Institut für Angewandte Physik

der Universität Bonn

Wegelerstraße 8
53115 Bonn

Raman spectroscopy
of single atoms

von
Igor Dotsenko

Diplomarbeit in Physik

angefertigt im
Institut für Angewandte Physik

vorgelegt der
Mathematisch-Naturwissenschaftlichen Fakultät
der Rheinischen Friedrich-Wilhelms-Universität
Bonn
im Oktober 2002

Referent: Prof. Dr. D. Meschede

Korreferent: Prof. Dr. E. Klempt
To my daughter, Alyona, who came into my life one
day after I had entered the lab...
Contents

Introduction 1

1 Theory 3
1.1 Stimulated Raman transitions . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Vibrational sidebands . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.1 Origin of vibrational sidebands . . . . . . . . . . . . . . . . . 8
1.2.2 Raman transitions between vibrational levels . . . . . . . . . . 10
1.3 Interference of two Raman transitions . . . . . . . . . . . . . . . . . . 14

2 Experimental setup 19
2.1 Magneto-optical trap . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Dipole trap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Magnetic field compensation . . . . . . . . . . . . . . . . . . . . . . . 27
2.4 Generation of the Raman laser beams . . . . . . . . . . . . . . . . . . 28

3 Raman spectroscopy 35
3.1 Experimental methods . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.1.1 State preparation . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.1.2 State-selective detection . . . . . . . . . . . . . . . . . . . . . 36
3.1.3 Experimental timing sequence . . . . . . . . . . . . . . . . . . 38
3.2 Spectroscopy of Zeeman sublevels . . . . . . . . . . . . . . . . . . . . 44
3.2.1 The linear Zeeman effect . . . . . . . . . . . . . . . . . . . . . 44
3.2.2 Analysis of the Raman spectra . . . . . . . . . . . . . . . . . . 45
3.3 Spectroscopy of vibrational levels . . . . . . . . . . . . . . . . . . . . 51
3.3.1 Experimental parameters . . . . . . . . . . . . . . . . . . . . . 52
3.3.2 Spectrum of the vibrational sidebands . . . . . . . . . . . . . 54

4 Rabi oscillations 57
4.1 Measurements of Rabi oscillations . . . . . . . . . . . . . . . . . . . . 57
4.2 Interference effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

Summary and outlook 63

Appendix 65

A Interaction of a three-level atom with three lasers 65

Bibliography 72

I
Introduction

The coherent control of quantum systems has recieved much attention and experi-
mental effort in recent years. Its applications range from the possibility of studying
fundamental aspects of quantum physics to quantum information processing. A
wide variety of physical systems and phenomena have been proposed as quantum
systems. Some of them are quantum dots, superconductive circuits, nuclear
magnetic resonance, trapped ions, atoms in optical lattices and cavity quantum
electrodynamics. The main advantages of neutral atoms as controllable quantum
systems are their weak interaction with their environment and a rich spectrum
of tools for the manipulation of all degrees of freedom. These have already been
proven to enable cooling to extremely low temperatures [Win79], the population
of a single internal and external state (e.g. Bose-Einstein condensation [And95]),
coherent manipulation in atom interferometry [Ber97], atomic clocks [Kit01], and
experiments with single atoms [Kuh01].

Quantum engineering of microscopic atomic systems requires control of all degrees

of freedom of isolated atoms: the number of particles, external degrees of freedom
and internal states [Chu02]. In our experiment an exact number of cold atoms
is provided by a single atom MOT. To isolate atoms we transfer them into an
optical dipole trap formed by a standing wave interference pattern. The position
and velocity of the atoms are controlled by moving the interference pattern
[Kuh01, Sch01]. This allows us to transport single atoms over macroscopic
distances with submicrometer precision. The next task is to manipulate the internal
degrees of freedom, i.e. the hyperfine states. Coherent population transfer among
these states can successfully be realized by stimulated Raman transitions [Ber98].

In the current work I will present the theoretical description of the coherent
properties of stimulated Raman transitions and describe the technique we use
for Raman spectroscopy of single atoms. We record Raman spectra of Zeeman
sublevels to develop methods of optical pumping into specific mF sublevels and to
precisely measure magnetic fields. Raman spectroscopy of vibrational sidebands
allows us to measure the oscillation frequency of atoms in the dipole trap and to

1
2 INTRODUCTION

drive transitions on different vibrational sidebands. Later this spectroscopy will be

used to measure the temperature of trapped atoms. To demonstrate the coherent
nature of the Raman transitions I will present Raman Rabi oscillations and report
about the theoretical prediction and experimental evidence of the presence of
interference of Raman transitions.

The long coherence time measured in the dipole trap allows us to use the super-
position of internal states of neutral atoms as qubits for quantum information pro-
cessing [Mon02]. A quantum gate operation [Ben00] can be implemented by si-
multaneous coupling of two atoms to the same mode of a high-finesse optical cavity
[Pel95, Bei00]. In the future we plan to deterministically position one or more
atoms in such a cavity. To perform cavity QED experiments a large and constant
atom-cavity coupling is preferable, which can be realized with a tightly confined
atom. To improve the localization of the trapped atom and to have complete con-
trol over its coupling to a cavity field, the atom has to be cooled down to its lowest
vibrational level. This can be achieved by using Raman sideband cooling techniques
[Mon95, Vul98, Kuk89]. The demonstrated capability to resolve the vibrational
sidebands as well as to coherently manipulate atomic states by means of Raman tran-
sitions will allow us to perform Raman sideband cooling of single atoms trapped in
a dipole trap.
Chapter 1

Theory

To carry out experiments on Raman spectroscopy and interpret the experimental

results we need a theoretical description of Raman transitions. In Sec. 1.1 I will
discuss the basics of two-photon coupling of atomic states and show coherent
properties of Raman transitions. The theory of Raman transitions will be extended
to the case of three laser beams in Sec. 1.3 to describe our experimental system.

The narrow linewidth of Raman transitions can be used to resolve atomic motional
sidebands. Sec. 1.2 is devoted to a discussion of vibrational sidebands of a tightly
bound atom in a harmonic potential well and the possibility to couple different
vibrational states by Raman transitions.

1.1 Stimulated Raman transitions

A Raman transition couples two atomic levels by the absorption of a photon from
one Raman beam (pump beam) and by stimulated emission of another one into the
other beam (Stokes beam). Fig. 1.1 shows a three-level atom and the laser field
consisting of two lasers in Λ-configuration. The pump laser couples levels |1i and
|2i, while the Stokes laser couples |2i and |3i. As a result, the states |1i and |3i
become coherently coupled by the Raman beams.

To avoid resonant excitation of the excited state, the detuning ∆ of the Raman
beams from the one-photon transition has to be much larger than the linewidth Γ
of the excited level. Both Raman beams are characterized by the Rabi frequency
Ωi , which is a frequency measure of the electric dipole interaction between an atom
and a laser field. If the optical frequencies of the Raman beams are ωP and ωS ,
respectively, the Raman detuning is defined as the detuning δ from the two-photon
resonance.

3
4 CHAPTER 1. THEORY

Figure 1.1: Three-level atom coupled by two lasers in Λ-configuration.

Basics of Raman transitions

To describe the coherent evolution of the Λ-system we perform a quantum-
mechanical treatment, i.e. we construct the Hamiltonian and solve the corresponding
Schrödinger equation. The time-dependent Schrödinger equation governing the time
evolution of the system reads
d
i~ Ψ(t) = Ĥ(t) · Ψ(t), (1.1)
dt
where Ĥ(t) = Ĥ 0 + V̂ (t) is the full Hamiltonian consisting of the unperturbed Hamil-
tonian Ĥ 0 , which defines the energy levels of an isolated atom, and the operator V̂ (t)
of the time-dependent interaction. The interaction of an electromagnetic field E(t) ~
with an atom is governed by the electric-dipole interaction:
V̂dip = −d̂ · E , (1.2)
where d̂ is the induced atomic dipole moment. The coupling strength of the inter-
action is characterized by the Rabi frequency [Met99]
r
d̂ · E I
Ω=− =Γ , (1.3)
~ 2I0
where Γ is the linewidth of the excited level, I0 is the saturation intensity and I
is the laser intensity. For the Raman pump and Stokes lasers we define the Rabi
frequencies as ΩP and ΩS , respectively.

The state vector Ψ(t) of the three-level system can be expressed as a superposition
of the eigenstates ψn of Ĥ 0 which are the three levels |1i, |2i and |3i:
Ψ(t) = C1 (t) · ψ1 + C2 (t) · ψ2 + C3 (t) · ψ3 . (1.4)
1.1. STIMULATED RAMAN TRANSITIONS 5

The coefficient Cn (t) is the probability amplitude which, squared, yields the popu-
lation of a state |ni. Using the rotating-wave approximation the Hamiltonian of the
system can be written [Sho90]
 
0 ΩP 0
~
Ĥ =  ΩP 2∆ ΩS  . (1.5)
2
0 ΩS 2δ

Eq. (1.4) and (1.5) result in a set of linear differential equations for the probability
amplitudes Ci :
1

 iĊ1 = Ω
2 P
C2
1
iĊ = Ω C1 + ∆ C2 + 12 ΩS C3 (1.6)
 2 2 P
1
iĊ3 = Ω
2 S
C2 + δ C 3

Since the population of the excited level |2i undergoes fast oscillations, we assume
that over many oscillations Ċ2 averages to zero. This approximation is known as
”adiabatic elimination” and reduces our system to an effective two-level system:

iĊ1 = Ω4∆P (ΩP C1 + ΩS C3 )

½
(1.7)
iĊ3 = −δC3 + Ω 4∆
S
(ΩP C1 + ΩS C3 ) ,

where the new Hamiltonian reads

à !
Ω2P ΩP ΩS
1 ∆ ∆
Ĥef f = ΩP ΩS Ω2S (1.8)
4 − 4δ
∆ ∆

The off-diagonal elements of the Hamiltonian are the coupling of the states |1i and
|3i by the Raman beams. The diagonal elements represent the light shift of the
corresponding levels due to the electric dipole interaction with the far red-detuned
lasers. The differential light shift for the resonant Raman transition is given by

Ω2S Ω2
δdif f = − P. (1.9)
4∆ 4∆
Solving (1.8) and considering that initially an atom is in state |1i we find the time
evolution of the system, i.e. the time dependence of the state populations

|C1 |2 = 1 − Λ · sin2 ( Ω20 t)

½
(1.10)
|C3 |2 = Λ · sin2 ( Ω20 t) ,
p
where Λ = Ω2R /Ω20 is the amplitude of the population oscillation, Ω0 = Ω2R + δ 2
is the generalized Rabi frequency and ΩR is the resonance Rabi frequency at which
6 CHAPTER 1. THEORY

the atomic population oscillates between these states when the Raman transition is
resonant
ΩP ΩS
ΩR = . (1.11)
2∆

The width of the transition is given by the range of the Raman detuning where the
oscillation amplitude is larger than 1/2, and is thus given by

∆ωpower = 2 ΩR . (1.12)

Since the Rabi frequency depends on the laser intensity, the dependence of the
transition linewidth on the laser power is known as power broadening of the
transition.

Another effect which broadens the width of the Raman transition is the Fourier limit
which is the dependence of the linewidth on the Raman pulse duration τ

2π
∆ωF ourier ≈ . (1.13)
τ

Scattering rate

Despite large detunings of the Raman lasers from the atomic one-photon resonance,
they can still off-resonantly excite the atomic transitions. A laser beam of intensity
I and detuning ∆ from the one-photon transition produces a scattering rate of

Γ I/I0
Γsc = , (1.14)
2 1 + I/I0 + ( 2∆
Γ
)2

where Γ is the linewidth and I0 is the saturation intensity. For large detunings, the
scattering rate becomes proportional to the laser intensity:
µ ¶2
ΓI Γ
Γsc = . (1.15)
2 I0 2∆

The scattering rate scales as ∆−2 , whereas the Rabi frequency is proportional to
∆−1 . Thus introducing a large detuning we can find a compromise between the
desirable reduction of the scattering rate and the unwanted reduction of the Rabi
frequency.
1.1. STIMULATED RAMAN TRANSITIONS 7

Raman transitions in a cesium atom

In our experiments we manipulate cesium atoms using the D2 -transition at 852 nm

between the 6S1/2 and 6P3/2 states. The hyperfine interaction splits the ground
state into two hyperfine levels F = 3 and F = 4. The hyperfine splitting is ∆ HFS =
2π · 9.19263177 GHz. To coherently manipulate the cesium ground-state hyperfine
levels we use the Raman transition as shown in Fig. 1.2. The typical detuning of
the Raman beams from the D2 transition is much larger than the linewidth of the
excited state ∆ = 2π · 13..300 GHz À Γ = 2π · 5.22 MHz.

Figure 1.2: Two-photon Raman transition connecting the two Cs hyperfine

ground states.

Since the cesium hyperfine ground levels have a long lifetime, the width of the
Raman transition is only given by power broadening and the Fourier limit. This
allows us to use Raman transitions for high-resolution spectroscopy.

The hyperfine states are degenerate with respect to spin orientations. The depen-
dence of the Rabi frequency on the initial mF sublevels and the polarizations of the
Raman beams is defined by

p ΩP ΩS p
ΩR, 0 = ΩR X(mF ) = X(mF ) , (1.16)
2∆
8 CHAPTER 1. THEORY

where the coefficient X(mF ) is given by [Mue01]

 1

 288
(4 + mF )(5 + mF ) : (π, σ − ), (σ + , π)

1
(4 − mF )(5 − mF ) : (π, σ + ), (σ − , π)


 288
X(mF ) = 1
h ¡ mF ¢ 2 i , (1.17)

 9
1 − 4
: (σ + , σ + ), (σ − , σ − )


0 : (π, π), (σ ± , σ ∓ )



where symbols in parentheses denote the polarizations of the two Raman lasers.

1.2 Vibrational sidebands

In this section I will introduce the concept of vibrational sidebands of a bound
atom and explain their origin and properties. To resolve them, a transition with a
narrow linewidth is required. A stimulated Raman transition induced by counter-
propagating Raman beams can serve for this purpose.

