡ ¢
Ŝz ≈
2i
We see that the quantum properties of the Stokes operators of light in this ap-
proximation solely are governed by the light in the y-polarized mode.
A.4 Stokes Operators and Losses

We finalize this appendix by calculating how the Stokes operators transform when
the light is subject to losses of magnitude 1 − η, i.e. the fraction η of the photons
survive. We continue to assume the approximation of a strong, x-polarized beam.
122 Appendix A - Polarization States of the Light Field
0
Ŝx,y,z Ŝx,y,z
√ √
â η â + 1 − η b̂
η
b̂ (vacuum)
Figure A.1: Losses are modeled by a beam splitter with reflection η. This at-
tenuates the input field â and admixes the vacuum component b̂ to the output
field.
As shown in figure A.1, we model the losses by a beam splitter with transmission
η which mixes a vacuum field b̂, b̂† with the incoming field â, â† . We denote
0
the incoming Stokes operators Ŝx,y,z , and the outgoing operators Ŝx,y,z . We
shall also assume that all Stokes operators are normalized to photons per second
(along the lines of App. C). A calculation shows that
Sx0 = ηSx ,
r
η(1 − η)Sx
Ŝy0 (t) = η Ŝy (t) + · V̂y (t), (A.15)
2
r
η(1 − η)Sx
Ŝz0 (t) = η Ŝz (t) + · V̂z (t).
2
We have introduced the vacuum operators V̂y (t) = b̂(t) + b̂† (t) and V̂z (t) =
−i(b̂(t) − b̂† (t)) which fulfil
D E D E
V̂y (t)V̂y (t0 ) = V̂y (t)V̂y (t0 ) = δ(t − t0 ),
D E D E (A.16)
V̂y (t)V̂z (t0 ) = − V̂z (t)V̂y (t0 ) = iδ(t − t0 ).
We see from (A.15) that all the Stokes operators are attenuated by a factor η
and that the quantum variables Ŝy and Ŝz in addition are mixed with an extra
noise source which must be there to preserve commutation relations.
123
APPENDIX B
Spins and Density Operators
Throughout this thesis we will find it useful to express spin operators in terms
of density operators. We will be specifically motivated for describing the spin
state of a collection of atoms in a hyperfine multiplet of the ground state of
cesium. In this appendix we consider a spin operator j within a multiplet such
that j2 = F (F + 1) with h̄ = 1. For cesium we have for the ground state F = 3
or F = 4. The connection between spin and density operators is very useful
for the derivation of the Hamiltonian (5.18) since we found density operators
convenient in the derivation process and spin operators more intuitive in the
final Hamiltonian. Also, as we discuss in Sec. 4.3, atoms are oriented along a
magnetic field leading to Larmor precession. Expressing density operators in
the energy eigen basis (along the x-axis) will reveal the importance of different
frequency components.
B.1 Quantization Along the z-axis
Describing the (2F + 1)-dimensional sub space of a hyperfine multiplet requires

first of all a choice of quantization axis. In the following we will take this as the
z-axis. In the eigen basis of ̂z we know from any elementary book on quantum
mechanics that
p
̂z |mi = m |mi and ̂± |mi = F (F + 1) − m(m ± 1) |mi ,
124 Appendix B - Spins and Density Operators
where ̂± = ̂x ± i̂y . In the language of density operators σ̂jk = |ji hk| we then
get
1 Xp
̂x = F (F + 1) − m(m + 1)(σ̂m+1,m + σ̂m,m+1 ), (B.1)
2 m
1 Xp
̂y = F (F + 1) − m(m + 1)(σ̂m+1,m − σ̂m,m+1 ), (B.2)
2i m
X
̂z = mσ̂m,m , (B.3)
m
Xp
̂+ = F (F + 1) − m(m + 1)σ̂m+1,m , (B.4)
m
Xp
̂− = F (F + 1) − m(m + 1)σ̂m,m+1 . (B.5)
m
By combining the above we can continue to find higher order moments of the
spin in terms of density operators, e.g.
X
̂2z = m2 σ̂m,m , (B.6)
m
Xp
̂2+ = (F − m)(F + m)(F + 1 + m)(F + 1 − m)σ̂m+1,m−1 , (B.7)
m
Xp
̂2− = (F − m)(F + m)(F + 1 + m)(F + 1 − m)σ̂m−1,m+1 . (B.8)
m
B.2 Quantization Along the x-axis
We will also consider spin operators expressed as a function of density operators

when quantized along the x-axis. The main motivation for this is the fact that we
experimentally orient the atomic spins along the x-direction such that h̂ x i ≈ F
(or the opposite direction with −F ). If F = 4 for instance we may in this case
assume that the only important density matrix operators are (when quantized
along the direction of orientation) σ̂44 , σ̂34 , σ̂43 , and σ̂33 since all the rest will be
much smaller. Furthermore, in this basis and in the presence of a magnetic field,
the diagonal elements will be constant, the first off-diagonal elements will rotate
with Larmor frequency Ω, the second off-diagonal elements at 2Ω, an so on.
The easiest way to express spin operators in the rotated basis along x is to
make a cyclic permutation of x, y, z-indices of the spin operators and then refer
B.3. Commutators 125
to known results of Sec. B.1. In this way we obtain for the spin components
X
̂x = mσ̂m,m , (B.9)
m
1 Xp
̂y = F (F + 1) − m(m + 1)(σ̂m+1,m + σ̂m,m+1 ), (B.10)
2 m
1 Xp
̂z = F (F + 1) − m(m + 1)(σ̂m+1,m − σ̂m,m+1 ), (B.11)
2i m
and we also list some higher order components like

1 Xp
̂y ̂z + ̂z ̂y = (F − m)(F + m)(F + 1 + m)(F + 1 − m)
2i m
× (σ̂m+1,m−1 − σ̂m−1,m+1 ) ≈ 0, (B.12)
1 X p
̂x ̂y + ̂y ̂x = F (F + 1) − m(m + 1)(2m + 1)(σ̂m+1,m + σ̂m,m+1 )
2 m
≈ σ jx (2F − 1)̂y , (B.13)
1 Xp
̂x ̂z + ̂z ̂x = F (F + 1) − m(m + 1)(2m + 1)(σ̂m+1,m − σ̂m,m+1 )
2i m
≈ σ jx (2F − 1)̂z , (B.14)
where σ jx is the sign of jx . The last terms above are approximations valid for
|h̂x i| ≈ F . The approximations of Eqs. (B.13) and (B.14) can be derived by
comparison to Eqs. (B.10) and (B.11) and maintaining only the most important
terms.
Under the same approximation we have ̂2x ≈ F 2 and ̂2y ≈ ̂2z ≈ F/2 which is
seen from the relation j2 = F (F + 1). Hence we can also state that
µ ¶
¡2 1
̂x − ̂2y ≈ ̂2x − ̂2z ≈ F F −
¢ ¡ ¢
. (B.15)
2
B.3 Commutators
When deriving equations of motion for spins we need to calculate commutators
between different spin operators. In this section we state the results for the most
important ones used in this thesis. Starting out from the well known
[̂x , ̂y ] = i̂z , (B.16)
and with the cyclic permutations thereof, we derive
̂x , ̂2− = +(̂x ̂z + ̂z ̂x ) − i(̂y ̂z + ̂z ̂y ),
£ ¤
(B.17a)
̂x , ̂2+ = −(̂x ̂z + ̂z ̂x ) − i(̂y ̂z + ̂z ̂y ),
£ ¤
(B.17b)
̂x , ̂2z = −i(̂y ̂z + ̂z ̂y ),
£ ¤
(B.17c)
126 Appendix B - Spins and Density Operators
̂y , ̂2− = −i(̂x ̂z + ̂z ̂x ) − (̂y ̂z + ̂z ̂y ),
£ ¤
(B.18a)
̂y , ̂2+ = −i(̂x ̂z + ̂z ̂x ) + (̂y ̂z + ̂z ̂y ),
£ ¤
(B.18b)
̂y , ̂2z = +i(̂x ̂z + ̂z ̂x ),
£ ¤
(B.18c)
̂z , ̂2− = −2̂2− ,

