Application of PDF methods to compressible turbulent ¯ows

Okay, in this article PDF methods are applied to compressible turbulent ¯ows.

B J Delarue as well as S P S B Pope.

128 Corner of Roberts and East Avenue, Ithaca, New York 14853, Sibley School of Mechanical
and Aerospace Engineering, Cornell University

~Received 30 September 1996; accepted 13 May 1997 ~

An example of a particle method is the so called probability density function ‘~PDF! The analysis
of a scalar ¯eld in a compressible turbulent flow is provided in the following section. The method is
applied to the turbulence in the compressible mixing layer, and comparison reveals good result
with experimental data. The derived PDF equation is applied by solving it through the
Lagrangian/Monte Carlo method. Thus, to consider the effects of compressibility in the analysis of
turbulent segregations, the IPTS PDF formulation is expanded to account for thermodynamic
properties like the pressure and the internal energy of the ¯uid. Thus it is seen that the mean
pressure which has been the focus of determination through intensive calculations in the last few
years is derived from the particle properties. It is thus not necessary to link the PDF solver with a
finite-volume type solver which may be more time for the computation of results. The stochastic
differential equations – SDE! the compressible particle – property models are derived from
second-order closures for compressed turbulence being valid for small turbulent – Mach number
flows only. These are carried out in decaying isotropic turbulence to examine the accuracy of the
PDF method in comparison to the Reynolds stress closures from which it is extrapolated as well
as in homogeneous shear flows where comparison is made with direct numerical simulation or
‘’DNS’’ data is conducted. The model is then used to simulate the plane compressible mixing layer
and reproduces the known behaviour where the rate of spreading reduces with increase in
compressibility.

ulence. Vital theoretical achievements have been obtained and discussed:1 Of course, one
should not forget that scientific research gives a set of experimental data which allows to
determine the nature of certain phenomena and their regularity, to find important characteristics of
new materials, as well as to obtain theoretical results and bring them to the discussion.
dinary PDE solvers unnecessary. The modelled transport
equation for the joint PDF of velocity and composition has
been successfully solved, using sets of stochastic particles
I. INTRODUCTION
with time-evolving properties to model ¯uid particles.21,22
Over the past few years, renewed interest in supersonic Recent works include the development of models for the
aircraft and high-speed combustion has emphasized the need for velocity-dissipation joint PDF.23,24 However, the majority of
research in the ®eld of compressible turbulence. Impor-tant applications is limited to low Mach number ¯ows and, with-out
theoretical results have been established and reviewed, 1 coupling to a ®nite-volume type solver to obtain quanti-
allowing a better understanding of the complex phenomena
involved in compressible turbulence. Extensive experimental a! Present
address: Electricite de France, Laboratoire National
work has been conducted, especially in the case of the plane d'Hydraulique, 78401 Chatou, France.
compressible mixing layer.2±9 In the turbulence modelling
community, the limitations of existing incompressible turbu-
lence models have been established.10 Second-order closures
have been designed to represent explicit compressibility ef-
fects, such as the compressible dissipation and the pressure±
dilatation correlation.11±18 The need for future research, in both
understanding and modelling the effects of compress-ibility on
turbulence, has been clearly established.10,19
For ¯ows involving combustion, probability density
function ~PDF! methods have demonstrated their ability to
treat the important processes of reaction and convection
exactly,20 making transport and reaction models used in or-