1.2.1 Origin of vibrational sidebands

Consider an atom with a single resonance in the absorption spectrum at ω0 . If the
atom is tightly bound in a harmonic potential where it oscillates at the frequency
Ωz , vibrational sidebands appear in the absorption spectrum. They are placed on
both sides of the carrier frequency ω0 separated by Ωz . The origin of these sidebands
can be explained both classically and quantum-mechanically.

Classical explanation
If an atom oscillates in the z direction with the oscillation frequency Ωz and the
amplitude A the equation of motion reads

z(t) = A sin(Ωz t) . (1.18)

Suppose, a laser with the frequency ωL and the electric field

EL (z, t) = E0 cos(ωL t − kL z)

is shined in along the z direction. The effective field E(t) at the position of the atom
is then phase-modulated with the modulation frequency Ωz :

E(t) = EL (z(t), t) = E0 cos(ωL t − kL A sin(Ωz t)) . (1.19)

1.2. VIBRATIONAL SIDEBANDS 9

Using the Bessel functions J l (m) with the modulation index m = kL A, equation
(1.19) can be decomposed into a series of cosine functions:
∞
X
E(t) = E0 J l (m) cos((ωL + lΩz )t) . (1.20)
l=−∞

This results in an absorption spectrum of the atom shown in Fig. 1.3. It consists of
the carrier frequency at ω0 and sidebands separated by Ωz . The height of the peaks
is given by the squared Bessel functions of the corresponding order.

Figure 1.3: Absorption spectrum of an atom oscillating with frequency Ωz in

an external potential. The resonance frequency of the atom at rest
is ω0 . A natural linewidth Γ = 0.1Ωz has been assumed.

Quantum-mechanical explanation
To explain the origin of the motional sidebands quantum-mechanically, we consider
a two-level atom with an energy separation E = ~ω0 of the ground |gi and the
excited |ei state. For a free atom only photons with frequency ω0 can induce a
transition between these states. In an external harmonic potential
1
U (z) = m Ω2z z 2 , (1.21)
2
the atomic motion is quantized leading to discrete vibrational states |ni with energies
µ ¶
1
En = n + ~Ωz . (1.22)
2
Consequently each electronic state |gi and |ei is dressed with the vibrational states
|n = 0, 1, 2...i and the transition |g, ni → |e, n + li produces a line in the absorption
spectrum at frequency ωsb = ω0 + l Ωz as shown in Fig. 1.4.
10 CHAPTER 1. THEORY

Figure 1.4: Two-level atom in a harmonic potential. Transitions to different vi-

brational levels result in the appearance of sidebands in the absorp-
tion spectrum. The sidebands can be resolved only if the transition
linewidth is much smaller than the oscillation frequency, Γ ¿ Ωz .

1.2.2 Raman transitions between vibrational levels

To resolve the vibrational sidebands, the linewidth of the transition Γ has to be
much smaller than the spacing between the vibrational levels, Ωz (see Fig. 1.4).
The cesium D2 -transition has a linewidth of Γ = 2π · 5.22 MHz À Ωz ≈ 2π · 500 kHz.
Thus we cannot resolve vibrational sidebands via a resonant excitation of the 6P 3/2
state.

Since the Cs hyperfine ground states F = 3 and F = 4 have a long lifetime, the width
of the Raman transition is limited only by power broadening and the Fourier limit.
Thus this transition can successfully be used to resolve the motional sidebands. The
vibrational levels coupled by the Raman beams are shown schematically in Fig. 1.5.

Lamb-Dicke regime
The Lamb-Dicke parameter controls the coupling between the light field and an
oscillating atom and is defined as

η = k L ez z 0 (1.23)
1.2. VIBRATIONAL SIDEBANDS 11

Figure 1.5: Raman transition coupling different vibrational levels.

whereq~kL is the momentum transfer, ez is a unit vector in the z direction, and

~
z0 = 2mΩ z
is the spread of the wave function of the oscillatory ground state.

The limit where η ¿ 1 is known as the Lamb-Dicke regime and is character-

ized by a strong localization of the atom compared to the optical wavelength.
For a two-photon process the momentum transfer between an atom and the laser
field is ~∆kL = ~kP − ~kS and the Lamb-Dicke parameter is given by η = ∆kL ez z0 .

For the counter-propagating Raman beams which couple the hyperfine states of a
trapped cesium atom oscillating at the frequency Ωz ≈ 2π · 500 kHz, the Lamb-Dicke
parameter is
r
ωrec
η = 2 ≈ 0.13 ¿ 1 , (1.24)
Ωz

where ωrec ≈ 2π · 2 kHz is the single photon recoil frequency for cesium given by

~kL2
ωrec = . (1.25)
2m

Coupling strength
The coupling strength between the vibrational states |ni and |mi is given by the
matrix element [Win79]

A n→m = hn| ei∆kL ẑ |mi . (1.26)

12 CHAPTER 1. THEORY

The position operator ẑ can be expressed in terms of the raising and lowering oper-
ators

ẑ = z0 ez (â + ↠) . (1.27)

With the Lamb-Dicke parameter η defined above the following relation holds

∆kL ẑ = η(â + ↠) . (1.28)

Using (1.28) the coupling strength can be calculated in terms of the generalized
Laguerre polynomial Lsk [Sch94]
s
2 /2 n< ! |n−m| |n−m| 2
An→m = e−η η Ln< (η ) , (1.29)
n> !

where n< and n> denote the smaller and larger values of n and m, respectively. In
the Lamb-Dicke regime, the following approximation is valid
s
η |n−m| n> !
An→m ≈ for n 6= m, (1.30)
|n − m|! n< !
2 2n + 1 2
An→n = e−η /2 Ln (η 2 ) ≈ 1 − η . (1.31)
2

Rabi frequency
The Rabi frequency of the population oscillation between the states |ni and |mi
depends on the coupling strength

ΩR, n→m = An→m · ΩR, 0 , (1.32)

where ΩR, 0 is the carrier Rabi frequency given by (1.16). Thus, in the Lamb-Dicke
regime the Rabi frequencies of the transitions on the first red and blue sidebands
are
√
ΩR, n→n−1 = η n · ΩR, 0 ,
√
ΩR, n→n+1 = η n + 1 · ΩR, 0 , respectively. (1.33)

This result can be used to transfer the total atomic population into the first red
sideband by a π-pulse. Then we can optically pump atoms into the initial hyperfine
state. By repeating these steps successively we can cool an atom into the lowest
vibrational state [Mon95].
1.2. VIBRATIONAL SIDEBANDS 13

Absorption rate
The absorption rate γ is proportional to the squared coupling strength. Thus the
absorption rates for transitions on the first-order sidebands are

γn→n−1 = η 2 n · γ0 ,
γn→n = (1 − η 2 (2n + 1)) · γ0 ,
γn→n+1 = η 2 (n + 1) · γ0 , (1.34)

where γ0 is the absorption rate of an atom at rest illuminated by the laser field of
the same intensity. One sees that in the Lamb-Dicke regime the sum of these three
absorption rates is unity and the absorption rates of transitions on higher-order
sidebands are negligible.

Eq. (1.34) shows that the absorption rate depends on the vibrational level and is
different for the red and blue sidebands. Besides, for the higher-order sidebands it
is reduced drastically

γn→n+k ∝ η 2|k| . (1.35)

If the atom is in the lowest vibrational state, the red sideband vanishes. The tem-
perature can be measured by comparing the heights the red and blue sidebands
in the spectrum. Eq. 1.34 shows how to experimentally find the mean vibrational
quantum number hni:
γn→n−1 n
=
γn→n+1 n+1
Ired
⇒ hni = , (1.36)
Iblue − Ired
where I is the intensity of the corresponding sideband in the recorded spectrum.
This method is valid as long as the transitions are not saturated.

Orientation of the Raman beams

A change of the vibrational state requires momentum transfer. In the configuration
of two copropagating Raman beams, the momenta of absorbed and emitted photons
are almost the same resulting in almost no net momentum transfer ∆kL ≈ 0.
Therefore the different vibrational states are not coupled by the lasers, so that
the atom cannot change its motional state and only transitions on the carrier are
possible. To excite different vibrational states, the two Raman beams must be
14 CHAPTER 1. THEORY

separated and shined at the atom from different directions.

Besides, if ∆kL points in the z direction, the Raman coupling is sensitive to motion
in the z direction only. The vibrational levels which correspond to the oscillation
in the radial direction are not excited.

The off-resonant scattering rate due to the Raman beams is independent of the
vibrational level occupied by the atom and can be treated in the same way as in
Sec. 1.1.

1.3 Interference of two Raman transitions

In the current experiment (see Sec. 2.4) the Raman beams are generated by a single
diode laser with the frequency ω2 . We use an EOM to phase-modulate the laser
beam and to produce two sidebands ω1 and ω3 with a frequency spacing close to
the hyperfine splitting of the 6S1/2 level (∆EOM ≈ ∆HFS = 9.192 GHz) (see Fig. 1.6)
with opposite phases.

Figure 1.6: Spectrum of our Raman beams

To better understand the interaction of all three Raman beams with an atom and to
study the influence of different experimental parameters on the coherent population
transfer a theory of this interaction has to be developed. For this purpose we
have first to find a theoretical model describing the current physical process and
1.3. INTERFERENCE OF TWO RAMAN TRANSITIONS 15

to perform the full quantum-mechanical treatment accompanied with reasonable

approximations.

Theoretical model

Our model system consists of three Raman beams interacting with a three-level
atom via the electric dipole interaction as shown in Fig. 1.7. States |1i and |3i
denote the cesium hyperfine levels (6S1/2 F = 3) and (6S1/2 F = 4), respectively,
and the state |2i is the 6P3/2 level. For simplicity the hyperfine states of the 6P3/2
level are considered to be degenerate since the Raman beams are far red-detuned
from the D2 -transition of Cs. Typical detunings ∆ are on the order of 13 . . . 300 GHz
and are much higher than the hyperfine splitting of the 6P3/2 level, which is about
0.6 GHz. The Raman detuning is given by δ.

Figure 1.7: Coupling of three Raman lasers with a three-level atom

In this system two Raman transitions can be resonant at the same time. One
transition involves ω1 and ω2 and the other ω2 and ω3 as pump and Stokes beams,
respectively. Since Raman coupling is a coherent process, these transitions interfere.
16 CHAPTER 1. THEORY

Interference of the Raman transitions and how to avoid it

The full quantum-mechanical treatment of the system is presented in Appendix A.
The main result is the presence of interference between the two possible Raman
transitions in the system, whose Rabi frequencies are
Ω1 Ω2 Ω2 Ω1
ΩR1 = and ΩR2 = respectively,
2∆ 2(∆ + ∆HFS )

where Ωi is the Rabi frequencies of the ith laser given by (1.3). Since the powers of
the phase-modulation sidebands ω1 and ω1 are equal, we suppose Ω1 = Ω3 .

The Rabi frequency ΩR at which the atom oscillates between the two ground states
strongly depends on the phase difference ∆ϕ
1
q
ΩR = Ω2R1 + Ω2R2 + 2ΩR1 ΩR2 · cos ∆ϕ . (1.37)
2
In the experiment the EOM phase-modulates the beam which results in the pro-
duction of sidebands with opposite phases. Thus ∆ϕ = π and the two Raman
transitions interfere destructively

Ωdestr
R = ΩR1 − ΩR2
Ω1 Ω2 ∆HFS
= · (1.38)
2∆ ∆ + ∆HFS
The destructive interference reduces the Rabi frequency significantly and for large
detunings ∆ the Rabi frequency scales as
1
ΩR ∼ . (1.39)
∆2
As we will see later in Sec. 4.2 the main disadvantage of the interference is not
the reduction of the Rabi frequency, but its instability. Since the vibrational
sidebands can only be excited with a non-zero angle between the Raman beams
(see Sec. 1.2.2), spatial separation of carrier and sidebands is required. The Raman
beams propagating along different optical paths get an arbitrary phase difference
∆ϕ which is very sensitive to all acoustic vibrations of optical elements leading to
big uncertainties in ΩR .

To avoid this uncontrollable interference, one of the two Raman transitions has to
be suppressed. For this purpose the frequency of the carrier is shifted with respect
to the sidebands. Additionally, the EOM is detuned from ∆HFS to make only one
of the two Raman transition resonant (see Fig. 1.8). To shift the carrier we can use
1.3. INTERFERENCE OF TWO RAMAN TRANSITIONS 17

Figure 1.8: Shifted carrier with respect to the sidebands. The EOM frequency
must be detuned to ∆EOM = ∆HFS + ∆AOM .

two AOMs installed for the carrier beam and the sidebands, whose frequencies differ
by
carrier
∆AOM = (∆AOM − ∆sidebands
AOM ).

The second transition becomes shifted by ∆off−res = 2∆AOM out of the two-photon
resonance. The modified coupling scheme with only one possible Raman transition
is presented in Fig. 1.9.

Since usual AOM frequencies are much smaller than ∆, the detunings of the Raman
beams from the one-photon resonances are considered to be unaltered so that the
Rabi frequency becomes
Ω1 Ω2
ΩR = . (1.40)
2∆
For the same reason the differential light shift is not changed by the introduction
of the AOM.

The presence of the interference between Raman transitions proves the coherent
properties of the Raman coupling and was confirmed experimentally (see Sec. 4.2).
18 CHAPTER 1. THEORY

Figure 1.9: Level scheme with one of the Raman transitions suppressed.
Chapter 2

Experimental setup

All experiments reported here were performed with single cesium atoms. To
coherently manipulate the internal atomic states, the atoms are stored in a dipole
trap which is a conservative potential allowing for long coherence times. Since the
dipole trap depth is only 3 mK the atoms have to be cooled below this temperature
before they are bound to the trap. A magneto-optical trap serves as a source of a
well-defined small number of cold cesium atoms. The traps are described in Sec. 2.1
and 2.2.

Raman transitions between the two hyperfine sublevels of the Cs ground state are
used to coherently manipulate these states. As shown in Sec. 1.1, the frequency
difference between the two Raman beams has to match the hyperfine splitting of
the Cs ground state with a stability down to the Hz level. The Raman beam setup
fulfilling these conditions is presented in Sec. 2.4.

