id
n
i a
rsit
4; p
stu
or
hin
nsis
important st
these experi
This can be
and the norm
mutual fricti
t�. The vor-
e-connected
d only by a
vanishes so
ex core size
PRL 94, 065302 (2005) P H Y S I C A L R E V I E W L E T T E R S week ending18 FEBRUARY 2005
conventional fluid [8]. Since the normal fluid is negligible
at very low temperatures, an important question arises:
even without the normal fluid, is ST still similar to CT or
not?
To address this question, we consider the statistical law
of CT [7]. The steady state for fully developed turbulence
is given by the healing length � � 1= g
. Although quan-
tized vortices at finite temperatures can decay through
mutual friction with the normal fluid, at very low tempera-
tures the vortices can decay only by two mechanisms; one
is the sound emission through vortex reconnections [11],
and the other is that vortices reduce through the Richardson
0031-9007=
atistical laws [7] of fully developed CT, so
ments show a similarity between ST and CT.
understood using the idea that the superfluid
al fluid are likely to be coupled together by the
on between them and thus to behave like a
superfluid velocity given by v�x; t� � 2r
�x;
ticity rotv�x; t� vanishes everywhere in a singl
region of the fluid; any rotational flow is carrie
quantized vortex in the core of which ��x; t�
that the circulation is quantized by 4�. The vort������p
Kolmogorov law. The Kolmogorov law is one of the most stant,
�x; t� is the condensate density, and v�x; t� is the
Kolmogorov Spectrum of Superflu
of the Gross-Pitaevskii Equatio
Michikazu Kobayash
Department of Physics, Osaka City Unive
(Received 30 November 200
The energy spectrum of superfluid turbulence is
equation. We introduce the dissipation term which w
to remove short wavelength excitations which may
inertial range. The obtained energy spectrum is co
DOI: 10.1103/PhysRevLett.94.065302
The physics of quantized vortices in liquid 4He is one of
the most important topics in low temperature physics [1].
Liquid 4He enters the superfluid state at 2.17 K. Below this
temperature, the hydrodynamics is usually described using
the two-fluid model in which the system consists of invis-
cid superfluid and viscous normal fluid. Early experimental
works on the subject focused on thermal counterflow in
which the normal fluid flowed in the opposite direction to
the superfluid flow. This flow is driven by the injected heat
current, and it was found that the superflow becomes
dissipative when the relative velocity between two fluids
exceeds a critical value [1]. Feynman proposed that this is a
superfluid turbulent state consisting of a tangle of quan-
tized vortices [2]. Vinen confirmed Feynman’s picture
experimentally by showing that the dissipation comes
from the mutual friction between vortices and the normal
flow [3]. After that, many studies on superfluid turbulence
(ST) have been devoted to thermal counterflow [4].
However, as thermal counterflow has no analogy with
conventional fluid dynamics, we have not understood the
relation between ST and classical turbulence (CT). With
such a background, experiments of ST that do not include
thermal counterflow have recently appeared. Such studies
represent a new stage in studies of ST. Maurer et al. studied
a ST that was driven by two counterrotating disks [5] in
superfluid 4He at temperatures above 1.4 K, and they
obtained the Kolmogorov law. Stalp et al. observed the
decay of grid turbulence [6] in superfluid 4He at tempera-
tures above 1 K, and their results were consistent with the
05=94(6)=065302(4)$23.00 06530
Turbulence: Numerical Analysis
with a Small-Scale Dissipation
nd Makoto Tsubota
y, Sumiyoshi-Ku, Osaka 558-8585, Japan
ublished 14 February 2005)
died numerically by solving the Gross-Pitaevskii
ks only in the scale smaller than the healing length
der the cascade process of quantized vortices in the
tent with the Kolmogorov law.
PACS numbers: 67.40.Vs, 47.37.+q, 67.40.Hf
of an incompressible classical fluid follows the
Kolmogorov law for the energy spectrum. The energy is
injected into the fluid at some large scales in the energy-
containing range. This energy is transferred in the inertial
range from large to small scales without being dissipated.
