a

r
X
i
v
:
0
9
0
1
.
3
2
1
8
v
1


[
c
o
n
d
-
m
a
t
.
o
t
h
e
r
]


2
1

J
a
n

2
0
0
9
Is the supersolid superuid?
Dimitar I. Pushkarov

Institute of Solid-State Physics,

Bulgarian Academy of Sciences,
Soa 1784, Bulgaria and
Fatih University, Istanbul 3400
An analysis of previous theories of superuidity of quantum solids is presented in
relation to the nonclassical rotational moment of inertia (NCRM) found rst in Kim
and Chan experiments. A theory of supersolidity is proposed based on the presence
of an additional conservation law. It is shown that the additional entropy or mass
uxes depend on the quasiparticle dispersion relation and vanish in the eective mass
approximation. This implies that at low temperatures when the parabolic part of the
dispersion relation predominates the supersolid properties should be less expressed.
PACS numbers: 67.80.Mg
Introduction
The experiments of Kim and Chan [1] breathed new life into the old idea of possible
superuidity of solids. A quantum solid possessing superuid properties has been called a
supersolid. Originally a supersolid should be a crystalline body where a nondissipative mass
current can occur. This should correspond to the superuid state of liquid helium (helium-II)
observed by Kapitsa, and explained theoretically rst by Landau. The superuidity is now
well studied and a number of eects has been found, predicted and explained. Between them
is the change of the rotational moment due to the fact that the superuid fraction cannot
be involved into rotation at velocities less than the critical one. The qualitative explanation
of such a behavior according to Landau is that at small velocities no excitations can be
generated. A successful hydrodynamical description is given by the so-called two-uid (or
two-velocity) hydrodynamics. From a mathematical point of view the new element in the

e-mail dipushk@issp.bas.bg
2
two-velocity hydrodynamic equations is the potentiality of the superuid velocity v
s
which
reads v
s
= with for the chemical potential in the frame where v
s
= 0. As a result,
a new vibrational mode, the second sound, appears. The phase transition into a superuid
state is well dened and the corresponding changes of the thermodynamic characteristics
are well investigated.
The quantum-mechanical consideration connects the superuidity with the Bose-Einstein
condensation (BEC). Later on, such kind of condensation in the momentum space was
observed in some gases as well. This is the reason to talk on a macroscopic quantum state
described by the condensate wave function.
In their works Kim and Chan have observed a nonclassical rotational moment (NCRM),
i.e. a rotational moment of inertia which changes its value with temperature in a way bearing
a resemble with the Kapitza experiments with rotating liquid helium. They argue that this
is enough to conclude that the body has been in a supersolid state and that the superuidity
has been nally observed in all three states of matter (gas, liquid and solid). Lately, the
term supersolid has become a synonym of a body with NCRM.
The rst reasonable question is whether the NCRM implies superuidity (supersolidity).
Is the supersolid state superuid or this is an evidence of a new phenomena, maybe more
interesting and famous than superuidity, but nevertheless, of dierent kind. The existing
experimental observations and theoretical analysis did not give an unambiguous answer yet.
Originally, the concept of supersolid appeared for a crystalline body inside which a nondis-
sipative (macroscopic) mass current can exist. First considerations (Penrose and Onsager,
Andreev and Lifshits, Legget, Chester etc.) had a crystalline bodies in mind. Defects in
such crystals are imperfections of the crystal lattice, or lattice with an ideal periodicity but
with a number of atoms less than the number of the lattice sites (Andreev-Lifshits). The
rst question is therefore if the experiments can be understood from such a point of view.
Most probably this is not the case.
Let us rst consider the validity of the Landau derivation of the critical velocity. In liquid
helium, the energy in the frame where the superuid velocity is zero can be written in the
form:
E = E
0
+P
0
v +
1
2
Mv
2
, P = P
0
+Mv (1)
3
The same relation for an elementary excitation (p) reads
E = (p) +pv +
1
2
Mv
2
(2)
where p is the momentum in the frame where v = 0. The least possible change in energy
due to the excitation created is (p) pv and should be negative in order to reduce the
energy of the system. This yields
v > (p)/p. (3)
It is worth noting that equation (1) is always valid because it follows directly from the
Galilean principle for macroscopic quantities. Relation (2) corresponds to the microscopic
characteristics of the elementary excitation. In a homogeneous and isotropic (Newtonian)
space these two relations coincide. However this is not the case in a crystalline solid where
quasiparticle states are classied with respect the quasimomentum, not momentum. Quasi-
momentum is simply a quantum number which apears due to the periodicity of the lattice.
There are no Galilean transformations for quasiparticle characteristics. The transformation
relations which replace the Galilean ones were derived in [2, 3, 4, 5]). The macroscopic
momentum (the mass ux) is not a mean value of the quasimomentum, but of the product
m

