NORDITA-2016-87
How long do particles spend in vortical regions in turbulent flows?
Akshay Bhatnagar,1, 2, ∗ Anupam Gupta,3, † Dhrubaditya Mitra,2, ‡ Rahul Pandit,1, § and Prasad Perlekar4, ¶
1
arXiv:1609.02601v1 [physics.flu-dyn] 8 Sep 2016
Centre for Condensed Matter Theory, Department of Physics,
Indian Institute of Science, Bangalore 560012, India.
2
Nordita, KTH Royal Institute of Technology and Stockholm University, Roslagstullsbacken 23, 10691 Stockholm, Sweden
3
Laboratoire de Gnie Chimique, Universite de Toulouse, INPT-UPS, 31030, Toulouse, France.
4
TIFR Centre for Interdisciplinary Sciences, 21 Brundavan Colony, Narsingi, Hyderabad 500075, India
We obtain the probability distribution functions (PDFs) of the time that a Lagrangian tracer
or a heavy inertial particle spends in vortical or strain-dominated regions of a turbulent flow, by
carrying out direct numerical simulation (DNS) of such particles advected by statistically steady, homogeneous and isotropic turbulence in the forced, three-dimensional, incompressible Navier-Stokes
equation. We use the two invariants, Q and R, of the velocity-gradient tensor to distinguish between
vortical and strain-dominated regions of the flow and partition the Q − R plane into four different
regions depending on the topology of the flow; out of these four regions two correspond to vorticitydominated regions of the flow and two correspond to strain-dominated ones. We obtain Q and R
along the trajectories of tracers and heavy inertial particles and find out the time tpers for which
they remain in one of the four regions of the Q − R plane. We find that the PDFs of tpers display
exponentially decaying tails for all four regions for tracers and heavy inertial particles. From these
PDFs we extract characteristic times scales, which help us to quantify the time that such particles
spend in vortical or strain-dominated regions of the flow.
PACS numbers: 47.27.-i,47.55.Kf,05.40.-a
Keywords: vortical regions; persistence time
I.
INTRODUCTION
The characterization of the statistical properties of
particles advected by a turbulent flow is a challenging
problem. Not only is it of fundamental interest in fluid
mechanics and non-equilibrium statistical mechanics, but
it also has applications in geophysical fluid dynamics
(e.g., raindrop formation in warm clouds [1–4]) and astrophysics (e.g., planet formation in astrophysical disks
[5, 6]). An important challenge here is to obtain the time
that such advected particles spend in vortical regions of
the flow. We build on our studies of persistence-time
statistics in two-dimensional (2D) fluid turbulence [7] to
develop a natural way of defining a time for which a particle stays in a vortical region in the three-dimensional
(3D) case. We illustrate how this is done for the case of
statistically steady, homogeneous, and isotropic fluid turbulence by studying turbulent advection of (a) neutrally
buoyant Lagrangian tracers (henceforth called tracers),
which move with the fluid velocity at the particle, and
(b) passive, heavy, inertial particles (henceforth heavy
particles), which are spherical particles that are heavier
than the carrier fluid and smaller than the Kolmogorov
length scale η, at which viscous dissipation becomes significant. The trapping of a tracer into a vortical region
is expected to give rise to very high values of particle ac-
∗ Electronic
address:
address:
‡ Electronic address:
§ Electronic address:
¶ Electronic address:
† Electronic
akshayphy@gmail.com
anupam@physics.iisc.ernet.in
dhruba.mitra@gmail.com
rahul@physics.iisc.ernet.in
perlekar@tifrh.res.in
celeration [8, 9]. The heavy particles are ejected from
vortices [10–15] hence they are preferentially found in
strain-dominated regions of the flow. This has been observed in direct numerical simulations (DNSs) by overlaying the positions of these particle on a pseudo-color plot
of the magnitude of the vorticity, in a two-dimensional
slice [16–18] through the simulation domain.