£ ¤
(B.19a)
̂z , ̂2+ = +2̂2+ ,
£ ¤
(B.19b)
̂z , ̂2z = 0.
£ ¤
(B.19c)
127
APPENDIX C
Continuous Description of Light

and Matter
We will in this section briefly summarize some important aspects of light/matter

interactions. This includes a continuum description of the quantized radiation
field and spatially continuous description of matter. The results in this appendix
are well known to many people (especially theorists) and we may ask, why then
use space to derive the results again? The reason is that (for an experimentalist)
it is difficult to quickly dig up all the results from the literature. Furthermore,
some of these results are so simple that nobody cared to write about it in a text
book! The results in this appendix should help giving an overview to some of the
theory needed in this thesis. The derivations are heavily inspired by [77, 78, 79].
C.1 Continuous Description of the Electromag-

netic Field
We start out from the quantized electric field (A.3) in one dimension. For peda-
gogic reasons we only care about one polarization mode, e.g. the x-polarization
of the electric field. Then we may write
r
X h̄ωλ ¡ ikλ z
+ â† e−ikλ z .
¢
E= âe (C.1)
2ε0 AL
λ
128 Appendix C - Continuous Description of Light and Matter
In this equation the electric field is quantized along a line of length L. Imposing
periodic boundary conditions implies a discrete resolution in k-space with reso-
lution ∆k = 2π/L. The resolution is going toward zero for L → ∞ and this is
the transition we want to make in the following. We need to make the change
X Z
∆k → dk.
To do so, we define the operator

âλ
â(k) = √ with k ≈ kλ . (C.2)
∆k
In this way Eq. (C.1) becomes
X√
r
h̄ωλ ¡
â(kλ )eikλ z + â† (kλ )e−ikλ z
¢
E= ∆k
2²0 L · A
λ
r
X h̄ωλ ¡
â(kλ )eikλ z + â† (kλ )e−ikλ z (C.3)
¢
= ∆k
4π²0 A
λ
r
h̄ω ¡
Z
â(k)eikz + â† (k)e−ikz ,
¢
→ dk (ω = kc).
4π²0 A
The operators â(k) and â† (k) are now continuous as a function of k and they have
units of square root meters. We will now derive the appropriate commutation
relation. Consider the sum
" #
âλ â†λ0
Xh i X Z
†
→ dk 0 â(k), â† (k 0 ) .
£ ¤
1= âλ , âλ0 = ∆k √ ,
λ
0 λ0 ∆k ∆k
These relations hold true whenever the sum of λ0 includes λ, or equivalently, for
all ranges of integration over k 0 including k. We conclude
â(k), â† (k 0 ) = δ(k − k 0 ).

£ ¤
(C.4)
In the continuous description the free field Hamiltonian (A.4) will turn into
µ ¶
1
Z
†
ĤR = dk h̄ck â (k)â(k) + . (C.5)
2
We see that â† (k)â(k)dk should be interpreted as number of photons which have
wave vector in the range [k, k + dk].
C.2 Spatial Description of the Electric field

Up until now the electromagnetic field has been described in reciprocal space,
i.e. in k-space. This is the natural way to deal with Maxwell’s equations. It is
C.2. Spatial Description of the Electric field 129
possible though to give a description in z-space by defining the Fourier transforms

Z ∞ Z ∞
1 1
â(z, t) = √ â(k, t)eikz dk and â† (z, t) = √ â† (k, t)e−ikz dk.
2π −∞ 2π −∞
(C.6)
Note, we now explicitly choose to work in the Heisenberg picture and we re-
mind ourselves of the time-dependence by making the operators a function of t.
Also, the last line of (C.3) assumes positive k-values. Making the Fourier trans-
form above from −∞ to ∞ is thus strictly not correct since it incorporates wave
vectors traveling in the negative z-direction. However, if we make the rotating
wave approximation in some band around a carrier frequency the error is negligi-
ble. Our particular way of defining the Fourier transform leads to the following
commutation relation for the spatial operators
â(z, t), â† (z 0 , t) = δ(z − z 0 ).
£ ¤
(C.7)
†
√
The dimension of â(z, t) and â (z, t) is 1/ length and the physical interpretation
is that â† (z, t)â(z, t)dz is the number of photons in the space between z and z+dz.
The time-dependence of â(z, t) follows Heisenberg’s equation of motion. We
decompose the full Hamiltonian into three parts Ĥ = ĤR + Ĥatoms + Ĥint where
ĤR describes the pure radiation field, Ĥatoms describes the matter independent
of light fields, and Ĥint is the interaction Hamiltonian. Since the light field
commutes with Ĥatoms we get
∂ 1 ∂ 1 1 h
Z Z i
ikz
â(z, t) = √ â(k, t)e dk = √ â(k, t), ĤR + Ĥint eikz dk.
∂t 2π ∂t 2π ih̄
Calculating the commutator with the pure radiation part ĤR leads to
h i Z
â(k, t), ĤR = dk 0 h̄ck 0 â(k, t), â† (k 0 , t) â(k 0 , t) = h̄ck â(k, t),
£ ¤
where (C.4) has been used in the last step. Since

1 ∂â(z, t)
Z
√ ikâ(k, t)eikz dk = ,
2π ∂z
the above equations reduce to
µ ¶
∂ ∂ 1 h i
+c â(z, t) = â(z, t), Ĥint . (C.8)
∂t ∂z ih̄
We have now derived a convenient way to describe how the light field is affected
by atoms through the interaction Ĥint . Now, if we restrict the radiation field to
a narrow band we get a strong motivation for the Fourier transformation of the
electric field into z-space. In this case in Eq. (C.3) we have ω ≈ ω0 where ω0 is
the carrier frequency of e.g. a laser beam. Then
Z r
h̄ω0 ¡
â(k, t)eikz + â† (k, t)e−ikz
¢
E = dk
4π²0 A
r (C.9)
h̄ω0 ¡ †
¢
= â(z, t) + â (z, t) .
2²0 A
We see that â(z, t) can directly be interpreted as the amplitude of the electric
field at position z at time t. A final remark in this section is that in free space
the time and space dependence is trivial, we may here exclude the z-dependence
altogether since â(z, t) = â(0, t − z/c). If we (relative to some point in space of
our choice) make the definition
√
â(t) = c · â(z, t), (C.10)
and the same for â† (t), we get operators that are normalized such that â† (t)â(t)
is the flux of photons at time t which is convenient for the description of a laser
beam incident on a detector for instance. For light in the vacuum state we have
â(t), â† (t0 ) = δ(t − t0 ).