k
with k for the quasimomentum. In addition, phonons in crystals have zero momentum
and do not transfer mass in contrast to the phonons in liquids. All this implies that the
Landau criterion (3) does not work in crystalline bodies. Its quasiparticle analogue should
look like
v
1
>
m

k
= m
ln
k
(4)
or
mv <
ln
k
. (5)
But, there is still a question what is, say, the phonon qusimomentum in the co-moving
(with the superuid fraction) frame. In addition, m = 0 for acoustic phonons. To avoid any
misunderstanding, let us stress again that whatever the dispersion relation of the elementary
excitations and the spectrum classication parameter (momentum or quasimomentum), the
macroscopic uxes have to obey Galilean relation (1).
Next, it is very important that the conservation laws (which are the basis of the hydrody-
namics) exist only in an inertial laboratory frame. And this laboratory frame is privileged,
not Galilean (see more details in Ref. [5]).
4
If one considers Bose-condensation of quasiparticles then such a condensate in a crystal
must be characterized by a value of the quasimomentum in a privileged frame.
Finally, the particles or quasiparticles (say vacancions) undergoing Bose-condensation
should interact weakly enough. It was shown [6] that the vacancion gas is most ideal near
the middle of the energy band, not in the bottom.
It is seen, therefore, that the situation in a crystalline body is completely dierent com-
pared to liquids and gases.
Nevertheless, the rst hydrodynamical theory of the superuidity of solids [8] was devel-
oped in a close analogy with the Landau theory of helium II. Andreev and Lifshits introduced
two velocities for a normal and a superuid fraction of the solid and apply the potentiality
condition for the superuid velocity, v
s
= . They used the Galilean invariance, so in the
frame, where the superuid component is in rest (v
s
= 0), the energy per unit volume is:
E = v
s
2
/2 +p.v
s
+, j = v
s
+p (6)
where j is the momentum per unit volume equal to the mass ow while p is the momentum
in the frame with v
s
= 0,
= (S, , w
ik
) (7)
is the internal energy as a function of the entropy, density and the distortion (not symmetric)
tensor
w
ik
=
u
i
x
k
.
The tensor of small deformations is as usually equal to
u
ik
=
1
2
_
u
i
x
k
+
u
k
x
l
_
and its trace equals the relative variation of the volume
u
ii
= w
ii
= V/V
A new point is that this trace is now not connected to the density variation with the known
relation, i.e.
w
ii
=

(8)
In this notation,
d = TdS +
ik
w
ik
+d + (v
n
v
s
)dp. (9)
5
A standard procedure follows based on the conservation laws:
+ div j = 0,
j
i
t
+

ik
x
k
= 0. (10)

S + div (Sv
n
+q/T) = R/T, (R > 0) (11)
v
s
+ = 0. (12)
The unknown quantities
ik
, , q, R have to be determined so as to satisfy the redundant
energy conservation law:

E + div Q = 0. (13)
The time derivative of E reads:

E = T

S +
ik
u
k
x
k
div j div
_
v
s
2
2
j
_
+j
v
s
2
2
(14)
(j v
n
) v
ni

ik
x
k
+v
n
v
s
div j
= div
_
j
v
s
2
2
+STv
n
+v
n
(v
n
p)
_
+T(

S + div Sv
n
)
+
ik
u
k
x
i
+ (j v
n
)
_

v
s
2
2
_
v
n

v
ni

x
k
{
ik
v
si
v
sk
+v
si
p
k
+v
sk
p
i
+ [ +TS + (v
n
v
s
)p +]
ik
} div j.
Here, a term of the form v
ni

kl
w
kl
x
i
is neglected as cubic in normal motion. With the
aid of conservation laws the time derivative of energy was written in the form [8]:

