221B Lecture Notes

Quantum Field Theory II (Bose Systems)

1 Statistical Mechanics of Bosons

1.1 Partition Function
As discussed in 221A, the path integral with the imaginary time gives you
the partition function of the system. Let us consider the partition function
of the free Schrodinger field theory, first the bosonic one. This calculation
shows non-trivially that the quantized Schrodinger field theory indeed con-
tains multi-body states automatically and is also useful from a practical point
of view.
The path integral in the particle quantum mechanics is given by the
summation over all possible paths in the configuration space xi (t). In the field
theory, (~x) is the canonical coordinate, and it can follow various paths
(~x, t). Therefore, the path integral in field theory is a summation over all
possible field configurations (~x, t). This discussion defines the path integral
Z
D(~x, t)D (~x, t)eiS/h , (1)
where the action is that of the field theory, such as
2
!
Z
h

S= h +
d~xdt i
(2)
2m
in the free case.
To calculate the partition function, we go to the imaginary time t = i ,
h2
" !#
Z
1 Z h Z
Z=
D(~x, )D (~x, ) exp +
d d~x h .
h
0 2m
(3)
We impose the periodic boundary condition (~x, t + h ) = (~x) for =
1/kT .
Because of the periodic boundary condition in and also in space due to
the box normalization, we can expand in Fourier series both in the imaginary
time as well as in space,
1 X
(~x, ) = 3/2 zp~,n ei~p~x/h e2in /h (4)
L p~,n

1
and simiarly for
1 X i~p~x/h 2in /h
(~x, ) = z e e . (5)
L3/2 p~,n p~,n

Then the exponent in the partition function Eq. (3) is

h2 p~2
! !
1 Z h Z
+
d d~x h zp~,n zp~,n . (6)
X
= 2in +
h
0 2m p
~,n
2m

Therefore, the path integral is simply a product of many many Gaussian

integrals
YZ 2 /2m)z z
dzp~,n dzp~,n e(2in+~p
Y
Z= ~,n p
p ~,n
= . (7)
p
~,n p
~,n
2in + ~p2 /2m

Now we use the infinite product representation of the hyperbolic functions

x2
!
Y sinh x
1+ 2 =
n=1 n x

x2
!
Y x
1+ = cosh . (8)
n=1 (2n 1)2 2

We use the first identity here. The partition function Eq. (7) can be rewritten
as

Y Y
Z =
p
~
~p2 /2m n=1 (2n)2 + (~p2 /2m)2

!1
Y Y (~p2 /2m/2)2
= 1 +
p
~
~p2 /2m n=1 (2n)2 n2

~p2 /4m
!
Y Y
= 2
p
~ n=1 (2n) ~p2 /2m sinh ~p2 /4m
Y 2
= c
p
~
sinh ~p2 /4m
2
Y e~p /4m
= c . (9)
p
~
1 e~p2 /2m

2
where c is an (infinite) overall constant, which is not important when evaluat-
ing various thermally averages quantities. This is a grand partition function
which contains summation over all possible number of particles but with no
chemical potential. One small difference from the conventional calculation is
that the path integral automatically includes the zero-point energy p~2 /4m,
which results in the exponential factor in the numerator. The reason why
there is this zero-point energy is because the Fock space is a collection of
(infinite number of) harmonic oscillators and each harmonic oscillator has
the zero-point energy. But the zero-point energy does not lead to any phys-
ical consequences (while the zero-point fluctuation does) and we can always
shift the energy of the system by an infinite constant p~ p~2 /4m to remove
P

the zero-point energy in the expression.

