You are on page 1of 20

A direct numerical simulation study of higher order statistics in a turbulent round jet

G. N. Taub, Hyungoo Lee, S. Balachandar, and S. A. Sherif

Citation: Physics of Fluids (1994-present) 25, 115102 (2013); doi: 10.1063/1.4829045


View online: http://dx.doi.org/10.1063/1.4829045
View Table of Contents: http://scitation.aip.org/content/aip/journal/pof2/25/11?ver=pdfcov
Published by the AIP Publishing

Articles you may be interested in


Enstrophy and passive scalar transport near the turbulent/non-turbulent interface in a turbulent planar jet flow
Phys. Fluids 26, 105103 (2014); 10.1063/1.4898208

Three-dimensional direct numerical simulation study of conditioned moments associated with front propagation in
turbulent flows
Phys. Fluids 26, 085104 (2014); 10.1063/1.4891735

A framework for epistemic uncertainty quantification of turbulent scalar flux models for Reynolds-averaged
Navier-Stokes simulations
Phys. Fluids 25, 055105 (2013); 10.1063/1.4807067

A detailed quantitative analysis of sparse-Lagrangian filtered density function simulations in constant and
variable density reacting jet flows
Phys. Fluids 23, 115102 (2011); 10.1063/1.3657085

Direct numerical simulation of transitional and turbulent buoyant planar jet flames
Phys. Fluids 16, 4443 (2004); 10.1063/1.1804974

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
PHYSICS OF FLUIDS 25, 115102 (2013)

A direct numerical simulation study of higher order


statistics in a turbulent round jet
G. N. Taub,1,a) Hyungoo Lee,2,b) S. Balachandar,1,c) and S. A. Sherif1,d)
1
Mechanical and Aerospace Engineering, University of Florida,
Gainesville, Florida 32611, USA
2
Fluid System Design Division, Korea Atomic Energy Research Institute,
Daedeok-Daero, Yuseong-gu, Daejeon 305-353, South Korea
(Received 31 January 2013; accepted 18 October 2013; published online 12 November 2013)

Up until recently direct numerical simulation (DNS) studies involving round turbu-
lent jets have focused on first and second order statistics and vortical behavior near
the source of the jet. The third order statistics necessary to compute the turbulent
kinetic energy and Reynolds stress transport equations have been examined using
LES studies. However, further examination with DNS is important as, on the subgrid
scale, LES uses models for Reynolds stress. In this study a DNS of a turbulent free
jet with a Reynolds number equal to ReJ = 2000 is computed using a second order
accurate, time splitting finite volume scheme. First, second, and third order statistics
are compared with previous experimental and numerical studies. All terms of the
turbulent kinetic energy balance are calculated directly. The results are compared
to experimental studies such as those of Hussein et al. [“Velocity measurements
in a high-Reynolds-number, momentum-conserving, axisymmetric, turbulent jet,” J.
Fluid Mech. 258, 31–75 (1994)], Panchapakesan and Lumley [“Turbulence measure-
ments in axisymmetric jets of air and helium. Part 1. Air jet,” J. Fluid Mech. 246,
197–233 (1993)], and others. The assumptions made by the various experimental
studies in order to solve the dissipation and pressure diffusion terms are discussed
and examined using the data from the current study. The Reynolds stress transport
equations are also calculated and discussed. Vortical structures are visualized by the
λci method and discussed along with entrainment of ambient fluid into the jet. The one
dimensional energy spectra in the azimuthal direction are calculated directly and are
also discussed. C 2013 AIP Publishing LLC. [http://dx.doi.org/10.1063/1.4829045]

I. INTRODUCTION
Axisymmetric turbulent free jets occur often in engineering applications and environmental
flows. As such, jets are often used as an archetype in the study of axisymmetric free shear layers. Many
researchers have examined this problem using analytical, experimental, and numerical techniques.1–6
Until recently most numerical studies have used Large Eddy Simulations (LES) or other models to
approximate the turbulent behavior of the jet flow. More recently it has become possible to perform
Direct Numerical Simulations (DNS) of turbulent jets where no modeling is used. Most of these
studies focus on the second order statistics, the formation of instabilities, and vortical structures in
the near field.7–11 Very few focus on the far field behavior or capture the third order moments needed
to examine the complete balance of turbulent kinetic energy and Reynolds stress, which is the focus
of this current work.
Schlichting3 is credited for some of the earliest work on turbulent shear flows including the
axisymmetric turbulent jet. He derived a similarity solution from the boundary layer equations and

a) gotaub@ufl.edu
b) hyungoo@kaeri.re.kr
c) bala1s@ufl.edu
d) sasherif@ufl.edu

1070-6631/2013/25(11)/115102/19/$30.00 25, 115102-1 


C 2013 AIP Publishing LLC

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-2 Taub et al. Phys. Fluids 25, 115102 (2013)

determined that the axial velocity of an axisymmetric turbulent free jet decays as z−1 away from the
source.
Wygnanski and Fiedler12 examined an air jet at a Reynolds number of ReJ = 105 using Hot-Wire
Anemometry (HWA), to measure various aspects of the first, second, and third order statistics. Later,
using more modern techniques, both Hussein et al.1 and Panchapakesan and Lumley2 performed
both HWA and Laser-Doppler Anemometry (LDA) experiments on jets with Reynolds numbers of
ReJ = 105 and ReJ = 1.1 × 104 , respectively. Both research groups presented detailed results of the
second and third order statistics as well as the turbulent kinetic energy balance and other second order
transport equations. Both Hussein et al.1 and Panchapakesan and Lumley2 made various assumptions
in order to solve for pressure diffusion and dissipation in order to close the turbulent kinetic energy
balance and other 2nd order transport equations. As a result there is some disagreement between the
results of the two experimental studies. Nevertheless, both papers are considered seminal works in
the experimental study of turbulent round jets.
Various researchers have examined turbulent jets numerically using LES techniques where
the effect of turbulence in the unresolved scales is modeled in order to approximate its effect
on the resolved scales. Olsson and Fuchs13 examined the transition from laminar flow near the
jet source to turbulent flow. In their 2006 paper Bogey and Bailly14 examined the influence of the
Reynolds number on the development of the flow by simulating jets with Reynolds numbers between
ReJ = 1700 and ReJ = 4 × 105 . In this study, it is suggested that self similarity is achieved in a
shorter axial distance away from the source for lower Reynolds number jets. Akselvoll and Moin15
performed simulations of co-annular jets in a confined space in order to study coaxial jet combustion.
Kung4 preformed an LES using an approximate deconvolution model of a jet with a Reynolds number
of ReJ = 2000 and compared his results with both experiments and DNS. He found that the LES
simulation overpredicted the turbulent intensities in the transitional region. Later in 2009, Bogey and
Bailly16 used LES techniques to examine a jet with a Reynolds number of ReJ = 1.1 × 104 . They
reported on the third order statistics, turbulent kinetic energy balance and Reynolds stress transport
equations and discussed possible sources for the disagreements found in the experimental papers.
As computer processing becomes both more powerful and economical, direct numerical
simulations have become more tractable. Boersma et al.7 examined the mean flow and second
order statistics of jets with a constant Reynolds number of ReJ = 2400, but with different inlet
velocity profiles. Lubbers et al.8 examined the self-similarity of the concentration of a passive scalar.
Babu and Mahesh9 compared the effect of entrainment of jets emanating from an orifice in a flat
surface with those emanating from a pipe or chimney. Brancher et al.10 focused on the effect of
breaking rotational symmetry by subjecting their simulated jets to stream-wise and azimuthal pertur-
bations. Freund11 examined noise sources in low Reynolds number turbulent jets. He reported on the
turbulent kinetic energy balance in the near field and ignored terms requiring third order statistics.
Wang et al.5 examined the third order moments of a free jet with a Reynolds number of ReJ = 4700,
but did not report on the turbulent kinetic energy balance nor Reynolds stress transport equations.
Muppidi and Mahesh17 studied a round turbulent jet with a cross flow and reported on the turbulent
kinetic energy balance.
While Bogey and Bailly16 examined the turbulent kinetic energy balance and Reynolds stress
transport equations using LES and Wang et al.5 examined the third order moments using DNS, at this
time, no study has reported on the behavior of the 2nd order transport equations using DNS, without
relying on modeling of turbulence or making assumptions to calculate difficult to measure terms. In
this current work we perform a DNS of a turbulent jet with a Reynolds number of ReJ = 2000. The
simulation is run for over 1000 time scales. First, second, and third order statistics are compared
with previous experimental and computational studies. The turbulent kinetic energy balance and all
terms of the Reynolds stress transport equations are calculated and discussed. Vortical structures are
visualized by the λci method18 and discussed along with entrainment of ambient fluid into the jet.
The one dimensional energy spectra in the azimuthal direction was calculated directly and is also
discussed.
We will begin with Sec. II where we will discuss the numerical methodology used in this study.
Our results will be presented in Sec. III. Final conclusions and future work will be discussed in
Sec. IV.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-3 Taub et al. Phys. Fluids 25, 115102 (2013)

