Trapped-ion physics lecture

Daniel Kienzler

June 29, 2021

2
Contents

1 Trap and motion 7

1.1 Trapping a charge in 1D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Trapping a charge in 3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.1 Laplace equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.2 Paul trap and pseudopotential approximation . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.3 Penning trap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 Linear Paul trap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.1 The Mathieu equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3.2 Typical imperfections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.3 Surface traps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4 Quantum harmonic oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4.1 Decoherence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4.2 Ion heating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.5 Multiple ions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.5.1 Decoherence in multi-ion crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.6 Dynamic operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.6.1 Motional excitation, coherent states, squeezed states . . . . . . . . . . . . . . . . . . . . 21
1.6.2 Trap arrays and transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2 Electronic structure 25
2.1 Zeeman effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2 Transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2.1 Electric dipole transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2.2 Magnetic dipole transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2.3 Electric quadrupole transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3 Different types of qubits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.1 Qubit decoherence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.2 Field-independent qubits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4 State preparation and dark states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.5 Readout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.6 Connecting qubit states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.6.1 Quadrupole transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.6.2 Magnetic dipole transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.6.3 Stimulated Raman transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.7 Two-level system, Rabi, Ramsey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.7.1 Rabi model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.7.2 Ramsey method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.7.3 Basic working principle of an atomic clock . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3 Spin-motion coupling 37
3.1 Atom-Field interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.1.1 Interaction with an atom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.1.2 Trapped ions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 Lamb-Dicke expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2.1 Physical interpretation of the LD parameter . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2.2 Laser beam geometry and microwaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.3 Transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.3.1 Carrier transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3
4 CONTENTS

3.3.2 First sideband transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3.3.3 Limit of approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3.4 Higher orders, full Lamb-Dicke expansion . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3.5 Transitions in 3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.3.6 Photon scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

4 Laser cooling 47
4.1 Doppler cooling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2 Sideband cooling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.4 Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.5 Cooling diagnostics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.6 Sympathetic cooling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

5 State dependent force 53

5.1 X-basis force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.2 Z-basis force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.3 Geometric phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

6 Spin-spin entanglement 57
6.1 Cirac-Zoller gate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.2 Mølmer-Sørenesen gate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
6.2.1 Error sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.2.2 Two-loop MS gate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6.2.3 Light shift gate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

7 Quantum computation 63
7.1 Some notes on trapped ion quantum information processing . . . . . . . . . . . . . . . . . . . . 63
7.1.1 Two level control / single qubit gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
7.1.2 Single-shot detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
7.1.3 Motional heating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
7.1.4 Imaging and lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
7.2 Approaches to scaling TIQI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
7.2.1 Long strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
7.2.2 QCCD architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
7.2.3 Photonic links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
7.3 Further technical aspects and solutions for scaling TIQI . . . . . . . . . . . . . . . . . . . . . . 68
7.4 Quantum error correction with trapped ions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

8 Quantum metrology 69
8.1 Quantum assisted metrology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
8.1.1 Doppler broadening and shifts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
8.1.2 Quantum Logic spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
8.2 Quantum enhanced metrology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
8.2.1 Motional squeezing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
8.2.2 Spin entanglement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
8.3 Atomic (ion) clocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
8.4 Probing physics beyond the standard model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
8.4.1 Gravitational shift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
8.4.2 Determination of fundamental constants . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
8.4.3 Drift of fundamental constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

9 Alternative techniques/additional topics 77

9.1 Encoding a qubit in the motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
9.2 Microwave entangling gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
9.3 Molecular ions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
9.4 Quantum simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
9.4.1 Quantum simulation with long ion strings . . . . . . . . . . . . . . . . . . . . . . . . . . 81
9.4.2 Simulators in two dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
CONTENTS 5

This script is built in many parts from the “Cavity QED and trapped ion physics” lecture script by J.P.
Home.
For this course the previous visit of the Quantum Optics lecture is required. The most relevant chapters
from that lecture’s script (402-0442-00L, HS2020, J.P. Home) are:
• Chapter 3: Atom-light interaction
• Chapter 4: Coupling of an Atom to many Field Modes

• Chapter 5: Density Matrix and Master Equation

• Chapter 10: Photon Momentum and Laser Cooling
• Chapter 11: Laser Cooling and Control of Bound Atoms

We will expand on chapters 10 and 11 in the context of trapped ions in this course.
6 CONTENTS
Chapter 1

Trap and motion

1.1 Trapping a charge in 1D

To trap a charged par tile in 1D we simply need a potential well - the simplest one is given by
1
φ1D (x) = U0 αx2 . (1.1)
2
Where we assume U0 to be a voltage on a far away electrode and α a constant taking into account the electrode
geometry. The potential energy of a trapped particle with electrical charge Ze (with e the elementary charge
and Z an integer) and mass m is then given by Zeφ. This is a harmonic potential and the solution is the
harmonic oscillator. Solving as usual we find an oscillatory motion

x(t) = A cos(ωt + φ) (1.2)

with amplitude A and phase φ of the motion given by the boundary conditions and the eigenfrequency, which
we will call trap frequency
r
ZeU0 α
ω= (1.3)
m
Or for some general potential
r
Ze d2 φ
ω= . (1.4)
m dx2

1.2 Trapping a charge in 3D

1.2.1 Laplace equation
For trapping a charged particle in 3D we might naively say we want an electric potential like this

φ3D (x, y, z) = U0 (αx2 + βy 2 + γz 2 ). (1.5)

But the issue is that we need to satisfy Laplace’s equation:

∂ 2 φ ∂ 2 φ d2 φ
0= + 2 + 2. (1.6)
∂x2 ∂y ∂z
For our example potential φ3D this means 0 = α + β + γ. To get this to zero we need at least in one direction
a negative second derivative (curvature), and that means anti-trapping. So for at least that one direction we
need to find a different mechanism. The two techniques are the Paul trap, which uses oscillating electric fields
and the Penning trap, which uses a combination of electric and magnetic fields. We look at both possibilities
in the next sections.

1.2.2 Paul trap and pseudopotential approximation

To understand the Paul trap, we first look at the basic behaviour of a particle in an oscillating electric field
(also rf field for radio frequency) in one dimension. This follows closely [22]. We assume we have an electric

7
8 CHAPTER 1. TRAP AND MOTION

field pointing along the x direction E(x, t) = E(x) cos(Ωrf t), where E(x) is a spatially dependent amplitude
and cos(Ωrf t) is the oscillation in time with frequency Ωrf . The equation of motion for our charged particle
of mass m, charge Ze, and velocity v is then given by
dv
m = ZeE(x) cos(Ωrf t). (1.7)
dt
We integrate this and find

ZeE(x)
v= sin(Ωrf t). (1.8)
mΩrf
The kinetic energy Ukin of the particle is then given by
 2
1 1 ZeE(x)
Ukin = mv 2 = sin2 (Ωrf t). (1.9)
2 2m Ωrf

And the time average over one period of the field’s oscillation is
 2
1 ZeE(x̄)
hUkin i = . (1.10)
4m Ωrf

Because we average over the oscillation we have replaced the spacial coordinate with also the average position
x̄. We deliberately left the possibility of the electric amplitude to have a spatial dependence. If the particle not
moves to an area of higher (lower) electric field amplitude it will gain (loose) kinetic energy. The ion will move
naturally to a position were it has a lower averaged kinetic energy. This might start to confuse you, because
it will also gain kinetic energy by moving towards this point of lower averaged kinetic energy. We should now
discriminate between a fast oscillatory motion (the one we averaged over) which we call micromotion and the
slow motion that is caused by a spatially varying electric field amplitude E(x) secular motion.
We can use this to create an effective confining potential by tailoring the electric field amplitude E(x) to
have a gradient (dE(x)/dx 6= 0) for x smaller and larger some position x0 and no gradient (dE(x)/dx = 0)
at x0 . This will result in a restoring force that drives the charged particle towards x0 . This force is also also
called ponderomotive force and the motion it’s causing is the previously mentioned secular motion.
We can now formulate an averaged/effective potential which is called pseudopotential. It’s usually given
to have the units of a regular electric potential but to distinguish it from real (non-averaged) potential we
give it the letter ψ:
 2
hUkin i |Z|e E(x̄)
ψ(x̄) = = . (1.11)
|Z|e 4m Ωrf

Note that it depends on the averaged position x̄, not x. This averaged position now only describes the secular
motion. The micromotion is averaged out and the pseudopotential looks like a static electric potential. But
don’t be fooled, once we look at the details we’ll see that it behaves quite different. This is quite a weird
potential. Contrary to a normal electrical potential this pseudopotential depends on the mass and charge of
the particle we’re looking at, and it depends on the electric field squared. This will all be important for the
details of trapped ion experiments. Also note that I write the absolute value of the charge number Z - the
pseudopotential is confining no matter if the particle is positively or negatively charged. We can of course now
also give a trap frequency - the secular trap frequency ωsec - that results from this confining pseudopotential
by simply treating it as a regular potential, so plugging it into Eq. 1.4. We get
r r
|Z|e d2 ψ |Z|e d2 E(x̄)2
ωsec = 2
= . (1.12)
m dx̄ 2mΩrf dx̄2
He we should pause and ask ourselves about the secular and the drive frequency and the approximation
we’ve made. Since we averaged over one cycle of the rf drive to get to the pseudopotential, it’s clear that
the secular motion needs to be much slower for the approximation to hold. In practice that means that the
pseudopotential approximation is only valid when ωsec  Ωrf . This is a very useful criterion but we’ll find a
mathematically better defined criterion in section 1.3.1.
To evaluate this further we need to know how E(x̄) looks like. For this we assume an oscillating potential
given by
1
φ(x, t) = Urf κ cos(Ωrf t)x2 (1.13)
2
1.2. TRAPPING A CHARGE IN 3D 9

Pseudopotential Micromotion

Figure 1.1: Pseudopotential cartoon

Where Urf the rf voltage amplitude on some (far away) electrodes and κ a constant taking into account the
electrode geometry. This results in the osculating electric field

E(x, t) = Urf κ cos(Ωrf t)x (1.14)

meaning that the E(x̄) we’re looking for is given by E(x̄) = Urf κx̄ and the secular frequency is
r
|Z|e d2 ψ |Z|eUrf κ
ωsec = = √ . (1.15)
m dx̄2 2mΩrf
Note that we’re have linear proportionality to the physical values here, where for a static potential we have a
square root! Keep in mind that there’s no real potential. A particles energy oscillates between kinetic energy
and potential energy while it oscillates in the potential well. Here, the conversion of energies is both times
kinetic energy, but it’s stored at different frequencies: the secular motion and the micromotion. Fig. 1.1 gives
a pictorial representation of this.
Let’s put some numbers to this to compare static and rf trapping using the potentials from Eq.’s 1.1
and 1.13. We’ll want a trap frequency of ω = 2π × 1 MHz for a 40 Ca+ ion (m = 40 u, Z = 1). We
choose Ωrf = 10 ωsec = 2π × 10 MHz to satisfy the pseudopotential approximation. We further assume the
electrode geometry gives the same contribution in both cases, so α = κ. It’s value is usually in the order
of 1/d2 with d the distance of the electrode to the ion. We assume a for us typical value of d = 200 µm so
α = κ = 2.5 × 107 µm. For these numbers we would have U0 = 0.65 V and Urf = 9.26 V. So already we see
we need higher voltage to get the same confinement from the pseudopotential as from the static confinement.
We might be inclined to use the linear scaling of the voltage of the pseudopotential vs. the square root scaling
for the static potential, so say, go to ωsec = 10 MHz which would need only a factor of 10 in Urf but a factor
100 in U0 . But thinking like this we’ve forgotten about satisfying the pseudopotential, so we also have to
scale Ωrf by 10, and that then means we need an additional factor of 10 on Urf to compensate and then have
gained nothing.
If we tailor an oscillating electric field in all three spatial directions we can create a confining potential in
3D. One possibility is using an electric potential of the form
 
1 1 2 1 2 2
φ(x, y, z, t) = Urf κ cos(Ωrf t) − x − y + z . (1.16)
2 2 2

This describes an oscillating quadrupole potential with Urf and κ defined as before. Note that it satisfies
Laplace’s equation but also provides the required form of the electric field for the pseudopotential. The
pseudopotential formed by this electric potential will confined a charged particle at x, y, z = 0 and leads to
oscillating motion around that point. So we get a 3D harmonic oscillator for the secular motion. This form
of trap we call a point trap. It provides an easy way to trap and the first traps but also some of today’s traps
were/are built like this. A fancy example is the fiber trap that is shown in Fig. 1.2 from the Sussex ion trap
group. For many applications that require to trap and control multiple ions there’s a (often) better choice:
The linear Paul trap which we will cover in depth later.
10 CHAPTER 1. TRAP AND MOTION

(c) Pseudopotential
(b) Drawing
(a) Photograph

Figure 1.2: Fiber cavity trap of the Sussex Ion Trap Cavity-QED and Molecular Physics Group as an example
of a point trap. Taken from [48] and [Link]

1.2.3 Penning trap

Before we get into the linear Paul trap and it’s nitty-gritty details that you’ll need to do and understand
experiments I want to explain the basics of the Penning trap which will we will not use much in the remainder
of the course, but it’s still important for you to understand its basic working principle. For this we return
to the quadrupole potential that we used for the Paul point trap in the previous section, just this time as a
static potential:
 
1 2 1 2 2
φ(x, y, z) = U0 κ − x − y + z . (1.17)
2 2
This will provide a confining potential well in the z-direction and anti-confinement in the x, y directions.
That means that the particle, starting close to x, y = 0, will be pushed outwards. If one now adds a magnetic
field B pointing along the z-direction it will cause the particles trajectory to curve in the x, y plane. If the
magnetic field is strong enough this will lead to two circular motions. The magnetron motion is slow and
follows a circular orbit around x, y = 0. The cyclotron motion is faster and makes smaller orbits around the
mean orbit of the magnetron motion. The 3D motion is sketched in Fig. 1.3. The three motional frequencies
(ωz axial, ωc cyclotron, ωm magnetron) are given by

r
2ZeU0 κ ZeB ωz2
ωz = , ωc = , ωm ≈ , (1.18)
m mc 2ωc
were c is the speed of light. For more details see [13].

1.3 Linear Paul trap

For a linear Paul trap confinement by the static electric potential is provided along z direction (also: axial
direction). This causes anti-confinement in the radial direction (x, y). This is counteracted by a strong rf
confinement in the radial direction. The exact shape of the potential will depend on the geometry of the
electrodes and is usually found by simulation. However, since our ions are usually well localized at the trap
center we can Taylor-expand the potential to second order, which gives a reasonable model. The dc1 and rf
potentials close to the center of a general trap are well described by the following expression

1
1
Udc αdc x2 + βdc y 2 + γdc z 2 + Urf cos(Ωrf t) αrf x2 + βrf y 2 + γrf z 2 .


φ(x, y, z, t) = (1.19)
2 2
Here Udc/rf are voltages on the trap electrodes and the parameters αdc/rf , βdc/rf , γdc/rf are factors taking
into account the geometry of the trap and have the dimension m−2 . For a linear Paul trap we have κdc =
−αdc /2 = −βdc /2 = γdc and κrf = αrf /2 = −βrf /2, γrf = 0. A typical geometry to produce such potentials is
the four-rod trap. Examples are shown in Fig. 1.4 and 1.5.
1 It is common to call the static confinement ‘dc’ - for direct current - which is nonsensical because no current is following to

provide static potentials - but nonetheless common nomenclature - maybe it originated from the fact that you need a dc power
1.3. LINEAR PAUL TRAP 11

(b) Potential

(a) Trap (c) Motion

Figure 1.3: Penning trap, electric potential and motion of a charged particle in a Penning trap. Taken from
[26].

(b) (c)

(a)

Figure 1.4: Old-school four-rod trap. (a) Taken from [91]. (b), (c): [Link]
Quadrupole_ion_trap.
12 CHAPTER 1. TRAP AND MOTION

(a)

(b)

(c)

Figure 1.5: Modern versions of the four-rod trap. (a) From the Oxford Trap Quantum Computing group:
“ ‘Blade’ ion trap in ultra-high vacuum system. If you look closely, you can see a single strontium ion in the
centre of the trap, between the tips of the two ‘needle’ electrodes. The tips of the needles are separated by
2.3mm. (Note that this is an actual photo of ion and trap, not a montage of separate photos; a long camera
exposure time, 30 seconds, was used to capture the faint fluorescence from the single ion.) [Photo credit -
David Nadlinger]”, (b) picture and (c) drawing of our monolithic ‘4-rod-trap’, which we want to use to do
high-precision quantum logic spectroscopy of H+2 . Dimensions are in mm.
1.3. LINEAR PAUL TRAP 13

1.3.1 The Mathieu equation

In section 1.2.2 we made the approximation that over the whole cycle of oscillation that the ion sits in the
same place. The real rigorous treatment is to write down equations for the force. In the following we will
show a more accurate description and will use the general potential of Eq. 1.19 but will point out how the
derived expressions behave for a linear Paul trap.
For u = {x, y, z}, and δdc/rf = {αdc/rf , βdc/rf , γdc/rf } respectively, the equation of force is

d2 u Ze ∂ψ Zeu
=− = (Udc δdc + Urf δrf cos(Ωrf t)) , (1.20)
dt2 m ∂u m
We define ζ = Ωrf t/2 and the so-called a- and q-parameters
4ZeUdc δdc
au = (1.21)
mΩ2rf
2ZeUrf δrf
qu = . (1.22)
mΩ2rf
Watch out, these depend on our definition of the potential and you’ll see other versions with varying factors
of two and signs. The definitions have to be such that we can rewrite the equation of motion to match the
form of the ‘standard’ Mathieu equation given by
d2 u
= (au − 2qu cos(2ζ))u. (1.23)
dζ 2
It’s known how to solve this: We use the Ansatz of a sum of terms oscillating with ζ
X X
x(ζ) = Aeiβx ζ C2n ei2nζ + e−iβx ζ C−2n e−i2nζ (1.24)
n n

One of the things about the solution to the Mathieu equations is we don’t constrain βx to be real. If βx is
imaginary we get exponential solutions, which are unstable. Meaning, that if you setup your trap parameters
wrong, the trap is unstable and you lose your ion. The equation is traditionally parameterized with a- and
q-parameters. The a-parameter is connected to the static potential, and the q-parameter to the RF potential.
For the linear Paul trap a and q will look the same (up to a sign) but for z, qz = 0 since we don’t
have an oscillating potential in the z direction. Linear Paul traps are usually operated in a regime where
2
|ax,y,z |, qx,y  1. This allows us to make a first-order approximation and we get the following solution for the
movement of the ion along axis u
 qu 
u(t) ' A cos (ωu,sec t) 1 − cos (Ωrf t) . (1.25)
2
We see that the motion is made up of two oscillating terms: one oscillating at the secular frequency ωsec given
by
r
Ωrf q 2 Ωrf
ωu,sec = βu = au + u . (1.26)
2 2 2
The second term (which for us has a much smaller amplitude than the first one) oscillates with
cos (ωu,sec t) cos (Ωrf t) = cos ((Ωrf + ωu,sec )t) + cos ((Ωrf − ωu,sec )t), (1.27)
so the micromotion frequency plus and minus the secular frequency. Since qz = 0, we get the expected trap
frequency of a static potential in the z direction which we already encountered in section 1.1.
r
√ Ωrf ZeUdc γdc
ωz,sec = az = . (1.28)
2 m
Looking at the radial direction and neglecting the static potential influence (ax,y ≈ 0) we get
Ωrf ZeUrf δrf
ωx,y,sec = qx,y √ = √ . (1.29)
2 2 2mΩrf
Which results in the same expression that we got for the pseudopotential approximation. It is common to
drop the label ‘sec’ on the trap frequencies and only distinguish the trap drive frequency by a subscript. So
for the rest of the script I’ll mean the secular frequency if I write e.g. ωx but will keep writing Ωrf for the
trap drive frequency.
supply to create the static voltage? I don’t know.
14 CHAPTER 1. TRAP AND MOTION

1.0

0.5
Position

0.0 Ωrf =5.7 ω

Ωrf =20.2 ω
-0.5

-1.0

0.0 0.5 1.0 1.5 2.0 2.5

Time

Figure 1.6: Ion trajectory for different Ωrf but same secular frequency ω based on Eq. 1.25.

Micromotion and secular motion

By evaluating Eq. 1.25 we can look at the trajectory of the ion including the micromotion. Fig. 1.6 shows
this.

Stability

There are certain regimes in which the trap is stable and unstable, so β being real or imaginary, respectively,
but also when we start to violate the assumptions we made for a and q. Fundamentally, one can show that
the Paul trap is stable for β ∈ {0, 1}. This seems to contradict my earlier statement of β required to be
real for a stable solution. If we’d do the full math we’d find that these are equivalent. We can draw a
stability diagram. In Fig. 1.7 an example is shown. One part of the trap stability we’ve already encountered
in the pseudopotential approximation, which we formulated as ωsec  Ωrf . We basically made the same
2
approximation when we limited us to qx,y  1 for the first order approximation of the solution to Mathieu’s
equation and this is the same stability boundary as β → 1. There the charged particle gets driven so strong
that the motion is amplified and thus the trap becomes unstable. The other boundary, β → 0, arises when
the pseudopotential is not strong enough to overcome an anti-confinement from the static potentials. Then
again the particle is not trapped.

Recap

With equation 1.29 we can formulate what we need for stable trapping with high trap frequencies:

• Low rf frequency to get high secular frequency, but also high rf frequency to get stable trapping. Here
we have to find a compromise.

• Low ion mass for high trap frequencies. That’s why D. J. Wineland did so many experiments using
9
Be+ , which is the lightest atomic ion for which laser cooling was demonstrated.

• High rf voltage for high trap frequencies. Why would we want to have high trap frequencies? (It helps
with most things. Technical noise sources often have a 1/f frequency dependence in their noise spectrum
and Doppler cooling will get us to lower n̄ if the trap frequencies are higher.

• What we have only briefly touched on: small trapping structures will increase δrf , so that also helps
too.
1.3. LINEAR PAUL TRAP 15

Figure 1.7: Stability diagram for a trap with αdc + βdc = −γdc and αrf = −βrf , γrf = 0, from [58].

1.3.2 Typical imperfections

Stray fields and micromotion
Micromotion is essential for trapping in a Paul trap but it’s also the bane of any ion trapper. In an ideal
trap micromotion (and it’s related motional energy) is optimally used for trapping with the pseudopotential,
i.e. the micromotion energy is identical to the energy of the secular motion (disregarding any effects of static
trapping fields). However, there are cases in which we have additional micromotion. They appear if the
oscillating electric field does not go to zero at the potential well minimum of the trap.
The basic effect of micromotion is just additional motion driven by the trap drive frequency Ωrf . This
driven motion leads to a lot of effects (typically harmful) that we’ll encounter in the rest of the course, so
we’ll keep coming back to this.
How does this non-zero electric field come about? One possibility is just by the geometry of the trap. If
we deviate from the standard 4-rod trap (or do a bad job in building one) these imperfections can cause an
oscillating electric field that does not go to zero and the desired location - this includes the axial direction
where we would have no rf field at all if we’d live in a perfect world (this is then called ‘axial micromotion’).
This even happens in a perfectly built Paul trap of finite length. For our linear Paul trap we’d assumed that
the rf potential in the z direction is zero. This is an approximation which can only be reached in the limit of
an infinitely long trap. A finite size leads to a weak axial field gradient (and thus pseudopotential); in that
case only in the very center in all three directions the electric field will vanish. Luckily this is a weak effect
in most traps.
Another cause of a non-vanishing oscillating electric field is a phase difference of the rf potentials on the
rf electrodes. This means we always have to make sure the electrical length of any individual connections to
the rf electrodes needs to be the same.
The most common cause of micromotion, however, are static electric fields that push the ion put of the rf
field null. This micromotion can be nulled by compensating any stray electric fields. This is done by either
using electrodes far away from the ions - those mostly create homogeneous fields and/or applying differential
voltages to the dc electrodes. Important is that the apparatus you build has the possibilities to cancel an
arbitrary stray field in the radial direction, so two electrodes or two sets of electrodes which produce fields in
different radial directions are required. An example is given in Fig. 1.8. The procedure of compensating stray
electric fields is often called (somewhat imprecise) ‘micromotion compensation’. These static electric stray
fields originate from various sources. Again imperfect trap geometry can be a cause but mostly it’s caused
by dielectric surfaces which are statically charged (often by the UV lasers that we use to control the ions).
In the literature you’ll sometimes read about ‘excess’ and ‘intrinsic’ micromotion. However, there is no
agreement which of the above types counts to which category, so be careful.
Almost all you need to know about micromotion (and a rigorous mathematical treatment) is in [8].
16 CHAPTER 1. TRAP AND MOTION

Figure 1.8: Micromotion compensation in our linear trap. The red arrow indicates the direction of the stray
field, the dark blue arrow indicates the direction of a field created by a differential voltage on the dc electrodes
and the light blue and green arrows indicate fields created by the further away shim electrodes in the matching
colors. A linear combination of any two will be enough to for compensation.

1.3.3 Surface traps

There is another important version of the linear Paul trap. It’s a planar trap where all electrodes are on
one surface. You can think of this as ‘unfolding’ the 4-rod trap onto a plane - as shown inf Fig. 1.9. Above
the so-called ‘5-wire’ geometry you still get a quadrupole then applying a voltage to the second and fourth
electrodes. This you can use for creating a pseudopotential for radial trapping by applying a rf potential to
those electrodes. The axial confinement is then provided by cutting the outer electrodes in segments (just as
in the 4-rod trap). Surface traps have several advantages over 3D geometries that make them quite attractive.
Their planar structure makes them very easy to fabricate with high precision - this can be done using standard
lithographic techniques. It also means that it’s much easier to interface with existing chip-technology. This
can greatly benefit scaling trapped ion quantum computers and we’ll discuss this in detail later.
Of course there are also downsides to the planar technology. The ion is not surrounded by electrodes
anymore which lead to weaker and shallower confinement (assuming similar trap sizes and applied voltages),
and this in turn necessitates smaller trap structures and thus a smaller ion-electrode distance (which we’ll see
in section 1.4.2 causes trouble). Being closer to the electrode structures does again have some advantages.
The field gradients one can create are much higher which has allowed, for instance, to perform quantum
computation using magnetic field gradients (which we’ll discuss later in the course).

1.4 Quantum harmonic oscillator

We want to do quantum physics with our ions and their motion in the trap, so we have to look at the motion
in the quantum regime. In the beginning this is fairly easy: Since the trap potentials are very harmonic
(i.e. the dominating terms are q 2 with q ∈ {x, y, z}) the motion of a single ion is simply described by a
three-dimensional quantum harmonic oscillator (QHO), so three independent oscillators with eigenfrequencies
matching the classical trap frequencies. Their description follows fully the textbook QHO that you know from
your QM lecture. For each you’ll have position and momentum operators q̂i , p̂i , annihilation and creation
operators âi , â†i with i ∈ {1, 2, 3} etc.
This is a good point to go to a textbook and remind yourself about the energy eigenstates of the QHO: Fock
states |n = 0, 1, 2...i, but also about thermal, coherent and squeezed states that you should have encountered
in quantum optics.
I did not specify in what direction these three oscillators move, i.e. what the axes of their fundamental
modes of motion are. Using only quadratic potentials you’ll find that there are three orthogonal, independent
oscillating, axes (so no coupling of the different motional modes). This is nothing specific to quantum physics,
of course, and details on this can be found in classical mechanics textbooks. For a linear Paul trap one mode
1.4. QUANTUM HARMONIC OSCILLATOR 17

Figure 1.9: Surface trap schematic. Left, top is a cut through our e-rod trap in the radial plane. The plus
sign marks the center of the qudrupole. The black arrows show how this structure can be unfolded into a
planar geometry. On the right side the field lines above the planar structure are drawn; again the plus sign
marks the center of the quadrupole.

will point along the axial direction, the other two will be in the radial plane, but their orientation will depend
on the trap geometry and the static voltages applied. For a four-rod trap they usually point along the diagonal
lines connecting the rf rods and the ground/dc rods. For the further parts of this course this just means that
the modes of motion are not x, y, z but some x0 , y 0 , z where x0 , y 0 are rotated by some angle with respect to
x, y.