2.1 Magneto-optical trap

A magneto-optical trap (MOT) serves as a source of cold cesium atoms. It traps

atoms from the background gas in a vacuum cell and cools them down to about
125 µK. The MOT consists of a laser field produced by three orthogonal pairs
of counter-propagating laser beams with orthogonal circular polarizations and a
quadrupole magnetic field produced by magnetic coils. The radiation force simulta-
neously pushes an atom to the center of the MOT and cools it at this position. The
magnetic field modifies the radiation pressure and produces a position dependent
force pointing towards the trap center [Bal00].

19
20 CHAPTER 2. EXPERIMENTAL SETUP

Principle of magneto-optical trapping

To cool atoms, a velocity-dependent force is necessary with a direction opposite

to the atomic momentum and which vanishes for stationary atoms. Two counter-
propagating laser beams which are red-detuned from the atomic resonance produce
such a velocity-dependent force in one dimension. An atom moving towards one
beam sees its frequency closer to the resonance due to the Doppler effect, i.e. blue
shifted, while the frequency of the other beam becomes further red-detuned. By
absorption of a photon preferentially from the blue shifted beam the atom gets
a momentum kick opposite to its velocity. Since the spontaneous emission is
isotropic, it does not change the average momentum of the atom and the total
momentum transfer is opposite to the direction of the atomic motion. This results
in an effective friction force cooling the atom along the direction of the laser beams.
This process is called Doppler cooling.

To cool atoms in all three dimensions, three orthogonal pairs of counter-propagating

laser beams are used. This configuration is called optical molasses because of the
isotopic effect in friction force. The Doppler temperature TD , the lowest temperature
which can be reached by this method, is determine by the equilibrium of the Doppler
cooling and the heating by the spontaneous emission process

Γ
k B TD ≡ ~ ≈ kB · 125 µK , (2.1)
2

where kB is the Boltzmann constant.

Trapping of atoms at a specific position requires a position-dependent force. For

this purpose a quadrupole magnetic field is added which vanishes at the center of
the optical molasses and increases linearly in all directions.

The principle of operation is explained in Fig. 2.1 with the example of an atom
having an optical transition between the levels F = 0 and F = 1. The degeneracy
of the upper level with respect to mF is lifted in the presence of the magnetic field
by the Zeeman effect and the splitting is proportional to the distance of the atom to
the magnetic zero point. If the red-detuned counter-propagating laser beams have
opposite circular polarizations, a stationary atom displaced from the zero point to
the left absorbs a photon preferentially from the laser beam which pushes it back to
the center of a MOT. The same effect occurs if an atom is displaced into the other
direction.
2.1. MAGNETO-OPTICAL TRAP 21

(a) (b)

Figure 2.1: Principle of operation of a magneto-optical trap. (a) Simplified

atomic level scheme in a one-dimension gradient field interacting
with two red-detuned laser beams. (b) A three-dimensional MOT.

Figure 2.2: Level scheme of the hyperfine structure of the cesium D2 -transition.

Magneto-optical trap for cesium atoms

In our experiment we manipulate cesium atoms using the D2 -transition at

λ = 852 nm (see Fig. 2.2). For the Doppler cooling a closed transition is needed
to make the cooling process continuous. Therefore the transition F = 4 → F 0 = 5
is used. The cooling laser is red-detuned by several line-widths from the cooling
transition.
22 CHAPTER 2. EXPERIMENTAL SETUP

Despite being far blue-detuned from the transition F = 4 → F 0 = 4, the cooling

laser can still excite this transition with the probability of about 1/1000. After
an atom is excited to F 0 = 4 it can spontaneously decay into the level F = 3
and would be lost for the cooling process. Thus we need a repump laser to pump
atoms back to the F = 4 level. For this purpose we use a laser resonant with the
transition F = 3 → F 0 = 4.

A schematic view of the MOT is presented in Fig. 2.3 (a). Magnetic field coils in
anti-Helmholz configuration produce the quadrupole magnetic field. The magnetic
field gradient is 350 G/cm at a current of 16 A. This localizes the atoms to about
30 µm.

Figure 2.3: (a) The scheme of the vacuum apparatus, the MOT setup and the
imaging optics. The third pair of the MOT beams is perpendicular
to the image plane. (b) The fluorescence seen by the APD depends
on the number of atoms trapped in the MOT. The steps up and
down in the fluorescence correspond respectively to the capture and
loss of an atom.

The MOT operates inside a glass cell connected to a vacuum chamber with an ultra-
high vacuum of < 10−10 mbar. The density of the esium background gas in the cell
is controlled by means of a valve connecting a Cs reservoir with the vacuum chamber.

By monitoring the fluorescence at the MOT position, we count the number of atoms
using the single atom fluorescence level as shown in Fig. 2.3 (b). The fluorescence
2.2. DIPOLE TRAP 23

light from the atoms is collected by a diffraction-limited objective [Alt02] with a

numerical aperture N A = 0.29, spatially filtered by a pinhole and projected onto
an avalanche photodiode (APD) with a quantum efficiency of 50 %.

2.2 Dipole trap

After atoms are trapped and cooled in the MOT, they are transferred into an
optical dipole trap. In contrast to the dissipative MOT, a dipole trap produces a
conservative trapping force resulting in long coherence times [Gri00].

Optical dipole traps rely on the electric dipole interaction of an atom with far-
detuned light
1
Udip (r) = − hd · Ei , (2.2)
2
where E is the electric field and d is the induced atomic dipole moment. For the
laser frequency ω and the atomic resonance frequency ω0 we define the effective laser
detuning ∆0
1 1 1
0
= + . (2.3)
∆ ω − ω0 ω + ω0
On condition that ω0 À |∆0 | À Γ we get simple expressions for the dipole potential
and the scattering rate [Gri00]
~Γ I(r) Γ
Udip (r) = , (2.4)
8 I0 ∆ 0
µ ¶2
Γ I(r) Γ
Γsc (r) = . (2.5)
8 I0 ∆0

Here, I(r) = 2²0 c|E(r)|2 denotes the laser intensity, I0 is the saturation intensity
and Γ is the linewidth of the atomic transition.

The equation (2.4) presents the ac Stark effect which is the energy shift of an atomic
state produced by the interaction with a far-detuned laser. In the case of a red-
detuned laser (∆0 < 0) the interaction energy is negative. This leads to the creation
of a potential well where an atom can be trapped as shown in Fig. 2.4.

Optical dipole trap

Our dipole trap consists of two counter-propagating Gaussian laser beams with equal
intensities, frequencies and parallel linear polarizations [Sch01, Kuh01]. The laser
24 CHAPTER 2. EXPERIMENTAL SETUP

Figure 2.4: AC Stark effect. A red-detuned laser coupled to a two level atom
shifts the atomic states. A beam with a Gaussian profile produces
a potential well where the atom can be trapped.

beams interfere and the standing wave interference pattern produces a position-
dependent light shift of the Cs ground state (see Fig. 2.5). This forms a periodic
trapping potential with a period of a half a wavelength

w20 h 2ρ2 i
U (z, ρ) = U0 · 2
· exp − 2
· cos2 (kz) , (2.6)
w(z) w(z)

where w2 (z) = w20 (1 + z 2 /zR2 ) is the beam radius with waist w0 and Rayleigh length
zR = πw20 /λ. The depth of the trap is given by

~Γ P Γ
U0 = . (2.7)
2 πw20 I0 ∆

Here, Γ = 2π · 5.2 MHz is the natural linewidth of the cesium D2 -line,

I0 = 1.1 mW/cm2 is the saturation intensity, P is the total laser power. For
alkalis the effective laser detuning ∆ is given by [Gri00] ∆−1 = (∆−1 −1
1 + 2∆2 )/3,
where ∆i is the detuning from the Di transition.

The light source for the dipole trap is a Nd:YAG laser with a wavelength of
λ = 1064 nm. This results in an effective detuning of |∆| ≈ 107 Γ. Both beams
are focused into the MOT down to about w0 = 20 µm with a Rayleigh length of
zR = 1.2 mm. The typical total power is P = 4 W. For these parameters the trap
depth equals |U0 | = 3 mK and the photon scattering rate is 32 photons/s.

To transfer atoms between the MOT and the dipole trap without loss we superimpose
both traps for several milliseconds.
2.2. DIPOLE TRAP 25

Figure 2.5: An atom trapped in an optical dipole trap formed by the standing
wave interference pattern.

Oscillation frequencies in a dipole trap

Since the trap depth is one order of magnitude larger than the Doppler tempera-
ture, atoms are placed deep in the potential well and the trapping potential can be
approximated by the harmonic potential
1
Uharm (z) = mΩ2z z 2 . (2.8)
2
where m is the mass of the atom. For a cesium atom (mCs = 2.2 · 10−25 kg) stored
in the dipole trap the axial oscillation frequency equals
s
2|U0 |
Ωz = 2π ≈ 2π · 570 kHz. (2.9)
mCs λ2
The oscillating atom has a discrete energy spectrum of the equidistant vibrational
levels separated by ~Ωz . The mean value of the vibrational quantum number n̄ can
be found from
µ ¶
1
En = ~Ωz n̄ + = k B TD
2
k B TD 1
⇒ n̄ = − ≈ 4. (2.10)
~Ωz 2
26 CHAPTER 2. EXPERIMENTAL SETUP

In the radial direction the dipole trap size is given by the beam diameter and the
radial oscillation frequency reads
s
4|U0 |
Ωrad = ≈ 2π · 7 kHz . (2.11)
mCs w20

Differential light shift of the hyperfine ground states

The Cs ground state consists of two hyperfine levels F = 3 and F = 4 separated by
a frequency of ∆HFS = 2π · 9.19 GHz. The dipole trap laser produces an ac Stark
shift of both hyperfine levels. Since the light shift depends on the detuning from
the atomic resonance, the shifts of these two levels are different. This results in the
appearance of a differential light shift (see Fig. 2.6)

~ δdif f ,YAG = Udip, F =3 − Udip, F =4 . (2.12)

Figure 2.6: Scheme of the differential light shift of the Cs ground levels produced
by Nd:YAG laser (greatly exaggerated).

Using (2.7) we get

Γ P Γ
δdif f ,YAG = 2
, (2.13)
2 πw0 I0 ∆00

µ ¶
1 1 ∆HFS 2∆HFS
where = − + . (2.14)
∆00 3 ∆1 (∆1 + ∆HFS ) ∆2 (∆2 + ∆HFS )
2.3. MAGNETIC FIELD COMPENSATION 27

For our experimental parameters the differential light shift due to the Nd:YAG laser

δdif f ,YAG / 2π = −9.2 kHz . (2.15)

Later in Sec. 3.2.2 this value will be compared with the differential light shift
induced by the Raman beams.

2.3 Magnetic field compensation

In order to perform optical pumping, the quantization axis of the system has to be
well-defined. For this purpose a magnetic field has to be set in a given direction
with a given strength. This requires both the compensation of stray magnetic
fields, e.g. of the Earth and magnetized objects near the MOT and the use of a
guiding magnetic field.

To compensate DC magnetic fields as well as to apply the guiding magnetic field,

three pairs of orthogonal magnetic coils are placed around the MOT. By controlling
the three coil currents we can create a magnetic field of arbitrary direction and
amplitude up to 1 G.

The influence of static magnetic fields on the optical pumping was used to coarsely
compensate them. To achieve this we illuminate the Cs background gas in the
vacuum cell with a σ + -polarized probe beam, which is resonant with the transition
F = 3 → F 0 = 2 (see Fig. 2.7). In the absence of a magnetic field the direction of
the probe beam defines the quantization axis and the σ + -polarized light transfers
the population into the mF = 3 and mF = 2 sublevels of the F = 3 state, which
are dark states. This leads to a reduction of the fluorescence. A magnetic field
component perpendicular to this quantization axis mixes these mF states with
other bright states and increases the fluorescence light.

The atoms used here are not trapped and only fluoresce while flying through
the probe laser beam. Since the interaction time is on the order of 100 ns, the
off-resonant excitation of atoms to the F 0 = 3 level with further decay into the
F = 4 level is insignificant. Therefore a repumping beam was not necessary.

The experimental procedure consists of optimizing the current of the compensation

coils to minimize the fluorescence rate on the APDs. Shining in the probe beam
along the z axis of the MOT we can compensate the x and y components of the
28 CHAPTER 2. EXPERIMENTAL SETUP

F’ = 5
4
6P3/2 3
2

+
s

F=4

6S1/2
F=3
mF=1 2 3

dark states

Figure 2.7: Optical pumping into dark mF states. A presence of a magnetic field
component perpendicular to the quantization axis mixes the mF
states become mixed with other mF states and the total fluorescence
increases.

residual magnetic field, shining it along the y axis we compensate the z component.

The static magnetic field near the glass cell before compensation was measured to be
about 45 µT. After the compensation the magnitude of the field was estimated to be
less then 2 µT. The precision of the described method was limited by the resolution
of the current sources and the fluctuations of fluorescence. Fine compensation can
be performed by means of Raman spectroscopy using the Zeeman splitting of the
mF sublevels [Rin01], see Sec. 3.2.

2.4 Generation of the Raman laser beams

To drive the two-photon transition between the cesium hyperfine ground states
F = 3 and F = 4 and to perform the Raman spectroscopy, two Raman beams are
needed fulfilling the following requirements:

• the frequency difference of the Raman beams has to match the hyperfine split-
ting of the Cs ground state (9.2 GHz);

• to record the Raman spectrum the frequency difference has to be scanned over
2.4. GENERATION OF THE RAMAN LASER BEAMS 29

a range of a few MHz with less than a Hz resolution and a similar absolute
frequency accuracy has to be provided;
• it should be possible to set the optical detuning to any value between 10 and
300 GHz to eliminate spontaneous scattering due to the Raman beams (see
Sec. 1.1);
• the powers of the Raman laser beams have to be reasonably stable to make
the Rabi frequency sufficiently constant to allow coherent manipulation of the
atom;
• the Raman beams have to be spatially separated to resolve the motional side-
bands (see Sec. 1.2).
Two laser beams with a frequency separation of 9.2 GHz can be produced by means
of different methods, e.g.:
• two phase-locked diode lasers with a frequency difference of 9.2 GHz [Sch96];
• a single diode laser with direct current modulation at 4.6 GHz [Rin99] or
9.2 GHz [Lau84, Aff00], generating frequency-modulation sidebands at a dis-
tance of the modulation frequency, and using carrier and sideband at 9.2 GHz,
or −1st and +1st sidebands at 4.6 GHz;
• a single laser and an electro-optic modulator (EOM) with a modulation fre-
quency of 4.6 GHz or 9.2 GHz, generating phase-modulation sidebands.
In our experiment we use the third possibility to produce optical sidebands at
9.2 GHz from the carrier frequency. Carrier and sidebands are then spatially
separated by an interferometer. For fast switching and an additional control of
frequencies and powers of the Raman beams, acousto-optical modulators (AOMs)
are installed in both beams.