Lz

z
y
Ly
x
D

Lx

FIG. 1. Schematic of the computational domain.

II. METHODOLOGY
We assume a jet emerging from a circular pipe of diameter, D. Fluid exiting the circular pipe
entrains ambient fluid above. Figure 1 shows the computational domain which is a rectangular region
with dimensionless length, width, and height of Lx , Ly , and Lz .
The quantities r̃ , z, and  θ denote the radial, axial, and azimuthal directional vectors in the
cylindrical coordinate system. The quantities  x, 
y, and 
z are the directional vectors in the Cartesian
coordinate system. The origin of the coordinates is positioned at the center of the jet inlet. The velocity
vector is u and p is the pressure. The “tilde” denotes dimensional variables. The fluid properties, such
as kinematic viscosity, ν, and the constant density, ρ 0 , are kept constant for purposes of this analysis.
The variables are scaled by the jet inlet diameter, D, and the jet inlet velocity, UJ . The subscript
‘‘J’’ denotes the conditions
 at the jet inlet. The resulting dimensionless variables are xi =  xi /D,
ui = u i /U J , p = p / ρ0 U J2 , and t = U Jt/D. A top hat initial velocity profile, UJ is assumed.
With appropriate non-dimensionalization, the incompressible governing equations are
∇ · u = 0, (1)

∂u 1 2
+ u · ∇u = −∇ p + ∇ u, (2)
∂t Re J
where the Reynolds number is defined as
UJ D
.
Re J = (3)
ν
When discussing second and third order terms we will use the convention that averaged first
order quantities are capitalized and perturbations are primed. Thus, Reynolds decomposition of
a fluctuating variable can be expressed as
x = X  + x  . (4)
The governing equations were solved numerically on a Cartesian grid with grid points clustered
close to the jet inlet. The domain size was chosen so that effects due to confinement in the lateral
direction are minimized. We consulted previous DNS studies where similar or larger Reynolds
numbers were examined, such as Boersma et al.7 and Aissia et al.19 as two examples, to determine
the size of our computational domain and grid resolution. The domain size and grid resolution used
were 25 × 25 × 45 (Lx × Ly × Lz ) and 251 × 251 × 550 (Nx × Ny × Nz ), respectively.
In the axial direction, the discrete function,
 
1 1
ξk = tanh ψk · log(A z ) , (5)
2 2

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-4 Taub et al. Phys. Fluids 25, 115102 (2013)

where the ψ k are Nz evenly spaced discrete points ranging from zero to one and Az = 1.76 is the
stretching parameter, which was utilized as a stretching function. The values ξ k were then scaled by
the constant co-efficient, L z /ξ Nz , in order to generate the cell center locations in the axial direction
ranging from zero to Lz . In this way the grid size in the axial direction ranged from 0.029D at the
jet inflow boundary to 0.128D at the outflow boundary. Fifty of the Nx and Ny grid points in the x
and y directions are reserved for the jet core, a square region of the x − y plane centered along the
jet axis, with each side of the square region having length D. Within this core region, grid points
are evenly spaced. Outside this core region the size of each grid cell is increased by a factor of
Ax = Ay = 1.037 from next grid cell closer to the jet axis or

X i+1 = A x X i and Yi+1 = A y Yi . (6)

In this way the grid cells in the x and y directions range from 0.02D in the core region to 0.44D at
the lateral boundaries. 
At the inflow boundary, a top-hat profile is used for the axial velocity, uz , for r = x 2 + y 2
< 1/2. Uniformly distributed random fluctuations ranging between ±0.05UJ were added to the top
hat inlet velocity profile. The purpose of these fluctuations is to speed up the transition to fully
turbulent behavior and is not expected to accurately simulate inflow turbulence. In this sense the
inflow is closer to a top-hat inflow (see Refs. 1 and 2) than a turbulent pipe flow. The lower boundary
outside the jet inlet, r ≥ 1/2, is considered as a no-slip wall. We allow the fluid to freely enter or
leave the lateral boundaries (x = ± 12 L x , y = ± 12 L y ) by applying Neumann boundary conditions. A
convective boundary condition with a constant convection coefficient is applied to the axial velocity
at the jet outlet (z = Lz ).
A second-order accurate finite volume scheme is employed for the spatial discretization of the
governing equations on a non-staggered grid. A fractional step method is used for time advancement.
In the advection-diffusion step, the nonlinear terms are treated explicitly using a second-order Adams-
Bashforth scheme and the diffusion terms are treated implicitly with the Crank-Nicholson scheme.
The pressure Poisson equation is solved as a pressure correction step. A final divergence-free velocity
is obtained at each time-step with a pressure correction step. The numerical technique is the same
as that used by Lee and Balachandar20 and Taub et al.,21 where further description and validation of
the code is discussed.
Message Passing Interface (MPI) architecture was used in order to enable parallel processing.
The simulation was run on 32 processors utilizing the High Performance Computer (HPC) center at
the University of Florida.

