ch3_interaction_effects

3 Interaction Effects
Learning Objectives
• Define Bose–Einstein condensation in an interacting boson system in terms of macro-

scopic occupation.
• Introduce the Gross–Pitaevskii equation and its applicable conditions.
• Derive the hydrodynamic equations and various physical consequences of the hydrody-
namics with or without harmonic trap.
• Introduce sound waves and emphasize the difference between wave behavior and
diffusive behavior.
• Introduce the criterion for superfluidity and determine the superfluid critical velocity.
• Introduce two-fluid hydrodynamics.
• Introduce the Bogoliubov theory.
• Introduce various concepts related to quasi-particles, including ground state as a vacuum
of quasi-particles, the zero-point energy of quasi-particles, quasi-particle lifetime, and
vacuum fluctuation.
• Introduce healing length as the characteristic length scale for low-energy physics of an
interacting Bose condensate.
• Summarize the difference between an interacting and noninteracting Bose condensate.
• Introduce Bragg spectroscopy.
• Show that the Bogoliubov theory fails in one dimension due to the infrared divergence.
• Discuss the basic idea of the Bethe-ansatz solution and the Tonks–Girardeau limit.
• Introduce scale-invariant quantum gases.
• Introduce the Josephson effect and self-trapping as two different transport phenomena
driven by phase coherence.
• Discuss the conjugate relation between the relative phase fluctuation and the relative
particle number fluctuation and show that strong enough phase fluctuation renders the
system into a Fock state.
• Discuss an example of quantum measurement that can project two independent conden-
sates to a coherent superposition state.
• Discuss interference effects in density and density–density correlations.
• Discuss the relation between the instability of the Schödinger Cat state and the stability
of symmetry breaking.
73
https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

3.1 Bose–Einstein Condensation
Before we discuss an interacting system, let us first review the concept of Bose–Einstein
condensation (BEC) in a free boson system. Considering free bosons with dispersion k =
2 k2 /(2m) in three dimensions, at high temperatures, each mode with a given wave vector
k is populated by
1
nk = , (3.1)
e( k −μ)/(kB T) − 1
where μ is the chemical potential and is negative at high temperature. In statistical mechan-
ics, we always consider the thermodynamic limit that N → ∞ and V → ∞ with density
n = N/V fixed. It is important to note that the population fraction n k /V at each mode
vanishes in the thermodynamic limit. This is always true as long as the chemical potential
μ is negative. As temperature decreases, μ increases. Then the question is whether there
exists a critical temperature, denoted by Tc , at which the chemical potential μ will increase
to zero. If such a Tc exists, by setting μ = 0, Tc is given by
∞ √
V 1 V(mkB Tc )3/2 zdz
N= 3
d k /(k T ) = √ . (3.2)
(2π )3 e k B c −1 2π 2 3 0 ez −1
Because the integration in the r.h.s. of Eq. 3.2 is finite, Eq. 3.2 is equivalent to 1/λ3 ∼
n ∼ 1/d3 , where d is the mean interparticle spacing and 2 /(mλ2 ) = kB T/(2π) defines
the thermal de Broglie wave length. The physical meaning of Eq. 3.2 is that the thermal
de Broglie wave length is comparable to the mean inter-particle distance d, or equivalently
speaking, kB Tc is comparable to the degenerate energy 2 /(md2 ). It is the same condition
as fermions entering quantum degeneracy.
Below Tc , μ cannot further increase to be positive for free bosons, and μ should retain
zero. In this case, the occupation at k = 0 mode N0 should be considered separately, that is,

V 1
N = N0 + d3 k /(k T) . (3.3)
(2π ) 3 e k B −1
It is easy to show that N0 /N = 0 in the thermodynamic limit, and this is taken as the Bose–
Einstein condensate (BEC). That a population fraction at a certain mode is nonzero in the
thermodynamic limit is called the macroscopic occupation. From the discussion of BEC
in a free bosons system, we learn that the Bose condensation is a transition from one sit-
uation that population fractions at all modes vanish in the thermodynamic limit to another
situation that at least one mode is macroscopically occupied. In fact, as we will summarize
at the end of Section 3.3, many properties are different between an interacting BEC and
a noninteracting BEC. In order to define BEC in an interacting system, it is important to
first identify which property of a noninteracting BEC should be regarded as the essential
defining property of a BEC, and we should capture and generalize this defining property
to an interacting system. Here we argue that one should take this macroscopic occupation
as the defining property of a BEC. Nevertheless, in a free system, it is straightforward to

75 Bose–Einstein Condensation
define the occupation in single-particle eigenmodes. Hence, the question is how to prop-
erly define occupation in an interacting system, and the answer to this question leads to the
concept of the off-diagonal long-range order (ODLRO) [183].
General Definition of BEC. Now we present a general definition of BEC in an interacting
system. First of all, let us introduce the concept of the density matrix. When the many-
body system is in a pure state with the many-body wave function (r1 , r2 , . . . , rN ), the
one-body density matrix is defined as

ρ(r, r ) = N ∗ (r, r2 , . . . , rN )(r , r2 , . . . , rN )d3 r2 . . . d3 rN . (3.4)
If the system is in a mixed state with probability ps in a many-body wave function

s (r1 , r2 , . . . , rN ), the one-body density matrix is defined as

ρ(r, r ) = N ps s∗ (r, r2 , . . . , rN )s (r , r2 , . . . , rN )d3 r2 . . . d3 rN . (3.5)
s
Alternatively, in a second quantized form, it can be equivalently defined as

ˆ † (r)(
ρ(r, r ) = ˆ r ) , (3.6)
where ˆ † (r) and (r)
ˆ are boson creation and annihilation operators at position r,
respectively. This one-body density matrix ρ(r, r ) can be diagonalized and decomposed as

ρ(r, r ) = Ni ψi∗ (r)ψi ( r ). (3.7)
i
Therefore, the eigenvector ψi defines the wave function of each mode, and Ni defines
the single-particle occupation of each mode. With this definition of occupation of single-
particle modes, we can now introduce the definitions of BEC in an interacting system as
follows:
• If, for all Ni , limN→∞ Ni /N = 0, we call it a normal phase.
• If there is one and only one Ni , limN→∞ Ni /N = 0, we call it a simple BEC.
• If there are more than one Ni , limN→∞ Ni /N = 0, we call it a fragmented BEC.
For the latter two cases, we say the system has ODLRO [183].
Similarly, we can introduce a higher-order density matrix. For instance, a two-body
density matrix can be defined as
ˆ † ( r 1 )
ρ(r1 , r2 , r1 , r2 ) = ˆ † (r2 )(r
ˆ 1 )(
ˆ r2 ) . (3.8)
Usually we consider the situation that r1 ≈ r2 ≈ r and r1 ≈ r2 ≈ r ; we can similarly
decompose the two-body density matrix as

ρ(r1 , r2 , r1 , r2 ) = Ni ψi∗ ( r1 ≈ r2 ≈ r)ψi (r1 ≈ r2 ≈ r ). (3.9)
i
In this way, we can define a boson pair condensate when there exists one or more Ni that
satisfy limN→∞ Ni /N = 0. In many cases, a fragmented BEC defined by the one-body
density matrix can be a simple BEC defined by the higher order density matrix [123]. We
will come back to revisit this in Section 3.5.

In ultracold atom experiments, a BEC is achieved after the laser cooling and the evap-
orative cooling. The first measurement proving Bose condensation is the time-of-flight
measurement of the momentum distribution [5, 44], which experimentally determines the
onset of BEC by this property of macroscopic occupation. When the harmonic trap is sud-
denly released, atoms acquire a velocity v = k/m where k is the momentum of atoms
before turning off the trap. Afterward, these atoms flight in free space with this velocity.1
After long enough time, when the size of the initial cloud can be ignored compared with
the distance that atoms have traveled, the initial momentum distribution can be revealed by
measuring the distance that atoms have flighted and dividing the distance by the time that
they have flighted. In other words, the spatial distribution after the time of flight, as shown
in Figure 3.1 as an example, reveals the momentum distribution before turning off the trap.
Two typical sets of measurements from the first BEC experiments are shown in Figure 3.1.
At higher temperature, this reveals the momentum distribution of thermal bosons. And
below certain temperature, there is a sudden onset of peak which expands much slower
(a) (b) Velocity (cm/s)

4.71
–5 0 5
4.25 0.76 MHz
4.23
0.68 MHz
4.21
Opacity Density
4.19
vevap(MHz)
0.63 MHz
4.16
0.53 MHz
4.11
0.5
0.46 MHz
4.06 0
0.0 0.5 1.0
300 µm Position (mm)
tFigure 3.1 Time-of-flight measurements of momentum distribution of a Bose gas. The onset of a sharp peak in the momentum
distribution marks the Bose–Einstein condensation transition. The number at each curve is called νevap , which labels
the radio-frequency used in the evaporative cooling. The smaller this number is, the cooler is the gas. (a) is reprinted
from Ref. [5], and (b) is reprinted from Ref. [44].
1 If there are other lasers to create optical lattices, or spin-orbit coupling, or other effects, as we will discuss in
later chapters, they will also be turned off at the same time when the trapping laser is turned off. This flighting
velocity k/m is not necessarily the same as the velocity of atoms before turning off the trap, and the latter
is defined as v = ∂E k /∂k, where E k is the single-particle dispersion in the presence of optical lattices, or
spin-orbit coupling or other effects. That is to say, the time-of-flight measures initial momentum distribution
not the initial velocity distribution.

77 Bose–Einstein Condensation
compared with the thermal components, because these atoms are the condensed part and
their expansion dynamics are no longer driven by the kinetic energy. The expansion dynam-
ics of the condensed part will be discussed in Section 3.2. As shown in Figure 3.1, below
Tc , this component increases, and the atom number inside this peak becomes a fraction of
the total number of atoms. This macroscopically occupied component is considered as the
Bose condensate, and the ratio between atoms in this component and the total atom number
is called the condensate fraction.
Below we will first focus on a simple BEC. For a simple BEC, since one eigenvalue
of the density matrix (say, N0 ) is much larger than any others, we can make a bold
approximation for the density matrix as
ρ(r, r ) N0 ψ ∗ (r)ψ(r ). (3.10)
Here N0 is of the order of N, and ψ(r) is the corresponding eigenvector. Eq. 3.10 means a
long-range correlation because ρ(r, r ), or equivalently speaking, the correlation function
√
ˆ ) , does not vanish even when |r−r | is taken to infinity. Here N0 ψ(r) is called
ˆ † (r)(r
the condensate wave function. The focus below is to discuss the equations that govern the
ground state and the dynamical behaviors of this condensate wave function. To this end,
we need to work out a microscopic theory, and we should make a proper approximation
for the underlying many-body wave function. Since here we consider our approximation
equation 3.10 of ρ(r, r ) is the key for BEC, let us ask an inverse question of what kind of
many-body microscopic state can give rise to such a one-body density matrix. In fact, at
least we can come up with the following two ways to satisfy Eq. 3.10.
• We can assume the system is a pure state and the many-body wave function is a product
state as

N
(r1 , . . . , rN ) = ψ(ri ). (3.11)
i=1
It is easy to verify that the wave function of Eq. 3.11 reproduces the density matrix given
by Eq. 3.10.
• We can assume that the wave function is a coherent state given by
√
ˆ † ( r)
d3 r N0 ψ(r)
| = e |0 , (3.12)
√
ˆ
and using the property of the coherent state, we have (r)| = N0 ψ( r)| . Then it
is straightforward to show that
ˆ ) = N0 ψ ∗ (r)ψ( r ).
ˆ † (r)(r (3.13)
Before ending this part, we shall make two remarks:
• Though these two microscopic descriptions look quite different, they give the same one-
body density matrix with ODLRO. As we will show in the following two sections, our
derivation of the hydrodynamic theory ultilizes the first description, and the Bogoliubov
theory is based on the second description. We will see that these two different theories
give the same low-energy excitation spectrum.

