c~ Computer methods

in applied
mechanics and
engineering
ElSEVIER Comput. Methods App1. Mech. Engrg. 161 (1998) 229-243

A unified finite elementformulation for compressibleand

incompressibleflows using augmentedconservationvariables
s. Mittala, T. Tezduyarb,*
"Department of Aerospace Engineering, Indian Institute of Technology, Kanpur, UP 208 016, India
bDepartment of Aerospace Engineering and Mechanics, Anny High Performance Computing Research Center, University of Minnesota,
1100 Washington Avenue South, Minneapolis, MN 55415, USA

Received 6 November 1997

Abstract

A unified approach to computing compressible and incompressible flows is proposed. The governing equation for pressure is selected
based on the local Mach number. In the incompressible limit the divergence-free constraint on velocity field determines the pressure, while it
is the equation of state that governs the pressure solution for the compressible flows. Stabilized finite element formulations, based on the
space-time and semi-discrete methods, with the 'augmented' conservation variables are employed. The 'augmented' conservation variables
consist of the usual conservation variables and pressure as an additional variable. The formulation is applied to various test problems
involving steady and unsteady flows over a large range of Mach and Reynolds numbers. @ 1998 Elsevier Science S.A. All rights reserved.

1. Introduction

Computationalmethodsto solve flow problemsfall mainly into two categories:(a) methodsfor compressible
flows and (b) methodsfor incompressibleflows. Thesetwo casesof methodsare quite different from eachother
with respect to the choice of variables, issuesrelated to numerical stability and choice of solvers.Various
researchersin the past have proposedideas for a unified approachto compressibleand incompressibleflows.
Turkel [1] suggesteda preconditioning method to acceleratethe convergenceto a steady state for both the
compressibleand incompressibleflow equations.Hauke and Hughes [2] and Hauke [3] presenteda finite
elementformulation for solving the compressibleNavier-Stokes equationswith different setsof variables.They
also showedthat in the context of primitive or entropy variables,the incompressiblelimit is well behavedand
therefore,one formulation can be usedfor solving both compressibleand incompressibleflows.Weissand Smith
[4] proposeda preconditioningtechniquein conjunction with a dual time-stepprocedureto computeunsteady
compressibleand incompressibleflows with density-basedvariables.Karimian and Schneider[5] presenteda
collocatedpressure-based method that works in both compressibleand incompressibleregimes.
In this article we presentan alternate,unified approachto computingcompressibleand incompressibleflows
using 'augmented' conservationvariables formulation. Compressibleflows have been computedby several
researchersin the past with the conservationvariablesformulation [6-13]. It was shown by Hauke [3], that in
the incompressiblelimit, the Euler Jacobiansfor the formulation employing conservationvariablesis not well
behaved.In the incompressiblelimit, since density becomesconstant,some of the coefficients must go to
infinity to accommodatefinite variations in pressure.It has also been shown by Panton [14] that in the

* Corresponding author.

0045-7825/98/$19.00 @ 1998 Elsevier Science S.A. All rights reserved.

PII: SO045-7825(97)00318-6
230 S. Mittal, T. Tezduyar I Camput. Methods Appl. Mech. Engrg. 161 (1998) 229-243

incompressiblelimit, the stateequationdegeneratesto a result accordingto which the densityof a fluid particle
is constant.This relation along with the massbalanceequationleads to the divergence-freeconstraint on the
velocity field that can be used to determine the pressurein incompressibleflows. The formulation that we
proposein this article is basedon the philosophythat pressureis determinedby the equationof statewhen the
flow is compressible,whereasit is determinedby the divergence-freeconstraintwhen the flow is incompress-
ible. To this end we employ the 'augmented'conservationvariableswhich consist of the usual conservation
variables(density, momentaand energy) and pressureas an additional variable.
We begin by reviewing the governingequationsfor compressibleand incompressiblefluid flow in Section2.
The equationsare cast in a non-dimensionalform and a parameterz, basedon the local Mach number, is
introduced that governs the choice of equations for compressibleand incompressibleflows locally in the
computationaldomain. The stabilized space-time variational formulation of these equationsin terms of the
augmentedconservationvariablesis presentedin Section 3. The SUPG (streamline-upwind/Petrov-Galerkin)
stabilization technique [6,7,9,13,15,16]is employed to stabilize our computationsagainst spurious numerical
oscillations.In Section4 we presentsomenumericalresults to test the performanceof the proposedformulation.
We begin with the computationof the shock-reflectionproblem that involves three flow regionsseparatedby an
oblique shockand its reflection from a wall. The exact solution for this problemis known and is comparedwith
the computedsolution. Next, supersonicflow past a cylinder at Mach 2 and Re 2000 is computedwith the
unified formulation and with the compressibleflow formulation basedon the equationof state.Finally, results
are presentedfor unsteadyflow past a cylinder at Re 100. These computationsare carried out for different
subsonicMach numbersincluding the incompressiblelimit.