III. RESULTS
A Reynolds number of ReJ = 2000 was selected in order to be large enough to exhibit turbulent
behavior but small enough as to still be tractable by DNS. The simulation ran for over 1.3 × 105
time steps or approximately 1000 time units, (τ = D/UJ ) beyond the time it took the simulation
to become statistically steady state. Averages of turbulent variables were calculated azimuthally as
well as temporally. Running the simulation over this number of time steps allowed for convergence
of the third order moments necessary for measuring turbulent diffusion of turbulent kinetic energy
 2
and Reynolds stress. Figure 2 compares the radial profiles of u 3  2
z , u z u r , and u z u θ , averaged
over 350τ and 1000τ . In all three cases the profiles averaged over 350τ are similar to those averaged
over 1000τ , giving us confidence that after 1000 time scales the third order statistics have converged
to the degree necessary for the current study.

A. First and second order statistics


Figure 3 depicts the decay of centerline axial velocity of the jet in the current study compared to
experimental studies by Hussein et al.1 and Panchapakesan and Lumley,2 the LES study by Bogey
and Bailly,16 and the DNS study performed by Kung.4 As can be seen in the figure the reciprocal
of the centerline velocity scales linearly with respect to z. Here the constant coefficient of decay is

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-5 Taub et al. Phys. Fluids 25, 115102 (2013)

τ
θ τ

FIG. 2. Third order statistics of current study averaged after 1000 time scales (solid lines) compared to similar statistics
averaged after 350 time scales (dashed lines).

found to be Cd = 5.4 where Cd can be defined as


Uc (z) Cd D
≈ . (7)
UJ (z − z v )
This compares well with other past DNS studies such as Wang et al.5 who found Cd = 5.5, and
experimental studies such as Hussein et al.,1 Cd = 5.8, Wygnanski and Fiedler,12 Cd = 5.4, and
Panchapakesan and Lumley,2 Cd = 6.06. The LES study by Bogey and Bailly16 yielded a slightly
higher value of Cd = 6.4. The term, z v in approximation (7) is the virtual origin which was found to
be z v = 1.3. Hussein et al.1 fit their data with z v = 4 whereas Panchapakesan and Lumley’s2 data
did not require a virtual origin, (z v = 0).
The radial profiles of the mean axial and radial velocity at two different axial locations, z = 15
and 25, are plotted in Figure 4 and compared to previous experimental and LES studies. Here the
radial coordinate is
r
η= . (8)
z − zv
The axial velocity corresponds well to a Gaussian curve fit,

U (η) ln( 12 ) 2
= exp η , (9)
Uz,c  η1/2
2

where η1/2 , the jet half width in terms of the similarity coordinate, is found to be η1/2 = 0.094.
This is the same value that was measured by Hussein et al.,1 and similar to the value obtained by
Panchapakesan and Lumley,2 η1/2 = 0.096. Bogey and Bailly16 reported a value of η1/2 = 0.087. The
difference in both decay rate and jet width may be due to differences in initial velocity profile. Bogey
and Bailly16 enforced an inlet velocity profile based on the hyperbolic tangent function suggesting

10 5
Hussain, Curve fit (Re=10 , Exp)
Pan. & Lumley, Curve fit (Re=1.1x104, Exp)
Bogey and Bailly (Re=11,000, LES)
8
Kung (Re=2,000, DNS)
Present (Re=2,000, DNS)
UJ/<Uz,c>

0
0 5 10 15 20 25 30
z

FIG. 3. Initial inlet velocity, UJ , over mean centerline axial velocity, Uz, c , of current study (filled in circles) compared to
Hussein et al.1 (solid line), Panchapakesan and Lumley2 (dashed line), Kung4 (open squares), and Bogey and Bailly16 (open
diamonds).

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-6 Taub et al. Phys. Fluids 25, 115102 (2013)

η η
FIG. 4. Comparison of current study’s time averaged velocity field at two different radial distances, z = 15, 20 with curve
fits of experimental results by Hussein et al.1 (solid lines), Panchapakesan et al.2 (dashed lines), Wygnanski and Fiedler12
(dashed-dotted lines), and LES results by Bogey and Bailly16 (open diamonds).

fully turbulent pipe flow at the jet source. Hussein et al.1 reported a nearly top hat profile at the inlet,
similar to the profile used in the current study.
The maximum positive radial velocity in the interior of the jet is found to be approximately
Ur +max /Uz, c  ≈ 0.02. This is slightly larger than 0.018 found by Panchapakesan and Lumley2
or 0.015 found by Wygnanski and Fiedler.12 The maximum inward radial velocity of entrained
fluid entering the jet is found to be Ur −max /Uz, c  ≈ −0.026 and is found at η = 0.23 slightly
further away from the jet axis and slightly larger in magnitude than measured by Panchapakesan and
Lumley2 and Wygananski and Fiedler12 who agreed on a value of Ur −max /Uz, c  = 0.025 located
at η = 0.21.
Figure 5 compares the evolution of the axial and radial components of the turbulent veloc-
ity intensity along the jet axis with a previous DNS study by Kung,4 the LES results by Bogey
and Bailly16 and an experimental study by Amielh et al.6 of a jet with a Reynolds number of
ReJ = 2.1 × 104 . The numerical studies predict relatively constant turbulent intensity above the
transitional region, whereas the experimental results, where larger Reynolds numbers were exam-
ined, suggests a more gradual approach to some asymptotic value. This further corroborates the
study of Bogey and Bailly14 who also suggested that low Reynolds number turbulent jets develop
more quickly and reach similarity in a shorter axial distance away from their source than jets with
larger Reynolds numbers. Indeed the LES study by Bogey and Bailly16 with a Reynolds number of

FIG. 5. Comparison of turbulent intensity along the jet axis found in the current study (solid lines) with that of the previous
DNS study by Kung4 (dashed lines), LES results by Bogey and Bailly16 (dashed-dotted lines), and an experimental study by
Amielh et al.6 (symbols).

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-7 Taub et al. Phys. Fluids 25, 115102 (2013)

Re = 11000 achieves a near constant centerline turbulent intensity farther from the source than the
two lower Reynolds number DNS studies and closer to the source than the higher Reynolds number
experimental study by Amielh et al.6
In the current study random perturbations were added to the initial top hat axial velocity profile
at each time step. This was done in order to reduce the axial length necessary to initiate turbulent
behavior and was not meant to simulate turbulence at the inflow. It can be seen in Figure 5 that in
the present simulations this initial random perturbation dies out rather quickly. Kung4 addressed the
perturbation in his DNS simulation by incorporating an inflow generator which simulated a turbulent
pipe flow at the jet source. Bogey and Bailly16 introduced perturbations in the form of divergence
free vortex rings of nine different modes in the shear layer near the inflow boundary. As such the
turbulent intensities reported by Kung4 and Bogey and Bailly16 are more similar to that found in the
experimental study of Amielh et al.6 than the current study near the jet’s source. However, the three
numerical techniques have similar results further away from the source, which is the area of interest
in the current study.
Figure 6 compares the radial profiles of the turbulent intensity components scaled by the mean
axial velocity at the centerline with previous studies. The profile data from the current study is at a
distance, z = 15, away from the jet inlet. The present results compare well with the experimental
results of Amielh et al.,6 Hussein et al.,1 and Panchapakesan and Lumley2 as well as the numerical
results of Kung4 and Bogey and Bailly.16 As noted by previous researchers, Wang et al.5 for example,
the maximum of both the axial Reynolds normal stress u 2 z  and the Reynolds shear stress term,
u z u r  are not along the axis, but rather at a radial distance slightly less than the jet’s half width
where there is a strong interaction between the jet and the ambient fluid. Panchapakesan and Lumley2
noticed that this is also true of the tangential Reynolds normal stress term, u 2 θ  and this can also be
seen in the current study.

