Structural Studies of Decaying Fluid Turbulence: Effect of Initial Conditions

Chirag Kalelkar
Centre for Condensed Matter Theory, Department of Physics,
Indian Institute of Science, Bangalore 560012, India.

We present results from a systematic numerical study of structural properties of an unforced,

incompressible, homogeneous, and isotropic three-dimensional turbulent fluid with an initial energy
spectrum that develops a cascade of kinetic energy to large wavenumbers. The results are compared
with those from a recently studied set of power-law initial energy spectra [C. Kalelkar and R. Pandit,
Phys. Rev. E, 69, 046304 (2004)] which do not exhibit such a cascade. Differences are exhibited
in plots of vorticity isosurfaces, the temporal evolution of the kinetic energy-dissipation rate, and
the rates of production of the mean enstrophy along the principal axes of the strain-rate tensor.
A crossover between non-cascade-type and cascade-type behaviour is shown numerically for a
specific set of initial energy spectra.

PACS numbers: 47.27.Gs

I. INTRODUCTION large wavenumbers; however, structural properties had

not been probed.
Studies of structural properties (in space) of a tur- In this paper, we present results from a pseudospectral
bulent fluid may lead to an understanding of dynami- direct-numerical simulation (DNS) of the unforced, in-
cally significant processes in turbulence. Homogeneous compressible, three-dimensional Navier-Stokes equations
and isotropic turbulence, within the context of statis- and contrast spatial results obtained from an initial en-
tically steady flows, has been the subject of extensive ergy spectrum that develops a cascade (Section II), with
laboratory[1] and numerical[24] studies. These studies results from the power-law initial spectrum (Section III)
show that the turbulent fluid has a kinetic energy spec- mentioned above. Plots of vorticity isosurfaces are found
trum with a wavenumber range that exhibits a power-law to differ, with distinct filaments in regions of intense vor-
with exponent equal to 5/3, in accordance with a phe- ticity, lacking in the case with power-law initial spectrum.
nomenological theory due to Kolmogorov[5, 6]. In the The temporal evolution of the kinetic energy-dissipation
statistical steady state, visualization of the vorticity field rate and the rates of production of the mean enstrophy
reveals that regions of intense vorticity are organized in along the principal axes of the strain-rate tensor, are also
slender filaments[14]. The vorticity vector is observed found to differ. However, preferential alignment of the
to preferentially align[7] with the eigenvector associated vorticity vector with the intermediate eigenvector of the
with the intermediate eigenvalue of the strain-rate ten- strain-rate tensor is found to be unchanged. We also
sor of the fluid. By contrast, systematic spatial studies compare some classical results on mean enstrophy pro-
of decaying turbulence are scarce and restricted to nu- duction in decaying turbulence. We show numerically
merical studies[8, 9] of vorticity isosurfaces with initial that a crossover between the above-mentioned cases may
energy spectra of the type that develop a cascade of ki- be realised with an appropriate choice of initial condi-
netic energy to large wavenumbers. tions.
The Navier-Stokes equations in vorticity form are
A recent study[10], investigated the decay of unforced,
incompressible, homogeneous, and isotropic three- Di 2 i
dimensional magnetohydrodynamic turbulence from = Sij j + , (1)
power-law initial conditions. The study was a general- Dt xj xj
ization of results[11] obtained for the corresponding fluid
where D/Dt t + vj j is the material derivative,
case. In particular, it was shown both analytically and
is the kinematic viscosity, Sij 1/2(j vi + i vj ) is the
numerically, that for the power-law initial energy spec-
strain-rate tensor, and i ijk j vk is the vorticity (ijk
trum E(k, t0 ) k (k = |k| is the magnitude of the wave
is the Levi-Civita tensor), i, j, k = 1, 2, 3, with a sum-
vector and t = t0 is the choice of virtual origin of time),
mation implicit over repeated indices. We enforce the
the kinetic energy E(t) was found to decay as t1 and
incompressibility condition i vi = 0.
the integral length scale L(t) was found to grow as t0.5 .
Such power-law initial conditions are of interest in the
astrophysical context of the decay of primordial energy
II. CASCADE-TYPE SPECTRUM
spectra[12]. It was shown numerically[10], that such an
initial energy spectrum does not develop a cascade to
A. Numerical Method

We use a pseudospectral method[13] to solve Eqs. (1)