2. The governingequations

Let {},tC ~n'd and (0, T) be the spatial and temporaldomains,respectively,where nodis the numberof space
dimensions,and let 1; denotethe boundaryof ,{},t'The spatial and temporalcoordinatesare denotedby x and t,
The Navier-Stokes equationsgoverning the fluid flow, in conservationform, are
iJp
iJt-+V'(pu)=O on,{},tfor(O,T), (1)

-iJ(pu)
at
+ v. (puu) + Vp - V. T =0 on ll, for (0, T) ,
(2)

a(pe) + V. (peu)+ V. (pu) -V. (Tu)+ V. q = 0 on n, for (0, T)

at (3)

Here, p, U, p, T, e, andq are the density,velocity, pressure,viscousstresstensor,total energyper unit mass,and

heat flux vector, respectively.The viscous stresstensor is define~as
T = ,u«Vu) + (VU)T)+ }"(V. u)1 . (4)

where ,u and },.are the viscosity coefficients.It is assumedthat ,u and },. are related by
2
},.= -3 ,u . (5)

Pressureis relatedto the other variablesvia the equationof state.For ideal gases,the equationof stateassumes
the special form
p=(y-l)pi, (6)
where y is the ratio of specific heats,and i is the internal energy per unit masswhich is related to the total
energy per unit massand kinetic energy as

(7)

The heat flux vector is defined as

 S. Mittal, T. Tezduyar I Comput. Methods Appl. Mech. Engrg. 161 (1998) 229-243 231

q = - KVO, (8)
where K is the heatconductivity and fJ is the temperature.the temperatureis relatedto the internal energyby the
following relation:

(J=~ (Q)
Cv
where Cv is the specific heat of the fluid at constantvolume. For an ideal gas
R
Cv =-:y-=-T' (IV)

whereR is the ideal gas constant.Prandtl number(Pr)' assumedto be specified,relatesthe heat conductivity of
the fluid to its viscosity accordingto the following relation:

K= JLCp
p , (11)
r
where Cp is the specific heat of the fluid at constantpressure.For an ideal gas

C = - yR
p y-l

In the limit of incompressibleflows, i.e. when the Mach number approacheszero, the above-mentionedset of
equationsassumea new form. It can be shown [14], that the stateequationalong with the massbalanceequation
lead to the following relation:
pV'u =0. (13)
Using the relation, Eqs. (1), (2) and (3) can be modified for incompressibleflows as
iJp
iJt-+ V. (pu) - pV' u = 0 on fl, for (0, T) ,
(14)

a(pu) + v. (puu) - puV. u + Vp -v. T=O on [}" for (O,T),

at

a(pe)
-ae-+ V. (peu) - peV.u + V. (pu) -pV' u -v. (Tu) + V. q = 0 on {It for (0, T). (16)

In this situation, the viscous stresstensor, given by Eq. (4) can be rewritten as
T = jL«Vu) + (VU)T) . (17)

It is possibleto combinethe two setsof governingequationsfor the compressibleand incompressibleflows and

expressthem in terms of non-dimensionalvariables.The non-dimensionalvariablesthat we chooseare
x u
u* =-:::- t* = tU~
L ' p*=Lp~
x*=L' U",'
p-p~
(8 - 8~)Cp
p*= 2 .
8* =
p~U~ U2~
whereall the quantitieswith the subscript'00' refer to the free-streamvaluesof the flow variables.The
governingequationsin the non-dimensionalvariablesthat are valid over the entire range of compressibleand
incompressibleflows are
iJp* + V*.
at*" (p*u*) - (1 - z)p*V*. u* = 0 on [J~ for (0, T*), (19)

iJ(p*u*) + v* . (p*u*u*) - (1 - z)p*u*V* . u* + V* p* - V* . T* = 0 on [J~ for (0, T*) ,

at
232 S. Mittal, T. Tezduyar I Comput. Methods Appl. Mech. Engrg. 161 (1998) 229-243