FIG. 6. Comparison of the radial profiles of Reynolds stress terms with previous studies. Here the results of Hussein et al.1
(solid thin lines) and Panchapakesan and Lumley2 (thin dashed lines) are represented by a Curve Fits (CF) of their respective
experimental data. The thicker lines represent the LES results by Bogey and Bailly16 (dashed-dotted lines), the DNS study
performed by Kung4 (dashed lines), and the current DNS study (solid lines).

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-8 Taub et al. Phys. Fluids 25, 115102 (2013)

X
X
X
X
θ
X
X
X
X π

FIG. 7. Contributions to the total momentum integral by various terms.

If the streamwise pressure gradient can be ignored then it is easy to show that the initial
momentum flux,

D/2

M J = 2π u 2z r dr , (10)
0 z=0

is conserved. However, Hussein et al.1 suggested that the streamwise pressure gradient should not
be ignored and by using the cross-stream momentum equation to eliminate the pressure term they
derived
M J = M(z) + , (11)
where M(z) is referred to as the momentum integral and is defined as


2 1  2 2

M(z) = 2π Uz  + u z  −
2
u r  + u θ  r dr (12)
0 2
and

∞ z
z
1 ∂   1 ∞
= 2π − Uz Ur  + u z u r  r 2 dr + Ur 2r dr
2 ∂z 0 0 2 0 0

∞  z
z

1 1 2 1 2
+ 2π ∇ Ur  r 2 dr + ∇ Uz r dr dz  . (13)
2 0 Re J 0 0 0 Re J
By a combination of scaling analysis and using the similarity solution to solve the integrals in
Eq. (13), Hussein et al.1 determined that is negligible compared to MJ . Therefore, it was suggested
that M(z) ≈ MJ is an important test to confirm the accuracy of any model of an axisymmetric jet in
an infinite environment.
The current study supports both the suggestion that the streamwise pressure gradient should be
considered and that is negligible compared to MJ . In the current DNS where the initial top hat
axial velocity was set to uz = 1 with random perturbations of ±5%, the initial momentum flux is
MJ ≈ π /4. Figure 7 displays each term of Eq. (12) throughout the entire domain and compares M(z)
to MJ . M(z) is found to be within ±5% of MJ throughout the domain. After the flow becomes fully
turbulent for approximately z > 10, the comparison improves so that M(z) is within ±1% of MJ . This
suggests that is not negligible, near the jet source, however its value is still quite small compared to
M(z). The mean flow momentum flux accounts for 93% of the total momentum integral in the fully
turbulent region. The total momentum flux including the mean flow and fluctuating velocity in the
axial direction is approximately 10% larger than the momentum integral, M(z), in the same region.

B. Third order statistics


Derivatives of third order moments play a dominant role in the turbulent diffusion of turbulent
kinetic energy and Reynolds stress. Figure 8 compares the radial profiles of the third order moments
of the current study with the experimental measurements of Hussein et al.1 and Panchapakesan and

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-9 Taub et al. Phys. Fluids 25, 115102 (2013)

<u’3z > <u’zu’2r > <u’zu’2θ >


.008 .004 .004

.006
.002 .002
.004
.000 .000
.002

.000 -.002 -.002


0 0.05 0.1 0.15 0.2 0.25 0 0.05 0.1 0.15 0.2 0.25 0 0.05 0.1 0.15 0.2 0.25
r/z r/z r/z

<u’ru’2z > <u’3r > <u’ru’2θ >


.004 .004 .004

.002
.002 .002
.000

-.002 .000 .000


0 0.05 0.1 0.15 0.2 0.25 0 0.05 0.1 0.15 0.2 0.25 0 0.05 0.1 0.15 0.2 0.25
r/z r/z r/z

FIG. 8. Third order statistics of current study (solid lines) compared to experimental studies performed by Hussein et al.1
(squares) and Panchapakesan and Lumley2 (circles), the LES study by Bogey and Bailly16 (dashed-dotted lines), the DNS
study by Wang et al.5 (dashed lines), normalized by Uz, c 3 .

Lumley,2 the LES by Bogey and Bailly16 and the DNS study by Wang et al.5 Terms including odd
powers of u θ will converge to zero and are not presented. Hussein et al.1 and Panchapakesan and
Lumley2 did not report on the u r u 2 θ  term which plays a significant role in the diffusion of the
turbulent kinetic energy as well as the u 2 θ  component of Reynolds stress.
 2
u z u r2 , u r u 2
z , and u u
z θ  are negative near the centerline of the jet. As discussed in
Panchapakesan and Lumley,2 this has a significant impact on the turbulent transport of kinetic
energy. u r u 2 θ  which is responsible for the radial flux of the tangential component of kinetic energy,
does not have this feature, and was found to be near zero but positive near the jet centerline.
There is some disagreement between the results of Hussein et al.1 and Panchapakesan et al.2 in
the u 3 3
z  and u r  terms, where in both cases Hussein et al. reports a significantly larger magnitude.
1

All three numerical studies compare well with Panchapakesan et al.2 in the case of the u 3 z  term. In
the case of the u r3  term the two DNS studies agree better with Hussein et al.1 while the LES study
by Bogey and Bailly16 agrees better with Panchapakesan and Lumley.2

C. Turbulent kinetic energy balance


By performing Reynolds decomposition of the momentum equations, subtracting off the RANS
equations, taking the dot product with the fluctuating velocity vector and taking the average, the
transport equation for turbulent kinetic energy, k, can be derived as


∂k ∂k 1 ∂      ∂   
0 = − Uz  + Ur  − r p ur + p uz
∂z ∂r r ∂r ∂z
 
  ∂ Uz  ∂ Ur    ∂ Uz   2  ∂ Ur  Ur  2
− u z u r + + u 2 + u + u θ 
∂r ∂z z
∂z r
∂r r
 
1 ∂ r  2   1 ∂  2   1 ∂u i 2
− q ur + q uz −   . (14)
r ∂r 2 2 ∂z Re J ∂ x j

Equation (14) is also referred to as the turbulent kinetic energy balance. Here

1  2   2   2 
k= u z + ur + u θ (15)
2

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-10 Taub et al. Phys. Fluids 25, 115102 (2013)

FIG. 9. Comparison of current study’s (symbols) turbulent energy balance with experimental results of Hussein et al.1 (thin
solid lines) and Panchapakesan and Lumley2 (dashed lines) and the LES study by Bogey and Bailly16 (dashed-dotted lines).
Values are scaled by the mean centerline axial velocity cubed over the axial distance from the jet inlet.