The inertial range is believed to be sustained by the self-
similar Richardson cascade in which large eddies are bro-
ken up into smaller ones, having the Kolmogorov law
E�k� � C�2=3k�5=3: (1)
Here the energy spectrum E�k� is defined as E � RdkE�k�,
where E is the kinetic energy per unit mass and k is the
wave number from the Fourier transformation of the ve-
locity field. The energy transferred to smaller scales in the
energy-dissipative range is dissipated by the viscosity with
the dissipation rate, which is identical with the energy flux
� of Eq. (1) in the inertial range. The Kolmogorov constant
C is a dimensionless parameter of order unity.
In CT, the Richardson cascade is not completely under-
stood, because it is impossible to definitely identify each
eddy. In contrast, quantized vortices in superfluid are defi-
nite and stable topological defects. A Bose-Einstein con-
densed system yields a macroscopic wave function
��x; t� � �������������
�x; t�p ei
�x;t�, whose dynamics is governed by
the Gross-Pitaevskii (GP) equation [9,10]
i
@
@t
��x; t� � ��r2 ��� gj��x; t�j2 ��x; t�; (2)
where � is the chemical potential, g is the coupling con-
2-1 2005 The American Physical Society
cascade process and eventually change to elementary ex- we used a spatial resolution �x � 0:125 and V � 323,
where the length scale is normalized by the healing length
�. With this choice, �k � 2�=32. Numerical time evolu-
tion was given by the Runge-Kutta-Verner method with the
time resolution �t � 1� 10�4.
To obtain a turbulent state, we start from an initial
configuration in which the condensate density
0 is uni-
form and the phase
0�x� has a random spatial distribution.
Here the random phase
0�x� is made by placing random
numbers between �� to � at every distance � � 4 and
connecting them smoothly (Fig. 1).
The initial velocity v�x; t � 0� � 2r
0�x� given by the
initial random phase is random; hence, the initial wave
function is dynamically unstable and soon produces homo-
geneous and isotropic turbulence with many quantized
vortex loops.
To confirm the accuracy of our simulation, we calculate
PRL 94, 065302 (2005) P H Y S I C A L R E V I E W L E T T E R S week ending18 FEBRUARY 2005
30
citations at the healing length scale. In any case, dissipation
occurs only at scales below the healing length. Therefore,
ST at very low temperatures, consisting of such quantized
vortices, can be a prototype to study the inertial range, the
Kolmogorov law, and the Richardson cascade.
There are two kinds of formulation to study the dynam-
ics of quantized vortices: one is the vortex-filament model
[12], and the other the GP model. By using the vortex-
filament model with no normal fluid component, Araki
et al. studied numerically a vortex tangle starting from a
Taylor-Green flow, thus obtaining the energy spectrum
consistent with the Kolmogorov law [13]. By eliminating
the smallest vortices whose size is comparable to the
numerical space resolution, they introduced the dissipation
into the system. Nore et al. used the GP equation to
numerically study the energy spectrum of ST [14]. The
kinetic energy consists of a compressible part due to sound
waves and an incompressible part coming from quantized
vortices. Excitations of wavelength less than the healing
length are created through vortex reconnections or through
the disappearance of small vortex loops [11,15], so that the
incompressible kinetic energy transforms into compress-
ible kinetic energy while conserving the total energy. The
spectrum of the incompressible kinetic energy is tempo-
rarily consistent with the Kolmogorov law. However, the
consistency becomes weak in the late stage when many
sound waves created through those processes hinder the
cascade process [16–18].
Our approach here is to introduce a dissipation term that
works only on scales smaller than the healing length �.
This introduction of the dissipation term supposes some
possible dissipation like the emission of elementary exci-
tations which can happen only in smaller scales than the
healing length � even at T � 0 K [11,15]. The cascade
process in larger scales must be independent of the detail of
the dissipative mechanism at small scales. This dissipation
removes not vortices but short wavelength excitations, thus
preventing the excitation energy from transforming back to
vortices. Compared to the usual GP model, this approach
enables us to more clearly study the Kolmogorov law.