£ ¤
(C.11)
C.3 Continuous Matter Operators

When describing the interaction of light with matter it is convenient to express
both the electric field and the atoms as a function of the space coordinate z. To
illustrate this we assume that we have a collection of two-level atoms coupled
to the light by a dipole transition. The generalization to more atomic levels is
obvious. We will denote the ground and excited state of the atom by |gi and |ei,
respectively.
Making the dipole approximation, the Hamiltonian describing the interaction
is [77] X
Ĥint = − dj · E(Rj ), (C.12)
j
where dj = −erj is the dipole operator for the j’th atom and Rj is the location
of the j’th atom. If we still consider light linearly polarized along x and make
the rotating wave approximation, the interaction Hamiltonian is
r
h̄ω0 X ³ ∗ (j) (j) †
´
Ĥint = d σ̂eg â(zj , t) + dσ̂ge â (zj , t) . (C.13)
2²0 A j
We sum over atoms, zj is the position of the j’th atom, and we have introduced
the (dimensionless) density operator σ̂ge = |gi he|. The dipole moment is defined
by d = e hg| x |ei. The above Hamiltonian is well known from text books (see
(j)
e.g. [80]), for the j’th atom the operator σ̂eg may change the atomic state from
|gi to |ei while the operator â(zj , t) annihilates a photon at position zj . The
strength of this process is governed by the dipole moment d∗ .
Instead of assigning a number j to every atom it is convenient to use the
position z as the index. If ρ is the density of atoms we may define σ̂ge (z, t) such
that
1 X
(j)
σ̂ge (z, t) = σ̂ge (t), (C.14)
ρAdz
zj ∈[z,z+dz]
C.3. Continuous Matter Operators 131
where A is the transverse extent of the atoms. Note that since ρAdz is the number
of atoms in the slice [z, z +dz] the operator σ̂ge (z, t) is the (dimensionless) density
operator for an atom picked out from this slice. With this definition, Eq. (C.13)
can be written
r Z L
h̄ω0 ¡ ∗
d σ̂eg (z, t)â(z, t) + dσ̂ge (z, t)â† (z, t) ρAdz,
¢
Ĥint = (C.15)
2²0 A 0
where the integral runs over the sample length L. We have now obtained a
convenient way to describe the interaction Hamiltonian for light coupled to a
collection of two-level atoms. For instance, propagation equations through the
sample are easily found by combining (C.7), (C.8) and (C.15),
µ ¶
∂ ∂
+c â(z, t) = −igρAσ̂ge (z, t), (C.16)
∂t ∂z
p
where g = ω0 /2²0 h̄Ad. Together with the Heisenberg equation of motion
for σ̂ge (z, t) we have the coupled quantum Maxwell-Bloch equations describing
light/matter interactions. To find the time evolution of the density operator
σ̂ge (z, t) we must calculate commutators of density operators. From Eq. (C.14)
we can in general show that
1
[σ̂µν (z, t), σ̂µ0 ν 0 (z 0 , t)] = (σ̂µν 0 (z, t)δνµ0 − σ̂µ0 ν (z, t)δν 0 µ ) δ(z − z 0 ). (C.17)
ρA
A final remark in this section is the extension of the above into the spin language.
If we consider density operators among ground state levels in a hyperfine multiplet
as in App. B we can express for instance the ̂z operator as a function of z, i.e.
X
̂z (z, t) = mσ̂m,m (z, t). (generalized B.3)
m
This can be done with all spin operators. We still have dimensionless operators
with usual expectation values (i.e. h̂z (z, t)i is in the range −F to F ). The
commutation relation will be modified in the following way
1
[̂x (z, t), ̂y (z 0 , t)] = i̂z (z, t) · δ(z − z 0 ), (C.18)
ρA
that is the commutation relations from Sec. B.3 all hold but with the addition of
the z-dependence and the factor δ(z − z 0 )/ρA.
133
APPENDIX D
Dipole Matrix Elements
In this thesis we need to calculate the strength of different optical transitions

between the ground state 6S1/2 and the excited state 6P3/2 multiplets in cesium.
This is to a great extent a problem of tensor algebra and some of the following
equations can be found in e.g. [81].
D.1 Calculating Matrix Elements

The energy levels of a cesium atom are well described by the total angular mo-
mentum F = J + I where J is the total electronic angular momentum and I is
the nuclear spin. However, the dipole interaction of an optical transition only
interacts with the electronic degrees of freedom and we end up with the problem
of calculating matrix elements of the kind hαJIF mF | rq |α0 J 0 IF 0 m0F i where rq is
a spherical tensor component acting on the J-part of the matrix element, I = 7/2
is the nuclear spin, J = 1/2 is the ground state electron angular momentum and
J 0 = 3/2 is the excited state electronic angular momentum. This situation is well
known and the result is
hαJIF mF |r−q | α0 J 0 IF 0 m0F i = (D.1)

√ F0
½ ¾
0
+m0F −mF F 1
(−1)F hF mF 1q|F 0 m0F i 2F + 1 hα0 J 0 krk αJi ,
J0 J I
where q = 0, ±1 is the tensor component index, hF mF 1q|F 0 m0F i is a Clebsch-

Gordan coefficient, the curly brackets is a 6j-symbol, and hα0 J 0 krk αJi is the
reduced matrix element.
134 Appendix D - Dipole Matrix Elements
We wish to calculate the latter in terms of the inverse life time of the 6P3/2
levels. The easiest is to consider the upper F 0 = 5, m0F = 5 state which can
only decay by spontaneous emission to the lower F = 4, mF = 4 state. We will
first calculate the probability of emitting a photon along the direction k with
polarization e(j) , where we parametrize k and e(j) , j = 1, 2 by
     
k sin θ cos φ sin φ − cos θ cos φ
k =  k sin θ sin φ  , e(1) = − cos φ , e(2) =  − cos θ sin φ  . (D.2)
k cos θ 0 sin θ
Note that these three vectors are mutually orthogonal and real. To go on we
need the interaction Hamiltonian between the radiation field and the atom. This
is r
X h̄ωλ (j) ³ ´
Ĥint = er · eλ âλj eikλ ·r + â†λj e−ikλ ·r (D.3)
2²0 V
λ,j
which is simply minus the scalar product of the quantized radiation field (A.1)
and the dipole operator d = −er of one atom. The two polarizations are indexed
by j and λ is running over all directions in space.
With the initial state |ii = |α0 J 0 IF 0 m0F i |nkj = 0i characterizing the up-
per state with J 0 = 3/2, F 0 = 5, m0F = 5, and no photons in the mode
propagating along k with polarization along e(j) , and the final state |f i =
|αJIF mF i |nkj = 1i where J = 1/2, F = 4, mF = 4, and one photon is present
in the mode kj, the transition matrix element is given by (in the dipole approx-
imation)
r
D ¯ ¯ E h̄ωλ D ¯ ¯ E
f ¯Ĥint ¯ i = e αJIF mF ¯e(j) · r¯ α0 J 0 IF 0 m0F
¯ ¯ ¯ ¯
2² V
r0 (D.4)
h̄ωλ 0 0 (j)
= −e hαJI44 |r−1 | α J I55i (e · e+1 ).
2²0 V
In the last equality we use the fact that only the r−1 component will contribute
to the ∆mF = +1-transition and that we can write r = −e+1 r−1 +e0 r0 −e−1 r+1 .
Now, the transition rate Γ is found by Fermi’s golden rule
2π ¯¯D ¯¯ ¯ E¯2
Γ= ¯ f ¯Ĥint ¯ i ¯ ρ(E). (D.5)
¯ ¯
h̄
The density ρ(E) of final photonic states is
V · E 2 · dΩ V · ω 2 · dΩ
ρ(E) = = . (D.6)
(2πh̄c)3 (2πc)3 h̄
Thus we find the rate into the solid angle dΩ along k with polarization e(j)
2π h̄ωe2 ¯¯ (j) V ω 2 dΩ
µ ¯2 ¶
0 0 2
Γ(Ω) = ¯e · e+1 ¯ |hαJI44 |r−1 | α J I55i|
¯
h̄ 2²0 V (2πc)3 h̄
3 2
(D.7)
ω e ¯ ¯ 2 2
¯ (j) 0 0
= ¯e · e+1 ¯ |hα J krk αJi| dΩ,
¯
32π 2 h̄c3 ²0
D.2. Dipole Coupling Constants 135
where we in the last step used Eq. (D.1) with appropriate values for F , mF , F 0 ,
m0F , J. and J 0 (the Clebsch-Gordan coefficient is unity while the 6j-symbol is
1/6). Now, the total decay rate is found by summing over j = 1, 2 and integrating
Ω over 4π. By using the definitions (D.2) and (A.6) we find
i cos θ
e(1) · e+1 = √ eiφ and e(2) · e+1 = √ eiφ , (D.8)
2 2
and we deduce
¯
¯ (1)
¯2 ¯ ¯2 1 + cos2 θ
¯e · e+1 ¯ + ¯e(2) · e+1 ¯ = ⇒
¯ ¯ ¯
2
Z µ¯ ¯2 ¯ ¯2 ¶ (D.9)
¯ (1) 8π
¯e · e+1 ¯ + ¯e(2) · e+1 ¯ dΩ = .
¯ ¯ ¯
4π 3
With this result at hand we have the total decay rate
e2 ω 3 2
γ= |hα0 J 0 krk αJi| , (D.10)
4π²0 h̄c 3c2
and isolating the reduced matrix element we get

2
¯ 6P3/2 krk 6S1/2 ¯2 = 3c γ .
¯ ®¯
(D.11)
αω 3
Here γ = 2π · 5.21MHz is the FWHM line width of the 6P3/2 states in cesium
measured in radians per second, and ω is the angular frequency of the 852nm
D2-line. c is the speed of light and α is the fine structure constant. All matrix
elements can now be deduced in absolute units from (D.1) and (D.11).
D.2 Dipole Coupling Constants