• The density matrix itself does not infer the global phase of the condensate wave function.
However, once a condensate wave function is chosen, we have to choose a fixed global
phase. Nevertheless, the energies of wave functions with different global phases are
degenerate, and we will discuss the connection to the gapless phonon mode in the next
section.
3.2 Hydrodynamic Theory
Now we consider an interacting many-boson system, and we start from the pseudo-
potential model described in Section 2.2:

N
p̂2i 4π 2 as ∂
Ĥ = + V( ri ) + δ(rij ) rij . (3.14)
2m m ∂rij
i=1 i<j
We evaluate the energy expectation value of this Hamiltonian under the wave function
Eq. 3.11. Here we should note that the wave function Eq. 3.11 remains regular at short-
distance between two atoms, and it does not obey the short-range behavior of two-body
wave function as we discussed in Sec 2.2, which requires that the wave function diverges
as 1/rij when ri approaches rj . The question is that how we reconcile the inconsistency
between this many-body wave function of Bose condensation and the requirement for two-
body wave function at short distance. To this end, we should emphasize that the assumption
of Bose condensation wave function is only able to reproduce the approximate form of the
density matrix Eq. 3.10, which ignores all the modes that are not macroscopically occupied.
In other words, this microscopic wave function only captures the mode that is macroscopi-
cally occupied. Since this macroscopically occupied mode must be a low-lying mode, this
is equivalent to say that this trial wave function only captures the low-energy and the long
wavelength physics, and the wavelength should be much larger than the interparticle spac-
ing. Therefore, it is natural that this wave function fails to capture short-range and high
energy physics at the scale comparable or shorter than the interparticle spacing. We shall
always keep this in mind that the following theory based on this assumption can only be
applied to the length scale larger than interparticle spacing.
Since the wave function Eq. 3.11 is regular at short distance between any two particles,
∂
∂rij rij will not play any role and it is straightforward to calculate the energy as

E ∗2 ∇ 2 (N − 1) 4π 2 as
= d r ψ (r) −
3
+ V( r) ψ(r) + |ψ( r)| .
4
(3.15)
N 2m 2 m
√
Minimizing the energy with respect to ψ ∗ , and redefining Nψ as ψ, one obtains

2 ∇ 2
− + V(r) ψ + U|ψ|2 ψ = μψ, (3.16)
2m

79 Hydrodynamic Theory
where μ is the chemical potential. The dynamical version of this equation is given by
2 2
∇ ∂ψ
− + V(r) ψ + U|ψ|2 ψ = i . (3.17)
2m ∂t
Here U denotes 4π2 as /m. Throughout this and next chapter, we always consider posi-
tive as and U > 0, unless specifically stated. Here U > 0 means repulsive interaction
between atoms, as we have discussed in Box 2.1. Eq. 3.16 and Eq. 3.17 are known as
the Gross–Pitaevskii equation, often short-noted as the GP equation. These equations are
also called nonlinear Schrödinger equations. Mathematically, they differ from the single-
particle Schrödinger equation because of the presence of the nonlinear term U|ψ|2 ψ. Here
we should also remark the physical difference between this equation and the single-particle
Schrödinger equation, although both of them are equations for single-particle wave func-
tion.2 The single-particle Schrödinger equation describes a system with only one particle
alone, and here the nonlinear Schrödinger equation describes a system with macroscopic
number of particles occupying a single-particle mode, where the nonlinear term presents
the interaction effects between these atoms. Below, we will mostly focus on the interaction
effects. First of all, for a uniform system with V(r) = 0, the density of the ground state
should be uniform, therefore we have the chemical potential μ = Un0 , where n0 = |ψ|2
is the mean condensate density. This already differs from the non-interacting case where μ
is always zero for a BEC. To further discuss the dynamical behaviors and excitations, we
proceed to introduce the hydrodynamic equations.
√
Hydrodynamic Equation. We decompose ψ as neiθ , where n is the density and θ is
√
the phase. Both n and θ are functions of space and time. Substituting ψ = neiθ into the
time-dependent GP equation Eq. 3.17, and after eliminating eiθ from both sides, the real
part of the equation gives

∂θ 2 1 2 √ 1 2
=− − √ ∇ n + mv s + V(r) + Un , (3.18)
∂t 2m n 2
where v s = ∇θ/m is the superfluid velocity. Taking the derivative ∇ at both sides, we
obtain

∂v s 2 1 2 √ 1 2
m = −∇ − √ ∇ n + mvs + V(r) + Un . (3.19)
∂t 2m n 2
The first term in the bracket of the r.h.s. of Eq. 3.19 is called the quantum pressure, and
other terms are in turn the kinetic energy, the trapping energy and the interaction energy.
Since the gradient of energy is force, the physical meaning of Eq. 3.19 is nothing but
F = ma in the classical mechanics. Thus Eq. 3.19 is also called the Newton equation. The
imaginary part gives
√
∂ n √ √
=− 2(∇ n) · (∇θ) + n∇ 2 θ , (3.20)
∂t 2m
and by using the definition of vs , it leads to
∂n
+ ∇ · (nv s ) = 0. (3.21)
∂t
2 Here single-particle wave function means the wave function has a single spatial coordinate r as its variable.

This equation means that the change in the local density should be equal to the net flux of
current. This equation is also called the continuity equation. Eq. 3.19 and Eq. 3.21 form
the zero-temperature superfluid hydrodynamic equations.
Hydrodynamic equations refer to equations that govern dynamics of a fluid. Here we
should make an important remark between the hydrodynamics and the BEC. We have
shown that a BEC with interaction naturally leads to the hydrodynamic behavior, but on the
other hand, hydrodynamics does not always require Bose condensation. Strong interactions
can also lead to the hydrodynamic behavior. In normal fluids the strong interaction due to
high density leads to hydrodynamic behavior. In dilute gas strong interaction can arise
from Feshbach resonances. Experiments have also observed the hydrodynamic behavior in
strongly interacting ultracold atomic gases nearby a Feshbach resonance, even when the
system is not cold enough to become a superfluid [129].
Sound Velocity. Let us first consider a uniform system with V(r) = 0, the ground state
has a uniform density n0 and v s = 0. We can expand n = n0 + δn, and simplify
the hydrodynamic equations by focusing on near-equilibrium and low-energy dynamics.
Because of being near equilibrium, we can only keep the leading order of δn and v s , and
because of being low-energy dynamics, we only need to keep the leading order of k and
ignore the higher-order derivative terms. Thus, the hydrodynamic equations can be greatly
simplified as
∂v s
m = −U∇δn (3.22)
∂t
∂δn
= −∇(n0 v s ). (3.23)
∂t
Here the density fluctuation and the phase fluctuation are locked together,3 which gives
rise to a single low-energy mode described by
∂ 2 δn Un0 2
= ∇ δn. (3.24)
∂t 2 m
This equation contains both second-order time and spatial derivatives, which is a wave
equation.
Here we should note the general difference between a wave equation and a diffusion
equation. Generally, considering an observable W,
∂ 2W
= c2 ∇ 2 W, (3.25)
∂t2
is called a wave equation, such as Eq. 3.24, where c is a constant velocity. It is easy to
show that if W obeys such a wave equation, when a profile is created in W(r), the shape
of the profile will keep unchanged as time evolves, and its center propagates in space with
the velocity c. That is also the reason why waves can carry information. In nature, sound
wave, electromagnetic wave and gravitational wave all obey wave equations, and they all
3 It will be different in the relativistic case where density and phase modes are decoupled at the lowest order, as
we will discuss in Section 8.1.

can carry information. In contrast, a diffusive equation contains a first-order time derivative
and a second-order spatial derivative, which can be generally written as W,
∂W
= D∇ 2 W, (3.26)
∂t
where D is usually called diffusion constant. If W obeys such a diffusion equation, when a
profile is created in W(r), the shape of the profile will gradually smear out as time evolves.
In a normal state, heat and Brownian motion are examples of diffusive motions. Below,
when we discuss the two-fluid hydrodynamics, we will discuss that the propagation of
the heat changes from the diffusive behavior to the wave behavior across the superfluid
transition.
The wave equation Eq. 3.24 gives rise to sound wave, or phonon mode, as the low-energy
excitations, and the phonon mode has a linear dispersion as

ω = Un0 /m|k|, (3.27)
√
where the phonon velocity c = Un0 /m. The sound wave describes the propagation of
a density deviation from the equilibrium density. Experimentally, one can use a focused
laser beam to create either a local density dip or a density hump at the center of a BEC,
and then watch the motion of this density dip or hump [8]. The results are shown in Figure
3.2. From this measurement one can see that the shape of the density dip or hump does
not change and its location moves from the center toward the edge of the cloud, from
which one can deduce the sound velocity. By repeating such measurements with different
densities, one can obtain the relation between the sound velocity and the density, as shown
in Figure 3.2(d), which verifies Eq. 3.27.
This low energy gapless excitation with linear dispersion is also called the Goldstone
mode. Note that the density and the phase are coupled through the continuity equation, the
phonon mode can be also be viewed as an excitation of spatially twisting phase of the con-
densate wave function. In the long wave length limit by taking k → 0, the excitation turns
to a uniform rotation of the phase of the condensate wave function. As discussed in Section
3.1, since two condensate wave functions with different global phases are degenerate, the
excitation energy vanishes in the long wave length limit.
Superfluidity and Critical Velocity. The phonon mode with linear dispersion is the only
mode at the lowest energy. As we will discuss here, this has very dramatic consequence.
Considering an impurity with mass m0 moving inside a condensate with velocity vi , friction
occurs when this impurity can be scattered to another velocity vf and the momentum is
transferred into an excitation of condensate with momentum q. Here we remark that, for a
system with Galilean invariance, this is equivalent to a system moving with velocity vi in
the presence of a static impurity. However, for a systems without Galilean invariance, the
two cases are not equivalent, which leads to two distinct critical velocities. We will discuss
such an example in the spin-orbit coupled BEC in Section 4.5.
Let us now focus on the situation with a moving impurity in a static BEC. When this
linearly dispersive mode is the only low-lying excitation, the momentum conservation and
the energy conservation together give

tFigure 3.2 Measurement of the phonon velocity in a BEC. (a–b) A local density dip or a hump is created by a focused laser beam
in the center of a BEC, which propagates toward the edge of a BEC. (c) Propagating of the local density modulation at
different times, from which one can deduce the sound velocity. (d) Measured sound velocity as a function of atom
density. Reprinted from Ref. [8]. A color version of this figure can be found in the resources tab for this book at
cambridge.org/zhai.
m0 vi = m0 vf + q (3.28)
m0 v2i m0 v2f
= + c|q|. (3.29)
2 2
Replacing vf = vi − q/m0 in Eq. 3.29, we obtain
q2
vi · q − c|q| = . (3.30)
2m0
Therefore, if |vi | < c, Eq. 3.30 cannot be satisfied. That means if the velocity of a moving
impurity is smaller than the phonon velocity, it cannot be scattered and there is no friction
for its motion. With the Galilean invariance, it is equivalent to say, when the fluid moves
with a velocity smaller than c, there is no friction. This phenomenon is known as superflu-
idity. The upper bound vc for the velocity of a moving fluid without friction is called the
superfluid critical velocity. Thus, the sound velocity here equals to the superfluid critical
velocity. For a more general isotropic quasi-particle dispersion E(|q|), the similar argument
leads to a general condition for the critical velocity given by


E(|q|)
vc = min . (3.31)
|q|
This is known as the Landau criterion for the superfluid critical velocity. From the Lan-
dau criterion it is also very clear that for a noninteracting BEC, the dispersion remains
quadratic and the critical velocity is zero. In other words, a noninteracting BEC is not a
superfluid. Bose condensation and interaction together lead to superfluidity. We will apply
this criterion again when we discuss the Fermi superfluid in Section 6.2.
Soon after the BEC is achieved, the existence of a critical velocity is also experimen-
tally confirmed [146]. Experimentally, they scan a focused beam back and forth in a BEC,
as shown in Figure 3.3(a). Then, after a certain scanning time, they turn off the scan and
let the condensate thermalize again. After that they measure the increasing of the thermal
fraction in the BEC. If there exists a critical velocity, because there is no friction and no
excitation when the scanning velocity is below the critical velocity, the thermal fraction
will not change. But when the scanning velocity is above the critical velocity, thermal frac-
tion increases as the velocity increases. In this experiment [146], they have tried different
scanning frequencies and amplitudes. When they plot the thermal fraction as a function of
the scanning velocity, as shown in Figure 3.3(b), the measurements with different frequen-
cies collapse nearly into the same curve, which clearly displays the behavior of a critical
velocity.
Thomas–Fermi Distribution. Now we consider the physical consequences of the hydro-

dynamic equation in the presence of a trapping potential V(r). By ignoring the quantum
tFigure 3.3 Experimental measurement of the critical velocity of a BEC. (a) Schematic of a local stirring of the condensate with a
fixed velocity v. (b) Thermal fraction after certain duration of stirring as a function of the velocity, for different stirring
amplitudes and frequencies. The dashed line indicates the critical velocity. Reprinted from Ref. [146].

pressure term in the Newton equation, the equilibrium solution gives vs = 0 and the
equilibrium density distribution
μ − V(r)
n0 (r) = (3.32)
U
for the regime with μ > V(r), and n0 (r) = 0 for μ < V(r). The equation μ = V(R)
determines the boundary of a BEC. For a harmonic trap, the density profile is an inverted
parabolic function, which is also called the Thomas–Fermi distribution. Note that μ = Un0
is the equation-of-state for a uniform system, the result of Eq. 3.32 can be interpreted as
the local density approximation, that is, we simply replace the chemical potential μ in the
equation-of-state of a uniform system by a “local chemical potential” μ(r) = μ − V(r).
Furthermore, the total number conservation equation is

d3 rn0 (r) = N, (3.33)
r⊂R
which determines μ for each given N, and fixes the entire density profile. For instance,
considering an isotropic harmonic trap mω02 r2 /2, one can find 4π mω02 R5 = 15UN, and
thus R ∝ N 1/5 .
Ignoring the quantum pressure term is crucial for obtaining the Thomas-Fermi distribu-
tion. Is this approximation justified, or consistent with the resulting density profile? Here,
with Eq. 3.32, let us do a “back-of-envelope” estimation. Using the result R ∝ N 1/5 , we
can estimate that the total interaction energy is proportional to UnN ∝ UN 2 /R3 ∝ UN 7/5 ,
and the total harmonic trapping energy is ∼ R2 N ∝ N 7/5 . For large number of atoms, since
these two terms have the same dependence on N, they can balance each other. However,
the quantum pressure term is originated from the density inhomogeneity, and is given by
N/R2 ∝ N 3/5 . Thus, when N is sufficiently large, the quantum pressure energy is always
much smaller compared with both the interaction energy and the harmonic trapping energy,
and therefore, the quantum pressure term can be safely ignored in the hydrodynamic
equation. This is a significant difference between the interacting and the noninteracting
cases. In the noninteracting case, the ground state density profile is determined by the bal-
ance between the harmonic trap potential and the kinetic energy, which gives a Gaussian
distribution. However, in the interacting case with large number of bosons, the ground
state density profile is determined by the balance between the interaction energy and the
harmonic trapping energy, which results in an inverted parabola.
Low-Energy Modes in a Harmonic Trap. Now we consider the low-energy excitations
describing the density fluctuation on top of the equilibrium Thomas-Fermi distribution. Let
us write
μ − V(r)
n(r, t) = n0 (r) + δn(r, t) = + δn( r, t), (3.34)
U
and to the linear order of δn, one obtains