1.4.1 Decoherence
There are two decoherence mechanisms: One is dephasing the other relaxation. In other fields the respective
(de-)coherence times are often called T1 and T2 times but are not commonly used in trapped ion literature.
For the QHO dephasing is a fluctuation of the trap frequency in time; relaxation is, in trapped ions, heating
of the motional state to a thermal state, or in other words, relaxation to the temperature of the electric
fields trapping the ion. Mathematically we describe these as jump operators (Lindblad operators) in a Master
equation in Lindblad. For heating we get the operators
p √
L̂1 = Γ(n̄ + 1) â, L̂2 = Γn̄ ↠, (1.30)

where Γ is the coupling rate (not equivalent but connected to the heating rate n̄˙ derived in the following
section) and n̄ is the average thermal occupation of the reservoir determined by its temperature. This might
look like a process that should lead to net-cooling, since the rate of heating is slightly smaller than the rate
of cooling. But don’t forget that the QHO is bounded at lower energy, so in the end this leads to heating.
For dephasing we have the Lindblad operator
√
L̂ = Γ â† â (1.31)

Nice paper on all this: [86]. What are the sources of these noise processes? Dephasing of the motion of single
ions is mostly caused by fluctuating/drifting voltages on the trap electrodes or similar processes that change
the confinement (and thus the trap frequencies) on slow timescales. For multiple ions and additional source
is discussed later. The sources of heating are discussed in detail in the next section.

1.4.2 Ion heating

When working with ions we want them cold, typically close to the quantum ground state |n = 0i. One issue
that comes up in ion traps is heating of that motion through noisy electrical fields. The noise of our electric
potentials will have a frequency component at the frequencies of the motional modes and will thus create an
oscillating potential at that frequency. This is driving the motion, causing excitation. We usually express
the strength of this heating in number of motional quanta gained (on a certain mode of motion) per unit
18 CHAPTER 1. TRAP AND MOTION

time: ṅ. In general this heating rate should be much lower than any relevant interaction or coherence time
that we require from our system, otherwise experiments in the quantum regime are hindered by this form of
decoherence.
We’ll now look at how to describe how the ion is reacting to a noisy electric field. This follows [85] and
references therein. A Hamiltonian picture for the effect of an electric field on the ion’s motion is

Ĥ = −qE(t)x̂ (1.32)

where E(t) is a noisy electric field as a function of time and q is the charge of the ion. Using first-order
perturbation theory we look at the time derivative of the matrix element c01 (this is just like Fermi’s golden
rule)
dc01
= −iqE(t)/h̄ h1| x̂ |0i e−iωm t (1.33)
dt
with ω the trap frequencypof the mode we’re looking at, so the energy difference of |0i and |1i. For the QHO
we know that h1| x̂ |0i = h̄/(2mωm ). Integrating in time we find
r Z t
q h̄ 0
c01 (t) = − E(t0 )eiωm t dt0 (1.34)
h̄ 2mω 0
We could now look at the amplitude of this to get the probability of exciting the ion from the ground state |0i
to the first exited state |1i, but first we need to remember that the electric field is noisy and of course in the
end we’re not interested in a single instance in time but an averaged property. For this we have to average
of the electric field and also watch out to do the integration correctly. If you’ve dealt with noise before this
might look familiar:
Z ∞
2 q2
|c01 | t = hE(t)E(t + τ )it eiωm τ dτ (1.35)
2mh̄ωm −∞
The integral is basically the Fourier transform of the correlation function for the electric field. The spectral
density of the electric field (fluctuations) for a frequency ω is defined as
Z ∞
SE (ω) = 2 hE(t)E(t + τ )it eiωτ dτ (1.36)
−∞

So in the end we find

d|c01 (t)|2 q2
n̄˙ = = SE (ωm ) (1.37)
dt 4mh̄ωm
˙ You see that the heating rate goes down with higher mass and trap
This is what we call the heating rate n̄.
frequency (not accounting for any frequency dependence of SE , but see below). We can connect this to the
˙
Lindblad description used in the previous section for heating. Then Γ = n̄/n̄ with Γ the coupling rate and
n̄ the temperature of the reservoir as defined in the previous section and n̄˙ the heating rate defined above.
What is the temperature/n̄ of our reservoir? I don’t know. It usually does not matter since n̄  1 and
we’re mostly doing experiments close to the ground state. I’m not aware of experiments trying to determine
n̄.
Where does this heating come from? One obvious issue are the voltage sources that we use to apply our
trapping potentials, but those we can get good enough (and filter them) that they do not matter anymore.
So back in the days when researcher measured the heating rates they expected the fundamental source to be
Johnson–Nyquist noise (often just ‘Johnson noise’). This is basically the noise caused by thermal electrons
in the conductors (our golden trap electrodes) scattering with atoms or other electrons. However, the noise
levels and found in experiments is typically four orders of magnitude higher than what is to be expected from
Johnson noise. This heating was thus named ‘anomalous’ and a lot of research has been performed to find
it’s source. What we know is:
• The distance with the electrode-ion distance d of the noise spectral density is SE ≈ 1/d3.5 (for Johnson
noise we would expect 1/d2 )
• The frequency dependence is SE ∝ 1/ω (for Johnson–Nyquist noise we’d expect no dependence of SE
on ω).
• Lower trap temperature results in lower heating rates [23]. If ion traps are cooled to 10 K the heating
rate drops by approx two order of magnitude.
1.5. MULTIPLE IONS 19

(a)
(b)

Figure 1.10: Heating rates in dependence of (a) ion-electrode distance (from [23]) and (b) trap temperature
(from [18]).

• If the trap surfaces are ‘cleaned’ in situ (without opening the vacuum chamber after cleaning) by Argo-
ion bombardment the heating rate drops by approx two order of magnitude and surface contaminants
which seem to be mostly hydrocarbons are removed [36].

This is a still unsolved puzzle. The current understanding is that contaminants on the trap electrode surfaces
move around and through this cause electric field noise. They can be cleaned of but redeposit if a trap is taken
to atmosphere. The strong distance scaling hurts surface traps a lot and they are often used in cryogenic
setups to push the heating rate down to acceptable levels. A review on this topic is given in [14]. Fig. 1.10
shows heating rate dependencies for temperature and electrode-ion distance.

1.5 Multiple ions

We can put multiple ions into a single confining potential. They will repel each other due to the Coulomb
force given by

1 q1 q2
F = (1.38)
4π0 r2

with qi the charge and r the ions’ distance. Two ions will align each other such that they sit along the
direction of weakest confinement. Here the linear Paul trap shows it’s use: By making the axial confinement
weaker than the radial the ions line up along the trap axis. This can be done for more than two ions, but for
them to form a linear chain (instead of some bunched up ellipsoid) the axial trap frequency has to be much
lower than the radials. Fig. 1.11 shows some examples. If the ions are put in a tight enough trap they couple
strongly through their Coulomb interaction. Strong coupling here means that the energy exchange frequency
of the ions is (roughly) in the same order of magnitude as the trap frequencies for a single ion in the same
potentials. This means that their motion is not well described anymore by the trap frequencies that a single
ions would have in the same potentials. One switches to a normal mode picture which describes the motion
as collective modes of the ion chain/crystal instead of individual modes per ion. Normal mode description
has the advantage that (at least in lowest order) the normal modes don’t exchange energy. An ion crystal of
N ions will have 3N normal modes which can again be treated as uncoupled harmonic oscillators. For two
ions the normal modes are shown and named in Fig. 1.12. This is quite an extensive topic of it’s own so we
don’t go into much more detail here. However, one important thing to point out, is that the coupling of the
ions in a trap goes down if the masses of the ions is too different. For me a rule of thumb is a factor of four
difference still works, but of course this depends on details and the application. So if we want to put two
different elements in a trap (and as you will see there are good reasons to do this), resulting in what we call
mixed species ion crystals, they better have similar mass, otherwise they don’t couple well. We come back to
what we can do with this later. If you’re interested in how the motion of these mixed species crystals behave
and how to calculate them you can have a look in Jonathan’s review paper on that [37].
20 CHAPTER 1. TRAP AND MOTION

Figure 1.11: Ion crystals. Here the number of ions is successively increased, keeping the trap potentials
constant. Above a certain number of ions the linear chain collapses into a Coulomb crystal in the form of a
zeppelin. From [35].

Figure 1.12: Normal modes for two and three-ion chains. Left axial for two ions, center radial (one plane)
for two ions, right axial for three ions. It’s no mistake that there’s no arrow on the second ion in the stretch
mode for three ions: if the masses of all ions are identical the center ion does not participate in that mode.
1.6. DYNAMIC OPERATIONS 21

1.5.1 Decoherence in multi-ion crystals

Heating of ions is different in multi-ion crystals. For the COM modes of a single-species chain we get an
amplification of the heating by the number of ions N because now N charges couple to the noisy electric field
and all ions couple maximally to the COM modes. This is obviously bad one we come to scaling trapped ion
quantum information...
The stretch or rather out-of-phase mode of a same-species ion chain however will exhibit only minimal
heating. Because every ion is moving with the same amplitude but opposite direction to it’s nearest neighbours
the collective motion cannot be exited by a fluctuation electric field but requires an electric field gradient. This
leads to very low heating rates of out-of-phase modes. This suggests that we should always work with stretch
modes but there is another issue: The Kerr effect. The ion-ion distance of a two-ion pair strongly determines
the stretch mode frequency. This distance will increase if the ions move radially in opposite directions. This
makes the axial stretch mode frequency sensitive to the radial rocking modes’ amplitudes. If the rocking
modes are in a thermal state this will cause a broadening of the stretch mode frequency and thus an increased
decoherence for motion of that mode.
For multi-species ion chains the situation gets more complicated and the details have to be determined by
calculation [37].

1.6 Dynamic operations

We want to now investigate what we can do using the trap to manipulate our ions’ (quantum) states.

1.6.1 Motional excitation, coherent states, squeezed states

If we change the voltages on the trap fast compared to the oscillation frequencies of the ion we will change its
motional state. We can distinguish two (extreme) regimes: One where we make an instantaneous change in
the potential (much faster than one oscillation period). The way to find what the ion’s state is doing would
be just to take the original state (maybe expressed by a sum of eigenstates of the initial potential) and put it
in the new potential and let it evolve. If let’s say we simply add a homogeneous electric field E to a quadratic
potential well αx2 this would simply shift the potential well minimum by a certain amount:
φ(x) = αx2 − Ex (1.39)
Turning on the electric field shifts the potential well minimum from x = 0 to xshift = E/(2α). What does
that mean for the state of the ion? It gets displaced. For this we can use the known displacement operator
D̂(β), where the displacement amplitude is given by
xshift
β= (1.40)
2x0
p
with x0 the ground state wave packet extent x0 = h̄/(2mω). This method we actually used some years ago
to make huge coherent states [2].
Another option to displace the ion is to use a field oscillating at the motional frequency of the ion. This
is then a driven/forced harmonic oscillator and the displacement will depend on the field amplitude and how
long it was turned on. The Hamiltonian for this in the interaction picture with respect to the ion’s motion is
given by
g
Ĥ = ih̄ (âe−iδt + ↠eiδt ), (1.41)
2
where g is the coupling strength and δ is the detuning of the oscillating electric field from the ion’s oscillation
frequency. This we sometimes call ’tickling’ the ion and can be used to determine the trap frequencies: If the
displacement gets large enough the speed of the ion is so large that it’s scattering rate is reduced due to the
Doppler shift and we see that the ion fluorescence counts drop as a response.
We can further use this method of modulating or switching the trap potential to create squeezed states
of the motion. This can be done by modulating the potential strength at twice the frequency of the ion’s
oscillation This is what we call a ‘parametric’ drive. This has been done recently using a surface trap
(making use of the strong field gradients that can be produced) [15]. In the supplemental materials (https:
//[Link]/content/suppl/2019/06/19/364.6446.1163.DC1) is a nice derivation of the
basic physics! The Hamiltonian for this drive is (here without detuning)
g
Ĥ = ih̄ (â2 + â†2 ). (1.42)
2
Here g is the parametric coupling strength. The square of the creation/destruction operators should signal
to you squeezing!
22 CHAPTER 1. TRAP AND MOTION

(a)
(b) (c)

Figure 1.13: Segmented traps: (a,b) 3D ‘wafer-style’ trap, (b) segmented surface trap. Both TIQI group.

Figure 1.14: The position of a harmonic well positioned at the origin (blue) can be displaced (orange) by the
addition of a uniform electric field (green).

1.6.2 Trap arrays and transport

We can modify our 4-rod trap to have more than one segment, giving us an array of traps along the axial
direction. Fig. 1.13 shows such segments for a 3D and a surface trap. Why would we do this? One useful
application is that we can start to move ions around in or trap, from one segment to another. This can be
used to take some ions out of the laser beam and put other into the laser beam, isolate the loading zone
(which might be compromised in it’s performance by the loading process) from a zone in which we want to
do high-precision or high-fidelity experiments. The probably biggest goal however is to enable scalable trap
structures for quantum information where we want to move ion-qubits around in a quantum processor to
couple them to other ion-qubits. We’ll get back to how this works in detail later in the course. For now we
want to look at transport operations form a general point of view.
To Transport ions we apply (slowly) varying voltages to the dc electrodes to shift the axial potential (ideally
keeping the radial potential the same) along the axis. If we do this adiabatically (the ion displacements needs
to be slow compared to the ions oscillation in the trap) the ion will not get exited and we can shift the ion
around without changing it’s motional state.
Let’s look at this in more detail. The Hamiltonian due to a harmonic trapping well is simply that of
the quantum harmonic oscillator. Upon adding a uniform electric field E(t) along the trap axis z, with time
dependent amplitude, the full Hamiltonian is
p̂2 1
Ĥ = + mω 2 x̂2 + ZeE(t)x̂ (1.43)
2m 2
which as Figure 1.14 illustrates simply corresponds to a harmonic well displaced by some distance xshift (t) =
ZeE(t)/(mω 2 ), so it can be rewritten as
p̂2 1 2
Ĥ(t) = + mω 2 (x̂ − xshift (t)) . (1.44)
2m 2
The additional term proportional to x2shift is only an energy offset and does not matter. In what follows, we
will solve Schrödinger’s equation for this Hamiltonian. To simplify matters, we first introduce states |χ(t)i
which travel along with the position xshift (t) of the minimum of the potential well, i.e.
 r   r 
mω mω
|χ(t)i = D̂ − xshift (t) |φ(t)i ↔ |φ(t)i = D̂† − xshift (t) |χ(t)i (1.45)
2h̄ 2h̄
1.6. DYNAMIC OPERATIONS 23

where |φ(t)i is the motional state of the ion in the stationary reference frame of the laboratory. Note that
from here onwards, we will omit the argument of the displacement operator D̂ for improved readability. Next,
we want to determine the Hamiltonian Ĥm governing the time evolution of the motional state in the reference
frame traveling with the potential well

Ĥm |χ(t)i = ih̄∂t |χ(t)i = ih̄∂t D̂ |φ(t)i (1.46)

 
= ih̄ ∂t D̂ |φ(t)i + D̂ih̄∂t |φ(t)i (1.47)

In order to simplify this expression, we need to evaluate

∂t D̂(β(t)) = ∂t exp β(t)↠− β ∗ (t)â = ∂t exp β(t)(↠− â)

(1.48)
†

†
†


= exp β(t)(â − â) â − â β̇ = D̂(β(t)) â − â β̇ (1.49)

where we first used the fact that the displacement β(t) = − mω

p
2h̄ xshift (t) is real and then used the chain rule
to evaluate the derivative. Inserting this into the previous expression, we get
r
mω
Ĥm |χ(t)i = ih̄(− ẋshift (t))(↠− â)D̂ |φ(t)i + D̂H(t) |φ(t)i (1.50)
2h̄
r
mω
= ih̄(− ẋshift (t))(↠− â) D̂D̂† |χ(t)i + D̂H(t)D̂† |χ(t)i (1.51)
2h̄
| {z } | {z } | {z }
−ẋshift (t)p̂ id Ĥ0 =h̄ω (↠â+ 21 )

From here, we go to the interaction frame picture of Ĥm with respect to Ĥ0 , resulting in
r
mω
ẋshift (t) ↠eiωt − âe−iωt .


Ĥm,I (t) = −ih̄ (1.52)
2h̄
This Hamiltonian is time-dependent, but it commutes with itself at all times, i.e.

[ĤI (t0 ), ĤI (t00 )] = 0 (1.53)

We can thus simply integrate Ĥm,I (t) over time to get the propagator Ûm,I (t):

i t
 Z 
0 0
Ûm,I (t) = exp − Ĥm,I (t )dt (1.54)
h̄ 0
mω t mω t
 r Z  r Z  
0 iωt0 0 † 0 −iωt0 0
= exp − ẋshift (t )e dt â + ẋshift (t )e dt â (1.55)
2h̄ 0 2h̄ 0
Defining
t
r Z
mω 0
αI (t) = − ẋshift (t)eiωt dt0 (1.56)
2h̄ 0

we recognize that the propagator corresponds to a displacement operator

Ûm,I (t) = exp αI (t)↠− αI∗ (t)â = D̂(αI (t)).

(1.57)

If we leave the interaction picture and go back to the reference frame moving together with the minimum of
the harmonic well, we get

Ûm (t) = D̂(α(t)) (1.58)

with
mω −iωt t
r Z
0
α(t) = − e ẋshift (t0 )eiωt dt0 . (1.59)
2h̄ 0

While arriving at this result required quite some work, it is easy to interpret: The effect of the transport
process on the motional state of the ion is simply a displacement. For transport without significant effect
on the motional state of the ion, we should transport slowly compared to the time it takes for one trap
oscillation. However, the equation also tells us that fast transport is also feasible - it just requires better
timing and control. Experimentally transport with low motional excitation has been demonstrated both for
slow (‘adiabatic’) and fast (‘diabatic’) transport [10, 89].
24 CHAPTER 1. TRAP AND MOTION

Figure 1.15: Illustration of the splitting process. At the top, you can see the five DC electrodes used for
splitting. a), b) and c) show the various steps during splitting (see main text).

Merging and splitting

The separation of a string of ions trapped in a single harmonic well into two sub-strings both stored in
individual harmonic wells is quite simple conceptually. Using DC electrodes that are segmented along the
trap axis, we can create a potential along the trap axis that looks like

U (x, t) = α(t)x2 + β(t)x4 , where β>0 (1.60)

Initially, we make the quadratic term controlled by α large such that it dominates over the quartic term
governed by β, essentially creating the single harmonic well. This corresponds to section a) in Fig. 1.15.
Over time, we then modify the voltages applied to the DC electrodes such that the quadratic term governed
by α goes from positive to negative. When α = 0, the ions are only loosely confined due to the quartic term,
corresponding to section b) of Fig. 1.15. Finally, when α < 0, we have two separate trapping wells. We can
then manipulate these individually. For merging of ion strings, the same procedure is used in reverse. As in
transport the speed of the operation matters for the final motional state of the ions and both adiabatic and
diabatic splitting and recombination are possible [10].
Chapter 2

Electronic structure

What do we need from our ions?

• Low mass for: high trap frequencies (helps with heating rates and the laser-ion interaction is stronger).
Watch out. This reasoning might have guided the early ion experiments from D. Wineland an co-workers
but gets softened today. Many groups now use 171 Yb+ which is quite heavy in comparison, but other
aspects make it useful (due to the mass the lifetime in the trap is much longer and the lasers are more
convenient than for 9 Be+ ).

• Simple electronic structure for Doppler cooling, fluorescence detection and state preparation.

• A good two-level system to encode a qubit.

Because 9 Be+ and 40 Ca+ ions are used in our group (and around the world) and are of they have the
typical features of trapped ions I’ll discuss everything using them as examples. Their level diagrams are shown
in figure 2.1.
Let’s start at the beginning. We’re working mostly with singly ionized ions, so I’ll limit the discussion
to that. The electronic structure of (atomic) ions is not fundamentally different than that of atoms. The
transitions frequencies are usually higher, so the wavelengths shift to the blue because the electrons are
stronger bound, so you’ll see a lot of UV lasers when working with trapped ions. In neutral atoms experiments
are most often done with alkali atoms. Why? Because they have only an single, unpaired electron in the
outermost shell (first group elements) and thus have a simple electronic structure. In (singly ionized) ions
it is most common to work with earth-alkali elements (second group). Since the ions miss one electron they
are isoelectronic to their neighbours in the first group, so they also have only one electron in the outermost
shell and again have a simple energy structure. The lowest energy state will have zero angular momentum
(L = 0, S1/2 orbital) and for light ions/atoms (9 Be+ , 24 Mg+ ) the first exited state will be L = 1 (P orbital).
For heavier ions (40 Ca+ and heavier) we get low-lying D orbitals (L = 2, and others the heavier we go) that
are lower than the P orbital in energy. This complicates some things but can also offer great benefits. If we
consider that this outermost electron still has it’s spin degree of freedom we will get two levels in the ground
state that will be split by the Zeeman effect in a magnetic field that we apply for that purpose. Through
spin-orbit coupling (finestructure) the P and D states will already be split into P1/2 and P3/2 and D3/2 and
D5/2 manifolds, which will again split in the magnetic field into 2J levels. If the ion’s nucleus also has a spin
(as is the case for 9 Be+ , its nuclear spin is I = 3/2) we’ll also have to take the arising hyperfine structure
into account. So for 9 Be+ we get 8 levels in the ground state (S). With 9 Be+ we don’t have a choice, there
are no other stable isotopes readily available that would have a lower I. For calcium the range is much larger
([Link] In our group we use 40 Ca+ (I = 0), but other
groups use often 43 Ca+ (I = 7/2). I’ll explain later why.

2.1 Zeeman effect

In almost all experiments we’ll turn on a quantization field, a B field that splits the degeneracy of the fine-
/hyperfine structure. You probably remember, the energy dependence on the magnetic field is

δE = µB gJ B, (2.1)

with the Bohr magneton µB , the gJ the g-Factor (or Landé-Factor) which takes into account the different
magnetic coupling of spin and orbital angular momentum. The coupling gives approx. 1 to 2 MHz/G energy

25
26 CHAPTER 2. ELECTRONIC STRUCTURE

(b)

(a)

Figure 2.1: Level diagrams for (a) 9 Be+ at B = 10 G and 40 Ca+ at 119 G (sorry, weird choice, still have to
get the numbers for low-field Ca). Frequency numbers give splitting of manifolds (in ω/(2π)).

shift. Often the magnetic field is chosen low, approx. 5 G, such that the levels split by more than what any
light shifts could cause from the lasers. This makes sure our states and their quantization axis (the direction
of the B field) are always well defined.

2.2 Transitions
2.2.1 Electric dipole transitions
Looking at the level diagram for 9 Be+ it is in principle simple. We have (electric) dipole transitions between
S and P (∆L = 1). In 40 Ca+ we also have these transitions, but additionally we’ll see that we have dipole
allowed transitions between P and D. These transitions typically have linewidth of approx. Γ ≈ 2π × 20 MHz,
so lifetimes of τ = 1/Γ ≈ 7 ns.
The first thing we want to do is to scatter photons, i.e. see our ions. For this we tune a laser in resonance
with a dipole transition. For 9 Be+ this is simple, we’ll usually take a laser that connects S1/2 to P3/2 . Because
we split our levels (and in the end we only want to scatter photon of certain states), we’ll choose a certain
polarization for our light. In 9 Be+ we choose σ+ polarization and tune the wavelength so that it connects the
stretch states, S1/2 , F = 2, mF = 2 and P3/2 , F = 3, mF = 3 , so states where all the angular momentum
is aligned with the quantization axis (so S1/2 , F = 2, mF = −2 is the other stretch state for the electronics
ground state of 9 Be+ ). I’ve written here F = 4, mF = 4 for states in P3/2 . In principle, these are not the
good quantum numbers, because the hyperfine interaction for the P3/2 manifold is much weaker. However,
the stretch states have a nice property: They don’t mix with other states. All the different components of the
total angular momentum point into one direction, no matter what the B field and their coupling among each
other is. I’ve chosen the F, mF notation so that it’s clear that we couple states with ∆mF = 1, so requiring
σ+ polarization of the beam. We should also not that from the P3/2 , F = 3, mF = 3 state there is no other
state in the S1/2 manifold to decay into, all other transitions would have ∆mF < 1. That means that the
population will constantly cycle between these two states but never go somewhere else. This is what we cal a
cycling transition. This is extremely nice because we can rapidly scatter photons, nearly at the decay rate of
the upper state. Keep in mind that polarization is never perfect, so with some probability we’ll excite some
other level in the P3/2 manifold that can then decay into other S1/2 states.
In 40 Ca+ it’s not quite as easy. We can again tune a laser with σ+ polarization to the S1/2,+1/2 ↔ P3/2,+3/2
2.3. DIFFERENT TYPES OF QUBITS 27

transition. But as we excite to the P3/2,+3/2 level the electron can decay also to the D3/2 and D5/2 manifolds.
In 40 Ca+ we chose a different approach: We typically use the S1/2 ↔ P1/2 transition, the polarization does
not matter much. From P1/2 the electron can still decay to D3/2 , with a branching ratio of approx 1:20
compared to the decay to S1/2 . To not loose that population we use a second laser tuned to the D3/2 ↔ P1/2
transition that serves as a ‘repumper’, putting back any population getting lost from the S1/2 ↔ P1/2 cycle.
These are the laser configurations that we’ll use for Doppler cooling, discussed later.

2.2.2 Magnetic dipole transitions

One can also use magnetic fields to couple states. This coupling appears in the multipole expansion of the
atom’s/ion’s charge distribution which we cut at zeroth order for the dipole approximation. Typically this
will couple to the electron spin and flip it. The transition matrix elements for this are given by hf | µ · B |ii
with the magnetic dipole moment operator µ and the oscillating magnetic field B. Being second order in the
multipole expansion, this is a weak interaction, so the vacuum field that would cause spontaneous decay is
mostly irrelevant, leading to very long lifetimes for the fine/hyperfine states of the electronic ground state (I
think theoretical estimates say approx. 50 years for F = 1 states in 9 Be+ . But we can create decent oscillating
magnetic fields using rf or microwave currents to perform coherent spin flips. Considering only transitions
within state manifolds, the election rules are simply ∆mF or J = 0, ±1. Rabi frequencies are in the order of
Ω ≈ gS µB Bh̄ with gS the electron spin’s g-factor.

2.2.3 Electric quadrupole transitions

Additional to the magnetic dipole interaction, in the first order of the multipole expansion of the ion’s charge
distribution the electric quadrupole appears. Here the electron is coupling to the gradient of the electric field
instead of the magnitude as for the dipole transitions. One can estimate that the quadrupole term is a factor
(ka0 )2 weaker than the dipole term with k the light’s wavevector and a0 the Bohr radius, so for 729 nm about
2 × 10−7 [78]. Selection rules for quadrupole transitions are ∆L = 2, ∆mL = 0, ±1, ±2 with the magnetic
quantum number selection rule details depending on the polarization and magnetic field directions in rather
complicated ways [78]. Due to the weaker coupling we again gain in lifetime of the excited state. For the
states of the D manifolds in 40 Ca+ we get a lifetime of τ ≈ 1.2 s and thus a linewidth of Γ ≈ 2π × 130 mHz.
This makes these transitions also viable qubit choices.

2.3 Different types of qubits

We can in principle choose any two level to encode a qubit, but of course we want the best one available.
We usually distinguish two main classes of qubits. In one we encode the qubit in the electronic ground state
using different Zeeman-/Hyperfine- sublevels. This is called a ‘Zeeman’ or ‘Hyperfine’ qubit - this is what
we use in 9 Be+ and sometimes in 40 Ca+ . They have the big advantage that they do not decay in a relevant
duration, limiting any decoherence to dephasing. Their readout is usually not the best since any readout
by fluorescence detection needs to be able to distinguish sublevels in the same manifold. The other class of
qubits are ‘optical’ qubits. Here The qubit gets encoded in sublevels of different electronic states which are
connected by an optical transition, hence ‘optical’ qubit. In 40 Ca+ the typical choice is the S1/2 ↔ D5/2
transition. Here the upper state can decay within a second and this starts to matter for experiments that
have very high fidelities or very long sequences.