± 2 +
In Sec. 5.2 we specifically need to calculate numbers like (gF,m;F 0 ,m0 ) and (gF,m;F 0 ,m0 )·
− ±
(gF,m;F 0 ,m0 ), where gF,m;F 0 ,m0 is defined in the discussion around Eq. (5.5). The
calculation is straightforward from the above results and from the Clebsch-
Gordan coefficients and 6j-symbols of Tab. D.1. We get for the squared coupling
constants


 240m2 ∓ 1200m + 1440 F = 3, F 0 = 2
−315m2 ∓ 315m + 3780 F = 3, F 0 = 3


2

cγ λ 1 75m2 ± 675m + 1500 F = 3, F 0 = 4

± 2
(gF,m;F 0 ,m±1 ) = 2 .
4A 2π 3360   35m ∓ 245m + 420 F = 4, F 0 = 3
−147m2 ∓ 147m + 2940 F = 4, F 0 = 4



112m2 ± 1232m + 3360 F = 4, F 0 = 5

(D.12)
For the cross terms we get for F = 3
+ − cγ λ2 p
(gF,m−1;F 0 ,m ) · (gF,m+1;F 0 ,m ) = (3 + m)(4 + m)(3 − m)(4 − m)
4A 2π 
240 F = 3, F 0 = 2 (D.13)
1 
× −315 F = 3, F 0 = 3 ,
3360 
75 F = 3, F 0 = 4
and for F = 4 we have
+ − cγ λ2 p
(gF,m−1;F 0 ,m ) · (gF,m+1;F 0 ,m ) = (4 + m)(5 + m)(4 − m)(5 − m)
4A 2π 
35 F = 4, F 0 = 3 (D.14)
1 
× −147 F = 4, F 0 = 4 .
3360 
112 F = 4, F 0 = 5
It is worth mentioning a few sum rules. We easily find

X
± 2 cγ λ2 8 ∓ m
(gF,m;F 0 ,m±1 ) = for F = 3, (D.15)
4A 2π 4
F0
X
± 2 cγ λ2 8 ± m
(gF,m;F 0 ,m±1 ) = for F = 4, (D.16)
4A 2π 4
F0
X
+ −
(gF,m−1;F 0 ,m ) · (gF,m+1;F 0 ,m ) = 0 for F = 3 and F = 4. (D.17)
F0
D.2. Dipole Coupling Constants 137
hF, m, 1, 1|F 0 , m + 1i
F’=2 F’=3 F’=4 F’=5
√ √ √
(2−m)(3−m) (3−m)(4+m) (4+m)(5+m)
F=3 √
42
− √
24
√
56
0
√ √ √
(3−m)(4−m) (4−m)(5+m) (5+m)(6+m)
F=4 0 √
72
− √
40
√
90
hF, m, 1, 0|F 0 , mi
F’=2 F’=3 F’=4 F’=5
√ √
(3−m)(3+m) (4−m)(4+m)
F=3 − √ √m √ 0
21 12 28
√ √
(4−m)(4+m) (5−m)(5+m)
F=4 0 − √ √m √
36 20 45
hF, m, 1, −1|F 0 , m − 1i
F’=2 F’=3 F’=4 F’=5
√ √ √
(2+m)(3+m) (4−m)(3+m) (4−m)(5−m)
F=3 √
42
√
24
√
56
0
√ √ √
(3+m)(4+m) (5−m)(4+m) (5−m)(6−m)
F=4 0 √
72
√
40
√
90
F0
½ ¾
F 1
3/2 1/2 7/2
F’=2 F’=3 F’=4 F’=5
p p p
F=3 − 1/28 3/112 − 5/336 0
p
F=4 0 1/12 − 7/432 1/6
Table D.1: Relevant Clebsch-Gordan coefficients and 6j-symbols for the 6S 1/2 →
6P3/2 transition in cesium.
139
APPENDIX E
Photo Detection Theory
In this appendix we briefly summarize the most important aspects of photo de-
tection theory and we tailor the theory to be directly applicable to our needs
in this thesis. Photo detection is well covered in the literature, see e.g. [82, 83].
Our experiments are made with balanced detection setup as shown in Fig. E.1.
This is practically a homodyne measurement which has been described in [84]
and results of the sections below follow directly from the approach used in this
reference.
E.1 Stokes Vector Detection

Let us assume that the detectors depicted in Fig. E.1 are infinitely fast and that
they deliver a current pulse eδ(t − t0 ) if a photon is incident on the detector
at time t0 , where e is the unit electrical charge released by the photon. If the
detector quantum efficiency is denoted ηd we have
D E D E
hi1 (t)i = eηd â†+45◦ (t)â+45◦ (t) and hi2 (t)i = eηd â†−45◦ (t)â−45◦ (t)
D E
⇒ hi(t)iSy = hi1 (t) − i2 (t)i = 2eηd Ŝy (t) .
(E.1)
The Sy index reminds us of the detector setup as shown in Fig. E.1. The spectrum
Φ(ω) of the photo current i(t) is defined by
Z ∞
Φ(ω) = hi(t)i(t + τ )i eiωτ dτ, (E.2)
−∞
140 Appendix E - Photo Detection Theory
i1(t)
√
â+45◦ = (âx + ây )/ 2 i(t)
-
âx , ây λ/2
i2(t)
√
â−45◦ = (âx − ây )/ 2
Figure E.1: Detector system for measuring the Ŝy -component of light. The
incident light described by the field operators âx and ây is rotated by a λ/2-plate
and split on a polarizing beam splitter. The two photo diodes will measure the
±45◦ -components of the incident field and the differential photo current i(t) =
i1 (t) − i2 (t) is sensitive to Ŝy (t).
and to calculate this spectrum we need the correlation function

D E D E
hi(t)i(t + τ )iSy = e2 ηd φ̂(t) δ(τ ) + 4e2 ηd2 : Ŝy (t)Ŝy (t + τ ) : , (E.3)
where the colons denote normal and time ordering of field operators, i.e. the
Stokes operators Ŝy (t) and Ŝy (t + τ ) should be written in terms of field operators
â±45◦ , â†±45◦ or âx,y , â†x,y and all daggers should be moved to the left, and time
must increase toward the center. Here φ(t) ˆ is the total photon flux hitting the
two detectors. The first term of (E.3) describes the possibility to count the same
electron twice when asking about the current. It only contributes at τ = 0 for
an infinitely fast detector and is known as shot noise of the light. The second
term is responsible for correlations between photo electrons arising from other
reasons, i.e. Ŝy at different times may be correlated if the light beam in the past
passed through some medium that affected the light. The correlation function
reads in the x, y-basis for τ > 0
hi(t)i(t + τ )iSy = e2 ηd â†x (t)âx (t) + â†y (t)ây (t) δ(τ )
®
+ e2 ηd2 â†x (t)â†x (t + τ )ây (t + τ )ây (t)

© ®
+ â†y (t)â†y (t + τ )âx (t + τ )âx (t)

®
(E.4)
† †
®
+ âx (t)ây (t + τ )âx (t + τ )ây (t)
+ â†y (t)â†x (t + τ )ây (t + τ )âx (t) ,
®ª
and for τ < 0 we just interchange t and t + τ above. This expression is unap-
proximated but quite annoying to use since we from the light/matter interaction
equations get expressions for e.g. Ŝy and not for the normal and time ordered
field operators directly. But if we make the approximation of a strong linearly
polarized beam of light along the x-axis (see App. A.3) we have
1 ³ ´ 1 ³ ´
φ̂(t) = A2x , ây (t) = Ŝy (t) + iŜz (t) , â†y (t) = Ŝy (t) − iŜz (t) ,
Ax Ax
(E.5)
E.2. Our Experimental Case and Detection 141
and inserting this into (E.4) we get for τ > 0
hi(t)i(t + τ )iSy ≈ 2e2 ηd Sx δ(τ )

