Quantum Matter / Matéria Quântica

Solutions to Problem set 8 — Trapped BEC & weakly interacting Bose gas

Notation recap

Effective Hamiltonian for the weakly interacting Bose gas. In the lectures, we discussed
Bogoliubov’s approach to the uniform, weakly interacting Bose gas, with inter-particle interactions
modeled by a s-wave pseudopotential of the form U (ri , rj ) = s δ(ri − rj ). In particular, using
Bogoliubov’s macroscopic approximation, we obtained the effective many-body Hamiltonian,

N 2s X′  †  N s X′  
Ĥeff ≡ + ε̃k ak ak + a†−k a−k + a†k a†−k + ak a−k , (1)
2V V
kx >0 kx >0

where N → ∞ is the total number of particles and V ≡ Ld is the volume of the box containing the
gas; primed sums exclude the term k = 0; a sum with “kx > 0” means “sum over all wavevectors in
half of the whole k-volume” 1 ; and

Ns ~2 k2 Ns X X′
ε̃k ≡ εk + , εk ≡ , µ= , N̂ ≡ a†k ak = N0 + a†k ak . (2)
V 2m V
k k

Bogoliubov-Valatin transformation (BVT). The Hamiltonian (1) is diagonalized with the

transformation
( (
αk ≡ uk ak − vk a†−k ak ≡ u∗k αk + vk α†−k
↔
α†−k ≡ −vk∗ ak + u∗k a†−k a†−k ≡ vk∗ αk + uk α†−k

where uk and vk (which may be chosen real in this case) are given by
r r
ε̃k 1 ε̃k 1
uk = + , vk = − − . (3)
2Ek 2 2Ek 2

The Bogoliubov Hamiltonian then becomes diagonal in the new operators αk , with the form

gN X′ 1 X′
K̂ ≃ + Ek α†k αk + (Ek − ε̃k ) , (4)
2 2
k k

where the k-sum is over the entire k-space except k = 0, and the quasiparticle dispersion reads
q
Ns
Ek ≡ ε̃2k − g2 , g≡ . (5)
V

This treatment is appropriate to describe the ground-state of the weakly interacting Bose gas, or at
temperatures low enough that the statistical behavior is dominated by the low-energy excitations
described by the Hamiltonian (4) and the scenario N0 ≈ N remains valid.

X′
1
It is of course redundant to write , since kx > 0 already excludes the case k = 0. However, to emphasize
kx >0
that terms containing k = 0 are excluded, we shall use both the prime and “kx > 0” in the relevant expressions.

1
Problem 1 (BEC of harmonically trapped non-interacting bosons)

In the lectures, we discussed the ideal Bose gas (non-interacting Bosons) with the particles trapped
in a box potential. Here we consider the analogous problem of a non-interacting Bose gas, but now
trapped by a harmonic potential.

Consider a system of N non-interacting spinless bosons with mass m trapped in a isotropic 3D

harmonic potential with characteristic frequency ω. The Hamiltonian of the system is

N  2 
X p̂ i mω 2 r̂i2
Ĥ = +
2m 2
i=1

and the energy spectrum of a single particle in this potential is

3~ω
Enx ,ny ,nz = ~ω(nx + ny + nz ) + E0 , E0 ≡ , nj ∈ {0, 1, 2, . . . }.
2
a) If we write Enx ,ny ,nz = E0 + ~ω n, where n ≡ nx + ny + nz , we can associate some integer n
to each of the allowed energy eigenvalues of a single particle. Let D(E) ~ω be the number of
states with energy equal to E = E0 + ~ω n. Show it is given exactly by

(n + 2)(n + 1)
D(E) ~ω = .
2

b) Verify that, if E ≫ ~ω, the previous result yields a density of states, D(E), that can be
approximated as
E2 3E
D(E) ≃ + .
2(~ω)3 2(~ω)2
What is the condition for this approximation to be valid?

c) Show that we may obtain this last result for D(E) using the definition for the density of states
in the continuum,
∞
dΩ(E) X
D(E) = , Ω(E) ≡ θ(E − Enx ,ny ,nz ),
dE
nx ,ny ,nz =0

and approximating the sums over nj by integrals. In this expression, Ω(E) is the total number
of states with energy less or equal to E. Comment on why this approach yields the same result
as that derived using the explicitly quantized (discrete) spectrum of the previous question.