is the turbulent kinetic energy and


   2    2  
q 2 u i  = u 2 
z ui + ur ui + u θ ui . (16)
The five terms in brackets represent the contribution of advection, pressure diffusion, production,
turbulent diffusion, and dissipation, respectively. The contribution to turbulent diffusion from the
viscous term is negligible due to the relatively large Reynolds number used in the simulation and
is not included in Eq. (14). This simplification is also made for the Reynolds stress transport
equations (20) through (23) defined below in Subsection III D.
Figure 9 compares the turbulent kinetic energy balance of the current DNS simulation with the
sixth order polynomial best curve fits of the experimental results of Panchapakesan and Lumley2
and Hussein et al.1 as well as the LES results of Bogey and Bailly.16 The balance from the current
study was calculated at an axial distance of z = 15 above the jet inlet.
Previous experimental studies were unable to calculate each term of Eq. (14) directly and
made various assumptions regarding the dissipation and pressure terms in order to close the tur-
bulent kinetic energy equation. Hussein et al.1 discussed various alternatives for approximating
the dissipation term and eventually suggested that the approximation based on a jet being locally
axisymmetric,
 
1 5 ∂u z 2 ∂u z 2 ∂u θ 2 8 ∂u θ 2
dissi pation ≈   + 2  + 2 +   , (17)
Re J 3 ∂z ∂r ∂z 3 ∂r
was preferable to other approximations which assumed local isotropy. This assumption is explored
in Figure 10 where the radial profile of dissipation is compared with the right- hand side of ap-
proximation (17) using data from the current study. Hussein’s assumption is found to be fairly
accurate, as the right-hand side of Eq. (17) is found to have less than a 10% error when com-
pared to the full dissipation term. Hussein et al.1 solved for the pressure term by closing the
balance.
Panchapakesan and Lumley2 assumed the pressure diffusion term to be negligible and obtained
the dissipation term by closing the balance. In that study the equations for the self similar solutions
were used to calculate the axial and radial derivative of various quantities. Other authors, such as
Shiri22 who performed LDA experiments on a jet with a Reynolds number of ReJ = 4 × 104 ,
estimated the pressure term following the suggestion by Hussein et al.1 and Lumley23 that the
streamwise gradient of p  u z can be ignored and that
  1
− p  u r ≈ q 2 u r . (18)
5

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-11 Taub et al. Phys. Fluids 25, 115102 (2013)

FIG. 10. Comparison of dissipation of turbulent kinetic energy between the entire term of the current DNS study (thick solid
line), the value determined by approximation (17) (thick dashed line), the experimental studies by Panchapakesan et al.2
(dashed line), Hussein et al.1 with locally axisymmetric assumption (thin solid line), and Shiri22 (dashed dotted line) and an
LES study by Bogey and Bailly16 (thin dotted line).

Hussein et al.,1 Panchapakesan et al.,2 and Shiri22 assumed that


u r u 2 3
θ  ≈ u r  (19)
for the purpose of calculating the turbulent diffusion term.
The validity of the assumptions made in approximations (18) and (19) are examined in
Figure 11, which compares the radial profiles of the approximated quantities to the terms used
in the approximations, using data from the current DNS study. u r u 2 θ  is found to have
 a similar

shape, but only achieves a maximum value of 60% of that of the u r3  term. The − p  u r term
compares well to 15 q 2 u r  but achieves its maximum value somewhat closer to the jet axis.
In the current study all terms of Eq. (14) were calculated directly from the DNS results. The
error in turbulent kinetic energy balance, in the same scaling as the terms seen in Figure 9 is less
than 0.02 for all values of η.
There is some disagreement regarding the shape of the dissipation profile. Figure 10 compares
the dissipation profile of the current DNS with previous experimental and LES results. Hussein
et al.1 estimated dissipation using approximation (17) and found dissipation to be almost Gaussian
with a magnitude much greater than that found by the other experimental and numerical studies.
It is tempting to suggest that this calls into question the locally axisymmetric assumption used by
Hussein et al.1 as represented by approximation (17). However, the full dissipation term evaluated

.004 .003

.002
θ

.002 .001

.000

.000 -.001
0 0.05 0.1 0.15 0.2 0.25 0 0.05 0.1 0.15 0.2 0.25
r/z r/z
   1 2 
FIG. 11. Comparison of radial profiles of u r u 2 3
θ  to u r  (left) and − p u r to 5 q u r  (right) using data from the current
study.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-12 Taub et al. Phys. Fluids 25, 115102 (2013)

in the current study compares well with that obtained by using the approximation (17) suggesting
that this is not the cause of the disagreement. Another possibility is that the jet examined by
Hussein et al.1 was at a Reynolds number an order of magnitude larger than those examined by
other experimental and numerical studies discussed here. It has been suggested by Mi et al.24
and Bogey and Bailly14 that dissipation rates are dependent on Reynolds number even for very
high Reynolds numbers. Since Hussein et al.1 solved for the pressure term by closing the balance,
the disagreement in the pressure term between Hussein et al.1 and others is a direct consequence
of the disagreement in dissipation, and therefore may also be due to the difference in Reynolds
number.
Shiri,22 who did not calculate dissipation directly but rather relied on the isotropic assumption,
reported an off-axis peak for the dissipation term. Whereas, Panchapakesan and Lumley2 reported
a flat region near the center of the jet, as did the current study. Bogey and Bailly16 did not see this
flat region near the centerline but reported a magnitude for the dissipation term similar to that of
Panchapakesan and Lumley.2

D. Reynolds stress transport equations


Transport equations for the Reynolds stress terms can be derived in a similar fashion as the
turbulent kinetic energy balance. From symmetry all terms that involve odd powers of u θ are zero
and therefore only the normal stresses and the u z u r  shear stress term need to be considered. These
four transport equations are:

u 2z transport equation:


∂  2  ∂  2   ∂p 2 ∂Uz    ∂Uz 
0 = − Uz  u + Ur  u − 2u z  − 2u z  + 2u z u r 
∂z z ∂r z ∂z ∂z ∂r
 
∂u 3
z  1 ∂  2   2 ∂u z 2 ∂u z 2 1 ∂u z 2
− + r u z u r  −  +  + 2  . (20)
∂z r ∂r Re ∂z ∂r r ∂θ

u r2 transport equation:


∂  2  ∂  2  ∂ p ∂Ur  ∂Ur 
0 = − Uz  u r + Ur  u r − 2u r  − 2u r2  + 2u z u r 
∂z ∂r ∂r ∂r ∂z
 
∂u z u r2  1 ∂  3  u r u θ  2 ∂u r 2 ∂u r 2 1 ∂u r 2
− + r u r  − 2 −  +  + 2  . (21)
∂z r ∂r r Re ∂z ∂r r ∂θ

u 2θ transport equation:


∂  2  ∂  2  Ur   2   ∂ p
0 = − Uz  u + Ur  u +2 u θ − 2u θ 
∂z θ ∂r θ r ∂θ

  ∂Uθ    ∂Uθ  ∂  2 1 ∂  2 u r u 2
θ 
− 2u z u θ  + 2u r u θ  − u u  + r u r u θ  + 2
∂z ∂z ∂z z θ r ∂r r
 