The Mach number (M) is defined as the ratio of the flow speedto the speedof sound; M~ refers to the
free-streamMach number.In the aboveequationsz(M) E [0, 1] is a function of Mach numbersuchthat z(O)= 0
and z(M ~ Mc) = 1, whereMc is a 'cut-off Mach number,decideda priori. All the resultsreportedin this article
are with Mc = 0.3 and with the following definition of z:

M
M~M,
z=' Mc (23)
M>Mc
The non-dimensionalviscous stresstensor and heat flux vector are defined as

T* = -1 (V*u*) + (V*U*)T - -2 z(V* .u*)1 (24)

Re I 3

q* = --=- 1 V*(J* (25)

R~Pr
Here, Re is the ReynoldsNumber defined as
p~U~L
R.=
-C JL (26)
In the rest of the article we will work with the non-dimensionalvariablesonly and therefore,the superscript' * '
will be dropped.The governing equations(19)-(22) can be written in the augmentedconservationvariables

.. a~ aF; aE;+ B.!!!!. + SU = 0 on n, for (0, T) , (27)

+~ I ax; I ax.
I

where U = (p, pu., puv p, pe), is the vector of augmentedconservationvariablesand M is a diagonalmatrix

definedas M = diag(l, I, 1,0, I). The various terms involving z and (I - z) in Eqs. (19)-(22) contributeto the
terms involving Bj and S in Eq. (27). Fj and Ej are, respectively,the Euler and viscousflux vectorsdefinedas
u;p \
u;pu. + 5jlP
F,= ujpu2 + 5;2P (28)
0
uj(pe +p) I
0
Til
Ei= Ti2 (29)
0
-q '.+7
J ikU

Here, Ui and qj are the componentsof the velocity and heat flux vectors, respectively, and Tikare the components
of the viscous stress tensor. In the quasi-linear form, Eq. (27) is written as

au
Mat+(Aj+Bj)~-~ au a ( Kjj~ au ) +su=O on {},t for (O,T),
(30)
I I J

whprp
 S. Mittal, T. Tezduyar I Comput. Methods Appl. Mech. Engrg. 161 (1998) 229-243 233

of.

Ai= --!.au'
is the Euler JacobianMatrix, and Kij is the diffusivity matrix satisfying
au
K.. -;-- =E.
IJ ux. I
J

Correspondingto Eq. (30), appropriateboundary and initial conditions are chosen

3. Finite element formulation

In order to construct the finite element function spacesfor the space-time method, we partition the time
interval (0, T) into subintervalsIn = (tn,tn+,), where tn and tn+1 belong to an ordered series of time levels
0 = to< t, < ... < tN = T. Let {In = {J, and 1;:= r;. We define the space-time slab Q as the domainenclosed
n n n
by the surfaces{In' {In+'' and Pn' where Pn is the surfacedescribedby the boundary r; as t traversesIn' The
surfacePn is decomposedinto (Pn)g and (Pn)h with respectto the type of boundary condition (Dirichlet or
Neumann)being imposed.For eachspace-time slab we define the correspondingfinite elementfunction spaces
.:;'hand rho Over the elementdomain,this spaceis formed by using first-order polynomials in spaceand time.
Globally, the interpolation functions are continuousin spacebut discontinuousin time.
h n find Uh Er such that 'v' Wh E
The stabilized space-time formulation is written as follows: given (Uh)-,

f (aWh)
r,

f Wh.Mh_dQ+
auh
- -e.

B' au'
Qn at Qn aXj (-F~+E~)dQ+f Wh.( , ax;
T(A:)T ( -a;; ) . [ Mh- at j"
aWh
--, - auk
+ (A h + Bh) -
Q,
(
-- - a Kh
,
~\.-, ,-
\£&, ',aXi
'ax iia;;) +ShUh] dQ

- , ;. , ","
Qe "

( Wh . hh dP
(33)
J(Pn)h

Here, hh representsthe Neumannboundarycondition imposedand (Pn)h is the part of the slab boundarywith
suchconditions.thesolutionto (33) is obtainedsequentially
for all space-timeslabsQQ' QI' Q2"'" QN-I and
the computationsstart with
(U h-)0 = U 0'
(34)
where U 0 is the specified initial condition.