We estimate the time that a tracer or a heavy particle spends in a vortical or strain-dominated region of
the flow by using the following, well-established technique for distinguishing between these flow regions [19–
21]: At any point in the flow, the velocity-gradient matrix A has two invariants Q and R [19, 20] (in the incompressible case, that we consider, the trace of A is
zero everywhere). Depending upon the signs of R and
∆ = (27/4)R2 + Q3 , we can divide the Q − R plane into
four regions (Fig. 1); in two of these regions two eigenvalues of A are complex conjugates of each other; and
the topology of the local flow is vortical. The other two
regions of the Q−R plane corresponds to those points for
which all the three eigenvalues of A are real, and the local flow is strain-dominated. In our DNSs, we follow the
trajectories of tracers or heavy particles in time and calculate the velocity-gradient matrix A at the positions of
these particles. The signs of R and ∆ help us to identify
whether a particle lies in a vortical or a strain-dominated
region of the flow at a given instant of time. To obtain
statistics for the time scales over which such particles stay
in vortical or strain-dominated regions of the flow, it is
natural to use the following idea of persistence from nonequilibrium statistical mechanics: For a fluctuating field
φ, we find the probability distribution function (PDF)
Pφ (tpers ), which gives the probability that φ does not
2
change sign up to time tpers . Persistence times can also
be thought of as first-passage times [22].
Persistence has been studied in many non-equilibrium
systems, e.g., the simple diffusion equation with random
initial conditions [23], reaction-diffusion systems [24],
and fluctuating interfaces [25]. In many systems it has
been found that Pφ (tpers ) ∼ tpers −θ , as tpers → ∞, where
θ is called the persistence exponent [26]. This exponent
θ can be universal; it can be obtained analytically only
in a few cases; most often it is calculated numerically.
We refer the reader to Refs. [26, 27] for reviews of such
persistence problems.
In our DNS we calculate the PDF Pφ (tpers ) of the times
tpers for which tracers or heavy particles remain in vortical or strain-dominated region. We find that, in the
frame of tracers or heavy particles, these PDFs show exponentially decaying tails, from which we extract the decay times scales. Our study quantifies the dependence
of these time scales on the Stokes number St = τp /τη ,
with τp the particle-response or Stokes time and τη the
dissipation-scale time and provides, therefore, a natural
way of answering the following question: How long do
particles spend in vortical regions in turbulent flows?
The remainder of this paper is organized as follows.
In Sec. II we present the 3D Navier-Stokes equation, the
equations we use for the time evolution of tracers and
heavy particles, and the numerical methods we use to
solve these; in subsection II A we define the two invariants
Q and R, which we use to distinguish between vortical
and strain-dominated regions of the flow. Section III
is devoted to a detailed description of our results; and
Section IV contains concluding remarks.
II.
MODEL AND NUMERICAL METHODS
We perform a DNS of the incompressible, threedimensional, forced, Navier-Stokes (3D NS) equation
∂t u + u · ∇u = ν∇2 u − ∇p + f ,
∇ · u = 0,
(1)
(2)
TABLE I: Table of parameters for our DNS run with N 3
collocation points: ν is the kinematic viscosity, δt the time
step, Np is the number of tracers or heavy particles, kmax the
largest wave number, ǫ the mean rate of energy dissipation;
η = (ν 3 /ǫ)1/4 and τη = (ν/ǫ)1/4p
are the dissipation length and
time scales, respectively; λ = 2νE/ǫ is the Taylor microscale, where E is the mean energy of the
P flow, and Reλ is
E(k)/k
the Reynolds number based on λ, Il = k E
is integral
length scale, where E(k) is the energy spectrum of the flow,
and Teddy = Il /urms is the large eddy turn-over time, where
urms is the root-mean-squared velocity of the flow.
N
ν
δt
Np
Reλ
256 3.8 × 10−3 5 × 10−4 40, 000
43
ǫ
η
λ
Il
τη
−2
0.49 1.82 × 10
0.16
0.51 8.76 × 10−2
u(X) is the flow velocity at the position X, and dots
denote time differentiation. We consider mono-disperse
spherical particles, with radii rp ≪ η, material density
ρp much greater than the fluid density ρf , and a small
number density, so we neglect (a) the effect of the particles on the flow (i.e., we have passive particles) and (b)
particle-particle interactions. We also assume that, as
in several experiments, typical particle accelerations, in
strongly turbulent flows, exceed significantly the acceleration because of gravity. We also study the statistics of
tracers for which the equation of motion is
Ẋ = u(X).
Ẋ = V ,
1
V̇ =
[u(X) − V ] ,
τp
(3)
where X and V denote, respectively, the position and
velocity of the particle, τp is the particle-response time,
(4)
We solve Eqs. (3) and (4) by using an Euler scheme in
time to follow the trajectories of Np particles in our DNS.