2 ∂u θ 2 ∂u θ 2 1 ∂u θ 2
−  +  + 2  . (22)
Re ∂z ∂r r ∂θ

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-13 Taub et al. Phys. Fluids 25, 115102 (2013)

u z u r transport equation:

∂    ∂    ∂ p ∂ p
0 = − Uz  u z u r + Ur  u z u r − u r  + u z 
∂z ∂r ∂z ∂r

∂Uz  ∂Uz  ∂Ur    ∂Ur 
− u z u r  + u r2  + u 2  + u u 
∂z ∂r z
∂z z r
∂r
 2
∂ 2  1 ∂ u u 
− u z u r  + r u z u r2  − z θ
∂z r ∂r r
 
2 ∂u z ∂u r ∂u z ∂u r 1 ∂u z ∂u r
−  +  + 2  . (23)
Re ∂z ∂z ∂r ∂r r ∂θ ∂θ
Similar to the turbulent kinetic energy balance the five terms in brackets, for each Eqs. (20) through
(23), represent the contribution of advection, pressure diffusion, production, turbulent diffusion, and
dissipation, respectively. The contribution to turbulent diffusion from the viscous term is negligible
due to the relatively large Reynolds number used in the simulation and is not included in Eqs. (20)
through (23).
Figure 12 compares the Reynolds stress transport equations of the current study with previous
experimental studies by Hussein et al.1 and Panchapakesan and Lumley.2 In the current study all
terms used to calculate the Reynolds stress transport equations were calculated directly except the
pressure diffusion term which was found by closing the balance. In a departure from the method
they used in calculating the turbulent kinetic energy balance, when calculating the Reynolds stress
transport equations Panchapakesan and Lumley2 used assumptions based on small scale isotropy
to calculate the dissipation term and solved for pressure diffusion by closing the balance. Hussein
et al.1 assumed that the parallel part of the axisymmetric dissipation assumption used in  2calculating

the dissipation term of the turbulent kinetic energy equation
 2  should
 2  be allocated to the u z term and
that the perpendicular part should be allocated to the u r and u θ terms. That is
 
1 ∂u z 2 ∂u z 2
zz ≈   + 2  , (24)
Re J ∂z ∂r

 
1 2 ∂u z 2 ∂u θ 2 8 ∂u θ 2
rr ≈ θθ ≈   + 2 +   , (25)
Re J 3 ∂z ∂z 3 ∂r

and zr ≈ 0. As in the turbulent kinetic energy equation Hussein et al.1 solved for pressure diffusion
by closing the balance.
The two
 experimental 2  studies
 2compare
 fairly well when comparing the axial normal stress
equation, u 2z . For the u r and u θ equations, Hussein et al.,1 as in the turbulent kinetic energy
equation, finds larger values for dissipation than either Panchapakesan and Lumley2 or the current
study. It is tempting to suggest that the current study supports the approximation for dissipation
based on small scale isotropy over the axisymmetric dissipation approximation. However, as shown
in Figure 10, the axisymmetric dissipation assumption used by Hussein et al.1 appears valid when
calculated and compared with data from the current study. This suggests another possible cause of
the disagreement. When comparing the shear stress balance where no approximations for dissipation
was needed, the current study agrees more with the results from Hussein et al.1

E. Vortical structures and entrainment


Plourde et al.25 examined the effect of individual vortical structures of an axisymmetric shear
flow on the entrainment of ambient fluid into the flow. They related a local instantaneous entrainment
coefficient to the vortical structures in the flow which they visualized by using the λci technique. The
λci technique was suggested by Zhou et al.18 and Chakraborty et al.,26 where λci , is the imaginary
part of the complex eigenvalues of the local velocity gradient tensor. As discussed in their papers, if

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-14 Taub et al. Phys. Fluids 25, 115102 (2013)

FIG. 12. Reynolds stress transport equations of current study (symbols) compared to experimental results by Hussein et al.1
(solid lines) and Panchapakesan et al.2 (dashed lines).

at any point the velocity gradient tensor has all real eigenvalues, then the flow has zero swirl at that
point. If on the other hand the local flow is swirling, then the velocity gradient tensor will have two
complex conjugate eigenvalues, and the value of the imaginary part, λci , is a measure of the strength
of the local swirling motion.
Figure 13 presents the iso-surface of λci = 0.5 calculated at two different instants in time, t1 and
t2 . The value of λci = 0.5 was chosen so as to clearly see the important vortical structures. Both t1
and t2 occur after the jet had become statistically stationary and are instants included in the temporal
averaged data. t1 occurs approximately 350 time scales after the simulation was determined to be
statistically stationary and t2 was at the completion of averaging over 1000 time scales.
While the details of the individual vortical structures at t1 and t2 appear quite different, the
large scale behavior of turbulence in different regions of the jet are similar. Instabilities are seen
forming very close to the origin in the form of mostly varicose vortex rings. A close examination

FIG. 13. 3D iso-surfaces of λci = 0.5 calculated at two different instants of time.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-15 Taub et al. Phys. Fluids 25, 115102 (2013)

of these vortex rings reveals that the instabilities are slightly helical in nature. By z = 5, secondary
instabilities begin to form. In the region between 5 < z < 10, where the flow transitions to full
turbulent behavior, the vortical structures quickly widen. After the jet transitions to fully turbulent
behavior above z > 10, the region of vortical structures widens far more slowly. Beyond an axial
distance of approximately z > 17, the strength of the local swirling motion begins to visibly weaken,
and the structures where λci = 0.5 appear less closely packed.
In their seminal paper on plumes, Morton, Taylor, and Turner27 suggested that the mean radial
velocity at the edge of an axisymmetric boundary layer such as a jet or a plume is proportional to the
axial velocity. That is, Ur  = αUz . They suggested that the coefficient of entrainment, α M , could
be evaluated as
∞
d
Uz r dr
αM =   ∞ dz 0
1/2 . (26)
2 0 Uz 2r dr
Hussein et al.1 showed that for a self similar jet equation (26) is equivalent to
1/2
α M = I1 /2I2 , (27)
where


Uz 
I1 = 2 ηdη (28)
0 Uz,c 
and

∞  2
Uz 
I2 = 2 ηdη. (29)
0 Uz,c 
One can solve for I1 and I2 by using Eq. (9) as an approximation for Uz /Uz, c . Both the current
DNS study and the earlier experimental study of Hussein et al.1 agree on a jet half width of η1/2 =
0.094. As such, both studies agree on the value of the entrainment coefficient, α M = 0.081.
Pham et al.28 suggested an alternate approximation for the entrainment coefficient,
∞
d
Uz (z)r dr
α P (z) = dz 0
, (30)
δ(z)Uz,c 
where δ is the momentum balance length scale defined as
 ∞ 2
2 0 Uz (z)r dr
δ (z) =  ∞
2
. (31)
0 Uz (z) r dr
2