Electronic address: [email protected] numerically, in a cubical box of side 2 with periodic
 2

boundary conditions and 1923 Fourier modes. We do

2
10

not address issues pertaining to the scaling of higher- 0

10
order structure functions of velocity differences and be-
lieve that our spectral resolution is adequate for the types 10
2

of studies that we have carried out. For the tempo-

k5/3 E(k)
ral evolution, we use an Adams-Bashforth scheme (step 10
4

size t = 103 ) with double-precision arithmetic and set

= 105 (we include hyperviscosity h = 106 ). The 10
6

2
initial velocity field is taken to be v(k, t0 ) k 2 ek eik , 8
10
with k random variables distributed uniformly between
0 and 2. This corresponds to an initial energy spectrum 10
2 10
E(k, t0 ) k 4 e2k (with E(k, t) |v(k, t)|2 ), which
0 1 2
10 10 10
k
is a convenient choice that develops a cascade to large 1
wavenumbers (see below). We measure time in units of
0 0.9
the initial large-eddy turnover
P time 0 21/2 2/vrms (here
0
0 equals 4.23), vrms [h k |v(k, t0 )| i] is the root- 0.8

mean-square value of the initial velocity, with the dimen- 0.7

sionless time t/0 (t is the product of the number of

E / E0
0
steps and t). We define Re0 2vrms / to be the value 0.6

of the initial box-size Reynolds number (here Re0 equals 0.5

932965). Our results are obtained for times t0 t << t ,
0.4
where t is thePtime at which P the (growing) integral
scale L(t) h( k |v(k, t)|2 /k)/ k |v(k, t)|2 i becomes 0.3

of the order of the linear size of the simulation box. For 0.2
times t & t , finite-size effects which might well be non- 0 0.5

1 1.5

universal, modify the numerical results, and are not con- 50

sidered here. 45

In FIG. 1, we show some preliminary results that serve 40

as a check of our numerical method and parameter val- 35

ues (which were chosen to ensure linear stability of the 30

numerical scheme). Figure 1(a) shows on a log-log plot,
/ 0

25
the scaled kinetic energy spectrum k 5/3 E(k, ) as a func- 20
tion of the wavenumber k. On starting with the spec-
15
trum specified above, a cascade of energy is seen to large
10
wavenumbers. The plots are equispaced in time with a
temporal separation of = 0.24. The plot with open cir- 5

cles is calculated at cascade completion at dimensionless 0

0 0.5 1 1.5

time = c = 0.71, and shows a wavenumber range (for
1 . k . 10) that exhibits the well-known 5/3 power-
FIG. 1: (a) Log-log plot of the temporal evolution of the
law[8]. Upon cascade completion, the shape of the en-
scaled kinetic energy spectrum k5/3 E(k, ) as a function of
ergy spectrum does not change appreciably (except at
the wavenumber k at temporal separations of = 0.24, with
large wavenumbers where it falls), but the kinetic energy 2
E(k, t0 ) k4 e2k . The plot with open circles is calculated
decays. In FIG. 1(b), we plot the normalized kinetic at cascade completion, at dimensionless time = c = 0.71.
energy E( )/E0 as a function of the dimensionless time (b) Plot of the normalized kinetic energy E( )/E0 as a func-
, which is found to decay monotonically[8]. In FIG. 2
tion of the dimensionless time , with E(k, t0 ) k4 e2k .
1(c), we plot the normalized kinetic energy-dissipation (c) Plot of the normalized kinetic energy-dissipation rate
rate ( )/0 [(t) k k 2 |v(k, t)|2 ] as a function of the
P
( )/0 as a function of the dimensionless time , with
dimensionless time . The kinetic energy-dissipation rate 2
E(k, t0 ) k4 e2k .
peaks[8, 9] at = c , corresponding to cascade comple-
tion in the energy spectrum, and decreases thereafter.
The turbulence may be considered as fully developed at
= c and our spatial results (see below) will be calcu- B. Vorticity
lated at this instant of time.
The dynamics of vortex structures in a turbulent flow-
field is governed by the response of the vorticity vector to
the strain-rate tensor. In FIG. 2(a), we plot the normal-
ized probability distribution P (||) of || at cascade com-
pletion. The distribution is found to peak at || = 1.5
 3