The velocity-gradient matrix A is calculated at each grid
point by using spectral method. We use trilinear interpolation to calculate the components of u(X) and A at
the off-grid positions of the particles. Table II gives the
list of parameters we use in our DNS.
A.
where u, p, f , and ν are the velocity, pressure, force, and
kinematic viscosity, respectively. Our simulation domain
is a periodic box of length 2π. We solve the 3D NS equation by using the pseudo-spectral method with N 3 collocation points and the 2/3-dealiasing rule [28]. We use
a constant-energy-injection forcing scheme [29], with a
rate of energy injection ε. For time integration we use a
second-order, exponential Adams–Bashforth scheme [30].
Heavy particles obey the following equations [31, 32]:
kmax η
1.56
Teddy
0.49
Q − R invariants of the velocity-gradient tensor
We follow Ref. [20] to note that the velocity-gradient
matrix A has three invariants under canonical transformations, namely, P = T r(A), Q = −T r(A2 /2), and
R = −T r(A3 /3). Incompressibility yields P = 0, for all
the points in our domain. The nature of the eigenvalues
is determined by the signs of R and ∆ = (27/4)R2 + Q3 ,
the discriminant of the characteristic equation of A. This
allows us to classify each point in our flow into four regions, in the Q−R plane, as shown in Fig. 1. If ∆ is large
and positive, vorticity dominates the flow; if, in addition,
R < 0 (Region B), vortices are compressed, whereas, if
R > 0 (Region A), they are stretched. If ∆ is large and
negative, local strains are high and vortex formation is
not favored; furthermore, if R > 0 (Region D), fluid elements experience axial strain, whereas, if R < 0 (Region
Q
3
1.0
2.0
0.5
1.5
0.0
1.0
R
Region: B
Region: A
0.5
−0.5
−1.0
∆
0.0
Region: C Region: D
−0.6 −0.4 −0.2
0.0
R
0.2
0.4
0.6
FIG. 1: (Color online) The flow is topologically different for
values of Q and R that lie in the four regions shown in the
Q − R plane (after Ref. [20]); the black curve is the zerodiscriminant line ∆ = 0. Regions A and B are vorticitydominated regions; in region A vortices are stretched and in
region B they are compressed. By contrast, regions C and
D corresponds to strain-dominated or extensional regions; in
region C fluid elements experience biaxial strain, whereas, in
region D, they feel axial strain. The red dashed curve shows
a illustrative path, in the Q − R plane, as a tracer moves
through the fluid in our DNS.
−0.5
0
1
III.
RESULTS
From our simulations we find that the iso-surfaces of
vorticity have tubular shapes that are well-known from
DNSs of fully developed turbulence. The heavy particles
distribute themselves away from regions of high vorticity.
We consider the motion of ten species of particles:
tracers and nine heavy particles, with different values of
St. We inject Np particles of each species into the flow.
We collect data for averages after the system of particles
and the flow have reached a non-equilibrium, turbulent,
but statistically steady, state. It has been already observed by over laying positions of heavy particles on twodimensional contours of vorticity that the heavy particles
distribute themselves away from regions of high vorticity.
Here we look at a time-series of R and ∆ obtained along
the trajectory of a particle; a typical example of such a
time series for a tracer is shown in Fig. (2). The intersection of any one of these curves with the black, horizontal
line indicates the migration of a particle from one region
of the Q − R plane to another.
3
4
5
t/Teddy
FIG. 2: (Color online) Plots of R (blue) and the discriminant
∆ of the characteristic equation for the velocity-gradient tensor (green), calculated along the trajectory of a tracer as a
function of the dimensionless time t/Teddy . The intersection
of any one of these curves with the black horizontal line indicates the migration of a particle from one region of the Q − R
plane to another. R and ∆ are Nondimensionalized by Λ3
and Λ6 , respectively, where Λ = uη /η.
A.
C), they feel biaxial strain [20].