2.3.1 Qubit decoherence

Similar to the motional coherence there can be relaxation and dephasing processes. As mentioned for ground-
state qubits the natural relaxation is irrelevant, where for optical qubits it can matter if you’re good enough.
If we use Raman transitions (which we’ll discuss later) spontaneous photon scattering√can cause relaxation as
a dominant decoherence process. The Lindblad operator for relaxation would be L̂ = Γσ− with Γ the inverse
life time of the transition. Dephasing is usually the bigger issue. Here a main part are fluctuating magnetic
fields. They will cause
√ fluctuations of the energy difference of the qubit levels. The Lindblad operator for
this is given by L̂ = Γσz . The rate of decoherence Γ can be determined from experimental measurements
and will depend on the magnetic field noise spectral density. Watch out here: Depending on the frequency
components present in the noise a master equation is not the best description for this noise. More elaborate
methods like the Redfield equation would often be required to model the decoherence properly. Often we
however don’t care too much about what the spectrum is in detail. We just try to make things better. That
often means magnetic field shielding and stabilization. Another cause of dephasing is the phase noise of the
laser used to control the ion qubits.
28 CHAPTER 2. ELECTRONIC STRUCTURE
1e9
1.5

1.0

0.5
in GHz

0.0

0.5

1.0

0 100 200 300 400 500 600

B field (G)
Figure 2.2: This is the B field dependence of the individual states in the S1/2 ground state of 9 Be+ . At low
fields the Zeeman effect leads to a linear dependence of the level energies with the B field. Here F, mF are
good quantum numbers. At medium field the lines bends. If you find two that bend by the same amount
you can take that as a qubit. The two lowest ones are at approx. 120 G. In this regime there are no good
quantum numbers for the states. At high field the states group in two sets of four. Here the nuclear spin
starts to decouple from the electrons spin and good quantum numbers are S, mS and I, mI .

2.3.2 Field-independent qubits

There is a very powerful way to reduce the magnetic field dependence of a qubit by choosing the correct state
and magnetic field. In the beginning of this chapter I’ve noted the hyperfine structure arising for ions with a
nuclear spin, so there’s a term in the Hamiltonian proportional to I · J. There is a nice thing about it: At low
magnetic field good quantum numbers are F, mF , the total angular momentum and it’s orientation. And the
Zeeman effect will be proportional to mF . At stronger magnetic field the nuclear and Spin and the Spin-Orbit
angular momentum (J) decouple and good quantum numbers are I and J. This is analog to the Paschen-Back
effect where at very high magnetic field the electron spin and orbital angular momentum decouple. At an
intermediate magnetic field this leads to a non-linear magnetic field dependence of the states and and neither
F nor I and J are good quantum numbers. For some level pairs points can be found where the first-order
magnetic field dependence goes to zero. This allows us to get very long coherence times (seconds or more,
hard to measure).

2.4 State preparation and dark states

To control a qubit we first have to be able to initialize it, so prepare a target state. For this we typically use
optical pumping. Because this is already an interesting piece of physics and we will use this concept again
I briefly want to introduce the dark state here. Whenever we do state preparation we want this to be an
irreversible process of some sort. Why? Because we want to end the process in a specific state independent of
the initial state. Mathematically speaking, we’re starting in a mixed state and want to get into a pure state.
This cannot be achieved by unitary operations (only). One way would be to first collapse the mixed state
into a pure state and then transfer it to the target state. So we could first measure in which state the ion is.
This is a projective measurement and thus by nature non-unitary. Once we know that use unitary operations
to get it to the target state. For this however our measurement needs to preserve the state. This is rather
hard, so we have another, easier method. We create a process that keeps re-distributing population unless we
are in the dark state. If we are smart we can set up our process such that the dark state is the desired target
state.
For 9 Be+ this is fairly simple. We just turn on a σ+ polarized 313 nm beam that is resonant with the
S1/2 ↔ P1/2 transition. This drives the population to higher mF states but does not touch any population
2.5. READOUT 29

in the |F = 2, mF = 2i state, due to the selection rules. The |F = 2, mF = 2i state is the dark state of our
process. It does not scatter any photons, so it’s dark. We can directly make this one of our qubit states
(|0i = |F = 2, mF = 2i)and make for instance the |1i = |F = 1, mF = 1i state the other qubit state. If we
want to use a different qubit we can use unitary operations (Raman transitions, microwaves) to transfer the
population before we begin any computation.
For 40 Ca+ it’s slightly harder. Here we can also use σ+ polarized light tuned to the S1/2 ↔ P1/2 transition
to drive population to the mS = +1/2 state but we need an additional repumper beam (866 nm) to prevent
population to get lost in the D3/2 states.
In 9 Be+ the state preparation fidelities are typically limited by polarization impurity and/or state mixing
by the trap rf, resulting in an error of ≈ 0.5% (this is a SPAM error, State Preparation And Measurement,
it’s hard to disentangle the two). For 40 Ca+ this would be the same but there are advanced schemes using
dark resonances or the qudrupole S1/2 ↔ D5/2 transition reaching SPAM errors of ≈ 10−4 (possibly even
better).

2.5 Readout
Readout of the spin state is performed by ‘state dependent fluorescence’, meaning we want the ion to scatter
photons when it’s in one of the qubit states but not in the other, then record those photons with a photo
multiplier or camera. In the optical qubit in 40 Ca+ this is fairly easy. We turn on a laser beam resonant with
the S1/2 ↔ P1/2 transition (397 nm) together with the repumper (866 nm) - similar to the state preparation.
But this time we use π polarization for the 397 nm beam, such that photons are scattered no matter in which
state in S1/2 we are. This means that the S+1/2 state is our ‘bright’ qubit state. Naturally, if the ion is
in the other qubit state, some level in the D5/2 , this is not affected by the lasers we turned on, no photons
are scattered and we have our dark qubit state (not to be confused with the dark state I was talking about
before). This readout can be very good with errors below 10−4 .
For 9 Be+ it is in principle really easy - we simply turn on a σ + polarized beam resonant with the
S1/2 , F = 2, mF = 2 ↔ P3/2 , mJ = 3/2, mI = 3/2 transition, but only at 1̃ saturation intensity. This
will scatter predominantly photons from the S1/2 , F = 2, mF = 2 state, others are detuned. So we’ll choose
that as our qubit bright state. As qubit dark state we’ll choose the state that is furthest detuned and need
the most scattering events to reach the qubit bright state - so the S1/2 , F = 1, mF = −1 state. The issue is
now that if we detect for too long (or with too much power and power broaden the transition) the qubit dark
state will go bright - and there is nothing we can do against this. So to get a good readout in 9 Be+ we have
to be very efficient in our photon collection. Our qubits for encoding quantum information might be other
hyperfine levels than the detection states - in that case we have to first use coherent operations to transfer
one of the qubit states to the dark, the other to the bright state and then detect.
I want to point out that during detection we change the state of the ion - In 40 Ca+ the qubit state S+1/2
will end up in a mixed state of S+1/2 and S−1/2 , in 9 Be+ the qubit dark state will with some probability
scatter photons and end up in a different hyperfine level. Also, whenever we scatter photons we heat up the
ion - due to the photon recoil, so we also change the motional state. This has consequences: We need to
always re-prepare our spin and motion states at the begin of each experiment sequence. And if we want to
do some complicated experiment where we measure and then do something with the resulting state we have
to find a different way. We will get to such scenarios later in the lecture. You might have noticed: I did not
talk about the D5/2 state in 40 Ca+ - that’s the one which will be preserved and we used that in several of
our experiments.

2.6 Connecting qubit states

We already discussed the different two-level system choices that are available. Now we look at the tools that
we have to manipulate them. For this I now show you the typical two-qubit interaction Hamiltonian that
we implement. We’ll discuss it in detail later. For now I just wanted to point out the important physical
quantities to connect them to the tools that we have.

Ω
δ
Ĥ = σ− eiφ + σ+ e−iφ + σz . (2.2)
2 2
Here and in the following I’ll omit the h̄, then my energy units are in Hz which I like because I can think in
MHz but not in Joule. Here Ω is the Rabi frequency and δ is a detuning of our laser (or microwave) from the
qubit transition. Usually, we’ll be using this Hamiltonian with δ = 0 to implement single qubit gates - the
σ± terms will simply rotate the spin between up and down. Please note that I’ve explicitly put in the phase
30 CHAPTER 2. ELECTRONIC STRUCTURE

φ. Sometimes this is omitted but it’s quite essential for us. However we implement this Hamiltonian we need
control over Ω, φ and δ - and we need to be able to turn the Hamiltonian off and on - for all the tools below
this means changing Ω from zero to some value and back to zero - so forming a pulse.

2.6.1 Quadrupole transition

I’ve already mentioned electric quadrupole transitions. Here I just want to make a note on the laser require-
ments to drive such a transition. When driving a quadrupole transition with a laser there is an issue we
usually don’t encounter with dipole transitions: The laser’s linewidth will be broader than the transition’s. In
term’s of decoherence this could mean that the dephasing is dominated by the laser linewidth instead of the
magnetic field noise or the excited state’s lifetime. This means a lot of work has to go into the stabilization
of the laser. We do this by stabilizing the laser against a more stable reference. This reference for long
time scales is just the qubit transition itself. At shorter timescales a optical cavity with very stable mirror
distance serves as reference. By stabilizing the laser to the length of a cavity it is possible to achieve sub-Hz
laser linewidth. Due to the weak coupling of qudrupole transitions we also require quite high laser power
(≈ 1 mW). The Rabi frequency will be determined by the laser power and the beam diameter. The phase
will be determined by the phase of the laser pulses. This phase has initially some arbitrary value, but is quite
stable. This is fine because only the relative phases of the laser pulses matter. The absolute phase is basically
set by the first operation we do. How do we control the phase of the laser? Usually we use acousto-optic
modulators (AOMs). You can read up details about those devices if you’re interested. Suffice to say that
they take a rf frequency and add or subtract it from the laser beam. Turning on and off the rf signal allows
us to turn on and off the beam, producing pulses of arbitrary duration. The important thing here is that the
phase of the laser is also influenced by the rf signal. If we want to shift the phase of our operation on the
qubit we simply shift the rf phase that we apply to the AOM.

2.6.2 Magnetic dipole transitions

If we want to use a ground state qubit we need a way to connect the ground states. one way of doing this
is to simply use magnetic dipole transitions. I’ve introduced them before. How do we drive these? With an
oscillating magnetic field. We have two possible choices how to create that field. One (rarely used) is to use
the far-field, so basically creating a microwave beam outside of the vacuum chamber and directing it at the
ions inside the chamber. This is not very efficient since the trap structure and the vacuum chamber provide
quite good shielding against this. Another way is to use near-field oscillating fields. They can be created by
putting a wire loop relatively close to the ions and running a microwave/rf current through them. This is
very convenient because microwaves are a mature technology and can be created relatively easily and have
very good phase noise. So they will not limit the qubit decoherence. But using microwaves in qubit control
also has some issues. One is that the wavelength is very long (1 GHz is ≈ 30 cm), so it’s not easy to talk to
one qubit but not the other qubit. Other issues with regard to spin-motion coupling we’ll get to later. But
we’ll also discuss nice solutions to these issues that put microwaves back on the board for scaling TIQI.

2.6.3 Stimulated Raman transitions

To control ground state qubits there is an alternative: Stimulated Raman transitions. This was already
covered in the Quantum Optics lecture. So here just a brief review. We use two laser beams that have a
detuning of the ground state qubit splitting and have some common detuning ∆ from a dipole transition. This
leads to a coherent coupling of the two qubit levels though a virtual, excited state. A simplified expression
for the resulting Rabi frequency ΩRaman is given by

Ω1 Ω2
ΩRaman ≈ (2.3)
∆
where Ωi are the Rabi frequencies of the individual beams with respect to the detuned dipole transition. What
I do not take into account here (and below in the scattering rate) are the transition matrix elements - this here
is just a general scaling. You can think of stimulated Raman transitions as a microwave implemented with
lasers, with the advantage of a much shorter wavelength. In Raman transitions dephasing is not caused by the
phase noise of the laser itself. This is irrelevant when both beams come from the same laser and thus the noise
is common mode and drops out. Since the detuning of the beams from each other is usually done by acousto-
or electro-optics with a rf/microwave input this is also not an issue (same as in microwaves). One issue can
arrive from beam path fluctuations: If the two beams take different paths to the ion and those path lengths
are fluctuating this causes phase fluctuations at the ion. Another, more fundamental issue is spontaneous
photon scattering. The detuning of the beams from the dipole transition is not infinite, so there is a finite
2.7. TWO-LEVEL SYSTEM, RABI, RAMSEY 31

change for the exited state to be populated and scatter a photon. This then causes either a phase-reset of the
qubit if the electron decays to the same ground state level it stated from (Rayleigh scattering) or a spin flip
if if decays to the other qubit level (Raman scattering). The good thing is that these scattering rates (Γscat )
have a stronger dependence on the Raman detuning than the Rabi frequency


Γdip Ω21 + Ω22
Γscat ≈ , (2.4)
∆2
with Γdip the natural linewidth of the dipole transition that the beams are detiuned from. So, to achieve
high-fidelity control one requires large detunings (∆ ≈ 0.1 THz to 1 THz) and high laser intensities (∆ ≈ 1 mW
to 10 mW). The detuning and phase, δ and φ, that appear in eq. 2.2 are controlled by the detuning of the
Raman beams from each other and their phase difference. We again control the individual beams using AOMs.

2.7 Two-level system, Rabi, Ramsey

We’ll now look at how the two-level system works on an abstract level. For this we look at the two famous
experiment sequences: Rabi and Ramsey, named after their inventors. (I will use the terms two-level system,
spin and qubit interchangeably.) They were developed as spectroscopy protocols, used to determine the
properties of an atomic transition with the most important quantity of course the transition frequency. Or
course they also have a relevance for quantum information, Rabi pulses are our way of performing most
single-qubit operations.

2.7.1 Rabi model

As mentioned before, we will derive the proper ion-light interaction when looking at spin-motion coupling, for
now I just show you the result when the motion is not relevant. A Rabi experiment is simply using a laser or
microwave pulse to drive the spin continuously followed by readout of the spin state. So the general unitary
operation that we implement with a laser or microwave on our two level systems is given by the following
Hamiltonian written in the Schrödinger picture with respect to the atomic operators.
h̄ωa h̄ΩR
σ̂+ eiφ e−iωL t + σ̂− eiωL t e−iφ .


Ĥ = σ̂z + (2.5)
2 2
Here ωa is the transition frequency of our two level system (atom), ωL is the laser frequency, ΩR the Rabi
frequency and φ the phase of the laser. The coefficients for σ̂+ , σ̂− resemble a system which is in the
interaction picture with respect to a spin Hamiltonian
h̄ωL
Ĥ0 = − σ̂z , (2.6)
2
where the minus sign is deliberate. Instead of continuing with the time evolution present, we can therefore
move to an interaction picture of the Hamiltonian
h̄ωL
Ĥ00 = σ̂z , (2.7)
2
We must compensate for this term in our interaction Hamiltonian by subtracting it.
ΩR eiφ
 
h̄ −δ h̄δ h̄ΩR
ĤI = −iφ = − σ̂z + (cos(φ)σ̂x − sin(φ)σ̂y ) , (2.8)
2 ΩR e δ 2 2
where δ = ωL − ωa .
We could try to find eigenvalues and eigenvectors to understand this system. To illustrate an alternative
approach, we note that the exponential of any sum of the three Pauli operators has a simple form, due to
their commutation relations. If we expand the exponential

eiK(aσ̂z +bσ̂y +cσ̂x ) (2.9)

where a2 + b2 + c2 = 1, we find
K2
1 + iK(aσ̂z + bσ̂y + cσ̂x ) − (aσ̂z + bσ̂y + cσ̂x )2 − ... (2.10)
2!
The squared term gives

(aσ̂z + bσ̂y + cσ̂x )2 = a2 σ̂z 2 + b2 σ̂y 2 + c2 σ̂x 2 = 1 (2.11)

32 CHAPTER 2. ELECTRONIC STRUCTURE

because products of the form σ̂z σ̂x + σ̂x σ̂z = 0 (try it and see). This means that the expansion of the
exponential gives
K2 K3
1 + iK(aσ̂z + bσ̂y + cσ̂x ) − 1−i 1(aσ̂z + bσ̂y + cσ̂x ) + etc. (2.12)
2! 3!
which simplifies to
1 cos(K) + i(aσ̂z + bσ̂y + cσ̂x ) sin(K). (2.13)
For the Rabi problem, we need to find a common factor and normalized coefficients a, b, c. Since the sum of
the squared coefficients of σ̂x , σ̂y , σ̂z in this case is (δ 2 + Ω2R )/4, it makes sense to extract this factor such
Rt p
that we can write K = 0 dt0 Ω0 with Ω0 /2 = δ 2 + Ω2R /2, and the normalized coefficients are a = −δ/(Ω0 ),
b = ΩR sin(φ)/(Ω0 ) and c = ΩR cos(φ)/(Ω0 ). We can then use the form of equation 2.13 to find that
 
ΩR cos(φ) ΩR sin(φ) δ
ÛI (t) = 1 cos(Ω0 t/2) − i σ̂ x − σ̂ y − σ̂ z sin(Ω0 t/2). (2.14)
Ω0 Ω0 Ω0
In order to go back to the Schrödinger picture we note that the interaction picture state is given by
|ψ(t)iI = ÛI (t) |ψ(0)iI = ÛI (t) |ψ(0)i , (2.15)
so
|ψ(t)iS = e−iĤ0 t/h̄ |ψ(t)iI
= e−iωL σ̂z t/2 |ψ(t)iI
= (cos(ωL t/2)1 − i sin(ωL t/2)σ̂z ) |ψ(t)iI
 −iω t/2 
e L 0
= ÛI (t) |ψ(0)i
0 eiωL t/2
 −i(δ+ω )t/2 
e a
0
= ÛI (t) |ψ(0)i . (2.16)
0 ei(δ+ωa )t/2
For δ  ωa we can make a rotating-wave approximation to obtain
−i ΩΩR0 eiφ e−iδt/2 sin(Ω0 t/2)

 −iδt/2
cos(Ω0 t/2) + i Ωδ0 sin(Ω0 t/2)


e
ÛS (t) = . (2.17)
−i ΩΩR0 e−iφ eiδt/2 sin(Ω0 t/2) eiδt/2 cos(Ω0 t/2) − i Ωδ0 sin(Ω0 t/2)
Note that on resonance the unitary transformation in the Schrödinger picture takes the form of a rotation
matrix
Rδ=0 (θ, φ) = 1 cos(θ/2) − i (sin(φ)σ̂x + cos(φ)σ̂y ) sin(θ/2) (2.18)
−ieiφ sin(θ/2)
 
cos(θ/2)
= (2.19)
−ie−iφ sin(θ/2) cos(θ/2)
with the rotation angle given by θ = ΩR t. We then refer to pulses as “π” pulses if θ = π (this inverts the
spin) and “π/2” pulses for θ = π/2. Note that though for θ = 2π, the spin state is the same, there is a global
“−1” factor that occurs. Though this is general not important, for the first ion trap quantum logic gates, the
use of this phase was crucial.

Experiments with Rabi oscillations

The observation of Rabi oscillations resulting from the dynamics just described usually proceeds along the
following lines. First one of the energy eigenstates of the system is prepared, usually by optical pumping
(we will see specific examples later in the course). The oscillating control field for the Rabi oscillations is
then turned on for a fixed duration t at a given frequency, phase and field strength. After this field has been
switched off, a measurement is made of which energy eigenstate the system is in.
For a system prepared in lower spin state, after a time t, the probability that a measurement finds the
spin to be in the upper state is (using equation 2.17)
2
ΩR iφ −iδt/2
P (|↑i) = e e sin(Ω0 t/2)
2Ω0
Ω2R
 q 
2 1 2 2
= 2 sin ΩR + δ t . (2.20)
ΩR + δ 2 2
This is plotted as a function of time and detuning for a fixed Rabi frequency in figure 2.3. It is worth noting
that the oscillations are faster if the system is off-resonance, but that the contrast of the oscillations is then
reduced.
2.7. TWO-LEVEL SYSTEM, RABI, RAMSEY 33

1.0
/ R =0
/ R =1
0.8 / R =2

Pe 0.6
0.4

0.2

0.0
0.0 0.5 1.0 1.5 2.0
t R /2

1.0
t R= /2
t R=
0.8 t R =2

Pe 0.6
0.4

0.2

0.0
4 3 2 1 0 1 2 3 4
/ R

Figure 2.3: Top: Plot of probability to be in the excited state as a function of time for δ = 0 (dashed)
and δ = ΩR e (solid). Note the reduced contrast and faster oscillations of the latter. Bottom: Frequency
dependence, for t = π/ΩR R
e (dashed) and t = 2π/Ωe .
34 CHAPTER 2. ELECTRONIC STRUCTURE

Experimental issues
In order to probe a system using Rabi spectroscopy, we turn on the field for a duration t, let it evolve,
and then measure the spin system at the end. To get statistics, we either have to repeat the experiment
many times, or we must use many particles simultaneously. However note that we would like to do the same
experiment every time we repeat. This is a problem for experiments with a single quantum system, because
the laboratory environment does not stay constant in time - noise in external fields will lead to different
behavior on each repeat. Noise affecting a Rabi experiment includes frequency noise in the field oscillator,
frequency fluctuations of the atomic transition, and intensity fluctuations of the field oscillator.
For the use of many systems simultaneously, there is very often the problem that the systems don’t all sit
in exactly the same position, and thus the intensity of the field at each system is different. This can also be
true of the atomic transitions themselves, for instance, an inhomogeneous magnetic field will mean that each
atom has a different transition frequency.

2.7.2 Ramsey method

The Ramsey experiment is “A Molecular Beam Resonance Method with Separated Oscillating Fields” [77].
That was the original naming by Ramsey himself. In a Ramsey experiment we’re making two π/2 pulses,
separated by a wait time τ . It is one of the most important methods in quantum science. It is a simple
interference experiment on a quantum state superposition. It avoids some of the problems mentioned in the
previous section for the Rabi method, because for much of the evolution time of the system, the control field
is turned off.
The Ramsey method consists of applying two Rabi π/2 pulses separated by a time interval τ . A Rabi
pulse with ΩR e tπ/2 = π/2, phase φ1 and δ = 0 produces a spin rotation

−ieiφ1
 
1 1
R̂(π/2, φ1 ) = √ −iφ . (2.21)
2 −ie 1 1

When applied to an initial state |↓i, this produces the equal superposition
1
√ |↓i − ieiφ1 |↑i ,


(2.22)
2

which evolves in time according to the free evolution Hamiltonian Ĥ0 = h̄ωa σ̂z /2. The general evolution can
be written as a matrix as
 −iω τ /2 
e a 0
Ûfree (τ ) = , (2.23)
0 eiωa τ /2

where τ is the evolution time. After the interval of length τ , the state of the atom is

eiωa τ /2  
√ |↓i − iei(φ1 −ωa τ ) |↑i , (2.24)
2
i.e. the superposition has evolved in phase due to the energy difference between the two states.
Subsequently a Rabi pulse with the same duration and frequency as the first is applied, but now with
phase φ2 . This means that the total evolution looks like

Ûtot = R̂(π/2, φ2 )Ûfree (τ )R̂(π/2, φ1 ), (2.25)

which for a starting state |↓i, produces

1
P (↓) = (1 − cos(φ2 − φ1 + ωa τ )) . (2.26)
2
Now imagine that both control pulses were generated from the same frequency source, which runs with
frequency ωL (a laser, for example). Then the phase of the second pulse will be φ2 = φ1 − ωL τ . In this case
1
P (↓) = (1 + cos(δτ )) . (2.27)
2
Looking at this expression, we see that there are fringes (much as in the Young’s double-slit experiment)
which have minima when δτ = (2n + 1)π. The width of each fringe in frequency is thus proportional to 1/τ ,
i.e. a longer delay gives higher sensitivity to frequency shifts. The interpretation of this is that the Ramsey
experiment performs a phase comparison between the phase of the control pulses and the phase of the atom.
2.7. TWO-LEVEL SYSTEM, RABI, RAMSEY 35

1.0

0.8

Pe 0.6
0.4

0.2

0.0
100 50 0 50 100
/2 (Hz)
1.0

0.8

Pe 0.6
0.4

0.2

0.0
4 2 0 2 4
/2 (Hz)
Figure 2.4: Top: Plot of probability to be in the excited state as a function of detuning for a Ramsey
experiment with τ = 1 s and tπ/2 = 25 ms. Bottom: a close up of the central region, showing that our
approximate expression holds for δ  ΩR
e (we see effectively full contrast sinusoidal fringes).

In the derivation above, we made an approximation that the relative detuning matters in the phase evolu-
tion, but not in the Rabi pulse itself (these were assumed to be resonant). We can make this approximation
so long as the frequency difference δ  ΩR , i.e. so long as the Rabi pulses approximately do the π/2 rotations
that were expected. This can typically be arranged while retaining sensitivity to the phase shift, by making
τ  tπ/2 . The general case looks mathematically much more messy, involving the finite lengths of the Rabi
pulses. This introduces an envelope function due to the lack of dynamics when the Rabi pulses are far from
resonance. An example of a frequency scan with τ  tπ/2 is given in figure 2.4, with the second figure showing
that close to resonance the dynamics is that of the simplified form from equation 2.27.

Experiments with Ramsey fringes

The Ramsey method is the same as for Rabi oscillations. First we must prepare the atom by optical pumping,
and after the two pulses with a gap in the middle, we measure the probability to be spin up. This method
has the significant advantage that the spectroscopic sensitivity is to the phase of the control oscillator to the
atomic phase.

2.7.3 Basic working principle of an atomic clock

Very stable (long lifetime but also insensitive to external perturbations) atomic transitions can be used as
a clock, since a clock is nothing different than a stable frequency that’s being counted. How do we extract
what that frequency is? We perform spectroscopy (Rabi or Ramsey) on said transition, stabilizing the laser
or microwave frequency to it. This is a different point of view than before: Here we say: we don’t know what
the laser frequency is and simply define the unit of time as 1/f with f the atomic transition frequency. Using
36 CHAPTER 2. ELECTRONIC STRUCTURE

a frequency comb we can then turn the optical frequency into an rf signal that can actually be counted with
electronics.
Chapter 3

Spin-motion coupling

3.1 Atom-Field interaction

We want to control our ions using electromagnetic fields, so we should look at how their interaction works.
You should look at chapter 3 of J.P. Home’s Quantum optics script here. Things will look familiar, and this
may be somewhat of a repetition for you.
We first look at the coupling of a charge only to a classical field. We start with the charge-field Hamiltonian
in the Coulomb gauge
p̂2 q (q Â)2
Ĥ = − Â · p̂ + . (3.1)
2m m 2m
with the vector potential A(r̂) (the scalar electric potential φ is zero due to the Coulomb gauge).
We now have to figure out which terms are important to us: you’ll see in the problem sets that the last
term is much smaller than the others for the parameter ranges typically considered in trapped-ion physics.
Thus we neglect this term, and are left with a simple interaction which is linear in the field and the momentum
of the charge. We now apply this result to an atom, where for our purposes the charge is that of the most
weakly bound electron.