This simple calculation clearly shows the advantage of the Schrodinger
field theory: it sums up states with different number of particles automati-
cally.
The inclusionRof the chemical potential is obvious. Because the number
operator is N = d~x (~x)~x and the grand canonical emsenble is summed
with a factor of e(EN ) , the path integral Eq. (3) is modified to
Z
Z = D(~x, )D (~x, )
h2
" !#
1 Z h Z
exp +
d d~x h . (10)
h
0 2m
The result of the path integral is then
2 /2m)/2
Y e(~p
Z=c . (11)
p
~
1 e(~p2 /2m)
Note that Z = e with = pV = F N for the grand partition
function. Namely,
2
X e(~p /2m)/2
= kT ln
p
~
1 e(~p2 /2m)
p~2
" ! #
X 1 2
= + kT ln(1 e(~p /2m) ) . (12)
p
~
2 2m
The thermodynamic quantities are related by
F = U T S, (13)
= F N, (14)

3
and have the following differentials
dU = T dS pdV + dN, (15)
dF = SdT pdV + dN, (16)
d = SdT pdV N d. (17)
Therefore, the number is given by
2
e(~p /2m)
 " #
 1 X 1 1
X  
N = = + (~p2 /2m) = + (~p2 /2m) .
2 1e 1

T,V p
~ p
~
2 e
(18)
Apart from the zero-point piece 1/2, this is the familiar expression for the
occupation number for Bose-Einstein distribution.
The thermal average energy U is given by the Legendre transform U =
+ T S + N with S = /T |V, . We obtain
~2 1
X p 1
 
U= + . (19)
p
~
2m 2 e(~p2 /2m) 1
This must be a famililar expression to you, again except the zero-point energy
term that can be dropped without changing physical content.

1.2 BoseEinstein Condensate

One prime application of Schrodinger field theory is the BoseEinstein con-
densate (BEC).
The expression for the thermally averaged energy Eq. (19) has a problem
when the chemical potential is positive. The region of the momentum space
p~2 /2m < does not give meaningful result: it is negative. The explanation
you have heard back in college may be that this suggests that there is a
macroscopic number of bosons condensed in this momentum region. But
what is actually going on?
Let us go back to the partition function Eq. (10), but now with a -
function repulsive potential term
Z
Z= D(~x, )D (~x, )
h2
" !#
1 Z h Z 1
exp +
d d~x h .
h
0 2m 2
(20)

4
(We used to write = h 2 .) The action in the exponent permits the inter-
pretation that the chemical potential is a part of the potential term
1
V () = + (21)
2
Now we can ask the question what happens when > 0. The potential for
is minimized when it has a finite value:
i
r
= e , (22)

where is an arbitrary phase. The expectation value of the Schrodinger
field is the BoseEinstein condensate. The number density of particles in the
condensate is of course
N
= 3 = = , (23)
L
and this is the equation what determines the chemical potential.

Im
Re

Figure 1: The winebottle or Mexican hat potential given in Eq. (21) for
> 0.

When we saw that the quantized Schrodinger field theory gives multi-
particle states, we didnt ask the question what Schrodinger field meant before
the quantization. We were happy that we could successfully obtain the multi-
particle quantum mechanics. In the normal situation without a condensate,
the expectation value of vanishes. Therefore is definitely not classical
and genuinely quantum. The particles are quantum fluctuations around the
vanishing expectation value hi = 0. On the other hand, what we see here is

5
that the classical Schrodinger field can exist and describe the BoseEinstein
condensate.
Remember that the field operator is an annhilation operator. If it
has an expectation value in the ground state, the ground state cannot be
an eigenstate of the number operator. This point can be understood by
assuming that the only condensate is in the zero-momentum state. The
Hamiltonian for the zero-momentum mode with the delta-function potential
V (~x ~y ) = (~x ~y ) or equivalently V (~p ~q) = L13 is

H = a (0)a(0) + a (0)a (0)a(0)a(0). (24)
2L3
It is difficult to diagonalize this Hamiltonian. However, the following varia-
tional method can be used. The coherent state
f /2 (0)
|f i = ef ef a |0i, (25)

as discussed in 221A, is an eigenstate of the annihilation operator

a(0)|f i = f |f i, (26)

and the expectation value of the Hamiltonian for this state is

hf |H|f i = f f + f f f f. (27)
2L3
Here, we used the fact that hf |a (0) = hf |f . Minimizing it with respect to
the complex parameter f , the variational method suggests the approximate
ground state
s
L3 i 1 2 L 3
|f i with f= e , E = f f = . (28)
2 2
The energy is clearly lower than the vacuum |0i. This state obviously
shows an expectation value for the annihilation operator
s
L3 i
hf |a(0)|f i = e (29)

and hence also for the field operator
1 X i
r
p~
i~ x/
h
hf |(~x)|f i = hf |a(~
p )e |f i = e , (30)
L3/2 p~