2
Persistence times via Q and R
We follow the trajectory of each particle and calculate
the components of A and the values of Q and R at the
particle position as a function of time. In Fig. 3 we plot
contours of the joint PDFs of Q and R [P (Q, R)], on log
scales; we calculate these values of Q and R along the
trajectories of tracers and heavy particles, for different
values of St. These joint PDFs show that the tracers are
more likely to be in vorticity-dominated regions (region
above the black curve in the Q−R plane), as compared to
the heavy particles; in addition, the probability of finding
heavy particles in the vortical regions first decreases and
then increases, as we increase St.
We obtain the PDFs from our DNS as follows: (A)
In the Eulerian framework, by following the time evolution of Q and R at a fixed point (x, y, z) in space, we
determine the time tpers for which the flow at this point
remains in one of the four regions described above; (B)
in the Lagrangian framework we obtain the time tpers
for which a tracer resides in one of these regions; (C)
the same calculation as in (B) but for heavy particles.
For the Eulerian PDFs we use a superscript E, for tracer
PDFs a superscript L, and for heavy-particle PDFs a superscript I. For each of the four regions in the Q − R
plane, we use the subscript A, B, C, and D. For example,
PAI denotes the PDF of times tpers that a heavy particle
spends in the region A of the Q − R plane.
In Fig. 4 we show semi-log plots of the PDFs of tpers
for the four regions A, B, C, and D, which indicate that
these PDFs display exponentially decaying tails for large
4
1.2
1.2
6
6
0.6
0.0
−1.2
2
−1.2
−1.8
0
0
−3.0
−1
0
R
1
2
−3
−2
−1
0
R
1
2
R
1
2
3
0.6
0.0
4
−1.2
3
−0.6
Q
Q
2
2
−1.2
−1.8
0
−1.8
0
−2.4
−3.0
−2
−3.6
0
2
−0.6
−3.0
−1
R
1
1.2
0.0
4
−2.4
−2
0
6
−1.8
−2
−1
0.6
0.0
0
−2
1.2
−0.6
Q
−3
6
0.6
−1.2
−3.0
−3.6
3
1.2
2
−2.4
−2
−3.6
6
−3
−1.8
−3.0
−2
3
4
−1.2
0
−2.4
−3.6
−2
−0.6
2
−1.8
−2.4
−2
0.0
4
−0.6
Q
Q
−0.6
2
0.6
0.0
4
Q
4
−3
1.2
6
0.6
−2.4
−3.0
−2
−3.6
−3
−2
−1
0
R
1
2
−3.6
3
−3
−2
−1
0
R
1
2
3
FIG. 3: (Color online) Contour plots of the joint PDFs of Q and R, on log scales, calculated along the trajectories of particles
with different Stokes numbers, from top the left corner, (a) tracers, (b) St = 0.1, (c) St = 0.5, (d) St = 1.0, (e) St = 1.4, and
(f) St = 2.0. Q and R are Nondimensionalized by Λ2 and Λ3 , respectively, where Λ = uη /η. ∆ = 0 curve is shown by solid
black line, ∆ > 0 corresponds to vorticity dominated region and ∆ < 0 corresponds to strain dominated region.
values of tpers . We give the forms of these PDFs, for small
values of tpers , in the insets (lin-lin plots). We find that
these PDFs do not go to zero as tpers → 0. The qualitative natures of these PDFs, for small tpers , are similar
for regions A, C, and D, but not for region B. These
PDFs are obtained by computing the histograms and,
therefore, they suffer from binning errors. To overcome
these errors, we calculate the corresponding cumulative
PDFs, by using the rank-order method [33]. We denote
by QIA the cumulative PDF (CPDF) that follows from
PAI ; clearly,
PAI (tpers ) ≡
d
dtpers
QIA (tpers ).
(5)
In Fig. 5 we give semi-log plots of QIA (tpers ), for tracers
and heavy particles, in regions A (top right), B (top left),
C (bottom left), and D (bottom right). We observe that
all these CPDFs have exponentially decaying tails, from
which we extract the characteristic time scales Tα (α =
A, B, C, or D) that we list in Table II for all species of
particles. We also note that, in regions A and B, which
are vorticity dominated, TA and TB are largest for tracers; and they decrease as St increases. Furthermore, for
all species of particles, TB > TA . The time scale TC for
region C, which is strain-dominated, does not change significantly with St. The time scale TD for region D, where
axial strain dominates, assumes its lowest value for tracers; and it changes only marginally as St increases.