3.1.1 Interaction with an atom

An electron in an atom is bound, but we want to know what effect the electric field has on the center of mass
(c.o.m.) of the composite system, rather than just the electron itself. This is of particular importance in spin
motion coupling.
Here, we will look at the procedure for obtaining a simple atom-field interaction Hamiltonian from the
result in equation (3.1). We start from the Hamiltonian
q
Ĥ = Ĥa + Ĥt − p̂el · A(r̂el ) (3.2)
m
q
= Ĥ0 − p̂el · A(r̂el ) (3.3)
m
= Ĥ0 + Ĥ1 (3.4)
where p̂el and r̂el are the electron momentum and position respectively and m and q are the electron mass
and charge. The internal and atom translation Hamiltonians are
h̄ωa
Ĥa = σ̂z (3.5)
2
P̂2
Ĥt = . (3.6)
2M
M is the mass of the atom, and P̂ is the momentum of the center of mass of the whole atom. At present,
we see that the translational energy of the center of mass is only dependent on the kinetic term, that is, we
assume no confining potential - we will add that later.
The other part we’ll sometimes call ‘Interaction Hamiltonian’. This has no connection with the interaction
picture! It’s just saying ‘this is the part of the Hamiltonian describing the interaction. This might be poor
nomenclature but is common.
q
Ĥ1 = − p̂el · A(r̂el ) (3.7)
m

37
38 CHAPTER 3. SPIN-MOTION COUPLING

The eigenstates of the Hamiltonian Ĥ0 are given by products of eigenstates of the individual terms, i.e.
the energy eigenstates are given in a general way by

|atom motioni |atom internali = |Ki |ai , (3.8)

where |Ki is a momentum eigenstate for the atom corresponding to momentum h̄K. For simplicity, we will
consider only two atomic states |ai, which we call the excited and ground states and denote |ei and |gi.
Before moving any further, it is worth moving to a coordinate system involving the position of the center
of mass R̂ of the atom and the relative position r̂ of the electron from the nucleus. In this case, the position
of the electron is given by

r̂el = R̂ + r̂. (3.9)

On differentiating once with respect to time and multiplying by the electron mass m, we find
m
p̂el = P̂ + p̂, (3.10)
M
where p̂ is the relative momentum of the electron with respect to the nucleus, and we used the fact that the
reduced mass mM/(M + m) ' m. The term describing the electron-field interaction in the Hamiltonian of
equation (3.4) then becomes
q q m 
− p̂el · A(r̂el ) = − P̂ + p̂ · A(r̂el ). (3.11)
m m M
Let us first consider the field to be a plane wave field with wavevector k and polarization unit vector ec .
The vector potential for such a field is
 
A(r̂el ) = A0 ec ei(k·r̂el −ωL t) + e∗c e−i(k·r̂el −ωL t) (3.12)
 
= A0 ec eik·R̂ eik·r̂ e−iωL t + e∗c e−ik·R̂ e−ik·r̂ eiωL t (3.13)
 
= A0 ec eik·R̂ eik·r̂ e−iωL t + H.c. , (3.14)

where A0 is the amplitude of the field. This is a classical field. We could also derived this with a quantum field
and after the derivation choose to go to the classical limit (i.e. a strong field). Note that in equation (3.14)
we split up an exponential of a sum of operators into a product of exponentials of the individual operators.
We can do this because these two operators commute.
A Hamiltonian is defined by its matrix elements. The interesting matrix elements are those which give
transitions between different eigenstates of Ĥ0 and which are close to resonance. For this we assume some
initial and final state of the motion |Ki and |K0 i and the atomic states |gi, |ei

hK0 | he| Ĥ1 |Ki |gi . (3.15)

Combining the Hamiltonian in equation (3.4) and the form of the vector potential given in equation (3.14),

qA0 −iωL t m
hK0 | he| Ĥ1 |Ki |gi = − e hK0 | P̂ · ec eik·R̂ |Ki he| eik·r̂ |gi
m M
qA0 −iωL t
− e hK0 | eik·R̂ |Ki he| p̂ · ec eik·r̂ |gi
m
+ H.c. (3.16)

In the problem sets, you will take this expression and use it to show that the first of these terms is much
smaller than the second if the atom is non-relativistic. As a result, we can neglect the first term, and arrive
at a simpler form for the interaction Hamiltonian which is what we want. For the atom motion you will find
that
q m qh̄
− hK0 | P̂ · ec eik·R̂ |Ki = − K0 · ec δ(K0 , K + k), (3.17)
m M M
where δ() is the Dirac delta function.
For the internal state coupling you will expand

p̂ · ec eik·r̂ = p̂ · ec (1 + ik · r̂ + ...) (3.18)

3.2. LAMB-DICKE EXPANSION 39

and truncate at the zeroth order. This is the dipole approximation. If you wanted to evaluate the matrix
elements for the magnetic dipole or quadrupole transitions you’d keep the first order. Using this (and some
trick with the interaction picture) you’ll get

he| p̂ · ec eik·r̂ |gi = imωa he| r̂ |gi · ec . (3.19)

Using the matrix elements from equations (3.17) and (3.19), we can then express the Hamiltonian as

Ĥ1 = −iωa qA0 e−iωL t δ(K0 , K + k) |K0 i hK| ec · he| r̂ |gi σ̂+ + H.c.. (3.20)

We can further clean this result up a bit by defining a Rabi frequency as

2ωa qA0
Ω = −i ec · he| r̂ |gi . (3.21)
h̄
Remember that this is the expression for a Rabi frequency of a dipole transition. For electric quadrupole and
magnetic dipole transitions the matrix element would be replaced by he| (ec · r̂)(k · r̂) |gi. If you want to go
into more details I found this online book very nice: [84]. In any case we’ll get the cleaner looking version

h̄Ω −iωL t
Ĥ1 = e δ(K0 , K + k) |K0 i hK| σ̂+ + H.c.. (3.22)
2
This is our final result for a free atom.

3.1.2 Trapped ions

In the case studied above we considered a free atom. To extend this to a trapped atom (or ion) we can leave
the translational part of the Hamiltonian unevaluated (the arguments regarding relative sizes of terms remain
if the atom is non-relativistic).

h̄Ω −iωL t ik·R̂

Ĥ1 = e e σ̂+ + H.c. (3.23)
2
Most of the time we’ll only look at one motional mode. Then we’ll use the simplified expression (assuming
the mode we’re looking at is in the z direction, and dropping any z subscripts)

h̄Ω −iωL t ikz Ẑ

Ĥ1 = e e σ̂+ + H.c. (3.24)
2
Not that I’ll also simplified the exponent with the position operator. We’ll evaluate the motional part in the
next chapter.

3.2 Lamb-Dicke expansion

For this and the following chapters I’ll change notations. We’re now not concerned with the electrons mass
or position anymore. So for the ions position and mass we’ll use lower-case letters from now on.
At the end of the last chapter we made the first mode to the quantized motion of a trapped ion, but we
didn’t do this for the position operator yet. Let’s do that for our one-dimensional version 3.24

kz ẑ = kz z0 (↠+ â) = η(↠+ â), (3.25)

p
were z0 = h̄/2mωz is the ground state wavepacket size. We’ve now introduced the Lamb-Dicke parameter
η. Using the definitions of k and z0 we can also write is as
r
2π h̄
η= cos (θ), (3.26)
λ 2mωz
where θ is the angle of the z-axis and the k-vector/laser beam. With typical wavelengths for optical transitions
λ ≈ 500 nm and typical ground-state wavepacket sizes of z0 ≈ 10 nm it’s clear that η  1. This means we can
expand the exponential in equation 3.24 with respect to η. This is the Lamb-Dicke expansion. It is analog to
the multipole expansion for the electronic transition - just now for the motional part.

η2 †
eik·R̂ = 1 + iη(↠+ â) − (â + â)2 + ... (3.27)
2
40 CHAPTER 3. SPIN-MOTION COUPLING

We’ve only compared λ to the ground state wave packet size. For higher Fock states the wave packed size will
be larger - this leads to a second constrain for the expansion - the Fock state should be low enough.
q Formally
we would require η 2 (2n + 1)  1, which is combined with η  1 just saying 1  kzn = k hn| R̂2 |ni, with
zn the rms size of the wave packet of Fock state |ni.
Inserting the expansion to second order into the Hamiltonian
η2
 
h̄Ω
Ĥ1 = 1 + iη(â + â ) − (â + â ) + ... σ̂+ e−iωL t + H.c.
† † 2
(3.28)
2 2
We now move to the interaction picture with respect to spin and motion
h̄ωa
Ĥ0 = σ̂z + h̄ωz (↠â + 1/2) (3.29)
2
and get
η2
 
h̄Ω
ĤI = 1 + iη(âe−iωz t + ↠eiωz t ) − (âe−iωz t + ↠eiωz t )2 + ... σ̂+ eiωa t e−iωL t + H.c. (3.30)
2 2

3.2.1 Physical interpretation of the LD parameter

The LD approximation and the LD parameter can be interpreted in several ways. One is that the wavelength
is much longer than the ion’s wave packet size, so it can be ‘linearized’. And the Lamb Dicke parameter is
just (up to a factor of 2π) the ratio of the wave packet size and the wavelength.
The other interpretation is comparing the (quantized) photon and ion recoil energy. The recoil energy on
a specific mode of the ion’s motion when absorbing a photon pf momentum p = h̄k is
(h̄k cos(θ))2
Er = (3.31)
2m
where θ is again the angle between the k-vector and the mode of motion. We can now look at the fraction of
the recoil energy and the energy of a quantum of motion of the respective mode and find
Er h̄(k cos(θ))2
= = η2 . (3.32)
h̄ωz 2mωz

3.2.2 Laser beam geometry and microwaves

The LD parameter for a given mode will depend on the the wavelength that is used to address that transition
and the projection of the laser’s k-vector on the motional mode direction. Some examples:
• For 40 Ca+ , ωm = 2π×1 MHz the optical qubit with λ = 729 nm and the laser’s k-vector k = {0, 0, 2π/λ}
the LD parameters are ηx = ηy = 0 and ηz ≈ 0.1.
• For√40 Ca+ √, ωm = 2π × 1 MHz the optical qubit with λ = 729 nm and the laser’s k-vector k =
{0, 2π/λ, 2π/λ} (so an angle of 45 deg on both y and z modes) the LD parameters are ηx = 0
and ηy = ηz ≈ 0.07.
• For 9 Be+ , ωm = 2π × 1 MHz the hyperfine qubit (1.2 GHz) with λ = 0.25 m and the laser’s k-vector
k = {0, 0, 2π/λ} the LD parameters are ηx = ηy = 0 and ηz ≈ 6.0 × 10−7 .
This tells us that microwaves have basically no coupling to the motion. That can be good and bad, as we see
below.
What about Raman transitions? Here we have two k-vectors and what counts for the LD parameter is
their difference ∆k = k2 − k1 . This means we can ‘tune’ the LD parameter. The two extremal cases are
• Using counter-propagating Raman beams, so an angle of 180 degrees between the two beams. For
9
Be+ , ωm = 2π × 1 MHz the hyperfine qubit (1.2 GHz), λ ≈ 313 nm: ∆k = k2 − k1 = {0, 0, k2 − k1 } ≈
{0, 0, 4π/λ}; and the LD parameters are ηx = ηy = 0 and ηz ≈ 0.95. A huge LD parameter!
• Using co-propagating Raman beams, so an angle of 0 degrees between the two beams. For 9 Be+ ,
ωm = 2π × 1 MHz the hyperfine qubit (1.2 GHz), λ ≈ 313 nm and both beams parallel to the z mode.
For an exact result we need to take the frequency difference of the two Raman beams into account. This
is just the qubit splitting of 1.2 GHz which corresponds to a difference of the wavelengths of the two
beams of δλ = λ2 − λ1 ≈ 0.4 pm. This results in the same ‘effective wavelength’ as for the microwave
case λeff ≈ λ2 /δλ ≈ 0.25 m. So ∆k = k2 − k1 ≈ {0, 0, 2πδλ/λ2 } ≈ {0, 0, 2π/(0.25 m)} and the LD
parameters are the same as we had for the microwave: ηx = ηy = 0 and ηz ≈ 6.0 × 10−7 .
3.3. TRANSITIONS 41

This tells us that with Raman beams we can choose if we want to couple to the motion or not. The great
advantage compared to the microwave is that the Raman beams are still localized, so I can still address single
ions.

3.3 Transitions
The transitions that the Hamiltonian 3.30 produces occur when the exponential on certain terms vanish. Foe
now we’ll only take the zeroth and first order of the LD expansion and deal with the higher orders later. Only
considering the zeroth and first orders is usually called the LD approximation.

3.3.1 Carrier transition

We first look at the case that the laser is simply close to resonance with the atomic transition, i.e. ωa ≈ ωL
and drop all terms rotating with ωz or more. This gives us the Carrier Hamiltonian (in first order of the LD
expansion)
h̄Ω
Ĥc = 1m σ̂+ eiδt + H.c. (3.33)
2
with δ = ωa − ωL and the identity operator for the motion 1m . This is the resonant case of the Hamiltonian
we’ve already looked at when only considering spin operations. Nothing special here, this will only affect the
two-level system and leave the motion untouched.

3.3.2 First sideband transitions

When we look at the next leading term we find resonances for ωL = ωa ± ωz . These are what we call the first
blue and red sidebands. Their Hamiltonians are:
h̄Ωη h̄Ωbsb
Ĥbsb = σ̂+ ↠+ H.c. = σ̂+ ↠+ H.c. (3.34)
2 2
h̄Ωη h̄Ωrsb
Ĥrsb = σ̂+ â + H.c. = σ̂+ â + H.c. (3.35)
2 2
The second of these Hamiltonians might look familiar, it is the famous Janyes-Cummings-Hamiltonian that
appears in (circuit) cavity quantum electrodynamics, taking one excitation out of the spin and putting it into
the motion (or vice versa): |n, gi ⇐⇒ |n − 1, ei. The first one is what is sometimes called the anti-Janyes-
Cummings-Hamiltonian and is doing a somewhat surprising operation: exciting both the motion and the spin
(or taking it out of both): |n, gi ⇐⇒ |n + 1, ei. This one does not appear in (c)QED. Thinking about the
energy scales of the atomic transition and the motion both of these Hamiltonians actually seem to be weird.
Even the process of transforming excitation of the atom into excitation of the motion we seem to loose energy
and gain momentum. The important part you should not forget is that there is a laser field which can provide
the energy and be tuned to either resonance - in (c)QED that is not possible. For the momentum this does
not work out though - the momentum of the photon does not match the momentum of an ion’s Fock state -
and this issue of course also appears in the carrier Hamiltonian where the motion was completely unaffected.
So where does the momentum go? It’s swallowed by the other classical device that we have - the trap. This
is identical to the Mössbauer effect where recoil of photon emission of a single atom is absorbed by the solid
state crystal it’s embedded in.
What we should also note is that the Rabi√frequency of the first sideband Hamiltonians is suppressed by
η compared to the carrier - additional to the n dependence that comes from ↠, â.
At this point it is useful to introduce a pictorial representation of the Hilbert spare. We can draw a ‘state
ladder’ and indicate the different transitions as shown in figure 3.1. One very important point is that if we are
in |n = 0, gi and drive the blue sideband we get |n = 1, ei but if we drive the red sideband nothing happens.
The state ladder ends and there is no state for the red sideband to couple to. The same is happening for the
blue sideband when we start in |n = 0, ei. We can use this imbalance to find out√if we are in the motional
ground state, so use it to diagnose the motional state. More generally, due to the n dependence we can use
sidebands to analyse the populations of the motional state. If we do Rabi flopping on a sideband the spin
populations depend on n. Formally we can use the same steps that lead us to equation 2.20 when looking
at Rabi flopping. The difference that we have to make is take into account the probabilities for the different
Fock states. We assume we are ins some motional state described the general density matrix
X
ρm = ρi,j (3.36)
i,j
42 CHAPTER 3. SPIN-MOTION COUPLING

Figure 3.1: The state ladder. It shows the states coupled by the carrier transition (green), the red sideband
(red) and the blue sideband (blue) as well as their Rabi frequencies.

0.25 0.5 0.25

A) B) C)
0.2 0.4 0.2
0.15 0.3 0.15
p(n)

p(n)

p(n)
0.1 0.2 0.1
0.05 0.1 0.05
1 1 1
0 0 0
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
0.8 n 0.8 n 0.8 n

0.6 0.6 0.6

P(↓)

0.4 0.4 0.4

0.2 0.2 0.2

0 0 0
0 500 1000 1500 0 200 400 600 800 1000 0 100 200 300 400 500
Blue sideband pulse duration (µs)

Figure 3.2: Blue sideband flopping curves with inset populations from fits for motional states: A) coherent
B) squeezed C) squeezed-coherent. Taken from [51].

So the probability to be in Fock state n is given by Pn = ρn,n . (This is what we sometimes call ‘populations’,
contrary to ‘coherences’ which refers to the off-diagonal terms of the density matrix.) So for the outcome of
a Rabi experiment where the laser is tuned to the blue sideband transition, turned on for a duration t and
the initial spin state is |gi, the probability to be in |ei is given by

1 X 1 √
Pe = − Pn cos(Ωη n + 1t). (3.37)
2 i
2

You can convince yourself that this is the same (for a single n state) as the sin2 that was used in equation
2.20. For the red sideband, also starting in |gi, we’d get

1 X 1 √
Pe = − Pn cos(Ωη nt). (3.38)
2 i
2

Here you can see that for n = 0 no oscillation is happening. That means by doing such an experiment
(repeating multiple times with different pulse durations, always initializing the same motional state) we can
record a Rabi flopping curve and then use this to determine the populations of the motional state. For this
we could fit the model above or simply do a Fourier transform of the data. Examples are shown in figure
3.2. We regularly use this technique to diagnose what is going on with the motion of the ion. Why do we
use the spin for this? Because we cannot directly read out the motional state (or at least it would be much
harder, it is done for electrons on the quantum level by using electrons coupling to the trap electrodes, but
this requires elaborate electronics and I believe is not (currently) possible for the low motional frequencies
that we’re dealing with when trapping atoms). See e.g. [32].
3.3. TRANSITIONS 43

1.0

0.8

0.6
Pg

0.4

0.2

0.0
0 20 40 60 80 100
t (μs)

Figure 3.3: Simulated carrier Rabi flopping using the full LD expansion. The blue trace is for a thermal state
of n̄ = 1, the orange for n̄ = 10, both with η = 0.05.

3.3.3 Limit of approximations

It is worth thinking about what approximations we have made to arrive at these results, and when they break
down. First, in the Lamb-Dicke approximation, we assumed that i) η  1 and ii) η 2 (2n + 1)  1. The first
condition i) can be met by choosing suitable experimental parameters. If it is violated, it is obvious that
truncating the expansion of the Hamiltonian in powers of η is not justified anymore. The second condition
ii) can be fulfilled by sufficiently cooling the ion(s) before an experiment.
More subtly, by picking out only resonant terms, we have also made an additional approximation. For the
first order sidebands, the resonant terms have a strength of h̄Ωη. However, the terms off-resonant by eiωz t
have a strength of h̄Ω. Thus, as the laser power is increased, we will increasingly also off-resonantly drive
carrier transitions, such that we can no longer neglect those terms.

3.3.4 Higher orders, full Lamb-Dicke expansion

We can go higher in the expansion and find more resonances. Looking at the second order we find

h̄Ωη 2 2 −2iωz t
+ (↠)2 e2iωz t )2 + â↠+ ↠â σ̂+ eiωa t e−iωL t + H.c.


ĤLD,2nd = − â e (3.39)
4
There are three transitions here. The first two are the second blue (ωL = ωa + 2ωz ) and red (ωL = ωa − 2ωz )
sidebands. Adding or subtracting two quanta of motion for a spin flip from down to up. Note that these
are now suppressed by η 2 compared to the carrier. One at ωL = ωa - so the carrier frequency might be a
bit puzzling. This is a second-order correction to the carrier transition. If we evaluate the matrix elements
for initial and final motional states |ni, |mi we get the term (2n + 1) |ni hn|, so again it does not change the
motional state but it does change the coupling strength depending on the motional state. In the second order
LD expansion the carrier Hamiltonian is thus

h̄Ω
Ĥc LD 2nd = (1 − η 2 (2n + 1))1m σ̂+ eiδt + H.c. (3.40)
2
This essentially means that the carrier Rabi frequency is weakly dependent on the motional state. This is an
issue when you want to do very high fidelity single qubit gates/spin rotations. When the ion is in a thermal
state that leads to a statistical distribution for the carrier Rabi frequency and reduces the fidelity - a weak
decay in Rabi frequencies would be visible. Simulated Rabi flopping is shown in figure 3.3. This means that
if we want to perform very high fidelity spin operations we need to cool the motion to the ground state (or
use a microwave or co-propagating Raman beams).
44 CHAPTER 3. SPIN-MOTION COUPLING

In general, if we venture outside the Lamb-Dicke regime, a full expansion gives transition operators
 1/2
iη (âeiωz t +↠e−iωz t )
X
0 −η 2 /2 n< ! 0 0 0
e = |n i hn| e η |n −n| L|n
n<
−n| 2 −iωz (n −n)t
(η )e (3.41)
n> !
n,n0

|n0 −n|
where n< is the lower of the occupation numbers n0 and n, and n> is the larger of these two and Ln< (η 2 )
are the generalized Laguerre polynomials.
A summary of various transition Rabi frequencies is shown in Figure 3.4.

Figure 3.4: The relative transition strengths for the carrier and the first few sidebands depending on the
motional state n for η = 0.0465 for which the condition η 2 (2n + 1)  1 corresponds to n  230. The left side
shows the regime where the Lamb-Dicke approximation holds, and the right side shows how it breaks down
for higher n.

3.3.5 Transitions in 3D
What changes if we look at the full 3D motion of an ion? We basically have to evaluate our general Hamiltonian
3.23. I want to point out two important aspects:
When driving a e.g. red sideband transition on the z mode we essentially drive a carrier transition on the
other modes. The fully spelled out Hamiltonian is
h̄Ωηz
Ĥrsb = σ̂+ 1x 1y âz + H.c. (3.42)
2
The same is true for the carrier:
h̄Ω
Ĥc = σ̂+ 1x 1y 1z + H.c. (3.43)
2
This might seem trivial but since we’ve seen that the carrier term has a dependence on the motional state in
higher orders of the LD expansion this can affect operations. These modes and their coupling are sometimes
called ‘spectator modes’. Again, to do very high fidelities we want to either have small LD parameters for the
spectator modes or we also need to cool them to the ground state.
Another effect arises in the 3D case: Intercombination sidebands. We will find resonances with multiple
modes of motion involved e.g. we will see a resonance for ωL = ωa + ωz − ωx , a ‘red-x-blue-z sideband’ with
the resulting Hamiltonian
h̄Ωηx ηz
Ĥx,z = σ̂+ âx 1y â†z + H.c. (3.44)
2
This increases the motion of one and lowers the motion of the other mode. Note that this is suppressed by
ηx ηz .

3.3.6 Photon scattering

What happens if we emit a photon by spontaneous emission? There we can look at the Lindblad operators
for the scattering process. The details were figured out here: [20]. The short version is that when scattering
a photon the probability to change the motion by one quantum is suppressed by approximately η 2 compared
3.3. TRANSITIONS 45

to scattering ‘recoil-free’ - so just de-exciting the two-level system without changing the motion. Again this
is fully analog to the Mössbauer effect and any left-over momentum is absorbed by the trap. Why is it η 2
and not η as for the Rabi frequency? You can trace this back to the Rabi frequency being proportional to
the matrix element/wave function overlap (e.g. dipole matrix element) while the decay rate of a transition is
proportional to the absolute value squared.
46 CHAPTER 3. SPIN-MOTION COUPLING
Chapter 4

Laser cooling

In laser cooling we create a coupling between spin and motion that leads to cooling of the motion. On a
fundamental level, what we’re doing is using the fact that the spin state is in the ground state at room
temperature - there are not enough photons at 300 K to excite e.g. the optical qubit at 729 nm of 40 Ca+ .
So this first means: This spin does not need to be cooled - it just needs to be prepared by optical pumping
(as we looked at in spin state preparation). For the motion this is not the case - in ion traps we have noisy
electric fields that will excite the motion and lead to a thermal state, very far from the ground state of the
motion. The trick is now to couple the spin to the motion in such a way that it ‘swaps’ motional for spin
excitation which can then be easily quenched by repumping. If we do this correctly this leads to an effective
zero Kelvin bath of the motion.

4.1 Doppler cooling

I will not discuss Doppler cooling in detail because it was already covered extensively in Jonathan Home’s
Quantum Optics lecture. There are a couple details for trapped ions that are worth mentioning though.
First, Doppler cooling will usually not get us to the motional ground state. The Doppler cooling is limited
to (theoretically) an energy of h̄Γ/2. With typically dipole transition linewidths used for sideband cooling of
20 MHz this is still an order of magnitude larger than our typical trap frequencies of around 1 MHz, which
will limit the achievable n̄. In most cases Doppler cooling will bring us into the LD regime, making all further
operations using the coherent RSB, BSB interactions much easier.
Doppler cooling in an ion trap is different from the usually covered textbook examples because there is a
trap. If you’re coming from neutral atoms usually the laser cooling beams provide the trapping (a magneto-
optical trap) and need to be set up such that they come from at least 4 directions (6 is more typical). In
an ion trap this is not required. If we first consider one spacial dimension only: the cooling happens if the
ion is moving towards the red-detuned laser beam, if it’s moving away from if no (few) photons are scattered
and no cooling happens. But because the ion is trapped it will simply oscillate back and forth in the laser
beam, getting cooled in one half of every oscillation cycle. But there is no need to introduce another, counter-
propagating laser beam. In 3D we also don’t require three laser beams, we simple need the lasers k-vector to
have a (decent) projection onto all of the three normal modes such that they all get cooled. This principle
is shown in figure 4.1. This however introduces a detail that we need to watch out for: The normal modes
need to be well defined such that these projections are well defined. Let’s assume two of the normal modes
are degenerate. Then we cannot define normal mode directions. Worse, the interaction of the ion with the
cooling laser will break the degeneracy itself because it introduces a (small) force, leading to a normal mode
along the k-vector and one perpendicular to it - exactly the situation we need to avoid because now the
perpendicular mode does not have a projection on the k-vector and does not get cooled (it only gets heated
through the recoil). That’s why we always need to control the directions of the normal modes and have
sufficient splitting between them. What is sufficient? The splitting needs to be larger than the force of the
cooling laser. Typically tens to hundred kHz are enough.

4.2 Sideband cooling

The red sideband can be used to cool trapped ions very close to the motional ground state. In it, we use the
red sideband/Jaynes-Cummings Hamiltonian
h̄Ω
âσ̂+ + ↠σ̂−


Ĥrsb = (4.1)
2

47
48 CHAPTER 4. LASER COOLING

Figure 4.1: In Doppler cooling a mode of motion will only get cooled if the cooling laser k-vector has a
projection on the mode direction. Otherwise it will only get heated by the recoil.

which removes one quantum of motion at the expense of raising the spin and vice versa. However, it alone
can not remove any entropy, since it results in cyclic transitions. We therefore additionally use a laser to
drive a dipole allowed transition from the state |↑i to a state that rapidly decays, as illustrated in Figure 4.2
for the case of 40 Ca+ .

Figure 4.2: The levels of 40 Ca+ relevant for sideband cooling. The state |↑i has a long natural lifetime of
roughly one second, which makes it a good state for a qubit. However, for fast cooling, we need a faster
decay. We therefore couple it to a state with a lifetime of only a few nanoseconds using a laser at 854 nm
to resonantly drive a dipole-allowed transition. The re-preparation of the |↓i state is not shown here, but
required after the 854 nm repumping.

This results in an effective decay of the state |↑i with rate γ. This decay, together with the red sideband
results in an effective evolution as
dρ i γ
= − [ĤI , ρ] + (2σ̂− ρσ̂+ − σ̂+ σ̂− ρ − ρσ̂+ σ̂− ) (4.2)
dt h̄ 2
To further simplify matters, we can adiabatically eliminate the spin degree of freedom1 , assuming that Ω  γ.
This results in
dρm
= Γ 2âρm ↠− ↠âρm − ρm ↠â


(4.3)
dt
2
where Γ = 2Ω γ and ρm is the reduced density matrix of the motional part alone.
The complete process is illustrated in Figure 4.3. The red sideband always drives transitions from |↓, ni to
|↑, n − 1i. Due to the engineered dissipation, this state then quickly decays to |↓, n − 1i, effectively removing
one quantum of motion. This process is repeated over and over again, pumping all the population into the
|↓, 0i state, effectively cooling the ion to its motional ground state.
1 If you’re unfamiliar with this, it’s covered in e.g. [50], page 22.
4.3. IMPLEMENTATION 49

Figure 4.3: Illustration of resolved red motional sideband cooling. Due to the strong engineered spontaneous
decay of the |↑i state, the population is typically in the |↓i state. Driving the red sideband then removes one
quantum of motion at the expense of raising the spin. However, the engineered decay then lowers the spin
again, closing the cycle.

The Lamb-Dicke parameter also plays an important role in resolved sideband cooling. We can write it as
r !2
h̄ (h̄k)2 1 Erecoil
η2 = k cos(θ) = cos2 (θ) = . (4.4)
2mω 2m h̄ω Emotional

Since η  1, this tells us that the recoil energy is much smaller than the energy of a motional quantum. This
is good, because during every cycle of sideband cooling, we have to emit a photon via spontaneous decay,
heating up the ion by the recoil energy. However, since the recoil energy is much smaller than one motional
quantum that we removed with the red sideband, this means that we are cooling after all. The decay along
a sideband transition is suppressed by η 2 compared to the decay that does not change the motion - this is
recoil-free emission, again fully analog to the Mössbauer effect.