∂ 2 δn μ − V(r)
= ∇ ∇δn (3.35)
∂t2 m

μ − V(r) Un0 (r) 2
=∇ ∇δn + ∇ δn, (3.36)
m m

where the first term in Eq. 3.36 varies in space of the order of 1/R. Considering δn varying
in space with a typical wave vector k, and if k 1/R, the first term in Eq. 3.36 can be
ignored compared with the second term in Eq. 3.36, then, it yields a sound mode with ω =
√
c(r)k, where c(r) = Un0 (r)/m. This is consistent with the local density approximation
where sound propagates with a local sound velocity.
When k ∼ 1/R, the first term becomes important. In this regime, the collective
mode becomes a global motion of entire condensate. For simplicity, let us consider
a three-dimensional isotropic trap as V( r) = mω02 r2 /2, we can expand δn as δn =
Pl2nr (r/R)rl Ylm (θ , φ). Here Pl2nr (r/R) are polynomials of degree 2nr , which only con-
tain even-power terms and satisfy the orthogonality condition. This polynomial contains
n nodes in the radial direction. The solution of frequencies can be obtained as [167, 137]

ω(nr , l) = ω0 2n2r + 2nr l + 3nr + l. (3.37)
This should be compared with the spectrum of noninteracting harmonic oscillator where
the excitation spectrum is given by ω(nr , l) = ω0 (2nr + l). The difference is another
manifestation of the interaction effects.
Surface modes. Modes with nr = 0 are called the surface modes because the density
change δn mostly manifests itself around the √ boundary, as shown in Figure 3.4 (a) and
(b). Their frequencies are given by ω = ω0 l. For l = 1, the density changes as δn =
rY1m = x ± y, z. Adding such δn into the Thomas-Fermi distribution in a harmonic trap,
it is easy to see that this mode corresponds to a global shift of the density distribution, as
shown in Figure 3.4(a). It describes the center-of-mass oscillation of the entire condensate
in the harmonic trap. This is called the dipole mode and its frequency is the same as trap
frequency independent of the interactions. Figure 3.4(b) shows the case with l = 2 called
the quadrupole mode, which causes a quadrupole deformation of the surface. In Figure 3.5
we show experimental observation of the quadrupole mode, together with the hexadecapole
mode with l = 4. Other than l = 1, the frequencies of all surface modes are smaller than
that of their noninteracting counterpart.
tFigure 3.4 Schematic of the collective modes of a BEC. (a) The dipole model with nr = 0, l = 1. (b) The quadrupole mode
with nr = 0, l = 2. (c) The breathing mode with nr = 1, l = 0. The solid line denotes the equilibrium
boundary, and the dashed and dotted lines denote the boundary at T/4 and 3T/4, respectively. T is a period of the
collective mode oscillation.

tFigure 3.5 The observation of the surface modes. (a) Density profile at different times of the oscillation. The upper panel is the
quadrupole mode with l = 2, and the lower panel is the hexadecapole mode with l = 4. (b) Schematic of the
l = 4 mode. Reprinted from Ref. [132].
Compressional modes. Modes with nr = 0 are called the compressional mode. In partic-
ular, δn for the mode with nr = 1 and l = 0 has one node in the radial direction, which
corresponds to either increasing density at the center and decreasing the density at the edge,
or vice versa. This periodic change in the condensate size is called
√ the breathing mode, as
shown in Figure 3.4(c). The frequency of this breathing mode is 5ω0 .
Anisotropic Expansion. Now let us consider the time-of-flight expansion of a Bose

condensate. As we have discussed above, the time-of-flight expansion of the normal com-
ponent is a ballistically expansion with a velocity k/m. However, the expansion of the
Bose condensed component follows the hydrodynamic equations. From Eq. 3.19, one can
see that, when the trap is turned off, the acceleration of the velocity in the initial stage of
this expansion is mostly determined by the gradient of the local internal energy ∇(Un0 ).
Considering an anisotropic trap
1 1
V(r) = mω2 z2 + mω⊥ (x + y2 ),
2 2
(3.38)
2 z 2
and if ωz ω⊥ , then the size Rz along ẑ is much longer than the size R⊥ in the transverse
plane, that is,

2μ 2μ
Rz = 2
R⊥ = 2
. (3.39)
mωz mω⊥
Therefore, it is also easy to see that the gradient term ∂z n(r) along ẑ is smaller than the gra-
dient terms ∂x n(r) or ∂y n(r) in the transverse plane. According to Eq. 3.19, a larger gradient

term will give rise to a faster acceleration, and therefore, a larger cloud size after expan-
sion. Thus, the aspect ratio of the cloud size is inverted before and after the time-of-flight
expansion. This is sharply in contrast to the thermal cloud where the cloud size will finally
become isotropic after long time time-of-flight, because the momentum distribution of the
thermal atoms are isotropic. This inverted aspect ratio is a hallmark of the hydrodynamic
behavior, which has also been observed in the experiments of BEC [84, 163], and later also
in strongly interacting Fermi gases nearby a Feshbach resonance [129].
Two-Fluid Hydrodynamics. Now we return to the uniform situation but we concern

about the finite temperature effect. Let us now ask ourselves following question. From
the previous discussion of the hydrodynamic equations, we have obtained a sound veloc-
√
ity c = Un0 /m, where n0 is the condensate density. If we simply extend this formula
to finite temperature, such a velocity will vanish at the transition temperature Tc . On the
other hand, we know from our daily physical intuition that there is always a sound velocity
even in normal gas above Tc , and such velocity should always be finite at Tc . How shall we
reconcile this contradiction ? This contradiction seems to indicate that we have missed a
sound mode below Tc . To solve this problem, we need to introduce a hydrodynamic theory
for the superfluid at finite temperature, which is known as the two-fluid hydrodynamics.
Here two fluids means one superfluid component describing the macroscopically occu-
pied mode and another normal component describing particles in all the other excited
modes. Landau formulated and solved the two-fluid equations in the non-dissipative limit.
The two-fluid model is described by several variables including the superfluid density ns ,
the superfluid velocity vs , the normal density nn , the normal velocity vn and entropy S.
These quantities satisfy following four two-fluid hydrodynamic equations [100]
∂n
+ ∇j = 0, (3.40)
∂t
∂S
+ ∇(Svn ) = 0, (3.41)
∂t

∂vs 1
= −∇ μ + v2s , (3.42)
∂t 2
∂ji ∂ij
+ = 0. (3.43)
∂t ∂xj
Here the total density n is a sum of both the superfluid component and the normal com-
ponent as n = ns + nn . ns vanishes at the transition temperature, and nn vanishes at zero
temperature. j is the total current given by j = ns vs +nn vn . Eq. 3.40 is an equation for total
number conservation. Eq. 3.41 is the entropy conservation because the nondissipative limit
is considered here. Here the entropy is only carried by the normal component, and there-
fore the entropy flows with the velocity vn . Eq. 3.42 governs the motion of the superfluid
velocity vs , which in fact takes the same form as Eq. 3.19. Eq. 3.43 governs the motion of
the total current, where ij is momentum flux density tensor given by
ij = Pδij + ns vsi vsj + nn vni vnj , (3.44)

where P is the pressure, i and j label spatial directions, and vsi and vni are the i = x, y, z
component of vs and vn , respectively.
We will not further proceed the detailed derivation of solving the two-fluid hydrody-
namics. Here we just briefly mention that the consequence of the two-fluid hydrodynamics
is the existence of another sound mode. The physical meaning is that, the entropy, or the
heating, also propagates as a wave in the superfluid phase. These two waves of the heat and
the density in general are hybridized, which gives rise to the first and the second sounds
in a superfluid. This is a highly unconventional phenomenon and is sharply in contrast to
our daily intuition from a normal fluid, where the motion of heat is diffusive. Approaching
Tc from below, one of the sound waves becomes density wave, which survives above Tc .
The other sound wave becomes heat wave, which undergoes a transition from the wave
behavior to the diffusive behavior across Tc .
3.3 Bogoliubov Theory
Now we take an alternative approach to treat the excitations of a BEC. In the Bogoliubov
theory, we start with a second quantized Hamiltonian with renormalizable contact potential
introduced in Section 2.2:
g
Ĥ = ( k − μ)â†k â k + â†k1 â†k2 âk3 âk4 , (3.45)
2V
k k1 k2 k3 k4
with k1 + k2 = k3 + k4 , where k = 2 k2 /(2m). The Bogoliubov theory is a perturbative

approach. Let us first consider the noninteracting limit that bosons are all condensed in the
zero-momentum state. Following our discussion in Section 3.1, we take the unperturbed
wave function as the coherent state given by
√
N0 â†k=0
|G0 ∝ e |0 , (3.46)
and under this wave function, we have

G0 |âk=0 |G0 = G0 |â†k=0 |G0 = N0 . (3.47)
√
Hence, under this wave function, we can replace both the operator â†k=0 and âk=0 as N0 .
√
Since N0 is a large number, in the Bogoliubov theory, we perform a systematical expan-
√
sion of the interaction term in terms of N0 . The leading order is that all four k1 , . . . , k4
√
equal zero and all operators in the interaction term are replaced by N0 , which gives rise
to condensate mean-field energy
E g
= −μn0 + n20 , (3.48)
V 2
where n0 = N0 /V. Thus, it gives μ = gn0 . Next, if one sets three of four momenta
k1 , . . . , k4 to zero, the last momentum also has to be zero. Therefore, it does not lead to
new term.

89 Bogoliubov Theory
In the next order, one can take two of four momenta k1 , . . . , k4 to be zero, and replace
√
the operator â†k=0 or âk=0 as N0 . We shall also use μ = gn0 .4 Thus, the Bogoliubov
Hamiltonian for finite momentum modes becomes
gn0 † †
Ĥ = ( k + gn0 )â†k â k + â k â−k + â−k â k . (3.49)
2
k =0 k =0
Note that the zero-momentum state has been excluded from the momentum summation in
the Bogoliubov Hamiltonian. The second term is the most important process. It appears
that the total number of atoms are not conserved. In fact, it is the total number of atoms
in the finite momentum states that is not conserved, because we have treated the atoms in
the zero-momentum state separately as a condensate. The last term describes that a pair of
bosons can be scattered into finite momentum states from the zero momentum condensate,
or a pair of bosons with finite but opposite momenta can annihilate into the zero momentum
condensate.
It is straightforward to show that by introducing
†
â k = u k α̂ k − v k α̂–k , (3.50)
â†−k = −v k α̂ k + u k α̂–k
†
, (3.51)
and

1 k + gn0
u2k= +1 , (3.52)
2 Ek

1 k + gn0
vk =
2
−1 , (3.53)
2 Ek
the Bogoliubov Hamiltonian can be diganoalized as
1
Ĥ = E k α̂ †k α̂ k + (E k − ( k + gn0 )) , (3.54)
2
k =0 k =0
with

Ek = ( k + gn0 )2 − (gn0 )2 = k ( k + 2gn0 ). (3.55)
Here we should note that Eq. 3.50 and Eq. 3.51 are not a unitary transformation, since a
unitary transformation of â k and â†−k does not keep the boson commutative relation, but
Eq. 3.50 and Eq. 3.51 can.
Here we should also remark that it is a subtle issue of how to relate g to as in such
a perturbative calculation. The relation Eq. 2.31 between g and as derived from the two-
body calculation in Section 2.2 is valid up to all orders. When this relation is applied to
a many-body calculation, it should be used in a way that is consistent with the degree
of approximation in the many-body calculation. Here at the leading order, the order of
approximation in the many-body calculation corresponds to the first order in the ladder
summation of the two-body calculation discussed in Section 2.2, and therefore we simply
take g = U = 4π 2 as /m. Then, the spectrum becomes

E k = k ( k + 2Un0 ). (3.56)
4 This treatment is to enforce the Hugenholtz–Pines relation such that the excitation remains gapless [78].