A scheme of the Raman laser setup is presented in Fig. 2.8. In the following part of
the chapter the main elements of the setup are briefly described. Detailed informa-
tion about the components used can be found in [Mue01].

Light source
The laser beam at 852 nm is generated by a single-mode diode laser in Littrow-
configuration. The output power is about 30 mW. The choice of the output fre-
quency, i.e. the detuning from the D2 -transition of Cs, is realized by turning the
laser grating. The output frequency is monitored by a wavemeter with a precision of
0.3 GHz. Usual detunings used in our experiments are in the range of 13..300 GHz.
The frequency drift is less than 0.5 GHz per hour without further frequency stabi-
lization, which means that the detuning is stable up to 1 %.
30 CHAPTER 2. EXPERIMENTAL SETUP

Generation of 9.2 GHz sidebands

The EOM phase-modulates of the laser frequency ωC and generates sidebands at
the modulation frequency ∆EOM . So, if the electric field of the input laser beam has
the form
Ein = E0 cos(ωC t), (2.16)
then the phase-modulation at the frequency ∆EOM with the modulation index m
results in an output field given by

Eout = E0 cos(ωC t + m sin(∆EOM t))

∞
X
= E0 J l (m) cos(ωC + l · ∆EOM )t) , (2.17)
l=−∞

where the amplitude of the l-th sideband is given by the Bessel function J l (m) as a
function of the modulation index m. The modulation index of the EOM used here
equals [EOM01] · ¸
rad p
m = 0.05 2Peff R , (2.18)
V
where R = 50 Ω is the input impedance of the EOM and Peff is the input RF-power.

The EOM is driven by a synthesized sweeper Agilent 83751A .01-20 GHz and a
RF-power amplifier. The synthesizer is externally locked to an atomic clock 10 MHz
reference to provide an absolute accuracy of better than 1 Hz at 9.2 GHz.

To maximize the power of the sidebands, the EOM has to be driven with high input
RF-power. The highest possible value of the input power is limited by heating of
the EOM crystal which can even destroy it. Drifts of the EOM temperature result
in slight deflection of the Raman beams and, hence, change their alignment to the
subsequent elements (an interferometer, AOMs, optical fibers). To avoid the heating
and to make the EOM less sensitive to fluctuations of the ambient temperatures,
a water cooling plate was mounted to the side of the EOM. It helps to keep the
temperature low and constant and allows us to drive the EOM with up to 4 W of
input power. The spectrum of the output light is monitored by means of a monitor
cavity with a free spectral range of 1.5 GHz (see Fig. 2.9).

Separation of carrier and sidebands

There are various techniques to separate two beams of different frequencies. If the
wavelengths differ by at least several nm, dichroic mirrors, diffraction gratings or
prisms can be used to separate them. The usual methods to separate two laser
beams with a frequency difference on the order of several GHz are interferometrical
2.4. GENERATION OF THE RAMAN LASER BEAMS 31

methods.

In the present setup we use the so-called HDW interferometer [Hau00], a type of
Mach-Zehnder interferometer (see Fig. 2.10). By moving a retro-reflecting prism the
difference ∆S between the two optical paths S1 and S2 can be set arbitrarily. Two
beams with a frequency difference of ∆HFS are separated if
c
∆S = (N + 1/2) . (2.19)
∆HFS

To observe the quality of the beam separation, we use a monitor cavity. The best
separation achieved is about 90 %.

Active electronic stabilization of the interferometer is employed. Since the power

of the carrier is higher than the total power of the sidebands, the error signal can
be derived from the ”carrier” output of the interferometer by choosing the prism
position such that the maximum power comes out that way. For this purpose the
prism is glued to a piezo actuator (PZT) and the voltage on the PZT is modulated
by 10 kHz. The signal of a photodiode placed at the ”carrier” output of the inter-
ferometer is sent to a lock-in amplifier. The resulting error signal is fed back to the
PZT.

Acousto-optical modulators
To provide additional control of frequencies and powers of the Raman beams,
acousto-optical modulators in double-pass configuration are installed in each sepa-
rated beam. They shift the optical frequencies by 2 × (100..120) MHz depending on
the experimental needs. Besides, by changing the amplitude modulation voltage we
are able to set the desired time dependence of the Raman pulse power. Addition-
ally, AOMs act as fast optical switches with a rise time of 2 µs. For comparison, the
reaction time of our mechanical shutters is on the order of a millisecond.

Guiding the Raman beams to the dipole trap

The Raman laser setup and the dipole trap apparatus are located on different optical
tables. The Raman beams are sent through single-mode polarization-maintaining
optical fibers from one table to the other. The optical fibers act as spatial filters
making positions and shapes of the Raman beams insensitive to alignment of the
Raman laser setup.

The Raman beams are focused to a waist of about 90 µm. The spread of atoms
in the dipole trap is defined by the MOT size and is about 30 µm. To make the
32 CHAPTER 2. EXPERIMENTAL SETUP

powers of the Raman beams uniform through the spatial distribution of atoms in
the dipole trap, and thus the Rabi frequency of the Raman transition constant, the
Raman beams have to be focused exactly to the center of the MOT.

The typical procedure to align lasers onto the MOT is to observe a change in the
fluorescence. Being far-detuned, the Raman beams do not affect the atoms in the
MOT and cannot be used for the alignment. Thus we have to use an additional
alignment laser beam, which influences trapped atoms and is coupled together with
the Raman beams into the same fibers. The fiber ensures perfect overlap of both
beams. The alignment beam can be chosen to be resonant with the Cs transition
F = 4 → F 0 = 4. This beam depopulates the F = 4 level by pumping atoms to the
F = 3 level, and reduces the fluorescence in the MOT cooling transiiton. To make
the fluorescence reduction appreciable, the MOT repumper has to be attenuated.
The Raman beam alignment is considered to be optimal if the maximum possible
reduction of the fluorescence is obtained.
2.4. GENERATION OF THE RAMAN LASER BEAMS 33

Figure 2.8: Scheme of generation of the Raman beams.

34 CHAPTER 2. EXPERIMENTAL SETUP

Figure 2.9: Carrier frequency and sidebands generated by the EOM as seen
behind a scanning monitor cavity. The fraction of power in the
carrier and the 1st sidebands are about 51 % and 21 % respectively.

Figure 2.10: Separation of carrier and sidebands using the HDW interferometer.
Chapter 3

Raman spectroscopy

Since the linewidth of a Raman transition which couples two long-lived states is
determined only by the duration and the powers of the lasers, Raman transitions
can be used to perform high-resolution spectroscopy. In our experiment the Raman
beams couple the two hyperfine levels of the Cs ground state and allow us to resolve
features of the Cs level structure such as Zeeman sublevels.

In the present chapter I will describe the experimental methods to perform Raman
spectroscopy of single atoms and present Raman spectra recorded with two different
configurations of the Raman beams: co- and counter-propagating.

The co-propagating Raman beams allow to perform Doppler-free spectroscopy

since the simultaneous absorption and emission of photons in co-propagating
laser beams compensate for the Doppler shift [Rin01]. Since co-propagating
Raman spectroscopy is still sensitive to the magnetic fields and light shifts, it can
successfully be used to calibrate the apparatus.

In contrast, if the Raman beams are counter-propagating, the momentum exchange

between the Raman beams and the atom is maximized allowing to resolve motional
sidebands of the trapped atoms and to perform Raman sideband cooling [Mon95,
Vul98].

3.1 Experimental methods

The dependence of the transition probability on the Raman detuning yeild Raman
spectrum. To determine the transition probability we have to find a method both
to prepare the initial and to detect the final hyperfine state of an atom.

35
36 CHAPTER 3. RAMAN SPECTROSCOPY

3.1.1 State preparation

Optical pumping is used to prepare an atom in a specific hyperfine state. For this
purpose, the pumping transition is usually chosen such that the desired state is a
dark state. For instance, to pump an atom into the F = 4 state, one can use a
laser beam which is resonant with the transition F = 3 → F 0 = 4 (see Fig. 3.1).
After several scattering processes an atom will decay into the F = 4 state. In
the experiment we use the MOT lasers for state preparation. If we switch off the
repumping laser before the cooling laser during the transfer of atoms into the dipole
trap, we prepare the atoms in F = 3, and vice versa.

(a) Pumping to F = 3. (b) Pumping to F = 4.

Figure 3.1: Preparation of the initial atomic state. The MOT cooling laser
is used for optical pumping to the F = 3 level via off-resonance
excitation of the F 0 = 4 state. The MOT repumping laser pumps
atoms to the F = 4 level.

3.1.2 State-selective detection

After the atoms have been prepared in F = 4 we couple the hyperfine states by
the Raman beams and then detect the population in F = 3. For this purpose
we use an intense ”push-out” laser resonant with the nearly-closed transition
F = 4 → F 0 = 5, which pushes the atoms out of the dipole trap if they are in
the F = 4 level. The atoms in the F = 3 level, however, are unaffected by the
push-out laser and remain trapped. The number of remaining atoms is measured
by observing their fluorescence in the MOT.
3.1. EXPERIMENTAL METHODS 37

For both the ”pushing-out” process and the optical pumping into F = 3 (see
Fig. 3.1) we use the same transition although these two task have opposite
requirements: for the pushing-out all atoms in F = 4 should be heated out of the
trap before any pumping to F = 3 occurs. For the pumping into F = 3 the pimping
process would be completed before the atoms are heated so much that they leave
the trap.

To achieve this we use two different phenomena depending on the intensity of the
laser beam. For relatively low laser intensities we observe recoil heating of the
atoms. After some time they decay into the other hyperfine state without leaving
the trap. This regime is used for optical pumping into F = 3. But increasing the
laser power we can get into another regime where atoms are pushed out of the trap
using significantly less photons and are therefore not transferred to the other state.

The difference between these two cases is that the recoil heating increases the
atomic energy by the recoil energy Er = ~2mk on each absorption or emission of a
2 2

photon, while the pushing-out adds up all photon momenta pr = ~k to the total
momentum of the atom.

Assuming that to remove an atom from the dipole trap of depth U0 the recoil heating
and the pushing-out processes need nheat and npush photons, respectively, then
U0 = 2nheat · Er
(npush pr )2
U0 = = n2push · Er , (3.1)
2m
and the following relation holds
√
npush = 2nheat . (3.2)
With U0 ≈ 3 mK (see Sec. 2.2), nheat ≈ 15000 and npush ≈ 175. Thus to push the
atom out of the dipole trap we need much less scattering events and the probability
for the off-resonant scattering into the state F = 3 is much reduced.

To stay in the pushing-out regime, i.e. in order to add up all the photon momenta,
the atom has to preserve the direction of motion before it escapes the trap. For
this purpose the push-out laser has to be intense enough to remove an atom out of
the trap within one oscillation period: τpush < Ω1z . In contrast, heating occurs when
the process takes long compared to the oscillation period before it boils an atom
out of the trap: τheat > Ω1z , or if counter-propagating laser beams are used and the
average momentum transfer is zero. This is the case for our optical pumping to
F = 3 since we use the counter-propagating MOT laser beams.
38 CHAPTER 3. RAMAN SPECTROSCOPY

In the experiment the intense push-out laser is shined in perpendicular to the

dipole trap. In this direction the oscillation frequency is the lowest. To reduce
the number of photons necessary for the pushing-out and to prevent accidental
pumping to F = 3, we lower the dipole trap depth adiabatically to about 3 − 7 %
of the maximum depth prior to the state selective detection.

The experimental realization of state preparation and state selective detection is

presented in Fig. 3.2. Initially we trap and cool cesium atoms in the MOT. By
overlapping the MOT and the dipole trap we transfer all atoms from one trap to
the other. It was found that setting the MOT to full power during the loading
of the dipole trap cools the atoms deeper into the dipole trap and we lose less of
them when we lower the trap depth. During the transfer we switch off the MOT
lasers one after another to prepare atoms in one of the two hyperfine states. The
push-out laser is shined in for 1 ms and then all remaining atoms are transferred
back into the MOT. Comparing the initial fluorescence level with the final one and
taking into account the background due to the stray light of the MOT beams we
count the initial and final number of atoms and thus obtain information about the
efficiency of the state preparation and detection.

Atoms that have been prepared in the F = 4 state are lost after the state selective
pushing-out, less than 1 % of them survive. And vice versa, atoms prepared in
F = 3 survive with a probability higher than 95 %.

We studied the efficiency of this detection method for different polarizations of

the push-out laser. The σ + -polarized push-out beam excites the closed transition
(F = 4, mF = 4) → (F 0 = 5, mF = 5) and reduces the probability of the
off-resonant scattering to the F = 3 state. Nevertheless, the detection efficiency is
still high for the nearly-closed transition excited by the π-polarized push-out laser
when we additionally lower the dipole trap down to 1.5 %.

3.1.3 Experimental timing sequence

The experiment is controlled by a computer, which has 32 digital and 8 analog
output channels. The digital channels control laser shutters, send TTL pulses to
electronic devices such as frequency generators, serve as trigger signals, etc. The
analog channels apply modulation voltages to AOMs, EOMs, set currents in the
compensation magnetic coils, etc. The time resolution of the computer control is
2 µs.

To run an experiment, we create an appropriate timing sequence of the output

3.1. EXPERIMENTAL METHODS 39

(a) Preparation in F = 4. No atoms survive after the pushing-out.

(b) Preparation in F = 3. About 95 % of atoms are still trapped.

Figure 3.2: Experimental sequence for the state selective detection. The plot
shows the time dependence of the fluorescence in the MOT region,
accumulated over 10 repetitions. The upper bars show the switching
of the corresponding lasers. The push-out laser removes atoms in
F = 4 from the trap (a), but leaves atoms in F = 3 unaffected (b).
40 CHAPTER 3. RAMAN SPECTROSCOPY

channels. A typical experimental timing sequence for experiments on Raman

transitions is presented in Fig. 3.3. Horizontal bars correspond to digital channels
which control shutters of different lasers used in the experiment and switch magnetic
fields on and off.