6
consitent with Eq. (22). Even though the coherent state is not the true
ground state of the Hamiltonian, it is clearly close enough as suggested by
the classical minimum of the Schrodinger field.
Note that the condensate could be described in this formalism because
it allowed states with different number of particles in the same Hilbert
space. This state could never be described in the conventional multi-body
Schrodinger wave functions.

1.3 More on Coherent States

How good is the variational method in this case? To see this, let us go back
to the Hamiltonian Eq. (24) and act it on the coherent state Eq. (25). We
find
!

H|f i =
a a + 3 a a aa) |f i
2L
!
2
= a f + 3 a a f ) |f i (31)
2L

We drop the momentum index in this section. What we need to know now
is the action of the creation operator on the coherent state.
It is useful to look at the probability distribution in the number of parti-
cles in a coherent state.
2 2

 
n

f f /2 f

 fn  (f f )
n
2
P (n) = |hn|f i| = hn|e (a ) |0i = ef f
n
hn| |ni = ef f .
   
 n!   n!  n!
(32)
This is nothing but the Poisson distribution with the averange n = f f .
Therefore for a large N = f f , the fluctuation in the number is N = N
and hence the number of particles in the coherent state is determined more
and more accurately as N increases: N/N = 1/ N . Assuming N  1, the
number operator
a a should therefore return the value N up to corrections
of order 1/ N . What it means is that, in the limit of large N , the coherent
state is nearly an eigenstate of the creation operator such that

N |f i = a a|f i = f a |f i ' f f |f i + O(N )1/2 . (33)

In this limit,
the variational state Eq. (28) becomes exact up to corrections
of order 1/ N .

7
 What is interesting is the emergence of coherence at the expense of uncer-
tainty in the number. This is the reflection of what is called number-phase
uncertainty principle. We can define the phase operator by

a = ei N , (34)
which is clearly consistent with the definition of the number operator N = a a
is the number operator. This definition is singular when N = 0, but because
we are interested in states with a macroscopic number of particles in the
condensate N  1, let us ignore the subtlety that happens only when N = 0.
From the commutation relation [N, a] = a, we find
[N, ei ] = ei , (35)
which can be rephrased as

N =i . (36)

Therefore, we find the commutator
[N, ] = i. (37)
In analogy to the canonical commutation relation [x, p] = i h giving rise to
the uncertainty principle xp h /2, we find the number-phase uncertainty
principle
1
N . (38)
2
One can construct eigenstate of the phase within the cheat we did with
the subtlety with the N = 0 state. Consider

ein |ni.
X
|i (39)
n=1

We now act the phase operator ei = a 1N on this state, and find

i 1
ein a |ni
X
e |i =
n=1 n

1
ein
X
= n|n 1i
n=1 n

= ei ein |ni
X

n=0
i
= e (|i + |0i). (40)

8
Up to the subtlety with the state |0i, it is an eigenstate of the phase
operator. Note that in this state, the number is completely uncertain: the
states with different numbers are added together with equal weight (unity).
It is useful to picture what the number-phase uncertainty means.
In case
of a simple harmonic oscillator, one can write a = (x + ip)/ 2 h (Ive set
m = = 1). On the classical phase space (x, p), the number operator
N is the squared radius from the origin (times a half), while the phase
operator ei is nothing but the phase on the complex plane x + ip. The
uncertainty principle tells you that a quantum mechanical state occupies the
minimum area of 2 h on the phase space. The number eigenstate is therefore
approximately a daughnut with a radius 2 hN x2 + p2 2 h(N + 1).
The phase is completely uncertain in this case. On the other hand, the
phase eigenstate would correspond to a thin ray emanating from the origin
towards infinity. It has a well-defined phase, while the number is completely
uncertain. The coherent state is a compromisebetween the number and
phase. The number is uncertain
only by N = N and hence the relative
error is small N/N = 1/ N for N  1. That allows the phase to be also
relatively well determined ' 1/ N . It can be viewed as a patch around
a point on the phase space spread both along the radial and the angular
directions.
Note that the coherent state is not a ground state of the Hamiltonian
Eq. (24) but we regard it as a variationa ansatz. If this ansatz is better than
the number eigenstate has to be studied including the non-zero modes.