Here we should introduce the concept of quasi-particle, which is one of the most fun-
damental concepts in quantum many-body physics. In a uniform system, “quasi-particle”
refers to eigenmodes with well-defined energy-momentum dispersion relation. Because of
interactions, a quasi-particle is normally different from the constitution particle of the sys-
tem. For example, in this case the operator for the constitution particle of this system is â k
but the operator for quasi-particle is α̂ k . Because E k is always positive for all k, the ground
state should be a vacuum of all quasi-particles, that is, α̂ k |G = 0. This leads to the ground
state |G that is modified from |G0 by interactions, and
vk † †
√ − u k â k â−k
N0 â†k=0
|G = e e k =0
|0 . (3.57)
Up to this level of the Bogoliubov Hamiltonian, α̂ †k |G
creates an eigenmode with well-
defined dispersion relation E k . In contrast, the original boson operator â†k does not generate
an eigenmode, and |G is also not the vacuum of â k . That is to say, there will always be
finite population on these finite momentum states, which is called the quantum depletion as
we will discuss below. The last term in Eq. 3.54 represents a zero-point energy associated
with the quasi-particle at each momentum, and this vacuum energy will lead to the Lee–
Huang–Yang correction [105] discussed below.
We should also note that there are also residual interactions between quasi-particles,
which will give rise to a finite lifetime τ k for each quasi-particle. Thus, in order for a
quasi-particle to be well defined, we require /τ k E k . The finite lifetime leads to a
broadening of quasi-particle energy, which is given by /τ k . If this broadening is compa-
rable or even larger than the excitation energy itself, the excitation energy will be smeared
out by the broadening. Another way to understand the well-defined quasi-particle is to con-
sider a superposition between a vacuum state and a state with one quasi-particle, which will
undergo Rabi oscillation with period 2π/E k . /τ k E k also means that τ k 2π /E k .
That is to say, for a well-defined quasi-particle, one can observe many periods of Rabi oscil-
lations before it damps out. In the opposite limit when /τ k E k , the oscillation decays
even before finishing one period, and the quasi-particle is no longer well defined. Such
a system without quasi-particle description remains as a theoretical challenge in modern
quantum many-body theory.
The Healing Length. As shown in Figure 3.6, the quasi-particle dispersion Eq. 3.56
exhibits a linear dispersion ck at small k, consistent with what we have derived from the
hydrodynamic theory in Section 3.2. In this regime, u2k ∼ v 2k ∼ Un0 /(2ck), which means
that each quasi-particle mode contains equal contribution of particle and hole. This regime
is called the phonon regime where the excitation is collective. At large momentum the
asymptotic behavior of E k is k + Un0 , which is simply a Hatree–Fock energy shift to the
free-boson dispersion. At the same time u k → 1 and v k → 0, and α̂ k approaches â k . This
regime is therefore called the free-particle regime where the excitation behaves more like a
free-particle. There exists a characteristic momentum regime where the quasi-particle dis-
persion undergoes a crossover from the phonon regime to the free-particle regime. Roughly
speaking,
√ it is determined
√ by 2 k02 /(2m) = ck0 and k0 = 2mc/. This sets a length scale
ξ = 2/k0 = /( 2mc) and ξ is called the healing length. We have 2 /(2mξ 2 ) = Un0 .
In other words, for k 1/ξ , interaction energy dominates over the kinetic energy and that
is why the excitation is collective. And for k 1/ξ the kinetic energy dominates over the

tFigure 3.6 The Bogoliubov excitation spectrum. Eq. 3.56 is shown by the solid line and is compared with the linear dispersion
ck (dashed line) at small momentum and the free-particle dispersion with a constant offset k + Un0 (dotted
line) at large momentum. Here ξ is the healing length, and Eξ = 2 /(mξ 2 ). A color version of this figure can be
found in the resources tab for this book at cambridge.org/zhai.
interaction energy so that the excitation becomes free-particle like. The healing length ξ
is an important quantity that has a few important physical meanings. We will encounter it
again below.
The Lee–Huang–Yang Correction. The summation of the second term in Eq. 3.54 repre-
sents the zero-point energy of each quasi-particle mode. For a single harmonic oscillator,
even without any excitation, there is a zero-point energy that is half of the harmonic fre-
quency. Here it can be viewed as that there is a harmonic oscillator on each k mode, and
the summation of all zero-point energies gives rise to the so-called vacuum energy. Here,
this summation is given by
1 1 −(gn0 )2
(E k − k − gn0 ) = . (3.58)
2 2 E k + k + gn0
k =0 k
In the micro-canonical ensemble, the total energy density up to this order is given by
E gn2 1 −(gn)2
= + , (3.59)
V 2 2V E k + k + gn
k
where the first term comes from the mean-field interaction energy of the zero-momentum
condensate. Here we have used the fact that, for weakly interacting regime, the quantum
depletion is small as we will show below, and we take n0 ≈ n. We can see that the second

term diverges as k m/(2 k2 ). Therefore, we need to use the renormalization condition
to eliminate this divergency. However, since this energy contribution is only a perturbative

correction to the next order of mean-field, we shall only use the renormalization condition
up to the second-order scattering processes. More explicitly, we should use
U2 1
g=U+ (3.60)
V 2 k
k
in the first term of Eq. 3.59 and replace g with U in the second term of Eq. 3.59. This leads
to the total energy density given by

E Un2 (Un)2 1 1
= + − . (3.61)
V 2 2V 2 k E k + k + Un
k
The first term of Eq. 3.61 is the mean-field energy. The summation in the second term of
Eq. 3.61 now converges, which gives rise to the leading-order correction to the mean-field
energy. This result was first obtained by Lee, Huang and Yang using the pseudo-potential
model [105] and is now named as the Lee–Huang–Yang correction, and

ELHY Un2 128
= √ na3s . (3.62)
V 2 15 π
Hence, it is now clear that the LHY correction comes from the zero-point quantum fluc-
tuation of all quasi-particles modes. To observe the LHY correction, the dimensionless
parameter na3s cannot be too small. Using the tunability of the scattering length by Fes-
hbach resonance, one can make the LHY correction visible. In the past years, the LHY
correction has been observed in several ultracold atom experiments [126].
Here let us make a short comment on more general role of such a correction. When
the ground state is unique at the mean-field level, the correction is quantitative. Never-
theless, there are also cases that the correction is qualitative rather than quantitative. It
happens when the mean-field ground states have degeneracy, but such degeneracy is not
protected by an exact symmetry of the full Hamiltonian. This usually occurs in many frus-
trated models. When this happens, the mean-field energy alone cannot determine a unique
ground state, and this fluctuation energy will play a crucial role in selecting out the actual
ground state, because the fluctuation energy usually can lift the degeneracy between dif-
ferent degenerate mean-field states. Usually, the intuition is that the fluctuation energy is
smaller if the sound velocity is smaller. This is because for smaller sound velocity, the exci-
tation energy of the phonon mode increases slower, and a smaller excitation energy also
means a smaller contribution of the zero-point energy. This phenomenon of selecting out
the ground state by fluctuation energy is the order by disorder mechanism. Here the word
“order” refers to that a unique mean-field state is selected out, and the word “disorder”
actually means the fluctuation energy.
The Quantum Depletion. Below we will check two issues in order for the Bogoliubov
theory to be a self-consistent theory. First, we have mentioned that the Bogoliubov theory
is a perturbative approach, which treats the noninteracting Bose condensate as the unper-
turbed wave function. We should first check that the condensate remains as the dominate
component when interactions are turn on. In other words, when interactions are turned on,
there exists population on the finite momentum state. The total number of atoms populated

at finite momentum states is called the quantum depletion. In order for the Bogoliubov
theory to be valid, the quantum depletion should remain small compared with the total
number of atoms.
To compute the quantum depletion Ndp , we note that the ground state |G is a vacuum
of quasi-particle operator α̂ k but not for atom operator â k . Using Eq. 3.50 and 3.51, it can
be straightforwardly calculated that
†
Ndp = G| â k â k |G = v 2k , (3.63)
k =0 k
and the density of quantum depletion ndp = Ndp /V is given by

1 2 1 mc 3 1
ndp = vk = ∝ 3. (3.64)
V 3π 2 ξ
k =0
That is to say, the depleted bosons is of the order one in a volume of ξ 3 , which gives another
physical meaning for the healing length. Thus, the density of quantum depletion is much
smaller than the total density when ξ d, where d is the mean interparticle distance.
In other words, the ratio between the density of quantum depletion to the total density is
given by
ndp 8
= √ (na3s )1/2 , (3.65)
n 3 π
and ndp /n 1 in the weakly interacting regime when na3s 1. Hence, this assumption of
the Bogouliubov theory is valid in the weakly interacting regime.
The Beliaev–Landau Damping. Another thing to check is whether the Bogoliubov quasi-
particles are really well-defined quasi-particles. To this end, we note that the Bogoliubov
Hamiltonian only retains the quadratic terms in terms of the creation and annihilation
operators, and at this order the quasi-particles are well-defined eigenmodes. But all these
quasi-particles acquire a finite lifetime when we go beyond this order. Here, we only
replace one of momentum of k1 , . . . , k4 as zero momentum, which gives
√
g N0 † †
Vint = âk1 âk2 âk3 + â†k1 âk2 âk3 . (3.66)
V
k1 k2 k3
At this order, the Bogoliubov quasi-particle interact with each other.

Rewritting the boson operator into quasi-particle operators, Vint contains four differ-
ent types of terms like (i) α̂k†1 α̂k†2 α̂k3 , (ii) α̂k†1 α̂k2 α̂k3 , (iii) α̂k†1 α̂k†2 α̂k†3 and (iv) α̂k1 α̂k2 α̂k3 .
The first term (i) describes one quasi-particle decays into two, and the second term (ii)
describes two quasi-particles scatter and merge into one quasi-particle. The terms (iii) and
(iv) describe simultaneously creation or annihilation three quasi-particles. The energy con-
served on-shell processes (i) and (ii) give rise to a damping rate and a finite lifetime of the
quasi-particles. The off-shell processes (iii) and (iv) do not satisfy energy conservation, but
they together can produce a second-order perturbation. For instance, three quasi-particles
can first be simultaneously created from the vacuum and then all of them are annihilated
together, which gives a correction to the quasi-particle energy. This effect is called the
vacuum polarization.

Here, it is inspiring to compare the discussion here with the discussion of the hydrogen
atom in atomic physics.
• When we only consider the Coulomb interaction, as we have discussed in Section 1.1,
the energy levels are exact eigenstates. This is at the same level as the Bogoliubov
Hamiltonian.
• When we consider the spontaneous emission, electron can decay from one excited state
to a lower energy state by simultaneously emitting a photon, and such processes give
rise to level broadening for all excited states. This spontaneous emission process is very
much like processes (i) discussed here, which gives rise to a finite lifetime for all quasi-
particles.
• When we further consider the quantum electrodynamics, the vacuum polarization means
that an electron-positron pair can be created from the vacuum and then annihilate each
other. This process can modify the Coulomb energy and gives the famous Lamb shift
for the energy level of the hydrogen atom. The shift of quasi-particle energy due to
the second-order process combining (iii) and (iv) is similar as the effect of the vacuum
polarization.
Here let us focus on the quasi-particle lifetime due to processes (i) and (ii). First, let us
consider the zero-temperature case. We consider that a quasi-particle with momentum k is
excited on top of the vacuum, and then we monitor how long it will decay. In this case,
since there is no other quasi-particle, only the first term contributes to the quasi-particle
lifetime. This is known as the Beliaev damping. For instance, let us consider a term as
†
Mq,k−q,k α̂k−q α̂ †q α̂ k , (3.67)
where M is a matrix element made of u’s and v’s resulting from the expansion, and the
damping rate for quasi-particle with momentum k is given by the Fermi-Gorden rule as
1 2π
= |Mq,k−q, k |2 δ(E k − E q − Ek−q ). (3.68)
τ
q,k−q
Straightforward calculation of this integration will lead to /τ ∝ k5 when the momentum
k is in the phonon regime. Of course, the proportional constant depends on the interaction
strength. This strong suppression at small momentum is partly because of the restriction
of the phase space due to the energy conservation constraint, that is to say, due to the
energy conservation, |q| must be smaller than |k|. Here the most important point is that
k5 is much smaller than linear k for small k, which means that the energy level broaden-
ing due to quasi-particle lifetime is much smaller than the excitation energy itself, that is,
/τ k E k . This satisfies the condition for a quasi-particle to be well defined as discussed
above. Experimentally, the Bogoliubov quasi-particles with a fixed momentum can be first
excited by the two-photon Bragg pulse, as we will discuss below. And then by monitoring
the number of excited atoms, the collision section can be determined, and the experimen-
tal results are shown in Figure 3.7(a). Indeed, one can see a strong suppression at low
momentum.
At finite temperature, the second term (ii) can also contribute to damping. This is known
as the Landau damping. In fact, each term in (ii) can find a conjugate process in (i), thus,

tFigure 3.7 Experiments on the quasi-particle damping rate. (a) At low temperature, damping is dominated by the Beliaev
damping. The suppression of quasi-particle collision at low momentum is measured. Here the quasi-particles are first
excited by the Bragg pulse. Reprinted from Ref. [87]. (b) The damping rate of a low-energy collective mode as a
function of temperature. Reprinted from Ref. [84].
we consider a term like Kk+q,q,k (α̂ †k+q α̂ q α̂ k + h.c.). On one hand, a quasi-particle at
momentum k can decay to another momentum k + q by merging another quasi-particle
at momentum q, and this rate is given by
df k 2π
=− | K k + q, q, k|2 f k f q (1 + f k+q )δ(Ek+q − E q − E k ), (3.69)
dt q
where f k is the boson population at momentum k and 1+f k+q is due to the boson enhance-
ment factor. On the other hand, another quasi-particle with momentum k + q can decay
into two quasi-particles with momentum k and q, and the rate for this inverse process is
given by
df k 2π
= | Kk+q,q, k |2 (1 + f k )(1 + f q )f k+q δ(Ek+q − E q − E k ). (3.70)
dt q
So the net rate is given by

df k 2π
= | Kk+q,q, k |2
dt q
× [(1 + f k )(1 + f q )f k+q − f k f q (1 + f k+q )]δ(Ek+q − E q − E k ).