Figure 3.3: Timing sequence of different lasers and magnetic fields for experi-
ments on the Raman transitions. The times can vary depending on
the experiment.

Below I will explain in detail the purpose of each element in the sequence.

• The MOT cooling laser forms the optical molasses, whereas the MOT repump-
ing laser pumps atoms back to the cooling transition. A TTL signal from the
computer closes or opens mechanical shutters of these lasers. Closing one
of them later than the other one results in optical pumping of atoms into a
specific ground hyperfine state (see Fig. 3.1).

• The full MOT power is used either to cool atoms deeper into the dipole trap
while loading it or to quickly load new atoms into the MOT. The variable
attenuation of the MOT cooling laser is provided by an AOM.

• The MOT magnetic field can be set to two different levels: low and high.
The low one produces a low gradient of the magnetic field and, thus, enlarges
the capture cross-section. It is used together with the full MOT cooling laser
3.1. EXPERIMENTAL METHODS 41

power to collect new atoms from the Cs background vapour into the MOT
quickly. In contrast, the high field compresses the MOT and is used for the
normal operation of the MOT.
• The mechanical shutter of the Nd:YAG laser is used to switch the dipole trap
on and off.
• The guiding magnetic field defines the quantization axis in the system and
splits the mF states.
• For some experiments optical pumping into a specific mF sublevel is necessary.
For this purpose we use the optical pumping laser together with the repumping
laser. The frequencies and polarizations of these lasers depend on the target
state of the optical pumping.
• The Raman lasers induce the Raman transition with a definite detuning and
pulse duration. The detuning is set by the 9.2 GHz frequency synthesizer
which drives the EOM in the Raman laser setup. The duration is controlled
by AOMs serving as fast optical switches.
• If we want to scan the Raman detuning during the experiment, we have to
program the 9.2 GHz frequency synthesizer. For this purpose we enter the
initial and final output frequencies, and the number of frequency steps to be
used. From this data the synthesizer automatically calculates the step size and
changes the output frequency by this amount whenever it gets a TTL pulse
from the computer.
• The push-out laser performs the state selective detection as described in
Sec. 3.1.2.
• While the timer card is enabled the fluorescence detected by the avalanche
photodiode is recorded by a computer .
• The dipole trap depth is proportional to the Nd:YAG laser power which is
controlled by means of the AOMs placed in the Nd:YAG beam path. To
make the state selective detection more efficient by reducing the amount of
scattered photons before an atom is pushed out of the trap, we lower the
dipole trap depth before applying the push-out laser. We lower the trap depth
adiabatically in two steps: first to 7.5 % of its maximum depth where the
optical pumping and Raman transition are performed and then down to 1.5 %
where the push-out laser is applied. After the state selective detection is
performed we increase the dipole trap depth back to the maximum value. The
time dependence of the dipole trap depth is shown schematically in the last
row in Fig. 3.3.
42 CHAPTER 3. RAMAN SPECTROSCOPY

The experiment can be divided into three parts: the first step is to prepare a known
number of cold atoms in the F = 4 state; then the hyperfine states are coupled
with the Raman beams and the state selective detection is performed; the final task
is to count the remaining atoms.

To carry out the experiment, we first have to prepare the known number of cold
atoms. For this purpose we trap and cool atoms in the MOT. Then we switch on
the timer card and a computer starts to record the fluorescence rate. Since we can
measure the fluorescence rate of a single atom, we can easily deduce the number of
initially trapped atoms. To load the dipole trap we overlap it with the MOT for
about 20 ms and switch the MOT cooling laser to its full power. This cools atoms
deeper into the dipole trap. We do not use the full power of the cooling laser all
the time in order not to saturate the APD. Closing the cooling laser shutter 10 ms
before the MOT repumping laser shutter is closed, we prepare the atoms in the
F = 4 state. The next step is to switch off the MOT magnetic field and to apply
the guiding magnetic field. It takes about 50 ms for the MOT magnetic field to
decay. Then we adiabatically lower the dipole trap depth down to 7.5 % during
10 ms.

After the atoms are trapped in the dipole trap, the main part of the experiment
is carried out. If necessary, the optical pumping into a specific mF levels can be
applied as will be described later in Sec. 3.3.1. The power and duration of the
pumping laser have to be chosen not to heat the atoms significantly during the
pumping process. For instance, for some experiments these parameters were 20 nW
and 3 ms respectively. The repumping laser has to have much higher power (up to
1 µW) and last longer than the pumping laser to ensure that atoms do not stay in
the wrong hyperfine state. Next we apply the Raman lasers, lower the dipole trap
further and perform the state selective pushing-out.

The next task is to transfer the remaining atoms back into the MOT. For this
purpose we increase the dipole trap depth back to the maximum value, and switch
on the MOT. After the Nd:YAG beam is blocked, we observe the fluorescence level
caused by the remaining atoms. After the MOT magnetic field is switched off,
the atoms leave the trap and we measure the stray light level of the MOT lasers.
Together with the initial fluorescence level this information is enough to determine
the ratio of atoms transferred from the F = 4 state to F = 3 by the Raman beams.
The population transfer is given by
Nfinal − Nbackground
P = , (3.3)
Ninitial − Nbackground
where Ninitial and Nfinal are the fluorescence levels corresponding to the initial and
3.1. EXPERIMENTAL METHODS 43

the final number of atoms respectively, and Nbackground is the background level due
to the stray light.

Figure 3.4: Experimental sequence for a Raman spectroscopy experiment. The

only information obtained is the initial, final and background fluo-
rescence levels. The fluorescence is accumulated over 10 repetitions.

Fig. 3.4 shows the typical time dependence of the fluorescence signal for a Raman
spectroscopy experiment. The time scale is different than shown in Fig. 3.3. To get
better statistics, this measurement is repeated several times (usually 10 − 20) with
a high number of atoms (usually about 40 − 60 per shot). To load the MOT with
atoms after each repetition we lower the gradient of the MOT magnetic field and
set the power of the MOT cooling laser to its maximum power. In this regime the
MOT collects atoms faster and the duration of this intensive loading depends on
the desired number of atoms and on the amount of Cs background gas.

After all repetitions are done, a TTL pulse is sent to the 9.2 GHz frequency
synthesizer which steps the Raman detuning to the next value. Thus the next
repetition of the experiment corresponds to another Raman detuning. For each
set of repetitions we determine the ratio of atoms transferred to the F = 3 state.
The Raman spectrum is the dependence of the population transfer on the Raman
detuning.
44 CHAPTER 3. RAMAN SPECTROSCOPY

3.2 Spectroscopy of Zeeman sublevels

Raman transitions with copropagating Raman beams are insensitive to the atomic
motion and allow us to perform spectroscopy of internal states of atoms. Since m F
sublevels are very sensitive to the magnetic field, Raman spectroscopy of Zeeman
sublevels can be used to precisely measure magnetic fields. Besides, it shows the
efficiency of the optical pumping into a specific mF sublevel and allows us to measure
light shifts.

3.2.1 The linear Zeeman effect

In the absence of a magnetic field the mF sublevels of the Cs hyperfine states are
degenerate. A magnetic field B lifts this degeneracy and the energy shift of an m F
sublevel is given by the linear Zeeman effect

∆EZeeman, F = gF µB B mF , (3.4)

where gF is the Landé factor and µB is Bohr magneton. This corresponds to a

frequency shift of
∆νZeeman, F = ZF B mF , (3.5)

where ZF = gF µB /h is the Zeeman coefficient: Z4 = −351 kHz/G and Z3 =

350 kHz/G [Ste02] (see Fig. 3.5). If the Raman beams are σ + -polarized, only
Raman transitions with ∆mF = 0 can be excited. Thus, one expects to see a
spectrum consisting of 7 equidistant lines with a frequency separation of

∆νadjacent = 2 Z B , (3.6)

where Z = Z3 ≈ −Z4 .

Beam geometry

First of all, an orientation of the quantization axis and proper polarizations of the
laser beams have to be chosen. The dipole trap axis is horizontal and the guiding
field defining the quantization axis was chosen in the vertical direction. To provide
the σ + -polarization of the Raman beams with respect to the chosen quantization
axis, they are circularly polarized and shined in from above. The push-out beam is
always shined in from above, in the radial direction of the dipole trap. A schematic
view of the beam geometry is presented in Fig. 3.6.
3.2. SPECTROSCOPY OF ZEEMAN SUBLEVELS 45

Figure 3.5: Zeeman splitting of the Cs ground states in the presence of a mag-
netic field.

Figure 3.6: Beam geometry for the Raman spectroscopy of Zeeman sublevels.

3.2.2 Analysis of the Raman spectra

The Raman spectrum of the Zeeman sublevels is shown in Fig. 3.7. Here,
the total power of the Raman beams was 340 µW, the duration of the Raman
pulses was 10 µs and the detuning from the excited level was 13.7 GHz. Each
point is a result of 20 measurements with approximately 15 atoms per measurement.
46 CHAPTER 3. RAMAN SPECTROSCOPY

Figure 3.7: The measured Raman spectrum of the Zeeman sublevels.

The spectrum consists of 7 peaks corresponding to 7 possible transitions with

∆mF = 0. The peaks are equidistant with a frequency separation given by the
linear Zeeman effect. The absolute statistical error of the peak heights is about
2.5 %. The full widths at half maximum of the peaks is on the order of 100 kHz, so
the peaks have Fourier-limited widths given by the inverse of the pulse duration.

The entire spectrum is shifted to the red with respect to the literature value of the
hyperfine splitting because of the differential light shift of the ground levels due
to the Raman and Nd:YAG beams and the quadratic Zeeman effect which will be
discussed later.

Calibration of the magnetic compensation coils

Since the Raman spectroscopy is sensitive to the magnetic field, it can be used
to calibrate the compensation coils. Several spectra were recorded with different
magnetic fields applied. Fig. 3.2.2 shows the dependence of the peak positions on
the coil current producing the guiding magnetic field. The straight lines are linear
fits.

After precompensation of the magnetic field in the x and y directions as described

3.2. SPECTROSCOPY OF ZEEMAN SUBLEVELS 47

Figure 3.8: Zeeman splitting of mF sublevels for different values of the current
in the z magnetic coils.

in Sec. 2.3 the guiding field in the z direction is applied. The total magnetic field is
given by q
Btotal = Bpre,2 2 2
x + Bpre, y + (Bextern, z + αz Iz ) , (3.7)
where Bpre, i is the i-th component of the precompensated magnetic field. In the
z direction, the magnetic field Bz consists of the z component of the unknown
external field Bextern, z and the field induced by the z magnetic coil. For large values
of Btotal À Bx , By we can neglect the nonperfect magnetic field compensation and
write
Btotal ≈ Bextern, z + αz Iz . (3.8)
This linear dependence is presented in Fig. 3.9. The proportionality coefficient
between the current and the magnetic field for the z coil can be found as

dB dB d(∆νZeeman )
αz = = . (3.9)
dIz d(∆νZeeman ) dIz

The first term is the inverse of the Zeeman coefficient Z for the cesium ground states,
while the second is the result of the linear fit in Fig. 3.9. Taking into account the
error of the fit, we deduce that
mG
αz = (260 ± 10) . (3.10)
A
The maximum value of the magnetic field produced by the coils is limited by the
maximum current supported by them. Resistive heating of the coils limits the
48 CHAPTER 3. RAMAN SPECTROSCOPY

Figure 3.9: Dependence of the Zeeman splitting on the compensation current

of the z coil.

current to Imax = 5 A. Using (3.10) the maximum field is 1.3 G. Together with the
compensation of the Bextern ≈ 400 mG this leads to the maximum possible value of
the guiding field Bguiding ≈ 1 G. This results in a maximum spacing between the
adjacent peaks in the Zeeman spectrum of

∆νadjacent, max ≈ 0.7 MHz . (3.11)

Spontaneous Raman scattering

If the efficiency of the state selective detection is close to 100%, the background
of the Raman spectra, i.e. the residual population transfer far from any Raman
resonance, is determined only by the number of Raman laser photons spontaneously
scattered by atoms which then decay into the F = 3 level.

An atom in the F = 4 state can scatter any of the three Raman beams. Using (1.15)
the scattering rate is given by
Γ3
µ ¶
IS /2 IC IS /2
ΓRaman = + + , (3.12)
8I0 (∆ − ∆HFS )2 ∆2 (∆ + ∆HFS )2
where IC and IS are intensities of the carrier and sideband beams, respectively. Since
these intensities are proportional to the total power PRaman of the Raman beams,

ΓRaman ∼ PRaman . (3.13)

3.2. SPECTROSCOPY OF ZEEMAN SUBLEVELS 49

The multiplicity of the two ground-state hyperfine levels are 7 and 9, and the prob-
7
ability for an excited atom to decay into the F = 3 state is approximately 7+9 . So,
if the duration of the Raman beams is τ , the fraction of atoms scattered to F = 3
is approximately equal to
7
nscat ≈ ΓRaman · τ , as long as nscat ¿ 1 . (3.14)
7+9
The dependence of the background of the Raman spectrum on the total power of
the Raman beams was extracted from the spectra of the central line. Since for a
Raman pulse duration of 10 µs the spectral lines are too broad, it is difficult to
obtain the correct value for the background. Thus we used only spectra recorded
with a pulse duration of 100 µs. The data together with a linear fit are plotted in
Fig. 3.10. The linear fit supports (3.12).

Figure 3.10: Dependence of the off-resonant population transfer of atoms on

the total intensity of the Raman beams.

For a Gaussian beam with a diameter w, the peak intensity and total power are
connected via the relation
2P
I= . (3.15)
πw2
Since we know the powers of the Raman beams and the scattering rate, using (3.12)
and (3.15) we can estimate the waist of the Raman beams:

wRaman ≈ 80 µm . (3.16)
50 CHAPTER 3. RAMAN SPECTROSCOPY

This value of the waist will be used later to calculate the Rabi frequencies of the
Raman transitions.