For a self similar jet α P is equivalent to


1
αP =I2 , (32)
2
which, using the data from the current study, can be evaluated as α P = 0.044.
α P (z) and α M (z) can also be evaluated numerically, which enables us to examine the changes
in entrainment near the source of the jet where the entrainment coefficient is not expected to be
constant. Figure 14 plots α M and α P as a functions of z. Here the independent variable, z, is plotted
as the ordinate, so the behavior of entrainment in Figure 14 can be more readily compared to the
behavior of the vortical structures in Figure 13.
Between 0 < z < 2, entrainment increases rapidly from α M = α P = 0 to roughly α M = 0.041
and α P = 0.035. α M continues to grow at a slower rate whereas, α P becomes roughly constant
until z = 5, which is approximately of the same height where secondary instabilities can be seen to
form in Figure 13. Between 5 < z < 6 both definitions for the coefficient of entrainment decrease
briefly before beginning to steadily increase from 6 < z < 15, a region which roughly corresponds
to the region of the jet where vortical structures rapidly widen. As expected above z > 15, where
the jet has achieved self similarity both α M and α P are approximately constant ( α M ≈ 0.080 and
α P ≈ 0.044) and the region of vortical structures widens more slowly. The computed values are in
excellent agreement with those predicted from the self-similar solution.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-16 Taub et al. Phys. Fluids 25, 115102 (2013)

FIG. 14. Entrainment coefficient as defined by Morton et al.,27 α M (z), Eq. (26) (dashed line) and as defined by Pham et al.,28
α P (z), Eq. (30) (solid line) as a function of z.

F. Spectra
Various researchers5, 14, 25, 28, 29 have obtained insights into the behavior of the fluctuating velocity
components by examining the temporal spectra and using the Taylor hypothesis to relate the results
to one dimensional spectra in the axial direction. In the current study one dimensional spectra in the
azimuthal direction were obtained directly by calculating the fourier transform,


1
û i (r, k, z, t) = √ u i (r, θ, z, t)e−ikθ dθ. (33)
2π 0
Numerical evaluation of the fourier transform was with 256 evenly spaced points in the tangential
direction at a given location in the r − z plane. Instantaneous spectra are then averaged over time. In
all, the energy spectra were calculated at eight different points located at r = 0.6, 1.6 and z = 20, 25,
30, and 35. However, very little variation was seen between spectra examined at the different axial
positions.
Figure 15 plots the energy spectra of the axial, radial, and tangential components of velocity at
two different radial positions r = 0.6, 1.6 at z = 30. In the figure the frequency, k, has been divided
by the radial distance from the centerline so that the frequencies at both radial positions represent the
same wavelengths in the azimuthal direction. The straight lines in the figure represent k−5/3 and k−7 .
The −5/3 slope is relevant in the inertial range and here it can be seen that this behavior dominates
only for very low frequencies. This is similar to the results of Bogey and Bailly14 who compared
the one dimensional spectrum of the axial velocity along the centerline of several jets with varying
Reynolds numbers. A slope steeper than −5/3 is an indication that dissipation is playing a larger
role in the energy cascade. As dissipation becomes dominant at higher frequencies a flow dependent
steeper slope can be expected. For example, in buoyant jets and plumes a −3 slope is observed in
the dissipative regime.25, 28 The value of −7 observed in the dissipative regime of the current study
is identical to that found by Fellouah et al.29
At both locations the spectra of all three velocity components were found to have similar
behavior. This suggests that the small scale isotropic assumption is valid at these points.

FIG. 15. One dimensional velocity spectra in the azimuthal direction for each velocity component at two different locations
on the r − z plane.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-17 Taub et al. Phys. Fluids 25, 115102 (2013)

IV. CONCLUSION

A DNS study of an axisymmetric turbulent jet was completed. First and second order statistics
were found to agree well with previous studies in the similarity region of the jet. The moderate
Reynolds number jet simulated in the current study was found to achieve constant centerline turbulent
intensities closer to the inlet, than previous experimental studies where jets with larger Reynolds
numbers were examined. This agrees with an assertion by Bogey and Bailly14 that moderate Reynolds
number turbulent jets achieve similarity closer to the source than larger Reynolds number jets. Indeed,
the jet in the current study (ReJ = 2000) achieved self similarity within 15 inlet diameters from the
jet source, whereas the jet studied by Panchapakesan and Lumley2 (ReJ = 11 000) did not achieve
self similarity until z = 70.
The total momentum integral was calculated and M(z) was found to be within 1% of MJ for all
z values above z > 10. The momentum flux contributed by the mean flow was found to account for
approximately 93% of the total momentum integral.
The six non-zero components of the third order statistics were presented and compared to
previous studies. Previous experimental studies found that the cross terms, u z u r2 , u r u 2 z , and
u z u 2
θ  are negative near the centerline of the jet but did not report on the fourth cross term u  2
r u θ .
The current study agreed with the previous studies in this regard. The fourth cross term u r u 2 θ 
was also calculated but was found to be positive for all values of r/z. There is some disagreement
between the results of Hussein et al.1 and Panchapakesan and Lumley2 in the u 3 3
z  and u r  terms,
3
the current study compares well with Panchapakesan et al. in the case of u z  but agrees better
2

with Hussein et al.1 in the case of u r3 .


Whereas previous studies made use of various assumptions in order to approximate the pressure
diffusion and dissipation terms in the turbulent kinetic energy balance, in the current study each
term was calculated directly. The validity of these assumptions have been examined by comparing
the approximated terms with the values of the terms used in the approximation with data from the
current study. Perhaps due to the use of different approximations, there has been some disagreement
in the experimental studies, as to the shape of the dissipation profile. With Hussein et al.1 reporting
a Gaussian profile, others such as Shiri suggesting an off axis maximum value, and Panchapakesan
and Lumley2 suggesting a rather flat profile near the centerline. It is important to remember that none
of the experimental studies were able to calculate all dissipation terms directly. Panchapakesan and
Lumley2 ignored the pressure term and solved for dissipation by closing the balance. This appears
to be a valid assumption as the current study agrees well with their data, both in the shape of the
dissipation term and the relative size of the pressure diffusion term. An earlier LES study by Bogey
and Bailly16 also agrees with Panchapakesan and Lumley2 in the magnitude of the dissipation term
although that study did not report the nearly constant dissipation value near the jet centerline seen by
the earlier experimental study and this current study. While the current study does not agree with the
magnitude or shape of the dissipation term found by Hussein et al.,1 the axisymmetric dissipation
approximation they used to calculate dissipation was found to be valid when tested using data from
the current study. Some researchers including Mi et al.24 and Bogey and Bailly14 have suggested
that Reynolds number plays an important role in determining flow characteristics such as dissipation
rate, even for very high Reynolds number jets. As such, the disagreements in the discussed studies,
may have more to do with the large range of Reynolds numbers examined, than the various models
used to approximate the various terms of the turbulent kinetic energy balance.
In examining the Reynolds stress transport equations Panchapakesan and Lumley2 used different
assumptions than when calculating the turbulent kinetic energy equation. Here they approximated
dissipation using a small scale isotropic assumption and solved for pressure by closing the balance.  2 
Both Hussein et al.1 and Panchapakesan and Lumley 2
 2  generally
 2  agree on the form of the u z
balance as does the current study. The balance of the u r and u θ Reynolds normal stress terms in
the current study are similar to that of Panchapakesan et al.,2 but care should be taken in dismissing
the axisymmetric dissipation assumption suggested by Hussein et al.1 as this approximation appears
quite good when tested using data from the current study. There is also some disagreement between
the two experimental studies in regard to the magnitude  of the production and pressure diffusion
terms in the Reynolds shear stress transport equation, u z u r . Here the current study agrees more