ties such as the mean pressure, to ¯ows with weak pressure
gradients.
Only recently have PDF methods been applied to ¯ows
with pressure-induced density variations,25±27 requiring in
general ~except reference 27! coupling with a ®nite-volume
solver of some kind. The problem of determining the mean
pressure directly from the particle properties ~i.e., without a
®nite-volume solver! is closely related to the dif®culties en-
countered in trying to extend PDF formulation to complex
compressible ¯ows.
The objective of the present work is to extend the exist-
ing PDF models to compressible reacting ¯ows with arbi-
trary pressure gradients, with the aim of developing a stand-
alone method to solve for the joint PDF of all relevant ¯ow
variables, including the dissipation rate of turbulent kinetic
2704 Phys. Fluids 9 (9), September 1997 1070-6631/97/9(9)/2704/12/$10.00
Downloaded 22 Sep 2004 to 140.121.120.39. Redistribution subject to AIP license or copyright, see http://pof.aip.org/pof/copyright.jsp
energy, without coupling with a pressure algorithm. By do-
ing so we hope to exploit fully, in a simple and computation-
ally ef®cient way, the remarkable potential offered by PDF
methods to solve for complex turbulent reacting ¯ows.
For the de®nition of the method, henceforth, we restrict
ourselves to nonreacting ¯ows. It is believed, however, that
extension of the method to account for the reaction will be
straightforward. The method has been successfully imple-
mented for homogeneous compressible ¯ows ~hence it deals
with unsteady ¯ows without coupling with a ®nite-volume
solver, a feature displayed only by recent work using smoothed
particle hydrodynamics27!, and statistically station-ary
inhomogeneous ¯ows. A unique feature of the method is the
inclusion in the joint PDF of two extra thermodynamic
variables, namely the pressure and the internal energy. Hence,
all statistics of the ¯ow, including the mean pressure, can be
determined directly from the particle properties. A
© 1997 American Institute of Physics
®nite-volume type solver is therefore unnecessary. In addi- where «s 5n^v8i v8i& is the standard solenoidal dissipation,
tion, models have been derived from existing Reynolds and «d 54/3n^u 8i, i 2& is the so-called dilatation
stress closures to account for the compressible dissipation dissipation, which is clearly zero in incompressible ¯ows,
and the pressure±dilatation correlation. and strictly positive in compressible ¯ows, thus amounting to
In section II we review some general effects of com- extra dis-sipation. We see that in ~1! three explicit terms
pressibility on turbulent ¯ows, as well as second-order clo- arising from compressibility need to be modelled: «d , Pd ,
sures from which our PDF formulation is inspired. It is em- and the last term, which arises because the Reynolds
phasized that the purpose of this work is neither to develop averages ^u 9i& of Favre ¯uctuations are nonzero. The last
new models for compressibility effects on turbulence, nor to term is probably very small in ¯ows without large pressure
test existing models, but to develop an innovative and con- gradients and away from walls. We therefore choose to
sistent PDF formulation designed for high-speed combus- neglect it. We need, however, to model the other two.
tion. In section III we detail the PDF formulation which is For the dilatation dissipation, we use the model of
the object of this work, summarizing brie¯y the general idea Sarkar et al.11 relating «d to «s in the following way:
behind PDF methods, then de®ning our stochastic variables
and the corresponding stochastic differential equations «d 5C d M t2 «s , ~2!
~SDE!. In section IV we present results for homogeneous 2 Ä2 Ä
where M t 52 k / a is the turbulent Mach number squared, a
¯ows. In section V, we extend the model to inhomogeneous
being the mean speed of sound. This model was developed
¯ows. At this point a comparison is made between model
for M t !1. The constant C d is of order 1. Equation ~2!
results and experimental data in the supersonic mixing layer
relates the dilatation dissipation to the solenoidal dissipation.
case.
It has been argued11,14 that the energy cascade responsible
for the latter is moderately affected by compressibility,
II. MODELS FOR THE EFFECTS OF
therefore standard incompressible models for the solenoidal
COMPRESSIBILITY ON TURBULENCE dissipa-tion can be used in the present situation.
In compressible reacting ¯ows, density variations arise For the pressure dilatation, we use Zeman's model,15
because of variations in chemical composition, and of tem- which can be summarized in the following two equations:
perature and pressure ¯uctuations. While all these effects 1 D ^ p 8 2&
contribute to create nonzero dilatation rates, only the latter Pd 52 , ~3!
are termed compressibility effects. In the following discus- 2 g^p& Dt
sion, we consider an inert ¯ow and therefore ignore the ef- D ^ p 8 2& ^ p 8 2 &2 p e 2
fects of a chemical reaction on the density. 52 . ~4!
Consider the equation for turbulent kinetic energy in Dt ta
compressible ¯ow: The ®rst equation is valid in homogeneous turbulence for M t
!1, and for high Reynolds and PeÂclet numbers. The sec-ond
]k Ä ]k ] ^ p& equation is fully modelled, relying on the results of Sarkar et
5T 1 P 1P