Differential light shift

The fits in Fig. 3.2.2 have an offset. The shift of the spectra is determined by
the differential light shifts due to the Raman beams and the Nd:YAG laser and
by the quadratic Zeeman effect. The light shift induced by the Nd:YAG beams is
δYAG ≈ −9 kHz (see Sec. 2.2). The quadratic Zeeman effect for the cesium atom is
given by [Ste02] · ¸
mHz
∆νZeeman, quad = 43 2 · B2 . (3.17)
µT
For B = 1 G, ∆νZeeman, quad ≈ 450 Hz and can be neglected in comparison with the
light shifts.

(a) PRaman = 130 µW, τ = 10 µs (b) PRaman = 64 µW, τ = 100 µs

Figure 3.11: Central peak in the Zeeman spectrum, which corresponds to the
(F = 4, mF = 0) → (F = 3, mF = 0) transition, recorded for
different pulse durations.

To study the dependence of the differential light shift due to the Raman beams on
different experimental parameters we recorded several spectra of the central line
corresponding to the transition (F = 4, mF = 0) → (F = 3, mF = 0). We varied
the total power PRaman of the Raman beams and the duration τ of the Raman
pulses. Two typical spectra are shown in Fig. 3.11. The line widths for both pulse
3.3. SPECTROSCOPY OF VIBRATIONAL LEVELS 51

durations are approximately Fourier-limited.

The theoretical dependence of the differential light shift on the powers of the Raman
beams (A.27) shows the proportionality

δRaman ∝ PRaman . (3.18)

Figure 3.12: Differential light shift of the (F = 4, mF = 0) → (F = 3, mF = 0)

transition.

The experimental dependence of the differential light shift on the total Raman
power is presented in Fig. 3.12. The linear dependence predicted in (3.18) is
supported here. Measurements with different Raman pulse durations shows the
independence of the light shift on the pulse duration.

3.3 Spectroscopy of vibrational levels

The near-zero momentum transfer between the copropagating Raman beams and
an atom makes the Raman spectroscopy described so far insensitive to atomic
velocities. However, counter-propagating beams produce enough momentum
transfer to excite different atomic motional states and thus allow to record the
52 CHAPTER 3. RAMAN SPECTROSCOPY

spectrum of vibrational sidebands of trapped atoms as shown in Sec. 1.2.2.

Scanning the Raman detuning we excite transitions on different sidebands and mea-
sure the population transfer. The experimental methods are the same as those
described above in Sec. 3.1. The timing sequence is similar to the one described in
Sec. 3.1.3.

3.3.1 Experimental parameters

To obtain more atoms and to reduce statistical errors of the measurements, we first
optically pump atoms into a specific mF state of the F = 4 level. Then we induce
Raman transitions to different vibrational levels of the same mF state of the F = 3
level.

We choose to prepare the mF = 0 state, since it is insensitive to fluctuations of

the magnetic field, which broaden all other spectral lines. The optical transition
(F = 4, mF = 0) → (F 0 = 4, mF = 0) is forbidden. Thus a π-polarized laser
resonant with the transition F = 4 → F 0 = 4 pumps atoms into the dark
mF = 0 state (see Fig. 3.13). A π-polarized laser resonant with the transition
F = 3 → F = 4 serves as a repumping laser for the optical pumping.

Figure 3.13: Optical pumping into the mF = 0 state.

To separate the other Raman transitions far enough from the transition
(F = 4, mF = 0) → (F 0 = 4, mF = 0) so that the vibrational sidebands of
the different mF states do not overlap, the maximum possible guiding magnetic
field of about 1 G is applied. This field separates adjacent peaks in the Zeeman
spectrum by about 0.7 MHz (see Sec. 3.2.2).
3.3. SPECTROSCOPY OF VIBRATIONAL LEVELS 53

To record a spectrum of the axial vibrational levels without exciting the radial
oscillations, the Raman beams are set counter-propagating and parallel to the dipole
trap lasers. To couple the mF = 0 states the Raman beams are σ + − polarized and
consequently the guiding magnetic field is set parallel to the Raman beams. The
orientations of the lasers are shown in Fig. 3.14.

Figure 3.14: Orientation of laser beams and guiding field for Raman spec-
troscopy of the vibrational sidebands.

To decrease the off-resonant scattering due to the Raman beams, the Raman laser
was detuned by about 150 GHz to the red from the D2 -transition. For such a big
detuning the interference effect described in Sec. 1.3 becomes more significant (see
(1.38)). To avoid the interference, the AOMs of the Raman carrier and sideband
beams are operated at 105 and 115 MHz respectively as was proposed in Sec. 1.3.
The double-pass configuration of the AOMs increases the frequency spacing between
the carrier and the blue sideband by ∆AOM = 20 MHz. To match this spacing to
the hyperfine splitting of the Cs ground state, the EOM frequency is set to

∆EOM = ∆HFS − ∆AOM ≈ 9.172 GHz .

The powers of the Raman pump and Stokes beams were about 170 and 360 µW
respectively. The Raman pulse duration was 2 ms. To make the entire spectrum
including higher-order sidebands fit within 0.7 MHz, it was recorded in the dipole
trap lowered to 3 % of the maximal depth, which reduces the axial oscillation fre-
quency (2.9) by a factor of about 6. The optical pumping provides about 55 % of
the initially trapped atoms for the spectroscopy experiment.
54 CHAPTER 3. RAMAN SPECTROSCOPY

3.3.2 Spectrum of the vibrational sidebands

The Raman spectrum of the vibrational sidebands is presented in Fig. 3.15. Each
point in the spectrum is a contribution of about 400 atoms which leads to a relative
error of of the peak heights of about 7.5 %. The central peak corresponds to the
carrier transition. Peaks to the left and to the right are the transitions on the red
and the blue sidebands respectively. Several orders of sidebands are visible here.
The positions of the peak maxima are equidistant with a frequency separation of
Ωz = 70 kHz which corresponds to the axial oscillation frequency.

Figure 3.15: Raman spectrum of the vibrational sidebands. Central peak cor-
responds to the transition on carrier. Peaks to the left and to the
right are transitions on red and blue sidebands, respectively.

The width of the central line is 22 kHz. Since the Fourier limit due to the pulse
duration of 2 ms is only 0.5 kHz, the main contribution to the linewidth of the
carrier transition is power broadening. Using (1.12) we obtain a carrier Rabi
frequency of ΩR, 0 = 11 kHz.

The additional asymmetric broadening of the sidebands is due to the contribution

3.3. SPECTROSCOPY OF VIBRATIONAL LEVELS 55

of atoms with relatively high temperatures. Because of the anharmonicity of

the dipole trap the oscillation frequencies of hot atoms are lower than those of
cold atoms. The line shapes result from both the anharmonicity of the trapping
potential and the energy distribution of the atoms in the dipole trap.

The heights of the first two sidebands are comparable with the height of the carrier
peak, while the higher-order sidebands are hardly visible in the spectrum. This is
explained by the dependence of the Rabi frequency and, hence, the π-pulse duration
on the order of the sideband and the vibrational state as given by (1.32). If the
transition involves atoms with different Rabi frequencies, e.g. atoms occupying
different vibrational levels, after an interaction time much longer than the average
π-pulse duration the population transfer will reach 50 % of the available population.
In contrast, if the transition is driven during a time much shorter than the π-pulse
the population transfer will stay very low.

Using (1.32) and considering that the mean vibrational level is n̄ = 4 and the Lamb-
Dicke parameter is η = 0.12, we calculate the number Ni of π-pulses within the 2 ms
pulse duration for the transition between the vibrational levels |n̄i and |n̄ − ii

N0 ≈ 22 , N1 ≈ 7 , N2 ≈ 1.3 , N3 ≈ 0.16 , N4 ≈ 0.017 , etc. (3.19)

Thus 2 ms is not enough time to transfer significant population into the sidebands
of the orders higher than 2nd . If a pulse duration t is much smaller than the π-pulse
Tπ ,i , the population transfer is approximated by
¶2
π2 2
µ
π t
Pi ≈ = N . (3.20)
2 Tπ ,i 4 i

For the 3rd sideband, (3.20) gives a population transfer of P3 ≈ 3.4 %. Since the
corresponding peak in the spectrum is broadened by the anharmonicity of the
potential, the peak height has to be reduced to make the peak area constant.
Thus the height of the 3rd red sideband in the spectrum is only P3, exp = (1.7±0.6) %.

The transitions on the carrier and the first and second sidebands are saturated
since they contain many Rabi oscillation cycles and the population transfer has
to average to about 50 % of the number of atoms pumped to the mF = 0 state,
correspondingly to 27 % of the atoms initially prepared in the MOT. The sideband
peak heights are reduced compared to the carrier due to the spectral width of the
peaks.
56 CHAPTER 3. RAMAN SPECTROSCOPY

Figure 3.16: Dependence of the oscillation frequency on the dipole trap depth.
U0 denotes the maximal potential depth.

The oscillation frequency of the atoms in the dipole trap, i.e. the separation of
peaks in the recorded spectrum, depends on the depth U of the trapping potential
as given by (2.9)

Ωz ∝ U 1/2 . (3.21)

We recorded Raman spectra for different depths of the dipole trap from 100 %
to 3 % of the maximal value U0 and measured the frequency separation between
the peaks in the raman spectra. The result is plotted in Fig. 3.16. The X axis is
normalized to the squared maximal potential depth U0 . The straight line is the
linear fit and confirms (3.21).

The oscillation frequency of the atoms measured in the deep dipole trap, which
corresponds to a total power of the Nd:YAG laser of 4 W, is Ωz = (530 ± 20) kHz.
This is in agreement with the theoretically calculated value Ωz, theory = 570 kHz
(see Sec. 2.2).
Chapter 4

Rabi oscillations

The coherence properties of an optical transition involving a fast-decaying atomic

state are not easy to determine since spontaneous decay destroys the coherence and
the coherence time of this process is limited by the lifetime of the states involved.
To achieve long coherence times a Raman transition coupling two long-lived states
can be used.

Rabi oscillations between two coupled states prove the coherence of this coupling.
In the following chapter I will describe an experiment measuring Rabi oscillations
induced by the Raman beams and show the presence of the interference of the two
Raman transitions discussed in Sec. 1.3.

4.1 Measurements of Rabi oscillations

Rabi oscillations are oscillations of the atomic population between two coherently
coupled states. In our experiment we couple the cesium ground-state hyperfine lev-
els by a stimulated Raman transition and observe the oscillations of the population
transfer between these two states.

To measure the Rabi oscillations of the cesium atoms we first prepare them in
the F = 4 hyperfine state. Then, with variable duration, we apply Raman beams
resonant with the two-photon transition between the hyperfine states. Finally, we
state-selectively detect the atoms. The methods of state preparation and detection
are the same as those used in experiments on the Raman spectroscopy. The
experimental timing sequence is similar to that described in Sec. 3.1.3, but instead
of scanning the Raman detuning, after each measurement we increase the duration
of the Raman pulse.

57
58 CHAPTER 4. RABI OSCILLATIONS

To observe population oscillation between states which are insensitive to the

fluctuations of magnetic fields we optically pump atoms into the mF = 0 state. In
order not to excite the vibrational sidebands, we use copropagating Raman beams.
The orientations of other lasers and the guiding magnetic field are the same as in
Fig. 3.14.

The modulation frequencies of the AOMs of the Raman beams are chosen to sup-
press one of the two Raman transitions as described in Sec. 3.3.1. Consequently the
EOM is driven with the modulation frequency ∆EOM ≈ ∆HFS −20 MHz = 9.172 GHz.

The dipole trap and Raman lasers shift the hyperfine levels producing the differen-
tial light shifts given by (2.13) and (A.27). To determine the resonance frequency
of the transition (F = 4, mF = 0) → (F = 3, mF = 0) we first recorded a Raman
spectrum of the Zeeman sublevels in the presence of the Raman beams. The
Raman transition was induced in the dipole trap lowered to 7.5 % of its maximum
value. The powers of the Raman sideband and carrier beams were 240 µW and
360 µW respectively. The detuning from the D2 -transition was 150 GHz. At these
parameters the frequency of the transition between the mF = 0 states, i.e. the
position of the central line in the spectrum, was found to be 5 kHz red-detuned
with respect to ∆HFS .

Next we set the frequency difference of the Raman beams to the hyperfine resonance
and measured the dependence of the population transferred into F = 3 on the
duration of the Raman pulse. The result is plotted in Fig. 4.1. The line is a fit with
a fit function given by
µ ¶
Pmax
f (t) = 1 − exp(−γ t) · cos(2π ΩR t) + P0 , (4.1)
2

where Pmax denotes the amplitude of the oscillation and P0 is the offset due to
imperfect state preparation and detection. The cosine function represents the Rabi
oscillations, while the exponential shows the decoherence. Thus ΩR is the Rabi
frequency and τ = 1/γ is the decoherence time of the system.

The fit yields a Rabi frequency of ΩR = 13.4 kHz and thus a π-pulse duration of
Tπ = 36 µs. Using (1.40) and (1.17) for the given experimental parameters we get a
theoretical value of the Rabi frequency of ΩR, theor = 37.5 kHz. The deviation of the
experimental value from the theoretical expectation can be explained by imperfect
alignment of the Raman beams into the MOT which reduces the laser power, and
thus the Rabi frequency.
4.2. INTERFERENCE EFFECT 59

Figure 4.1: Rabi oscillations induced by the Raman lasers.

The states of the atom in the dipole trap are shifted by the ac Stark effect pro-
portional to the laser intensity at the atom’s position (Sec. 2.2). Thus, on average
the hot atoms experience smaller light shifts than the cold ones. This results in a
spread of Rabi frequencies and causes decoherence. The fit shown above yields a
decoherence time of about 2 ms which is much longer than Tπ . This allows us to
prepare any superposition of the hyperfine states

|Ψi = α|F = 3i + β|F = 4i . (4.2)

4.2 Interference effect

To experimentally prove the interference of the Raman transitions calculated in
Sec. 1.3 we recorded Rabi oscillations of the system with the two Raman transitions
being resonant. In other words, the frequency differences between each sideband
and the carrier equal the hyperfine splitting of the Cs ground state. To prepare such
a system we set the frequencies of both AOMs to the same value, i.e. 110 MHz and
the EOM frequency to match the light-shift corrected hyperfine splitting 9.192 GHz.
The recorded dependence of the population transfer on the pulse duration is shown
in Fig. 4.2.
60 CHAPTER 4. RABI OSCILLATIONS

Figure 4.2: Rabi oscillation in the presence of interfering Raman transitions

with fluctuating relative phases.