To solve the GP equation numerically with high accu-
racy, we use the Fourier spectral method in space with a
periodic boundary condition in a box with spatial resolu-
tion containing 2563 grid points. We solve the Fourier
transformed GP equation
i
@
@t
~��k; t� � �k2 �� ~��k; t� � g
V2
X
k1;k2
~��k1; t� ~���k2; t�
� ~��k� k1 � k2; t�; (3)
where V is the volume of the system and ~��k; t� is the
spatial Fourier component of ��x; t� with the wave number
k, numerically given by a fast Fourier transformation [19].
We consider the case of g � 1. For numerical parameters,
065
the total energy E, the interaction energy Eint, the quantum
energy Eq, and the kinetic energy Ekin [14]:
E�
R
dx����r2�g=2j�j2 �R
dx
; Eint � g2
R
dxj�j4R
dx
;
Eq �
R
dx�rj�j 2R
dx
; Ekin �
R
dx�j�jr
2R
dx
; (4)
and compare them with those given by the dynamics with
the different time resolutions �t � 2� 10�5 or different
spatial resolutions 5123 grids. For the comparison, we
estimate the relative errors F12�A� � j�hAi1 � hAi2�=hAi1j
and F13�A� � j�hAi1 � hAi3�=hAi1j at t � 12 for A � E,
Eint, Eq, and Ekin, where h i1, h i2, and h i3 are, respectively,
the values given by three simulations with different reso-
lutions: (i) �t�1�10�4 and 2563 grids, (ii) �t �
2� 10�5 and 2563 grids, and (iii) �t � 1� 10�4 and
5123 grids. Comparison is shown in Table I; the relative
errors are extremely small, which allows us to use
resolution (i).
Furthermore, the total energy is conserved by the accu-
racy of 10�10 in our simulation.
-π
π
0 32
φ0(x,y=0,z=0)
random number
x
λ
-
0 32
32 0
(x,y,z=0)
x
y
φ π
π
(a) (b)
FIG. 1 (color online). How to make the random phase
0�x�
with � � 4 (a) and one example of an initial random phase in a
cross section
0�x; y; 0� (b).
2-2
We introduced a dissipation term in Eq. (3) to remove
excitations of a wavelength shorter than the healing length.
The imaginary number i in the left-hand side of Eq. (3)
was replaced by �i� ��k� , where ��k� � �0��k� 2�=��
with ��x� being the step function. The effect of this dis-
sipation is shown in Fig. 2. We divide up to the kinetic
energy Ekin into the compressible part Eckin �R
dx��j�jr
�c 2=R dx
, due to sound waves, and the
incompressible part Eikin �
R
dx��j�jr
�i 2=Rdx
, due
c i
This figure shows that our ST satisfies the Kolmogorov
law Eikin�k� / k�5=3 for times 4 & t & 10, when the system
may be almost homogeneous and isotropic turbulence. We
also calculated the energy dissipation rate � � �@Eikin=@t
and compared the results quantitatively with the
Kolmogorov law. Figure 3(b) shows that � is almost con-
stant in the period 4 & t & 10, which means that the
Kolmogorov spectrum �2=3k�� is also constant in the
period. Without the dissipation term ��k�, the ST could
not satisfy the Kolmogorov law. This means that dissipat-
ing short wavelength excitations are essential for satisfying
the Kolmogorov law. Figure 3(c) shows the time develop-
ment of the exponent � for �0 � 0. In this case, � is
consistent with the Kolmogorov law only in the short
period 4 & t & 7; active sound waves affect the dynamics
of quantized vortices for t * 7. Such an effect of sound
waves is also clearly shown in Fig. 3(d). Compared with
Fig. 3(b), � is unsteady and smaller than that of �0 � 1 and
even becomes negative at times. When � is negative, the
energy flows backward from sound waves to vortices,
which may prevent the energy spectrum from satisfying
the Kolmogorov law for t * 7. According to Fig. 3(a), the
agreement between the energy spectrum and the
Kolmogorov law becomes weak at a later stage t * 10,
which may be attributable to the following reasons. In the
period 4 & t & 10, the energy spectrum agrees with the
Kolmogorov law, which may support that the Richardson
cascade process works in the system. The dissipation is
TABLE I. Dependence of F12 and F13 on E, Eint, Eq, and Ekin.
