You are on page 1of 53

Magnon dispersion in

ferromagnetic spinor
Bose–Einstein condensates

Sam Fischer

A thesis presented for the degree of


Bachelor of Science Advanced with Honours

School of Physics
Monash University
Australia

November 14, 2014


Magnon dispersion in ferromagnetic
spinor Bose–Einstein condensates

Sam Fischer

Supervisory panel:
Dr Lincoln Turner
Dr Russell Anderson
Dr Tapio Simula

Abstract

Magnons are elementary excitations to a magnetically-ordered condensed matter


system. The effective mass and dispersion relation for magnons in a ferromagnetic
spinor Bose–Einstein condensate have recently been measured [Marti, et al., Phys.
Rev. Lett. 113, 155302 (2014)], but the measured mass disagrees with both mean-
field and beyond mean-field predictions. In this thesis, I describe one-dimensional
mean-field simulations of spinor condensates with specifically engineered spin ex-
citations. By simulating multiple techniques to probe the effective magnon mass,
we observe a mass deviation from the mean-field result for linear excitations, which
increases with the strength of the spin excitation. However, the functional form of
this deviation is found to depend on the particular type of spin excitation.

ii
Acknowledgements

There are many people to whom I must express my gratitude. It is only with the
support of these people that I have managed to achieve this milestone.
Firstly I must thank Mr. Mark Ellis, my high school physics teacher. He was the first
of many to inspire me to pursue an understanding of the world around me, and I may not
have reached my current level of appreciation for science without his guidance and fine
teaching skills.
Next I must thank my current supervisors and mentors — Drs. Lincoln Turner, Russell
Anderson and Tapio Simula (or as I like to refer to them, “The Three Musketeers”). I don’t
think I could have possibly had a better team to work with all year on this project. With
the combination of Tapio’s deep theoretical insight, Lincoln’s practical perspective and
Russ’s hands-on assistance, you have all given me a fine education into ultracold quantum
gases and in how to conduct research in the most effective manner. I have appreciated
every minute of working with the three of you, and I look forward to continuing that work
with you all in the years to come.
I must also thank the PhD students of the School of Physics for providing such a
supportive working environment. I would especially like to thank Martijn, Lisa, Phil and
Shaun for their support with countless queries and computational problems.
Last but not least are my fellow Honours students. You’ve all made for an exhilarating
year, and I have appreciated sharing this journey with you. Special mentions must go
to Ando, Mithuna, and Andrew (the ‘honorary’ Honours student) for keeping me sane
towards the end.

iii
Contents
1 Introduction 1

2 Background theory 2
2.1 Basic condensate physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.1.1 The Gross–Pitaevskii equation . . . . . . . . . . . . . . . . . . . . . . . . 2
2.1.2 The Thomas–Fermi Approximation . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Condensate excitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2.1 The Bogoliubov–de Gennes equations . . . . . . . . . . . . . . . . . . . . 5
2.2.2 Phonons and collective modes . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Spinor condensates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3.1 Three component Gross–Pitaevskii equation . . . . . . . . . . . . . . . . . 8
2.3.2 Ground state phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3.3 Spin excitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3 Methods 12
3.1 PyGPE simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2 Spin rotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4 Magnon contrast interferometry 15


4.1 Coherent magnon optics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.2 Analytical investigation into standing spin waves . . . . . . . . . . . . . . . . . . 17
4.3 Magnon interferometry simulations . . . . . . . . . . . . . . . . . . . . . . . . . . 19

5 Dispersion of monochromatic magnons 23


5.1 Simulating magnon propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
5.2 Frames of reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.3 Magnon dispersion relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.4 Interpretation of observed magnon mass variation . . . . . . . . . . . . . . . . . . 30
5.5 Comparing masses for standing and running magnons . . . . . . . . . . . . . . . 30

6 Gaussian wavepackets of magnons 33


6.1 Measuring the effective mass from wavepacket dispersion . . . . . . . . . . . . . . 33
6.2 Wavepacket expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6.3 Dispersion of Gaussian spin tip profile . . . . . . . . . . . . . . . . . . . . . . . . 37

7 Conclusion 40

8 Appendices 41
A Simulations of density excitations in scalar condensates . . . . . . . . . . . . . . 41
B Monochromatic magnons imprinted using magnetic field gradients . . . . . . . . 42

9 References 44

iv
1 Introduction
Following the pioneering theory work by Bose [1] and Einstein [2] in the mid-1920’s, the
study of Bose–Einstein condensates (BECs) has attracted much scientific attention in the
last two decades, following their experimental realisation in ultracold atomic gases [3–5].
These quantum fluids [6, 7] incorporate aspects from almost all areas of physics, ranging
from quantum and statistical mechanics, to optical, atomic and condensed matter physics.
Spinor condensates [8, 9] are a particularly interesting variety of BECs, possessing a spin
degree of freedom that admits a rich assortment of magnetic phenomena [10], including
spin textures [11] and domain walls [12]. These characteristic features of spinor conden-
sates are built around a framework of elementary magnetic excitations known as magnons.
Proposed by Bloch [13] in 1930 to explain excitations of solid state ferromagnets, these
quasiparticle excitations quantise the collective propagation of spin disturbances through
a magnetised medium.
In a spinor condensate, magnons are predicted to disperse quadratically, like free parti-
cles [14,15]. This thesis investigates the magnon dispersion relation, and more specifically
the effective mass of magnons within the nonlinear environment of a spinor condensate.
Our investigation is motivated by the recently reported experimental measurement of
effective magnon mass in a ferromagnetic condensate by Marti, et al. [16], in which an
anomalously heavy magnon mass was observed. By performing simulations of spin exci-
tations in the mean-field regime, we find a strong correlation between the strength of the
spin perturbation and the effective magnon mass. While our mean-field results cannot
alone explain the findings of Marti, et al., we substantiate the extreme caution that must
be exercised in reporting an effective magnon mass outside the limit of small excitations.
Following a summary of the relevant background theory in Section 2, we describe
the computational methods used in our investigations in Section 3. This section also
explains the spin rotation formalism used throughout this thesis. Section 4 discusses the
recent experimental measurement of the effective magnon mass and dispersion relation,
including our analytical investigations and simulations to understand this result. We then
describe our own method for measuring magnon properties in Section 5, by simulating
the propagation of monochromatic magnons. We present simulation results that indicate
that the effective mass varies as a function of the magnon population, but this variation
is speculated to result from counter-propagating magnons associated with the alternate
ferromagnetic ground state. In Section 6, we employ a different approach for measuring
the effective magnon mass, by tracking the spread of a Gaussian wavepacket of magnons.
For this method, we observe a different functional form for the mass variation to that
found in Section 5, however both methods present identical results in the single magnon
limit.

1
2 Background theory
2.1 Basic condensate physics
The phase transition experienced by a dilute gas of bosons when cooled to sub-Kelvin
temperatures is referred to as Bose–Einstein condensation. The exploration of such ex-
treme conditions is vindicated by the unique properties associated with the condensed
state. The most remarkable of these properties is atomic coherence — a BEC can be con-
sidered as a coherent superposition of atoms, similar to the manner in which a laser is a
coherent superposition of photons [17]. This coherence makes many quantum mechanical
phenomena observable on a macroscopic scale [18].
Bose–Einstein condensation can occur only for bosons, as they are unencumbered
by the Pauli exclusion principle, and therefore capable of mutually occupying a single
quantum state. This phenomenon can be mathematically described by the Bose–Einstein
distribution,
1
N (ε) =   , (1)
exp kε−µ
BT
− 1

which describes the mean number N of bosons in a given energy state ε, where kB is
Boltzmann’s constant, T is the temperature and µ is the chemical potential [6, 19, 20].
This distribution shows that a macroscopic occupation number of the state nearest to the
chemical potential µ is allowed.
In ordinary, everyday conditions, each particle in a gas of bosons has a de Broglie
wavelength λdB so small compared to the interparticle separation that they are still dis-
tinguishable [21], and we do not observe condensation. However, as the temperature of
the gas is decreased, the de Broglie wavelengths of the particles increase, and the Bose–
Einstein statistics that describe the particles become more important. Once a critical tem-
perature Tcrit is reached, typically on the order of 100 nK [22], these wavelengths become
comparable in size to the average interatomic separation, and the individual wavepackets
begin to coherently superpose, forming a “giant matter wave” [17, 23]. This is the BEC
phase transition, shown in Figure 1, with the order parameter ψ for the transition often
referred to as a macroscopic wavefunction.

2.1.1 The Gross–Pitaevskii equation

At zero temperature, the behaviour of the condensate is well described by Gross–Pitaevskii


theory [6, 22]. While each of the N particles in the condensate experiences a potential
from the other N −1 particles, one can approximate this potential as a short-range contact
potential g0 δ(~r), were ~r is the interparticle separation [6]. The coupling constant

4π~2 a0
g0 = , (2)
m

2
Figure 1: Phase transition for Bose–Einstein condensation. At room temperature, a gas
of bosons may be modelled as a system of billiard balls. As the temperature of the
particles is decreased, their de Broglie wavelengths λdB increase and the classical billiard
ball description becomes unsuitable. At the critical temperature, λdB approaches the
average interatomic spacing and the particles condense into the BEC ground state. At
zero temperature, all particles have condensed out of the thermal cloud, leaving a ‘pure’
condensate. Reproduced from Reference [23].

characterises the collisional interaction strength in terms of particle mass m and the two-
body s-wave scattering length a0 [24].
This contact interaction term can be incorporated into the Schrödinger equation to
give the most basic mathematical description of a BEC, the (time-independent) Gross–
Pitaevskii equation (GPE):

~2 2
 
2
− ∇ + Vtrap (~r) + g0 |ψ (~r)| ψ (~r) = µ ψ (~r) . (3)
2m

where Vtrap (~r) is the external trapping potential1 and the eigenvalue µ is the chemical
potential [6,24]. Since the Hamiltonian of the GPE (left hand side of Eq. 3) depends upon
the solution ψ(~r) of the equation, the GPE is also referred to as the nonlinear Schrödinger
equation. Another key difference between the ‘conventional’ Schrödinger equation and
the GPE is that the squared modulus of ψ(~r) no longer gives the probability density of
1
In an experimental setting, these are typically harmonic oscillator potentials. In simulation, we can
also consider homogeneous potentials to observe the behaviour of condensates with a uniform density.

3
finding a single particle at position ~r, but rather the actual density of particles there2 .
This becomes apparent when we view the condensate wavefunction ψ as a function of the
condensate density ρ(~r) and phase φ(~r) according to
p
ψ(~r) = ρ(~r) eiφ(~r) . (4)
R
Using Eq. 4, we may also define the number of condensed particles as N = ρ(~r) d~r.
Gross–Pitaevskii theory assumes a zero temperature condensate, however for tem-
peratures well below Tcrit — when the condensed component of the gas is much larger
than the uncondensed thermal cloud — the zero temperature condensate remains a good
approximation for many realistic situations [25, 26]. For higher temperatures, various
other theories have been formulated to describe the effect of the thermal cloud. Some
of these approaches involve treating the uncondensed component as an additional po-
tential (Hartree–Fock–Bogoliubov and Popov approximations [6, 25]), considering macro-
scopic occupation of multiple energy eigenstates (the stochastic projected GPE [26, 27])
or describing the condensate and thermal cloud with coupled hydrodynamic equations
(Zaremba–Nikuni–Griffin formalism [25, 28]).

2.1.2 The Thomas–Fermi Approximation

While, in general, the GPE has no analytic solution3 , one can obtain an approximation
for ψ(~r) for condensates with a large atom number. In this regime, the kinetic energy is
small compared to the interaction energy, characterised by g0 |ψ (~r)|2 . Accordingly, we can
neglect the kinetic energy term in the GPE; this is known as the Thomas–Fermi approxi-
mation [6]. The corresponding expression for the condensate density in the Thomas–Fermi
regime [29] is given by

µ − Vtrap (~r)
 
2
ρ (~r) = |ψ (~r)| = max ,0 . (5)
g0

It should be noted that this approximation fails near the surface of the condensate cloud,
since the Thomas–Fermi density possesses a hard edge, whereas the true condensate
density smoothly decays to zero near the edges.

2
We have implicitly defined the condensate wavefunction ψ(~r) this way by normalising it to the particle
number N . By instead including the particle number in our definition of the GPE, the wavefunction may
be described identically as for the Schrödinger equation wavefunction [22].
3
Solutions exist in some trapping potentials, however none exist for a harmonic trap.