0
( denotes the standard deviation) and has an expo- 10

nential tail for || > 1.5 with a fit P (||) e|| ,

= 0.05 0.01, = 1.00 0.01 (error-bars from a 1

10

least-squares fit). In FIG. 2(b), we plot iso-|| surfaces

for the isovalue || = h||i (angular brackets denote a 2
10

P (||)
volume average) at cascade completion, which appear
to be crumpled sheet-like structures (found throughout 3
10
the isovalue range [h||i , h||i + ]). At early times
<< c , regions of intense vorticity (with the isovalue 4
10
|| = h||i + 2) were found to be sheet-like (see FIG.
3(a)). It is plausible that a small volume fraction[14] of 5
10
such structures, subject to shear and stretching, roll-up 0 2 4 6 8 10 12 14
|| /
to form the filaments visualised in FIG. 3(b) at cascade
completion. The filaments are found to be of diameter of
the order of the grid spacing, and have a contour length
that often extends nearly to the linear size of the sim-
ulation box[15]. We choose to quote dimensions of the
structures relative to the (fixed) box-size and the grid-
spacing, since both the dissipative and the integral length
scales vary in time, in decaying turbulence. Iso-|| sur-
faces in the range || < (h||i ) were not found to
exhibit any particular structure at cascade completion.
The above results are in accordance with earlier studies
of statistically steady[14] and decaying[8, 9] turbulence.

FIG. 2: (a) Semilog plot of the normalized probability distri-

C. Strain-Rate Tensor bution P (||) of || ( is the vorticity) at cascade completion,
2
with E(k, t0 ) k4 e2k . The dashed-line plot is a normalized
Gaussian distribution for comparison.
At each grid point, we compute the eigenvalues 1 ,
(b) Plot of iso-|| surfaces for the isovalue || = h||i at cas-
2 and 3 (with the convention 1 2 3 ) of the 2
cade completion, with E(k, t0 ) k4 e2k .
strain-rate tensor Sij , and the corresponding orthonor-
mal eigenvectors e1 , e2 , and e3 . In a constant-density
flow, incompressibility requires that 1 + 2 + 3 = 0,
with 1 > 0, 3 < 0, the sign of the intermediate eigen- and the preferential alignment of with e2 is consistent
value 2 being indeterminate. In FIG. 4(a), we plot the with corresponding observations in statistically steady[3]
normalized probability distribution P (i ) of the eigen- and decaying[7] turbulence.
values i at cascade completion. We find that 2 has
a positive mean. The statistically preferred ratio of the
D. Mean Enstrophy Production Rate
mean strain-rates h1 i : h2 i : h3 i was found to equal
4.9 : 1 : 5.9 at cascade completion[16].
In FIG. 4(b), we plot the normalized probability dis- In a classical work[17], Betchov derived an equation
tribution of cosine of the angle between and the eigen- for the rate of production of the mean enstrophy h 2 i[18]
vectors ei , at cascade completion. We observe that is in unforced, incompressible, homogeneous, and isotropic
preferentially aligned (or antialigned) with eigenvector e2 turbulence, viz.
(a peak-to-valley ratio 5.1 was found) corresponding * 2 +
to eigenvalue 2 of minimum relative mean strain-rate h 2 i 2 vi
= h1 2 3 i , (2)
(disregarding the sign in the mean ratio, which merely t 4 xj xk
indicates an extensional or compressional strain on the
fluid element). The vorticity is also found to be prefer- where the angular brackets denote a volume average.
entially perpendicular to eigenvector e3 corresponding to From Eq. (2), it is evident that production of h 2 i re-
the principal compressive eigenvalue 3 , while no specific quires a predominantly negative value of 1 2 3 in the
angular relationship is observed relative to eigenvector fluid. In FIG. 5(a), we plot the normalized probabil-
e1 . We could confirm an earlier result[3] that the prefer- ity distribution P (1 2 3 ) of 1 2 3 at cascade com-
ential alignment of with e2 is already pronounced dur- pletion and find that the distribution has a negative
ing the cascade process, much prior to the appearance mean (with a skewness equal to 7.83), a result that
of distinct filaments in regions of intense vorticity. The has already been noted in an early numerical study[19].
positive mean of 2 , the ratio of the mean strain-rates, Sheet-like vortex structures are expected to predominate
 4