F12�A� F13�A�
A � E 2:4� 10�12 6:3� 10�10
A � Eint 3:7� 10�12 8:8� 10�10
A � Eq 2:6� 10�12 6:9� 10�10
A � Ekin 5:1� 10�12 9:4� 10�10
PRL 94, 065302 (2005) P H Y S I C A L R E V I E W L E T T E R S week ending18 FEBRUARY 2005
to vortices, where rot�j�jr
� � 0 and div�j�jr
� � 0
[14,15]. Figure 2 shows the time development of E, Ekin,
Eckin, and Eikin in the cases of 2(a) �0 � 0 and 2(b) �0 � 1.
Without dissipation, the compressible kinetic energy
Eckin is increased in spite of conserving the total energy
[Fig. 2(a)], which is consistent with the simulation by Nore
et al. [14]. The dissipation suppresses Eckin and thus causes
Eikin to be dominant. This dissipation term works as a
viscosity only at the small scale. At other scales, which
are equivalent to the inertial range �k < k < 2�=�, the
energy does not dissipate.
We calculated the spectrum of the incompressible ki-
netic energy Eikin�k� defined as Eikin �
R
dkEikin�k�.
Initially, the spectrum Eikin�k� significantly deviates from
the Kolmogorov power law; however, the spectrum ap-
proaches a power law as the turbulence develops. We
assumed that the spectrum Eikin�k� is proportional to k��
in the inertial range �k < k < 2�=�, and determined the
exponent � by making a straight line fit to a log-log plot of
the spectrum Eikin�k� at each time. The time development of
� is shown in Fig. 3(a).
5
10
15
E
Ekin
Ekin
c
Ekin
i
5
10
15
E
Ekin
Ekin
c
Ekin
i
(a) (b)
00 1 2 3 4 5 6
t
00 1 2 3 4 5 6
t
FIG. 2. Time development of E, Ekin, Eckin, and Eikin. (a) Case
with �0 � 0; (b) case with �0 � 1.
06530
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8
0 2 4 6 8 10 12
η
5/3
η
t
-2
-1
0
1
0 2 4 6 8 10 12
ε
t
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8
0 2 4 6 8 10 12
η
5/3
η
t
-2
-1
0
1
0 2 4 6 8 10 12
ε
t
(a) (b)
(c) (d)
FIG. 3. Time dependence of the exponent of the spectrum �
(a) and the dissipation rate �. (a) � for �0 � 1. (b) � for �0 � 1.
(c) � for �0 � 0. (d) � for �0 � 0. In (a) and (c), the error bars
are the standard deviations of the data from the fit. The line � �
5=3 is shown to compare the results with the Kolmogorov law.
2-3
the energy and broadening the inertial range of the energy
spectrum. We will report on these results in near future.
We acknowledge W. F. Vinen for useful discussions.
M. T. and M. K. both acknowledge support by a Grant-in-
Aid for Scientific Research (Grants No. 15340122 and
No. 1505983, respectively) by the Japan Society for the
Promotion of Science.
[1] R. J. Donnelly, Quantized Vortices in Helium II
(Cambridge University Press, Cambridge, 1991).
[2] R. P. Feynman, in Progress in Low Temperature Physics,
edited by C. J. Gorter (North-Holland, Amsterdam, 1955),
0
32
32
32
x
y
z
-5
-4.5
-4
-3.5
-3
-2.5
-2
-0.5 0 1.1
Ekin
c(k)
Ekin
i(k)
Cε2/3k-5/3
(C=1)
L
og
E
(k
)
Log k
k=2 π/ξ
(a) (b)
FIG. 4. Vorticity rotv�x; t� and the energy spectrum at t � 6:0.