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-18 Taub et al. Phys. Fluids 25, 115102 (2013)

closely to the results of Hussein et al.1 Hopefully a better understanding of the transport of all four
Reynolds stress terms can be used to improve LES and RANS simulations which rely on modeling
these terms.
Vortical structures and entrainment were discussed and it was suggested that the behavior of the
large scale vortical structures affected the rate of entrainment.
The one dimensional velocity spectra in the azimuthal direction was calculated directly. The
−5/3 slope indicative of the inertial regime of turbulent flows was observed as well as a higher
wave number region of −7 slope which was also observed by Fellouah et al.29 to be present in the
dissipative regime of a pure axisymmetric jet.

ACKNOWLEDGMENTS

This study is part of a larger project in which the turbulent behavior of pure jets, plumes, and
forced plumes are compared. In the near future the DNS code is to be modified in order to examine
the turbulent behavior of multiphase forced plumes. This research is funded by the National Science
Foundation (NSF OISE-0968313) through the Partnership for International Research and Education
(PIRE).

1 J. Hussein, S. Capp, and W. George, “Velocity measurements in a high-Reynolds-number, momentum-conserving, axisym-

metric, turbulent jet,” J. Fluid Mech. 258, 31–75 (1994).


2 N. Panchapakesan and J. Lumley, “Turbulence measurements in axisymmetric jets of air and helium. Part 1. Air jet,” J.
Fluid Mech. 246, 197–233 (1993).
3 H. Schlichting, Boundary Layer Theory, 4th ed. (McGraw-Hill, 1960).
4 M. Kung, “Large eddy simulation of turbulent channel and jet flows using the approximate deconvolution model,” Ph.D.

thesis, Swiss Federal Institute of Technology, Zurich, Switzerland, 2007.


5 Z. Wang, P. He, Y. Lv, J. Zhou, J. Fan, and K. Cen, “Direct numerical simulation of subsonic round turbulent jet,” Flow,

Turbul. Combust. 84, 669–686 (2010).


6 M. Amielh, T. Djeridane, F. Anselmet, and L. Fulachier, “Velocity near-field of variable density turbulent jets,” Int. J. Heat

Mass Transfer 39, 2149–2164 (1996).


7 B. Boersma, G. Brethouwer, and F. Nieuwstadt, “A numerical investigation on the effect of the inflow conditions on the

self-similar region of a round jet,” Phys. Fluids 10, 899 (1998).


8 C. Lubbers, G. Brethouwer, and B. Boersma, “Simulation of the mixing of a passive scalar in a round turbulent jet,” Fluid

Dyn. Res. 28, 189–208 (2001).


9 P. Babu and K. Mahesh, “Upsteam entrainment in numerical simulations of spatially evolving round jets,” Phys. Fluids 16,

3699 (2004).
10 P. Brancher, J. Chomas, and P. Huerre, “Direct numerical simulations of round jets: Vortex induction and side jets,” Phys.

Fluids 6, 1768–1774 (1994).


11 J. Freund, “Noise sources in a low-Reynolds-number turbulent jet at Mach 0.9,” J. Fluid Mech. 438, 277–305 (2001).
12 I. Wygnanski and H. Fiedler, “Some measurements in the self-preserving jet,” J. Fluid Mech. 38, 577–612 (1969).
13 M. Olsson and L. Fuchs, “Large eddy simulation of the proximal region of a spatially developing circular jet,” Phys. Fluids

8, 2125–2137 (1996).
14 C. Bogey and C. Bailly, “Large eddy simulations of transitional round jets: Influence of the Reynolds number on flow

development and energy dissipation,” Phys. Fluids 18, 065101 (2006).


15 K. Akselvoll and P. Moin, “Large-eddy simulation of turbulent confined coannular jets,” J. Fluid Mech. 315, 387–411

(1996).
16 C. Bogey and C. Bailly, “Turbulence and energy budget in a self-preserving round jet: direct evaluation using large eddy

simulation,” J. Fluid Mech. 627, 129–160 (2009).


17 S. Muppidi and K. Mahesh, “Direct numerical simulation of round turbulent jets in crossflow,” J. Fluid Mech. 574, 59–84

(2007).
18 J. Zhou, R. Adrian, S. Balachandar, and T. Kendall, “Mechanisms for generating coherent packets of hairpin vortices in

channel flow,” J. Fluid Mech. 387, 353–396 (1999).


19 H. Aissia, Y. Zaouali, and G. El Salem, “Numerical study of the influence of dynamic and thermal exit conditions on

axisymmetric laminar buoyant jet,” Numer. Heat Transfer A 42, 427 (2002).
20 H. Lee and S. Balachandar, “Drag and lift forces on a spherical particle moving on a wall in a shear flow at finite Re,” J.

Fluid Mech. 657, 89–125 (2010).


21 G. N. Taub, H. Lee, S. Balachandar, and S. A. Sherif, “A study of similarity solutions for laminar swirling axisymmetric

flows with both buoyancy and initial momentum flux,” Phys. Fluids 23, 113601 (2011).
22 A. Shiri, “Turbulence measurements in a natural convection boundary layer and a swirling jet,” Ph.D. thesis, Chalmers

University of Technology, Göteborg, Sweden (2010).


23 J. Lumley, “Computational modeling of turbulent flows,” Adv. Appl. Mech. 18, 123–176 (1979).
24 J. Mi, M. Xu, and T. Zhou, “Reynolds number influence on statistical behaviors of turbulence in a circular free jet,” Phys.

Fluids 25, 075101 (2013).

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19
115102-19 Taub et al. Phys. Fluids 25, 115102 (2013)

25 F. Plourde, M. Pham, S. Kim, and S. Balachandar, “Direct numerical simulations of a rapidly expanding thermal plume:
Structure and entrainment interaction,” J. Fluid Mech. 604, 99–123 (2008).
26 P. Chakraborty, S. Balachandar, and R. Adrian, “On the relationships between local vortex identification schemes,” J. Fluid

Mech. 535, 189–214 (2005).


27 B. Morton, G. Taylor, and J. Turner, “Turbulent gravitational convection from maintained and instantaneous sources,”

Proc. R. Soc. London 234, 1–23 (1956).


28 M. Pham, F. Plourde, S. Kim, and S. Balachandar, “Large-eddy simulation of a pure thermal plume under rotating

conditions,” Phys. Fluids 18, 015101 (2006).


29 H. Fellouah, C. Ball, and A. Pollard, “Reynolds number effects within the development region of a turbulent round free

jet,” Int. J. Heat Mass Transfer 52, 3943–3954 (2009).

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.195.150.201 On: Tue, 28 Oct 2014 05:42:19

You might also like