(3.71)
Note that in the equilibrium distribution, if f k is given by the boson distribution function
denoted by f k0 , this rate vanishes because (1 + f k0 )(1 + f q0 )f k+q
0 = f k0 f q0 (1 + f k+q
0 ) if
Ek+q = E q + E k . This is known as the detailed balance condition. Hence, to study the
quasi-particle lifetime, we consider the situation that f k slightly deviates its equilibrium
value as f k = f k0 + δf k , then we have
dδf k 2π δfk
= | K k+q,q,k |2 (f k+q
0
− f q0 )δ(E k+q − E q − E k ) (3.72)
dt q

and the damping rate

1 2π
=− | Kk+q,q, k |2 (fk+q
0
− f q0 )δ(Ek+q − E q − E k ). (3.73)
τ q
Computation of this integration is quite involved. In the regime with ck < kB T < kB T ∗ , it
gives [144, 177]
4
1 T
∝ ck, (3.74)
τk T∗
where T ∗ is defined as kB T ∗ = Un0 . The Landau damping is less suppressed by the
momentum factor compared with the Beliaev damping, because this damping process is
through scattering with another thermally populated quasi-particle at high energy, and
therefore the phase space restriction from the energy conservation is less important. But the
Landau damping is still suppressed by the temperature factor because it requires thermal
population of other quasi-particles. In any case, again, the quasi-particle is well defined at
low temperature because /τ k is suppressed by a factor of (T/T ∗ )4 compared with E k , and
therefore, the level broadening is smaller than the excitation energy itself. In Figure 3.7(b),
we show the experimental measurement of how the damping rate of a low-lying excitation
depends on temperature, which can be well explained by the Laudau damping mechanism.
The Bragg Spectroscopy. Spectroscopy refers to a probe by a time-periodical perturba-

tion with varying frequency, from which one can investigate how the response of a system
changes as the frequency varies. Here we will discuss the Bragg spectroscopy as an exam-
ple of the spectroscopy measurement in ultracold atomic physics. This probe is a powerful
experimental tool to measure the quasi-particle properties discussed above. For the Bragg
spectroscopy, two lasers with same polarization, for instance, along ẑ, are applied to an
ultracold atom system, and the two lasers have different momenta k1 and k2 and different
frequencies ω1 and ω2 , as shown in Figure 3.8(a). The electric field is then given by
E = E1 eik1 r−iω1 t ẑ + E2 eik2 r−iω2 t ẑ, (3.75)
tFigure 3.8 Experiment of the Bragg spectroscopy. (a) Schematic of the experimental scheme for the Bragg spectroscopy
measurement. Two laser beams with different momenta and different frequencies are applied to the condensate.
Typical absorption images (b, d) before and (c, e) after the Bragg pulse. (d) and (e) are integrated density from (b) and
(c), respectively. A portion of atoms is transferred into momentum with finite q = k1 − k2 . Reprinted from
Ref. [163].

where E1 and E2 are intensities of the two lasers. With the previous discussion of the scalar
light shift in Section 1.3, these laser beams give a scalar light shift as
V(r) ∝ cos(k · r − ωt), (3.76)
where k = k1 − k2 and ω = ω1 − ω2 . In the second quantized form, this perturbation Ĥperp

acts as d3 rV(r)n̂(r), and more explicitly,
Ĥperp ∝ ρ̂ k e−iωt + h.c., (3.77)

where ρ̂ k = q â†q+k â q .
After applying the Bragg pulse, part of the atoms are transferred from zero to finite
momentum state, as shown in Figure 3.8(b). Using the perturbation theory, one can show
that the transition probability is proportional to Sρ k (ω). Here, for a general operator F̂,
SF (ω) is called the dynamic structure factor and is defined as
1 −Em /(kB T)
SF (ω) = e |n|F̂|m |2 δ(ω − (En − Em )), (3.78)
Z m,n
which reduces to

SF (ω) = |n|F̂|0 |2 δ(ω − (En − E0 )) (3.79)
n
at the zero temperature limit. SF (ω) is always positive. It is easy to show that the dynamic
structure factor defined in this way can also be written as temporal correlation function
given by

SF (ω) = dteiωt F̂(t)F̂(0) . (3.80)
For each fixed momentum k, one can vary ω and the resonant frequency of Sρ k (ω) deter-
mines the excitation dispersion E(k), and the width can be related to the quasi-particle
lifetime as /τ k .
A typical measurement of the Bogoliubov spectrum E(k) of Eq. 3.56 by the Bragg spec-
troscopy is shown in Figure 3.9(a). In the zoom-in plot, it is shown that the dispersion is
linear for the small momentum regime. When the momentum becomes larger than 1/ξ , the
spectrum becomes quadratic again and is parallel to 2 k2 /(2m). As shown in Figure 3.6(b),
E(k) − 2 k2 /(2m) indeed saturates to a constant value in large k. This constant value in fact
is the chemical potential of the system.
Interaction Effect on a BEC. In the above two sections, we have used both the hydrody-
namic theory and the Bogoliubov theory to address the interaction effects in a BEC. After
these discussions, we now understand better why we emphasize the macroscopic occupa-
tion as the essential ingredient and the defining property for a BEC. This is because many
other properties are actually different between a non-interacting BEC and an interacting
one. We summarize these differences in Table 3.1. Before ending this section, we should
also summarize that, in an interacting BEC, as the interaction strength increases, following
effects occurs:

Table 3.1 Comparison of physical properties between a noninteracting BEC and a weakly interacting BEC in
three dimensions
Noninteracting BEC Weakly interacting BEC
Chemical potential μ μ=0 μ = Un0 (mean-field level)
Low energy dispersion Quadratic Linear
Superfluidity No Yes
Density in harmonic trap Gaussian Inverted parabola
Expansion dynamics Ballistic Hydrodynamic
Low-energy modes in trap Equal spacing Unequal spacing
Quasi-particle Original bosons Bogoliubov quasi-particle
Quantum depletion at T = 0 Zero ∼ 1/ξ 3
Quasi-particle lifetime Infinite Finite
tFigure 3.9 Measurement of the Bogoliubov dispersion. (a) The Bogoliubov dispersion ω(k) (i.e., E(k) in Eq. 3.56) measured by
the Bragg spectroscopy. The inset shows the zoom-in plot of the small momentum regime. The dashed line shows
ω0 (k) = 2 k2 /(2m). (b) ω(k) − ω0 (k) as a function of k. Reprinted from Ref. [165].
• The mean-field interaction energy and the chemical potential increase.

• The phonon velocity and the superfluid critical velocity increase.
• The zero-point energy and the LHY correction increase.
• The healing length ξ becomes smaller.
• In the momentum space, the regime of collective excitation expands.
• The quantum depletion increases.
• The quasi-particle lifetime becomes shorter.

99 One-Dimensional Bosons
3.4 One-Dimensional Bosons
In this section, we will study interacting bosons in one dimension. In the literatures of
ultracold atomic physics, “one dimension” sometimes means “real one dimension” and
sometimes means “quasi one dimension.” We clarify these two different cases in Box 3.1.
Here we consider the situation of real one dimension. By the Bogoliubov theory, we can
see from Eq. 3.53 that the occupation on single-particle mode with momentum k is always
given by n k = v 2k , and in the small k regime, the momentum distribution behaves as
gn0
n k = v 2k →
. (3.81)
ck
This behavior holds for any dimension as long as that the Bogoliubov theory is valid.
In three dimensions, the integration over the entire momentum converges, which gives a
finite quantum depletion shown in Eq. 3.64. And as we have discussed in Section 3.3, this
depletion is small compared with the total density,
as2 long as the interaction is weak. How-
ever, in one dimension, it is easy to show that vk dk diverges as log when taking the
infrared cutoff to zero. As we discussed in Box 2.2, an infrared divergence means the
low-energy and long-range physics is not treated correctly. In this case, it means that in
one dimension, the low-energy quantum depletion diverges if one assumes a Bose con-
densate. Therefore, the Bogoliubov theory is no longer a self-consistent theory and fails
in one dimension. This difference between one dimension and three dimensions is essen-
tially due to their difference in the single-particle density-of-state. In three dimensions,
√
the low-energy density-of-state vanishes as when the energy approaches zero. In
Box 3.1 One Dimension and Quasi One Dimension
In the ultracold atom system, “one dimension” is achieved by applying a confinement potential in the trans-
verse direction, and this confinement potential is usually a harmonic trap with large trapping frequency
ω⊥ . In the ultracold atom literature, the term “one dimension” sometimes means “real one dimension” and
sometimes means “quasi one dimension.” By “real one dimension,” it means that the harmonic confinement
energy ω⊥ is much larger than the chemical potential μ, i.e., ω⊥ μ, such that the population in
the single-particle excited state of the transverse direction is negligible. In this case, the wave function in
the transverse direction is almost frozen to the ground state Gaussian wave function, as we discussed in the
confinement-induced-resonance in Section 2.5. Such a strong confinement can usually be achieved either by
using a strong two-dimensional optical lattice or by using atomic chip. As we discussed here, Bose condensa-
tion is absence in real one dimension. However, by “quasi one dimension,” for bosons, it is still a condensate
in threedimensions, but the Thomas–Fermi radius along the transverse direction is about the healing length
ξ , i.e., μ/(mω⊥2 ) ∼ ξ . Since 2 /(mξ 2 ) ∼ μ, this is equivalent to ω⊥ ∼ μ. This situation can
usually be satisfied by an anisotropic harmonic trap. In this case, the dynamics of the condensate is mostly
along the longitudinal direction, and therefore the system is called quasi-one-dimensional. The soliton in a
Bose condensate discussed in Section 4.1 belongs to this situation.

√
one dimension, the low-energy density-of-state diverges as 1/ when approaches zero.
Because of the large low-energy density-of-state in one dimension, Bose condensate is
strongly depleted. Note that this argument is independent of the strength of interactions.
That is to say, even for very weak interactions, the condensate will be destroyed inevitably.
Hence, we need to introduce an alternative theory to capture the interacting effects in one
dimension.
Bethe–Ansatz Solution. In Section 2.5 we have discussed that, unlike the three-
dimensional case, the δ-function potential is well defined in one dimension. There, with
a two-body problem, we have shown how to determine the interaction strength of the
δ function potential from the three-dimensional scattering length. Here we will discuss
that one-dimensional Bose gas interacting with the δ function potential is actually exactly
solvable [111].
The reason that such one-dimensional models are exactly solvable is in fact due to a
special feature of one dimension. Considering two atoms with position x1 and x2 , suppose
in the regime x1 < x2 , their momenta are respectively k1 and k2 , and the wave function
is eik1 x1 +ik2 x2 . In the regime when x1 > x2 , suppose their momenta are k1 and k2 , the
momentum conservation gives
k1 + k2 = k1 + k2 . (3.82)
And outside the interaction range, the energy is purely kinetic energy and therefore, for the
same eigenstate, it requires
k12 + k22 = k12 + k22 . (3.83)
In one dimension, it is easy to show that Eq. 3.82 and Eq. 3.83 together determine that
either (k1 , k2 ) = (k1 , k2 ) or (k1 , k2 ) = (k2 , k1 ). That is to say, in the regime x2 > x1 , the
momenta of two atoms are either the same as, or a permutation of two momenta as in the
regime x2 < x1 . This property is unique for one dimension and does not hold in higher
dimension.
The property can be straightforwardly generalized to a multiparticle situation. Consid-
ering N particles whose coordinates are labeled by x1 , . . . , xN , suppose that in the region
x1 < x2 < · · · < xN , the momenta of each particle are k1 , . . . , kN and the wave function
is given by ei i ki xi . In the same region and other spatial regions, if the many-body wave

function also contains other components such as ei i ki xi , it can be shown that {k1 , . . . , kN }
can only be a permutation of {k1 , . . . , kN }. Therefore, for a given set of {k1 , . . . , kN }, all
the wave functions ei i ki xi with {k1 , . . . , kN } being a permutation of {k1 , . . . , kN } form a
closed Hilbert space, whose dimensionality is always finite. In general, the many-body
wave function is a superposition of all of them, and the coefficients in the superposition
are determined by matching the boundary conditions when any two xi = xj . This will
eventually reduce the Schrödinger equation to a set of algebraic equations. In this sense,
this model is exactly solvable. The same method can be used to solve interacting spinful
fermions in one dimension [184].
Note that in the weakly interacting limit, the interaction energy is estimated by g1d n1d ,
where g1d is the one-dimensional interaction strength discussed in Section 2.5 and n1d is