E + div
__
v
s
2
2
+
_
j +STv
n
+v
n
(v
n
p) +q +(j v
n
)+
+ v
nk

ki

ik
u
k
_
=
= R +
ik
v
ni
x
k
+div (j v
n
) +
qT
T
+ (v
nk
u
k
)

ik
x
i
, (15)
This yields the following expressions for the uxes:

ik
= v
si
v
sk
+v
si
p
k
+v
nk
p
i
+ [ +TS + (v
n
v
s
)p +]
ik

ik
+
ik
, (16)
=
v
s
2
2
+ +. (17)
6
Q =
_
v
s
2
2
+
_
j +STv
n
+v
n
(v
n
p) +q
+ (j v
n
) +v
nk

ki

ik
u
k
(18)
and the dissipation function of the crystal is
R =
ik
v
ni
x
k
div (j v
n
)
qT
T
(v
nk
u
k
)

ik
x
i
(19)
We shall not write here the relations between
ik
, , q and (v
n
u) that follow from the
Onsager principle and the positivity of the dissipative function. The main consequence is
that with neglecting dissipation one has v
n
= u. The normal motion is therefore the motion
of lattice sites (which may not coincide with given atoms). The superuid ow could, hence,
be possible at a given (even not moving) lattice structure.
However, instead of (15) the time derivative

E can also be written in the form:

E + div
__
v
s
2
2
+
_
j +STv
n
+v
n
(v
n
p) +q +(j v
n
)+ (20)
+ v
nk

ki

ik
v
nk
_
=
= R +
ik
v
nk
x
i
+div (j v
n
) +
qT
T
+
ik

x
i
( u
k
v
nk
),
which leads to other expressions for uxes.
In this case the nondissipative theory yields:

E +div
__
v
s
2
2
+
_
j +STv
n
+v
n
(v
n
p) + (
v
2
s
2
)(j v
n
)+
+v
nk
_

ki
v
si
v
sk
+v
sk
p
i
+v
ni
p
k
(21)
[ +TS + (v
n
v
s
)p +]
ik
_
_
=
=
_

ki
v
si
v
sk
+v
sk
p
i
+ v
ni
p
k

[ +TS + (v
n
v
s
)p +]
ik
_
v
nk
x
i
+
+(
v
2
s
2
)div (j v
n
) +
ik
u
k
x
i
and hence,

ik
= v
si
v
sk
+v
si
p
k
+v
nk
p
i
+
_
+TS + (v
n
v
s
)p +
_

ik
(22)
= +
v
2
s
2
,
ik
= 0 !!! (23)
7
The procedure used is, therefore, not unique. The relation v
n
= u was not derived, but
presupposed. In fact, the consideration started as a three-velocity theory ( u, v
n
, v
s
) and
the identity u = v
n
follows from a condition the time derivative of the total energy be not
dependent on the

ik
x
k
which is not well grounded. Next, the conservation laws are written
in the system where v
s
= 0 and this is not the laboratory frame in which the lattice cites
are in their equilibrium positions.
That is why we turned to another approach based on our theory of the quasiparticle
kinetics and dynamics in deformable crystalline bodies [3, 4, 5, 7]. This theory works with
exact (in the frame of the quasiparticle approach) selfconsistent set of equations including
the nonlinear elasticity theory equation and a transport Boltzman-like equation valid in the
whole Brillouin zone of quasiparticles with arbitrary dispersion law. The theory is developed
for crystalline bodies subject to time-varying deformations and arbitrary velocities.
Partition Function and Thermodynamic Relations
Let us consider a gas of quasiparticles with dispersion law (k) at low temperatures,
when the frequency of normal processes is much larger that of the Umklapp processes, i.e.