D E D E Dh iE 
Ŝy (t)Ŝy (t + τ ) + Ŝy (t + τ )Ŝy (t) Ŝy (t + τ ), Ŝz (t)
+ 4e2 ηd2  − ,
2 2i
(E.6)
and for τ < 0 we interchange t and t + τ which obviously only has an effect on
the last term. We could as well have chosen to measure the Ŝz component of
light which would be implemented by replacing the λ/2-plate in Fig. E.1 with
a (suitably adjusted) λ/4-plate. Going through all the above results once again
would then yield (for τ > 0)
D E
hi(t)iSz = 2eηd Ŝz (t) , (E.7)
hi(t)i(t + τ )iSz ≈ 2e2 ηd Sx δ(τ )
D E D E Dh iE 
Ŝz (t)Ŝz (t + τ ) + Ŝz (t + τ )Ŝz (t) Ŝy (t), Ŝz (t + τ )
+ 4e2 ηd2  − .
2 2i
Equations (E.6) and (E.7) may be a little cumbersome but they have the ad-
vantage that they are expressed directly in terms of Stokes operators Ŝy and
Ŝz .
E.2 Our Experimental Case and Detection

We will go on with the above equations and show that in our experimental
conditions theyD can be E simplified.
D E First of all, if the light is in the coher-
ent state with Ŝy (t) = Ŝz (t) = 0, the y-polarized component of light is
in
the †vacuum ® state. In this case (C.11) holds and we can ® easily show that
ây (t)ây (t0 ) = δ(t − t0 ) while hây (t)ây (t0 )i = â†y (t)â†y (t0 ) = â†y (t)ây (t0 ) =
®
0. Under the approximation D of strong

E linearly
D polarized
E light along the x-
axis we calculate further Ŝy (t)Ŝy (t0 ) = Ŝz (t)Ŝz (t0 ) = Sx /2 · δ(t − t0 ) and
Dh iE
Ŝy (t), Ŝz (t0 ) = iSx · δ(t − t0 ). In this case we have for perfect detector
efficiency ηd = 1
hi(t)i(t + τ )iSy = hi(t)i(t + τ )iSz = 2e2 Sx δ(τ ), (E.8)
and we stress that this is only valid for a light in the coherent state incident on
a detector with 100% efficiency. We may assume that the light emitted from our
lasers are in the coherent state so the above considerations are valid for light
before an atomic sample, not after. The delta function δ(τ ) ensures that the
spectrum (E.2) is white, i.e. independent on frequency ω.
Now, we may also produce squeezed light experimentally, and the character-
ization and detection of this is a long story [83, 85, 60]. We make a simplified
description assuming that the bandwidth of squeezing is infinite, i.e. the spec-
trum (E.2) measured with squeezed light incident on a detector is independent
on frequency. This is a valid approximation under the circumstances of this
thesis since the bandwidth of squeezing is much larger than the dynamical band-
width of atoms interacting with the light. A result of this assumption is that
hi(t)i(t + τ )iSy ,Sz must be proportional to δ(τ ) and this is then also required for
the right hand side Dh of Eqs. (E.6)
iE and (E.7). For the commutator we then only
have the choice Ŝy (t), Ŝz (t0 ) = iSx δ(t − t0 ).
Now, we know that we experimentally would measure hi(t)i(t + τ )iSy = ²y ·
2
2e Sx δ(τ ) and hi(t)i(t + τ )iSz = ²z · 2e2 Sx δ(τ ) where ²y and ²z characterize the
noise of a Ŝy or Ŝz measurement relative to the case of coherent state light (E.8).
Putting all the above together we conclude that
Dh iE
Ŝyin (t), Ŝzin (t0 ) = iSx δ(t − t0 ), (E.9)
D E Sx
Ŝyin (t)Ŝyin (t0 ) = ²y · δ(t − t0 ), (E.10)
2
D E Sx
Ŝzin (t)Ŝzin (t0 ) = ²z · δ(t − t0 ), (E.11)
2
which is valid for all input fields before atomic samples encountered in this thesis,
squeezed or not. The squeezing parameters must fulfil ²y · ²z ≥ 1. If ²y < 1 and
²z > 1 we have squeezing in Ŝy and anti-squeezing in Ŝz , and vice versa.
The physical interpretation of the delta correlations of the above equations
is that polarization properties of photons measured at time t and of photons
measured an infinitely small time step later are completely independent of each
other. This is obvious for the ây mode in the vacuum state (²y = ²z = 1). For
the squeezed states photons must be correlated to each other, but the above just
states that correlated photons arrive in pairs at exactly the same time (i.e. for
squeezing in Ŝy with ²y < 1 two photons may be correlated such that they will
click in two different detectors of Fig. E.1 giving no contribution in the differential
photo current i(t)).
The situation is completely different if the light has passed an atomic sam-
ple with slow time dynamics, then Ŝy (t) and Ŝz (t + τ ) may posses information
about non-commuting observables of the atomic sample and (E.9) would not
hold. However, for a very specific case (which is encountered in our experiments)
of Ŝy , Ŝz transforming like
Z
Ŝyout (t) = Ŝyin (t) + dt0 L(Ŝzin (t0 ), . . .),
(E.12)
out in
Ŝz (t) = Ŝz (t),
in
where L can be a function of manyDhvariables including iE atomic
Dh variables and iEŜz
but excluding Ŝyin we easily find Ŝyout (t), Ŝzout (t0 ) = Ŝyin (t), Ŝzin (t0 ) =
E.2. Our Experimental Case and Detection 143
iSx δ(t − t0 ) and we end up with

³D E D E´
hi(t)i(t + τ )iSy = 2e2 Ŝyout (t)Ŝyout (t + τ ) + Ŝyout (t + τ )Ŝyout (t) ,
³D E D E´ (E.13)
hi(t)i(t + τ )iSz = 2e2 Ŝzout (t)Ŝzout (t + τ ) + Ŝzout (t + τ )Ŝzout (t) ,
which is valid for 100% effective detectors and for the very special case of (E.12).
To include less than unity efficiency detection is straightforward if needed. We
also note that including ηd < 1 in the description above or including it as a loss
along the lines of App. A.4 makes no difference. We see that all the trouble
caused by the normal and time ordering in photo detection theory in some cases
can be boiled down to a simple expression involving only Stokes operators.
We conclude this appendix with a useful result for the spectrum (E.2) of a
measurement of Ŝyout (or similarly Ŝzout ). This is close to what is known in the
literature as the Wiener-Khintchine√ R theorem [86]. If the Fourier transformed
Stokes operator Ŝyout (ω) = 1/ 2π Ŝyout (t)eiωt dt has a correlation function on
the form D E
Ŝyout (ω)Ŝyout (−ω 0 ) = f (ω)δ(ω − ω 0 ), (E.14)
then the spectrum (E.2) of the photo current i(t) can be found by Fourier trans-
forming (E.13) and we get
2e2
Φ(ω) = √ [f (ω) + f (−ω)] . (E.15)
2π
The front factor is irrelevant but the fact that there are two terms with opposite
signs on ω is a result of the normal ordering of light operators. This result is
used in Chap. 8.
145
APPENDIX F
The Quadratic Zeeman Effect
The quadratic Zeeman effect is well understood [87], we will outline the important
results in this appendix. An alkaline atom in an external magnetic field B is
described by the Hamiltonian
µJ µI
Ĥ = haI · J − J·B− I · B, (F.1)
J I
where J describes the angular momentum of the outermost electron, I is the
nuclear spin, a describes the strength of the magnetic dipole interaction between
the electronic and nuclear spin, and h is Planck’s constant. The magnetic moment
of the electron (for an s-electron with L = 0) is µJ = −1.0011596521869(41)µB .
The value for the nuclear moment in cesium is µI = 2.582025(4)µN . Thus, the
last term in (F.1) always gives a minor correction compared to the second term,
but the relative strength between the first and second terms depends on the
magnetic field B.
The exact solution for the energy E to the above Hamiltonian is
r
hνhfs µI hνhfs 4m
EF,m = − − Bm ± 1+ x + x2 , (F.2)
2(2I + 1) I 2 2I + 1
where ± is used for F = I ± 1/2, m is the magnetic quantum number (quantized
along the direction of the magnetic field), B = |B|, and the hyperfine splitting
νhfs relates to a by hνhfs = ha
2 (2I + 1). The parameter x describes the relative
strength between the Zeeman effect and the hyperfine splitting:
(−µJ /J + µI /I)B
x= . (F.3)
hνhfs
146 Appendix F - The Quadratic Zeeman Effect
For weak fields m describes the projection of the total angular momentum F =
I+J (the total angular momentum is denoted J in the rest of this thesis, with the
obvious possibility of confusion!). We see from (F.2) that for small field strengths
or very strong fields, the energy depends linearly on B. In the intermediate region
the situation is quite non-linear. All our experiments are performed in the weak
field regime with x ≈ 3 · 10−4 . Here a linear approximation is very good, but it
is important to calculate also the second order contribution.
We calculate the separation of adjacent sub-levels starting out by expand-
ing (F.2) to first order in the magnetic field strength B (leaving out the constant
shift independent of B). With the standard convention
(1)
EF,m = gF µB Bm, (F.4)
we get for cesium with nuclear spin I = 7/2