101 One-Dimensional Bosons
the one-dimensional density. The typical kinetic energy is given by 2 n21d /(2m). Hence, the
ratio of the interaction energy to the kinetic energy is given by
g1d n1d 2mg1d
γ = = . (3.84)
n21d /(2m)
2 2 n1d
It is interesting to note that the smaller the density, the larger the γ . The system is in a
strongly interacting regime when γ 1. This is in contrast to the three-dimensional case
where the high density regime is strongly interacting. In practices, there are several differ-
ent ways to achieve the γ 1 regime. One can tune g1d to be very large by utilizing the
confinement-induced-resonance as discussed in Section 2.5. And one can also achieve the
γ 1 regime by reducing the density [90]. Moreover, an alternative tool is to increase the
mass. The single-particle effective mass is defined through the single-particle dispersion,
which can be tuned by applying an extra optical lattice along the longitudinal direction, as
we will show in Section 7.1 [134].
In the limit γ 1, the ground state should first try to minimize the interaction energy. To
maximumly minimize the interaction energy of a δ-function repulsive potential, it requires
that the wave function vanishes when the coordinates of any two particles coincide. If we
write an antisymmetric wave function of two particles as
= eik1 x1 +ik2 x2 − eik1 x2 +ik2 x1 , (3.85)
this wave function satisfies the requirement = 0 when x1 = x2 . And it also requires
k1 = k2 for otherwise the wave function vanishes everywhere. However, this wave function
does not satisfy the Bose statistics. Hence, we should modify the wave function as
x

= |eik1 x1 +ik2 x2 − eik1 x2 +ik2 x1 | = sin (k1 − k2 ) , (3.86)
2
where x = x1 − x2 . This consideration can be generalized to a many-body situation and the
many-body wave function can be written as
⎡ ik1 x1 ⎤
e eik1 x2 . . . eik1 xN
ik x
⎢ e 2 1 eik2 x2 . . . eik2 xN ⎥
= Abs ⎢
⎣Det . . .
⎥ , (3.87)
... ... . . . ⎦
eikN x1 eikN x2 . . . eikN xN
where “Det” denotes Slater Determinant and “Abs” denotes taking an absolute value. The
Slater determinant is the many-body wave function of free fermions. However, because of
taking the absolute value of the Slater determinant, the momentum distribution is different
from that of free fermions and it does not display a sharp Fermi surface structure. Neverthe-
less, the momentum distribution is much more flat in the momentum space compared with
that of the Bose condensation case. In fact, it can be shown that there is no macroscopic
occupation in any modes, and the system is not a Bose condensate.
Since the interaction energy vanishes for this wave function, the total energy is given by
the kinetic energy alone, and

N
2 k 2 i
E= , (3.88)
2m
i=1

Box 3.2 Scale Invariant Quantum Gases
We consider the scaling transformation that the coordinates of all particles scale as ri → λri . A quantum
many-body system is called scale invariant if the total energy E scales as E → E /λ2 under the scal-
ing transformation. Obviously noninteracting Bose and Fermi gases are scale invariant, since the Hamiltonian

only contains the kinetic energy T = i 2 ∇i2 /(2m) that is scaled by 1/λ2 under the scaling transfor-
mation. For an interacting gas, if the total energy is proportional to the Fermi energy EF only, such a system
is also scale invariant because EF is always scaled by 1/λ2 under the scaling transformation. Such exam-
ples include the one-dimensional Tonks–Girardeau gas discussed here and the three-dimensional unitary
Fermi gas discussed in Chapter 6. Both are strongly interacting systems. There are also examples of weakly
interacting systems that are scale invariant, for instance, a weakly interacting two-dimensional Bose gas.
At the mean-field level, the interaction energy of weakly interacting bosons is always proportional to the
two-dimensional density n2d . In two dimensions, the density n2d also scales as 1/λ2 under the scaling trans-
formation, and therefore, both T and V scale as 1/λ2 . Hence, the total energy also scales with 1/λ2 .
However, in this system, the scale invariance will be broken when quantum effects beyond mean field are
included, which is known as an anomaly as in high-energy physics.
(a) γavg (b)

0 1 2 3 4 5
60 1600
50 1400
1200
40
T1D (nK)
1000
30 800
600
20 1
400
n(x)
10 200 0
0 –20 0
2x/λ
20
0
0 20 40 60 80 10–1 100
U0 (Erec) p (hk)
t
Figure 3.10 Experimental measurements of Tonks gases. (a) T1D measures the total energy of the system and is plotted as a
function of γ . The horizontal dashed line indicates the fermionized energy in the Tonks–Girardeau limit. The solid
line is theory based on the Bethe–Ansatz solution. The deviation is attributed to the system not being in a pure
one-dimensional regime, as indicated by the squares. (b) Momentum distribution with γ ≈ 200. The solid line is
theory based on the fermionized wave function in the Tonks–Girardeau limit. (a) is reprinted from Ref. [90], and (b) is
reprinted from Ref. [134]. A color version of this figure can be found in the resources tab for this book at
cambridge.org/zhai.
where ki = 2πn/L and L is the system size. Since all ki should be different from each
other, to minimize the total energy, ki should be as small as possible. Thus, n should take
all integers from −N/2 to N/2. This is reminiscent of filling the Fermi sea for a free Fermi
gas. It also says that when g1d → ∞, the ground state energy of such a strongly interacting

103 Phase Coherence and Fragmentation
Bose gas approaches the total kinetic energy of a free Fermi gas with same density. In
this sense, we state that the system is fermionized, and such system is called the Tonks–
Girardeau gas [174, 60]. This Tonks–Girardeau gas is one of the examples that are scale
invariant. There are other systems in ultracold atomic gases that are scale invariant, and we
summarize them in Box 3.2.
The first experimental evidences of the Tonks–Girardeau gas were reported in Ref. [90]
and Ref. [134], where the γ 1 regime is achieved either by reducing the density or
by increasing the effective mass. The main experimental observations are presented in
Figure 3.10. In Ref. [90], they observe that the energy of this one-dimensional system satu-
rates to that of a free Fermi gas with the increasing of γ . In Ref. [134], they observe that the
momentum distribution becomes more and more flat when γ increases. Both experimental
measurements agree with the theoretical calculations using the Bethe–Ansatz method and
agree with a fermionized wave function Eq. 3.87 in the large-γ limit.
3.5 Phase Coherence and Fragmentation
In previous sections, we mainly focus on the interaction effects on energy. In this section,
we focus on the interaction effects on phase coherence. Here “phase” actually means the
relative phase between different positions or different subsystems of the Bose conden-
sate. To this end, we consider a simple “toy model” of interacting bosons with two spatial
modes, and therefore, the “phase” simply means the relative phase between these two
spatial modes.
Double-Well Model. Let us consider interacting bosons in a double-well potential, as
shown in Figure 3.11. At each well, we only consider the lowest energy state, whose spatial
wave functions are denoted by ψ1 (r) and ψ2 (r), respectively. When the barrier is suffi-
ciently high and the tunneling is much weaker compared with the level separation in each
t
Figure 3.11 Illustration of the double-well model. The solid line is the double-well potential. Two dotted lines represent wave
function ψ1 (r) and ψ2 (r) of the two lowest energy states in each well. The horizontal black dashed lines indicate
the energy levels in absence of tunneling, which is much larger than the tunneling energy and the interaction energy
considered here. A color version of this figure can be found in the resources tab for this book at cambridge.org/zhai.

well, tunneling can hardly couple these two low-lying states to other excited levels. More-
over, when we include interactions, we assume that the interaction energy is also much
weaker compared with the level separations. Hence, we only need to consider the lowest
energy mode in each well. Although in practical situations, these conditions are hard to
be fully satisfied, this model can still serve as a sufficiently simple toy model to illustrate
many key physical concepts.
Because of tunneling, the single-particle ground state ψg (r) is given by
1
ψg (r) = √ (ψ1 (r) + ψ2 (r)) , (3.89)
2
and the first excited state e (r) is given by
1
ψe (r) = √ (ψ1 (r) − ψ2 ( r)) . (3.90)
2
For a symmetric double-well, the Hamiltonian possesses the parity symmetry of x ↔ −x.
Hence, both ψg and ψe respect this symmetry. And the symmetric superposition ψg is
the ground state, following from the Feynman’s argument that the ground state of a real
Hamiltonian should be nodeless.
We can expand the field operator as
ˆ
(r) = ψ1 (r)â1 + ψ2 (r)â2 , (3.91)
where â1 and â2 are boson annihilation operators in each well, and the second-quantized
Hamiltonian for this system can be given by
U !
Ĥ = −J â†1 â2 + â†2 â1 + n̂1 (n̂1 − 1) + n̂2 (n̂2 − 1) , (3.92)
2
where J denotes the single-particle tunneling rate between two wells, and U is the
interaction strength between atoms in the same well.
There are two ways to solve this model. First, we can take the mean-field approximation
as we discussed in Section 3.1. Assuming that all bosons are condensed in the same single-
particle state, we can write down the mean-field wave function as
1 α α N
|MF = √ cos e−iθ/2 â†1 + sin eiθ/2 â†2 |0 , (3.93)
N! 2 2
where the state is taken as a general superposition of two modes. It is straightforward to
show that for this wave function,
α α α α
â†1 â1 = N cos2 , â†2 â2 = N sin2 , â†1 â2 = N sin cos eiθ , (3.94)
2 2 2 2
and the energy of this mean-field state is given by
UN 2
E = MF |Ĥ|MF = −JN sin α cos θ + cos2 α. (3.95)
4
Since J > 0, it is easy to see that the kinetic energy always favors θ = 0. And for repulsive
interaction U > 0, both kinetic and interaction term favor α = π/2. Thus, as long as

U > 0, regardless of the value of U/J, the ground state determined by the mean-field
ansatz is always
† N
1 â1 + â†2
|MF = √ √ |0 . (3.96)
N! 2
This state means that all bosons occupy the same single-particle state, which is exactly the
single-particle ground state Eq. 3.89.
Let us consider the situation that the interaction is attractive and U < 0.5 When 2J >
|U|N, the minimum is still located at α = π/2. When 2J < |U|N, minimizing E with
respect to α will yield two degenerate minima at α0 = arcsin(2J/(|U|N)) and π − α0 ,
respectively. In the limit U → −∞, α → 0, and these two degenerate solutions approach
â†N †N
1 |0 and â2 |0 , respectively. It is easy to see that these two mean-field solutions break
the parity symmetry between left and right wells.
Another way to solve this model is to write down the most general form of the wave
function under the Fock bases as
"
N â†N 1 †N2
N 1 â2
| = l + l, − l = l √ |0 , (3.97)
2 2 N1 ! N2 !
l l
where N1 = N/2 + l and N2 = N/2 − l. Here since we take the two-mode approximation,
the Hilbert space dimension increases linearly with the number of atoms, and we can easily
solve this model up to a few thousands of atoms. Later we will compare the solution from
these two approaches.
Josephson Effects. Let us first consider the U > 0 regime. According to the mean-field
results, we have α = π/2 and θ = 0, thus N̂ = N̂1 − N̂2 = 0. That is to say, there are
equal number of atoms residing in each well and the relative phase between them is also
zero. Now we consider the situation when either α or N̂ deviates from this equilibrium
situation and study their dynamics.
First, we have
d N̂ 2J
= − Î , (3.98)
dt
and in this two-mode case, the current operator Î is defined as −i(â†1 â2 − â†2 â1 ). With
Eq. 3.94, we have −Î = −N sin α sin θ. We assume 2JN sin α is nearly a constant
throughout the dynamics, denoted by EJ following the convention in literatures, and we
will see that it is indeed the case in the two regimes that we will discuss below. Then, we
have
d N̂ EJ
= − sin θ. (3.99)
dt
In the weak tunneling regime, the phase dynamics of each mode is mainly governed by
the local interaction energy. Therefore, on one site the phase evolves as e−iUN1 (N1 −1)t/2 and
on the other site the phase evolves as e−iUN2 (N2 −1)t/2 . Thus the time evolution of phase
5 As a toy model with two discrete modes, let us disregard the issue that the condensate with attractive
interactions is unstable because of the negative compressibility.

difference is proportional to N. Also following the convention in literatures, we denote

the proportional coefficient as Ec and therefore, we have
dθ Ec
= N̂ . (3.100)
dt
From Eq. 3.99 and 3.100, one can already see that θ and N are two quantities that form
two coupled equations. Eq. 3.99 and 3.100 lead to
d2 θ Ec EJ
= − 2 sin θ, (3.101)
dt2
which coincides with the classical equation-of-motion of a simple pendulum. It is known
that there are two solutions for a classical pendulum, depending on the initial velocity.
Here these two solutions are respectively known as the Josephson oscillation and the
Self-trapping. We shall remark that both two phenomena are consequences of the phase
coherence and are absent in a normal gas.
• If the initial N̂ 0 and θ are sufficiently small, it gives a simple harmonic oscillation
with frequency
√
Ec EJ
ωJ = . (3.102)