4
2.2 Condensate excitations
In the previous section, we limited our discussions to condensates in their time-
independent ground states. It would be remiss to ignore the time-dependent behaviour
of the condensate, as the dynamical properties of these quantum fluids are affected by
phenomena such as thermal cloud interactions [25] and quantum depletion resulting from
particle interactions [30]. From an experimental perspective, condensate dynamics mani-
fest as a collective motion of the cloud, whereas theoretically they can be described by a
sum of elementary excitations.

2.2.1 The Bogoliubov–de Gennes equations

Here we demonstrate the simplest formalism for describing condensate excitations, the
Bogoliubov–de Gennes formalism, where we begin by asserting time dependence onto the
GPE. This gives the generalised time-dependent Gross–Pitaevskii equation [6],

~2 2
 
∂ 2
i~ ψ (~r, t) = − ∇ + Vtrap (~r, t) + g0 |ψ (~r, t)| ψ (~r, t) . (6)
∂t 2m

When the time-dependent wavefunction takes the form ψ (~r, t) = ψ (~r) e−iµt/~ , it can be
applied to (6) to recover the time-independent GPE.
The time-dependent GPE allows us to calculate the spectrum of elementary excitations
for the condensate [31], which is “an essential ingredient in calculations of thermodynamic
properties” [6]. We approach the problem from the perspective of linear response theory4 ,
where we treat the wavefunction as a sum of the ground state wavefunction and a time-
dependent perturbation:

ψ (~r, t) = (ψ (~r) + pk δψk (~r, t)) e−iµt/~ , (7)

where the perturbation takes the form

δψk (~r, t) = uk (~r)e−iωk t + vk∗ (~r)eiωk t . (8)

Here ωk is the eigenfrequency of the excitation mode with wavevector ~k and wavenumber
k = |~k|, and uk (~r) and vk (~r) are functions that quantify the amplitude of the particle-like
and hole-like excitations respectively5 [8, 24]. We also note that pk indicates the fraction
of condensed particles excited to a given excitation mode [32].
4
This is one of many possible approaches to calculate the excitation spectrum, such as deriving the
hydrodynamic equations for the condensate [24] or utilising the microscopic theory of the Bose gas, which
entails second quantisation of the boson field operators [22]. Each equivalent approach reveals different
insights into the nature of the condensate.
5
The amplitude of excitation represents the number of particles excited out of the condensate ground
state. This model for condensate excitation is valid only when the number of excited particles is far
outweighed by the total number of condensed particles.

5
Applying ansatz (7) to the time-dependent GPE allows us to derive a pair of coupled
equations in terms of uk (~r) and vk (~r),
! ! !
H0 − µ + 2g0 ψ 2 (~r) g0 ψ 2 (~r) uk (~r) uk (~r)
= ~ωk , (9)
−g0 ψ 2 (~r) − (H0 − µ + 2g0 ψ 2 (~r)) vk (~r) vk (~r)

where H0 = (−(~2 /2m)∇2 + Vtrap (~r)). These are known as the Bogoliubov–de Gennes
(BdG) equations [24, 31, 33], and they allow us to uncover a ‘periodic table’ of the ele-
mentary excitations for a Bose–Einstein condensate.
While, in general, these equations cannot be solved analytically to determine the
quasiparticle wavefunctions uk (~r) and vk (~r), they may be solved numerically6 [35–37].
Nevertheless, one can still determine the spectrum of energy eigenvalues analytically by
diagonalising the BdG equations [38], with the resulting eigenvalues taking the form
p
~ωk = k (k + 2g0 n), (10)

where n is the spatially uniform condensate density7 . This dispersion relation corresponds
to a free-particle spectrum k = ~2 k 2 /(2m) with an additional mean-field contribution [6].

2.2.2 Phonons and collective modes

For excitations of large momentum, condensate excitations are free-particle-like, following


an approximate dispersion ~ωk = k [24]. Physically, these excitations correspond to a
particle leaving the condensate ground state with a momentum of ~~k [6]. For low-lying
excitations (i.e. low k), however, the excitations are superpositions of particle-like and
hole-like excitations uk and vk respectively. In this regime, the dispersion relation takes
the form ωk = ck, where r
g0 n
c= . (11)
m
The linear dependence of ωk on k implies that low energy excitations propagate through
the condensate as phonons with speed c [39]. For low k, excitations are a collective
phenomenon, with their behaviour manifesting throughout the condensate at large. Each
excitation energy eigenvalue has an associated collective mode, with unique macroscopic
behaviour [40]. Below we list some of the lowest-lying collective modes and their observed
behaviour:
6
The solutions form a spectrum of excited states, and by projecting this set of states onto the con-
densate ground-state wavefunction, the time evolution of the condensate can then be determined [34]. In
Appendix A, we describe simulations in which we perform this projection, and observe one-dimensional
low-lying collective modes.
7
Note that this excitation spectrum was derived for a homogeneous condensate, as are all results given
henceforth that rely on a constant density n. The effects of the trapping potential are paramount to the
very nature of the condensate.

6
(i) Slosh mode: Also known as the dipole mode, in reference to the mode’s angular
momentum, this is the lowest energy collective excitation mode. It is observed as a
rigid-body centre of mass oscillation of the condensate [23].

(ii) Breathing mode: This mode manifests as a periodic and axially symmetric expansion
and contraction of the condensate radius [41, 42].

(iii) Quadrupole surface modes: Two degenerate modes which result in a periodic vari-
ation in the condensate’s cross-sectional aspect ratio, with the density profile of the
condensate varying between circular and elliptical [41–43].

(iv) Scissors mode: Manifested as an elliptically deformed condensate which oscillates


slightly about its equilibrium position [43]. The importance of the scissors mode
stems from the evidence it provides for the emergence of superfluidity in BECs
[44–47].

(v) Higher order surface modes: Many excitation modes of higher energy than those
listed (e.g. the hexadecapole mode) occur as standing waves along the circumference
of the condensate, which leave the centre of the condensate relatively unperturbed
[48]. These strong surface perturbations can lead to the development of topological
defects in the form of quantised vortices [49].

As an aside, we also introduce the concept of the healing length ξ. This characteristic
length scale represents the minimum distance over which the condensate can “heal” from
a perturbation8 . It also indicates the typical sizes of cores of the aforementioned vortices
[24].

2.3 Spinor condensates


In describing BECs so far, we have implicitly considered only ‘scalar’ condensates [10], i.e.
condensates comprised of particles in a single Zeeman state. For condensed particles with
spin F , each of the associated Zeeman states mF = 2F + 1 may be trapped simultaneously.
This affords the condensate an additional spin degree of freedom [8], which allows for spin-
exchange interactions [9], and merits the so-called ‘spinor’ condensate as a unique tool
for investigating magnetic ordering phenomena. While many of these phenomena can
be explored with spinor condensates of different atomic spins, the work described in this
thesis deals exclusively with spin-1 condensates, so we shall henceforth refer to these
specifically when describing “spinor” BECs.
8
Both density and spin perturbations have respective healing lengths ξn and ξs [50].

7
2.3.1 Three component Gross–Pitaevskii equation

The previously considered order parameter ψ only suitably described a single-component


condensate. A spinor condensate must instead be described by a vector wavefunction,
   p 
ψ−1 (~r) ρ−1 (~r)eiφ−1 (~r)
  p
Ψ (~r) =  ψ0 (~r)  =  ρ0 (~r)eiφ0 (~r)  , (12)
 
p
ψ+1 (~r) ρ+1 (~r)eiφ+1 (~r)

with each component, with spin projection mF = −1, 0 and +1 respectively, being de-
scribed by its own wavefunction component ψmF , density ρmF and phase φmF [51]. We
may simplify our spinor wavefunction under the single-mode approximation (SMA), in
which the spatial and spin degrees of freedom become decoupled [52]. This approxi-
mation reasonably describes a condensate with weak spin-dependent interactions9 whose
components all occupy the same region of space [53]. Under the SMA, we cast the wave-
function as a product of the total condensate density ρ(~r) and the magnetisation vector
ζ, a spinor which defines the population fraction in each spin component [54]:
p
Ψ (~r) = ρ(~r)ζ(~r). (13)

In addition to the spin-conserving interaction parametrised by g0 , spinor condensates


also allow for spin-exchange interactions. In this process, two mF = 0 particles may be
converted into an mF = ±1 pair, and vice versa10 [55]. This additional phenomenon
requires not one, but two coupling constants [51],

4π~2 a0 + 2a2
g0 = (14)
m 3

4π~2 a2 − a0
g2 = , (15)
m 3
where a0 and a2 are the scattering lengths for collisions of particles with total spin 0
and total spin 2, respectively. The coupling constants (14) and (15) allow for population
redistribution between the three spin projections [6], and hence for intricate spin dynamics
and textures. Without these coupling terms, a spinor BEC would behave no differently
to three distinct scalar BECs overlapping in space.
Magnetic fields heavily influence spinor condensates, since they induce Zeeman shifts to
the energy levels of each component. These shifts effectively determine which components
act as the ground state, and control the relative populations of each component [9, 51].
9
In comparison to the spin-independent interaction strength.
10
For spin-1 condensates, this population mixing takes place without atoms being lost from the trap,
however for other multicomponent condensates, analogous behaviours can be detrimental to sustaining
the condensed fraction [8, 51].

8
In the presence of a magnetic field with linear and quadratic Zeeman energies p and q
respectively, the ground state of the condensate may be described by a set of coupled
GPEs [8] given by

~2 2
 
g2
− ∇ + Vtrap (~r) + p + q + g0 ρ (~r) − g2 hFz i − µ ψ−1 + √ hF+ iψ0 = 0 , (16)
2m 2

~2 2
 
g2 g
− ∇ + Vtrap (~r) + g0 ρ (~r) − µ ψ0 + √ hF+ iψ+1 + √2 hF− iψ−1 = 0 (17)
2m 2 2
~2 2
 
g
− ∇ + Vtrap (~r) − p + q + g0 ρ (~r) + g2 hFz i − µ ψ+1 + √2 hF− iψ0 = 0, (18)
2m 2
where hF± i ≡ hFx i ± ihFy i and hFz i refer to the expectation values of the angular
momentum projection operators11 for a spin-1 particle,

1  ∗
ψ0 + ψ0∗ (ψ−1 + ψ+1 ) + ψ+1


hFx i = √ ψ−1 ψ0 (19)
2

i  ∗
ψ0 + ψ0∗ (ψ−1 − ψ+1 ) + ψ+1


hFy i = √ −ψ−1 ψ0 (20)
2
hFz i = |ψ−1 |2 − |ψ+1 |2 . (21)

2.3.2 Ground state phases

One important feature of spinor condensates is their ability to support multiple


magnetically-ordered ground state phases. The ground state phase is determined by
the strength of the terms in the spinor Hamiltonian that affect spin orientation, namely
the spin coupling constant g2 , and the linear and quadratic Zeeman interactions p and q
respectively [9]; these dependencies are illustrated in Figure 2. A spin-1 BEC may exist
in one of four spin ordering manifolds, known as the ferromagnetic, polar, antiferromag-
netic and broken-axisymmetric phases [8, 56]. The work in this thesis deals entirely with
ferromagnetic condensates, for which the spins are fully magnetised, i.e. mF |ζmF |2 = 1.
P

In the absence of a magnetic field, the energetically-favoured ground state for this phase
is equally represented by either ζ = (1, 0, 0)T or ζ = (0, 0, 1)T , where T is the transpose
operator.

11
We shall explicitly define these operators in Section 3.2, where they describe spin rotations on the
Bloch sphere.

9
(a ) c1 > 0 (b) c1 < 0
p/c 1 n p/| c 1| n
p=q

F F
p = q + 12 c 1 n
1
p2 = 2 c 1 n q P
AF BA
1/2 q/c 1n 2 q/|c 1|n
P
−1

F F
p2= q 2− 2|c 1| nq

Figure 2: Phase diagrams for spin-1 BEC with (a) positive, and (b) negative, spin-
dependent interactions c1 . In this figure, c1 refers to the constant we have named g2
and n refers to the condensate number density. The vertical and horizontal axes are the
linear and quadratic Zeeman energies p and q respectively, in units of c1 n. Here we show
the (F) ferromagnetic, (P) polar, (AF) antiferromagnetic, and (BA) broken-axisymmetric
phases, with thick lines marking the phase boundaries. Reproduced from Reference [50].