S ]x i
^r& 2^r&«2^u 9 &
i]x
1^r&U

al.,11,28 and stating that pressure variance relaxes to an

]t i d i

]^ ti j & equilibrium level p e on the acoustic time scale ta . These two

2 ]x j D, ~1! model quantities are de®ned in Zeman.15 The pressure±
dilatation is expected to be important in nonequilibrium ¯ows,
where
namely in ¯ows with a strong dependence on initial conditions,
for instance decaying isotropic turbulence. It is expected to be
T 52 ] 1 Ä
of lesser importance in equilibrium ¯ows, for
S2 D
^r&u 9u 9u 9 1^p 8u 9 &2^u 9t8 &

]x i j j i i j ij instance most free shear ¯ows without shocks. To date, most

is the transport term, models for Pd 12,15,16 are restricted to weakly inhomogeneous
Ä turbulence.
P 52 r
Ä
u 9u 9
]U i The models for both «d and Pd are restricted to ¯ows
^ & i j ]x j with M t !1. In our calculations, for example, in mixing lay-
is the production term, ers with free stream Mach numbers as high as 6.5, we never

Pd 5 K p8
]x i
]u 8i L encountered values of M t above 0.5, which has been consid-
ered to fall within the range of applicability of these models.
The above discussion is centered exclusively on model-
ling from the point of view of the turbulent kinetic energy
is the trace of the pressure-rate of strain correlation, also
equation, and not from the Reynolds stresses equation. The
called pressure dilatation, and « is the viscous dissipation.
models we have chosen will affect the turbulence in an iso-
In ~1!, the brackets correspond to Reynolds averages and the
primes to ¯uctuations about these averages, while the tildes tropic manner. Though it is known at this point 19,10 that com-
and double primes stand for Favre averages and ¯uctuations, pressibility also, and probably chie¯y, affects the turbulence
by modifying the deviatoric pressure-rate of strain correla-
respectively. It has been shown11,14 that the viscous dissipa-
tion so that the redistribution of energy does not function as
tion can be split into two terms:
well when the compressibility level is high, resulting in in-
«5«s 1«d , creased anisotropies in the normal stresses and decreased an-

Phys. Fluids, Vol. 9, No. 9, September 1997 B. J. Delarue and S. B. Pope 2705

Downloaded 22 Sep 2004 to 140.121.120.39. Redistribution subject to AIP license or copyright, see http://pof.aip.org/pof/copyright.jsp
 FDA 95-05, Cornell University, 1995.
6
W. Kollmann, ``The pdf approach to turbulent ¯ow,'' Theor. Comput. Fluid Dyn. 1, 249
4 ~1990!.
A. W. Vreman, N. D. Sandham, and K. H. Luo, ``Compressible mixing layer growth rate
and turbulence characteristics,'' J. Fluid Mech. 320, 235 ~1996!.
B. J. Delarue and S. B. Pope
5
Jayesh and S. B. Pope, ``Stochastic model for turbulent frequency,'' Tech-nical report,

Downloaded 22 Sep 2004 to 140.121.120.39. Redistribution subject to AIP license or copyright, see http://pof.aip.org/pof/copyright.jsp
33 36
K. N. C. Bray, P. A. Libby, and F. A. Williams, ``High speed turbulent combustion,'' S. B. Pope, ``Position, velocity and pressure correction algorithm for par-ticle method
in Turbulent Reacting Flows, edited by P. A. Libby ~Aca-demic, New York, 1994!, solution of the pdf transport equations,'' Technical report, FDA 95-06, Cornell University,
Chap 10. 1995.
34 37
D. W. Bogdanoff, ``Compressibility effects in turbulent shear layers,'' AIAA J. 21, 926
D. C. Haworth and S. B. Pope, ``A generalized Langevin model for tur-bulent
¯ows,'' Phys. Fluids 29, 387 ~1986!. ~1983!.
35 38
G. L. Brown and A. Roshko, ``On density effects and large structures in turbulent mixing
S. B. Pope, ``On the relationship between stochastic Lagrangian models of
turbulence and second-moment closures,'' Phys. Fluids 6, 973 ~1994!. layers,'' J. Fluid Mech. 64, 775 ~1974!.

Phys. Fluids, Vol. 9, No. 9, September 1997 B. J. Delarue and S. B. Pope 2715

Downloaded 22 Sep 2004 to 140.121.120.39. Redistribution subject to AIP license or copyright, see http://pof.aip.org/pof/copyright.jsp