To provide a clear answer to the question we pose in
the title of this paper, we must calculate the PDFs of
the time tpers for which heavy particles stay in vortical
regions of the flow. We do this by monitoring the sign
of ∆ along the trajectories of the particles, for ∆ > 0
in vorticity-dominated regions of the flow and ∆ < 0 in
strain-dominated ones. In Fig. 6 we shows the CPDFs
tpers for the cases where ∆ remain positive (left panel)
or negative (right panel), along the trajectories of tracers
(St = 0) or heavy particles; we find that these CPDFs
also have exponentially decaying tails. We extract the
time scales Tvortical and Tstrain , for particle residence in
vortical or strain-dominated regions of the flow, respectively, by fitting exponential functions to these tails. We
list these times in Table III for different values of St. We
observe that Tvortical decreases monotonically as St increases, whereas Tstrain first increases and then decrease.
Furthermore, the values of Tvortical and Tstrain indicate
that tracers and heavy particles, with small values of St,
stay longer in vortical regions of the flow than in straindominated ones, because the difference between Tvortical
and Tstrain is large here. By contrast, for heavy particles,
with high values of St, the difference between Tvortical and
Tstrain is insignificant, so these particles spend roughly
the same amount of time in vortical regions of the flow
as in strain-dominated ones.
IV.
CONCLUSIONS
Our DNS of tracers and heavy particles in statistically steady, homogeneous and isotropic turbulence in the
5
101
101
4
3
2
1
0.0 0.1 0.2 0.3 0.4 0.5
PDF
10-1
10-2
10-3
10-4
0
St = 0.0
St = 0.2
St = 1.0
St = 1.7
St = 2.4
1
10
2
3
Region: C
PDF
10-2
10-4
0.0
St = 0.0
St = 0.2
St = 1.0
St = 1.7
St = 2.4
0.5
4
tpers/Teddy
10-1
10-3
Region: A
10-1
10-2
5
6
7
25
20
15
10
5
00.0 0.1 0.2 0.3 0.4 0.5
10-4
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
tpers/Teddy
101
10
0
Region: D
10-1
10-2
10-3
1.0
tpers/Teddy
1.5
2.0
6
5
4
3
2
1
0.0 0.1 0.2 0.3 0.4 0.5
St = 0.0
St = 0.2
St = 1.0
St = 1.7
St = 2.4
10-3
101
0
100
PDF
Region: B
PDF
100
10-4
0
St = 0.0
St = 0.2
St = 1.0
St = 1.7
St = 2.4
1
2
10
8
6
4
2
00.0 0.1 0.2 0.3 0.4 0.5
3
tpers/Teddy
4
5
FIG. 4: (Color online) Semi-log plots of the persistence-time PDFs Pφ (tpers ) of the times tpers for the four regimes in the Q − R
plane, for different Stokes numbers; the inset shows Pφ (tpers ) for small tpers .
TABLE II: Values of characteristic time scales, Tα for all four
regions of Q − R plane (α = A, B, C, D), calculated in the
Eulerian frame for and in the frame of tracers and inertial
particles, by fitting Qα (tpers /Teddy ) ∼ exp(−tpers /Tα ) to the
cumulative PDFs of residence time.
TA /Teddy TB /Teddy TC /Teddy TD /Teddy
Eulerian
Tracers
St = 0.1
St = 0.2
St = 0.5
St = 0.7
St = 1.0
St = 1.4
St = 1.7
St = 2.0
St = 2.4
0.13
0.37
0.35
0.34
0.34
0.32
0.32
0.30
0.29
0.29
0.28
0.23
0.68
0.63
0.58
0.48
0.45
0.44
0.41
0.39
0.39
0.37
0.08
0.17
0.18
0.18
0.18
0.19
0.19
0.16
0.16
0.15
0.16
0.13
0.37
0.39
0.41
0.42
0.41
0.42
0.42
0.41
0.42
0.39
forced, 3D NS equation has helped us to explore how long
such particles spend in vortical regions of a turbulent flow
and in strain-dominated ones by combining properties of
the velocity-gradient tensor, which is well known in fluid
mechanics, and the notion of persistence times, which has
received considerable attention in non-equilibrium statistical mechanics. The Q and R invariants play a crucial
role in our analysis of PDFs and CPDFs of persistence
times, conditioned on the values of R and ∆. The exponential tails of these PDFs and CPDFs help us to extract
time scales that we identify with particle-residence times
in vortical or strain-dominated regions of the turbulent
flow. We hope that our detailed study of persistence-time
PDFs in 3D turbulent flows will lead to experimental
studies of such statistics for tracers and heavy particles.