2.3.3 Spin excitations

The BdG formalism established in Section 2.2 can also been applied to spinor condensates
to find their excitation modes [8, 14, 15]. Each spinor phase supports unique excitation
spectra, but we shall present the results only for the ferromagnetic case, in which we
obtain three excitation branches: a single phonon branch and two magnon branches. Their
dispersion relations are given in Table 1, and are shown diagrammatically in Figure 3.
The phonon branch is reminiscent of the dispersion relation for a scalar condensate,
in that the functional form also describes particles in the same spin state obeying lin-
ear dispersion for low k. The two magnon branches, however, exhibit quadratic disper-
sion, with nonzero energy gaps which depend on the applied magnetic field and spin-spin

Table 1: Dispersion relations for the three excitation branches for a spin-1 condensate.
Here we define the magnon dispersion m 2 2
k = ~ k /(2m∗ ), with m∗ being the effective
magnon mass. At high momentum, all three branches describe particle-like excitations
(i.e. excitations with quadratic dispersions), while the phonon branch exhibits linear
dispersion for low-lying excitations.

Excitation branch Quasiparticle ∆mF pDispersion relation


Density Phonon 0 k [k + 2(g0 + g2 )n]
Transverse spin Magnon 1 m
k +p−q
Quadrupolar spin Magnon 2 m
k + 2p − 2g2 n

10
12 50
(a) (b)
40

0 n
= o
ν hon
p
8 1/ξn
30
Ekν /c1 n

1/ξs on
gn
e ma
ers
nsv 20
tra 2
4 ν=
1/ξs
F
Eg,2 10
g non
ma 1
F ν=
Eg,1
0 0
0.0 0.5 1.0 1.5 2.0 2.5 0 2 4 6 8 10
k ξs k ξs

Figure 3: Dispersion relation for the ferromagnetic phase, where again c1 refers to g2 .
Subplots (a) and (b) focus on different ranges of k values, with the k axis given in terms
of the spin healing length ξs ; we also mark the density healing length ξn . The phonon
(black solid line), quadrupolar magnon (blue dashed line), and transverse magnon (red
dash-dot line) branches of the excitation spectra are shown, labelled with indices ν = 0,
F
1, 2, respectively. The magnon gaps are given by Eg,ν . Here we have used the parameters
p = 1.5g2 n, q = −g2 n, g0 = 20g2 and g2 > 0. Reproduced from Reference [50].

interaction strengths [50]. Transverse spin excitations describe ground state particles be-
ing ‘promoted’ to the mF = 0 state (i.e. ∆mF = 1), while quadrupolar spin excitations
(also referred to as the nematic magnon branch) describe particles excited from the fully
magnetised ground state to the opposite fully magnetised state (∆mF = 2)12 .
As free particles also exhibit quadratic dispersion, magnons from either branch can
be thought of as free particles trapped within the potential of the ferromagnetic ground
state. The speed of magnon dispersion within this medium is governed by the effective
magnon mass m∗ . Within the framework of mean-field theory, the effective magnon mass
is predicted to be exactly equal to the bare atomic mass of the condensed species, in the
limit of small excitations [14, 15]. When we consider the effects of quantum depletion
of uncondensed atoms, as described by beyond mean-field theory13 [30, 57], we obtain a
magnon mass increase of 0.3% from the bare atomic mass [58].
The effective magnon mass has recently been measured for transverse spin excitations
in a ferromagnetic spinor condensate [16]. The measured result featured a 3.3% deviation
from the bare atomic mass, a measurement that cannot be explained by mean-field and
beyond mean-field theories. Investigating this anomalously large mass for transverse
magnons is the focus of the remainder of this thesis.

12
We must emphasise that, as these excitations were derived using linear response theory (i.e. BdG
theory), they are only valid for small amplitude excitations to the condensate in its ground state.
13
This formalism, otherwise known as Beliaev theory, provides a second order correction to the Bogoli-
ubov excitation spectra.

11
3 Methods
In this section, we briefly explain the code used to simulate spin excitations for this
project14 . We also introduce the general theory of spin rotations, which is relevant to all
of the simulations described in this thesis.

3.1 PyGPE simulations


The simulations described in this thesis were performed using a mean-field GPE simulator
written in Python, referred to as “PyGPE”15 . We note that, by default, this simulator
operates in a frame rotating at the Larmor frequency, so that the linear Zeeman term p
is removed from the spinor Hamiltonian. The simulator takes as its input parameters of
the condensate, such as atom number and nonlinear interaction strength, as well as the
magnetic field and spin-independent trapping potential. An initial guess for the ground
state wavefunction, a fully magnetised ferromagnetic 87 Rb spinor condensate with all
atoms in either the mF = −1 or +1 Zeeman state, is then specified. The density profile
of this wavefunction is determined by the specified trapping potential geometry; either
harmonic or homogeneous.
We then evolve this initial wavefunction in imaginary time using a split-step Fast
Fourier Transform (FFT) technique to reduce the error associated with separately eval-
uating the kinetic and potential energies of the unitary time evolution operator [59, 60].
Imaginary time propagation – otherwise known as a Wick rotation – transforms the GPE
from a wave equation into a diffusion equation. Propagating in imaginary time damps out
the high energy components of the initial wavefunction, converging towards the ground
state.
To investigate spin dynamics in our simulations, we then apply a norm-preserving
spin rotation operator to our newly-found ground state. This unitary operation redis-
tributes the atom populations among the three spin components. The spin rotation can
be spatially-dependent or uniform in nature, depending on the particular spin excitation.
Following the spin tip, the condensate wavefunction is propagated in real time, once again
using the split-step FFT technique. The simulator returns the wavefunction evaluated at
each timestep, allowing post-simulation processing to be performed on these data.

14
In Appendix A, we also describe simulations performed with a different simulator, which investigate
density excitations to scalar condensates.
15
We thank Martijn Jasperse for writing the time propagation code for this simulator.

12
3.2 Spin rotations
The spinor representation of the condensate allows us to express spin rotations (equiv-
alent to transverse spin excitations for small rotations) in a convenient manner. The
magnetisation vector ζ, introduced in Section 2.3.1, describes the relative densities of
each component of the condensate as a vector on the Bloch sphere. For the ferromagnetic
condensate which we will be concerned with in this thesis, ζ is a spinor of unit length,
and can be described by spherical coordinates (θ, φ). In its ground state, a ferromagnetic
condensate will always be exclusively in either the mF = −1 or +1 Zeeman state, corre-
sponding to polar angles θ = 0 and θ = π respectively, and we can describe an excitation
of this ground state as a rotation of the magnetisation vector.
We begin by defining the angular momentum projection operators for a spin-1 particle
as      
0 1 0 0 −1 0 1 0 0
1  i 
Fx = √ 1 0 1, Fy = √ 1 0 −1, Fz = 0 0 0 . (22)
   
2 2
0 1 0 0 1 0 0 0 −1
We can use these operators to define any arbitrary rotation of the spinor ζ in spin-space.
Starting from the mF = −1 state, all of the spin rotations we perform in our work involve
a rotation of the condensate spin by some polar angle θ about the y-axis, followed by a
rotation about the z-axis by some azimuthal angle φ [61]. Applying such rotations to a
vector in spin-space entails applying the respective rotation operators to the magnetisation
vector ζ, giving an overall unitary rotation operator of
  
e−iφ cos2 θ
− √12 e−iφ sin θ e−iφ sin2 θ

2 2
U = e−iFz φ e−iFy θ = 
 
√1 − √12 sin θ 
 2
sin θ cos θ . (23)
e sin2
iφ θ √1 eiφ e−iφ cos2 2θ
 
2 2
sin θ

Applying this to the ferromagnetic mF = −1 ground state, ζ = (1, 0, 0)T , gives a rotated
spinor
    T
−iφ 2 θ 1 iφ 2 θ
ζ = e cos , √ sin θ, e sin . (24)
2 2 2
Taking the spin expectation values with respect to this new spinor16 , we readily confirm
that the rotation operator in Eq. 23 rotates the spin expectation value to a point on the
Bloch sphere described by spherical coordinates (θ, φ):

(hFx i, hFy i, hFz i) = (sin θ cos φ, sin θ sin φ, cos θ). (25)

As the mF = +1 state serves as an equally valid ferromagnetic ground state, we can


16
Under the single-mode approximation, the spin expectation values are computed as hFx,y,z i =
hζ|Fx,y,z |ζi [62].

13
similarly define a rotation matrix to rotate the spinor describing this state, ζ = (0, 0, 1)T .
However, to rotate the spinor to the same angle (θ, φ) on the Bloch sphere as for a rotation
starting from mF = −1, we must rotate about the y-axis by an angle (π+θ). Applying the
corresponding spin tip operator U = e−iFz φ e−iFy (π+θ) to the mF = +1 spinor ζ = (0, 0, 1)T
gives a rotated spinor identical to Eq. 24. Hence whether our ground state is defined
as mF = −1 or +1, we can perform rotations that result in an identical state. For the
simulations performed in this body of work, we have set our condensate to initially be in
the mF = −1 state.

14
4 Magnon contrast interferometry
In this section, we describe the recently reported measurement of the effective magnon
mass in a ferromagnetic condensate [16]. The measured mass was anomalously heavier
than mean-field and beyond mean-field predictions, with the cause of the discrepancy
noticeably unidentified by the authors of the article. To investigate the result, we describe
an analytic framework for the induced spin dynamics, which we implement in our own
simulations of the reported experiment.

4.1 Coherent magnon optics


The basic properties of magnons, such as their dispersion relation, effective mass m∗ , mag-
netic moment and energy gap, were recently determined experimentally in a ferromagnetic
spinor condensate by Marti, et al. [16]. For the purposes of this investigation, we shall
discuss only their results concerning magnon mass and dispersion. These measurements
were obtained using their so-called magnon contrast interferometer, which measures the
contrast of density fringes in a magnon standing wave, using state-selective absorption
imaging.
The experiment began with a ferromagnetic 87 Rb condensate that was initially po-
larised in the mF = −1 state, in a trapping geometry reminiscent of a surfboard, i.e. an
effectively two-dimensional condensate with two different perpendicular trapping frequen-
cies. A series of spatially-dependent spin rotations were then applied to the condensate.
This was achieved using two equal-frequency, circularly polarised lasers that intersected at
a small relative angle; the resulting standing wave intensity pattern was then shone onto
the gas. The standing wave exhibited a spatially and temporally modulated intensity,
with the spatial modulation giving a sinusoidal variation in intensity along the elongated
y-axis, and a temporal modulation with a frequency matching the resonance condition for
population transfer between the mF = −1 and 0 Zeeman states. The resulting spatially-
dependent rotation of the polar angle on the Bloch sphere is given by

θmax
θ(x, y) = (1 + cos(ky)), (26)
2

where θmax is proportional to the maximum intensity Imax of the interfered beam, and
k is the wavevector of the imprinted standing wave. The wavevector imprinted onto the
condensate depended on the relative angle between the two laser beams.
For weak excitations, i.e. |θmax |  1, the resulting population modulation effectively
created a standing spin wave, or as the authors state, “a ferromagnetic condensate excited
with coherent populations of magnons at three wavevectors”17 : ~0 and ±~k. This stand-
ing wave set the initial condition for time evolution of the mF = 0 state, whereby the
17
The trivial wavevector ~k = ~0 corresponds to a spatially uniform spin tip.