In this case, both the atom number imbalance, current and phase oscillate in time with
this frequency. This is known as the Josephson oscillation. Here we should remark that,
even initial N̂ 0 = 0, as long as θ is nonzero, a finite current will be induced. This is
counterintuitive and is strongly in contrast to a normal gas, where a macroscopic current
is not possible if there is no bias voltage or particle imbalance applied between two sides.
• If initial N̂ 0 is much larger than EJ /Ec , from Eq. 3.100, one finds that θ increases
linearly in time as tEc N 0 /. And if θ changes sufficiently fast, from Eq. 3.99, we
can see that at the zeroth order the current averages out and N̂ remains as a constant
N̂ 0 . This is also a counterintuitive result and also cannot happen in a normal gas. In
a normal gas, when a strong density imbalance is applied between two sides, the system
should gradually relax to the balanced situation. In contrast, here when the initial state
strongly deviates from the ground state with balanced population, the system retains this
far-from-equilibrium situation instead of relaxing back to the ground state. This regime
is known as the self-trapping.
Therefore, as initial N̂ 0 increases, the dynamics changes from the Josephson oscilla-
tion regime to the self-trapping regime. This has been demonstrated experimentally with a
BEC in double-well potential [4]. Figure 3.12(a) shows the case with small atom number
imbalance, where both the relative atom number and the relative phase oscillate around
zero. Figure 3.12(b) shows the case with large atom number imbalance, where one can see
that θ increases monotonically in time and the relative atom number is nearly a constant
around its initial value.
Fragmentation and Phase Fluctuation. Now we return to another method to solve the
Hamiltonian with the general wave function Eq. 3.97. First of all, in the noninteracting
limit U = 0, it is easy to see that the ground state is given by

t
Figure 3.12 Experimental observation of the Josephson oscillation and the self-trapping. Each figure shows how the atom numbers in two wells, the relative population
N̂ /N, and the relative phase θ change as a function of time t. (a) Josephson oscillation. (b) Self-trapping. Reprinted from Ref. [4]. A color version of this
figure can be found in the resources tab for this book at cambridge.org/zhai.
N
1 â†1 + â†2
| = √ √ |0 , (3.103)
N! 2
and this wave function is also the mean-field state for all positive U/J. We expand the wave
function Eq. 3.103 in terms of Eq. 3.97, and with the help of the Stirling’s approximation
log N! = N log N − N, we can obtain

e−2l /N
2
N!
l = ! ! . (3.104)
2N N2 + l ! N2 − l ! (π N/4)1/4
That is to say, l obeys a Gaussian distribution with the width σ 2 N being proportional to
N, and it is straightforward to show that ( N̂)2 ∼ N. This is actually a natural result of
the central limit theorem.
Now when U is finite, we consider the Schrödinger equation [71]
El = −Jl+1 l+1 − Jl l−1 + Ul2 l , (3.105)

√
where Jl = J (N/2 + l)(N/2 − l + 1). For sufficiently large and positive U, the wave
function is localized around l = 0. For l N, approximately we have Jl ≈ JN/2 as a
constant, and Eq. 3.105 is reduced to a discrete version of the single-particle Schrödinger
equation in a harmonic trap. The ground state solutions gives

1 −l2 /σ 2 N JN
l = e , σ N∝
2
. (3.106)
(π σ N /2)
2 1/4 U
Thus, as U increases, the width of the Gaussian distribution σ 2 N varies from being propor-
√
tional to N to being proportional to JN/U. The particle number fluctuation ( N̂)2 ∼
σ 2 N is strongly suppressed as U increases. This is illustrated in Figure 3.13. In the limit
JN/U 1, this state approaches a Fock state
t
Figure 3.13 Schematic of atom number fluctuations. Here we plot l as a function of l = (N1 − N2 )/2 for different regimes
of U/J. Here U/J < 0 and |U| J for (a), U > 0 but U J for (b), and U > 0 and U J for (c). A
color version of this figure can be found in the resources tab for this book at cambridge.org/zhai.

†N/2 †N/2
â1 â2
| = |0 . (3.107)
(N/2)!
This state becomes an eigenstate of N̂.
For noninteracting quantum state Eq. 3.103, the single-particle density matrix can be
easily computed as

â†1 â1 â†1 â2 N 1 1
ρ= = . (3.108)
â†2 â1 â†2 â2 2 1 1
This density matrix can be diagonalized as

1 0
ρ=N . (3.109)
0 0
That is to say, only the population on one of the modes is of the order N and the population
on the other mode is zero. According to the definition given in Section 3.1, the system is a
simple BEC. For strong repulsive interaction discussed above, because

(N/2 − l)(N/2 + l + 1)l+1 l ≈ (N/2)e−1/(2σ N ) ,
2
â†1 â2 = (3.110)
l
the single-particle density matrix is given by

e−1/(2σ
2
N 1 N)
ρ= −1/(2σ 2 ) . (3.111)
2 e N 1
Two eigenvalues are λ = (N/2)(1 ± e−1/(2σ N ), and both eigenvalues are of the order of N.
2
In the limit U → +∞ and σ N → 0, the quantum state becomes the Fock state Eq. 3.113,
and it is easy to see that the density matrix becomes

N 1 0
ρ= . (3.112)
2 0 1
According to the definition given in Section 3.1, the system is a fragmented condensate.
Note that the Fock state can also be viewed as a single condensate in terms of two-particle
density matrix, because this state can be written as
1 † † N/2
| = â â |0 , (3.113)
(N/2)! 1 2
and â†1 â†2 can be viewed as a boson pair operator.

To better understand what happens from the small U regime to the large U regime, we
revisit the mean-field description. We can expand the mean-field energy Eq. 3.95 around
its minimum α = π/2 and θ = 0, and it yields
JN 2 UN 2
E= θ + ( α)2 , (3.114)
2 4
where α = α − π/2. If we consider α and θ as a pair of conjugate variables like x̂
and p̂, Eq. 3.114 can be considered as a harmonic oscillator, and the width of the phase

√ √
fluctuation is given by σθ2 ∝ UN/J. When UN/J 2π, we can consider that the
phase can freely fluctuate between zero and 2π. Using the identity that
π †N/2 †N/2
1 â â2
dθ √ (â†1 eiθ/2 + â†2 e−iθ/2 )N |0 ∝ 1 |0 , (3.115)
−π N
2 N! (N/2)!
we can state that a strong enough phase fluctuation renders the system from a single con-
densate to a fragmented condensate with fixed relative particle numbers. We shall revisit
similar physics in Section 4.3 when we discuss spinor condensate.
It is also interesting to note that σ 2 N σθ2 ∼ N. It is actually better to state that the relative
atom number difference N̂ and the relative phase θ can be viewed as a pair of conjugate
variables. In the weakly interacting regime, there is a well-defined relative phase, meaning
that the phase fluctuation is strongly suppressed, therefore, the relative atom number fluc-
tuation is as large as ( N̂)2 ∼ N. As U increases, the relative atom number fluctuation is
gradually suppressed and the phase fluctuation increases. Eventually, in the strongly repul-
sive interaction limit, the relative atom number fluctuation is strongly suppressed such that
the system becomes a Fock state, but the phase fluctuation becomes very large such that
the phase coherence between two sides is completely lost.
Quantum Measurement. When talking about interference in a quantum system, the

double-slit experiment always the first one that occurs to one’s mind. In order to see an
interference pattern in the usual double-slit experiment, there are essentially two condi-
tions. First, the waves passing through two slits have to have well-defined relative phase.
Second, the particles have to be identical particles when they are detected. In other words,
the detectors do not know which slit the particle comes from. If the particle are labeled by
the path, the interference will disappear.
Decades ago, Anderson has asked the question that, if two superfluids have never met
before, whether they can interfere when they meet. This question can be studied in ultracold
atom experiments. The answer to this question has to involve the measurement process.
Let us prepare a Bose condensate in a double-well potential. If the barrier is shallow, the
system initially is a simple condensate with well-defined relative phase between two sides.
Then, we turn off the potential and let the system expand. When atoms from two sides met
and overlap, the density distribution will show interference pattern, similar as in the usual
double-slit interference experiment. However, if we start with sufficiently high barrier, the
system initially is a Fock state like Eq. 3.113. And as shown by Eq. 3.115, there is no well-
defined relative phase between two sides. The question is whether we can still observe the
interference pattern when the double-well potential is turned off and atoms overlap.
First of all, we consider an ensemble measurement, that is to say, we repeat the same
measurements under identical conditions and take an average over them. In this case, the
density is given by
n(r) = ˆ
ˆ † (r)(r) = (1∗ (r)â†1 + 2∗ (r)â†2 )(1 ( r)â1 + 2 (r)â2 ) , (3.116)
where 1 (r) and 2 (r) label the wave function evolving from the left and the right sides,
respectively. Note that â†1 â2 = â†2 â1 = 0 for the Fock state Eq. 3.113, Eq. 3.116
becomes

t
Figure 3.14 Experimental observation of interference pattern. A BEC is separated into two by a high barrier created by laser. When
these two initially separated BECs are released from the trap, and when they overlap after expansion, an interference
pattern can be observed in each run of measurements. Reprinted from Ref. [7]. A color version of this figure can be
found in the resources tab for this book at cambridge.org/zhai.
n(r) = â†1 â1 |1 (r)|2 + â†2 â2 |2 (r)|2 . (3.117)
It is obvious that no interference exists after the ensemble average.

However, if one actually performs such an experiment, one can observe the interference
pattern in each run of measurement, as one can see in Figure 3.14. As we will explain
below, this is a quantum measurement effect. Quantum measurement can be viewed as a
projection. As we have shown in Eq. 3.115, a Fock state can be viewed as a superposition of
single condensate with all possible relative phases. Thus, each measurement projects this
state into one of the states in the superposition with fixed relative phase. To illustrate how
this quantum measurement projects the Fock state into a state with fixed relative phase, we
present following two-step arguments [28].
• First, it is important to emphasize again that the atoms are identical particles, and when
one atom is detected, one cannot distinguish which side the atom comes from. When
the first atom is detected, without loss of generality, we can assume that this atom has
equal probability of coming from the left side or coming from the right side, with a
relative phase φ. The many-body wave function after the first atom being detected |1
is given by
"
â1 e−iφ/2 + â2 eiφ/2 N N
|1 = √ 2, 2
2
√ " "
N −iφ/2 N N
iφ/2 N N
= e 2 − 1, + e ,
2 2 − 1 . (3.118)
2 2

Now considering that the second atom is detected, this atom is also assumed to be in
a superposition state with a relative phase ζ , then, the many-body wave function |2
after the second atom being detected is given by
â1 e−iζ /2 + â2 eiζ /2
|2 = √ |1 (3.119)
2
" "
N −i(φ+ζ ) N
N
i(φ+ζ ) N N
∝ −1 e 2 − 2, 2 + e 2, 2 −2
2
"
N N N
+ 2 cos(φ − ζ ) − 1, − 1 , (3.120)
2 2 2
and
2 |2 = (N − 2) + 2N cos2 (φ − ζ ). (3.121)
This shows that 2 |2 has the largest amplitude when φ = ζ . That is to say, the
possibility is the largest when successively two atoms with the same relative phase are
detected.
• Second, if successively k-atoms are detected with same relative phase, we have
−iφ/2 k "
â1 e + â2 eiφ/2 N N
√ 2 2,
2
−iφ/2 k π † N
â1 e + â2 eiφ/2 â1 eiθ/2 + â†2 e−iθ/2
∝ √ dθ √ |0 (3.122)
2 −π 2
π †
â eiθ/2 + â† e−iθ/2 N−k
∝ dθ(cos(θ/2 − φ/2))k 1
√ 2 |0 . (3.123)
−π 2
It is easy to see that (cos(θ/2 − φ/2))k quickly approaches the delta function δ(θ − φ)
as k increases, because for any number x < 1, xk quickly approaches zero as k increases.
Therefore, such a measurement picks up a coherent state with fixed relative phase, that is,
−iφ/2 k " † N−k
â1 e + â2 eiφ/2 N N â1 eiφ/2 + â†2 e−iφ/2
√
2 2, 2 → √
2
|0 . (3.124)
From above discussion, we see that quantum measurement can project a Fock state
without phase coherence to a simple condense state with a well-defined relative phase.
However, this relative phase is randomly picked up by the measurement process, and
therefore, for different runs of experiment, the relative phases are different and one can
see different interference pattern. Hence, after averaging over many measurements, the
interference pattern is smeared out, which recovers the result of ensemble measurement
discussed above.
The Hanbury–Brown–Twiss Effect. For a Fock state, above discussions have shown that
although the ensemble measurement of density does not show interference pattern, the
identical boson nature of atoms enables that each individual measurement does show inter-
ference. Here we can further show that, also due to the nature of identical bosons, the

density-density correlation can also show an interference pattern even after the ensem-
ble average. This is known as the Hanbury–Brown–Twiss effect. We summarize whether
interference pattern can appear or not under different situations in the Table 3.2.
Following the same spirit of Eq. 3.116 and considering the density-density correlation
function, we have
ˆ
ˆ † (r)(r) ˆ † (r )(r
ˆ )
= n(r)n(r ) + â†1 â1 â†2 â2 (1∗ (r)2∗ (r )2 (r)1 (r ) + h.c.), (3.125)
where n(r) and n(r ) are given by Eq. 3.117. Here we also take â†1 â2 = â†2 â1 = 0
because of the absence of phase coherence in the initial state. The interference appears
in the second term of Eq. 3.125.
This effect has been first observed in ultracold atomic system by using the Mott insulator
phase of bosons in optical lattices [56]. We will discuss the Mott insulator in Section 8.1.
The basic idea is to use a deep optical lattice potential to localize atoms inside each well,
such that there is no phase coherence between wells. Then, after turning off the lattice, the
wave functions of atoms expand from different sites, and then they will overlap with each
other, as shown in Figure 3.15(a). As shown in Figures 3.15(b) and (d), when measuring the
density after the time-of-flight and averaging over a number of measurements, one cannot
see any signal of interference pattern in the density profile. However, when one analyzes
t
Figure 3.15 Experimental observation of the Hanbury–Brown–Twiss effect. (a) Illustration of the basic idea of this experiment.
Atoms are initially prepared in a deep optical lattice, and density distributions are measured after the time of flight.
(b) and (d) show the density after ensemble average, and (c) and (e) show the density-density correlation after the
ensemble average. (d) and (e) are integrated results of (b) and (c), respectively. Reprinted from Ref. [56]. A color
version of this figure can be found in the resources tab for this book at cambridge.org/zhai.