4.3 Implementation
SBC can be implemented either continuous, turning on the RSB and the repumping lasers at the same time,
or pulses, driving cycles of discreet pulses of RSB and repump. The continuous version can be very fast but
has the disadvantage that the repumper causes a Stark shift of the qubit levels that needs to be calibrated
and requires the repump laser power to be reasonably stable so that the qubit frequency is stable. Further, it
requires the Rabi frequency ratio of RSB and repumper to be roughly equal, otherwise the process becomes
very inefficient. If the RSB is too strong/the repumper too weak, the process is just slow, on average doing
a ≈ 2π rotation before the spin gets repumped. If the repumper is too strong an interesting bit of physics
happens: The quantum Zeno effect [42]. The repumper essentially continuously measures the spin state, and
if doing so strongly, this inhibits any evolution of the spin, suppressing the sideband cooling process.
In the pulses version of SBC these issues don’t show up which is why it’s the typical method used in the
lab unless speed is of high importance. Here, the repump pulse is just made long enough to repump the spin
to very high probability and the RSB pulse duration is usually adapted through the sequence such that first
π pulses for large n and at the end π pulses for the low n states are implemented. This helps to increase the
efficiency and thus reduce the number of RSB pulses that need to be performed.

4.4 Limits
We will now consider how this is different from Doppler cooling. The key thing is that the linewidth of the
transition. The key feature is the linewidth of the transitions as it compares to the trapping frequency. Most
trapping frequencies are on the order of several MHz. Typically the excited state linewidth is an allowed
dipole transition, which have linewidths on the order of Γ = 2π × 20 MHz. If you try to excite this system
with a laser, you have a very broad line, and all of the sidebands sit under this line. If you look at this system,
what is violated about the conditions that we assumed for sideband cooling?
50 CHAPTER 4. LASER COOLING

I wrote down the decay of the atom Γ, and I wrote down the Hamiltonian. What did I assume when I
wrote down the Hamiltonian? We picked out particular terms from a bigger Hamiltonian. We assumed that
we can choose a particular frequency for our laser such that we can pick out only the red-sideband terms. If I
have a very broad linewidth on my transition, then I am in trouble, because there is no way to address ONLY
the red sideband transitions, ωL = ωatom − ωz because the system doesn’t spend enough time in the excited
state. The effective spectrum that it sees is broadened by the natural linewidth. Nature doesn’t let me couple
to just the RSB Hamiltonian, I have to couple to the ones that remove motion, that increase motion, I get
no specificity.
Sideband cooling only works in a certain regime, one in which the linewidth is less than the trap frequency,
Γ < ωz . This tells us that the transition must be a narrow linewidth transition, and it must be driven fairly
slow, because the Rabi frequency must also be smaller than the trap frequency. If this is not the case the
carrier transition will get excited off-resonantly. With a small LD parameter this will always be the case to
some extend and can limit the final temperature/fidelity of the preparation of the motional ground state.
Another limitation (and if all other things are done well the typical limitation) is the heating rate from
anomalous heating. Eventually, the heating and cooling rates will simply balance and determine the lowest
reachable temperature.

4.5 Cooling diagnostics

Once we have cooled to the ground state we can try to estimate how well we are in the ground state and
measure the motional heating rate by first cooling to the ground state, waiting different times and finally
measuring the motional state. To measure the motional state we could do red or blue sideband flopping and
using a fit or Fourier transform to determine the populated Fock states as described in section 3.3.2. This
would be fairly tedious as it requires a good amount of data. There is a simpler method that we can use
when we assume that the motion is in a thermal state, which is often a good approximation, after cooling
and especially after heating for some time. For a thermal state we can simply measure the contrast of the
red and blue sideband when using the same pulse durations. We would get (preparing the spin state in
|↓i) the spin populations Prsb (↑) for the red sideband and Pbsb (↑) for the blue sideband, with their ratio
R = Prsb (↑)/Pbsb (↑). It can be shown that the average phonon number n̄ is then given by [91]

R
n̄ = . (4.5)
1−R
Obviously, when the motional ground state would be perfectly prepared we had Prsb (↑) = 0, Pbsb (↑) > 0 for
any red/blue sideband pulse duration and get R = 0 and thus n̄ = 0. This is well illustrated by the spectrum
shown in figure 4.4. This is a very useful, fast tool to determine the thermal occupation of a motional mode.

4.6 Sympathetic cooling

We can co-trap multiple ions in the same potential and if the trap frequency is high enough (i.e. the ions are
pushed close enough together) their motion is strongly coupled and does not belong to one or the other ion
but is shared. We can then cool this shared motion using only one of the two ions instead of both. If we had
trapped the same species this would be pointless in most instances, but if we trap different species, they will
have different wavelengths at which we drive their various transitions. That means we can scatter photons
(and thus laser cool) on one species while the internal state of the other species is unaffected by this - it’s
not resonant to the light from the laser or the light emitted by the cooling ion. This can be immensely useful
and in fact is the basis for the major proposal for scaling trapped ion quantum computation and is used for
one of today’s best atomic clocks. It enables us to cool an ion species with another, so if we can’t cool some
ions species because it does not have a suitable electronic structure we can co-trap a useful cooling ion and
still get the ion of interest cold. Also, we can store quantum information in one ion species and (re-)cool the
ions’ motion with another co-trapped ion species without destroying the stored quantum information in the
first species.
The technique certainly has it’s limits: the motion of the ions needs to be couple strongly to have to
cooling work efficiently. The main limit that creates is that we need to use sufficiently similar ion masses
for the different species, since the trap frequencies scale with mass and the coupling gets less strong if the
frequencies are too different. How different is too different depends on the application. Sympathetic cooling
+
was recently demonstrated for mass ratios of 2/9 in our lab (H+ 9
2 and Be ).
4.6. SYMPATHETIC COOLING 51

Figure 4.4: To my embarrassment I don’t remember out of which publication this figure was copied. I hope
the author(s) forgive me this oversight. The spectrum is taken with the same pulse duration, measuring the
spin excitation after the coherent pulse, which frequency is changed to cover the carrier transtion and the first
motional sidebands of one mode of motion. As is nicely shown here (black dots are data, solid line is probably
a fit) the spectrum shows an overdriven carrier, a close to optimal driven blue sideband and no excitation for
the red sideband, because the motion is in the ground state.
52 CHAPTER 4. LASER COOLING
Chapter 5

State dependent force

We already discussed driving the ion’s motion with oscillating electric fields equal to the trap frequency. This
created a coherent state of the motion. Now we want to do something similar using spin-motion coupling and
using a laser as a tool. In the end what we’ll get is a state-dependent force that we can use to entangle two
(or more) ion spins. For now however we discuss the basic interactions on a single ion.
We have two fundamental choices here. We first discuss the the commonly used today and the somewhat
out-of-fashion version after that.

5.1 X-basis force

The more modern version could be called Møllmer-Sørensen coupling, after (some) of the inventors. There
we’re actually several people coming up with this at a similar time.
For this we start by writing down the Hamiltonian for applying a blue and red sideband at the same time
h̄Ωη
Ĥ = (âσ̂+ − ↠σ̂− ) + H.c. (5.1)
2
Here the sidebands have the same strength and are on resonance and I’ve chosen a specific phase for the red
and blue sideband to introduce the minus sign - this will just make things easier later, but you can do the
full math with the phases general to find the general result. We can rewrite this as
h̄Ωη
ĤMS = (â − ↠)σ̂x , (5.2)
2
where we made the simplification of taking Ω to be real. Forgetting for a moment about the σ̂x we see that
the motional term â + ↠will be a displacement once we form the evolution operator with the magnitude of
the displacement given by the duration we apply the Hamiltonian for and the Rabi frequency Ω
     
−iηΩt † ηΩt ηΩt † ∗ †

ηΩt
exp (â − â ) = exp (−i) exp â − â = exp (−i) exp αâ − α â = D̂ α = (5.3)
2 2 2 2
This is a force on the motion.
Now we look at how the σ̂x modifies this. If we write the σ̂x operator in the x-basis we have

σ̂x = |+i h+| − |−i h−| (5.4)

So if apply the Hamiltonian ĤMS  on an ion in state |+i the force will be unmodified, but if we apply it on
the |−i state we get a force D̂ α = − ηΩt 2 , which is pointing into the opposite direction. And that’s it:
state-dependent force.
We can already use this and apply it to an √ ion prepared in its more conventional energy eigenstates, so in
the z-basis. Starting in state |↑, n = 0i = 1/ 2(|+, n = 0i + |−, n = 0i) we get the state
1
√ (|+, αi + |−, −αi) (5.5)
2
which is an entangled state of spin and motion, where the position of the ion (in phase space!, left or right
from the origin), is entangled with the spin state. This is often called a ‘cat state’ because it is quite similar
to Schrödinger’s cat - the position of the ion is basically the cat (dead or alive) - the classical part and the
spin is the quantum part - the radioactive isotope in a superposition state of decayed and not decayed.

53
54 CHAPTER 5. STATE DEPENDENT FORCE

5.2 Z-basis force

To implement a state-dependent force in the z basis we first have to remind ourselves about the light shift.
The light shift of a far-detuned laser beam will perturb an atomic transition as δLS = Ω2 /∆, were Ω is the
Rabi frequency of the laser on the transition(s) that it’s detuned from by ∆. If we chose two S1/2 hyperfine
levels in 9 Be+ and a laser beam of some power, polarization and detuning from the P3/2 levels they will have
different Rabi frequencies - and thus light shifts - depending on the polarization of the beam (and their energy
difference if it’s comparable to the detuning ∆). We can write the resulting Hamiltonian as
ĤLS = h̄σ̂z δLS (5.6)
The trick is now to make a spatially varying light shift by forming a standing wave using two lasers with the
same frequency. The interference pattern of the laser will modulate the Stark shift as δLS,SW = 2δLS cos (∆k · r)
with ∆k = k2 − k1 the differencek-vector of the two laser beams and r the location of an atom in space.
I’ve omitted any offset phase here. We can simplify by assuming the laser geometry is such that ∆k has
only a component along z, so we get δLS,SW = 2δLS cos (∆kz). Since there is now a potential gradient in the
z direction, this will result in a force, and since the light shift is different for our two qubit states we get a
qubit-state dependent force which we could write as
ĤLS,SWF = h̄2δLS cos (∆kz)σ̂z ẑ (5.7)
This operator z is not in the rotating frame, right now that force is static, so it will not do anything to our
trapped ion except offset it’s position. We want however a force oscillating at the motional frequency, to be
able to excite the motion. This can be easily achieved by modulating the standing wave. The probably easiest
way to do this is create a ‘running-standing wave’ - for this we detune one of the laser beams by the ion’s
trap frequency ωz from the other. The light shift becomes
δLS,RSW = h̄2δLS cos (∆kz) cos (ωz t) (5.8)
Going into the interaction picture with respect to the motion we finally get
ĤSDF,z = h̄2δLS cos (∆kz)σ̂z (↠+ â), (5.9)
which is now a state-dependent force in the z basis. Again, I left out any detunings from the motion, which
can easily be put in.

5.3 Geometric phase

We can now play games with these forces. We can, as mentioned, form Schrödinger-cat-like states, but we
can also just use the force like we used the excitation of a resonant electric field on an ion - to displace the
ion. For this we have to initialize the spin state in an eigenstate of the force’s basis. In any case what we
can do now is implement a displacement operator (state-dependent or independent) D̂(α). I want to now
point describe how we can use this to accumulate a phase, a so-called geometric phase. Assume we do two
displacements D̂(α) and D̂(β) subsequently. Then we can write the total displacement as
∗
D̂(α)D̂(α) = D̂(α + β)eiIm(αβ )
(5.10)
You can derive this with using the Baker-Campbell-Hausdorff formula. Let’s assume we do four displacements
in a row with α1 = α, α2 = iα, α3 = −α, α4 = −iα, so going along a square in phase space with edge length
α. For simplicity we assume α to be real. The resulting displacement should add to zero:
2 2 2
D̂(α1 )D̂(α2 )D̂(α3 )D̂(α4 ) = D̂(α + iα)e−iα D̂(−α − iα)e−iα = D̂(0)e−2iα . (5.11)
So, not quite zero - the displacement is zero, but we picked up a phase factor φ = −2iα2 . This is the geometric
phase. Right now it’s of no use - there is no system (of interest) to compare to that did not pick up this
phase. We’ll see however that this geometric phase is essential for implementing a geometric phase gate to
entangling two ion spins. We could have also gotten a geometric phase by using an oscillating electric field,
but since those displacements are not state-dependent we can’t really do much with them. This step-wise
way or creating a geometric phase (and using it for entangling ions) goes back to Milburn [67, 68].
Instead of creating this geometric phase in this step-wise manner we can also simply introduce a detuning
δ from the motion. Looking at this in with the MS-Hamiltonian (but it’s really analog in the LS Hamiltonian)
we get
h̄Ωη
ĤM S = (âe−iδt − ↠eiδt )σ̂x . (5.12)
2
5.3. GEOMETRIC PHASE 55

When we turn on this force, the ion will describe a circular trajectory in phase space. The loop will close for
a duration t = 2π/δ and the accumulated geometric phase would be the area of the circle. You’ll look at this
in detail in the problem set.
56 CHAPTER 5. STATE DEPENDENT FORCE
Chapter 6

Spin-spin entanglement

The initial though for entangling two ions was to use discreet sideband operations - the resulting entangling
gate is the Cirac-Zoller gate, after it’s inventors [21]. It was immensely important to bring trapped-ion
quantum computing of the ground - it was however almost never used since it has severe limitations. What
was and is still used a lot are geometric phase gates instead. There are two versions, one that is acting in
the spin x basis, which is the famous Mølmer-Sørensen gate and another which is sometimes called the light
shift gate, which acts in the z basis. Well first cover the Cirac-Zoller gate, then the geometric phase gates,
concentrating on the MS gate.
The fundamental goal of all theses gates is to implement a spin-spin interaction between the two ions.
But since the spins couple directly only very weakly, the approach here is to use the motion to mediate this
interaction. In the Cirac-Zoller gate this can be seen very clearly because of it’s discreet nature, in the MS
gate it’s more a transient effect. In both cases the motion is used as a ‘quantum bus’ to couple the spins. The
motion of trapped ions is in general not as stable as the spin. Coherence times for the spins can be far longer
than for the motion which is affected by motional heating and trap frequency fluctuations. So when using
the motion as a bus we have to keep in mind that we ideally don’t want to keep information in the motion
for very long. So in principle we should make sure we’re fast with these ‘quantum bus’ operations and that
no residual entanglement between motion and spin remains once the operation is done.

6.1 Cirac-Zoller gate

For the Cirac-Zoller gate we assume we have a three-level system in our ion, which we can control coherently.
We’ll call the three levels |↓i, |↑i, |Ai. Our goal is to implement a phase gate between the qubit states |↓1,2 i,
|↑1,2 i of the two ions, using the auxiliary level |A2 i and one of the ion’s shared motional modes with states
|n = 0, 1, ...i. The unitary for a two-qubit phase gate is given by
 
1 0 0 0
0 1 0 0
Ĝ = 
0 0 −1 0
 (6.1)
0 0 0 1

acting on the two-qubit state-vector {|↑1 ↑2 i , |↓1 ↑2 i , |↑1 ↓2 i , |↓1 ↓2 i} We assume the ions’ motion is initialized
in the ground state |0i and general qubit states α |↑1 i + β |↓1 i and γ |↑2 i + δ |↓2 i are encoded in their electronic
states. The procedure goes as follows

1. The initial state of the full system is |ψ0 i = |0i (α |↑1 i + β |↓1 i)(γ |↑2 i + δ |↓2 i)

2. Transfer spin 1 into motion using RSB π pulse on ion 1 (|↓1 , ni ↔ |↑1 , n − 1i):
|ψ1 i = |↓1 i (α |1i + β |1i)(γ |↑2 i + δ |↓2 i)

3. To get a -1 phase perform 2π rotation on red sideband of auxiliary transition of ion two (|↓2 , ni ↔
|A2 , n − 1i):
|ψ2 i = |↓1 i (αγ |↑2 i |1i − αδ |↓2 i |1i + βγ |0i |↑2 i + βδ |0i |↓2 i)

4. Transfer motion back to spin 1 with RSB π pulse on ion 1:

|ψ3 i = |0i (αγ |↑1 i |↑2 i − αδ |↑1 i |↓2 i + βγ |↓1 i |↑2 i + βδ |↓1 i |↓2 i)

This results in the wanted operation. Note that we used the motional state, but at the end of the operation
the motion is not entangled anymore with the spin.

57
58 CHAPTER 6. SPIN-SPIN ENTANGLEMENT

Why was the Cirac-Zoller gate not really used? For one thing: It requires single-ion addressing which was
challenging, but to day is done quite often. So why not use it today? Because the fidelity of the operation
is directly linked to the fidelity of the motional ground state initialization. Since the motion can be cooled
to a thermal state of only ≈ 0.01 quanta (and this means the ground state is populated with a probability of
≈ 0.99) this set the limit on the fidelity we can achieve. We will see that for the MS gate things scale much
better.

6.2 Mølmer-Sørenesen gate

The MS gate simply turns on the MS Hamiltonian with a small detuning δ from the motional mode we use
(this is the same detuning I described in the Geometric Phase section 5.3. Note that this means we have
to put in a red sideband with ωL,rsb = ω0 − ωz − δ and a blue sideband with ωL,bsb = ω0 + ωz + δ. The
detuning δ can be positive or negative but has to applied as described. With the detuning on we again are
making circles in phase space, returning to the origin at times t = 2π/δ and accumulating a geometric phase
proportional to the area of the circles. If both ions are in the same x eigenstate the only thing that changes
from the single-ion description is that the force is twice as strong (we’re driving the motion with two ions
instead of one). If the two ions are in different x eigenstates the forces cancel, no displacements happens and
no geometric phase gets accumulated.
You will derive the propagator for the MS gate in the problem set - I’ll just give it here as a result:

ÛMS (t) =D̂(α(t)(σ̂x ⊗ 1 + 1 ⊗ σ̂x ))e−i2β(t)(1⊗1+σ̂x ⊗σ̂x ) (6.2)

iΩ −iδt/2
α(t) = − e sin (δt/2) (6.3)
δ
 2
Ω
β(t) = (δt − sin (δt)) (6.4)
2δ

Here Ω ≡ Ωrsb/bsb . We see that there is a displacement part α(t) and a phase that is accumulated depending
on β(t). We now have to tune the parameters t δ and Ω correctly to get what we want. First, we should tune
the displacement such that after the MS gate the motion is back to were it started, so α(t) = 0. This happens
at t = 2π/δ and multiples of that. This results in
 2
π Ω
β= (6.5)
2 δ

So for the accumulated phase (for states |+, +i, |−, −i) we have
 2
Ω
φ = −2π (6.6)
δ
√
We see that this is the area of a circle with radius r = Ω/( 2δ). Setting Ω = δ/2 we get the classic MS
operation

ÛMS |+1 , +2 , ni → − i |+1 , +2 , ni (6.7)

ÛMS |+1 , −2 , ni → |+1 , −2 , ni (6.8)
ÛMS |−1 , +2 , ni → |−1 , +2 , ni (6.9)
ÛMS |−1 , −2 , ni → − i |−1 , −2 , ni (6.10)

Or written for only the spin states in the x basis

 
−i 0 0 0
0 1 0 0
ÛMS =
0 0 1
 (6.11)
0
0 0 0 −i

This is not identical to the Cirac-Zoller (phase gate) that we looked at before (or the classical cNOT) but
again this is a universal two-qubit gate and adding some single qubit rotations can be turned into either of
those two. Importantly, it is independent of the motional state. You will derive this in detail in the problem
set. What I’ve sketched here (and what you’ll derive in the problem sets) is only valid in the LD regime. The
independence on the initial motional state is not given outside of the LD regime.
6.2. MØLMER-SØRENESEN GATE 59

We will now write the MS gate in the z basis and look at its behaviour there. In the z basis ({|↑↑i , |↑↓i , |↓↑i , |↓↓i})
the MS gate is given by
 
1 0 0 1
1 0 1 −i 0
ÛMS = √ eiπ/4  0 −i 1 0 .
 (6.12)
2
1 0 0 1

We can now immediately see that this is an entangling gate. If it is acted on |↑↑i we get out the Bell state
(|↑↑i + |↓↓i)/2. The full evolution is shown in figure 6.1.

Figure 6.1: Data and fits for the evolution of two ions prepared in |↑↑i. The x axis is the MS interaction
duration, the y axis shows the different spin populations. Black: P (↑↑), red: P (↑↓ + ↓↑), blue: P (↓↓). The
MS gate time would here be at ≈ 50 µs where we get the Bell state (|↑↑i + |↓↓i)/2.

6.2.1 Error sources

When performing the MS gate of course all parameters need to be calibrated correctly, and because t, δ and Ω
are linked this is actually a bit tedious. What can happen if their wrong? We could pick up the wrong phase
for the |++i and |−−i states - this would mean we’re implementing a different gate. We could also either
not close the loop in phase space or overshoot and start the next - this would leave the motion entangled
with the spin. That’s bad because the motion usually has a shorter coherence time than the spin and we
thus get decoherence (and of course it would be a nightmare to take the rest of the populating in he motion
into account... Of course we can try to do a perfect job in calibration, the issue in the end will not be our
calibration but the noise in our experiment. Error sources that can be explained easily in our understanding
of the MS gate are: fluctuations of the motional frequency and the Rabi frequency. A fluctuation of the
motional frequency is equivalent to a wrong detuning δ, so will cause closing-the-loop errors. Fluctuations of
the Rabi frequency will lead to the wrong accumulated phase.
60 CHAPTER 6. SPIN-SPIN ENTANGLEMENT

Other error sources are a bit harder to grasp intuitively. Among them are qubit frequency fluctuations
(spin decoherence) and motional heating.
Another error source would be that we violate the initial assumptions: That we are in the LD regime
and that we drive only the sideband transitions. I will get later to workarounds for violating the LD regime.
Excitation of the carrier transition (not only the sidebands) is quite common. This is simply cause by the
much stronger Rabi frequency of the carrier than the sidebands. The standard way to counter this issue is
to perform pulse shaping. Here the turn-on and turn-off of the MS gate pulse is done smoothly instead of
using a square pulse. This will suppress higher frequencies in the frequency spectrum of the pulse (Fourier
transform of the pulse shape).

6.2.2 Two-loop MS gate

I’ll discuss now a popular variation of the MS gate that can help us suppress some errors. To make the MS
gate work we picked a specific combination of parameters: t = 2π/δ and Ωrsb/bsb = δ/2. This was to ensure
that we pick up a phase −i for the spin states |++i and |−−i and that the loop in phase space is closed after
the operation and thus the spin is not entangled with the motion anymore. We could have made a different
choice (and the original MS proposal did so) - we could do more than one loop to accumulate the required
phase instead of just one. This is what we typically call multi-loop gates. The simplest version is obviously
a two-loop gate. For this we set t = 4π/δ - this gives us two loops and we now have to figure
√ out what to
set Ω to so that we still get the correct phase. From equation 6.6 we find that Ω = δ/(2 2). What is the
advantage of using two loops? There are several:
• The loops in phase space will be smaller - so we excite to lower n states - and that means we don’t have
to full the LD approximation as rigorously.
• We can put a spin-echo pulse between the two loops (doing one MS pulse for the first loop, doing a
pi-pulse on the spins, then the second MS pulse for the second loop). This can reduce spin decoherence.
• We can swap some phases in the laser pulses for the second loop so that the second loop goes into the
other direction in phase space compared to the first. This can help with loop-closing errors.
Let’s have a look at the phases on the MS gate. The phases of the MS gate are the phases of the RSB
and BSB laser tones φRSB and φBSB . We can reformulated these phases as
φRSB − φBSB
φm = (6.13)
2
φRSB + φBSB
φS = (6.14)
2
where φm we call the motional phase. When we swap that by π the force in phase space points into the
opposite direction. We can use that to swap the direction of the second loop of the MS gate. This will still
accumulate the phase positive, so adding to the first loop, but in case we did not close the loop in the first
loop (because we drove too long or short), we now drive back along the same path in phase space in reverse
and end up at our starting position. This gets rid of the loop closing error. This technique is the same as
using spin echo on the spin - just here on the motion. It’s sometimes called ‘motional echo’ but also ‘Walsh
pulse’ - details on the technique can be found here [34]. φS is the spin phase and can also be swapped to
produce the same effect, but on the spin - so a spin echo, in case you have an offset frequency between the two
ions (could be caused by a magnetic field gradient) or spin decoherence. More on the phases in entangling
gates can be found here [57].
You will look at these techniques also in the problem sets.
What is the price we paid for these two loops? Either the full gate duration will be longer or we have to
use a larger detuning and higher Rabi frequency (so we need more laser power). This can of course introduce
more errors or make certain errors impact us stronger. In the experiment we’d have to study in detail what
implementation gives us the best gate - depending on our error sources.

6.2.3 Light shift gate

The light shift gate (sometimes also called ‘wobble gate’) is also a geometric phase gate and uses the motion
much in the same way as the MS gate. The first implementation was demonstrated here [59]. The big
difference is that it does not create the state-dependent forces in the x basis like the MS but in the z-basis
by using state-dependent ac Stark shifts. It is performed with two laser beams with frequencies ω1 = ω0 − ∆
and ω2 = ω0 − ∆ + ωm + δ, respectively, where ω0 is (usually) the transition frequency of a dipole transition
from which the beams are detuned by a large detuning δ such that the upper-state excitation is minimal and
6.2. MØLMER-SØRENESEN GATE 61

only a light shift is created on two ground state qubit levels. The polarization of the two beams are set such
that the differential shift of the qubit levels is maximized. As with the Raman beams we discussed before,
the beams are arranged such that their difference k-vector has a projection on the motional mode that is
used for the gate. If that would be all we’d have a standing wave causing a position-dependent Stark shift, as
depicted in figure 6.2. Depending on where the ions would sit in the beam and depending on their internal
state they’d get pushed to one side or another by a certain constant amount. But since the two beams have
a relative detuning that matches the frequency ωm of one motional mode of the two ions plus the small gate
detuning δ, the standing wave oscillates with this frequency and drives the motion (close to) resonantly. This
will cause the same loops in motional phase space as for the MS gate and again the gate parameters are tuned
such that we get a phase i after the loop closes. But on what states? As mentioned the gate acts on the
z basis, because that is all the Stark shift does. We can now tune the distance of the ions (by tuning the
trap frequency) such, that we get the forces on the two ions have the same magnitude and either point in the
same or opposing directions. If they point into the same direction they we can excite the COM mode, if they
oppose each other we can excite the stretch mode.
We now still have to deal with the different forces resulting from different spin states. As mention, they
can be tuned by the polarization of the beams and are usually set up such that for, say, the |↓↑i and |↑↓i we
get the required phase but for |↓↓i and |↑↑i we get the identity.

Figure 6.2: Cartoon of the state- and position-dependent force created by a standing wave as used in the light
shift gate. In reality the forces for the two different qubit states will not be pointing in opposing directions -
they just will have different strength but will point in the same direction if the qubit states are encoded in
the hyperfine manifold.
62 CHAPTER 6. SPIN-SPIN ENTANGLEMENT
Chapter 7

Quantum computation

The traditional quantum information processing with trapped ions uses the electronic states for the infor-
mation storage and computation, most of the time simply using qubits, so two-level systems. We’ve already
covered at length how to coherently control the two level systems that are available to us in the usual ion
species that we use. For quantum computation we need full control over these two level systems: preparation,
readout, universal coherent control in form of a set of universal operations - in our case single qubit rotations
and the MS- or light-shift-gate. To build a universal quantum computer we will need a lot of qubits, meaning
we have to scale our few-ion experiments to much larger size. Currently its hard to say what size is required
for a universal, error-corrected quantum computer, but save to say that the number is (probably far) beyond
100 qubits. The details are covered in other courses.
We first discuss some minor aspects of trapped-ion quantum information (scaling), then look at the main
approaches to scaling trapped ion systems.