15
(a) (c) 0 20 40 60 80 100
0.00

ω/ωfree − 1
B −0.02
−0.04
−0.06

24
25

ω/2π (Hz)
22

(ω(k)−ω(0))/2π (Hz)
20

20
15 0 2 4 6
(b) N m = 0 (104)
0.3
F
10
Contrast

0.2
5
0.1

0.0 0
0 50 100 150 200 250 0 20 40 60 80 100
Time (ms) k/2π (mm−1)

Figure 4: Experimental measurement of the magnon dispersion relation, obtained using


magnon contrast interferometry. (a) A standing wave of magnons was initially imprinted
onto the condensate, and allowed to evolve. The spatially periodic tilting of the magneti-
sation vector was observed as a striped pattern in the mF = 0 density, with this density
being proportional to the magnon density for small tip angles θ. For the condensate
shown, the imprinted standing wave had wavevector k = 2π/(15.4 µm), with a scalebar
50 µm wide shown for comparison. (b) The periodic spatial modulation of the magnon
density (open black circles) oscillated with a frequency of ωcontrast = 2(ω(k) − ω(0)) (fit-
ted red curve), measured by the contrast of the mF = 0 density images. Over many
cycles, the modulation amplitude decayed, and the contrast settled to a nonzero value,
possibly due to interactions with the thermal cloud. (c, bottom) The magnon dispersion
relation, obtained using contrast curves such as (b). The contrast oscillation frequencies
(black points) were calibrated for finite magnon densities, with such a calibration shown
in the inset. The fitted quadratic dispersion relation (red dashed line) lies below the free-
particle result (black line), indicating a heavier effective magnon mass than predicted.
(c,top) Fractional variation of data from the m∗ = mRb dispersion frequency. Reproduced
from Reference [16].

periodically tilted spins rotated back and forth sinusoidally. In real space, this appeared
as though the striped mF = 0 density pattern periodically faded to a uniform density
and back again (see Figure 4 (a)). Measurements of the contrast of this density pattern
were taken over time, revealing not only a cosinusoidal variation in the contrast (in sync
with the periodic density variations) as shown in Figure 4 (b), but a decay in contrast
to some mean value. The lead author speculated in their PhD thesis [63] that this decay
was a result of inhomogeneous broadening, though we suspect this may be a result of
interactions with the surrounding thermal cloud.
The truly remarkable feature of this experiment is that the contrast measurement
is the only observation required to obtain a dispersion relation, and hence an effective
mass, for the magnons in their interferometer. By measuring the contrast oscillation

16
frequency ωcontrast = 2(ω(k) − ω(0)) for a given imprinted wavevector ~k, a single point on
the dispersion relation could be probed, and by taking repeated measurements at different
magnon wavevectors, a full view of the dispersion relation18 was uncovered, as shown in
Figure 4 (c).
The authors also noted a variation in the magnon dispersion frequency with an in-
creased maximum tip angle θmax . As the mF = 0 population (which is proportional to
the number of magnons) was increased, the collective dispersion frequency of the spin
wave also appeared to increase as shown in the inset of Figure 4 (c). To account for this,
they ascribed a linear trendline to these data and extrapolated down to zero magnon den-
sity. All ωcontrast measurements were scaled according to the fit, such that their presented
dispersion frequencies are lower than their observed values for θmax > 0.
The effective magnon mass m∗ was determined by fitting the corrected contrast val-
ues to the expected magnon dispersion relation ω(k) = ~k 2 /2m∗ . Their obtained value of
m∗ = 1.033(2)stat (10)sys × mRb far exceeds the predicted masses from mean-field (1× mRb )
and beyond mean-field (1.003 × mRb ) theory [58]. One potential cause for this mass dis-
crepancy is the introduced frequency correction; the lead author noted [63] the dispersion
frequency shift to be due to nonlinear interactions in the condensate. In Section 4.3,
we shall see the effects of nonlinear interactions in our own simulations of the magnon
contrast interferometer experiment.

4.2 Analytical investigation into standing spin waves


We shall now theoretically investigate the magnon contrast interferometry experiment
from the perspective of the spin wave Fourier modes in one dimension. We begin by
noting that the spinor describing the experiment is given generally by Eq. 24, as we
derived in Section 3.2. The squared modulus of the spinor gives the fraction of the total
condensate population in each of the three Zeeman states, and multiplying these by the
total density gives the densities in each component19 :

|ψ−1 (x, t)|2 |ζ−1 (x, t)|2


      
cos4 2θ
ρ~ (x, t) =  |ψ0 (x, t)|2  = ρ(x)  |ζ0 (x, t)|2  = ρ(x)  12 sin2 θ  . (27)
     

|ψ+1 (x, t)|2 |ζ+1 (x, t)|2 sin4 2θ




In the limit of small, spatially uniform θ, the resulting mF = 0 density, ρ0 , is given by

1 1
ρ0 (x) = ρ(x) sin2 θ ≈ ρ(x)θ2 for |θ|  1. (28)
2 2
18
This method was incapable of measuring a gap in the dispersion relation at ω(0), since it measured
the recoil frequency ω(k) − ω(0).
19
Without loss of generality, the initial phase φmF of each component can be set to zero for this spin
excitation.

17
For the interferometry experiment, the spatially-varying spin tip angle depended on
the imprinted magnon wavevector k0 as

θmax
θ(x) = (1 + cos(k0 x)). (29)
2

Substituting this into Eq. 28, we find an mF = 0 density of


2
θmax
ρ0 (x) = ρ(x) (3 + 4 cos(k0 x) + cos(2k0 x)) . (30)
16

This shows us the Fourier components of the mF = 0 density for the magnon interferom-
etry experiment, at spatial frequencies 0, ±k0 and ±2k0 .
Armed with this knowledge, we set out to illustrate two of the authors’ claims:

(i) the magnon density is proportional to the density of atoms in the mF = 0 component,
and

(ii) their standing wave comprises magnons of three wavevectors, ~0 and ±~k.

To demonstrate the first claim, we identify that an elementary spin excitation, or


magnon, may be interpreted as a wave of transverse magnetisation F⊥ , or a transverse spin
wave [58]. The magnon density is hence given by the squared modulus of the transverse
magnetisation expectation value,

|hF⊥ i|2 = |hΨ|Fx |Ψi|2 + |hΨ|Fy |Ψi|2 = ρ(x) sin2 θ = 2ρ0 (x), (31)

which is twice the mF = 0 density (Eq. 28). Hence we may refer to mF = 0 density and
magnon density synonymously.
Now if we assume that a superposition of magnons were created at spatial frequencies
0 and ±k0 , then their momentum space distribution is

f (k) = δ(k − k0 ) + b δ(k) + δ(k + k0 ), (32)

where b quantifies the relative amplitude between the 0 and ±k0 modes. For such a spin
excitation, the corresponding magnon density is simply a weighted sum of plane waves
that evolve with time [58] according to
2
X
|hF⊥ i|2 = 2ρ(x) ei(kx−ω(k)t) f (k) . (33)


k

Evaluating this expression using Eq. 32, we obtain a transverse spin density of

b2
 
2
|hF⊥ i| = 2ρ(x) 1 + + cos(2k0 x) + 2b cos(k0 x) cos((ω(k0 ) − ω(0))t) . (34)
2

18
We can compare this to the mF = 0 density given in Eq. 30 to show that we have
synthesised a magnon population within the mF = 0 density with our spin rotation in
Eq. 29. Equating expressions (30) and (34) at the initial time, we can solve for b to get
b = 2, giving us a magnon evolution that follows

|hF⊥ i|2 = 2ρ(x) (3 + cos(2k0 x) + 4 cos(k0 x) cos((ω(k0 ) − ω(0))t)) . (35)

This important result shows us that the time evolution of the magnon density reveals
the magnon dispersion relation, as the Fourier power (given by the squared amplitudes
of the terms in Eq. 35) at wavevector k0 evolves as cos2 ((ω(k0 ) − ω(0)) t). Measuring the
spatial Fourier power of the mF = 0 density at wavevector k0 at a range of different times
allows the oscillation frequency ωcontrast to be determined; repeating this procedure for
many wavevectors probes the dispersion relation.
Figure 5 shows the time dependence of the transverse spin density given by Eq. 35
for a homogeneous total density ρ(x). One should note that the contrast oscillations in
the colour plot retain a fixed amplitude as time proceeds, unlike in the actual experiment
where the contrast envelope decayed to a fixed value. In reference to this comparison,
we must stress that the transverse spin density shown here exists in the non-interacting
limit, and is therefore not reflective of the magnon density within the condensate. To
demonstrate this, we may apply a Thomas–Fermi density profile ρ(x) to Eq. 35, in which
case we observe the same oscillatory behaviour as in Figure 5, but with a envelope identical
to the form of the density profile. Within an actual condensate, spatially-dependent
nonlinear interactions drive dynamics that are less uniform, as we shall see in the next
section.

4.3 Magnon interferometry simulations


Using similar parameters as in the magnon contrast interferometry experiment, we per-
form simulations20 of the spin dynamics in PyGPE to determine what effect the nonlinear
interactions of the condensate have on magnon propagation. Our one-dimensional ge-
ometry is sufficient to model the effectively two-dimensional experiment reported, as the
magnon propagation occurred only along a single axis. We use two distinct trapping po-
tentials in our simulations: a harmonic potential as used in the actual experiment, and a
homogeneous potential that provides a uniform condensate density21 .
Starting with a harmonically trapped ferromagnetic mF = −1 ground state (see
Figure 6 (a)), we perform a spatially modulated spin tip as in Eq. 29, which transfers
20
Our results for these simulations are presented qualitatively, as the mechanism for mathematically
describing the contrast was not clear from the article [16]. This mechanism was recently revealed in the
lead author’s thesis [63], however we did not have sufficient time to implement this analysis.
21
The walls of the homogeneous potential are not infinitely sharp, but taper off to zero with a width
of the order of the condensate healing length.

19
Figure 5: Colour plot depicting the evolution of the magnon density distribution. The false
colour scale has been normalised to the peak magnon density (red). The Fourier mode
at spatial frequency k0 can be seen in the regions where the density reaches a maximum,
while the 2k0 mode is visible at times when the peak density has subsided. These times
correspond to the vanishing amplitude of the k0 mode. For both a homogeneous and
Thomas–Fermi density profile, the standing wave continues oscillating indefinitely with
no decay process in the model.

a small population into the mF = 0 and +1 states22 (see Figure 6 (b)). We use a maxi-
mum tipping angle θmax = 11.5◦ , corresponding to the stated mF = −1 and 0 populations
for the experimental run shown in Figure 4. Following time propagation, the locally
normalised mF = 0 densities for both trapping potentials can be seen in Figure 7.
In the homogeneous trap (Figure 7 (a)), the mF = 0 atomic density evolves similarly
to the predicted magnon density as shown in Figure 5, with some slight inhomogeneities
likely resulting from the walls of the trap not being infinitely sharp. In the harmonically
trapped case (Figure 7 (b)), we also observe a periodic standing wave evolution, but after a
few cycles, we see a diminished density at the centre of the condensate. In the experiments
by Marti, et al., the contrast settled to a finite value, yet our mean-field simulations
indicate that the contrast continues to vary with time, with an enhanced contrast at the
edges of the condensate. Even more peculiar is the observed contrast revival seen at
400 ms, where the original standing wave profile returns. This modulated extinction and
revival in observed contrast repeats itself periodically as a diamond pattern, in addition
to the underlying periodicity of standing magnon propagation; such an observation was
not reported to have occurred in the contrast interferometry experiment.

22
For the tipping amplitude we employ, the population fraction transferred into the +1 state is so low
that, whether spin-exchange interactions are active or not, there is no discernible effect on the densities
of the other two components.

20
(a) (b)

Figure 6: Density profiles for a harmonically trapped condensate in our magnon contrast
interferometry simulations, prior to time propagation. (a) Initial condensate ground state,
with all atoms in the mF = −1 state. (b) Condensate densities after applying a spatially-
dependent spin tip. The small population fraction transferred into the mF = 0 component
is visible as the small density grating (green), with the concomitant depletion of the initial
state (blue) just visible.

The modulation occurs as a result of self-interactions between condensed particles.


While Figure 5 shows the propagation of the free-particle-like magnon modes, condensed
particles do not (strictly speaking) act like free particles; the self-interactions add an ad-
ditional potential. For a harmonically-trapped condensate, the high density of particles
in the centre of the trap create stronger interactions there, driving the modulation. We
do not observe such a modulation in the homogeneous trap since the particle density, and
hence self-interaction strength, is constant along the condensate. The observed modula-
tion demonstrates how large a role nonlinear interactions play into condensate dynamics.
We shall observe further effects due to nonlinear interactions in the following sections.

21
(a)

(b)

Figure 7: Locally normalised mF = 0 densities (i.e. ρ0 (x)/ρ(x)) for (a) homogeneous,


and (b) harmonic trapping potentials. The false colour scales for the densities have been
normalised to the peak mF = 0 density (red). On the spatial axes, we have excluded
regions for which the atom number goes to zero.