Any acoustical vibrations of optical elements randomly change the phase of a

light field. Thus the separated Raman beams have the phase difference ∆ϕ
(A.29) fluctuating randomly and resulting in random interference (A.28). Thus,
Fig. 4.2 shows the average over Rabi oscillations for all possible interferences: from
destructive to constructive. All of them start at zero and their sum averages quickly
to constant value.

In the absence of this interference effect we would still expect to see Rabi oscillations
but with a Rabi frequency being about two times greater than before since then
the two Rabi frequencies would just sum up. Thus, this measurement proves the
presence of the interference of the Raman transitions.

To observe the presence of only the destructive interference, we recorded the Rabi
oscillations induced by the non-separated Raman beams. For this purpose we
disabled the HDW interferometer by blocking its retro-reflecting prism. In this case
the Raman beams are not separated and the light exits the interferometer at only
4.2. INTERFERENCE EFFECT 61

one output.

The measured Rabi oscillations in the presence of destructive interference are

shown in Fig. 4.3. The Rabi frequency is ΩR = 740 Hz. The powers of the pump
and Stokes beams were 53 µW and 26 µW respectively and the resonance of the
transition between mF = 0 states was found to be 130 Hz red-detuned from ∆HFS .

Figure 4.3: Rabi oscillations in the presence of the interference of two Raman
transitions.

Let us define the reduction factor r as a ratio of the Rabi frequencies in the absence
(1.40) and presence (1.38) of the destructive interference:
ΩR ∆ + ∆HFS
r = destr
= ≈ 17.3 . (4.3)
ΩR ∆HFS
To compare the Rabi frequencies obtained experimentally both with and without
the interference we have to take into account the fact that the powers of the pump
and Stokes beams were different in these two experiments. The Rabi frequency of
the electric-dipole interaction depends on the light intensity
r
I
Ω = Γ , (4.4)
2I0
62 CHAPTER 4. RABI OSCILLATIONS

where for the Gaussian beam the intensity is given by

2P
I = . (4.5)
πw2
Thus, the reduction factor is given by
s
Ω0R PP00 PS00
rexper = ≈ 3.3 , (4.6)
Ω00R PP0 PS0

where we mark the parameters of the ”no interference” and ”interference” cases
with one and two primes respectively.

The fact that the theoretical reduction factor is bigger than the experimental one
can be a consequence of the imperfect separation of the Raman beams. If the AOMs
of the Raman setup have different modulation frequencies, the imperfect separation
results in a light field consisting of six beams instead of three. It produces several
more possible Raman transitions which interfere destructively and reduce the Rabi
frequency.
Summary and outlook

The goal of my diploma work was to develop methods to perform Raman spec-
troscopy of single neutral atoms stored in an optical dipole trap and to demonstrate
coherent manipulation of internal states with optical techniques.

As part of my work I improved the experimental setup for the creation and
manipulation of Raman laser beams which was built up before [Mue01]. Raman
transitions between the cesium hyperfine states require two separate laser beams,
whose frequency difference matches the hyperfine splitting. For this purpose we
use a single diode laser and generate sidebands at 9.2 GHz by an electro-optical
modulator.

To determine whether a Raman transition took place, we have to prepare the

initial state and detect the final state after the Raman transition has been
induced. A method to detect the hyperfine state of a single atom was developed.
For this purpose a resonant probe laser heats and pushes the atoms out of the
dipole trap if they are in the upper hyperfine state. The atoms in the lower
hyperfine state, however, are unaffected by the probe laser and remain trapped.
The initial and final number of atoms is measured by observing their fluorescence
in the MOT. The efficiency of this state-selective detection method is close to 100 %.

The Raman spectroscopy allowed us to perform high-resolution spectroscopy of

single atoms trapped in the dipole trap. Raman spectra of Zeeman sublevels
were recorded to develop methods of optical pumping into specific mF sublevels,
to precisely measure magnetic fields, to calibrate the apparatus and to measure
differential light shifts.

One of the most important results of my work was Raman spectroscopy of vibra-
tional sidebands of trapped atoms. The counter-propagating Raman beams allowed
us to resolve vibrational sidebands and to directly measure the oscillation frequency
in the dipole trap. Later this spectroscopy will be used to measure the temper-
ature of trapped atoms by comparing the heights of the first red and blue sidebands.

63
64 Summary and outlook

To study the coherent properties of Raman transitions Rabi oscillations were

recorded. They showed a coherence time much longer than a single Rabi cycle.
This allows us to coherently manipulate the hyperfine states by the Raman coupling.

Our system consists of three Raman beams interacting with a three-level atom and
producing two possible Raman transitions. I performed a full quantum-mechanical
treatment of this system by solving the corresponding Schrödinger equation. The
calculation revealed interference between the two Raman transitions. By comparing
the Rabi oscillation induced by one and two Raman transitions this interference
effect was experimentally confirmed. To avoid the interference, the sidebands
generated by the 9.2 GHz EOM were shifted with respect to the carrier to suppress
one of the Raman transitions.

To perform cavity QED experiments we plan to deterministically inject one or

more atoms into a high-finesse cavity . In order to provide for a large and constant
atom-cavity coupling, we have to better localize the trapped atom inside the cavity
mode. For this purpose the atom has to be cooled down to its lowest vibrational
level. Besides, the cooling increases the lifetime of atoms in the dipole trap and the
coherence time. The cooling of the tightly bound atom can be achieved by using
Raman sideband cooling [Kuk89, Mon95, Vul98, Ker02]. Our capability to
resolve the motional sidebands as well as to coherently transfer atomic population
between atomic states using Raman transitions will allow us to perform Raman
sideband cooling of trapped atoms.

Raman transitions can also serve for coupling of internal and external atomic de-
grees of freedom. For example it was proposed to use Raman transitions to map a
superposition of the hyperfine states into a superposition of the vibrational levels
[Mon97, Kin99], thus allowing for coherent control also over the atomic motion.
Appendix A
Interaction of a three-level atom
with three lasers

Our model system consists of three Raman beams interacting with a three-level atom
via the electric dipole interaction (see Sec. 1.3). The level scheme is shown in Fig. 1.7.
Here ωi denote the laser frequencies, δ is the Raman detuning considered here to
compensate for light shifts. Below I present the quantum mechanical treatment of
this interaction.

Figure A.1: Coupling of three Raman lasers with a three-level atom

65
66 APPENDIX A.

Theoretical basis
To perform a quantum-mechanical treatment of a system, it is necessary to
construct its Hamiltonian and to solve the corresponding Schrödinger equation.
This provides a description for the time-dependent probability amplitudes.

The time-dependent Schrödinger equation governing the time evolution of the system
reads
d
i~ Ψ(t) = Ĥ(t) · Ψ(t), (A.1)
dt
where Ĥ(t) = Ĥ 0 + V̂ (t) is the full Hamiltonian consisting of the unperturbed
Hamiltonian Ĥ 0 , which defines the energy levels of the atom, and the operator V̂ (t)
of the time-dependent interaction. Ψ(t) is a state vector, which can be expressed as
a superposition of the eigenstates ψn of Ĥ 0 :
X
Ψ(t) = ψn · Cn∗ (t). (A.2)
n

The coefficient Cn∗ (t) is the probability amplitude which, squared, yields the popula-
tion of a state |ni. Substitution of (A.2) into (A.1) and the use of the orthogonality
of the basis vectors ψn give us the set of equations for the probability amplitudes:

d ∗ 0 ∗
X
i~ Cm (t) = Em Cm (t) + Vmn (t)Cn∗ (t) , (A.3)
dt n

0
where Em is an eigenvalue of Ĥ 0 and Vmn (t) = hψm |V̂ (t)|ψn i is the matrix element
of the interaction operator.

Application to the three-level atom

By extracting time-dependent phases from the probability amplitudes we get some
flexibility in presenting the expression for the state vector. Thus the time-dependent
state vector of a three-level atom may be written as
3
X
Ψ(t) = ψn · Cn (t)e−iξn (t) , (A.4)
n=1

where ξn (t) is a time-dependent state phase. Its presence does not change quantum-
mechanical observables but the proper choice of the phase allows to significantly
simplify the expressions encountered in the calculations.
INTERACTION OF A THREE-LEVEL ATOM WITH THREE LASERS 67

The interaction of the electromagnetic radiation with the atom is primarily consid-
ered to be the electric dipole interaction:

V̂dip (t) = −d̂ · E(t). (A.5)

Here dˆ is the operator of the electric dipole moment, and the electric field consisting
of three monochromatic waves is represented as
3
X
E(t) = Ek cos(ωk t − ϕk ). (A.6)
k=1

The matrix element of the electric dipole interaction is given by

Vmn (t) = Vnm (t) = −dmn · E(t), with dmn = hψm |d̂|ψn i. (A.7)

Calculation of dnm and, consequently, of Vnm is in general a nontrivial problem

since the basis state ψn of a multi-level atom is not easily calculated. To avoid this
inconvenience we assume d12 = d32 and introduce the Rabi frequency for each laser
beam as
d12 · Ek
Ωk = − . (A.8)
~
This definition leads to
3
X
Vmn (t) = Vnm (t) = ~Ωk cos(ωk t − ϕk ). (A.9)
k=1

Note that the dipole matrix elements d11 , d13 = d31 , d33 vanish as those couple
states with the same parity. Thus only V12 = V21 and V32 = V23 are the nonvanishing
matrix elements of the interaction and the Hamiltonian of the chosen model has the
following matrix form

E10
 
V12 (t) 0
Ĥ =  V21 (t) E20 V23 (t)  (A.10)
0
0 V32 (t) E3

Using (A.4) we derive the time-dependent Schrödinger equation in the form similar
to (A.3):
3
d ˙
X
i~ Cm (t) = −~ξm (t) Cm (t) + Hmn (t) Cn (t) ei(ξm (t)−ξn (t))
dt n=1
3
− ~ξ˙m (t)] · Cm (t) +
X
0
= [Em Vmn (t) Cn (t) ei(ξm (t)−ξn (t)) . (A.11)
n=1
68 APPENDIX A.

Now it is practical to combine equations (A.11) for the probability amplitudes

with the Hamiltonian (A.10) and write the matrix form of the time-dependent
Schrödinger equation

E10 − ~ξ̇1 (t) V12 (t) · ei(ξ1 −ξ2 )

    
C (t) 0 C1 (t)
d  1
i~ C2 (t)  =  V21 (t) · e−i(ξ1 −ξ2 ) E20 − ~ξ̇2 (t) V23 (t) · e −i(ξ 3 −ξ 2 )   C2 (t)  (A.12)
dt C3 (t) 0 V32 (t) · ei(ξ3 −ξ2 ) E30 − ~ξ̇3 (t) C3 (t)

Consequently the time-dependent Schrödinger equation as a set of differential equa-

tions with respect to the probability amplitudes Cn now reads

d
i~ C1 (t) = [E10 − ~ξ˙1 (t)] · C1 (t) (A.13)
dt
~
+ [Ω1 exp(−iω1 t + iϕ1 ) + Ω1 exp(iω1 t − iϕ1 )
2
+Ω2 exp(−iω2 t + iϕ2 ) + Ω2 exp(iω2 t − iϕ2 )
+Ω3 exp(−iω3 t + iϕ3 ) + Ω3 exp(iω3 t − iϕ3 )] · C2 (t) · ei(ξ1 −ξ2 )

d
i~ C2 (t) = [E20 − ~ξ˙2 (t)] · C2 (t) (A.14)
dt
~
+ [Ω1 exp(−iω1 t + iϕ1 ) + Ω1 exp(iω1 t − iϕ1 )
2
+Ω2 exp(−iω2 t + iϕ2 ) + Ω2 exp(iω2 t − iϕ2 )
+Ω3 exp(−iω3 t + iϕ3 ) + Ω3 exp(iω3 t − iϕ3 )] · C1 (t) · e−i(ξ1 −ξ2 )

~
+ [Ω1 exp(−iω1 t + iϕ1 ) + Ω1 exp(iω1 t − iϕ1 )
2
+Ω2 exp(−iω2 t + iϕ2 ) + Ω2 exp(iω2 t − iϕ2 )
+Ω3 exp(−iω3 t + iϕ3 ) + Ω3 exp(iω3 t − iϕ3 )] · C3 (t) · e−i(ξ3 −ξ2 )

d
i~ C3 (t) = [E30 − ~ξ˙3 (t)] · C3 (t) (A.15)
dt
~
+ [Ω1 exp(−iω1 t + iϕ1 ) + Ω1 exp(iω1 t − iϕ1 )
2
+Ω2 exp(−iω2 t + iϕ2 ) + Ω2 exp(iω2 t − iϕ2 )
+Ω3 exp(−iω3 t + iϕ3 ) + Ω3 exp(iω3 t − iϕ3 )] · C2 (t) · ei(ξ3 −ξ2 )
INTERACTION OF A THREE-LEVEL ATOM WITH THREE LASERS 69

Interaction picture
To eliminate the diagonal elements of the Hamiltonian the interaction picture of
probability amplitudes is applied. In this case the state phases ξn are set to

 ξ˙1 = E10 /~


ξ˙ = E20 /~ (A.16)
 ˙2 0
ξ3 = E3 /~ − δ

and the corresponding phase differences are

ξ1 − ξ2 = −(E20 − E10 )/~ · t

½
= −ω21 t
(A.17)
ξ3 − ξ2 = −(E20 − E30 )/~ · t − δ · t = −ω23 t − δ · t,

where ω21 and ω23 are frequencies of the transitions |1i → |2i and |3i → |2i re-
spectively. According to the level scheme (see Fig. A.1) the following equalities
hold


 ω1 − ω2 = ∆HFS − δ
ω2 − ω3 = ∆HFS − δ




ω21 − ω1 = ∆




ω21 − ω2 = ∆ + ∆HFS − δ

(A.18)

 ω21 − ω3 = ∆ + 2∆HFS − 2δ
 ω23 − ω1 = ∆ − ∆HFS



ω − ω2 = ∆ − δ

 23



ω23 − ω3 = ∆ + ∆HFS − 2δ

Using (A.16), (A.17) and (A.18) the differential equations (A.13), (A.14) and (A.15)
can be rewritten accordingly.