Our work is a natural generalization of a similar study
for tracers [7] in two-dimensional, statistically steady,
homogeneous and isotropic turbulent flows. In twodimensions, instead of Q and R, we must use the OkuboWeiss parameter Λ ≡ det(A). This study has found that
6
0
0
10
10
−1
Cumulative PDF
10
−2
10
−3
10
−4
10
0
−1
2
−3
10
−4
3
4
tpers/Teddy
5
6
Region: A
0
0
1
2
tpers/Teddy
3
0
10
10
−1
−2
10
−3
10
−4
10
0
0.5
−1
−2
10
−3
10
−4
10
Region: C
1
tpers/Teddy
1.5
St = 0
St = 0.2
St = 1
St = 1.7
St = 2.4
10
Cumulative PDF
St = 0
St = 0.2
St = 1
St = 1.7
St = 2.4
10
Cumulative PDF
−2
10
10
Region: B
1
St = 0
St = 0.2
St = 1
St = 1.7
St = 2.4
10
Cumulative PDF
St = 0
St = 0.2
St = 1
St = 1.7
St = 2.4
0
Region: D
1
2
tpers/Teddy
3
4
FIG. 5: (Color online) Semi-log plots of the cumulative persistence time PDFs (obtained by the rank-order method) for the
four regimes in the Q − R plane, for different values of the Stokes number.
TABLE III: Values of the characteristic time scales, for the
vortical (∆ > 0) and strain dominated (∆ < 0) regions, calculated in the frame of tracers and heavy particles for different
values of St.
Tvortical /Teddy Tstrain /Teddy
Tracers
St = 0.1
St = 0.2
St = 0.5
St = 0.7
St = 1.0
St = 1.4
St = 1.7
St = 2.0
St = 2.4
1.44
1.11
0.97
0.73
0.71
0.64
0.59
0.56
0.55
0.55
0.54
0.59
0.59
0.62
0.63
0.60
0.59
0.57
0.55
0.51
the PDF of the persistence time τ , for a Lagrangian particle in vortical regions, displays a power-law tail, i.e.,
−θ
P Λ (τ− ) ∼ τ−
, where the exponent θ ≃ 2.9 [7]. By
contrast, we show that the residence-time PDFs in 3D
turbulent flows display exponentially decaying tails, for
all species of particles and for all four regions in the Q−R
plane. The most likely reason for this qualitative difference of persistence-time PDFs (power-law as opposed to
exponential tails) in 2D and 3D fluid turbulence is that,
in the 3D case, the velocity-gradient tensor A always has
one real eigenvalue, so tracers and particles can escape
more easily from vortical regions than they can in 2D
turbulent flows. However, we must also note that the
extent of the power-law region seen in the 2D study [7]
increases with the Reynolds number. The Reynolds numbers that we can achieve in our 3D DNS is significantly
lower than that in 2D. Therefore, very-high-resolution,
large-Reynolds-number DNSs of 3D turbulence with trac-
7
100
10-2
10
10-2
10-3
10-3
10-4
10-5
St = 0.0
St = 0.2
St = 1.0
St = 1.7
St = 2.4
-1
Cumulative PDF
-1
Cumulative PDF
10
100
St = 0.0
St = 0.2
St = 1.0
St = 1.7
St = 2.4
10-4
0
2
4
6
8
tpers/Teddy
10
12
14
16
10-5
0
1
2
3
4
tpers/Teddy
5
6
7
FIG. 6: (Color online) Semi-log plots of the cumulative persistence-time PDFs (obtained by the rank-order method) for vortical
(∆ > 0, left panel) and strain-dominated (∆ < 0, right panel) regions, for different values of the Stokes number St (the plot for
tracers is labeled by St = 0). From the slopes of the tails of these PDFs we extract the times Tvortical and Tstrain , for particle
residence in vortical or strain-dominated regions of the flow, respectively,.
ers and particles are required to confirm the absence of
power-law tails in persistence-time PDFs here.