22
5 Dispersion of monochromatic magnons
While the magnon interferometry experiment [16] probed the magnon dispersion relation
using interference of a standing spin wave, their result of an anomalous effective magnon
mass fell short on explanation. The work described in this section aims to investigate
whether the discrepancy can be understood in terms of mean-field physics. In order to
investigate this claim, we set out to probe the dispersion relation in PyGPE using our own
technique. Since PyGPE simulates only mean-field phenomena, we are able to investigate
basic magnon propagation without the distractions of magnetic dipolar interactions and
beyond mean-field effects.
An an alternate method for computing effective magnon mass to the interferometry
experiment, we instead chose to simulate and analyse the dynamics of a single, monochro-
matic spin wave. This technique23 allows us to meaningfully measure magnon propagation
for any range of polar spin tip angles θ. More fundamentally, it allows us to understand
the behaviour of isolated spin excitations, rather than superpositions thereof. Single spin
waves could also allow for investigating the interactions between spin waves and domain
walls, and may provide a pathway to probing the nematic magnon branch.

5.1 Simulating magnon propagation


Our protocol for imprinting spin waves begins by considering a ferromagnetic ground state
condensate (i.e. all atoms in the mF = −1 state) in a harmonic trap. Once again, we
work in a frame rotating at the Larmor frequency (which we shall refer to as the ‘static’
frame), and switch off any spin-exchange interactions (by setting a2 = a0 ) for simplicity.
To our ground state, we perform a spin rotation of a spatially uniform polar angle θ,
followed by a spatially-dependent azimuthal angle rotation of φ(x) = kx. An example of
the resulting densities and phases for each component is shown in Figure 8. This gives
each component of the condensate a linear phase gradient, which constitutes a spin wave
with a single wavevector k, rather than a magnon standing wave as created using the
magnon contrast interferometry technique.
After propagating the condensate (with its imprinted spin wave) in time according to
the GPE, the spatially-dependent azimuthal angle φ can be measured using

hFy i
 
φ(x, t) = arctan . (36)
hFx i

Plotted on space and time axes, we obtain an azimuthal angle colour plot such as that
shown in Figure 9 (d). At early times in the simulation, the azimuthal angle φ follows
23
Besides the technique described in this section, we also describe a second technique for created
monochromatic magnons in Appendix B, by utilising magnetic field gradients.

23
(a) (b)

Figure 8: Initial density (a) and phase (b) profiles for each component of a condensate
imprinted with a monochromatic spin wave, as viewed from the static frame. Here we
have chosen tip angle θ = 60◦ and wavevector k = 2π/(25 µm). The linear phase gradients
resulting from the imprint cause the three components to separate over time, as can be
observed in Figures 9 (a) through (c).

plane wave dispersion according to

φ(x, t) = kx − ωt, (37)

where ω is the observed magnon dispersion frequency. To explicitly measure ω for a given
wavevector k, we consider a single point in space, chosen as the centre of the conden-
sate, x = 0 µm, as it is the most uniform. This corresponds to taking a vertical trace of
φ(x = 0 µm, t), and as would be expected from Eq. 37, the gradient of this curve gives
the dispersion frequency (see Figure 9 (e)). While in theory the dispersion frequency
should stay constant with time (indicative of plane wave dispersion), two practical effects
prevent this idealisation. Both of these effects result from the spin wave imprint being ac-
companied by the introduction of phase gradients, which cause an effective Stern-Gerlach
separation between the components of the condensate.
Firstly, the phase gradients drive the components of the condensate up the walls of
the harmonic trap, setting up a sloshing motion. This results in an oscillating magnon
dispersion frequency that varies with the sloshing motion. Secondly, when the phase
gradients have acted long enough to displace the components in space, the dispersion
frequency reading becomes contaminated by complicated mean-field dynamics. Our ana-
lytical treatment of the condensate depends on it being well described by the single-mode
approximation (as introduced in Section 2.3.1), and this approximation is no longer valid
when the components of the condensate no longer overlap in space.

24
(a) (b) (c)

(d) (e)

Figure 9: Observables for propagation of a spin wave with wavevector k = 2π/(25 µm)
imprinted onto a condensate tipped to a global tip angle of θ = 60◦ . (a,b,c) Densities
of the three condensate components as a function of space and time. The components
gradually move apart due to the phase gradients imposed by imprinting the spin wave.
The false colour scales for the density of each component have been normalised to the
same peak density. (d) Azimuthal angle colour plot. Each horizontal trace gives the
azimuthal angle of spins at each point along the condensate at a given instant of time. At
the edges of our spatial grid (|x| > 50 µm), numerical error is induced due to a vanishing
atom population in those regions. (e) Magnon dispersion frequency, measured as a trace of
the colour plot about the centre of the condensate. The monotonic decrease in dispersion
frequency before 30 ms is a result of sloshing of the components in the trap, whilst the
oscillations after 30 ms occur due to component separation.

25
To avoid the effects of both the sloshing motion and component separation, the initial
dispersion frequency is used, as the spin wave disperses as a plane wave (i.e. according
to Eq. 37 with constant ω) only at the start of the simulation. Since we only require the
initial dispersion frequency, simulations can be ran over a very short time, and we can
gather data very rapidly.

5.2 Frames of reference


As in countless areas of physics, it is instructive to consider a system in different frames
of reference. Up until this point, we have been considering the condensate in a frame
rotating at the Larmor frequency, for which the imprinted spin wave causes the different
spin components to separate. The magnitude of the imposed phase gradients can be
seen in the phase prefactors for each component in Eq. 24 — the mF = −1 component
has a prefactor of e−iφ , and moves to the left with momentum ~k, i.e. with momentum
proportional to the gradient of the phase φ = kx. Similarly the mF = +1 state moves to
the right with momentum ~k, while the mF = 0 component remains stationary.
In these simulations, the measured magnon dispersion frequency depends on the frame
of reference it is being measured in. It is therefore instructive to also measure the dis-
persion frequency from the frame of the condensate centre of mass, a frame for which the
total wavevector of the condensate,
X Nm
F
kBEC = ∇φmF , (38)
mF
Ntot

vanishes. Here we have defined NmF as the number of atoms in each component, and
P
Ntot = mF NmF . Note that the centre-of-mass frame still rotates at the Larmor fre-
quency, but also incorporates the translation of the condensate due to the imposed phase
gradients.
To determine the gauge transformation to arrive in this frame, we multiply our spinor
from Eq. 24 by an arbitrary global phase factor ei , giving a new spinor24 of
   i  T
i−iφ 2 θ e i+iφ 2 θ
ζ= e cos , √ sin θ, e sin . (39)
2 2 2

This gives the new phase for each component to be ((x) − φ(x), (x), (x) + φ(x)). Now
since we want to choose a gauge for which kBEC = 0, we search for an  for which this
condition holds. The corresponding gauge takes the form (x) = φ(x) cos θ. Hence if we
multiply our spinor by this phase prefactor before time propagation, we obtain a magnon
dispersion frequency as viewed from the frame of the centre of mass. It should also be
noted that the magnon wavevector k is preserved following this Galilean transformation.
24
It should be noted that this spinor leaves the spin expectation values of Eq. 25 unchanged.

26
5.3 Magnon dispersion relations
Whichever frame we are in, we can measure the dispersion frequency ω for a given magnon
wavevector k that we imprint according to φ(x) = kx. By changing the imprinted wavevec-
tor over many simulations, the dispersion relation can be obtained for a given polar tip
angle θ, as shown in Figure 10 for θ = 0.01◦ . In the static frame, we find that the obtained
dispersion frequency varies quadratically with wavevector k, as given by

~ 2
ω= k . (40)
2m∗

Note that the dispersion frequency shift resulting from the magnon energy gap p − q [50]
is removed by using reference frames that rotate at the Larmor frequency, and numeri-
cally setting the quadratic Zeeman shift q to zero. Under these conditions, the observed
dispersion frequency describes gapless free-particle-like propagation of magnons. We also
note that, in the centre-of-mass frame, we measure an inverted dispersion relation relative
to that found in the static frame.

25

20
ω(k)/2π (Hz)

15

10

0
0
ω/ωfree−1 (10−6 )

2
4
6
0 20 40 60 80 100
−1
k/2π (mm )

Figure 10: (top) Dispersion relation calculated for θ = 0.01◦ in the static frame. Our
data (black points) are fitted quadratically (red dashed line) to determine the magnon
mass. In the limit of small tip angles, the magnon dispersion frequency approximates the
free-particle dispersion frequency ωfree (black line). The variation of the data from the fit,
as seen in the residuals (bottom), occurs for spin waves with wavelengths of order of the
size of the condensate or longer.

27
The effective magnon mass m∗ is determined from the measured dispersion relation by
fitting Eq. 40 to our simulation data. For small tip angles, e.g. θ = 0.01◦ , we observe that
the effective magnon mass approaches the bare atomic mass mRb ; this is consistent with
the predictions of mean-field theory [58]. However, a ‘deviation’ from the bare atomic
mass occurs, with this deviation varying as a function of the global spin tip angle θ. This
tip angle dependence is visually noticeable in the observed dispersion relations – for a
fixed spin tip angle, the dispersion relation is characterised by a unique curvature, as
Figure 11 demonstrates. The decreasing curvature with spin tip angle may be interpreted
as an increase in the effective magnon mass. In the case of θ = 90◦ , the dispersion
frequency becomes zero for all wavevectors, effectively freezing the imprinted spin wave
and implying an infinite effective magnon mass. As the tip angle increases past 90◦ , the
dispersion frequency becomes negative25 , and the spins precess in the opposite direction.

25

20
ω(k)/2π (Hz)

15

10

00 20 40 60 80 100
−1
k/2π (mm )

Figure 11: Dispersion relations measured in the static frame for θ = 0.01◦ , 30◦ , 60◦ ,
75◦ and 90◦ , descending from the free-particle dispersion (black line). The vanishing
dispersion frequency for all wavevectors at θ = 90◦ corresponds to the observation of a
‘frozen’ spin wave.

25
When viewed from the static frame. Beyond 90◦ , the dispersion relation is positive when viewed from
the centre-of-mass frame.

28
The relationship between magnon mass and tip angle is shown in Figure 12, with the
results for both the static and centre-of-mass frames shown. The measured negative mass
simply indicates magnon dispersion to the left of our system, in contrast with positive
mass indicating rightwards dispersion. Here the results for both frames are equal and
opposite: both predict the same magnon mass, but in a given frame the magnon moves
in the opposite direction to when it is viewed in the other frame. In both frames, we find
that the observed magnon dispersion frequency ω for a given magnon wavevector k scales
with cos θ 26 . This coincides with the observed functional form of the mass variation,
given by m∗ /mRb = 1/ cos θ.

8
Static frame data
6 CoM frame data
4
2
m ∗/mRb

0
2
4
6
80 20 40 60 80 100 120 140 160 180
θ (degrees)

Figure 12: Magnon mass deviation from the atomic mass with tip angle. Each data
point in this figure is generated from a single dispersion relation, such as those shown
in Figure 11. The symmetry between results from the static and centre-of-mass (CoM)
frames indicates consistency between frames. We have also fitted our data to ±1/ cos θ
relationships for the static and centre-of-mass frames (blue and green curves) respectively,
with excellent agreement. The observed negative mass simply indicates magnon disper-
sion to the left of our system, when viewed through a given frame. As the polar angle
approaches θ = 180◦ , the mF = +1 state acts as the dominant ferromagnetic ground
state, giving a symmetry about θ = 90◦ .

26
That is, ω = ωfree cos θ, where ωfree ≡ ~k 2 /(2mRb ).

29
5.4 Interpretation of observed magnon mass variation
In the limit of small excitations, we have found that the effective magnon mass that we
measure for monochromatic magnons matches the predictions of mean-field linear response
theory. This theory is not valid for describing large perturbations, and thus it is rather
peculiar that the effective magnon mass we measure as a function of the spin tip angle
(m∗ /mRb = 1/ cos θ) follows such a simple functional form for all tip angles. To address
this mass deviation, we begin by noting that it is directly related to the magnetisation

hFz i = |ψ−1 |2 − |ψ+1 |2 = cos θ. (41)

Since the magnetisation relates to the densities of the mF = −1 and +1 states, we


conjecture that the mass deviation is related to the emerging finite populations in all
three spin components. Since collective excitations are emergent phenomena, it should
come as no surprise that the behaviour of magnons, which are collective excitations, is
intimately tied to the behaviour of the bulk condensate.
A single magnon is described by an infinitesimal spin rotation that moves a single atom
from the mF = −1 ground state into the mF = 0 state. Relative to one another, these
components move apart with a momentum par particle of ~k, which can be interpreted
as a magnon with wavevector k propagating as a free particle with effective mass mRb .
As the spin tip angle θ is increased, population is also transferred into the mF = +1
component, with each particle moving with momentum ~k in the opposite direction to
the ground state. This is interpreted as a second spin excitation mode, propagating against
the mode associated with the mF = −1 component. The strengths of these excitation
modes are proportional to the mF = −1 and +1 densities, which determine the net spin
wave dispersion frequency. For example, when the densities are equal (i.e. θ = 90◦ ), the
strengths of the opposing spin excitation modes are also equal, and we observe no net
spin precession.