0.6
3
2 1

0.4

P( i ) i
0.2

0
15 10 5 0 5 10 15
i / i

1.8

1.6

1.4

1.2
e3

P [ cos ( , e ) ]
i
1

0.8

e
0.6 1

0.4

0.2 e
2

0
1 0.5 0 0.5 1
cos ( , ei)

FIG. 3: (a) Plot of iso-|| surfaces for the isovalue || = FIG. 4: (a) Plot of the normalized probability distribution
h||i + 2 ( is the standard deviation) at dimensionless time P (i ) of the eigenvalues i of the strain-rate tensor Sij , i =
2 2
<< c , with E(k, t0 ) k4 e2k . 1, 2, 3, at cascade completion, with E(k, t0 ) k4 e2k .
(b) Plot of iso-|| surfaces for the isovalue || = h||i + 2 at (b) Plot of the normalized probability distribution of cosine of
2
cascade completion, with E(k, t0 ) k4 e2k . the angle between and the eigenvectors ei of Sij at cascade
2
completion, with E(k, t0 ) k4 e2k .

if h1 2 3 i < 0[17, 20], which accords with FIG. 2(b).

Alternatively, h1 2 3 i > 0 would imply a predominance isfied to a greater degree during the cascade process, and
of filamentary vortex structures. In FIG. 5(b), we plot for dimensionless times & c asymptotes to the value
h1 2 3 i as a function of the dimensionless time . We 0.57. Dynamical effects and homogeneity suggest an
find that the magnitude of h1 2 3 i peaks at cascade inequality that is stronger than the one due to Betchov.
completion ( = 0.71), which is consistent with the di-
mensionless time at which the kinetic energy-dissipation
rate is a maximum (cf. FIG. 1(c)). In homogeneous tur- III. POWER-LAW SPECTRUM
bulence, the term h1 2 P 3 i can be shown[17] to equal
the expression hi j Sij i = i hi ( ei )2 i. In FIG. 5(c), A. Numerical Method
we plot hi ( ei )2 i, the rates of production of the mean
enstrophy along eigenvectors ei , and find that the pro- We use the numerical scheme, spectral resolution, step
duction rate is largest along eigenvector e2 correspond- size, viscosities, and boundary conditions as specified in
ing to eigenvalue 2 of minimum relative mean strain-rate Section II(A). The initial velocity field is taken to be
(see above), a remarkable result that appears to be new. v(k, t0 ) k 1/2 eik , with k random variables distributed
From FIG. 5(b) we observe that the mean enstro- uniformly between 0 and 2, which corresponds to the
phy production rate h1 2 3 i has an upper bound. In initial energy spectrum E(k, t0 ) k (with E(k, t)
the same work[17], Betchov invoked formal mathemati- |v(k, t)|2 ). Here, the dimensionless time 0 equals 13.9
cal inequalities and incompressibility to derive the upper and the initial box-size Reynolds number Re0 equals
bound
350181. Our results are obtained for times t0 t << t ,
1 and we choose to calculate our spatial results at dimen-
|h1 2 3 i| h(21 + 22 + 23 )3/2 i. (3) sionless time = c = 0.71[22] which is equal to the di-
3 6
mensionless time at which the kinetic energy-dissipation
In FIG. 6, we plot inequality (3)[21] as a function of the rate peaks, on starting with a cascade-type spectrum as
dimensionless time and find that the inequality is sat- shown in FIG. 1(c).
 5

1
10

0
10

0.6
1
10
P( )
1 2 3

2
10

B
3 0.4
10

4
10

5
10 0.2
50 40 30 20 10 0 10 20 0 0.2 0.4 0.6 0.8 1 1.2 1.4
( 123 ) /
140

FIG.
6: Plot of Betchovs inequality B
120
3 6|h1 2 3 i|/h(21 + 22 + 23 )3/2 i 1 as a function of
2
100 the dimensionless time , with E(k, t0 ) k4 e2k (open
circles) and E(k, t0 ) k (asterisks).
< 123 >

80

60
E( )/E0 and the normalized kinetic energy-dissipation
40
rate ( )/0 , as a function of the dimensionless time
20
(plotted on a linear scale in order to compare with FIGs.
1(b) and (c)). The kinetic energy E( ) is found to decay
0
0 0.5 1 1.5
as a power-law (on a log-log plot) with an exponent equal

to 0.91 0.04, with error-bars from a least-squares fit.
120
i=1
i=2
The exponent is theoretically predicted[11] to equal 1,
100 i=3 and we believe the discrepancy is due to the low spectral
80
resolution of our DNS. The normalized kinetic energy-
dissipation rate ( )/0 does not exhibit a peak (cf. FIG.
60
1(c)) and decays monotonically.
< ( e ) >
2
i