The contour in (a) is 98% of maximum vorticity. The energy
PRL 94, 065302 (2005) P H Y S I C A L R E V I E W L E T T E R S week ending18 FEBRUARY 2005
caused mainly by removing short wavelength excitations
emitted at vortex reconnections. However, the system at
the late stage t * 10 has only small vortices after the
Richardson cascade process, being no longer turbulent.
The energy spectrum, therefore, disagrees with the
Kolmogorov law of � � 5=3. And emissions of excitations
through vortex reconnections hardly occur, which reduces
greatly the energy dissipation rate �.
Spatially, the system appears to develop full turbulence.
Shown in Fig. 4(a) is the spatial distribution of vortices.
The vorticity plot suggests fully developed turbulence. The
energy spectrum in Fig. 4(b) agrees quantitatively with the
Kolmogorov law. We thus conclude that the incompress-
ible kinetic energy of ST without the effect of short wave-
length excitations satisfies the Kolmogorov law.
This inertial range 0:20 & k & 6:3 is larger than those in
the simulations by Araki et al. [13] and Nore et al. [14].
The Kolmogorov constant is estimated to be C ’ 10�0:5 �
0:32, being just smaller than that in classical turbulence.
In conclusion, we investigate the energy spectrum of ST
by the numerical simulation of the GP equation.
Suppressing the compressible sound waves which are cre-
ated through vortex reconnections, we can make the quan-
spectrum was obtained by making an ensemble average for 20
different initial states The solid line refers to the Kolmogorov
law in Eq. (1).
titative agreement between the spectrum of incompressible
kinetic energy and Kolmogorov law. Our next subjects are
making the steady turbulence by introducing injection of
06530
2-4
Vol. I, p. 17.
[3] W. F. Vinen, Proc. R. Soc. London A 240, 114 (1957); 240,
128 (1957); 240, 493 (1957).
[4] J. T. Tough, in Progress in Low Temperature Physics,
edited by C. J. Gorter (North-Holland, Amsterdam,
1955), Vol. VIII, p. 133.
[5] J. Maurer and P. Tabeling, Europhys. Lett. 43, 29 (1998).
[6] S. R. Stalp, L. Skrbek, and R. J. Donnelly, Phys. Rev. Lett.
82, 4831 (1999).
[7] U. Frisch, Turbulence (Cambridge University Press,
Cambridge, 1995).
[8] W. F. Vinen, Phys. Rev. B 61, 1410 (2000).
[9] E. P. Gross, J. Math. Phys. (N.Y.) 4, 195 (1963).
[10] L. P. Pitaevskii, Sov. Phys. JETP 13, 451 (1961).
[11] M. Leadbeater, T. Winiecki, D. C. Samuels, C. F.
Barenghi, and C. S. Adams, Phys. Rev. Lett. 86, 1410
(2001).
[12] K. W. Schwarz, Phys. Rev. B 31, 5782 (1985); 38, 2398
(1988)
[13] T. Araki, M. Tsubota, and S. K. Nemirovskii, Phys. Rev.
Lett. 89, 145301 (2002).
[14] C. Nore, M. Abid, and M. E. Brachet, Phys. Rev. Lett. 78,
3896 (1997); Phys. Fluids 9, 2644 (1997).
[15] S. Ogawa, M. Tsubota, and Y. Hattori, J. Phys. Soc. Jpn.
71, 813 (2002).
[16] N. G. Berloff, Phys. Rev. A 69, 053601 (2004).
[17] N. G. Berloff and B. V. Svistunov, Phys. Rev. A 66,
013603 (2002).
[18] J. Koplik and H. Levine, Phys. Rev. Lett. 71, 1375 (1993);
76, 4745 (1996).
[19] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P.
Flannery, Numerical Recipes in Fortran 77 (Cambridge
University Press, Cambridge, 1985).