7.1 Some notes on trapped ion quantum information processing

7.1.1 Two level control / single qubit gates
We’ve already covered two-level coherent manipulation. I just wanted to point out how to efficiently perform
single-qubit gates for trapped ions. First are operations that change the populations in the energy eigenbasis.
On the Bloch sphere those correspond to rotations around x or y. A notation common in trapped ions is this

R̂(θ, φ) = cos (θ/2)1 − i sin (θ/2) cos (φ)σx − i sin (θ/2) sin (φ)σy (7.1)
−i sin (θ/2)e−iφ
 
cos (θ/2)
=
−i sin (θ/2)eiφ cos (θ/2)

Here θ is the amount of rotation and φ is the angle that determines around which axis the rotation is
performed. A π/2 rotation about the x (y) axis would be given by θ = π/2, φ = 0 (φ = π/2). This is a useful
notation because it decouples the natural control knobs that we have. We can directly connect them to the
physical quantities from eq. 2.2: θ = 2Ωt with t the duration of our pulse. The phase φ is identical in both
equations. Here we of course assume that our rotation is fully on resonance, so δ = 0 (no detuning of the
laser from the qubit transition).
If we want to implement a phase shift of our qubit we could use a light shift/AC Stark shift/AC Zeeman
shift. You get a light shift of δLS = Ω/(4δ) if Ω  δ. Then this would implement

e−iφ/2
 
0
R̂z (φ) = e−iφ/2 (1 + σz )/2 + eiφ/2 (1 − σz )/2 = (7.2)
0 eiφ/2

with φ = δLS t and again t is the duration of the pulse. There is however a much simpler solution. A useful
identity is

R̂(θ, φ)R̂z (φ0 ) = R̂z (φ0 )R̂(θ, φ − φ0 ) (7.3)

This means we can simply ‘account’ for the phase shift by shifting all following operations accordingly. At
the end of a quantum computation computation we’ll read out the populations of the qubit and the phase
shift does not matter for those, so any phase shift immediately before any measurement we can simply drop.

63
64 CHAPTER 7. QUANTUM COMPUTATION

7.1.2 Single-shot detection

It is worth emphasising that for (traditional) quantum error correction, ancilla measurements (so detection
of the spin state of some ions but not all of them) and feedback conditioned on the outcome during the
computation is required. This readout needs to be ‘single-shot’, i.e. we need to have a high-fidelity result
of the ancilla ion’s state (|↑i or |↓i) after only one detection. This is possible with trapped ions by simply
choosing a photon count threshold to separate bright and dark histograms for a single ion.
One issue is that these measurements happen during the computation, so some ions still carry quantum
information that is not to be disturbed. As we scatter photons of the ancilla ions (and turn on a laser to do
so) scattered photons might hit other ions if they are close enough. One could simply separate the ions one
wants to read out from the other ions so that the probability of causing errors is very low. This could be done
by transporting them away.
There is however another, more elegant, solution: We can use two different ion species for computational
ions and ancillas. This way the wavelength at which the ancillas scatter photons is not resonant with transi-
tions of the qubit ions and they are not disturbed. We will also see that this mixed-species approach is useful
for other issues in TIQIP.

7.1.3 Motional heating

When long computations are performed re-cooling of the motion during the computation will be required.
Otherwise we would not be in the LD regime anymore and the MS gates would not work (well). But cooing
faces the same issue as detection of ancilla ions - we need to scatter photons. Again one solution is to use
a different species of ions for cooling to allow cooling while keeping quantum information intact in the other
species. Of course the qubit and cooling ions need to have decent Coulomb coupling so that the cooling of
the cooling species also cools the qubit ions’ motion.

7.1.4 Imaging and lasers

To focus laser beams to get small laser spot to address ions individually (and safe laser power) and to image
ions efficiently (so collect many of the photons an ion emits during detection) so that detection is fast and
high fidelity either the emitter/detector needs to be close to the ion or it needs to be large, simply because
a tightly focused laser beam expands rapidly (Gaussian optics will tell you that) and the emitted light of
an ion will simply spread over all of the 4π solid angle. In the past this meant that outside of the vacuum
chambers that hold the ions very large and complex lens systems were mounted to address and/or image few
ions. These lens systems of course only work very well on the optical axis and can only to some extent be
optimized to also work off-axis. Eventually the field of view is always limited - meaning ions to far from the
optical axis cannot be imaged cleanly or addressed cleanly. Several groups are already working on integrating
optics and detectors into ion traps or have them close to the ion traps. Integrated photonic waveguides have
been utilized to address ions [64], and detectors have been integrated into traps, too [83]. More detail in the
lecture slides.

7.2 Approaches to scaling TIQI

7.2.1 Long strings
We first discuss the approach I would call ‘long strings’ or ‘long chains’. Here the idea is just to add more
ions to a single trapping potential. It mostly goes back to [94]. To perform single- and two-qubit gates on
the ions (and read them out and prepare the ancillas individually), we need to be able to singly address them
with laser beams. Going two a large 3D Coulomb crystal will make this impossible. The idea here is to form
one long string of ions in the axial direction of a linear Paul trap and address the ions with tightly focused
laser beams from the radial direction.
To form a long string of ions we need to keep the axial confinement sufficiently low compared to the radial
- otherwise the strong collapse to a 2D or 3D crystal. This will get harder the longer the string is when
using a harmonic potential for axial confinement because the inner ions get pushed together harder and will
pop out radially as we add more ions. It’s possible to calculate the required confinement for a linear string:
ωx,y /ωz > 0.77N/ log N , where ωx,y (ωz ) are the radial (axial) trap frequencies and N is the number of ions
[94].
Because the ions get pushed together strongly in the center of the string it gets very hard to singly address
them with high fidelity. We would be off much better with a constant ion-ion spacing. This is possible
by creating a (slightly) anharmonic potential [94] which is now commonly done in experiments. The ideal
7.2. APPROACHES TO SCALING TIQI 65

potential required is then approximated to give a sufficient result by expanding it to 4th or 6th order and
implementing that with a segmented trap with 5 or 7 electrode segments, respectively. It is also possible
(and done) to use some ions at each of the string that are not used for computation but just to compress the
string. Only the inner ions will then have a sufficiently equidistant spacing as suggested in [94]. (The cooling
of these outer ions as suggested in [94] is not performed and I think is anyway unrealistic.)
MS gates are usually using the radial modes of motion in the long-string approach. This has two reasons:
The axial motion will have very low frequencies and thus very high heating rates which would lead to too
bad gate fidelities. The individual addressing of the ions is best done with laser beams perpendicular to the
axial direction which leads to no coupling to the axial motion, only the radial. Radial modes however bunch
together strongly with too low frequency-splitting between them to address only single modes of motion for
reasonable MS gate times (axial modes have an equidistant, large spacing, this issue is also discussed in more
detail in [94]). This means that the gates on radial modes need to be performed on many modes at once with
the difficulty that all the loops in phase space for the different modes also need to close at the same time.
This can be achieved by complex amplitude or frequency modulation of the gate pulses but adds considerable
complexity (see e.g. [54] for a recent implementation and references therein).
These are the basics of the long string approach. There are several issues however that will likely prevent
this approach from being sufficient for the construction of a full quantum computer from just one single,
extremely long string of ions.
As more ions get added to the string we get more motional modes (we get 3N modes for N ions). These
will have to be cooled and they act as spectator modes. Cooling can be performed using cooling techniques
that are broadband and fast but get to low temperature - resolved sideband cooling wound not work. EIT
cooling and polarization-gradient cooling can be used to achieve this [71, 56, 25, 45, 46]. Even with this
additional cooling the system will get much less clean than what we’re used to from few-ion systems and it
will be very difficult to keep high gate fidelities with large numbers of ions.
Since the ions are very close together long strings will require a second species for ancilla-readout and
cooling during the computation.
The nice thing about the long string approach is that it is relatively ‘easy’ (compared to other methods)
to get to number of 10 ions, and has allowed a lot of nice progress and demonstrations in the past years.
It’s currently not clear to what number this can be pushed before gate fidelities or other issues become too
limiting. While the required ion traps are fairly simple, a lot of work is required for the single-ion addressing
and readout and also the characterization of the now much-less clean system.
A recent result from a IONQ demonstrated 11 qubits in a long chain with two-qubit gate fidelities of
95%-99% [93].

7.2.2 QCCD architecture

The so-called qccd (quantum charged coupled device) architecture goes to the other extreme of the long
string approach. It was proposed by D. J. Wineland and coworkers [91, 49]. Here the number of ions in
a single trapping potential is kept minimal - roughly two computational ions and two cooling/ancilla ions
in one potential well max. This is enough to do 1 MS gate on two ions, and has the advantage over the
long-string approach that the system is very clean and thus you get very good gate fidelities. To scale the
system the ions are shuttled through a large trap array structure and recombined with other ions to again
do the required operations. In principle then no single-ion addressing is needed. Here the complexity lies in
the trap and transport part - which is also hard. The qccd architecture is in some sense the ’gold standard’
of scaling TIQI, it is however extremely challenging, mostly because our capabilities of building precise traps
with complex structures is limited (but improving) and because surfaces charge up statically over time (often
from the laser we use), causing changing electric fields and curvatures that require compensation.
The long-string approach and the qccd architecture are not mutually exclusive. It is possible that a large-
scale device might use aspects of both approaches. Potentially having longs strings for some computation
block that afterwards get split up and reconfigured/redistributed in a qccd device.
A recent, very nice demonstration of the qccd architecture was performed by Honeywell, one of the
companies now getting into trapped ion quantum information [74].

7.2.3 Photonic links

The third approach I’d like to discuss are ’photonic links’, originally proposed here [24]. Here ions in separate
traps are entangled by a probabilistic operation and combined with quantum teleportation this can be used
to do computation. This approach is seldom seen as a sufficient approach to scaling TIQIP but will likely
be required to both get the long strings and even the qccd approach to scale to useful sizes. The idea then
66 CHAPTER 7. QUANTUM COMPUTATION

Figure 7.1: (Some) electronic levels of the ytterbium ion.

would be to build a long string or qccd apparatus to its maximum size and then connect multiple copies by
photonic links.
For the photonic link approach we need do discuss some new concepts because it uses techniques that we
have not discussed thus far.

Remote entanglement

Our first goal is to create entanglement between two remote ions. Let’s assume we have the electronic level
structure as depicted in figure 7.1 and we prepare the state 2 P1/2 , F = 0 for both ions. We then use a laser
to excite both ions to the 2 P1/2 , F = 1, mF = 0 state at the same time. The ions get excited and decay
again to one of the four ground states. By doing so, each ion emits a photon. Let’s look at the cases were
the ions emit a σ̂+ or a σ̂− photon (which will have somewhat different frequencies too, due to a magnetic
field splitting the levels), decaying to the 2 S1/2 , F = 1, mF = ±1 , which we’ll label |↑i1,2 , |↓i1,2 . Each ion’s
electronic state is then entangled with the polarization (and frequency) of its emitted photon (which we label
|+i1,2 , |−i1,2 ). The states of the ion-photon pairs are

1
ψ1,2 = √ (|↑, +i + |↓, −i). (7.4)
2

We can set up optics such that the two photons overlap on a 50:50 beam splitter. A beam splitter has two
‘modes’ - horizontal and vertical - with each an input and an output port (which we’ll label 1,2 for the input,
3,4 for the output). The typical optical device (and the labels we’ll use) is depicted in figure 7.2. The beam
splitter interaction of course appears also in other physical system, not just optics (also in the motion of
trapped ions) and will have the same quantum description. A photon (or laser beam) of mode 1 entering the
beam splitter will be partially coupled to mode two and the final state will be a superposition of the photon
exiting the output ports of both modes of the beam splitter. How does a beam splitter work in the quantum
regime? The beam splitter Hamiltonian is given by

h̄ †
ĤBS = (â â2 + â1 â†2 ). (7.5)
2 1
As you can see it swaps populations from a mode 1 to a mode 2 and vice versa. To get how the beam splitter
changes out photon states we need to get the evolution operator for the beam splitter - which would say we
need to know for what duration the beam splitter Hamiltonian is applied. Of course with a real beam splitter
(for instance a half-silvered mirror) we can’t choose that, we can only choose the thickness of the silver coating
which determines the reflectivity and thus the coupling strength
√ of the two mode. For the required 50:50
beam splitter the product of time and coupling strength is 2 with the resulting evolution operator
 
i
ÛBS = exp √ (â†1 â2 + â1 â†2 ) (7.6)
2
7.2. APPROACHES TO SCALING TIQI 67

(a) (b)

Figure 7.2: Optical beamsplitter. (a) Photo from Thorlabs. (b) Drawing with input and output ports.

To understand what these do to our single photons (Fock states) we first use them to transform the creation
and destruction operators
ÛBS â†1 ÛBS
†
= â†1 + iâ†2 (7.7)
ÛBS â†2 ÛBS
†
= iâ†1 + â†2 (7.8)
Note the phase shift. A general Fock state entering the beam splitter (on both input ports) can be written as

|ni1 |mi2 = |n, mi = (â†1 )n (â†2 )m |0, 0i (7.9)

After the beam splitter this state transforms to
ÛBS |n, mi = ÛBS (â†1 )n ÛBS
†
ÛBS (â†2 )m ÛBS
†
ÛBS |0, 0i = (ÛBS â†1 ÛBS
† n
) (ÛBS â†2 ÛBS
† m
) ÛBS |0, 0i (7.10)
In our example we have two photons, one entering on each input port, so their initial state is given by |1, 1i.
Plugging this into our formula we get the classic result
ÛBS |1, 1i = |2, 0i + |0, 2i (7.11)
which you have probably seen before - photon bunching, or Hong-Ou-Mandel (HOM) effect - the photons
leave through the same port. One property that still distinguishes us from the case we want to look at is
that for the HOM effect to happen, the photons need to be identical - so in our case they need to have the
same polarization! If they have different polarization they don’t interact and no photon bunching happens.
we now put detectors after the output ports and look at the clicks we get from them. The detectors will not
make a difference if one or two photons hit them at the same time (they’ll only click once). In case we have
identical photons, they bunch and we get one click from one of the detectors. These events we’ll throw away.
So what does the state look like for cases where the emitted photons are different and pass the beam splitter
(this time we also look at the ions’ states)? The initial state is
1
|ψi = √ (|↑, +i1 |↓, −i2 + |↓, −i1 |↑, +i2 ) = |↑↓i |+i1 |−i2 + |↓↑i |−i1 |+i2 (7.12)
2
where the mode and ion labels are identical. This is the original state, just with the cases removed where the
ions emitted photons with the same polarization. Once the photons passed the beam splitter we get
1
|ψ 0 i = √ (|↑↓i (|+i1,3 + i |+i1,4 )(i |−i2,3 + |−i2,4 ) + |↓↑i (|−i1,3 + i |−i1,4 )(i |+i2,3 + |+i2,4 ) (7.13)
2
where I kept the original ion labels (1,2) and added labels
√ for the output ports√(3,4). We can sort the terms
in singlet and triplet spin states |Si = (|↑↓i − |↓, ↑i)/ 2, |T i = (|↑↓i + (|↓↑i)/ 2 and find
|ψ 0 i = |Si (|+1,3 −2,4 i − |−1,4 +2,3 i) + i |T i (|+1,3 −2,3 i |+1,4 −2,4 i). (7.14)
68 CHAPTER 7. QUANTUM COMPUTATION

Figure 7.3: The basic circuit for state teleportation. The state of the first qubit gets teleported through by
use of an entangled pair as a resource to qubit three. The double lines indicate classical communication.
Depending on the measurement outcomes for qubit 1 and 2 the final two single-qubit operations on qubit 3
are conditioned.

The triplet state appears again with both photons in one arm - so we can’t distinguish it from the one-click
events that we got before - and thrown out. So these go out too. The singlet state however appears with
one photon in each arm. This is our heralding signal! When we see simultaneous clicks in the detectors we
know we have a singlet state of the ions, which is an entangled state. In the lab we simple repeat the protocol
until we see this outcome. We now created remote entanglement. This is however not in itself useful for
computation yet. By combining this with quantum teleportation we can use it to do computation.
The here described scheme is only one of many - all work roughly similar but might have better (or worse)
success rates. The key thing to remember here is that the process is probabilistic. Only in some instances
we get entanglement that we can use. On top of that you’ll have technical limitations - not all the photons
the ions emit can be collected, the detectors will not have unit quantum efficiency. A very nice, recent work
from the Oxford ion trap quantum computing group gets to a Bell state fidelity of 94% with a rate of 182 s−1
(success probability 2.18 × 10−4 - so they try very often and fast) [82].

Teleportation
In quantum state teleportation we transfer a quantum state from one qubit to another using an entangled
pair of qubits as a resource. This has many applications. Here it would be used to copy a state of a qubit
from one trap to another through the photonic link. The circuit is shown in figure 7.3. State teleportation is
somewhat inefficient in the context of computation and a more efficient approach is gate teleportation [30],
where a gate operation is performed between two remote qubits using an entangled pair as a resource. Gate
teleportation has recently been implemented for the first time with trapped ions here [90] ;) and already
improved here [93].

7.3 Further technical aspects and solutions for scaling TIQI

Please refer to the TIQIP 2 slides.

7.4 Quantum error correction with trapped ions

Please refer to the slides (for now).
Chapter 8

Quantum metrology

Metrology is (strictly speaking) the “the scientific study of measurement” (Wikipedia etc.) and establishes a
system of physical units. It basically goes back to the French revolution and the precursors of the ‘système
international’ (SI) or ‘International System of Units’. We will not be looking into this, suffice to mention that
the SI is currently being re-defined to base all units on fundamental physical quantities. Of course the atomic
clocks that we’ll discuss provide the definition of the second.
However, the term ‘metrology’ is used in much broader context today, where it mostly refers to measuring
some constants of nature. This could be particle masses, electronic transition frequencies of atoms, or exotic
properties like a dipole moment of the electron.
The term ‘Quantum Metrology’ is even more loosely used. It basically refers to preforming metrology in
the quantum regime (whatever that means).
I would like to distinguish two aspects that I call ‘quantum assisted metrology’ and ‘quantum enhanced
metrology’, were the first describes measurement techniques that enable us to measure better or at all using
techniques that in some way make use of quantum physics. This is fairly loosely defined. The latter category
describes something much more tangible: The enhancement of the sensitivity of a measurement beyond the
classically possible. We’ll discuss these two aspects in the following, using some examples from trapped ion
metrology.

8.1 Quantum assisted metrology

As mentioned, this is a rather broad category. Here I would sort in experiments that make use of techniques
that somehow make the measurement possible or enable it. The first is maybe a bit boring but might give
you a feeling for one of the mayor systematics in spectroscopy: the Doppler shift and laser cooling to reduce
it.

8.1.1 Doppler broadening and shifts

The Doppler shift, of course, will shift the transition frequency as the particle that we interrogate moves
relative to the source of the radiation (laser). For particle moving with velocity v (relative to the laboratory
frame), where vk is the velocity component along the laser beam direction (positive here is towards the
source/observer) some transition frequency f0 would appear shifted to f
vk
f = f0 γ(1 − ) (8.1)
c
p
where c is the speed of light and γ = 1 − (v/c)2 . Because we’re dealing with trapped particles (and the
probe times are typically long compared to the trap frequency) the velocity is averaged over the trap cycle.
Taking this into account, we can express the relative frequency shift (f0 − f )/f and expand in orders of v/c
2
f0 − f 1 vk v2 vk
= −1= − 2
+ + O(v/c)3 (8.2)
f γ(1 − vk /c) c 2c c

The first term of the expansion should look familiar, it’s the regular Doppler shift that you know from your
undergrad courses. Since the trap averages over the direction, there will be no shift from that term. But
there will of course be a Doppler broadening. If we assume that we can resolve the motional sidebands of
the transition we can use the framework we developed to estimate that by simply using a thermal state and
calculating the Rabi frequencies for the different (sideband) transitions. I’ve done that for some thermal

69
70 CHAPTER 8. QUANTUM METROLOGY

1.0
n=10 nbar=0
0.8 n=1000 nbar=100
0.8
nbar=1000
0.6
0.6

Pup
Ω

0.4
0.4

0.2
0.2

0.0
0.0
-20 -10 0 10 20 -20 -10 0 10 20

i th sideband i th sideband

(a) (b)

Figure 8.1: Doppler broadening. The calculations for the plots are done with a LD parameter of η = 0.1. (a)
Rabi frequency for a discreet Dock state for different sidebands. (b) Spin population after probing a transition
of an ion in a thermal state for the ideal (n = 0) carrier pi-time.

states and the sidebands in figure 8.1. For a transition where the motional sidebands cannot be resolved the
same principle applies, just that there will be an envelope of the natural linewidth. You can notice two things
here: Obviously it’s better to be cold, to not have broadening, but also if the temperature is constant, it’s
better to have a higher trap frequency so that n̄ is smaller.
This was the first-order Doppler broadening and apart from cooling you could also get rid of it using
two-photon transitions for the spectroscopy. They work similar to Raman transitions and can by choosing
counter-propagating beams the LD parameter can be zero. There is however a more problematic shift related
to the motion of the ion - the second order Doppler shift - which cannot be canceled other than getting rid of
the thermal excitation. It is described by the second and third term in the expansion and as you can see there
is a square before the averaging of the velocity v, so it does not average to zero! This is a relativistic effect
and can be explained with the time dilation a moving system experiences, hence it’s also called ‘time dilation
shift’. There is nothing you can do about it than cooling and accounting for it by very precisely measuring
any motion that you have.
It’s important to note that this shift will not only be caused by thermal motion but also by micromotion.
So there will be an additional contribution to the shift from trap imperfections or stray electric fields [7].
What’s the ‘quantum metrology’ part here? The cooling. By putting the system close to the ground
state and being able to resolve the carrier from the motional sideband transitions we can get rid or reduce
the impact of the thermal (classical) motion, so it’s a classic (albeit somewhat trivial) example how we can
measure better (reduce systematic errors of the measurement) by going to the quantum regime.

8.1.2 Quantum Logic spectroscopy

The other aspect of quantum assisted metrology is that it sometimes enables a measurement. The prime
example for this is ‘quantum logic spectroscopy’ which was first proposed in the 1990s and implemented in
2005 [91, 79]. Let’s assume we have some atom or molecule that we would like to perform spectroscopy on
because it is fundamentally interesting. In our group we have an experiment were we would like to perform
spectroscopy on the H+ 2 molecular ion because it is possible to calculate it from first principles, meaning by
comparison of theory and experiment we can (assuming the theory is correct) determine the fundamental
constants that enter the theory to predict the transition frequencies that we measure in the experiment - in
case of H+ +
2 this would be for instance the proton-to-electron mass ratio. The issue of H2 is that we cannot
easily detect its internal state because it does not have a cycling transition to scatter photons. Remember:
This is basically the case for most ions (atomic and molecular): We picked the easy-to-control ions from the
second group of the periodic table - having a cycling transition is not a given. The idea of quantum-logic
spectroscopy is to co-trap a ‘logic’ ion (which has a cycling transition) with the ‘spectroscopy’ ion of interest
and use quantum logic operations to perform the readout of the spectroscopy ion through the logic ion. The
logic protocol is quite similar to the Cirac-Zoller gate.
We assume we have co-trapped 9 Be+ and H+ 2 , such that their motion is strongly coupled and thus shared.
Figure 8.2 shows a picture of a 9 Be+ ion that has an (invisible) co-trapped H+ 2 next to it. After performing
some spectroscopy pulse (Rabi, Ramsey) on a transition in H+ 2 , it is in a superposition state of two internal
states α |↑H2 i + β |↓H2 i. The goal is now to read out the state, i.e. determine α, β. We prepare the 9 Be+ ion
in the spin down state |↓Be i and a mode of motion that both ions couple well to in its ground state |0i. The
8.2. QUANTUM ENHANCED METROLOGY 71

Figure 8.2: A 9 Be+ ion (the bright spot), co-trapped with a (invisible) H+
2 ion. An unremarkable picture but
a big success for us.

full initial state of the system is then

(α |↑H2 i + β |↓H2 i) |↓Be i |0i . (8.3)

We apply a red sideband pulse on the H+

2 and get

(α |↓H2 |1ii + β |↓H2 i |0i) |↓Be i , (8.4)

And follow this with a red sideband pulse on 9 Be+ resulting in

(α |↓H2 i |↑Be i + β |↓H2 i |↓Be i) |0i . (8.5)

When we now read out the spin state of 9 Be+ in the usual way (and repeat to get statistics, always preparing
the same initial state) we can determine |α| and |β|, so we have a method to measure the state of H+2 indirectly.
This method was first demonstrated for 9 Be+ and 27 Al+ as logic/spectroscopy ions, respectively. 27 Al+ is a
very good ion for an atomic clock and is today used (with quantum logic spectroscopy) in one of the most
accurate clocks [12].
Quantum logic spectroscopy immensely powerful. It allows us to control any ion species, regardless of its
internal structure. It was used for molecular ions and also highly-charged ions [92, 65]. On top of the readout,
sympathetic cooling allows the reduction of Doppler shifts and simply by measuring the protocol can be used
to prepare a pure initial state of the spectroscopy ion, in case we also lack this capability, as is the case in
H+2.

8.2 Quantum enhanced metrology

For the quantum enhancement there are two typical cases to look at (for us): Squeezing of the motion and
entanglement of spins.

8.2.1 Motional squeezing

The ground state |0i (and any displaced ground state) is a so-called ‘minimal-uncertainty state’ - i.e. it fulfils
the Heisenberg uncertainty relation with an equal sign

σp σx = h̄/2, (8.6)

and, since the ground state is circular σp = σx There is no way around Heisenberg, but we can choose
to distribute the uncertainty unequally instead of equally. That’s exactly what the squeezed state is doing
with it’s squeezed and anti-squeezed quadratures. This can be used to make a measurement of for instance
the displacement of one quadrature more precise at the expense of making the measurement of the other
quadrature less precise. This is what we’ve basically demonstrated in [60] and was also demonstrated (maybe
a more useful fashion) in [15]. Of course squeezing of bosonic states was used a lot with photons.