5.5 Comparing masses for standing and running magnons


In the magnon contrast interferometry experiment [16], the authors of the paper found
that as the mF = 0 population increased, the measured dispersion frequency also in-
creased as a result of nonlinear interactions [63]. As described previously in Section 4.1,
they calibrated for finite magnon populations by extrapolating the observed dispersion
frequency down to zero magnon density.
We also encounter a frequency shift in our monochromatic magnon simulations as the
global tip angle θ is increased. As described above, we conjecture that our frequency
shift results from having a finite mF = +1 population, which is unavoidable if we are to

30
continue considering a ferromagnetic condensate27 . The presence of mF = +1 atoms may
have had a similar effect in the interferometer, as the lead author noted that “at larger
angles, the mF = −1 and mF = +1 images [both] show clear fringes” [63].
We directly compare both frequency shifts in Figure 13, and note two key differences
between the results of Marti, et al. and our simulations. The first of these refers to the
measured dispersion frequencies for infinitesimal spin rotations. Our simulations conform
to the mean-field result in the |θ|  1 limit, i.e. m∗ = mRb . On the other hand, the linear
fit used by Marti, et al. would suggest a lower dispersion frequency, and hence heavier
effective magnon mass, than is theoretically predicted by the linear excitation theory in
the mean-field limit. The calibration data presented in the article could be considered to
broadly fall within the mean-field prediction for small excitations, as the data featured
a wide spread and no error bars. These features make us question the validity of the
ascribed linear fit, and hence, their assertion of an anomalously heavy magnon mass.

24.0
23.5
23.0
22.5
ω/2π (Hz)

22.0
21.5
21.0
20.5
20.00 1 2 3 4 5
NmF =0 ( 104 atoms)
Figure 13: Dispersion frequency shift for finite mF = 0 population at wavevector k =
2π/(10.1775 µm). Our data (red points) shows a linear decrease from the free-particle
result, whereas the magnon interferometry results (blue points and fit, as taken from inset
of Figure 4 (c)) [16] would suggest a linear increase. The free-particle dispersion frequency
(black line) is shown for comparison. We note that the experimentally determined data
points feature a wide spread and no error bars, which makes us question of validity of the
ascribed linear fit used for their calibration.

27
To remove the mF = +1 population would change the length of the global magnetisation vector
for our condensate, meaning that we would no longer be considering only transverse spin waves in a
ferromagnetically ordered gas, but a combination of transverse and nematic spin excitations.

31
The second key difference between the results refers to the different gradients of the
linear fits describing the effective mass away from small tip angle limit. Our observed dis-
persion frequencies decrease with increasing mF = 0 population, contrary to the frequency
increase as found for standing spin waves. Assuming the linear fit to the interferometer
data to be valid, the difference may be a result of each method exciting a different su-
perposition of magnon excitations: the interferometer excites a magnon standing wave,
while our simulations describe monochromatic magnons travelling in one direction. We
again note that the quasiparticle representation of spin excitations becomes less mean-
ingful as the excitation amplitude increases. Hence the different functional forms may be
of no significance, with only the effective mass measured for small tip angles carrying any
importance. This idea is motivated by the results reported in the next section, where we
probe the effective magnon mass using a different mechanism. This new method excites
a superposition of magnon modes, and implies an effective mass variation (equivalent to
a dispersion frequency shift) that assumes yet another functional form.

32
6 Gaussian wavepackets of magnons
In this final section, we describe the dispersion of a magnon wavepacket as a means for
measuring the effective magnon mass, as proposed in Reference [58]. A similar experiment
has been performed [16], in which a Gaussian wavepacket of mF = 0 atoms was created
from a fully magnetised mF = −1 condensate. In that experiment, a relative force
was induced between the mF = −1 and 0 components via a magnetic field gradient,
and, by measuring the acceleration of the wavepacket, the magnon magnetic moment
was determined. While we have also simulated a Gaussian wavepacket of magnons, our
simulations measure their effective mass by tracking their rate of expansion. This method
warrants comparison to our previous studies of monochromatic magnons, and allows us
to draw conclusions about the nature of measuring the effective magnon mass outside of
the small excitation limit.

6.1 Measuring the effective mass from wavepacket dispersion


The theory in this section has been adapted from Reference [58].
A Gaussian wavepacket of magnons is represented in momentum space by a Gaussian
superposition state with spectral weighting

f (k) = exp −d20 (k − k0 )2 ,



(42)

where d0 is the Gaussian RMS width of the wavepacket in coordinate space and k0 is
the wavevector for the centre of mass of the wavepacket. Assuming that this wavepacket
disperses in accordance with the quadratic magnon dispersion ω = ~k 2 /(2m∗ ), then we
can compute the time evolution of the wavepacket with Eq. 33. We predict the transverse
spin density to evolve as
 
2
2π − (x − ~k0 t/m∗ ) 
|hF⊥ i|2 ∝ q exp    . (43)
~2 t2 ~2 t2
2d20 1 + 4d40 m2∗
2
2d0 1 + 4d4 m2
0 ∗

This indicates that the magnon wavepacket moves with a group velocity of vg = ~k0 /m∗
and expands with a time-dependent width given by
s
~2 t2
d(t) = d0 1+ . (44)
4d40 m2∗

Hence by measuring either the centre-of-mass position or width of the wavepacket as


a function of time, the effective magnon mass can be determined. We choose to mea-
sure the width of an expanding wavepacket, since this measurement can be performed
for a wavepacket at the centre of the trap with zero group velocity. This allows us to

33
avoid inducing any asymmetric distortion of the wavepacket that occurs for a travelling
wavepacket in a harmonic trap.
It must be noted that Eq. 33 was derived in the limit of small tip angles, corresponding
to a low spin excitation amplitude. In our simulations, we nonetheless explore the regime
of large tip angles to see

(i) if the Gaussian expansion predicted in Eq. 44 persists, and if so

(ii) how the effective mass depends on the tip angle.

6.2 Wavepacket expansion


To imprint a magnon wavepacket, we once again initialise our condensate in the fully
magnetised mF = −1 state, and perform a spatially-dependent polar angle rotation of
s !
x2

θ(x) = arcsin 2A exp − 2 , (45)
2d0

where A is the amplitude of the spin tip, given as a fraction of the initial mF = −1
density. Since the mF = 0 density transforms according to ρ0 (x) = 21 ρ(x) sin2 θ, the above
spin rotation gives a corresponding density of

x2
 
ρ0 (x) = A ρ(x) exp − 2 , (46)
2d0

with such a density profile shown in Figure 14.

1000
mF = -1
800 mF = 0
mF = 1
Density (atoms / µm)

600

400

200

080 60 40 20 0 20 40 60 80
x (µm)

Figure 14: Gaussian wavepacket of magnons imprinted into a fully magnetised condensate.
This initial wavepacket has RMS width d0 = 2 µm and amplitude A = 0.15, which
corresponds to a maximum tip angle of θmax = 33.2◦ at the centre of the wavepacket.
Imprinting the wavepacket causes an associated depletion from the mF = −1 component.

34
Figure 15: (top) mF = 0 density, shown after 0, 0.5, 1 and 2 Rayleigh times. The initial
wavepacket is identical to that shown in Figure 14. Our simulated data (coloured dots) are
appropriately fitted to Gaussian profiles (coloured fits), with low absolute error (bottom).

When propagated in time, we observe in Figure 15 that the mF = 0 density disperses


according to Eq. 44. By allowing the wavepacket to expand for one Rayleigh time28

tR ≡ 2d20 m∗ /~ (i.e. the time taken for the wavepacket to expand to 2 times its original
width), we fit Eq. 44 to our data via nonlinear regression to extract an effective mass (see
Figure 16).
We may determine the effective mass in this manner for wavepackets of various initial
widths and amplitudes, though only a certain range of these parameters give physically
interpretable results:

– For an initial RMS width d0 less than 2 µm, the spatial grid used in our simulations
(0.28 µm per element) is insufficient to resolve the imprinted Gaussian, hence a
meaningful dispersion does not occur for such wavepackets. Similarly we find that
wavepackets with initial widths greater than 4 µm do retain a Gaussian profile, but
do not expand according to Eq. 44 prior to reaching one Rayleigh time. As such,
we cannot impute an effective mass using these results.
28
The Rayleigh time is analogous to the Rayleigh range used in Gaussian optics.

35
Simulation data
4.0 Nonlinear regression fit
Fit for m ∗ = mRb

Gaussian width (µm)


3.5

3.0

2.5

2.0
0 5 10 15 20
Time (ms)

Figure 16: Gaussian widths as a function of time, up to 2 Rayleigh times. Our data (black
dots), when fitted to a Gaussian dispersion curve (red dashed curve), gives an effective
magnon mass. The dispersion curve corresponding to a magnon mass equal to the bare
atomic mass (black curve) is shown for comparison. Note that the effective magnon mass
is calculated for a fit extending to one Rayleigh time, whilst this shows the equivalent for
2 Rayleigh times.

– While there is no lower limit to the Gaussian amplitude, any wavepacket with A ≥
0.5 becomes analytically meaningless. This can
√be understood
 if we consider the tip
angle of the wavepacket peak, θmax = arcsin 2A ; any amplitude A greater that
29
0.5 will have no solution .

Within the valid range of widths, the effective mass at a given tip angle varies by
no more than 0.15%. Thus the effective mass depends only weakly on the width of the
wavepacket, in a way that is not accounted for in Eq. 44. We also find the effective mass
to depend on tip angle, as in Section 5.3. However, unlike the case of monochromatic
magnons, we see that the effective mass decreases with increasing maximum tip angle
(see Figure 17). In comparing our results, we must emphasise that the two individual
simulations of Sections 5 and 6 feature inherently different spin excitations. They also
measure the effective magnon mass in entirely different ways: our monochromatic magnon
simulations measure the dispersion relation for individual magnon wavevectors, while our
Gaussian wavepacket simulations track the expansion of a packet of magnons of various
wavevectors. The different functional forms for magnon mass deviation are likely a conse-
quence of the different excitation regimes, i.e. for the magnon wavepacket, the interaction
of magnons from different regions of the magnon excitation spectrum could lead to an
overall mass decrease.
29
We avoid using an amplitude of exactly 0.5, as this imprints mF = −1 and +1 densities with sharp
edges in the centre of the condensate.

36
1.05

1.00

0.95
m ∗/mRb

0.90

0.85

0.80 0 10 20 30 40 50 60 70
θmax (degrees)

Figure 17: Magnon masses measured from the dispersion of Gaussian mF = 0 density
profile for peak tip angles θmax with amplitudes up to 0.45. The effective mass is very
weakly dependent on the initial wavepacket width; m∗ varies by less than 0.15% for d0
between 2 µm and 4 µm.

6.3 Dispersion of Gaussian spin tip profile


As Eq. 44 was strictly derived for the limit of small spin tip angles (|θ|  1), we sought
to determine the evolution of Gaussian wavepackets imprinted into θ, as opposed to the
mF = 0 density. For small tip angles, wavepackets imprinted into both θ and the mF = 0
density should be effectively Gaussian, and hence follow Gaussian dispersion. This is an
experimentally relevant scenario, as Gaussian profiles in θ may be easier to imprint onto
a condensate. In these simulations, we determine the effective magnon mass using the
same procedure as in Section 6.2, but we instead imprint a polar angle rotation of

x2
 
θ(x) = θmax exp − 2 . (47)
2d0

The corresponding mF = 0 density transform, again taking the form ρ0 (x) = 21 ρ(x) sin2 θ,
gives a ‘pseudo-Gaussian’ density profile30 , with similar transforms to the initial mF = −1
and +1 states also giving pseudo-Gaussians.
Following time propagation of the initial condensate profile, we can determine the
polar tip angle θ at any given point along the condensate using

hFz i
 
θ(x, t) = arccos , (48)
hF i
30
The mF = 0 density is effectively Gaussian in the limit of small θ.