The clustering of heavy particles in 3D fluid turbulence has been characterized by calculating a correlation
dimension, which decreases first as St increases (for small
St), thus indicating clustering; but this dimension reaches
a minimum value near St ≃ 0.7, and then increases to
a value ≃ 3 (i.e., a uniform distribution with insignificant clustering) as St increases beyond 0.7 [13]. This can
be understood in terms of singularities (caustics) in the
(particle) velocity gradient field, see e.g., Refs. [34, 35]
for a review. The intuitive picture of clustering because
of ejection from vortices is not enough to understand the
clustering. Nevertheless, we observe by plotting the joint
PDFs of Q and R as measured along the trajectories of
heavy particles (Fig. 3): the probability of finding the
heavy particles in the vortical regions first decreases and
then increases, as we increase St. However, the characteristic times scales that we have calculated for such
particles in vortical structures behave differently, insofar
as they do not show such a clear, non-monotonic dependence on St (see Tables II and III).
Our DNS supports and quantifies the qualitative argument that heavy particles spend less time than tracers
in vortical regions in 3D turbulent flows. However, the
residence time scales depend only weakly on St, over the
range we have in Tables II and III. Surprisingly, these
characteristic times scales are comparable to the largeeddy turnover time. The values of these time scales can
be used as input parameters in developing a model for
the dynamics of the particles in turbulent flows. If the
same characteristics time scales are calculated for Eulerian grid points, we find that they are about one-tenth of
the large-eddy turnover time, i.e., they are of the same
order as our Kolmogorov time scale. We see from the tails
of the cumulative PDFs in Fig. 6 that some of the particles can reside inside vortical regions for times that are
much longer than Teddy . Therefore, we need to run our
DNSs for very long times to get good statistics. The results we present here have been obtained by running our
DNSs for roughly 80Teddy. With such long runs, it is not
possible to carry out very-high-resolution DNSs, at high
Reynolds numbers and to obtain reliably the Reynoldsnumber dependence of persistence times.
One of the many longstanding questions in turbulence
concerns the lifetime of vortices. Clearly, to measure the
lifetime of a vortex we must have a precise definition of
a vortex, which is, in itself, still controversial, (see, e.g.,
Ref. [36]). One of the several different criteria used to
define a vortex, called the Q-criterion, is precisely the
condition ∆ > 0 that we have used. If we use this condition to define a vortex, then the time a tracer particle
spends in a vortex can be considered as a measure of the
lifetime of a vortex itself. Therefore, with this interpretation, we have provided an answer to the old question:
What is the typical lifetime of vortical structures? The
cumulative probability distribution of the lifetime of a
vortex, in homogeneous and isotropic turbulence, given
in Fig. (6), has an exponential tail, which allows us to
define a characteristic lifetime for a vortex; we give this
lifetime in Table III. Other criteria for the definition of
vortical regions can be used to measure the lifetime of
vortices; and these may yield results that are different
from those in Table III. An interesting attempt has been
made to measure the PDF of the life time of vortical
structures in Ref. [37] by using a DNS of light bubbles.
This study lacked a precise definition of a vortex and had
much smaller run times than those in our DNSs. Nevertheless, the characteristic lifetime of vortices, obtained
in this study of Ref. [37], are roughly equal to those we
8
find.
An alternative way to define a vortical region (as opposed to a vortical point) is: “to be a compact region
of vorticity, possibly unbounded in one direction, surrounded by irrotational fluid. Strictly speaking, the viscosity has to vanish for this definition to make sense, but
we suppose that the viscosity is very small, and we allow
transcendentally small vorticity outside the vortex ...”
(this quotation is from Ref. [38]). By using the lifetime
of vortical regions, in a model for vortex tubes, Mori [39]
has argued that the characteristic dimension of vortical
regions increases as a power-law in time, with a universal
exponent equal to 3/2. As the Q-criteria is applicable to
a point, but not to a region, we cannot comment on this
result.
V.
(DM), Knut and Alice Wallenberg Foundation (DM and
AB) under project Bottlenecks for particle growth in
turbulent aerosols (Dnr. KAW 2014.0048), and Council of Scientific and Industrial Research (CSIR), University Grants Commission (UGC), and Department of Science and Technogy (DST India) (AB and RP). We thank
SERC (IISc) for providing computational resources. PP
and RP thank NORDITA for hospitality under their Particles in Turbulence program; DM thanks the Indian Institute of Science for hospitality during the time some of
these calculations were initiated.