1.4 Excitations above BoseEinstein Condensate

The fascinating aspect of BoseEinstein condensates is that a macroscopic
number of particles behave collectively as a coherent matter wave. Starting
from the classical picture of particles, it is definitely a highly quantum me-
chanical phenomenon. On the other hand, from the point of view of the field
theory formulation, the classical field describes the coherent matter wave
while its quantization gives ordinary particles. Two approaches are therefore
the opposite.
Going back to the real-time action
" #
Z
h
S= h

dtd~x i + , (41)
2m 2

9
we write down the EulerLagrange equation for the classical field (~x, t)

h
h
i + = 0. (42)
2m
q
The expectation value = / we discussed already is a solution to this
classical equation of motion with minimum energy.
Another (exact) important solution to this classical equation of motion
is given by
i~k~xit
r
(~x, t) = e . (43)

Substituting it into the equation of motion, it is easy to obtain the dispersion
relation
2~k 2
h
h
= . (44)
2m
This solution represents the flow of the BoseEinstein condensate (e.g., a
superfluid flow of 4 He). Coherence of flowing condensate can be seen clearly
in the interference pattern shown in Fig. 2.
In order to see that it describes the superfluid, namely there is no viscosity,
we study microscopic excitations above the condensate. Remember that the
viscosity is caused because the overall flow of the fluid causes microscopic
excitations, namely heat, and the energy/momentum is transferred from the
bulk flow to microscopic excitations. We would like to see that the bulk flow
of BoseEinstein cannot create such microscopic excitations below a critical
velocity.
We study the fluctuation around the static expectation value from this
equation of motion. The field can fluctuate both in the density and the phase.
We parameterize them by

r 
= + ei (45)

where both and are real-valued fields. By plugging this parameterization
into the equation of motion Eq. (42), we obtain

2
h
hi +
h h
i ~ i
( + 2() ~ + (hi + )(()
~ 2 + i))
2m
+(hi + ) (hi + )3 = 0. (46)

10
Figure 2: Interference pattern of two expanding condensates observed after
40 ms time-of-flight, for two different powers of the argon ion laser-light sheet
(raw-data images). The fringe periods were 20 and 15 m, the powers were
3 and 5 mW, and the maximum absorptions were 90 and 50%, respectively,
for the left and right images. The fields of view are 1.1 mm horizontally by
0.5 mm vertically. The horizontal widths are compressed fourfold, which en-
hances the effect of fringe curvature. For the determination of fringe spacing,
the dark central fringe on the left was excluded. Taken from M. R. Andrews,
C. G. Townsend, H.-J. Miesner, D. S. Durfee, D. M. Kurn, and W. Ketterle,
Science, 275 (1997) 637-641.

11
This non-linear equation cannot be solved in general. However, if we are
interested in small fluctuations, we can linearize the equation, i.e., drop all
terms quadratic in the fluctuation or higher. Then the linearized equation is
quite simple:

2
h
i hi +
h h ( + hii)) 2 = 0. (47)
2m
Since the real and imaginary parts of the equation must both be satisfied,
we find two coupled equations

2
!
h
2 hi = 0
+h (48)
2m
2
h
h
+ hi = 0. (49)
2m
Taking h (/t) of the first equation and substituting it into the second one,
we find
2 h
! 2
2 h
h + 2 = 0. (50)
2m 2m
Now using the Fourier modes sin((Et p~ ~x)/
h), we find the dispersion
relation of the fluctuation
p~2 p~2
!
2
E(~p) = 2 + . (51)
2m 2m