40
i

20
B. Vorticity
0

20
In FIG. 8(a), we plot the normalized probability dis-
40
0 0.5 1 1.5
tribution P (||) of || ( is the vorticity) at = c .
The distribution peaks at || = 1.5 as in FIG. 2(a),
however it does not exhibit a stretched-exponential tail.
FIG. 5: (a) Semilog plot of the normalized probability dis- In FIG. 8(b), we plot iso-|| surfaces for the isovalue
tribution P (1 2 3 ) of 1 2 3 at cascade completion, with || = h||i at = c , which appear to be shredded sheet-
2
E(k, t0 ) k4 e2k . The dashed-line plot is a normalized like structures (observed throughout the isovalue range
Gaussian distribution for comparison. [h||i , h||i + ]). At early times << c , regions of
(b) Plot of h1 2 3 i (see Eq. (2)) as a function of the di-
2 intense vorticity (with the isovalue || = h||i + 2) were
mensionless time , with E(k, t0 ) k4 e2k . found to be structure-less blobs of length of the order
(c) Plot of the rates of production of the mean enstrophy of the grid spacing (see FIG. 9(a)), in contrast to the
hi ( ei )2 i along the eigenvectors ei of Sij , as a function of
2 sheet-like structures in FIG. 3(a). At = c , isosurfaces
the dimensionless time , with E(k, t0 ) k4 e2k . of intense vorticity (in FIG. 9(b)), appear to be roughly
ellipsoidal in shape with semiaxes of the order of the grid
spacing and differ markedly from the coherent filaments
in the cascade-type case shown in FIG. 3(b).
In FIG. 7(a), we show on a log-log plot, the kinetic en-
ergy spectrum E(k, ) as a function of the wavenumber
k. The plots are equispaced in time with a temporal
separation of = 0.24 and the plot with open circles is C. Strain-Rate Tensor
calculated at = c . The spectrum does not cascade to
large wavenumbers or exhibit a wavenumber range with In FIG. 10(a), we plot the normalized probability dis-
a 5/3 power-law at any stage of temporal evolution. In tribution P (i ) of the eigenvalues i of the strain-rate
FIGs. 7(b) and (c), we plot the normalized kinetic energy tensor Sij at = c . We find that 2 has a positive
 6

0 0
10 10

2
10 1
10

4
10
2
10

P (||)
E(k)

6
10
3
10
8
10

4
10
10
10

5
12 10
10 0 1 2 3 0 2 4 6 8 10
10 10 10 10
k || /

0.9

0.8

0.7

0.6
E / E0

0.5

0.4

0.3

0.2

0.1

0
0 0.5 1 1.5

0.9 FIG. 8: (a) Semilog plot of the normalized probability distri-

0.8
bution P (||) of || ( is the vorticity) at dimensionless time
= c , with E(k, t0 ) k. The dashed-line plot is a normal-
0.7
ized Gaussian distribution for comparison.
0.6
(b) Plot of iso-|| surfaces for the isovalue || = h||i at
0
/

0.5 = c , with E(k, t0 ) k.

0.4

0.3

0.2 antiparallel) to e2 (with a reduced peak-to-valley ratio

0.1 4.1) and perpendicular to e3 , while no specific angular
0
0 0.5 1 1.5
relationship is observed relative to e1 (though we note a
small probability of alignment at cos(, e1 )=1).