ACKNOWLEDGMENT
This work has been supported in part by Swedish Research Council under grant 2011-542 and 638-2013-9243
[1] R. A. Shaw, Annual Review of Fluid Mechanics 35, 183
(2003).
[2] W. W. Grabowski and L.-P. Wang, Annual Review of
Fluid Mechanics 45, 293 (2013).
[3] M. Pinsky and A. Khain, Journal of aerosol science 28,
1177 (1997).
[4] G. Falkovich, A. Fouxon, and M. Stepanov, Nature 419,
151 (2002).
[5] P. J. Armitage, Astrophysics of Planet Formation (Cambridge University Press, Cambridge, UK, 2010).
[6] I. De Pater and J. J. Lissauer, Planetary sciences (Cambridge University Press, ADDRESS, 2015).
[7] P. Perlekar, S. S. Ray, D. Mitra, and R. Pandit, Physical
review letters 106, 054501 (2011).
[8] F. Toschi et al., Journal of Turbulence N15 (2005).
[9] L. Biferale and F. Toschi, Journal of Turbulence N40
(2005).
[10] J. K. Eaton and J. Fessler, International Journal of Multiphase Flow 20, 169 (1994).
[11] L. R. Collins and A. Keswani, New Journal of Physics 6,
119 (2004).
[12] H. Yoshimoto and S. Goto, Journal of Fluid Mechanics
577, 275 (2007).
[13] J. Bec et al., Physical review letters 98, 084502 (2007).
[14] L. Biferale et al., Physics of Fluids (1994-present) 17,
115101 (2005).
[15] M. R. Maxey, Journal of Fluid Mechanics 174, 441
(1987).
[16] L. Biferale et al., Progress in Turbulence II (Springer,
ADDRESS, 2007), pp. 207–212.
[17] J. Bec, H. Homann, and S. S. Ray, Physical review letters
112, 184501 (2014).
[18] M. Cencini et al., Journal of Turbulence N36 (2006).
[19] M. Chong, A. E. Perry, and B. Cantwell, Physics of Fluids A: Fluid Dynamics (1989-1993) 2, 765 (1990).
[20] B. J. Cantwell, Physics of Fluids A: Fluid Dynamics
(1989-1993) 5, 2008 (1993).
[21] A. Perry and M. Chong, Annual Review of Fluid Mechanics 19, 125 (1987).
[22] S. Redner, A guide to first-passage processes (Cambridge
University Press, ADDRESS, 2001).
[23] S. N. Majumdar and C. Sire, Physical review letters 77,
1420 (1996).
[24] E. Ben-Naim, Physical Review E 53, 1566 (1996).
[25] J. Krug et al., Physical Review E 56, 2702 (1997).
[26] A. J. Bray, S. N. Majumdar, and G. Schehr, Advances in
Physics 62, 225 (2013).
[27] S. N. Majumdar, Current Science 77, (1999).
[28] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A.
Zang, Spectral methods in fluid dynamics (PUBLISHER,
ADDRESS, 1988).
[29] G. Sahoo, P. Perlekar, and R. Pandit, New Journal of
Physics 13, 013036 (2011).
[30] A. Bhatnagar, Ph.D. thesis, Indian Institute of Science
Bangalore, 2016, unpublished thesis.
[31] R. Gatignol, Journal de Mécanique théorique et appliquée 2, 143 (1983).
[32] M. R. Maxey and J. J. Riley, Physics of Fluids (19581988) 26, 883 (1983).
[33] D. Mitra, J. Bec, R. Pandit, and U. Frisch,
Phys. Rev. Lett 94, 194501 (2005).
[34] A. Pumir and M. Wilkinson, arXiv preprint
arXiv:1508.01538 (2015).
[35] K. Gustavsson and B. Mehlig, Advances in Physics 65,
1 (2016).
[36] J. Jeong and F. Hussain, Journal of fluid mechanics 285,
69 (1995).
[37] L. Biferale, A. Scagliarini, and F. Toschi, Physics of Fluids 22, 065101 (2010).
[38] D. Pullin and P. Saffman, Annual review of fluid mechanics 30, 31 (1998).
[39] H. Mori, Progress of theoretical physics 65, 1085 (1981).