In Fig. 3, beautiful measurement of the excitation spectrum in a trapped

BEC is shown. Because the BEC is trapped, it resides in a potential; it
changes the dispersion relation above slightly. Nonetheless for a given trap
there is no free parameter and the theory fits the data beautifully.
For small momentum in Eq. (51), we find that the energy is linear in mo-
mentum. We identify this limit as the sound wave. Note that the field is
fluctuation in the density = = (hi + )2 , and is related to the plane
wave of by Eq. (48). Therefore the wave is indeed a progation of den-
sity fluctuation, which justifies the interpretation. The sound speed is then
directly read off from the dispersion relation Eq. (51) for small momentum

E 2 
c2s = 2  = . (52)
p~ p~0 m


12
Figure 3: The measured excitation spectrum (k) of a trapped Bose-Einstein
condensate. The solid line is the Bogoliubov spectrum with no free parame-
ters, in the local density approximation (LDA) for = 1.91 kHz. The dashed
line is the parabolic free-particle spectrum. For most points, the error bars
are not visible on the scale of the figure. The inset shows the linear phonon
regime. Taken from J. Steinhauser et al ., Phys. Rev. Lett., 88, 120407
(2002).

After quantization, this becomes a quasi-particle (elementary excitation of a

collective system) called phonon with the energy E = cs |~p|.
On the other hand, at large momentum, the dispersion relation Eq. (51)
can be approxiimated as

p~2
E(~p) ' + + O(~p2 )1 (53)
2m
and hence it is the same as the single particle excitation except the offset
= c2s m. This is called the excitation in the free-particle regime.
In the case of liquid 4 He, the interaction is quite strong and the linearized
analysis fails. The dispersion relation rises linearly in the phonon-regime but
it turns around the develops a minimum called roton (see Fig. 4). As
far as I know, there is no first-principle calculation of this spectrum. The
interaction is too strong for the perturbation theory to be valid to make
reliable quantitative predictions.
The linear dispersion in the phonon regime is important because it is the
very reason for superfluidity. Suppose the condensate is flowing with velocity
~v past a macroscopic obstruction of mass M . It is more convenient to go to

13
Figure 4: Taken from D.G. Henshaw and A.D.B. Woods, Phys. Rev., 121,
1266 (1961).

14
the reference frame where the condensate is at rest while the obstruction is
moving. The question is if the obstruction can transfer its kinetic momentum
and energy to create a quasi-particle. If it can, it will keep creating more and
more quasi-particles, which eventually thermalize due to their own interac-
tions. In other words, the original kinetic energy became partially heat, and
the velocity slows down, namely the fluid is viscous. To see if a quasi-particle
can be created, we look at energy- and momentum-conversation laws
1 1 ~0 2
M~v 2 = M v + E(~p), (54)
2 2
M~v = M v~0 + p~. (55)

Substituting the second equation to the first, we find

p~2
0 = ~v p~ + + E(~p). (56)
2M
Because the mass M is a macroscopic mass, we can ignore the second term
in the r.h.s. of the equation. Therefore, in order for a quasi-particle to be
crated, we need to satisfy
E(~p) = ~v p~. (57)
If E(~p) is quadratic in p, as for a normal particle, one can always find a
solution for this equation. However, if E(~p) is linear in p, as is the case here,
we do not find a solution unless v > cs . (Actually, in the presence of the
roton, you find a solution at a smaller v.) Therefore the moving obstruction
cannot create a quasi-particle and hence there is no viscosity. This is why
BoseEinstein condensate of 4 He atoms becomes a superfluid.