µ ¶
1 µI −µJ /J + µI /I
gF = − ±
µB I 2I + 1
( (F.5)
0.250390 for F = 4
= .
−0.251194 for F = 3
These two numbers differ in magnitude by approximately 0.3%. Thus we have a

slightly higher separation between levels for the case of F = 3 than for F = 4.
To calculate the quadratic Zeeman shift, it will suffice to do the approximation
µI = 0. In this case we may write the first order expansion of (F.2) as hνL ≡
J /J
Em+1 − Em ≈ −µ 2I+1 · B, and we then easily derive to second order
µ ¶
Em+1 − Em νL
= νL 1 − (2m + 1) . (F.6)
h νhfs
This equation describes the transition frequency between the m’th and the (m +
1)’th level. As is described in Sec. 7.1 we perform spectroscopy on transitions
with ∆m = ±1, see Fig. 7.2(b). The separation νQZ caused by the quadratic
Zeeman effect between two lines will thus be
2νL2
νQZ = . (F.7)
νhfs
Most of our experiments have νL in the vicinity of 325kHz corresponding to a
magnetic field of a little less than 1 Gauss. With the cesium hyperfine splitting
being νhfs = 9.1926GHz we get a quadratic Zeeman splitting of 23Hz.
147
APPENDIX G
Spin Decay and Langevin Forces
In this appendix we derive the correlation function for the Langevin noise terms
F̂y and F̂z used in Chap. 8. These Langevin forces are necessary in presence of
the decay terms of Eqs. (8.3) and (8.4). We may take these two equations and
leave out the terms from the coherent evolution of Heisenberg equations, i.e.
∂ ôut
J (t) = −ΓJˆy (t) + F̂y (t), (G.1)
∂t y
∂ ôut
J (t) = −ΓJˆz (t) + F̂z (t). (G.2)
∂t z
We assumeDthe reservoir E to have no memory, i.e. the correlation functions can

be written F̂i (t)F̂j (t ) = kij δ(t − t0 ). Integrating suitable combinations of the
0
above equations over a small time step ∆t will lead to the different kij ’s. Starting
with kyz we get to first order
D E D E Dh iE
i Jˆx = i Jˆx (t + ∆t) = Jˆy (t + ∆t), Jˆz (t + ∆t)
Dh iE Z t+∆t Z t+∆t Dh iE
= (1 − 2Γ · ∆t) Jˆy (t), Jˆz (t) + F̂y (t0 ), F̂z (t00 ) dt0 dt00
t t
D E
= i(1 − 2Γ · ∆t) Jˆx + (kyz − kzy )∆t
D E
⇒ (kyz − kzy ) = 2iΓ Jˆx ,
(G.3)
148 Appendix G - Spin Decay and Langevin Forces
where we have assumed the macroscopic mean value of Jˆx to be independent

on time for a well oriented sample. Note, that the quantity (kyz − kzy ) changes
sign if the macroscopic spin reverses direction. We could also have calculated the
anti-commutator above,
¿h i À ¿h i À
Jˆy (t + ∆t), Jˆz (t + ∆t) = (1 − 2Γ · ∆t) Jˆy (t), Jˆz (t) + (kyz + kzy )∆t.
+ +
(G.4)
But since Jˆy Jˆz + Jˆz Jˆy = (Jˆ+
2
− Jˆ−
2
)/2i and the expectation value of Jˆ+
2
or Jˆ−
2
is
zero for a completely polarized sample we have kyz + kzy = 0 to a high precision.
We proceed with kzz ,
D E D E Z t+∆t Z t+∆t D E
Jˆz2 (t + ∆t) = (1 − 2Γ · ∆t) Jˆz2 (t) + F̂y (t0 )F̂z (t00 ) dt0 dt00
t t
D E
ˆ2
= (1 − 2Γ · ∆t) Jz + kzz ∆t
D E
⇒ kyy = kzz = 2Γ Jˆz2 = ΓJx ,
(G.5)
where we again assume time independence of the variance of Jˆz (it is given by
its steady state value which in our case means the coherent spin state since Γ is
governed by optical pumping to this state. The last equality reflects that. We
also stated the same result for kyy which is derived in a similar fashion. We
conclude this appendix by repeating the results,
D E D E
F̂y (t)F̂y (t0 ) = F̂z (t)F̂z (t0 ) = Γ · |Jx | · δ(t − t0 ),
D E D E (G.6)
F̂y (t)F̂z (t0 ) = − F̂z (t)F̂y (t0 ) = iΓ · Jx · δ(t − t0 ),
where Jx is the mean value of the macroscopic spin Jˆx counted positive along
the x-axis.
149
Bibliography
[1] J. M. Radcliffe, J. Phys. A 4, 313 (1971).

[2] W. K. Wootters and W. H. Zurek, Nature 299, 802 (1982).
[3] N. Gisin, G. Ribordy, W. Tittel, and H. Zbinden, Rev. Mod. Phys. 74, 145
(2002).
[4] M. Hillery, Phys. Rev. A 61, 022309 (2000).
[5] T. C. Ralph, Phys. Rev. A 61, 010303 (2000).
[6] N. J. Cerf, M. Levy, and G. V. Assche, Phys. Rev. A 63, 052311 (2001).
[7] C. Silberhorn, T. C. Ralph, N. Lütkenhaus, and G. Leuchs, Phys. Rev. Lett.
89, 167901 (2002).
[8] F. Grosshans and P. Grangier, Phys. Rev. Lett. 88, 057902 (2002).
[9] F. Grosshans et al., Nature 421, 238 (2003).
[10] L. K. Grover, Phys. Rev. Lett. 79, 325 (1997).
[11] P. W. Shor, in Proceedings of the 35th Annual Symposium on the Founda-
tions of Computer Science (IEEE Computer Society Press, Los Alamitos,
California, 1994), p. 124.
[12] D. P. DiVincenzo, Phys. Rev. A 51, 1015 (1995).
[13] J. I. Cirac and P. Zoller, Phys. Rev. Lett. 74, 4091 (1995).
[14] C. Monroe et al., Phys. Rev. Lett. 75, 4714 (1995).
[15] F. Schmidt-Kaler et al., Nature 422, 408 (2003).
[16] L. M. K. Vandersypen et al., Nature 414, 883 (2001).
[17] C. H. Bennett and D. P. DiVincenzo, Nature 404, 247 (2000).
150 BIBLIOGRAPHY
[18] S. Lloyd and S. L. Braunstein, Phys. Rev. Lett. 82, 1784 (1999).
[19] S. L. Braunstein, Phys. Rev. Lett. 80, 4084 (1998).
[20] S. Lloyd and J. J. E. Slotine, Phys. Rev. Lett. 80, 4088 (1998).
[21] T. C. Ralph, W. J. Munro, and G. J. Milburn, Quantum Computation

with Coherent States, Linear Interactions and Superposed Resources, quant-
ph/0110115.
[22] S. Massar and E. S. Polzik, Phys. Rev. Lett. 91, 060401 (2003).
[23] M. Brune et al., Phys. Rev. Lett. 77, 4887 (1996).
[24] C. Monroe, D. M. Meekhof, B. E. King, and D. J. Wineland, Science 272,