Adiabatic elimination of the excited level

To reduce our three-level system to an effective two-level system, we adiabatically
eliminate the excited state. We expect the population of the state |2i to be very
small compared to the other states but fast varying, since the detunings from the
one-photon resonance are considerable (∆ À ∆HFS and ∆ À Ωi ). Thus, C1 (t)
and C3 (t) are approximated as constant in comparison with dtd C2 (t) and integrate
(A.14). Then we put the resulting solution for C2 (t) into the expressions for dtd C1 (t)
and dtd C3 (t) and get two equations with only two variables C1 (t) and C3 (t).

Rotating-wave approximation
To eliminate the high-frequency components, i.e. exponentials oscillating at the sum
of optical frequencies, we assume that the state populations are constant during an
70 APPENDIX A.

optical period. As long as the evolution of the state populations are supposed
to be slow compared to the hyperfine splitting frequency, terms oscillating with
∆ or even ∆HFS can be neglected as well. Thus, we neglect all fast-oscillating
terms. This approximation is known as the rotating-wave approximation. Denoting
∆ϕ21 = ϕ2 − ϕ1 and ∆ϕ32 = ϕ3 − ϕ2 we get

1 Ω21 Ω22 Ω23

· ¸
d
i C1 (t) = − + + · C1 (t) (A.19)
dt 4 ∆ ∆ + ∆HFS − δ ∆ + 2∆HFS − 2δ
· ¸
1 Ω1 Ω2 Ω2 Ω3
− exp(i∆ϕ21 ) + exp(i∆ϕ23 ) · C3 (t)
4 ∆ ∆ + ∆HFS − δ
· ¸
d 1 Ω1 Ω2 Ω2 Ω3
i C3 (t) = − exp(−i∆ϕ21 ) + exp(−i∆ϕ23 ) · C1 (t) (A.20)
dt 4 ∆ ∆ + ∆HFS − δ
Ω21 Ω22 Ω23
· ¸
1
− + + · C3 (t)
4 ∆ − ∆HFS + δ ∆ ∆ + ∆HFS − δ

+ δ · C3 (t)

Reduced two-level system

After adiabatic elimination of the second level and application of the rotating wave
approximaiton we now have an effective two-level system with the time-dependent
Schrödinger equation µ ¶ µ ¶
d C1 C1
= i Ŵ . (A.21)
dt C3 C3
Here the interaction matrix is
Ω2 Ω22 Ω23
 
Ω1 Ω2 i∆ϕ21 Ω2 Ω3
1  ∆1 + ∆+∆HFS −δ + ∆+2∆HFS −2δ ∆ e + ∆+∆ HFS −δ
ei∆ϕ32 
Ŵ = Ω21 Ω22 Ω23
4 Ω1 Ω2 −i∆ϕ21 Ω2 Ω3
∆ e + e−i∆ϕ32
∆+∆HFS −δ ∆−∆HFS +δ + ∆ + ∆+∆HFS −δ − 4δ
(A.22)
As an abbreviation the following symbolic notation will be from now on used
µ ¶
1 A B
Ŵ = (A.23)
4 C D
It is straightforward to show that (A.21) is identical to the set of equations
(
C̈1 (t) + i A+D
4
Ċ1 (t) + AD−BC
16
C1 (t) = 0
4i A (A.24)
C3 (t) = − B Ċ1 (t) − B C1 (t),

which can easily be solved if we suppose that the atom is initially in level |1i:
½
C1 (0) = 1
C3 (0) = 0.
INTERACTION OF A THREE-LEVEL ATOM WITH THREE LASERS 71

Rabi oscillation
To find the probability for occupying a specific state one should square the corre-
sponding probability amplitude. Squaring the solutions of (A.24) gives us the time
evolution of the population similar to the two-beam case described in Sec. 1.1:
|C1 (t)|2 = 1 − Λ · sin2 ( Ω2R t)
½

|C3 (t)|2 = Λ · sin2 ( Ω2R t) ,

where
4BC
Λ = (A.25)
(A − D)2 + 4BC
1p
and Ω0 = (A − D)2 + 4BC. (A.26)
4
We see that the probability for finding the atom on each hyperfine level oscillates
at frequency Ω0 . The amplitude of the oscillation Λ is maximal and equal to 1, if
A = D, i.e. if the two-photon resonance condition (also called Raman resonance) is
fulfilled. Assuming δ ¿ (∆, ∆HFS ) this leads to
Ω21 Ω22 Ω23
µ ¶
∆HFS
δ= + + . (A.27)
4 (∆ − ∆HFS )∆ ∆(∆ + ∆HFS ) (∆ + ∆HFS )(∆ + 2∆HFS )
This expression represent the differential light shift caused by the three Raman
lasers acting on each hyperfine level. This light shift has to be considered in the
experiment to provide the two-photon resonance condition.

The resonant Rabi frequency is then

s
1 Ω21 Ω22 Ω22 Ω23 Ω1 Ω22 Ω3
ΩR = + + 2 · cos ∆ϕ , (A.28)
2 ∆2 (∆ + ∆HFS )2 ∆(∆ + ∆HFS )

where ∆ϕ ≡ ∆ϕ32 − ∆ϕ21 = ϕ1 + ϕ3 − 2ϕ2 . (A.29)

If Ω1 = Ω3 , the expression (A.28) shows the interference between the two possible
Raman transitions in the system, whose Rabi frequencies are
Ω1 Ω2 Ω2 Ω1
ΩR1 = and ΩR2 = respectively.
2∆ 2(∆ + ∆HFS )
The Rabi frequency ΩR at which the atom oscillates between the two levels strongly
depends on the phase difference ∆ϕ
q
ΩR = Ω2R1 + Ω2R2 + 2ΩR1 ΩR2 · cos ∆ϕ . (A.30)
72 APPENDIX A.

The discussion of the interference effect as well as a method how to avoid it are
presented in Sec. 1.3.

Power broadening
The amplitude of the population oscillation Λ can be rewritten from (A.25)

Ω2R
Λ = , (A.31)
Ω2R + δR2

where δR is the Raman detuning when light shifts are taken into account. The
amplitude drops down to 1/2 at δR = ΩR . Thus the width of the Raman transition
due to power broadening is given by

∆ωpower = 2 ΩR . (A.32)

This result is the same as for the two-beam case described in Sec. 1.1.
Bibliography

[Aff00] C. Affolderbach, A. Nagel, S. Knappe, C. Jung, D. Wiedenmann and

R. Wynands, Nonlinear spectroscopy with a vertical-cavity surface-emitting
laser (VCSEL), Appl. Phys. B 70, p. 407–413 (2000)

[Alt02] W. Alt, An objective lens for efficient fluorescence detection of single atoms,
International Journal for Light and Electron Optics 113(3), p. 142–144 (2002)

[And95] M. H. Anderson, J. R. Ensher, M. R. Matthews, C. E. Wieman

and E. A. Cornell, Observation of Bose-Einstein Condensation in a Di-
lute Atomic Vapor , Science 269, p. 198 (1995)

[Bal00] V. I. Balykin, V. G. Minogin and V. S. Letokhov, Electromagnetic trap-

ping of cold atoms, Rep. Prog. Phys. 63 (2000)

[Bei00] A. Beige, D. Braun, B. Tregenna and P. L. Knight, Quantum Comput-

ing Using Dissipation to Remain in a Decoherence-Free Subspace, Phys. Rev.
Lett. 85, p. 1762 (2000)

[Ben00] C. H. Bennett and D. P. DiVincenzo, Quantum information and compu-

tation, Nature 404, p. 247–255 (2000)

[Ber97] P. R. Berman, Atom Interferometry, Academic Press (1997)

[Ber98] K. Bergmann, H. Theuer and B. W. Shore, Coherent population transfer

among quantum states of atoms and molecules, Reviews of Modern Physics
70(3), p. 1003–1025 (1998)

[Chu02] S. Chu, Cold atoms and quantum control , Nature 416, p. 206–210 (2002)

[EOM01] Practical Uses and Applications of Electro-Optic Modulators, technical report,

New Focus, Inc. (2001)

[Gri00] R. Grimm, M. Weidemller and Y. Ovchinnikov, Optical dipole traps for

neutral atoms, Adv. At. Mol. Opt. Phys. 42, p. 95 (2000)

[Hau00] D. Haubrich, M. Dornseifer and R. Wynands, Lossless beam combiners

for nearly equal laser frequencies, Rev. Sci. Inst. 71(2), p. 225–230 (2000)

73
74 BIBLIOGRAPHY

[Ker02] A. J. Kerman, Raman Sideband Cooling and Cold Atomic Collisions in Op-
tical Lattices, dissertation, Stanford University (2002)

[Kin99] B. E. King, Quantum State Engineering and Information Processing with

Trapped Ions, dissertation, University of Colorado (1999)

[Kit01] J. Kitching, L. Hollberg, S. Knappe and R. Wynands, Compact atomic

clock based on coherent population trapping, Electron. Lett. 37, p. 1449–1451
(2001)

[Kuh01] S. Kuhr, W. Alt, D. Schrader, M. Mueller, V. Gomer and

D. Meschede, Deterministic Delivery of a Single Atom, Science 293, p. 278–
280 (2001)

[Kuk89] J. R. Kuklinski, U. Gaubatz, F. T. Hioe and K. Bergmann, Adiabatic

population transfer in a three-level system driven by delayed laser pulses, Phys.
Rev. A 40(11), p. 6741–6744 (1989)

[Lau84] K. Y. Lau, C. Harder and A. Yariv, Direct modulation of the semiconduc-

tor laser at frequencies higher than 10GHz by low-temperature operation, Appl.
Phys. Lett 44, p. 273–275 (1984)

[Met99] H. J. Metcalf and P. van der Straten, Laser Cooling and Trapping
(1999)

[Mon95] C. Monroe, D. M. Meekhof, B. E. King, S. R. Jefferts, W. M. Itano,

D. J. Wineland and P. Gould, Resolved-Sideband Raman Cooling of a
Bound Atom at the 3D Zero-Point Energy, Phys. Rev. Lett. 75(22), p. 4011–
4014 (1995)

[Mon97] C. Monroe, D. Leibfried, B. E. King, D. M. Meekhof, W. M. Itano

and D. J. Wineland, Simplified quantum logic with trapped ions, Phys. Rev.
A 55(4), p. 2489–2491 (1997)

[Mon02] C. Monroe, Quantum information processing with atoms and photons, Nature
416, p. 238–246 (2002)

[Mue01] M. Mueller, Ramankühlung einzelner Atome, diplomarbeit, Universität

Bonn (2001)

[Pel95] T. Pellizari, S. A. Gardiner, J. I. Cirac and P. Zoller, Decoherence,

Continuous Observables, and Quantum Computing: A Cavity QED Model ,
Phys. Rev. Lett. 75, p. 3788 (1995)

[Rin99] J. Ringot, Y. Lecoq, J. C. Garreau and P. Szriftgiser, Generation

of phase-coherent laser beams for Raman spectroscopy and cooling by direct
current modulation of a diode laser , Eur. Phys. J. D 65, p. 285–288 (1999)
BIBLIOGRAPHY 75

[Rin01] J. Ringot, P. Szriftgiser and J. C. Garreau, Subrecoil Raman spec-

troscopy of cold cesium atoms, Phys. Rev. A 65 (2001)

[Sch94] F. Schwabl, Quantum Mechanics, Springer, second edition (1994)

[Sch96] O. Schmidt, R. Wynands, Z. Hussein and D. Meschede, Steep dispersion

and group velocity c/3000 in coherent population trapping, Phys. Rev. A 53(1),
p. 27–30 (1996)

[Sch01] D. Schrader, S. Kuhr, W. Alt, M. Müller, V. Gomer and

D. Meschede, An Optical Conveyor Belt for Single Neutral Atoms,
[Link].B (2001)

[Sho90] B. W. Shore, The Theory of Coherent Atomic Excitation (1990)

[Ste02] D. A. Steck, Cesium D Line Data, Los Alamos National Laboratory (2002)

[Vul98] V. Vuletic, C. Chin, A. J. Kerman and S. Chu, Degenerate Raman

Sideband Cooling of Trapped Cesium Atoms at Very High Atomic Densities,
Phys. Rev. Lett. 81(26), p. 5768–5771 (1998)

[Win79] D. J. Wineland and W. M. Itano, Laser cooling of atoms, Phys. Rev. A

20(4), p. 15291540 (1979)
Acknowledgements

During the last year I had the pleasure of studying and working with a number of
wonderful people who deserve thanks for their help and support during my research.

First of all I would like to thank my supervisor, Prof. D. Meschede, who first
revived my interest in the life of photons and atoms by his quantum optics class
and later provided the opportunity to have fun playing with single atoms in the
lab. I would like to acknowledge Prof. E. Klempt for his very kindwillingness to be
the coreferent of my master thesis.

I am very grateful to our ”single atoms” Ph.D. students Stefan Kuhr, Wolfgang Alt
and Dominik Schrader, who are one of the most creative, optimistic and charming
people I have associated with. Thank you for sure guiding me through my thesis.
Thank you for gratis sharing all secrets of our laboratory. Thank you for being
walking encyclopedias of optics and electronics which are ready to answer every my
ignorant question. Thank you for your continuous readiness to spend a time (which
you usually do not have) helping me with the experimental setup. Thank you for
reading again and again all preliminary versions of the current thesis. Thank you
for your tireless efforts in making me be a real physicist.

My great thanks to Dr. Viktor Gomer who first introduced me to the fascinating
world of single atoms and help me not to be afraid of mathematics. I am also
grateful to Dr. Arno Rauschenbeutel for stimulating discussions which always
throw light on physical phenomena which I did not understand before.

I an grateful to my ”colleagues” - our Diploma students Yevgen Miroshnichenko and

Wenjamin Rosenfeld with whom I shared the work in the lab, fruitful discussions
and many cheerful moments in the past year.

I would like to acknowledge all the members of the group and the institute
administrative staff for the good time we shared, for friendly atmosphere you keep
in the institute, for your assistance and kindness.

Thanks to my parents. Whatever success I have in my life is due to the beginnings

which they gave me.

And finally, I would like to thank my beloved wife, Ira, for being my strongest
support and for your patience in waiting for me after our numerous night measure-
ments.

77