1
n

1
U
.
The distribution function n
k
(k, r, t) corresponds to S
max
with conserved energy E, quasi-
particle density n, quasimomentum K and momentum (mass ow) j dened, respectively,
as:
S(r, t) =
_
s[n
k
] dk, (24)
where s[n
k
] = (1 +n
k
) ln(1 +n
k
) n
k
ln n
k
E(r, t) =
_

k
n
k
dk (25)
n(r, t) =
_
n
k
dk (26)
K(r, t) =
_
kn
k
dk (27)
j(r, t) = m
_

k
k
n
k
dk, dk =
1
(2)
3
dk
1
dk
2
dk
3
(28)
8
This yields
n
k
(k, r t) =
_
exp
_

k
V.k mW.(
k
/k)
T
_
1
_
1
with V, W and for Lagrangian multipliers. Varying S yields
TS = E V.KW.j n
= E TS V.KW.j n
and respectively
d = SdT K.dVj.dWnd
The nondissipative equations involve the following conservation laws:
n + div J = 0, J = j/m (29)
j
i
t
+

ik
x
k
= 0, (30)
K
i
t
+
L
ik
x
k
= 0, (31)

S + div F = 0,

E + div Q = 0 (32)
To second order with respect to velocities V and W one has:
J
i
= n
0
V
i
+n
ij
W
j
, (33)
where
n
ij
=
_
n
0
k

il
(k) dk,
il
(k) = m

2

k
k
i
k
l
(34)
The local drift velocity U = j/mn
0
is then
U
i
= V
i
+
n
il
n
0
W
l
(35)
The mass ux is, therefore, not collinear to any of velocities V and W. Analogously,
K
i
=
il
V
l
+mn
0
W
i
, (36)

il
=
_
k
i
k
l
n
k

k
dk =

2

V
i
V
l

T,,W

1
il
=

2
E
K
i
K
l

S,n,j
(37)
9
To second order in velocities the diagonal terms of the quasimomentum ux tensor coincide
with the thermodynamic potential (T, V, W, ) [3, 4, 5]:
L
ij
=
_
k
i

k
k
j
n
k
dk =
0

ij
(38)
and the momentum ux tensor has the form:

il
=
il
= T
_
ln(1 +n
0
k
)
il
(k) dk (39)
The energy ux is:
Q
i
(r, t) =
_

k

k
k
i
n
k
dk = W
0
V
i
+ (TS
il
+n
il
)W
l
(40)
where
S
il
=
_
s[n
0
k
]
il
dk (41)
and W
0
= E
0

0
is the enthalpy at V = W = 0. Hence, W
0
V is the energy ux known
from the classical hydrodynamics, and there are additional terms due to the supersolid
behavior. The full hydrodynamic system consists then of four equations:
n +ndiv U = 0, (42)
mn

U
i

il
x
l
= 0 (43)

is

sl
x
l
n

0
x
i
+n
2
(
il

il
)

W
l
= 0 (44)

E +W
0
div U+TS
_
S
il
S

n
il
n
_
W
l
x
i
= 0 (45)
where

il
=
ik
n
kl
/(n
0
)
2
(46)
Taking into account the thermodynamic identity:
dE = T dS +dn +V.dK+W.dj (47)
the energy conservation law can be replaced by entropy equation:

S +S div U+S
_
S
il
S

n
il
n
_
W
l
x
i
= 0 (48)
It is seen that the mass ux and the entropy ux have dierent velocities both in magnitude
and direction. This means that a mass ux without entropy transport can take place. This
implies an existence of a superuid density
s
.
10
In the case of quadratic dispersion law

il
= mn
0

ik
,
S
il
S
=
n
il
n
,
il
=
il
, j
i
=
il
K
l
(49)
and the additional entropy ux vanishes. This implies that the superuid eects should be
negligible at very low temperatures where the excitations with parabolic dispersion relation
predominate.
Let us now rewrite the hydrodynamic set in terms of Landau superuid theory in order
to see better the analogy.
Cubic crystal
Let us rst consider, for simplicity, a cubic crystal. Then, to the second order with respect
to velocities one has
J(r, t) = nV+ W, Q(r, t) = W
0
V+ (p +q)W (50)

ik
= p
ik
, L
ik
=
0

ik
, F = SV+p
T
W (51)
where
p(r, t) =
1
3
m
_ _

k
k
_
2
n
0
k
dk,
m
m

0
(52)
(r, t) =
1
3
m
_

2

k
k
2
n
0
k
dk,
m
m

n (53)
q(r, t) =
1
3
m
_

k

k
k
2
n
0
k
dk,
m
m

E
0
(54)
and the following relations take place:
p

=
_
p

_
T
= , p
T
=
_
p
T
_

=
p +q
T
(55)
The meaning of the quantities involved can be seen from their limiting expressions in the
eective mass (m