Table 3.2 On interference pattern of density and density-density correlation measurements

Single measurement Ensemble measurement
n(r) Yes No
n(r)n(r ) Yes Yes
Note: “Yes” and “No” denote the presence and the absence of interference pattern,
respectively.
Box 3.3 Symmetry Breaking
Let us first introduce two concepts of the symmetry group of the wave function and the symmetry group of
the Hamiltonian. The symmetry group of the Hamiltonian is the group of operations that keep the Hamilto-
nian invariant, and the symmetry group of the wave function is the group of operations that keep the wave
function invariant. When we say that a state breaks symmetry, it means that the symmetry group of the wave
function is a subgroup of the symmetry group of the Hamiltonian. A direct consequence of symmetry breaking
is degeneracy. For the single-particle case, we have emphasized that the ground state has no degeneracy, and
therefore, it cannot break symmetry. However, symmetry breaking is a common phenomenon in the ground
state of a many-body system, the reason for which will be explained in this section. In the Landau theory of
phase transition, different phases are characterized by different symmetry groups of the wave function. In
Section 4.5, we will discuss an example of a Bose system with various symmetry-breaking phases, and we
shall discuss how symmetry breaking helps us determine the properties of phase transitions. In Section 8.2,
we will discuss an example of a Fermi system with various symmetry-breaking phases.
the density-density correlation from these images, the interference peaks are clearly visible,
as shown in Figures 3.15(c) and (e).
The Schrödinger Cat and Spontaneous Symmetry Breaking. Now we turn into the
discussion of attractive interaction with U < 0. As mentioned above, when |U|N >
2J, the mean-field energy has two degenerate minima respectively located at α0 =
arcsin(2J/(|U|N)) and π − α0 . In the limit U → −∞, these two solutions approach â†N 1 |0
and â†N
2 |0 .
In Box 3.3, we have introduced the concept of the symmetry group of the wave function
and the symmetry group of the Hamiltonian, and the important concept of symmetry break-
ing. Here the Hamiltonian has a Z2 symmetry of exchanging â1 and â2 . The mean-field
states for very negative U do not respect this Z2 symmetry, and consequently, the mean-
field ground states are two-fold degenerate. It is noted in Box 3.3 that symmetry breaking is
a common phenomenon in nature. However, we also note that, for a real Hamiltonian such
as this case, the ground state should be unique and cannot has degeneracy. How should we
resolve this paradox ?
Indeed, if we solve the Hamiltonian Eq. 3.105 with the general wave function Eq. 3.97,
the system should have a unique ground state, and the corresponding l respects the Z2
symmetry and exhibits a double peak distribution

l ∝ e−(l−l0 /2)
2 /(2σ 2 )
+ e−(l+l0 /2)
2 /(2σ 2 )
, (3.126)
as shown in Figure 3.13. In the limit U → −∞, the ground state wave function, denoted
by |+ , is a superposition as
1
+ = √ (|N, 0 + |0, N ) , (3.127)
2
which is also known as the NOON state. These two states |N, 0 and |0, N in the super-
position are macroscopically distinct, because one corresponds to all atoms in the left
well and the other corresponds to all atoms in the right well. Such a superposition of
macroscopically distinct states is well known as the Schrödinger Cat state.
The Schrödinger Cat state has the same one-body density matrix as the Fock state
discussed above, that is,

N 1 0
ρ= , (3.128)
2 0 1
therefore it is also a fragmented state. However, in contrast to the Fock state, its particle
number fluctuation is ( N̂)2 ∝ N 2 . It also differs from the Fock state in terms of two-
body density matrix. The Fock state can be viewed as a pair condensate in terms of two-
body density matrix, but the Schrödinger Cat state cannot.
However, why is such a Schrödinger Cat state very rare in nature ? In fact, in this model,
there exists the first excited state |− whose energy is very close to the ground state, and
the wave function is given by
1
|− = √ (|N, 0 − |0, N ) . (3.129)
2
One can show that the energy difference between |+ and |− exponentially decreases
as the atom number N increases. Thus, even for N being of the order of ten, this energy
separation is already extremely small. This is because, in order to couple |N, 0 state to
|0, N state, there have to go through N-steps of single-particle tunneling as
|N, 0 → |N − 1, 1 → |N − 2, 2 → · · · → |1, N − 1 → |0, N , (3.130)
and during this process, the interaction energy keeps increasing. Such a process is called
macroscopic quantum tunneling. Therefore, the tunneling rate is strongly suppressed as N
increases, and consequently, the energy splitting between |+ and |− is strongly sup-
pressed. In fact, this is not a specific feature of this particular model. It applies universally
to all Schrödinger Cat states. We recall that the Schrödinger Cat state refers to superpo-
sition of macroscopically distinct states, and if two states are macroscopically distinct, it
means that there are macroscopic number of degrees of freedom that are different between
these two states. In nature, nearly all physical systems possess locality, which means each
term can only change few number of degrees of freedom. Hence, in order to couple two
macroscopically distinct states, one needs macroscopic number steps. Therefore, the cou-
pling coefficient becomes extremely small. For instance, let us consider two low-lying
ferromagnetic states whose spins point to different directions. In order to couple them, one
needs to flip each spin from one direction to another direction, and the intermediate states

are not ferromagnetic, which have higher energies. Hence, such a coupling will be strongly
suppressed as the number of spins increases.
Because of this extremely weak coupling between macroscopically distinct states due
to the macroscopic quantum tunneling, a Schrödinger Cat state is very unstable in nature.
For instance, in this case, if we add infinitesimal small energy difference between the left
and the right sides, that is, (â†1 â1 − â†2 â2 ), which is inevitable in reality, the energy dif-
ference between |N, 0 and |0, N is N. This energy difference can easily overwhelm the
exponential small coupling between |N, 0 and |0, N . Hence, the actual ground state is
either |N, 0 or |0, N . Another way to view this problem is to consider a dynamical pro-
cess. For instance, suppose that initially all atoms are condensed in one of the well because
of this energy difference, when we turn off the energy difference such that the Hamilto-
nian restores the Z2 symmetry, the system should oscillate between two wells to restore the
symmetry. However, because of the extremely small tunneling rate, the oscillation period is
extremely long, which can easily exceed the lifetime of our universe for system with large
number of particles. In case of ferromagnetism, the presence of an infinitesimal magnetic
field can pin the spins to certain direction and breaks the SU(2) spin rotational symmetry.
When the infinitesimal small magnetic field is turned off, it also takes extremely long time
to rotate all spins from one to another direction and to restore the SU(2) spin rotational
symmetry.
In summary, we have shown that the instability of the Schrödinger Cat state and the
stability of the symmetry-breaking phenomenon in nature are two sides of the same coin.
From the discussion above, it is important to realize that symmetry breaking is a unique
phenomenon in many-body system with large number of degrees of freedom.
Exercises
3.1 Discuss the Bose–Einstein condensation temperature for free bosons at one and two
dimensions.
3.2 Calculate the two-body density matrix ρ(r, r ) for noninteracting bosons above and
below the Bose–Einstein condensation temperature.
†
3.3 Show that if we define α̂ k and α̂−k as a unitary transformation of â k and â†−k , α̂ k
†
and α̂−k do not obey the boson commutative relation.
3.4 Show that the wave function Eq. 3.57 is the ground state wave function of the
√
Bogoliubov Hamiltonian that satisfies G|âk=0 |G = N0 and α̂ k |G = 0.
3.5 Show that the detailed balance condition
(1 + f k )(1 + f q )f k+q = f k f q (1 + f k+q ) (3.131)
can be satisfied if f k satisfies the Bose distribution
1
fk = (3.132)
−1 eβ E k
and E k + E q = Ek+q . Also show that the detailed balance condition
(1 − f k )(1 − f q )f k+q = f k f q (1 − f k+q ) (3.133)

117 Exercises
can be satisfied if f k satisfies the Fermi distribution

1
fk = (3.134)
eβ E k + 1
and E k + E q = Ek+q .
3.6 Calculate Sρ k (ω) with the Bogoliubov Hamiltonian and show the Feynman relation
2 k2 /(2m)
Sρ k (ω) = . (3.135)
Ek
3.7 Considering two atoms in a one-dimensional ring with size L, compute the ground
state energy as a function of g1d .
3.8 Verify Eq. 3.94 for the mean-field state Eq. 3.96.
3.9 Numerically solve the Hamiltonian Eq. 3.92 for a finite number of particles, and
discuss the wave function, density matrix, and relative particle fluctuations in three
regimes (i) U > 0 and U J; (ii) U > 0 and U J; and (iii) U < 0 and |U| J.
In regime (iii), also discuss the excitation gap between the first excited state and the
ground state as a function of total particle number N.
3.10 Discuss the Hanbury–Brown–Twiss effect for noninteracting fermions and compare
its difference with noninteracting bosons.

ch3_interaction_effects

Uploaded by

Copyright:

Available Formats

ch3_interaction_effects

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

ch3_interaction_effects

Uploaded by

Copyright:

Available Formats

3 Interaction Eﬀects

• Define Bose–Einstein condensation in an interacting boson system in terms of macro-

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

3.1 Bose–Einstein Condensation

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

If the system is in a mixed state with probability ps in a many-body wave function

Alternatively, in a second quantized form, it can be equivalently defined as

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

(a) (b) Velocity (cm/s)

4.25 0.76 MHz

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

ρ(r, r )  N0 ψ ∗ (r)ψ(r ). (3.10)

Before ending this part, we shall make two remarks:

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

3.2 Hydrodynamic Theory

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

Thomas–Fermi Distribution. Now we consider the physical consequences of the hydro-

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

Anisotropic Expansion. Now let us consider the time-of-flight expansion of a Bose

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

Two-Fluid Hydrodynamics. Now we return to the uniform situation but we concern

ij = Pδij + ns vsi vsj + nn vni vnj , (3.44)

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

3.3 Bogoliubov Theory

with k1 + k2 = k3 + k4 , where k = 2 k2 /(2m). The Bogoliubov theory is a perturbative

and under this wave function, we have

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

and the density of quantum depletion ndp = Ndp /V is given by

At this order, the Bogoliubov quasi-particle interact with each other.

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

So the net rate is given by

× [(1 + f k )(1 + f q )f k+q − f k f q (1 + f k+q )]δ(Ek+q − E q − E k ).

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

and the damping rate

The Bragg Spectroscopy. Spectroscopy refers to a probe by a time-periodical perturba-

E = E1 eik1 r−iω1 t ẑ + E2 eik2 r−iω2 t ẑ, (3.75)

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

V(r) ∝ cos(k · r − ωt), (3.76)

where k = k1 − k2 and ω = ω1 − ω2 . In the second quantized form, this perturbation Ĥperp

Ĥperp ∝ ρ̂ k e−iωt + h.c., (3.77)

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

• The mean-field interaction energy and the chemical potential increase.

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

3.4 One-Dimensional Bosons

Box 3.1 One Dimension and Quasi One Dimension

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

 = eik1 x1 +ik2 x2 − eik1 x2 +ik2 x1 , (3.85)

https://doi.org/10.1017/9781108595216.004 Published online by Cambridge University Press

ρ(r, r ) N0 ψ ∗ (r)ψ(r ). (3.10)

ij = Pδij + ns vsi vsj + nn vni vnj , (3.44)

where k = k1 − k2 and ω = ω1 − ω2 . In the second quantized form, this perturbation Ĥperp

= eik1 x1 +ik2 x2 − eik1 x2 +ik2 x1 , (3.85)

El = −Jl+1 l+1 − Jl l−1 + Ul2 l , (3.105)