1.5 Bogoliugov Transformation

How do we describe the quasi-particle excitation with the operator language?
To study this, we write down the Hamiltonian in the momentum space using
formulae in the previous lecture note,

p~2
!
1
a(~p) a(~p)+ 3 a (~p4 )a (~p3 )a(~p2 )a(~p1 )p~1 +~p2 ,~p3 +~p4 .
X X
H=
p
~
2m 2L p
~1 ,~
p2 ,~
p3 ,~
p4
(58)

15
 q
Because we took the coherent state for a(0) = /L3/2 as the variational
ansatz, the Hamiltonian for the non-zero modes is
p~2
!
1 X 
a(~p) a(~p)+ a(~p)a(~p) + a (~p)a (~p) + 4a (~p)a(~p) ,
X
H=
p
~
2m 2 p~
(59)
where interaction terms proportional to among non-zero modes are omitted.
Note that the second term does not vanish in the weak-coupling limit 0
for fixed (i.e., fixed sound speed) because of the condensate /, while
the interaction terms vanish in the same limit. Therefore, it makes sense
to retain only the terms above and drop the interaction terms to study the
behavior of the non-zero modes. This Hamiltonian is rather peculiar because
it has a term with creation operators only or annihilation operators only. In
other words, this Hamiltonian no longer conserves the number of particles
because of the lack of the phase invariance a(~p) ei a(~p).
Bogoliubov found a way to diagonalize this Hamiltonian. Among creation
and annihilation operators, both a(~p) and a (~p) change the momentum of
the state by ~p, either by annihilating momentum p~ or creating momentum
~p. Because the number conservation is violated, creation and annihilation
operator can now mix, as long as they share the same momentum. Therefore,
we can consider the Boliubov transformation

b(~p) = a(~p) cosh + a(~p) sinh , (60)

b(~p) = a(~p) sinh + a(~p) cosh . (61)

The point is that the new operators defined this way also satisfy the same
commutation relation [b(~p), b (~q)] = p~,~q and can be regarded as new creation
and annihilation operators. By suitably choosing the parameter , we can
make Hamiltonian Eq. (59) not to have terms bb or b b . Choosing
~2
p
2m
+
cosh 2 = r  , sinh 2 = r  , (62)
~2
p ~2
p ~2
p ~2
p
2m 2m
+ 2 2m 2m
+ 2

we obtain the Hamiltonian

v v !
p~2
u 2 ! !
X up~ p~2 1 1 u p~2 p~2
u
H= t + 2 b(~p) b(~p) + + t + 2 .
p
~
2m 2m 2 2m 2 2m 2m
(63)

16
We used the fact that the summation over p~ includes ~p and combined both
contributions to simplify the expression. The ground state of this Hamilto-
nian is clearly the state annihilated by the new annihilation operators b(~p),
and excitations are created by b(~p) . The excitation energy for the creation
operator b(~p) agrees with that obtained from the classical analysis Eq. (51).
How is the ground state b(~p)|gi = 0 related to the original Fock states?
It is easy to show that the unitarity operator
a(~
p) )
U (~p) = e(a(~p)a(~p)a(~p) (64)

relates two sets of operators

U (~p)a(~p)U (~p) = b(~p). (65)

Therefore, the state |g(~p)i annihilated by b(~p) is written as

a(~
p) )
|g(~p)i = U (~p)|0i = e(a(~p)a(~p)a(~p) |0i. (66)

This state is different from the coherent state because it does not give an
expectation value of the annihilation operator
q a(~p), but it has a pair-wise
~2 p
p ~2
condensate hg(~p)|a(~p)a(~p)|g(~pi = /2 2m ( 2m + ) 6= 0. Therefore, it is
fair to say that not only the zero mode is condensed in BoseEinstein conden-
sate, non-zero modes are also condensed when is sizable, i.e., p~2 /2m < .
This is precisely the momentum range where the naive formula for the occu-
2
pation number n(~p) = 1/(e(~p /2m) 1) is ill-defined (negative).
Another interesting point is that there is an additional negative constant
in the Hamiltonian Eq. (63). The variational ansatz for the full Hamiltonian
is Y
|f i |g(~p)i, (67)
p
~
q
where |f i with f = / is the coherent for the zero mode and |g(~p)i is the
Bogoliubov transformed ground state defined in Eq. (66). The constant term
contributes to the expectation value of the full Hamiltonian in the variational
method, and makes the variational state have lower energy than the number
eigenstate.

17