8.2.2 Spin entanglement

[11, 9] We’ve discussed before that in a Rabi or Ramsey experiment we are limited in our frequency sensitivity
by the probe time T , leading to a Fourier limited measurement (in the absence of noise). I’d like to expand
a little on this by introducing the Allen deviation.
72 CHAPTER 8. QUANTUM METROLOGY

When we perform spectroscopy we have three parameters: The probe time T , the number of particles
(ions) we measure at the same time N and the averaging time τ over which we repeat the measurement.
By doing a measurement we get an estimate of the difference of the laser’s frequency to the ion’s transition
frequency. We express this as a relative quantity
ωLaser − ωIon
y= (8.7)
fIon

For a single measurement we get y, averaging several measurements over τ we get hy(τ )i. We can now calculate
the statistical variance of hy(τ )i and how it behaves as we increase τ . For this we calculate the variance of
blocks of averaged data hy(τ )ii . We assume we took K total measurements and chop them up into M blocks
of measurements, where each block is the average over a period τ . The so-called Allan variance - named after
one of it’s inventors, David W. Allan - given as
M −1
1 X 2
σy2 (τ )

= hy(τ )ii+1 − hy(τ )ii (8.8)
2(M − 1) i=1

and the more common quantity derived from this is the Allan deviation σy (τ ) which describes your knowledge
of the frequency difference after a duration of continuous measurement τ . The Allen deviation for averaging
perfect Ramsey measurements can be derived as [44]

1
σy (τ ) = √ , (8.9)
ω0 N T τ

with ω0 = ωIon . Perfect Ramsey measurements means that you use all the time you have for measuring,
so not dead time. This is fairly unrealistic, since we would also spend time for laser cooling etc., also there
is no decoherence of any sort and the π/2 pulses of the Ramsey are of negligible duration compared to the
measurement duration T . This ideal uncertainly is also called ‘quantum projection noise’ - it arises from the
fact that we measure a system √ that only gives us one bit of information per particle and interrogation. The
Fourier limit is basically in the T term. Think again about the state of the ion in the Ramsey measurement:
We initially set up a superposition state
1
|ψi = √ (|↑i + |↓i) (8.10)
2

which will pick up a relative phase (relative to the phase of the laser) exp (iδt) with δ = ω0 − ωLaser , so for a
free evolution time of T we get
1
|ψi = √ (|↑i + eiδT |↓i) (8.11)
2

which resulted in the signal (spin population)

1
P (↓) = (1 + cos(δT )) . (8.12)
2
In the Allan deviation we see that also the number of particles can be used to reduce the uncertainty -
naturally: it’s more measurements
√ in the same amount of time. Since these ions are uncorrelated the reduction
in uncertainty goes with N .
But we could also use correlated spin states, so entanglement. For M ions we define the ‘all spins up/down’
state
M
Y
|⇑i = ⊗ |↑i i (8.13)
i
M
Y
|⇓i = ⊗ |↓i i (8.14)
i

Let’s assume we have the following so-called ‘Green-Horne-Zeilinger’ (GHZ) state

1
|ψi = √ M
(|⇑i + |⇓i). (8.15)
2
8.3. ATOMIC (ION) CLOCKS 73

If we let that state evolve we get

M
1 Y 1
|ψi = √ M
(|⇑i + ⊗eiω0 T |↓i i) = √ M
(|⇑i + eiM ω0 T |⇓i), (8.16)
2 i 2
where we assume that the frequency of each ion is the same (ω0 ), resulting in the phase evolution of the ions
to adds up linearly. In reference to a laser we would get
1
|ψi = √ (|⇑i + eiM δT |⇓i), (8.17)
2
If we now read out we get the signal
1
P (⇓) = (1 + cos(M δT )) . (8.18)
2
√
This signal is M times as sensitive as the single-ion signal and factor M more sensitive than the uncorrelated
M -ion case. This results in the Allen deviation for M entangled spins
1
σy (τ ) = √ . (8.19)
ω0 M T τ
While for two ions this advantage seems marginal, ten entangled ions are as good as 100 untangled ions!
How does this ‘amplification’ come about? First, it’s important to not get confused about correlated
spins and uncorrelated spins/experiments. You √ could still run N experiments with M correlated spin in
parallel and would get an additional factor 1/ N in the Allen deviation (of course you would have to built
N experiments...). You can notice that the Allen deviation also has an 1/ω0 factor. When preparing an
entangled state we[re essentially making a ‘super-spin’ with a frequency splitting of ωsuper = N ωo . On a
‘super’-Bloch sphere it behaves just like a single spin, where in the case of uncorrelated spins we have just
multiple, independent Bloch spheres, one for each physical spin. So in the Allen deviation we could equally
well write ωsuper instead of N ωo .
Right now we’re still missing the tools to carry out such an ‘entangled’ Ramsey experiment. We’ll look
at this again for two ions only but it’s identically for more ions. Of course we start with the ions in |↓↓i.
We can then use an MS gate to get to the initial Bell state |↓↓i + |↑↑i. Then we let the system evolve for
T and follow with another MS pulse that then creates the previously mentioned signal. What you might be
confused about is what the phase of the laser is in the MS case. Recalling the phase definition I introduced
in the two-loop MS gate (see section 6.2.2), the phases of the MS gate are the phases of the RSB and BSB
laser tones φRSB and φBSB . We can reformulated these phases as
φRSB − φBSB
φm = (8.20)
2
φRSB + φBSB
φS = (8.21)
2
where φm we call the motional phase - it determines which direction we drive out in phases space but dos not
matter here; φS is the spin phase and is what you would have to scan to get the typical Ramsey fringe pattern
for the entangled case. This has been demonstrated several times. A notable demonstrations was done here
[70] with up to 14 entangled ions. This technique can also be combined with QLS [52].
We should be aware that the formula for the Allen deviation for entangled qubits is again only valid for
perfect Ramsey experiments. Unfortunately, the gain that we get from Heisenberg scaling goes away if the
measurement is limited by decoherence, which it often will be if we are trying to perform the best measurement
we can. If Heisenberg scaling will improve a measurement or clock depends on the exact noise spectrum and
details of the experiment and considerable work has been done to investigate this [38]. It is also possible to
go beyond GHZ states, which can in part recover the Heisenberg advantage, even with decoherence [3].

8.3 Atomic (ion) clocks

We now look a bit closer into atomic clocks as the prime an example of (quantum) metrology with trapped
ions. Among the currently best clocks in the world are: A single-aluminum ion clock (using a co-trapped
magnesium or calcium ion and QLS) [12], a ytterbium single-ion clock [40], strontium and ytterbium neutral
atom lattice clocks. A nice, recent review of optical atomic clocks: [61].
They are mostly characterized by two numbers: Their frequency uncertainty uf (so their accuracy) which
is unusually given in relative terms (uf /f ), because that is what matters for a clock, and their instability, so
74 CHAPTER 8. QUANTUM METROLOGY

Figure 8.3: Level structure of the 27

Al+ ion. Taken from [43].

precision, which is given as the Allen deviation σy (τ ). The frequency uncertainty tells you how accurate the
clock can tell time. The Allen deviation tells us how long the clock needs to average measurements for to
−18
−15
√ uncertainty of all these clocks is uf /f ∼ 10 , but they
reach a certain accuracy. The relative frequency have
√
very different Allen deviation: σy (τ ) 10 τ , while the ’fastest’ neutral atom clock has σy (τ ) 3 × 10−17 τ
[62]. This reflects the fact that the neutral atom clocks work with many atoms (3000 in the case of [62] - but
the ion clocks only with a single ion.
Let’s have a closer look into the ion clocks, using the NIST 27 Al+ QLS clock as example. Brewer et al. use
a Al+ ion co-trapped with a 24 Mg+ ion. The 24 Mg+ ion has basically the same level structure as 9 Be+ and
27

is used as the logic ion for QLS. The aluminium ion has a level structure as shown in figure 8.3. In principle
it would be possible to read out the 27 Al+ ion directly, but the wavelength for the cycling transition is with
167 nm quite hard to build a laser for and its linewidth is 2π × 224 MHz, so the Doppler temperature would
be quite large. The QLS approach gets around those limitations, as discussed before.
There are several dipole forbidden transitions. The S ↔3 P 1 transition has a linewidth of 2π × 522 Hz
and the S ↔3 P 0 transition 2π × 8 mHz. The S ↔3 P 0 is the clock transition, so the transition the laser gets
stabilized to. The S ↔3 P 1 is a useful transition because it its narrow enough to resolve motional sidebands
but does decay in a reasonable amount of time.
The Al clock is run as follows: The motion is cooled using the 24 Mg+ ion, then the clock transition is
probed (in this case via a simple Rabi pulse, ∼ 100 ms) putting the system into the state

|↓Mg i |n̄i (α 1 S0 + β 3
P0 ). (8.22)

Since the clock probe pulse takes quite some time the motion will have heated up during that to some extent
and is now cooled to the ground state as a first step of the QLS protocol, using the 24 Mg+ ion.

|↓Mg i |n = 0i (α 1 S0 + β 3
P0 ). (8.23)
1 3
This is followed by a blue sideband pulse on the SAl ↔ P1 transition resulting in

|↓Mg i (α |n = 1i 3 P1 + β |n = 0i 3 P0 ), (8.24)

followed by a red sideband pulse on magnesium

|n = 0i (α |↑Mg i 3 P1 + β |↓Mg i 3 P0 ). (8.25)

Subsequently reading out the magnesium state then projects the state into one of the two states |↑Mg i 3 P1
or |↓Mg i 3 P0 and we know what state the 27 Al+ ion in. To improve readout fidelity we can now simply wait
until the 3 P1 state has decayed back down to the ground state and repeat the QLS sequence to increase
readout fidelity. After that the clock probe is repeated and so on. Here the QLS sequence is slightly different
from what we looked at before but the principle is the same, only that it can be carried out multiple times
due to the convenient 3 P1 state.
Like every clock the 27 Al+ clock has a detailed uncertainty budget. It’s evaluation and reduction is the
main work of building and improving it. The error budget of Brewer et al. is shown in figure 8.4. The
8.3. ATOMIC (ION) CLOCKS 75

Figure 8.4: Error budget of the most recent 27

Al+ QLS clock [12].

table gives first the shifts of the unperturbed clock transition. They of course need to be known - the
time standard would be the unperturbed transition frequency, not some shifted value that depends on the
experiment apparatus. The table then gives the uncertainties on these shifts, which result together (sum of
squares) together in the overall uncertainty of the clock. Let’s look at the individual entries.

Excess micromotion The largest impact on the uncertainty is the excess micromotion. the micromotion
can only be canceled to some finite value - the rest is then typically caused by trap imperfections (for instance
phase offsets between the rf electrodes). This micromotion is motion, so causes a (second order) Doppler shift
of the clock transition frequency as we have discussed.

Blackbody radiation (BBR) The thermal radiation that surrounds the 27 Al+ ion will shift the clock
transition. The 27 Al+ ion has a very low BBR shift but it does start to matter now.

Quadratic Zeeman This is the Zeeman shift caused be the magnetic field (ac and dc) - the linear
Zeeman is canceled by alternating the clock probe pulses between the stretch states (the hyperfine levels
1
S0 , mF = ±5/2 ↔ 3 P0 , mF = ±5/2 ).

Secular motion This is the residual secular motion after ground state cooling and the added motion due
to heating during the clock probe, again causing a second order Doppler shift.

Background gas collisions The 27 Al+ -24 Mg+ ion crystal will be hit periodically by background gas
molecules (mostly H2 ). This causes the ion pair to de-crystallize and get (relatively) large distance from the
trap rf null, again causing lots of micromotion, causing a strong, uncontrolled Doppler shift that basically
randomizes the final spin populations of 27 Al+ .

First order Doppler In this paper the authors find that they have a small first-order Doppler shift, caused
by the ion slowly moving in one direction during the clock probe pulse. By alternating two counter-propagating
laser beam directions for the probe the shift can be canceled but of course there is a residual uncertainty due
to an uncertainty because the probe beams cannot be guarantied to be perfectly counter-propagating.

Clock laser Stark This is the (small) light shift induced by the clock laser.

AOM phase chirp AOMs are not perfect and can cause a fluctuation of the laser light as they are turned
on and warm up.

Electric quadrupole The electric quadupole of the Paul trap can cause a shift of the clock frequency.
To my knowledge the micromotion (the biggest uncertainty) could be further reduced in the already built
next iteration of the clock, so the 27 Al+ clock might be getting even better.
The main challenge for the 27 Al+ clock (and really any ion clock) is that only one ion is used and thus
the averaging is so incredibly slow. The ultimate test of a clock is a clock comparison. The goal (and that
has been done in the past) is to built two identical 27 Al+ clocks and compare
√ them down to their estimated
level of uncertainty. With the slow averaging of σy (τ ) = 1.2 × 10−15 / τ it would take 19 days of constant
76 CHAPTER 8. QUANTUM METROLOGY

measurements to compare two clocks with a relative frequency uncertainty of 9.4 × 10−19 as exhibited here.
And of course these are experiments and things go wrong and the up-time of one experiment is probably only
50% even if it is baby sited continuously. This emphasises the need for more clock ions, and indeed several
research groups are working towards this goal. The issue is, here again as in quantum information scaling
the system! This is mostly hindered by micromotion because adding more ions means going away from the
rf null (here we are so sensitive that also axial micromotion is an issue), causing micromotion. With better
engineered traps or smart choices of the trap rf frequency it might be possible to tackle these issues [76, 5].
If then also entanglement could be added to the clock’s advantage we would have a truly marvelous quantum
clock!

8.4 Probing physics beyond the standard model

8.4.1 Gravitational shift
This is not really physics beyond the standard model but I think still very interesting to discuss. Since an
optical clock (on earth) is in a gravitational potential its frequency gets red shifted. This is of course a
prediction of general relativity and was tested numerous times, also with atomic clocks. 1979 for instance
H-Maser on board of a rocked was shot 10 000 km high and compared with a maser on earth to measure the
gravitational induced frequency difference [88].
Today’s atomic clocks are so accurate that this can be done in the lab. One famous demonstration was
done in the NIST ion storage group with two 27 Al+ clocks, where one was 30 cm higher than the other one
and frequency difference was measured between the clocks [19].
The fractional gravitational red shift is given by
δf g∆h
= 2 (8.26)
f0 c

with the height difference ∆h, the gravitational acceleration g = 9.81 m s−1 and the speed of light c. A 1
meter height difference results in a 1.1 × 10−16 relative frequency shift. So today’s clocks are sensitive to
height differences of 1 cm - which will need to be taken into account in clock comparisons.

8.4.2 Determination of fundamental constants

For certain, simple systems or certain properties it is possible to calculate fundamental constants from mea-
surement values of for instance, molecular (rovibrational) transitions. This of course provides accurate mea-
surements that will then be of further use, but it is often also possible to test some predictions of theories
this way. For instance high-precision spectroscopy of the HD+ molecular ions has lead to the most precise
determination of the proton-electron mass ratio recently, but is also able to put bounds on fifth forces [73, 53].
Another notable example is the measurement of the electron electric dipole moment (EDM). The Standard
Model of particle physics predicts a very low, but non-zero value. Extensions like Supersymmetry however
predict much larger values. Recent measurements using molecules (neutral and ions) are now able to provide
upper bounds that already contradict some minimal supersymmetric models [16, 4].

8.4.3 Drift of fundamental constants

Some extensions to the standard model of particle physics predict drifts in the natural constants like the fine
structure constant α or the proton-to-electron mass ratio µ [87]. Because the transitions frequencies of atoms
and molecules have a dependence on their values the atomic clock transition frequencies would also drift in
time. To make quantitative predictions its necessary to know what the proportionality between the values is,
i.e. what is the clock’s transition frequencies sensitivity to a change in some fundamental constant. This is
much easier to derive than to calculated the absolute transition frequencies from first principles. When the
proportionality is known the clock can be compared to another clock with a lower dependence on the same
constant resulting in a frequency ratio measurement. This measurement is then repeated (often one year after
the first) and a drift can be estimated. So far all these measurements result in no significant drift, so far
always resulting in upper bounds for the drifts, which are of course equally important results [29, 39].
Chapter 9

Alternative techniques/additional
topics

9.1 Encoding a qubit in the motion

Instead of encoding quantum information in two-level systems it is of course possible to use higher-dimensional
Hilbert spaces. One possibility is to use a quantum harmonic oscillator. To encode a qubit in an oscillator
was proposed in 2001 [31] and is now known as GKP codes after their inventors. Since a qubit is encoded
(so still only storing the same amount of information as in a two level system) it seems pointless. Why use
a larger Hilbert space (so also a more complex system)? The redundancy of the larger Hilbert space can
directly be used for logical encoding and error correction, providing a first layer of protection. That means
in one oscillator you can encode a logical qubit were you would need ∼ 7 two-level systems to do the same.
Encoding of and computation with (single) GKP qubits has recently been demonstrated using trapped ions
[28] and superconducting microwave cavities [17].
Details of oscillator codes are covered in the QIP II lecture. I’m going to only give a very superficial
introduction here and explain how the states can be prepared in the lab with trapped ions.
The theoretical GKP states are using a “comb” structure in phase space, a superposition of infinitely
many, equidistant, infinitely squeezed states. The two qubit states are differentiated by a shift of their comb
pattern by half the comb period. In math: We define a state infinitely squeezed in the p quadrature with
position α along the q quadrature as |S, αi. The logical qubit states are then defined as
∞
X
|0iL ≡ |S, iδi (9.1)
i=−∞
X∞
|1iL ≡ |S, (i + 1/2)δi , (9.2)
i=−∞

with δ the spacing of the comb teeth. A pictorial representation of the states is showing in figure 9.1.
In practice it is of course not possible to make a superposition of these infinitely squeezed infinitely many
states. It is however also not required. The GKP code can be fault tolerant even with imperfect states
[27]. The Pauli operations (and corresponding stabilizers) are simply displacements of the states. E.g. a X
π rotation would simply be a displacement of the state by half the period δ. Figure 9.2 summarizes the
operations.
How do we prepare such a state in a trapped ion oscillator? The first step is to produce a squeezed state
of the ground state. We do this by using the dissipative state preparation technique [51] as discussed in the
ground state cooling chapter, but using a unitary method would also be possible [63, 15]. The final motional
state would be |S, 0i - a squeezed state at the origin.
In the next step we start to create superpositions of these states by making cat states. We start with
the previous state, also including the internal state now, prepared in the spin down state |↓i |S, 0i = (|+i −
|−i) |S, 0i. We act a state-dependent force in the x-basis on this (MS drive) and get the squeezed cat state
[60]

|+i |S, αi − |−i |S, −αi = |↑i (|S, αi + |S, −αi) + |↓i (|S, αi − |S, −αi) (9.3)

We need to get rid of the entanglement with the spin. As you can see form the result re-written in the
z-basis a ‘positive’ and a ‘negative’ cat are entangled there with the spin states. By measuring the spins state

77
78 CHAPTER 9. ALTERNATIVE TECHNIQUES/ADDITIONAL TOPICS

Figure 9.1: Principle of the theoretical GSK states. From [31].

Figure 9.2: Summary of the GKP operations.

9.2. MICROWAVE ENTANGLING GATES 79

we can collapse the system into one of those and thus get rid of the entanglement with the spin but still
have superposition state in the motion. We’re using 40 Ca+ for these experiments. If the detection projects
the system in the bright (|↓i) state we scattered a substantial amount of photons and the motion will have
decohered due to the recoil. If instead the dark state (|↑i) comes up the spin is collapsed but no photons were
scattered. Thus the (collapsed) motional state is preserved. We can used this to do heralding. If we measure
the dark state after making the cat we continue the experiment, if we measure bright we start over. The next
step is to make superpositions of the superposition, so applying the SDF again with the same displacement
amplitude. We would get the state

|+i (|S, 2αi + |S, 0i) − |−i (|S, 0i + |S, −2αi) = |↑i (|S, 2αi + |S, −2αi) + |↓i (|S, 2αi + 2 |S, 0i + |S, −2αi)
(9.4)

The |↑i cat is of no use to us, but the |↓i starts to look a bit like a grid state. Performing a carrier π rotation
to flip the spin we repeat the heralding protocol and end up with the state

|↑i (|S, 2αi + 2 |S, 0i + |S, −2αi). (9.5)

This can be repeated until the desired number of superpositions is reached. We stop after this step for the
|0iL and one more for the |1iL state.

9.2 Microwave entangling gates

When covering the LD approximation we found that magnetic dipole transitions in the GHz regime have
extremely small LD parameters (η ∼ 10−6 ). This makes them essentially not couple to the motion and
thus they can’t be used for spin-motion coupling and thus entangling gates. And while they can be used to
perform very high-fidelity single-qubit gates, their long wavelength also means that it’s not directly possible
to address individual ions, so microwave usually cause spin rotations on all ion in the trap. There are a couple
ways around this. We’ll discuss here only one approach for solving either issue, although others exist. Both
solutions are addressed similarly: Rather than using free-space microwaves (where the wavelength is fixed to
be very long) it’s possible to use the near field instead. For spin-motion coupling this means engineering a very
strong oscillating microwave gradient - this essentially artificially shortens the wavelength that enters the LD
parameter and allows to perform spin-motion coupling. One way to do this is by using a surface trap (where
the ion can be trapped very close to the surface) and have an electrode carry a microwave current. This
creates an oscillating magnetic field around the electrode that falls of radially - thus a gradient. In principle
this would be enough, however the spacial gradient for realistic electrode-ion distances (∼ 30 µm) is still small
compared to the amplitude of the field oscillation. This causes the carrier to be too strongly off-resonantly
driven and compromising any spin-motion coupling. This can be solved by a “three-wire geometry” shown in
figure 9.3, where the field is nulled by a suitable geometry and current phase and amplitude, enabling a point
with an oscillating gradient with zero field. This has been used to perform high-fidelity entangling gates with
microwaves [33].
A similar technique was used for single ion addressing, using wire loops in a geometry where a field at
the location of the addressed ion is non-zero but does fall of rapidly with distance. Additionally cancellation
fields can be applied at the positions of other ions, further suppressing any unwanted rotations on spectator
ions. Figure 9.4 shows the electrode structure that was used in [6].
While microwaves gradients can be used to do entangling gates and addressed single qubit operations,
you should keep in mind that laser cooling and detection of the internal state cannot be performed using
microwaves. For any type of cooling it’s necessary to perform a non-unitary operation. Dissipation is required
to extract entropy. To do this efficiently we need a dipole transition that decays fast enough to have a sufficient
cooling speed. The magnetic dipole transitions of the hyperfine structure of the electronic ground states have
extremely long lifetimes, so cannot be used. The same of course is correct for detection: We rely on scattering
many (thousands) of photons to read out the internal state with high fidelity, and again we need a dipole
transition to do that on experimental timescales.

9.3 Molecular ions

Molecular ions are still ions and the trapping and the motion in the trap works just the same as for atomic
ions (or any other charged particle). The real difference is the same for neutral and ionic molecules - their
internal structure is much more complicated than that of atoms. The additional degrees of freedom (vibration
and rotation) add a hard to control level structure. this has several reasons: Unlike the (translational) motion
of the ions in the trap the rotation and vibration of a molecule is anharmonic and thus the energy spacing
80 CHAPTER 9. ALTERNATIVE TECHNIQUES/ADDITIONAL TOPICS

(a)

(b)

Figure 9.3: (a) Three-wire geometry for creating a magnetic field gradient without a field (not sure where this
picture is taken from...). (b) Surface trap structure with three tracks for the microwave currents (in yellow).
Picture from [72].

Figure 9.4: Electrode structure for addressing single ions with microwaves [6].
9.4. QUANTUM SIMULATION 81

of the levels is not equidistant. That means for cooling such a motion it is not possible to use something
parallel to sideband cooling with e.g. a “red” rotational sideband (since there will be distinct transitions with
different frequencies between different rotational or vibrational levels. Also, there are almost no ions with
closed cycling transitions that are needed for cooling or detection. It is possibly (with many lasers) to still
perform cooling, but is is much more involved. Examples with ions are [80, 81].
What kind of transitions are the vibrational and rotational transitions? Similar to the electronic structure
that depends on the properties of the molecule. For the simplest molecules, diatomic molecules there are two
classes: Hetero-nuclear molecules which consist of two atoms with different masses (e.g. CaH+ ) and homo-
nuclear molecules with the same masses (e.g. H+ 2 ). Since hetero-nuclear molecules form a dipole (the electron
is stronger attracted by one core than the other) rotational and vibrational transitions are dipole allowed.
However, they do not feature the short lifetimes, tens of MHz linewidth that atomic dipole transitions have.
Due to the poor overlap of the rovibrational molecular wave functions lifetimes are more in the ∼ 10 − 100 ms
regime. Also their frequencies are not optical, but rather in the single THz range for fundamental rotational
transitions and tens of THz for the fundamental vibrational transitions (“higher sidebands” called overtones
can also be driven (and decay) but those are then suppressed stronger). The THz regime is technically quite
challenging, because radiation sources (e.g. quantum cascade lasers) are more experimental than standard
lasers, complicating control of molecular ions further. Also the THz range of the rotation is the same range as
the standard room-temperature thermal radiation. This causes the rotation to be constantly heated to room
temperature (or what ever the environment temperature is).
Homo-nuclear molecules (due to their symmetric charge distribution) have no dipole moment and thus
rovibrational transitions can only be driven by higher orders (quadrupole, magnetic dipole), and the lifetimes
of the rovibrational levels can be extremely long (some levels in H+2 have lifetimes of up to 80000 years). This
makes them harder to experiment with (nothing ever decays) but the narrow linewidth of the transitions are
of course beneficial for spectroscopy.
Molecules offer more variety (there are just so many) and more/different sensitivities to certain ef-
fects/properties (for instance the electron dipole moment or proton-electron mass ratio discussed before).
That makes them interesting for metrology. The complex level structure makes them harder to control but
the large Hilbert space also offers opportunities - for instance a recent publication proposed to encode some-
thing akin to the oscillator GKP states in molecules [1]. Also their variety could be important - the many
available transitions (and thus transition wavelengths) could offer a high-fidelity ion-photon interface at wave-
lengths in the telecom range (1550 nm - were optical fibers have the lowest loss and a lot of advanced optical
technology exists), where atomic ions usually don’t have transitions.

9.4 Quantum simulation

In quantum simulation the general idea is to let go of some of the requirements to build a universal, fault-
tolerant quantum computer and gain easier scaling to larger number of constituents (not necessarily qubits).
The loss of universality means that not all problems can be solved on a quantum simulator. A typical
implementation is the implementation of a certain Hamiltonian and study its ground state or some dynamics.
The Hamiltonian can be of fundamental interest, mimicking some solid-state system, or could be mapped
onto a computational problem and the quantum simulator could, in principle, provide a quantum advantage
solving such a problem.
A recent review covers trapped-ion quantum simulators in-depth ([69]. We’ll discuss the basics here, first
looking at the implementation of the ”natural” trapped-ion Hamiltonian, the Ising Hamiltonian, its continuous
implementation in long ion strings, trotterization of Hamiltonians for more general simulations and discussing
2D extensions to the long string simulators.

9.4.1 Quantum simulation with long ion strings

Implementation
The Hamiltonian we want to implement with the simulator is the Ising Hamiltonian of N spins
N
X N
X
ĤIsing = Ji,j σ̂x,i σ̂x,j − B σ̂y,i (9.6)
j<i i

were the first term describes (two-body) σ̂x σ̂x couplings of individual spins (i,j) with varying couplings per
pair (Ji,j ) and the second term describes global single-spin rotations, just as a homogeneous magnetic field
would act on a system of spins.
82 CHAPTER 9. ALTERNATIVE TECHNIQUES/ADDITIONAL TOPICS

Implementing the second term in a many ions is trivial - you just need to do global carrier rotations. The
second term is a bit trickier. It’s similar to the MS interaction - and this is how we implement it, just a bit
different than encountered before for two ions.
If we look at the mode structure of transverse (radial) modes of a long string (see figure ??) we see a
lot of modes bunched up together. If we would try to implement a MS interaction on only one mode we
would need to put the gate detuning δ to a very small value to only excite one mode of motion. Assume
we can do that and we use the COM mode of one of the radial directions. We would get, for the correct
gate parameters, an effective σ̂x -σ̂x coupling across the chain, where every ion pair is coupled with the same
strength - so we would get a global J = Ji,j . We did not discuss the details of the MS interaction on many
ions so far. One thing to keep in mind, and also explaining why I talk about pairwise interactions is the fact
that the underlying interaction of the σ̂x -σ̂x coupling of the MS Hamiltonian is the Coulomb interaction - so
a two-body interaction! In fact it’s hard to get a native more-than-3 body interaction with trapped ions.
We looked at the MS gate only in it’s ‘modern’ form - using a single loop in phase space to do the gate.
For this we hat the gate time t = 2π/δ and Ω = δ/2. This means we only get the σ̂x -σ̂x interaction when
we close the loop, the interaction ad intermediate times is quite different. We do this because it’s efficient
but the original proposal was a different one. If you choose instead t  2π/δ and Ω  δ/2 you get a lot of
small loops in phase space and the phase accumulation that is needed for an entangling gate is rather slow.
The advantage is however that the entanglement with the motion is always small (so motional decoherence
matters less, that is why the two-loop gate can help with bad motion) and the spin evolution is much closer to
the effective σ̂x -σ̂x interaction at all times. Using the MS interaction like this we can now implement the Ising
Hamiltonian with a global J. We can however also enter another regime. By increasing the gate detuning
δ more modes of motion will get excited by the MS interaction. Because we do small loops we don’t have
to worry anymore that they close at the same time as we would have to for fast MS gates. The effective
σ̂x -σ̂x couplings then get to be more localized. That’s an effect from the superposition of all the forces of all
the modes combined. As has been shown [75], depending on the gate detuning the coupling strengths can be
given as

J0
Ji,j = (9.7)
|i − j|α

with 0 < α < 3 and δ ∝ α. The global all-to-all coupling case is given by α → 0. As α gets larger the
interaction becomes more local with α → 3 describing pure nearest-neighbour interactions.
This gives us all the tools needed for the first experiments. Note that so far we only use global inter-
actions/beams, so the technical difficulty is not as high as when we would have to do single-ion addressing
to do discreet gates on individual ions. We also don’t need to do in-sequence detection and feedback, no
sympathetic cooling etc. so this is a lot easier (but not easy!) than universal quantum computation.