d) At thermal equilibrium, N and the chemical potential, µ, are related by the condition
∞  3  2
X 1 kB T 3 kB T
N= = N0 + g3 (z) + g2 (z). (6)
nx ,ny ,nz =0 eβ(Enx ,ny ,nz −µ) − 1 ~ω 2 ~ω

Obtain the result in the right-hand side, where N0 is the number of particles in the ground

2
 state, z ≡ eβ(µ−E0 ) and gs (z) is the polylogarithm function:
Z ∞ ∞
1 xs−1 X zl
gs (z) ≡ dx = , gs (1) = ζ(s). (7)
Γ(s) 0 z −1 ex − 1 ls
l=1

e) Since the number of bosons in the ground state is given by

z solving for z N0
N0 = −−−−−−−−−−−→ z= .
1−z N0 + 1

Therefore, in the thermodynamic limit (N ≫ 1) we may determine the BEC transition tem-
perature, Tc , by setting z = 1 in eq. (6) and ignoring the term N0 . Show that in the limit
N → ∞, Tc and the temperature dependence of the condensate fraction are given by
 1  3
N 3 N0 (T ) T
kB TC = ~ω , =1− .
ζ(3) N Tc

f) Given the previous result, justify the following statement: as we enter the BEC state (T . Tc ),
the condensate fraction grows twice faster when confined by an harmonic trap than if contained
by a cubic “box” without any additional confining potential.

g) Show that, for T > Tc , the chemical potential is approximately given by

 
T
µ(T ) ≃ E0 − 3kB T log , (N ≫ 1),
Tc

which is consistent with the fact that µ → E0− as T → Tc+ .

Hint — This is the leading-order approximation. Understand how it justifies using an approximate form for
the function gs (z).

Solution

a) (...)

b) (...)

c) (...)

d) (...)

e) (...)

f) (...)

g) (...)

3
Problem 2 (uniform weakly interacting Bose gas)

Consider the approach to the weakly interacting Bose gas summarized by the results in eqs. (1) to
(4) at the start of this problem set.

a) Use the Bogoliubov-Valatin transformation (3) and explicitly derive the diagonal form (4),
starting from the original Hamiltonian in eq. (1).

b) Show that the Bogoliubov quasiparticle spectrum has the asymptotic behavior

~c k, for small k Ns
Ek ≃ , g≡ .
εk + g, for large k V

Obtain the sound speed c in terms of the original parameters and identify the characteristic
wavevector, kc , that defines the “small k” and “large k” regimes.

c) Compute the amount of condensate depletion, N − N0 , at T = 0 due to interactions. Show

that, in the 3D thermodynamic limit, we obtain
 
N − N0 hN̂ex i 1 X′ † 1 X′ ε̃k 1 1 ~
= = hak ak i = − ≃ 2 3, λc ≡ .
V V V V 2Ek 2 3π λc mc
k k

d) Without explicitly computing the final result, verify that the analogous calculation of (N −
N0 )/V at finite (but small) temperatures:

i) remains finite in 3D;

ii) diverges in 2D but is finite at strictly T = 0;

iii) diverges at any temperature in 1D.

e) Let |Ψgs i be the ground state of the many-body Bogoliubov Hamiltonian (4). Its defining
property is being the vacuum of the Bogoliubov operators: αk |Ψgs i = 0, for all k 6= 0. Show
that this is fulfilled by the state
 
X′
|Ψgs i = N exp  wk a†k a†−k  |φ0 i,
kx >0

with an appropriate choice of the coefficients wk . In this expression, N is a normalization

constant, wk are constants, and |φ0 i is a normalized many-particle state with occupation only
in the k = 0 state (the state where BEC occurs; the state φ0 doesn’t need to be specified in
this question). Obtain wk explicitly in terms of the coefficients uk and vk of the Bogoliubov
transformation.
Hint — It’s helpful to remember that, given a function F with derivative F ′ , then [ a, F (a† ) ] = F ′ (a† ),
which provides a simple way to evaluate bosonic commutators with such structure.

4
 f) Show that the normalization constant above is
Y′ 1
N = , (uk : coefficient of the BVT).
uk
kx >0

g) Given the expression above for |Ψgs i, describe in qualitative terms the nature of the Bogoliubov
ground state:

i) How can we describe it in terms of the single-particle occupations?

ii) Which wavevectors (or single-particle energies) more strongly modify |Ψgs i relative to
|φ0 i?

iii) Is Ψgs given above consistent with the expected ground-state in the non-interacting limit
(s → 0)?