37
1.00
d0 =2µm
0.99 d0 =3µm
0.98 d0 =4µm
0.97
m ∗/mRb

0.96
0.95
0.94
0.93
0.92 0 20 40 60 80
θmax (degrees)

Figure 18: Magnon masses measured from the dispersion of polar angle Gaussians with
initial width d0 and amplitude θmax . The measured mass decreases from the bare atomic
mass as both d0 and θmax increase.

p
where we calculate the expected total spin density hF i = hFx i2 + hFy i2 + hFz i2 . By
calculating the expectation values of the spin operators at each timestep, we observe how
θ evolves with time. Curiously, we find that the polar angle along the condensate retains
its Gaussian profile as it spreads, even for large wavepacket amplitudes. This is rather
unexpected, in that θ is ‘constructed’ from the densities of the three spin components,
none of which are Gaussian themselves, and hence would not be expected to necessarily
follow a corresponding dispersion. In any case, we can still use the dispersion of the polar
angle Gaussian to measure the magnon mass as before, with one caveat. Since Eq. 44
describes the dispersion of a wavepacket of mF = 0 atoms, as opposed to a Gaussian polar
angle profile, the mass is now measured by tracking a wavepacket width of
s
~2 t2
d(t) = d0 1+ , (49)
d40 m2∗

up to a new Rayleigh time of tR ≡ d20 m∗ /~. We note that the mass can only meaningfully
measured when the polar angle profile stays Gaussian, a situation that fails when the
polar tip angle of the Gaussian peak exceeds 90◦ . Beyond this angle, the mF = 0 density
takes the initial form of two peaks, a profile which would not be expected to disperse in
an even approximately Gaussian fashion.

38
For this scenario, the measured magnon mass is proportional to both the maximum
tip angle θmax and the initial wavepacket width d0 . This is surprising, as the Gaussian
wavepackets in mF = 0 density (described in the previous section) had a negligible de-
pendence on the wavepacket width. Even more concerning is that, even in the small θ
limit, the Gaussians in the spin tip profile do not all converge to give the same effective
magnon mass. These findings are inconsistent with the results of Section 6.2, and leads us
to believe that these simulations possess an unresolved error, which given ample time we
would have addressed. Besides this issue, the effective masses measured as a function of
θmax follow a similar trend to Figure 17 for the mF = 0 wavepacket. Another important
consideration is that the effective mass deviations found here are less extreme than those
in Section 6.2; Figure 17 shows a maximum deviation from the bare atomic mass of 15%,
whereas in Figure 18 we see an maximum deviation of 5% at the same maximum tip angle
of θmax = 72◦ .

39
7 Conclusion
This project has studied how magnons propagate in the mean-field regime, and specifi-
cally how the effective magnon mass varies with the strength and type of spin excitation.
In Section 4.3, we qualitatively found that the form of the trapping potential affects
dispersion of standing spin waves, with the spatially varying nonlinear interactions in
a harmonically trapped condensate leading to a modulation in standing spin wave dy-
namics. Following this, we probed the dispersion relation for transverse spin excitations
by simulating monochromatic magnons in a ferromagnetic spinor condensate. The ef-
fective magnon mass was found to vary cosinusoidally with respect to the bare atomic
mass, which we conjectured to be a result of counter-propagation of magnons associated
with the mF = −1 and +1 ferromagnetic ground states. Finally, we measured the magnon
mass by tracking the expansion of a Gaussian wavepacket of magnons, and additionally by
measuring the dispersion of the magnetisation profile along the condensate. Significantly,
the effective mass decreased with increasing tip angle, in contrast to the simulations of
monochromatic magnons. Furthermore, the dependence of the effective magnon mass on
spin tip angle was much weaker when probed using dispersing wavepackets of magnons.
Importantly, we note that in the limit of small spin tip angles (|θ|  1), all simulated
experiments, with the exception of one31 , predict the effective magnon mass to be equal to
the bare atomic mass of the condensed species, in accordance with linear excitation theory
in the mean-field regime. Additionally, we have found that, in the mean-field picture, the
effective mass varies with increasing magnon population, with the variation appearing to
depend on the spectral distribution of induced spin excitations. As such, any experiment
probing magnon dispersion beyond the limit of small excitations must calibrate for the
observed mass deviation to determine the true effective mass of a single magnon. Our
work cannot explain the anomalously heavy magnon mass measured by Marti, et al. using
mean-field theory alone, and this motivates further investigations outside of the mean-
field regime, e.g. by introducing long-range magnetic dipolar interactions or quantum
depletion effects into our simulations. We also treat our work as sufficient background to
commence investigations into the nematic spin excitation branch.

31
The simulations described in Section 6.3 predicted the effective magnon mass to approximate the
bare atomic mass rather than to equal it, but we again note the unresolved error in those results.

40
8 Appendices
A Simulations of density excitations in scalar condensates
At the beginning of this project, our aim was to numerically solve the BdG equations
for a spinor condensate and apply the resulting spin excitation wavefunctions to the
condensate ground state. This would form a general theoretical framework for applying
arbitrary combinations of excitations to a condensate. Our initial model was built for
a scalar condensate in Matlab, which solved the BdG equations for a single-component
condensate, hence determining the quasiparticle wavefunctions for density excitations.
The model begins by taking an initial one-dimensional condensate wavefunction with a
Gaussian density profile and finding its ground state. This is achieved by propagating the
condensate wavefunction in imaginary time, using a unitary time evolution operator. The
ground state wavefunction ψ(x) is then used in formulating the BdG matrix (Eq. 9). From
this, the quasiparticle wavefunctions uk (x) and vk (x), and their respective eigenvalues ωk ,
can be found numerically.
The time dependence of the condensate can then be observed by projecting a linear
combination of excited states onto the ground state [24] according to
X
pk uk (x)e−iωk t + vk∗ (x)eiωk t ,

ψ(x, t) = ψ(x) + (50)
k

and propagating this state forward in time. While any linear combination is allowed, the
fundamental excitations are found when a nonzero excitation amplitude pk is assigned to
only a single excitation mode k; these give the collective modes of the scalar condensate.
For example, populating only the first excited state32 excites the slosh mode of the con-
densate, which is observed as a condensate moving back and forth periodically. Similarly,
populating only the second mode allows the breathing mode to be observed. Using this
basis set of quasiparticle excitation wavefunctions, any arbitrary excitation to the con-
densate can be constructed. However, their amplitudes pk must be small (O(10−3 )), as
the BdG theory of linear excitations breaks down when the number of excited particles
in the condensate is of the order of the total condensate number.

32
This state has an energy of ~ω, where ω is the one-dimensional trapping frequency.

41
B Monochromatic magnons imprinted using magnetic field gra-
dients
Upon embarking on our initial studies of monochromatic magnons, we set about the task
using a different spin wave imprinting technique. This method more closely resembles
how we would perform such a feat in an experimental setting33 , as opposed to the ‘hand
of God’ style imprinting we describe in Section 5. In this protocol, we begin with a
ground state spinor condensate simulated in PyGPE, and perform the follow imprinting
procedure:

(1) Perform a spatially uniform spin tip θ, as was initially performed for simulations
described in Section 5.

(2) Begin time propagation under a magnetic field gradient34 . The field gradient causes
the spins to Larmor precess at different rates along the length of the condensate. In the
presence of this field gradient, the azimuthal angle φ takes the form φ(x, t) = kx − ωt,
but now the wavevector k is time-dependent, and determined by the strength ∂B ∂x
and
duration tgrad of the field gradient:

∂B
k = −γ tgrad . (51)
∂x

Here γ is the gyromagnetic ratio for 87 Rb, which in the mean-field regime is equal to
the gyromagnetic ratio for magnons [16].

(3) Continue time propagation and switch off the field gradient. This locks in a magnon
of a single, time-independent wavevector that propagates freely.

(4) Apply a second field gradient of opposite sign to the initial gradient, in order to
‘unwind’ the spin wave. As the azimuthal angles of the spins all realign, a strong hFx i
signal can be measured.

For experimentally-created condensates, we are unable to directly measure the magnon


dispersion frequency, but we can instead infer magnon properties by measuring the spin
hFx i along the entire condensate over time35 . To further emulate lab conditions, we now
include a background magnetic field of 2 kG, to impose Larmor precession upon the spins.
In order to measure an hFx i signal that varies with magnetic field, we must also switch
on the quadratic Zeeman shift.
33
In fact, this method was originally conceived as a feasibility study for experimentally measuring the
effective magnon mass for monochromatic magnons.
34
The gradient is antisymmetric about the origin of our geometry, i.e. the centre of the condensate
experiences only the background magnetic field.
35
Such a measurement technique has recently been conceived at Monash’s spinor BEC lab.

42
Figure 19 shows our main observables for the simulation: (a) the azimuthal angle
colour map and (b) the time-dependent global hFx i measurement36 . For the particular
simulation shown, we apply a field gradient of 6 G/m for the first 10 ms, and allow for
free magnon propagation (no field gradient) for 10 ms. We then apply a reverse gradi-
ent (−6 G/m) for 20 ms, which unwinds the spin wave, then rewinds it in the opposite
direction, as seen for the final 10 ms of the simulation.

(a)

(b)

Figure 19: Observables for monochromatic magnons generated using magnetic field gra-
dients. Here we have used a background magnetic field of 2 kG and field gradients of
±6 G/m. (a, top) Azimuthal angle colour plot. Dashed lines indicate borders between
stages of the simulation, which are described in the text above. (b, bottom) Global hFx i
measurement. All spins are initially aligned parallel, but gradually misalign due to the
field gradient, until no signal is measured. When a reverse gradient is applied, the spins
realign, giving the ‘gradient echo’ signal centred at 30 ms.

36
We could have instead measured hFy i along the condensate, however both measurements give the
same information.

43
9 References
[1] S. N. Bose. Plancks Gesetz und Lichtquantenhypothese. Zeitschrift für Physik, 26,
178 (1924). doi: 10.1007/BF01327326.

[2] Albert Einstein. Quantum Theory of the Monatomic Ideal Gas. Sitzungsberichte
der Preussischen Akademie der Wissenschaften, Physikalisch-mathematische Klasse,
page 261 (1924).

[3] M. H. Anderson, J. R. Ensher, M. R. Matthews, C. E. Wieman, and E. A. Cornell.


Observation of Bose–Einstein Condensation in a Dilute Atomic Vapor. Science, 269,
198 (1995). doi: 10.1126/science.269.5221.198.

[4] K. B. Davis, M. O. Mewes, M. R. Andrews, N. J. van Druten, D. S. Durfee, D. M.


Kurn, and W. Ketterle. Bose–Einstein Condensation in a Gas of Sodium Atoms.
Physical Review Letters, 75, 3969 (1995). doi: 10.1103/PhysRevLett.75.3969.

[5] C. C. Bradley, C. A. Sackett, J. J. Tollett, and R. G. Hulet. Evidence of Bose–Einstein


Condensation in an Atomic Gas with Attractive Interactions. Physical Review Let-
ters, 75, 1687 (1995). doi: 10.1103/PhysRevLett.75.1687.

[6] Christopher Pethick and Henrik Smith. Bose–Einstein Condensation in Dilute Gases.
Cambridge University Press (2002).

[7] P. G. Kevrekidis, D. J. Frantzeskakis, and R. Carretero-González. Basic Mean-


Field Theory for Bose–Einstein Condensates. In Professor Panayotis G. Kevrekidis,
Professor Dimitri J. Frantzeskakis, and Professor Ricardo Carretero-González, edi-
tors, Emergent Nonlinear Phenomena in Bose–Einstein Condensates, number 45 in
Atomic, Optical, and Plasma Physics, pages 3–21. Springer Berlin Heidelberg (2008).

[8] Yuki Kawaguchi and Masahito Ueda. Spinor Bose–Einstein condensates. Physics
Reports, 520, 253 (2012). doi: 10.1016/j.physrep.2012.07.005.

[9] J. Stenger, S. Inouye, D. M. Stamper-Kurn, H.-J. Miesner, A. P. Chikkatur, and


W. Ketterle. Spin domains in ground-state Bose–Einstein condensates. Nature, 396,
345 (1998). doi: 10.1038/24567.

[10] Dan M. Stamper-Kurn and Wolfgang Ketterle. Spinor condensates and light scat-
tering from Bose–Einstein condensates. arXiv:cond-mat/0005001 (2000).