Ground state studies

Often finding the ground state of a given Hamiltonian can be an important task and cannot be performed
for a strongly coupled many-body system on a classical computer. The idea here is to use the simulation to
“show” us what the ground state is. The most common technique for this is adiabatic ramps, here discussed
again using the example of the Ising Hamiltonain. For the second term of the Ising Hamiltonian we can easily
prepare the ground state - it’s just a global π/2 σ̂x rotation. How does that help us to get to the ground
state of the full Ising Hamiltonian? By adiabatic evolution. If we start with the global B-field interaction
strong compared to the two-body couplings (B  Ji,j ) then the ground state can be well approximated by
the ground state of the B-field interaction alone. We prepare that, turn on the interaction with B  Ji,j and
then slowly turn down the strength of B. If this is done slowly enough (adiabatically), so that any excitation
of an excited state is suppressed the system always stays in the ground state of the current Hamiltonian,
ending up in the ground state of the σ̂x -σ̂x coupling and thus in an interesting entangled ground state. This
was shown several times. A nice study is for instance [41] with 11 ions.

Dynamics

It is of course also possible to study the dynamics of such a system by preparing an initial state and letting
the system evolve. This was for instance done by the Innsbruck ion group by exiting only one ion in a chain
and letting this excitation evolve. This of course required single-ion addressing. A nice study like that was
performed by Jurcevic et al. who showed the spreading of quantum information across a long ion chain in
different coupling regimes [47].
9.4. QUANTUM SIMULATION 83

Beyond the simplest Ising Hamiltonian

To implement more complex or less ‘native’ Hamiltonians one can go back to a more discretized approach
that at times almost resembles classic, gate-based, quantum information processing. Assume we want to
implement the Hamiltonian

Ĥtot = Ĥ1 + Ĥ2 . (9.8)

To reduce complexity it might be nice to break up the full Hamiltonian into parts, first apply Ĥ1 , then Ĥ2 .
This would mean we can write the evolution operator as

Ûtot (t) = exp (i(Ĥ1 + Ĥ2 )t/hbar) = exp (iĤ1 t/hbar) exp (iĤ2 t/hbar) (9.9)
h i
But if two parts do not commute ( Ĥ1 , Ĥ2 6= 0) this is of course not possible. In this case the Suzuki-Trotter
expansion can save us, which states that we can write the evolution operator as above even if the operators
do not commute for short times, so strictly τ → 0
N
Y
Ûtot,ST (t) = exp (i(Ĥ1 + Ĥ2 )t/h̄) ≈ exp (iĤ1 τ /h̄) exp (iĤ2 τ /h̄) (9.10)
j

for τ small compared to the characteristic times of Ĥ1 and Ĥ2 and t = N τ . This technique is often called ’Trot-
terization’ (sadly forgetting about Suzuki). The technique has been used quite a lot. Here are fundamental
study for a quantum simulator can be found [55].

9.4.2 Simulators in two dimensions

Obviously, the physics gets richer (and less predictable) if one changes from 1D strings to 2D arrays. The
currently only implementation that reaches large numbers of ions in 2D is using a Penning trap at the Ion
Storage Group at NIST. There entanglement among 200 ions has been shown [? ].
Other approaches to scaling two 2D are:
• Large-scale Paul-trap geometries - here the issue is, as so often the micromotion.
• Micro-Paul trap arrays [66] (Freiburg, Germany and NIST Boulder, USA).

• Micro-Penning trap arrays (our group).

• Optical trapping of ions (our group).
In all these approaches the challenge is somewhat different to the quantum computer scaling that was addressed
by the qccd architecture. Here we want a constantly coupled system, so many ions that are close enough to
each other at all times to couple via their Coulomb interaction.
84 CHAPTER 9. ALTERNATIVE TECHNIQUES/ADDITIONAL TOPICS
Bibliography

[1] Victor V. Albert, Jacob P. Covey, and John Preskill. Robust encoding of a qubit in a molecule. Phys.
Rev. X, 10:031050, Sep 2020.

[2] J. Alonso, F. M. Leupold, Z. U. Solèr, M. Fadel, M. Marinelli, B. C. Keitch, V. Negnevitsky, and J. P.

Home. Generation of large coherent states by bang–bang control of a trapped-ion oscillator. Nature
Communications, 7(1):11243, 2016.

[3] A. André, A. S. Sørensen, and M. D. Lukin. Stability of atomic clocks based on entangled atoms. Phys.
Rev. Lett., 92:230801, Jun 2004.

[4] V. Andreev, D. G. Ang, D. DeMille, J. M. Doyle, G. Gabrielse, J. Haefner, N. R. Hutzler, Z. Lasner,

C. Meisenhelder, B. R. O’Leary, C. D. Panda, A. D. West, E. P. West, X. Wu, and ACME Collaboration.
Improved limit on the electric dipole moment of the electron. Nature, 562(7727):355–360, 2018.

[5] Kyle Arnold, Elnur Hajiyev, Eduardo Paez, Chern Hui Lee, MD Barrett, and John Bollinger. Prospects
for atomic clocks based on large ion crystals. Physical Review A, 92(3):032108, 2015.

[6] D. P. L. Aude Craik, N. M. Linke, T. P. Harty, C. J. Ballance, D. M. Lucas, A. M. Steane, and D. T. C.

Allcock. Microwave control electrodes for scalable, parallel, single-qubit operations in a surface-electrode
ion trap. Applied Physics B, 114(1):3–10, 2014.

[7] D. J. Berkeland, J. D. Miller, J. C. Bergquist, W. M. Itano, and D. J. Wineland. Laser cooled mercury
ion frequency standard. Phys. Rev. Lett., 80:2089, 1998.

[8] D. J. Berkeland, J. D. Miller, J. C. Bergquist, W. M. Itano, and D. J. Wineland. Minimization of ion

micromotion in a Paul trap. Journal of Applied Physics, 83(10):5025–5033, 1998.

[9] J. J. Bollinger, Wayne M. Itano, D. J. Wineland, and D. J. Heinzen. Optimal frequency measurements
with maximally correlated states. Phys. Rev. A, 54:R4649–R4652, Dec 1996.

[10] R. Bowler, J. Gaebler, Y. Lin, T. R. Tan, D. Hanneke, J. D. Jost, J. P. Home, D. Leibfried, and D. J.
Wineland. Coherent diabatic ion transport and separation in a multizone trap array. Phys. Rev. Lett.,
109:080502, Aug 2012.

[11] Samuel L. Braunstein. Quantum limits on precision measurements of phase. Phys. Rev. Lett., 69:3598–
3601, Dec 1992.

[12] S. M. Brewer, J.-S. Chen, A. M. Hankin, E. R. Clements, C. W. Chou, D. J. Wineland, D. B. Hume,

and D. R. Leibrandt. 27 al+ quantum-logic clock with a systematic uncertainty below 10−18 . Phys. Rev.
Lett., 123:033201, Jul 2019.

[13] Lowell S. Brown and Gerald Gabrielse. Geonium theory: Physics of a single electron or ion in a penning
trap. Rev. Mod. Phys., 58:233–311, Jan 1986.

[14] M. Brownnutt, M. Kumph, P. Rabl, and R. Blatt. Ion-trap measurements of electric-field noise near
surfaces. Rev. Mod. Phys., 87:1419–1482, Dec 2015.

[15] S. C. Burd, R. Srinivas, J. J. Bollinger, A. C. Wilson, D. J. Wineland, D. Leibfried, D. H. Slichter, and

D. T. C. Allcock. Quantum amplification of mechanical oscillator motion. Science, 364(6446):1163–1165,
2019.

[16] William B. Cairncross, Daniel N. Gresh, Matt Grau, Kevin C. Cossel, Tanya S. Roussy, Yiqi Ni, Yan
Zhou, Jun Ye, and Eric A. Cornell. Precision measurement of the electron’s electric dipole moment using
trapped molecular ions. Phys. Rev. Lett., 119:153001, Oct 2017.

85
86 BIBLIOGRAPHY

[17] P. Campagne-Ibarcq, A. Eickbusch, S. Touzard, E. Zalys-Geller, N. E. Frattini, V. V. Sivak, P. Reinhold,

S. Puri, S. Shankar, R. J. Schoelkopf, L. Frunzio, M. Mirrahimi, and M. H. Devoret. Quantum error
correction of a qubit encoded in grid states of an oscillator. Nature, 584(7821):368–372, 2020.

[18] J Chiaverini and JM Sage. Insensitivity of the rate of ion motional heating to trap-electrode material
over a large temperature range. Physical Review A, 89(1):012318, 2014.

[19] C. W. Chou, D. B. Hume, T. Rosenband, and D. J. Wineland. Optical clocks and relativity. Science,
329(5999):1630–1633, 2010.

[20] J. I. Cirac, R. Blatt, P. Zoller, and W. D. Phillips. Laser cooling of trapped ions in a standing wave.
Phys. Rev. A, 46:2668–2681, Sep 1992.

[21] J. I. Cirac and P. Zoller. Quantum computations with cold trapped ions. Phys. Rev. Lett., 74(20):4091–
4094, 1995.

[22] H.G. Dehmelt. Radiofrequency spectroscopy of stored ions i: Storage. In D.R. Bates and Immanuel
Estermann, editors, Advances in Atomic and Molecular Physics, volume 3 of Advances in Atomic and
Molecular Physics, pages 53 – 72. Academic Press, 1968.

[23] L. Deslauriers, S. Olmschenk, D. Stick, W. K. Hensinger, J. Sterk, and C. Monroe. Scaling and suppression
of anomalous heating in ion traps. Phys. Rev. Lett., 97:103007, Sep 2006.

[24] L.-M Duan, Boris Blinov, D. Moehring, and Christopher Monroe. Scalable trapped ion quantum com-
putation with a probabilistic ion-photon mapping. Quantum Information & Computation, 4:165–173, 05
2004.

[25] S. Ejtemaee and P. C. Haljan. 3d sisyphus cooling of trapped ions. Phys. Rev. Lett., 119:043001, Jul
2017.

[26] Philip Ekstrom and David Wineland. The isolated electron. Scientific American, 243(2):104–121, 1980.

[27] C. Flühmann. Encoding a qubit in the motion of a trapped ion using superpositions of displaced squeezed
states. Ph.D. thesis, Eidgenössische Technische Hochschule Zürich, 2019.

[28] C. Flühmann, T. L. Nguyen, M. Marinelli, V. Negnevitsky, K. Mehta, and J. P. Home. Encoding a qubit
in a trapped-ion mechanical oscillator. Nature, 566(7745):513–517, 2019.

[29] R. M. Godun, P. B. R. Nisbet-Jones, J. M. Jones, S. A. King, L. A. M. Johnson, H. S. Margolis,

K. Szymaniec, S. N. Lea, K. Bongs, and P. Gill. Frequency ratio of two optical clock transitions in
171
yb+ and constraints on the time variation of fundamental constants. Phys. Rev. Lett., 113:210801,
Nov 2014.

[30] D. Gottesman and I. L. Chuang. Quantum teleportation is a universal computational primitive. Nature,
402:390, 1999. quant-ph/9908010.

[31] Daniel Gottesman, Alexei Kitaev, and John Preskill. Encoding a qubit in an oscillator. Phys. Rev. A,
64:012310, Jun 2001.

[32] D. Hanneke, S. Fogwell Hoogerheide, and G. Gabrielse. Cavity control of a single-electron quantum
cyclotron: Measuring the electron magnetic moment. Phys. Rev. A, 83:052122, May 2011.

[33] T. P. Harty, M. A. Sepiol, D. T. C. Allcock, C. J. Ballance, J. E. Tarlton, and D. M. Lucas. High-fidelity

trapped-ion quantum logic using near-field microwaves. Phys. Rev. Lett., 117:140501, Sep 2016.

[34] D. Hayes, S. M. Clark, S. Debnath, D. Hucul, I. V. Inlek, K. W. Lee, Q. Quraishi, and C. Monroe.
Coherent error suppression in multiqubit entangling gates. Phys. Rev. Lett., 109:020503, Jul 2012.

[35] Johannes M. Heinrich. A Be+ Ion Trap for H+

2 Spectroscopy. Ph.D. thesis, SORBONNE UNIVERSITE,
2018.

[36] D. A. Hite, Y. Colombe, A. C. Wilson, D. T. C. Allcock, D. Leibfried, D. J. Wineland, and D. P. Pappas.

Surface science for improved ion traps. MRS Bulletin, 38(10):826–833, 2013.

[37] J. P. Home. Chapter 4 - quantum science and metrology with mixed-species ion chains. In Paul R. Berman
Ennio Arimondo and Chun C. Lin, editors, Advances in Atomic, Molecular, and Optical Physics, vol-
ume 62 of Advances In Atomic, Molecular, and Optical Physics, pages 231 – 277. Academic Press, 2013.
BIBLIOGRAPHY 87

[38] S. F. Huelga, C. Macchiavello, T. Pellizzari, A. K. Ekertand M. B. Plenio, and J. I. Cirac. Improvement

of frequency standards with quantum entanglement. Phys. Rev. Lett., 79(20):3865–3868, 1997.

[39] N. Huntemann, B. Lipphardt, Chr. Tamm, V. Gerginov, S. Weyers, and E. Peik. Improved limit on a
temporal variation of mp /me from comparisons of yb+ and cs atomic clocks. Phys. Rev. Lett., 113:210802,
Nov 2014.

[40] N. Huntemann, C. Sanner, B. Lipphardt, Chr. Tamm, and E. Peik. Single-ion atomic clock with 3×10−18
systematic uncertainty. Phys. Rev. Lett., 116:063001, Feb 2016.

[41] R Islam, C Senko, WC Campbell, S Korenblit, J Smith, A Lee, EE Edwards, C-CJ Wang, JK Freericks,
and C Monroe. Emergence and frustration of magnetism with variable-range interactions in a quantum
simulator. Science, 340(6132):583–587, 2013.

[42] W. M. Itano, D. J. Heinzen, J. J. Bollinger, and D. J. Wineland. Quantum zeno effect. PhysRevA,
41:2295, 1990.

[43] Wayne Itano, James Bergquist, Anders Brusch, Scott Diddams, Tara Fortier, Thomas Heavner, Leo Holl-
berg, David Hume, Steven Jefferts, Luca Lorini, Thomas Parker, Till Rosenband, and Jason Stalnaker.
Optical frequency standards based on mercury and aluminum ions. Number 6673. Proceedings of the
SPIE, Conference 6673, 2007-09-12 2007.

[44] Wayne M Itano, James C Bergquist, John J Bollinger, JM Gilligan, DJ Heinzen, FL Moore, MG Raizen,
and David J Wineland. Quantum projection noise: Population fluctuations in two-level systems. Physical
Review A, 47(5):3554, 1993.

[45] Elena Jordan, Kevin A. Gilmore, Athreya Shankar, Arghavan Safavi-Naini, Justin G. Bohnet, Murray J.
Holland, and John J. Bollinger. Near ground-state cooling of two-dimensional trapped-ion crystals with
more than 100 ions. Phys. Rev. Lett., 122:053603, Feb 2019.

[46] M K Joshi, A Fabre, C Maier, T Brydges, D Kiesenhofer, H Hainzer, R Blatt, and C F Roos. Polarization-
gradient cooling of 1d and 2d ion coulomb crystals. New Journal of Physics, 22(10):103013, oct 2020.

[47] P. Jurcevic, B. P. Lanyon, P. Hauke, C. Hempel, P. Zoller, R. Blatt, and C. F. Roos. Quasiparticle
engineering and entanglement propagation in a quantum many-body system. Nature, 511(7508):202–
205, July 2014.

[48] Ezra Kassa, Hiroki Takahashi, Costas Christoforou, and Matthias Keller. Precise positioning of an
ion in an integrated paul trap-cavity system using radiofrequency signals. Journal of Modern Optics,
65(5-6):520–528, 2018.

[49] D. Kielpinski, C. Monroe, and D. J. Wineland. Architecture for a large-scale ion-trap quantum computer.
Nature, 417:709–711, 2002.

[50] D. Kienzler. Quantum Harmonic Oscillator State Synthesis by Reservoir Engineering. Ph.D. thesis, ETH
Zürich, 2015.

[51] D. Kienzler, H.-Y. Lo, B. Keitch, L. de Clercq, F. Leupold, F. Lindenfelser, M. Marinelli, V. Negnevit-
sky, and J. P. Home. Quantum harmonic oscillator state synthesis by reservoir engineering. Science,
347(6217):53–56, 2015.

[52] D. Kienzler, Y. Wan, S. D. Erickson, J. J. Wu, A. C. Wilson, D. J. Wineland, and D. Leibfried. Quantum
logic spectroscopy with ions in thermal motion. Phys. Rev. X, 10:021012, Apr 2020.

[53] I. V. Kortunov, S. Alighanbari, M. G. Hansen, G. S. Giri, V. I. Korobov, and S. Schiller. Proton–electron

mass ratio by high-resolution optical spectroscopy of ion ensembles in the resolved-carrier regime. Nature
Physics, 2021.

[54] K. A. Landsman, Y. Wu, P. H. Leung, D. Zhu, N. M. Linke, K. R. Brown, L. Duan, and C. Monroe.
Two-qubit entangling gates within arbitrarily long chains of trapped ions. Phys. Rev. A, 100:022332,
Aug 2019.

[55] B. P. Lanyon, C. Hempel, D. Nigg, M. Müller, R. Gerritsma, F. Zähringer, P. Schindler, J. T. Barreiro,

M. Rambach, G. Kirchmair, M. Hennrich, P. Zoller, R. Blatt, and C. F. Roos. Universal digital quantum
simulation with trapped ions. Science, 334(6052):57–61, 2011.
88 BIBLIOGRAPHY

[56] Regina Lechner, Christine Maier, Cornelius Hempel, Petar Jurcevic, Ben P. Lanyon, Thomas Monz,
Michael Brownnutt, Rainer Blatt, and Christian F. Roos. Electromagnetically-induced-transparency
ground-state cooling of long ion strings. Phys. Rev. A, 93:053401, May 2016.
[57] P. J. Lee, K-A. Brickman, L. Deslauriers, P. C. Haljan, L0M. Duan, and C. monroe. Phase control of
trap ion quantum gates. Journal of Optics B, 7:S371, 2005.
[58] D. Leibfried, R. Blatt, C. Monroe, and D. Wineland. Quantum dynamics of single trapped ions. Rev.
Mod. Phys., 75:281–324, 2003.
[59] D. Leibfried, B. DeMarco, V. Meyer, D. Lucas, M. Barrett, J. Britton, W. M. Itano, B. Jelenkovic,
C. Langer, T. Rosenband, and D. J. Wineland. Experimental demonstration of a robust and high-fidelity
geometric two ion-qubit phase gate. Nature, 422:412–415, 2003.
[60] Hsiang-Yu Lo, Daniel Kienzler, Ludwig de Clercq, Matteo Marinelli, Vlad Negnevitsky, Ben C. Keitch,
and Jonathan P. Home. Spin-motion entanglement and state diagnosis with squeezed oscillator wavepack-
ets. Nature, 521(7552):336–339, 05 2015.
[61] Andrew D. Ludlow, Martin M. Boyd, Jun Ye, E. Peik, and P. O. Schmidt. Optical atomic clocks. Rev.
Mod. Phys., 87:637–701, Jun 2015.
[62] G. Edward Marti, Ross B. Hutson, Akihisa Goban, Sara L. Campbell, Nicola Poli, and Jun Ye. Imaging
optical frequencies with 100 µHz precision and 1.1 µm resolution. Phys. Rev. Lett., 120:103201, Mar
2018.
[63] D. M. Meekhof, C. Monroe, B. E. King, W. M. Itano, and D. J. Wineland. Generation of nonclassical
motional states of a trapped atom. Phys. Rev. Lett., 76:1796, 1996.
[64] Karan K. Mehta, Colin D. Bruzewicz, Robert McConnell, Rajeev J. Ram, Jeremy M. Sage, and John
Chiaverini. Integrated optical addressing of an ion qubit. Nature Nanotechnology, 11(12):1066–1070,
2016.
[65] P. Micke, T. Leopold, S. A. King, E. Benkler, L. J. Spieß, L. Schmöger, M. Schwarz, J. R. Crespo López-
Urrutia, and P. O. Schmidt. Coherent laser spectroscopy of highly charged ions using quantum logic.
Nature, 578(7793):60–65, 2020.
[66] Manuel Mielenz, Henning Kalis, Matthias Wittemer, Frederick Hakelberg, Ulrich Warring, Roman
Schmied, Matthew Blain, Peter Maunz, David L. Moehring, Dietrich Leibfried, and Tobias Schaetz.
Arrays of individually controlled ions suitable for two-dimensional quantum simulations. Nature Com-
munications, 7:ncomms11839, June 2016.
[67] G. J. Milburn. Simulating nonlinear spin models in an ion trap. quant-ph/9908037, 2000.
[68] G. J. Milburn, S. Schneider, and D. F. James. Ion trap quantum computing with warm ions. Fortschr.
Physik, 48:801–810, 2000.
[69] C. Monroe, W. C. Campbell, L.-M. Duan, Z.-X. Gong, A. V. Gorshkov, P. W. Hess, R. Islam, K. Kim,
N. M. Linke, G. Pagano, P. Richerme, C. Senko, and N. Y. Yao. Programmable quantum simulations of
spin systems with trapped ions. Rev. Mod. Phys., 93:025001, Apr 2021.
[70] Thomas Monz, Philipp Schindler, Julio T. Barreiro, Michael Chwalla, Daniel Nigg, William A. Coish,
Maximilian Harlander, Wolfgang Hänsel, Markus Hennrich, and Rainer Blatt. 14-qubit entanglement:
Creation and coherence. Phys. Rev. Lett., 106:130506, Mar 2011.
[71] Giovanna Morigi, Jürgen Eschner, and Christoph H. Keitel. Ground state laser cooling using electro-
magnetically induced transparency. Phys. Rev. Lett., 85:4458–4461, Nov 2000.
[72] C. Ospelkaus, U. Warring, Y. Colombe, K. R. Brown, J. M. Amini, D. Leibfried, and D. J. Wineland.
Microwave quantum logic gates for trapped ions. Nature, 476:181–184, 2011.
[73] Sayan Patra, M. Germann, J.-Ph. Karr, M. Haidar, L. Hilico, V. I. Korobov, F. M. J. Cozijn, K. S. E.
Eikema, W. Ubachs, and J. C. J. Koelemeij. Proton-electron mass ratio from laser spectroscopy of hd+
at the part-per-trillion level. Science, 369(6508):1238–1241, 2020.
[74] J. M. Pino, J. M. Dreiling, C. Figgatt, J. P. Gaebler, S. A. Moses, M. S. Allman, C. H. Baldwin, M. Foss-
Feig, D. Hayes, K. Mayer, C. Ryan-Anderson, and B. Neyenhuis. Demonstration of the trapped-ion
quantum ccd computer architecture. Nature, 592(7853):209–213, 2021.
BIBLIOGRAPHY 89

[75] D. Porras and J. I. Cirac. Effective quantum spin systems with trapped ions. Phys. Rev. Lett., 92:207901,
May 2004.
[76] Karsten Pyka, Norbert Herschbach, Jonas Keller, and TanjaE. Mehlstäubler. A high-precision segmented
paul trap with minimized micromotion for an optical multiple-ion clock. Applied Physics B, 114(1-2):231–
241, 2014.
[77] Norman F. Ramsey. A molecular beam resonance method with separated oscillating fields. Phys. Rev.,
78:695–699, Jun 1950.
[78] Ch. Roos. Controlling the quantum state of trapped ions. Ph.D. thesis, Universität Innsbruck, Austria,
2000.
[79] P. O. Schmidt, T. Rosenband, C. Langer, W. M. Itano, J. C. Bergquist, and D. J. Wineland. Spectroscopy
using quantum logic. Science, 309:749–752, 2005.
[80] T Schneider, B Roth, H Duncker, I Ernsting, and S Schiller. All-optical preparation of molecular ions in
the rovibrational ground state. Nature Physics, 6(4):275–278, March 2010.
[81] P. F. Staanum, K. Højbjerre, P. S. Skyt, A. K. Hansen, and M. Drewsen. Rotational laser cooling of
vibrationally and translationally cold molecular ions. Nature Physics, 6:271–274, 2010.
[82] L. J. Stephenson, D. P. Nadlinger, B. C. Nichol, S. An, P. Drmota, T. G. Ballance, K. Thirumalai, J. F.
Goodwin, D. M. Lucas, and C. J. Ballance. High-rate, high-fidelity entanglement of qubits across an
elementary quantum network. Phys. Rev. Lett., 124:110501, Mar 2020.
[83] S. L. Todaro, V. B. Verma, K. C. McCormick, D. T. C. Allcock, R. P. Mirin, D. J. Wineland, S. W.
Nam, A. C. Wilson, D. Leibfried, and D. H. Slichter. State readout of a trapped ion qubit using a
trap-integrated superconducting photon detector. Phys. Rev. Lett., 126:010501, Jan 2021.
[84] Andrei Tokmakoff. Book: Time Dependent Quantum Mechanics and Spectroscopy (Tokmakoff), 12 2020.
[Online; accessed 2021-03-23].
[85] Q. A. Turchette, D. Kielpinski, B. E. King, D. Leibfried, D. M. Meekhof, C. J. Myatt, M. A. Rowe, C. A.
Sackett, C. S. Wood, W. M. Itano, C. Monroe, and D. J. Wineland. Heating of trapped ions from the
quantum ground state. Phys. Rev. A, 61:063418, 2000.
[86] Q. A. Turchette, C. J. Myatt, B. E. King, C. A. Sackett, D. Kielpinski, W. M. Itano, C. Monroe, and D. J.
Wineland. Decoherence and decay of motional quantum states of a trapped atom coupled to engineered
reservoirs. Phys. Rev. A, 62:053807, Oct 2000.
[87] Jean-Philippe Uzan. Varying constants, gravitation and cosmology. Living Reviews in Relativity, 14(1):2,
Mar 2011.
[88] R. F. C. Vessot, M. W. Levine, E. M. Mattison, E. L. Blomberg, T. E. Hoffman, G. U. Nystrom, B. F.
Farrel, R. Decher, P. B. Eby, C. R. Baugher, J. W. Watts, D. L. Teuber, and F. D. Wills. Test of
relativistic gravitation with a space-borne hydrogen maser. Phys. Rev. Lett., 45:2081–2084, Dec 1980.
[89] A. Walther, F. Ziesel, T. Ruster, S. T. Dawkins, K. Ott, M. Hettrich, K. Singer, F. Schmidt-Kaler, and
U. Poschinger. Controlling fast transport of cold trapped ions. Phys. Rev. Lett., 109:080501, Aug 2012.
[90] Yong Wan, Daniel Kienzler, Stephen D. Erickson, Karl H. Mayer, Ting Rei Tan, Jenny J. Wu, Hilma M.
Vasconcelos, Scott Glancy, Emanuel Knill, David J. Wineland, Andrew C. Wilson, and Dietrich Leibfried.
Quantum gate teleportation between separated qubits in a trapped-ion processor. Science, 364(6443):875–
878, 2019.
[91] D. J. Wineland, C. Monroe, W. M. Itano, D. Leibfried, B. E. King, and D. M. Meekhof. Experimental
issues in coherent quantum-state manipulation of trapped atomic ions. J. Res. Natl. Inst. Stand. Technol.,
103:259–328, 1998.
[92] Fabian Wolf, Yong Wan, Jan C. Heip, Florian Gebert, Chunyan Shi, and Piet O. Schmidt. Non-destructive
state detection for quantum logic spectroscopy of molecular ions. Nature, 530(7591):457–460, 2016.
[93] K. Wright, K. M. Beck, S. Debnath, J. M. Amini, Y. Nam, N. Grzesiak, J. S. Chen, N. C. Pisenti,
M. Chmielewski, C. Collins, K. M. Hudek, J. Mizrahi, J. D. Wong-Campos, S. Allen, J. Apisdorf,
P. Solomon, M. Williams, A. M. Ducore, A. Blinov, S. M. Kreikemeier, V. Chaplin, M. Keesan, C. Monroe,
and J. Kim. Benchmarking an 11-qubit quantum computer. Nature Communications, 10(1):5464, 2019.
90 BIBLIOGRAPHY

[94] Shi-Liang Zhu, C. Monroe, and L.-M. Duan. Trapped ion quantum computation with transverse phonon
modes. Phys. Rev. Lett., 97:050505, Aug 2006.