Solution

a) (...)

b) The quasiparticle spectrum is given by

~2 k 2
q
Ns
Ek ≡ ε̃2k − g 2 , ε̃k ≡ εk + g, g≡ , εk ≡ ,
V 2m
so s 
q 
2 2
p ~2 4mg
Ek = (εk + g) − g = εk (εk + 2g) = k2 k2 + 2 .
2m ~

We can define the quantity kc ≡ 4mg/~2 , with dimensions of wavevector, and write
s  2
~2 k 2 kc
Ek = 1+ .
2m k

The behavior in opposite limits of k relative to kc is

r
~2 p 2 2 ~2 kc k g
k ≪ kc : Ek ≃ kc k = = ~k .
2m 2m m
s  2  
~2 k 2 kc ~2 k 2 kc2 ~2 k 2
k ≫ kc : Ek = 1+ ≃ 1+ 2 = + g.
2m k 2m 2k 2m

We conclude that
4mg 4mN s
kc ≡ 2
= 2
~ ~ V
is a characteristic wavevector, proportional to the interaction strength s, which defines the two regimes
( r r
~c k, k ≪ kc g Ns
Ek ≃ , c≡ = ,
εk + g, k ≫ kc m mV

where c is the coefficient with dimensions of velocity for the linear dispersion at low energies.

5
c) Since we must compute
N − N0 1 X′ †
= hak ak i,
V V
k

where h· · · i is the expectation value in the ground state (the vacuum of the operators αk ), we must express
the operators in terms of αk using the BV transformation, according to which
r r
ε̃k 1 ε̃k 1
ak = uk αk + vk α†−k , uk = + , vk = − − .
2Ek 2 2Ek 2

Hence,
  
ha†k ak i = h uk α†k + vk α−k uk αk + vk α†−k i = u2k hα†k αk i + vk2 hα−k α†−k i = vk2 ,

and we establish the first part of the result:

 
N − N0 1 X′ 2 1 X′ ε̃k
= vk = −1 .
V V 2V Ek
k k

The second part consists in computing this sum in the thermodynamic limit as an integral:
Z Z ∞
N − N0 1 X′ 2 1 2 1
= vk ≃ vk dk = vk2 k 2 dk.
V V (2π)3 2π 2 0
k

To compute this integral we will note that vk2 is a monotonic decreasing function of k with asymptotic
behavior (
1
2 4, k→∞
vk ∝ k , vk2 −−−−→ 0, vk2 −−−−→ +∞,
1 k→∞ k→0
k , k → 0

so we can partially integrate:

Z ∞ by parts Z ∞   Z ∞  
vk2 k 3 ∞ 1 dvk2 1 dvk2
vk2 2
k dk = − 3
k dk = − k 3 dk.
0 3 0 3 0 dk 3 0 dk

Now, since q  
1 ε̃k
Ek = ε̃2k − g 2 and vk2 = −1 ,
2 Ek
we have
        
dvk2 1 dε̃k 1 dEk ε̃k 1 dε̃k 1 dEk ε̃k
= − = −
dk 2 dk 2E dk Ek2 2 dk Ek dε̃k Ek2
      
1 dε̃k 1 ε̃2 dε̃k −g 2 g 2 ~2 k
= − k3 = 3 =− .
2 dk Ek Ek dk 2Ek 2mEk3

6
 Replacing this in the integral,
Z ∞ Z ∞ Z ∞
g 2 ~2 k4 g 2 ~2 k4
vk2 k 2 dk = 3 dk = 3/2 dk
0 6m 0 Ek 6m 0
 2 2
k
( ~2m + g)2 − g 2

 ~2 k2
y set u = +g
2m
 3/2   Z ∞
g 2 ~2 2m m (u − g)3/2
= du
6m ~2 ~2 g (u2 − g 2 )3/2
 3/2 Z ∞
g 2 2m du
=
6 ~2 g (u + g)3/2
 3/2 " #∞
g 2 2m 2
= −
6 ~2 (u + g)1/2 g
 3/2  1/2
g 2 2m 2
=
6 ~2 g
2  mg 3/2
= .
3 ~2
Putting everything together, we arrive at
Z ∞  1/2
N − N0 1 1  mg 3/2 1 ~2
= vk2 k 2 dk = = 2 3, λc ≡ .
V 2π 2 0 3π 2 ~ 2 3π λc mg
pg q
Ns
Recalling from question (b) that we defined c ≡ m = mV , the wavelength λc can thus be written as