REMARKS
(1) In the variational fonnulation given by Eq. (33), the first threeintegralsand the right-handside constitute
the Galerkin fonnulation of the problem. Both the Euler and viscous flux tenns are integratedby parts.
This fonn of the variational fonnulation ensuresthat, in the presenceof shocks,the methodgives right
jump conditionsand shocklocation. The Neumannboundarycondition at the outflow boundaryinvolves
the nonnal componentsof the stressvector and momentumflux. In the limit of incompressibleflows
(wherethe pressureis specifiedonly upto a constantand one needsto define a datum pressure)this fixes
the pressureat the boundary.To computeflows that involve free surfacesthe weak fonn given by Eq.
(33) has to be modified by carrying out the integration-by-partof the time-dependenttenn.
(2) The first seriesof element-levelintegralsin Eq. (33) are addedto the variational fonnulation to stabilize
the computationsagainstnumerical instabilities. In the advection-dominatedrange, thesetenns prevent
the node-to-nodeoscillationsof the flow variables.In the limit of incompressibleflows, the inclusion of

~
234 S. Minai, T. Tezduyar I Comput. Methods Appi. Mech. Engrg. 161 (1998) 229-243

these terms allows one to employ equal-order-interpolationfor velocity and pressure.The choice of
stabilizationcoefficient 'Tis quite different for compressibleand incompressibleflows. In the contextof a
unified formulation, one would like to design the stabilization coefficient 'T such that it reducesto the
appropriatedefinitions in the two limits. The secondseriesof elementlevel integralsin Eq. (33) are the
shock capturing terms that stabilize the computationsin the presenceof sharpgradients.The coefficient
of shock-capturingoperator,fJ is sameas defined in [13]. The stabilizationcoefficient 'T is definedas
'T = max[O, 'Ta- 'TpJ

where 'Tais a diagonal matrix defined as 'Ta= diag(T), T), T), T2' T),

7',= +
*) 2) -1/2
(36)

(37)

A= Reu~3
(38)
Reu> 3

In the aboveequationsReuis the cell Reynoldsnumber,c is the wave speed,Iluhllis the flow speed,and h
is the elementlength. £(M) E [0, 1] is a function of Mach numbersuchthat £(0) = 0 and £(M ~ Mc) = 1.
All the results reportedin this article are with the following definition of £:
'-N:) Z
M~M,
z=
M>M,
whereMc = 0.1. Matrix T/3is subtractedfrom Tato accountfor the shock-capturingterm as shownin Eq.
(35). It is defined as

= ~~wi/3
'1",8 M (40)

Considerthe computationof compressibleflow past a solid body using the unified formulation as in Eq.
(33). In the regionswhere the machnumberis low, for examplein the boundarylayer, the flow is almost
incompressibleand density assumes,approximately,the free-streamvalue. Under theseconditions the
continuity equation, Eq. (19), behaveslike an advection equation for density. Farther away from the
boundarylayer, the flow is in the compressibleregimeand the densityvariationsare quite significant.The
term in Eq. (33) that involves Nh providesnumericalstability to the densityfield in the region wherethe
above-mentionedtransition takes place. This term is not neededin the formulation if one is seeking
solutionsgovernedby either the compressibleor the incompressibleflow equationsonly. The matrix Nh
is a diagonal matrix defined as Nh = diag(l, 0, 0, 0, 0).
Ah
In the stabilizing terms in Eq. (33), componentsof the A k matrix are defined as
Ah h
[A k ].',J.=[A k ].'.J.+z8 I,k J.5 C
+ 1 8.
(41)

where C is a constant. This term provides stability to the computations for compressible flows. Notice
that such a term is not explicitly added to the formulation in terms of the conservation variables; a similar
term is already presentin the definition of A ~. In the caseof augmentedconservationvariablessuch a
term is not presentin the original definitionof A ~. It has been our experience,with augmented
conservationvariablesformulation, that in the absenceof this term the velocity field developsoscillations
that grow with time. In our computationswe chooseC = 2.
 S. Mi/tal, T. Tezduyar I Comput. Methods Appl. Mech. Engrg. 161 (1998) 229-243 235