[11] Gary Ruben, Michael J. Morgan, and David M. Paganin. Texture Control in a
Pseudospin Bose–Einstein Condensate. Physical Review Letters, 105, 220402 (2010).
doi: 10.1103/PhysRevLett.105.220402.

44
[12] L. E. Sadler, J. M. Higbie, S. R. Leslie, M. Vengalattore, and D. M. Stamper-Kurn.
Spontaneous symmetry breaking in a quenched ferromagnetic spinor Bose–Einstein
condensate. Nature, 443, 312 (2006). doi: 10.1038/nature05094.

[13] F. Bloch. Zur Theorie des Ferromagnetismus. Zeitschrift für Physik, 61, 206 (1930).
doi: 10.1007/BF01339661.

[14] Tin-Lun Ho. Spinor Bose Condensates in Optical Traps. Physical Review Letters,
81, 742 (1998). doi: 10.1103/PhysRevLett.81.742.

[15] Tetsuo Ohmi and Kazushige Machida. Bose–Einstein condensation with internal
degrees of freedom in alkali atom gases. Journal of the Physical Society of Japan,
67, 1822 (1998). doi: 10.1143/JPSJ.67.1822.

[16] G. Edward Marti, Andrew MacRae, Ryan Olf, Sean Lourette, Fang Fang, and Dan M.
Stamper-Kurn. Coherent Magnon Optics in a Ferromagnetic Spinor Bose–Einstein
Condensate. Physical Review Letters, 113, 155302 (2014). doi: 10.1103/Phys-
RevLett.113.155302.

[17] Wolfgang Ketterle. Nobel lecture: When atoms behave as waves: Bose–Einstein
condensation and the atom laser. Reviews of Modern Physics, 74, 1131 (2002). doi:
10.1103/RevModPhys.74.1131.

[18] W. Ketterle, D. S. Durfee, and D. M. Stamper-Kurn. Making, probing and under-


standing Bose–Einstein condensates. arXiv:cond-mat/9904034 (1999).

[19] A. M. Guenault and Tony Guenault. Statistical Physics. Springer (2007).

[20] Daniel V. Schroeder. An Introduction to Thermal Physics. Addison Wesley (2000).

[21] Jason Remington Ensher. The First Experiments with Bose–Einstein Condensation
of 87 Rb. PhD thesis (1998).

[22] Anthony J. Leggett. Bose–Einstein condensation in the alkali gases: Some funda-
mental concepts. Reviews of Modern Physics, 73, 307 (2001). doi: 10.1103/RevMod-
Phys.73.307.

[23] Dallin Durfee and Wolfgang Ketterle. Experimental studies of Bose–Einstein con-
densation. Optics Express, 2, 299 (1998). doi: 10.1364/OE.2.000299.

[24] Franco Dalfovo, Stefano Giorgini, Lev P. Pitaevskii, and Sandro Stringari. Theory of
Bose–Einstein condensation in trapped gases. Reviews of Modern Physics, 71, 463
(1999). doi: 10.1103/RevModPhys.71.463.

45
[25] Allan Griffin, Tetsuro Nikuni, and Eugene Zaremba. Bose-Condensed Gases at Finite
Temperatures. Cambridge University Press (2009).

[26] P.B. Blakie, A.S. Bradley, M.J. Davis, R.J. Ballagh, and C.W. Gardiner. Dynamics
and statistical mechanics of ultra-cold Bose gases using c-field techniques. Advances
in Physics, 57, 363 (2008). doi: 10.1080/00018730802564254.

[27] S. J. Rooney, P. B. Blakie, and A. S. Bradley. The stochastic projected


Gross–Pitaevskii equation. Physical Review A, 86 (2012). doi: 10.1103/Phys-
RevA.86.053634.

[28] E. Zaremba, T. Nikuni, and A. Griffin. Dynamics of Trapped Bose Gases at Fi-
nite Temperatures. Journal of Low Temperature Physics, 116, 277 (1999). doi:
10.1023/A:1021846002995.

[29] Dan M. Stamper-Kurn. Peeking and poking at a new quantum fluid: Studies of
Bose–Einstein condensates in magnetic and optical traps. PhD thesis (2000).

[30] T. D. Lee, Kerson Huang, and C. N. Yang. Eigenvalues and Eigenfunctions of a Bose
System of Hard Spheres and Its Low-Temperature Properties. Physical Review, 106,
1135 (1957). doi: 10.1103/PhysRev.106.1135.

[31] N. Bogolubov. On the theory of superfluidity. J. Phys., 11, 23 (1947).

[32] Tapio Simula. Collective dynamics of vortices in trapped Bose–Einstein condensates.


Physical Review A, 87, 023630 (2013). doi: 10.1103/PhysRevA.87.023630.

[33] Allan Griffin. Conserving and Gapless Approximations for an Inhomogeneous Bose
Gas at Finite Temperatures. Physical Review B, 53, 9341 (1996). doi: 10.1103/Phys-
RevB.53.9341.

[34] Jacob Lyman Roberts. Bose–Einstein Condensates with Tunable Atom-atom Inter-
actions: The First Experiments with 85 Rb BECs. PhD thesis (2001).

[35] M. Edwards, R. J. Dodd, K. Burnett, and C. W. Clark. Zero-temperature, mean-


field theory of atomic Bose–Einstein condensates. National Institute of Standards
and Technology (1996).

[36] Mark Edwards, P. A. Ruprecht, K. Burnett, R. J. Dodd, and Charles W. Clark.


Collective excitations of atomic Bose–Einstein condensates. Physical Review Letters,
77, 1671 (1996). doi: 10.1103/PhysRevLett.77.1671.

[37] P. A. Ruprecht, Mark Edwards, K. Burnett, and Charles W. Clark. Probing the
linear and nonlinear excitations of Bose-condensed neutral atoms in a trap. Physical
Review A, 54, 4178 (1996). doi: 10.1103/PhysRevA.54.4178.

46
[38] Yvan Castin. Bose–Einstein condensates in atomic gases: simple theoretical results.
arXiv:cond-mat/0105058 (2001).

[39] L. Landau. Theory of the Superfluidity of Helium II. Physical Review, 60, 356 (1941).
doi: 10.1103/PhysRev.60.356.

[40] S. Stringari. Collective excitations of a trapped Bose-condensed gas. Physical Review


Letters, 77, 2360 (1996). doi: 10.1103/PhysRevLett.77.2360.

[41] D. S. Jin, J. R. Ensher, M. R. Matthews, C. E. Wieman, and E. A. Cornell. Collective


Excitations of a Bose–Einstein Condensate in a Dilute Gas. Physical Review Letters,
77, 420 (1996). doi: 10.1103/PhysRevLett.77.420.

[42] M.-O. Mewes, M. R. Andrews, N. J. van Druten, D. M. Kurn, D. S. Durfee, C. G.


Townsend, and W. Ketterle. Collective Excitations of a Bose–Einstein Condensate
in a Magnetic Trap. Physical Review Letters, 77, 988 (1996). doi: 10.1103/Phys-
RevLett.77.988.

[43] Alexander Altmeyer. Collective oscillations of an ultracold quantum gas in the BEC-
BCS crossover regime. PhD thesis, University of Innsbruck (2007).

[44] D. Guéry-Odelin and S. Stringari. Scissors Mode and Superfluidity of a Trapped


Bose–Einstein Condensed Gas. Physical Review Letters, 83, 4452 (1999). doi:
10.1103/PhysRevLett.83.4452.

[45] O. M. Maragò, S. A. Hopkins, J. Arlt, E. Hodby, G. Hechenblaikner, and C. J.


Foot. Observation of the Scissors Mode and Evidence for Superfluidity of a Trapped
Bose–Einstein Condensed Gas. Physical Review Letters, 84, 2056 (2000). doi:
10.1103/PhysRevLett.84.2056.

[46] T. P. Simula, M. J. Davis, and P. B. Blakie. Superfluidity of an interacting trapped


quasi-two-dimensional Bose gas. Physical Review A, 77, 023618 (2008). doi:
10.1103/PhysRevA.77.023618.

[47] U. Al Khawaja and H. T. C. Stoof. Nonlinear coupling between scissors modes


of a Bose–Einstein condensate. Physical Review A, 65, 013605 (2001). doi:
10.1103/PhysRevA.65.013605.

[48] R. Onofrio, D. S. Durfee, C. Raman, M. Köhl, C. E. Kuklewicz, and W. Ketterle.


Surface Excitations of a Bose–Einstein Condensate. Physical Review Letters, 84, 810
(2000). doi: 10.1103/PhysRevLett.84.810.

[49] T. P. Simula, S. M. M. Virtanen, and M. M. Salomaa. Surface modes and vortex for-
mation in dilute Bose–Einstein condensates at finite temperatures. Physical Review
A, 66, 035601 (2002). doi: 10.1103/PhysRevA.66.035601.

47
[50] L. M. Symes, D. Baillie, and P. B. Blakie. Static structure factors for a spin-1 Bose–
Einstein condensate. Physical Review A, 89, 053628 (2014). doi: 10.1103/Phys-
RevA.89.053628.

[51] D. M. Stamper-Kurn and W. Ketterle. Spinor Condensates and Light Scattering


from Bose–Einstein Condensates. In R. Kaiser, C. Westbrook, and F. David, editors,
Coherent atomic matter waves, number 72 in Les Houches - Ecole d’Ete de Physique
Theorique, pages 139–217. Springer Berlin Heidelberg (2001).

[52] E. A. Ostrovskaya, M. K. Oberthaler, and Y. S. Kivshar. Nonlinear Localization of


BECs in Optical Lattices. In Professor Panayotis G. Kevrekidis, Professor Dimitri J.
Frantzeskakis, and Professor Ricardo Carretero-González, editors, Emergent Nonlin-
ear Phenomena in Bose–Einstein Condensates, number 45 in Atomic, Optical, and
Plasma Physics, pages 99–130. Springer Berlin Heidelberg (2008).

[53] S. Yi, Ö. E. Müstecaplıoğlu, C. P. Sun, and L. You. Single-mode approximation


in a spinor-1 atomic condensate. Physical Review A, 66, 011601 (2002). doi:
10.1103/PhysRevA.66.011601.

[54] Jochen Kronjager. Coherent Dynamics of Spinor Bose–Einstein Condensates. PhD


thesis, University of Hamburg (2007).

[55] E. M. Bookjans, A. Vinit, and C. Raman. Quantum Phase Transition in an Antifer-


romagnetic Spinor Bose–Einstein Condensate. Physical Review Letters, 107, 195306
(2011). doi: 10.1103/PhysRevLett.107.195306.

[56] Dan M. Stamper-Kurn and Masahito Ueda. Spinor Bose gases: Symmetries, mag-
netism, and quantum dynamics. Reviews of Modern Physics, 85, 1191 (2013). doi:
10.1103/RevModPhys.85.1191.

[57] Eric Braaten and Agustin Nieto. Renormalization effects in a dilute Bose gas. Phys-
ical Review B, 55, 8090 (1997). doi: 10.1103/PhysRevB.55.8090.

[58] Nguyen Thanh Phuc, Yuki Kawaguchi, and Masahito Ueda. Beliaev theory of
spinor Bose–Einstein condensates. Annals of Physics, 328, 158 (2013). doi:
10.1016/j.aop.2012.10.004.

[59] Juha Javanainen and Janne Ruostekoski. Symbolic calculation in development of


algorithms: split-step methods for the Gross–Pitaevskii equation. Journal of Physics
A: Mathematical and General, 39, L179 (2006). doi: 10.1088/0305-4470/39/12/L02.

[60] Barry I. Schneider and Lee A. Collins. The discrete variable method for the solution
of the time-dependent Schrödinger equation. Journal of Non-Crystalline Solids, 351,
1551 (2005). doi: 10.1016/j.jnoncrysol.2005.03.028.

48
[61] J. J. Sakurai. Modern Quantum Mechanics. Addison Wesley, revised edition (1993).

[62] D. S. Hall. Multi-Component Condensates: Experiment. In Professor Panayotis G.


Kevrekidis, Professor Dimitri J. Frantzeskakis, and Professor Ricardo Carretero-
González, editors, Emergent Nonlinear Phenomena in Bose–Einstein Condensates,
number 45 in Atomic, Optical, and Plasma Physics, pages 307–327. Springer Berlin
Heidelberg (2008).

[63] George Edward Marti. Scalar and Spinor Excitations in a Ferromagnetic Bose–
Einstein Condensate. PhD thesis, University of California, Berkeley (2014).

49

You might also like