1131 (1996).
[25] A. Furusawa et al., Science 282, 706 (1998).
[26] D. Bouwmeester et al., Nature 390, 575 (1997).
[27] D. Boschi et al., Phys. Rev. Lett. 80, 1121 (1998).
[28] N. Lütkenhaus, J. Calsamiglia, and K. A. Suominen, Phys. Rev. A 59, 3295

(1999).
[29] Y. H. Kim, S. P. Kulik, and Y. Shih, Phys. Rev. Lett. 86, 1370 (2001).
[30] M. Fleischhauer and M. D. Lukin, Phys. Rev. Lett. 84, 5094 (2000).
[31] D. F. Phillips et al., Phys. Rev. Lett. 86, 783 (2001).
[32] C. Liu, Z. Dutton, C. H. Behroozi, and L. V. Hau, Nature 409, 490 (2001).
[33] A. V. Turukhin et al., Phys. Rev. Lett. 88, 023602 (2002).
[34] Y. H. Shih and C. O. Alley, Phys. Rev. Lett. 61, 2921 (1988).
[35] Z. Y. Ou, S. F. Pereira, H. J. Kimble, and K. C. Peng, Phys. Rev. Lett. 68,
3663 (1992).
[36] P. G. Kwiat et al., Phys. Rev. Lett. 75, 4337 (1995).
[37] C. Schori, J. L. Sørensen, and E. S. Polzik, Phys. Rev. A 66, 033802 (2002).
[38] J. S. Bell, Physics 1, 195 (1965).
[39] A. Aspect, J. Dalibard, and G. Roger, Phys. Rev. Lett. 49, 1804 (1982).
[40] E. Hagley et al., Phys. Rev. Lett. 79, 1 (1997).
[41] C. A. Sackett et al., Nature 404, 256 (2000).

BIBLIOGRAPHY 151
[42] D. J. Wineland et al., Phys. Rev. A 46, R6797 (1992).
[43] G. Santarelli et al., Phys. Rev. Lett. 82, 4619 (1999).
[44] J. Hald, J. L. Sørensen, C. Schori, and E. S. Polzik, Phys. Rev. Lett. 83,
1319 (1999).
[45] A. Kuzmich, L. Mandel, and N. P. Bigelow, Phys. Rev. Lett. 85, 1594 (2000).
[46] H. J. Briegel et al., Lect. Notes Comput. Sc. 1509, 373 (1999).
[47] S. Haroche, M. Brune, and J. M. Raimond, Phil. Trans. Roy. Soc. A 355,
2367 (1997).
[48] G. C. Bjorklund, M. D. Levenson, W. Lenth, and C. Ortiz, Appl. Phys. B

32, 145 (1983).
[49] O. Schmidt, K. M. Knaak, R. Wynands, and D. Meschede, Appl. Phys. B

59, 167 (1994).
[50] M. A. Bouchiat and J. Brossel, Phys. Rev. 147, 41 (1966).
[51] E. B. Alexandrov et al., Laser Phys. 6, 244 (1996).
[52] E. B. Alexandrov et al., Phys. Rev. A 66, 042903 (2002).
[53] P. W. Milonni and J. H. Eberly, Lasers (John Wiley & Sons, New York,
1988).
[54] D. Kleppner, N. F. Ramsey, and H. M. Goldenberg, Phys. Rev. 126, 603

(1962).
[55] H. J. Metcalf and P. van der Straten, Laser Cooling and Trapping (Springer,
New York, 1999).
[56] W. Happer, Rev. Mod. Phys. 44, 169 (1972).
[57] D. Tupa and L. W. Anderson, Phys. Rev. A 36, 2142 (1987).
[58] A. Kuzmich, N. P. Bigelow, and L. Mandel, Europhys. Lett. 42, 481 (1998).
[59] Y. Takahashi et al., Phys. Rev. A 60, 4974 (1999).
[60] E. S. Polzik, J. L. Sørensen, and J. Hald, Appl. Phys. B 66, 759 (1998).
[61] A. Einstein, B. Podolsky, and N. Rosen, Phys. Rev. 47, 777 (1935).
[62] L. Duan, J. I. Cirac, P. Zoller, and E. S. Polzik, Phys. Rev. Lett. 85, 5643
(2000).
[63] L. Duan, G. Giedke, J. I. Cirac, and P. Zoller, Phys. Rev. Lett. 84, 2722
(2000).
152 BIBLIOGRAPHY
[64] R. Simon, Phys. Rev. Lett. 84, 2726 (2000).
[65] M. G. Raymer, A. C. Funk, B. C. Sanders, and H. de Guise, Phys. Rev. A

67, 052104 (2003).
[66] F. Hong-yi and J. R. Klauder, Phys. Rev. A 49, 704 (1994).
[67] I. Bouchoule and K. Mølmer, Phys. Rev. A 66, 043811 (2002).
[68] A. DiLisi and K. Mølmer, Phys. Rev. A 66, 052303 (2002).
[69] C. Henkel and M. Wilkens, Europhys. Lett. 47, 414 (1999).
[70] C. H. Bennett et al., Phys. Rev. Lett. 70, 1895 (1993).
[71] L. Vaidman, Phys. Rev. A 49, 1473 (1994).
[72] S. L. Braunstein, C. A. Fuchs, H. J. Kimble, and P. van Loock, Phys. Rev.

A 64, 022321 (2001).
[73] F. Grosshans and P. Grangier, phys. Rev. A 64, 010301 (2001).
[74] S. L. Braunstein and H. J. Kimble, Phys. Rev. A 80, 869 (1998).
[75] A. Kuzmich and E. S. Polzik, Phys. Rev. Lett. 85, 5639 (2000).
[76] J. Fiurasek, Private communication.
[77] C. Cohen-Tannoudji, J. Dupont-Roc, and G. Grynberg, Introduction to

Quantum Electrodynamics (John Wiley & Sons, Inc., New York, 1989).
[78] K. J. Blow, R. Loudon, S. J. D. Phoenix, and T. J. Shepherd, Phys. Rev. A

42, 4102 (1990).
[79] M. Fleishhauer and T. Ricter, Phys. Rev. A 51, 2430 (1995).
[80] C. Cohen-Tannoudji, J. Dupont-Roc, and G. Grynberg, Atom-Photon In-

teractions (John Wiley & Sons, Inc., New York, 1992).
[81] A. R. Edmonds, Angular momentum in quantum mechanics (Princeton Uni-

versity Press, Princeton, 1974).
[82] C. W. Gardiner and P. Zoller, Quantum Noise, 2nd ed. (Springer, Heidel-
berg, Germany, 2000).
[83] Z. Y. Ou, C. K. Hong, and L. Mandel, J. Opt. Soc. Am. B 4, 1574 (1987).
[84] B. Huttner and Y. Ben-Aryeh, Phys. Rev. A 40, 2479 (1989).
[85] S. F. Pereira, X. Min, H. J. Kimble, and J. L. Hall, Phys. Rev. A 38, 4931
(1988).
BIBLIOGRAPHY 153
[86] P. Meystre and M. Sargent III, Elements of Quantum Optics (Springer-

Verlag, Berlin, 1991).
[87] N. F. Ramsey, Molecular Beams (Oxford University Press, Amen House,

London EC4, 1956).
Erratum
In the following we list errors in the thesis Entanglement and Quantum In-
teractions with Macroscopic Gas Samples by Brian Julsgaard, University of
Aarhus, October 2003. This erratum has been updated August 28th, 2005.
1 Interaction Hamiltonian
There is a mistake of a factor of two in the derived effective Hamiltonian
Eq. (5.18). This error arises from a mistake in the adiabatic elimination
procedure of Sec. 5.2. I am grateful to Klemens Hammerer for finding the
error. Below we will explain it in detail.
1.1 A Spin-1/2 Toy Model