FIG. 7: (a) Log-log plot of the temporal evolution of the ki-

netic energy spectrum E(k, ) as a function of the wavenum-
ber k at temporal seperations of = 0.24, with E(k, t0 ) k. D. Mean Enstrophy Production Rate
The plot with open circles is calculated at dimensionless time
= c = 0.71. In FIG. 11(a), we plot the normalized probability dis-
(b) Plot of the temporal evolution of the normalized kinetic tribution P (1 2 3 ) of 1 2 3 at = c . As in FIG.
energy E( )/E0 as a function of the dimensionless time , 5(a), we find that the distribution has a negative mean
with E(k, t0 ) k. (a relatively smaller skewness equal to 3.90 was found).
(c) Plot of the normalized kinetic energy-dissipation rate
In FIG. 11(b), we plot the value of h1 2 3 i (see Eq.
( )/0 as a function of the dimensionless time , with
E(k, t0 ) k. (2)) as a function of the dimensionless time and find
that it decreases monotonically, in contrast to FIG. 5(b).
In FIG. 11(c), we plot hi (ei )2 i, the rates of production
of the mean enstrophy along eigenvectors ei as a function
mean. The statistically preferred ratio of the mean eigen- of the dimensionless time , and find that the production
values h1 i : h2 i : h3 i was found to equal 4.6 : 1 : 5.6, rates decrease monotonically along all the principal axis
which differs only marginally from that obtained in the directions, in remarkable contrast to FIG. 5(c). In FIG.
cascade-type case. In FIG. 10(b), we plot the normal- 6, we plot inequality (3) as a function of the dimension-
ized probability distribution of cosine of the angle be- less time , and find that the inequality monotonically
tween and the eigenvectors ei of Sij at = c . As in approaches the same asymptotic value 0.57 as in the
FIG. 4(b), we observe that is preferentially parallel (or cascade-type case.
 7

0.5

3 2 1

0.4

0.3

i
P ( ) /
i
0.2

0.1

0
10 5 0 5 10
/
i i
1.5

e3

P [ cos ( , ei) ]
e
1
0.5

e2

0
1 0.5 0 0.5 1
cos ( , e )
i

FIG. 9: (a) Plot of iso-|| surfaces for the isovalue || = FIG. 10: (a) Plot of the normalized probability distribution
h||i + 2 ( is the standard deviation), at dimensionless time P (i ) of the eigenvalues i of the strain-rate tensor Sij , i =
<< c , with E(k, t0 ) k. 1, 2, 3, at dimensionless time = c , with E(k, t0 ) k.
(b) Plot of iso-|| surfaces for the isovalue || = h||i + 2 at (b) Plot of the normalized probability distribution of cosine of
= c , with E(k, t0 ) k. the angle between and the eigenvectors ei of Sij at = c ,
with E(k, t0 ) k.

E. Crossover
isosurfaces, the temporal evolution of the kinetic energy-
dissipation rate, and the rates of production of the mean
From FIGs. 1(c) and 7(c), it is evident that some rep-
enstrophy along the principal axes of the strain-rate ten-
resentative set of initial energy spectra may be utilised to
sor. However, preferential alignment of the vorticity vec-
exhibit a crossover from non-cascade-type to cascade-
tor with the intermediate eigenvector of the strain-rate
type behaviour, with the kinetic energy-dissipation rate
tensor is found to be unchanged. A crossover between
serving as a diagnostic. q In particular, one may choose
non-cascade-type and cascade-type behaviour is shown
the set E(k, t0 ) k q ek with parameter q. In FIG. 12,
numerically for a specific set of initial energy spectra.
we plot ( )/0 as a function of the dimensionless time
Our study shows that decaying turbulence is strikingly
(with 0 calculated using the initial energy spectrum
different for different types of initial conditions. Initial
with q = 0.4). On varying q in the range [0.4, 0.8], we
conditions that lead to a cascade yield features similar to
find that ( )/0 exhibits a smooth crossover at q 0.6
those obtained in the statistically steady case. However,
from non-cascade-type to cascade-type behaviour.
power-law initial conditions that do not lead to an energy
cascade exhibit several qualitatively new features which
we have described above.
IV. CONCLUSIONS

To summarize, we have presented results from a sys- Acknowledgments

tematic numerical study of structural properties of an un-
forced, incompressible, homogeneous, and isotropic tur- The author thanks T. Kalelkar, R. Pandit, and R.
bulent fluid with an initial energy spectrum that devel- Govindarajan for discussions, D. Mitra for the code,
ops a cascade to large wavenumbers. The results are SERC (IISc) for computational resources, and CSIR (In-
contrasted with those from the power-law initial spec- dia) for financial support.
tra investigated in Refs. [10, 11], which do not exhibit
such a cascade. Differences are noted in plots of vorticity
 8