~
λc ≡ .
mc

d) Zero temperature. At T = 0, the amount of depletion in d dimensions is given by

Z Z
N − N0 1 X′ 2 1 2 Sd−1 ∞ 2 d−1
= vk ≃ vk dk = vk k dk,
V V (2π)d (2π)d 0
k

where Sd is the area of the unit d-sphere. Since

q
ε̃k 1
Ek = ε̃2k − g 2 and vk2 = − ,
2Ek 2

then, asymptotically, (
1
k4 , k→∞
vk2 ∝ .
1
k, k→0

The integral above is clearly convergent at k → ∞ for d ≤ 3. At k → 0, the integrand is

vk2 k d−1 ∝ k d−2 ,

so the integral only converges if

d − 2 > −1 −→ d > 1.

This means that Bogoliubov’s model of the weakly interacting Bose gas predicts that the BEC is stable
at T = 0 with respect to quantum fluctuations for d > 1. In d = 1, the BEC is predicted to be unstable
against quantum fluctuations.

7
 Nonzero temperature. For finite temperature, we need to go back and re-compute the thermal ex-
pectation value:
  
ha†k ak iT = h uk α†k + vk α−k uk αk + vk α†−k iT = u2k hα†k αk iT + vk2 hα−k α†−k iT

= u2k hα†k αk iT + vk2 hα†−k α−k iT + vk2

u2k + vk2
= + vk2 .
eβEk − 1

The new contribution at T 6= 0 comes from the first term in the updated result for ha†k ak i:
Z Z
N − N0 1 u2k + vk2 u2k + vk2 d−1
= (· · · ) + dk ∝ (· · · ) + k dk.
V | {z } (2π)d eβEk − 1 eβEk − 1
contrib. at T =0

As k → 0, the integrand of the second term behaves as

u2k + vk2 d−1 k −1 d−1

k −−−→ k = k d−3 .
eβEk − 1 k→0 k

Therefore, the condensate depletion due to thermal fluctuations diverges for d < 3. Bogoliubov’s model
thus predicts that, while the BEC can remains stable for d = 3 (up to some finite critical temperature), it
doesn’t survive thermal fluctuations in lower dimensions.

In summary, we found that Bogoliubov’s model predicts that:

• For d = 3, there is a finite critical temperature, Tc , below which we have a BEC.

• For d = 2, the critical temperature is zero; while stable at strictly T = 0, the BEC is destabilized for
arbitrarily small T > 0.

• For d = 1, there is no BEC; quantum fluctuations are sufficiently strong to suppress it even at T = 0.

e) We must verify that the given state is the vacuum of all the operators αk . Since operators with different
k commute, we may factor the exponential of the sum as a product of exponentials:
!
X′ Y′  
† †
|Ψgs i = N exp wk ak a−k |φ0 i = N exp wk a†k a†−k |φ0 i.
kx >0 kx >0

(Note that this is possible because the primed sum involves only half of the vectors in k space, and this
guarantees that −k is never equal to some other q; therefore, all the terms in the sum commute among
themselves.) To compute αk |Ψgs i, recall that

αk = uk ak − vk a†−k .

Since αk involves only ak and a†−k , it commutes with all the factors appearing in Ψgs , except the one
involving k,  
Y′    
αk |Ψgs i = N  exp wq a†q a†−q  αk exp wk a†k a†−k |φ0 i,
qx >0, q6=k

8
 so that we need only evaluate the last part:
     
αk exp wk a†k a†−k |φ0 i = uk ak − vk a†−k exp wk a†k a†−k |φ0 i
   
= uk ak exp wk a†k a†−k |φ0 i − vk a†−k exp wk a†k a†−k |φ0 i


y use the property [a, F (a† )] = F ′ (a† ) in the first term
   
= uk wk a†−k exp wk a†k a†−k |φ0 i − vk a†−k exp wk a†k a†−k |φ0 i
   
= uk wk − vk a†−k exp wk a†k a†−k |φ0 i.