) approximation.
In the notation of Landau theory: F = SV
n
, V = V
s
and the system of equations
(4246) takes the form:
n +n
s
div V
s
+n
n
div V
n
= 0,

S +Sdiv V
n
= 0, (56)

K+ST +n = 0,
j
t
+p
T
T +p

= 0 (57)
11
where n
n
= S
p

p
T
, n
s
= n n
n
and the number density ux is
J = n
s
V
s
+n
n
V
n
(58)
Second Sound
One has in variables , T, V
s
, V
n

T + +n
s
.V
s
+n
n
.V
n
= 0

T + +S.V
n
= 0
ST +n +
s

V
s
+
n

V
n
= 0
nST +nn
n
=
n
n
s

V
s
+
n
n
n

V
n
= 0
where
=
n
T

=
S

T
, =
n

T
, =
S
T

(59)

n
= mnS/p
T
, =
1
3
_
k
2
n
0
(1 +n
0
) dk,
s
=
n
(60)
K = V +mnW =
s
V
s
+
n
V
n
(61)
For quasiparticles with quadratic dispersion law n
s
=
s
= 0:

T + +n.V
n
= 0 (62)

T + +S.V
n
= 0 (63)
ST +n +
s

V
s
+

V
n
= 0 (64)
If the number of quasiparticles is not conserved

2
0
(q) =
TS
C
v

q
2
(65)
If the number of quasiparticles is conserved

2
(q) =
2
0
(q)
_
_
1
n
S
_
2
+
C
v
n
2
TS
2
_
=
_
T
C
v
_
s
v
0
_
2
T

v
0
_
T
_
q
2

=
_
P
n
_
s
n

q
2
(66)
12
where s = S/n, v
0
= 1/n .
Quasiparticles with non-quadratic dispersion law
If the number of quasiparticles is not conserved

2
(q) = (1 +

)
TS
2
C
v

n
q
2
,

n
,

=

s

n
,
n
=
n
s
n
n
(67)
If the number of quasiparticles is conserved

2
1
(q) =
n
_

2
0
(q, =
n
)
1
n
n
TS
C
v
(P/T)
2
n
(P/n)
s
q
2
_
(68)

2
2
(q) = (1
n
)
2
(q, =
n
) +
n

2
0
(q, =
n
)
2
1
(q) (69)
2
n
TS

n
C
v
_
P
T
_
n
q
2

2
(q) >
1
(q) (70)
Conclusion
It is shown that the theory of superuidity of solids should not be a replica of the Lan-
dau theory of superuidity. For crystalline bodies a two-velocity theory of supersolidity is
presented with accounting of the quasimomentum conservation law. Such a theory cannot
be applied to disordered systems, glasses etc.
Acknowledgements
The author thanks Professor V.Kravtsov for the inviation to the Abdus Salam ICTP,
Trieste, where this work was submitted. Acknowledgements are also given regarding the
partial nancial support from the National Science Fund, Contract F-1517
[1] E. Kim and M.H.W. Chan, Nature, 427 225 (2004); E. Kim and M.H.W. Chan, Science 305
1941-44 (2004); E. Kim and M.H.W. Chan, J.Low Temp. Phys. 138 859 (2005)
[2] A.F.Andrev and D.I.Pushkarov, - Sov.Phys.JETP 62 (5) 1087-1090 (1985)
[3] D.I.Pushkarov, - Quasiparticle theory of defects in solids, World Scientic, Singapore, 1991
[4] D.I.Pushkarov, Defektony v kristallakh (Defectons in crystals - quasiparticle approach to quan-
tum theory of defects) in Russian, Nauka, Moscow 1993
13
[5] D.I. Pushkarov, - Phys.Rep. 354 411 (2001)
[6] D.I.Pushkarov, - Phys.Stat.Sol.(b) 133 525 (1986)
[7] D.I.Pushkarov and R. Atanasov, Phys.Scripta 42 481 (1990)
[8] A.F.Andreev and I.M.Lifshitz - Zh.Eksp.Teor.Fiz. 56 (12) 2057 (1969)