It is unnecessary going through the calculations of Sec. 5.2 with the full level
structure and propagating optical fields. We shall consider a single spin-1/2
atom as depicted in Fig. 1 interacting with only two light modes, â+ and â− .
The interaction Hamiltonian for this system is (in the frame rotating with
the laser field)
Ĥint = h̄∆(σ̂33 + σ̂44 ) + h̄g(ã†+ σ̃14 + σ̃41 ã+ + ã†− σ̃23 + σ̃32 ã− ), (1)
where g is a real parameter describing the interaction strength and the atomic
operators σ̃ij and light operators ã± have already been written in the slowly
|3i |4i
∆
ωat
ωL â+ â−
|1i |2i
Figure 1: The level structure of a simple spin-1/2 model with optical frequen-
cies and detunings included. The propagation direction of the light mode
coincides with the quantization axis of the atomic states and we decompose
the light into the two polarization modes â+ and â− .
1
varying form (see Eq. (5.9)). From this Hamiltonian we may derive the
equations of motion. For the ground state operators we find
∂ σ̂11 1 h i
= σ̂11 , Ĥint = −ig(ã†+ σ̃14 − σ̃41 ã+ ),
∂t ih̄
∂ σ̂22
= −ig(ã†− σ̃23 − σ̃32 ã− ), (2)
∂t
∂ σ̂12
= −ig(ã†− σ̃13 − σ̃42 ã+ ).
∂t
To proceed we need approximate results for the atomic operators on the right
hand side of these equations. For σ̃14 we find
∂ σ̃14 g
= −i∆σ̃14 − igã+ (σ̂11 − σ̂44 ) ⇒ σ̃14 ≈ − ã+ σ̂11 , (3)
∂t ∆
where the adiabatic elimination is carried out by setting the time derivative
equal to zero. We also neglect the excited state population σ̂44 . For σ̃13 we
have
∂ σ̃13 g
= −i∆σ̃13 − ig(ã− σ̃12 − ã+ σ̃43 ) ⇒ σ̃13 ≈ − ã− σ̃12 , (4)
∂t ∆
where the excited state coherence σ̃43 was neglected. In similar manners we
find
g g
σ̃23 ≈ − ã− σ̂22 and σ̃24 ≈ − ã+ σ̂21 . (5)
∆ ∆
Now we insert the approximations (3)-(5) into the equations of motion (2)
and obtain
∂ σ̂11 ∂ σ̂22
= = 0,
∂t ∂t (6)
∂ σ̂12 ig 2 †
= (ã− ã− − ã†+ ã+ )σ̃12 .
∂t ∆
These equations are reproduced by the effective ground state Hamiltonian
h̄g 2 ³ ´
Ĥeff = − σ̂11 ã†+ ã+ + σ̂22 ã†− ã− . (7)
∆
If we instead (as was done in Sec. 5.2) insert the approximations (3)-(5)
directly into the un-approximated Hamiltonian (1) we obtain an effective
Hamiltonian twice as big which is wrong since the correct Eq. (6) would then
not be reproduced.
2
1.2 Implications on the Experimental Results
The erroneous factor of two discussed above is of course inherited to many
equations in the thesis chapters following the derivation of the interaction
Hamiltonian. In Eq. (6.15) the correct interaction parameter is
γ λ2
a=− a1 , (6.15)
8A∆ 2π
and whenever we encounter an expression proportional to a (or a1 , a2 ) we
must reduce the result by a factor of two. It is not convenient to list all the
cases here but one important equation is the estimation of the atomic to shot
noise ratio κ2 in the entanglement generation experiment. We find
9.3 · P [mW] · T [ms] · a1 (∆) · θDC [deg]
κ2 = . (9.51)
∆blue [MHz]
On page 106 (and in different papers published after the Ph.D. thesis, the
error was found in the summer of 2005) we actually discuss a discrepancy
of a factor of approximately two in the observed atomic to shot noise ratio.
With the error found there is now a much better agreement between theory
and experiment.
2 Simple Errors
There is a mistake in Eq. (5.19). For F = 3 the sign of a1 has to be reversed
such that we have
µ ¶
1 45 21
a1 = − − 80 → −1 (8)
56 1 + ∆24 /∆ 1 + ∆23 /∆
Updates
Below we list publications which are a result of work performed after the
Ph.D. thesis. Hence, this should not be considered as a part of the thesis but
as a supplement. I hope it will be useful to the reader.
• D. V. Kupriyanov, O. S. Mishina, I. M. Sokolov, B. Julsgaard, and
E. S. Polzik, Phys. Rev. A 71, 032348 (2005). This publication treats
the higher order effects as discussed in Chap. 6. It also coveres the noise
properties of the higher order terms. Available at quant-ph/0411083.
3
• B. Julsgaard, J. Sherson, J. I. Cirac, J. Fiurasek, and E. S. Polzik,
Quantum Memory for Light, Nature 432, 482 (2004). Also available at
quant-ph/0410072. This publication describes an experiment carried
out with a setup quite similar to the entanglement experiments in the
thesis.
• J. Sherson, B. Julsgaard, and E. S. Polzik, Distant Entanglement of

Macroscopic Gas Samples in Decoherence, Entanglement and Infor-
mation Protection in Complex Quantum Systems, eds. W. M. Akulin,
A. Sarfati, G. Kurizki, and S. Pellegrin (Springer, Dordrecht, 2005).
Also available at quant-ph/0408146. This publication describes very
well the new entanglement procedure as discussed in Sec. 10.4.

Brian Julsgaard

Uploaded by

Copyright:

Available Formats

Brian Julsgaard

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Brian Julsgaard

Uploaded by

Copyright:

Available Formats

Entanglement and Quantum

Danish National Research Foundation

This thesis is submitted to the Faculty of Science at the University of

List of Publications vii

1 Outline of the Thesis 1

2 Atoms and Light 3

5 The Effective Interaction Hamiltonian 33

7 Atomic State Characterization 49

8 Recording Quantum Fluctuations 67

9 Entanglement, Theoretical Approach 79

10 Experimental Generation of Entangled States 95

11 Quantum Information Protocols 109

12 Summary and Outlook 117

A Polarization States of the Light Field 119

B Spins and Density Operators 123

C Continuous Description of Light and Matter 127

D Dipole Matrix Elements 133

E Photo Detection Theory 139

F The Quadratic Zeeman Effect 145

G Spin Decay and Langevin Forces 147

[I] B. Julsgaard, A. Kozhekin, and E. S. Polzik, Experimental long-lived en-

[II] C. Schori, B. Julsgaard, J. L. Sørensen, and E. S. Polzik, Recording Quan-

[III] J. L. Sørensen, B. Julsgaard, C. Schori, J. Hald, and E. S. Polzik, Quantum

[IV] E. S. Polzik, B. Julsgaard, J. Sherson, and J. L. Sørensen, Entanglement and

[V] J. L. Sørensen, B. Julsgaard, C. Schori, and E. S. Polzik, Quantum limits

[VI] E. S. Polzik, B. Julsgaard, C. Schori, and J. L. Sørensen, in The Expand-

[VII] B. Julsgaard, C. Schori, J. L. Sørensen, and E. S. Polzik, Atomic Spins as a

October 2003, Brian Julsgaard.

Outline of the Thesis

Atoms and Light as Interesting

2.1 The Atomic Ground State Spin

(a) (c) (d)

(a) x z (b) (c)

x-pol y-pol Ŝy

Macroscopic Stokes vector

2.2 Polarization States of Light

2.3 Interaction Between Atomic Spins and

(a) (c) (d)

Probe 852nm Atomic Atomic

Ŝyout (t) = Ŝyin (t) + aSx Jˆz (t), (6.11)

Ŝzout (t) = Ŝzin (t). (6.12)

3.1 Why Quantum State Engineering

complex number ψ(x), a vast amount of complex numbers. We could also

3.2 Technological Implementations

• Quantum cryptography is an area of physics dealing with secret com-

• Quantum computation is a big motivation for exploiting and controlling

• Quantum communication is the transport of (unknown) quantum states

• Quantum memory is the process of storing and retrieving a quantum

• Entanglement generation is a key ingredient for quantum teleportation

• Precision measurements is also a motivation for the study of quantum

3.3 Continuous Versus Discrete Systems

matter is (in the dipole approximation) basically governed by the Hamiltonian

• Yet another possibility is our approach.

purposes of quantum computation, quantum cryptography, and error correction.

In this chapter we discuss general aspects of our experiments. This includes

4.1 Laser Systems

the F = 4, mF = 4 state is a dark state if this laser has σ+ -polarization.

(a) (b) l/2 l/2 To experiment

EOM Cesium l/4

4.2 Atomic Vapour Cells