Clearly, the action of the operator a†−k exp(wk a†k a†−k ) cannot be zero since it involves only creation
operators. The only way to obtain zero is if the prefactor is explicitly zero, which determines how the
coefficient wk must be chosen. Since this calculation applies to all k 6= 0, it follows that we must have
vk
wk = .
uk

f) We must compute the normalization constant, N , by requiring hΨgs |Ψgs i = 1. With the given expression
for Ψgs we have
! !
hΨgs |Ψgs i X′
∗
X′ † †
= hφ0 | exp wk ak a−k exp wk ak a−k |φ0 i
N2
kx >0 kx >0
Y′ Y′
 
= hφ0 | exp wq∗ aq a−q exp wk a†k a†−k |φ0 i.
qx >0 kx >0

Each exponential separately generates an infinite number of states covering all possible numbers of particle
pairs with opposite momenta, as we can easily see with a Taylor expansion:
∞
  X (wk )n † † n
exp wk a†k a†−k |φ0 i = (ak a−k ) |φ0 i
n=0
n!
∞
!n !n
X
n a†k a†−k
= (wk ) √ √ |φ0 i
n=0 n! n!
∞
X
= (wk )n |n(k) i|n(−k) i|φ0 i,
n=0

where |n(k) i represents the normalized Fock state with n bosons in the single-particle orbital with wavevec-
tor k. The product over k then becomes

Y′   ∞ X
X ∞
(k ) (−k ) (k ) (−k )
exp wk a†k a†−k |φ0 i = · · · (wk1 )n1 (wk2 )n2 · · · |n1 1 , n1 1 , n2 2 , n2 2 , . . . i|φ0 i,
kx >0 n1 =0 n2 =0

9
 where we used ki to label all the distinct wavevectors and ni for the respective occupations. We thus have
∞ ∞ ∞
! ∞
!
hΨgs |Ψgs i X X
2n1 2n2
X
2n1
X
2n2
= · · · |wk1 | |wk2 | · · · = |wk1 | |wk2 | ···
N2 n1 =0 n2 =0 n1 =0 n2 =0


y sum each geometric series
  
1 1
= ···
1 − |wk1 |2 1 − |wk1 |2
Y′ 1
= .
1 − |wk |2
kx >0

Recalling from the previous question that wk = vk /uk , the normalization constant becomes
s s
Y′ p Y′ |vk |2 Y′ |uk |2 − |vk |2
N = 1 − |wk |2 = 1− 2
=
|uk | |uk |2
k >0 k >0 k >0
 x x x

 2 2
y remember that |uk | − |vk | = 1 (always in a Bogoliubov transformation)
s
Y′ 1
=
|uk |2
kx >0


y since uk > 0
Y′ 1
= .
uk
kx >0

g) From the solution of the previous two questions:

i) It is clear that the Bogoliubov ground state consists of a linear combination of pairs of particles of
opposite momenta (k, −k), in addition to the particles in the lowest energy orbital represented by
|φ0 i. The pairs span all momentum values (excluding k = 0) and the number of (k, −k) pairs runs
from zero to infinity.

ii) However, we saw above that this linear combination is not arbitrary! The weight of each contribution
is governed by the coefficients wk , which are given by
r q
vk ε̃k − Ek Ns
wk = =− , Ek ≡ ε̃2k − g 2 , ε̃k ≡ εk + g, g≡ .
uk ε̃k + Ek V

We note that
wk < 0, lim wk = −1, lim wk = 0− ,
k→0+ k→∞

and |wk | decreases with increasing |k|. Recalling that

!
X′
|Ψgs i = N exp wk a†k a†−k |φ0 i,
kx >0

this means that the contributions are progressively suppressed with increasing wavevector: pairs of
particles with higher single-particle energy εk are represented with less weight/probability in that
linear combination. Alternatively, the wavevectors that more strongly modify |Ψgs i relative to |φ0 i
are those near k = 0.

Another way to say this is that Ψgs differs from φ0 because interactions change the nature of the

10
 ground state, which now includes zero-momentum pairs (k + (−k) = 0) even at T = 0. Therefore,
the number of particles with a given momentum is not sharply defined but has some finite variance,
even at T = 0. This effect is the origin of the designation “quantum fluctuations”: in this case, quan-
tum fluctuations in the occupation of a given single-particle state, in the sense that those occupations
are not sharply defined and have an associated uncertainty.

iii) Without inter-particle interactions, we expect the ground state to consist of all the particles occupying
the k = 0 orbital. This is naturally captured by the expression we obtained for Ψgs because, if we
turn the interactions off,

lim wk = 0, lim uk = lim N = 1, −→ |Ψgs i = |φ0 i,

g→0 g→0 g→0

which, according to the definition given for φ0 , is a state where all occupations are zero, except for
the k = 0 orbital, precisely as expected for the ground state without interactions.

11