1
10

0
10

1
10
P( )
1 2 3

2
10

3
10

4
10

5
10
30 25 20 15 10 5 0 5
( 123 ) /
4

3.5

2.5
< >
1 2 3

1.5

0.5

0
0 0.5 1 1.5

5
i=1
i=2
4 i=3

2
< ( e )2 >
i

1
i

3
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

FIG. 11: (a) Semilog plot of the normalized probability distri-

bution P (1 2 3 ) of 1 2 3 , with E(k, t0 ) k. The dashed-
line plot is a normalized Gaussian distribution for comparison.
(b) Plot of h1 2 3 i (see Eq. (2)) as a function of the di-
mensionless time , with E(k, t0 ) k.
(c) Plot of the rates of production of the mean enstrophy
hi ( ei )2 i along the eigenvectors ei of Sij , as a function of
the dimensionless time , with E(k, t0 ) k.

[1] S. Douady, Y. Couder, and M. Brachet, Phys. Rev. Lett., (1991).

67, 983 (1991); E. Villermaux, B. Sixou, and Y. Gagne, [4] J. Jimenez, A. Wray, P. Saffman, and R. Rogallo, J. Fluid
Phys. Fluids, 7, 2008 (1995). Mech., 255, 65 (1993).
[2] E. Siggia, J. Fluid Mech., 107, 375 (1981); Z. She, E. [5] A. Kolmogorov, Dokl. Akad. Nauk. SSSR, 31, 538 (1941)
Jackson, and S. Orszag, Nature, 344, 226 (1990). [Proc. R. Soc. London, Ser. A, 434, 15 (1991)].
[3] A. Vincent and M. Meneguzzi, J. Fluid Mech., 225, 1 [6] A. Monin and A. Yaglom, Statistical Fluid Mechanics,
 9

4
q=0.40 [11] P. Ditlevsen, M. Jensen, and P. Olesen, Physica A, 342,
q=0.60
3.5 q=0.70 471 (2004).
q=0.80
[12] P. Olesen, Phys. Lett. B, 398, 321 (1997).
3
[13] S. Dhar, A. Sain, and R. Pandit, Phys. Rev. Lett., 78,
2.5 2964 (1997).
[14] In Ref. [4] it has been shown that filamentary structures
0
/

2
in regions of intense vorticity contain only a small per-
1.5 centage of the total enstrophy.
[15] In Ref. [9], filamentary structures are visualized in a 5123
1
DNS of decaying turbulence.
0.5
[16] We note that in Refs. [7], the strain-rate eigenvalue ratio
was found to equal 3 : 1 : 4 based on the most probable
0
0 0.1 0.2 0.3 0.4 0.5 values, as opposed to the mean.

[17] R. Betchov, J. Fluid Mech., 1, 497 (1956).
[18] The analogous equation for the mean-squared strain-rate
FIG. 12: Plot of the normalized kinetic energy-dissipation hs2 i hSij Sij i is redundant since hs2 i = h 2 i/2 in homo-
rate ( )/0 as a function of the dimensionless time , with geneous, incompressible flows. The probability distribu-
q
E(k, t0 ) kq ek , 0.4 q 0.8.
p
tion P (|s|), |s| Sij Sij was found to resemble P (||).
[19] R. Betchov, Phys. Fluids, 18, 1230 (1975).
[20] A. Townsend, Proc. R. Soc. London, Ser. A, 208, 534
edited by J. Lumley, (MIT Press, Cambridge, 1975), Vol. (1951).
2. [21] Betchov in fact states that the inequality (3) cannot
[7] W. Ashurst, A. Kerstein, R. Kerr, and C. Gibson, Phys. be measured by conventional methods (p. 499, Ref.
Fluids, 30, 2343 (1987); A. Tsinober, E. Kit, and T. [17]) and introduces a weaker inequality which he tests
Dracos, J. Fluid Mech., 242, 169 (1992). through hot-wire anemometry. Clearly, there is no diffi-
[8] K. Yamamoto and I. Hosokawa, J. Phys. Soc. Jpn., 57, culty in directly testing inequality (3) in a DNS.
1532 (1988). [22] At dimensionless time = c , the Reynolds number Re
[9] I. Hosokawa, S. Oide, and K. Yamamoto, J. Phys. Soc. were found to be of the same order of magnitude in both
Jpn., 66, 2961 (1997). cascade-type and power-law cases.
[10] C. Kalelkar and R. Pandit, Phys. Rev. E, 69, 046304
(2004).