(3) The sixth integral enforcesweak continuity of the velocity field acrossthe space-time slabs.
(4) If one is interestedin strictly incompressibleflows, Eq. (33) can still be used. However, to reducethe
computationalcost, one can prescribethe density field in the entire domain and therefore,not solve for it.
Additionally, if one is not interestedin the temperaturefield, one can drop the energyequationtoo. This
is possiblebecausein the incompressiblelimit, the energyequationis decoupledfrom the rest of the flow
equations.
(5) The variational formulation in terms of the augmentedconservationvariableshas an advantagethat any
changein the equation of state can be incorporatedin the implementationwith very little effort. For
example,to include the real gas effects in the formulation, one needsto modify the stateequationonly.
However, the implementationis not so straightforwardif one usesthe conservationvariables.
(6) We also implemented this unified approach in the context of a semi-discreteformulation. In the
incompressiblelimit, Eq. (21) reducesto the divergencefree condition on the velocity field. It behaves
like a constraint equation and one looks for a pressurefield such that the velocity field satisfiesthe
divergencefree condition at each time level. In an implementationfor time-accuratecomputationof
incompressibleflows, the divergencefree equationand the pressureterms are evaluatedat the n + 1 time
level while the rest of the termsare evaluatedat n + 1/2 time level. This ensuresthat the velocity field at
eachtime level is divergencefree. On the other hand, if the divergencefree equationis evaluatedat the
n + 1/2 time level, the velocity field computedat later times may not be divergencefree and one cannot
ensure the stability of computations.Therefore, in the semi-discreteimplementation of the unified
formulation for time-accuratecomputations,the pressureand the terms involving divergenceof velocity
are evaluatedat n + 1 time level while the other terms are evaluatedat n + 1/2 time level. We have
computed unsteady solutions using, both, the space-time and semi-discreteimplementationsof our
formulation. The results obtained from the two implementationsare almost indistinguishable.In this
article, therefore,we report the resultsonly from the space-time implementation.It must be pointed out
that the computationswith the space-time method are substantiallymore expensivethan the ones with
the semi-discretemethod. However, the space-time method allows one to compute flows involving
moving boundariesand interlaces.

4. Numericalexamples

Most of the computationsreportedin this article were carried out on the Digital 3000/300 AXP work-station
at liT Kanpur. Some were computedon the CRAY C90 at Networking Computing Servicesin Minneapolis,
Minnesota. For the space-time implementations,the finite-elementbasis functions are bilinear-in-spaceand
linear-in-time, and 2 X 2 X 2 Gaussianquadratureis employedfor numerical integration.The calculationswith
the semi-discreteimplementation,basedon bilinear finite-elementbasisfunctions,give almostindistinguishable
resultsas the onesfrom the space-time method.In this article, only the results computedwith the space-time
methodare shown. The nonlinear equationsystemsresulting from the finite-elementdiscretizationof the flow
equationsare solved using the GeneralizedMinimal RESidual (GMRES) technique [17] in conjunction with
diagonal and block-diagonalpreconditioners.

4.1. Shock-reflectionproblem

This two-dimensional,inviscid, steadyprobleminvolves three flow regionsseparatedby an oblique shockand

its reflection from a wall as shown in Fig. 1.
It is a standardbenchmarkproblem and for more details the interestedreaderis referred to the work by Le
Beau and Tezduyar [9] and Shakib [18]. The motivation for this computationis to establishconfidencein our
formulation and its implementationfor computing flows involving shocks. The computationaldomain is a
rectangularregion of dimensions4.1 in the x direction and 1.0 in the y direction. The meshconsistsof 60 X 20
rectangularelements.At the left boundary,flow data correspondingto Mach 2.9 is prescribed:
236 S. Mittal, T. Tezduyar / Comput. Methods Appl. Mech. Engrg. 161 (1998) 229-243

'M=2.9
p=1
Region 1 u I =1
(42)
U2 =0
,0 = 0

M=2.378

M=2.9 M=1.942

Region 1 Region3

Fig. I. Shock-reflection problem: problem description.

At the top boundary,the flow conditions that are specifiedcorrespondto Mach 2.3781 and an incident shock
angle of 29°:

'M = 2,3781
p = 1.7
Region 2 Ul = 0,9033 (43)
Uz = -0.1746
.f} = 0.07685
At the lower boundary,the componentof velocity normal to the wall is assigneda zero value. The computations
begin with a uniform Mach 2.9 flow in the domain and continuetill the steady-statenorm dropsbelow a certain

Fig. 2. Shock-reflection problem: density and pressure fields for the steady-state solution.
 S. Mittal. T. Tezduyar / Comput. Methods Appl. Mech. Engrg. 161 (1998) 229-243 237

value. It should be pointed out that in this problem, the local Mach number in the entire domain is always
greaterthan the 'cut-off Mach numberdefinedin Eq. (23) and, therefore,z = 1 everywhere.This implies that
the pressureis determinedby the equationof state for perfect gas everywherein the domain.
Fig. 2 showsthe density and pressurefields in the domain for the steady-statesolution. Comparedin Fig. 3
are the densityfields for the computedand exactsolutionsat y = 0.25. Our resultscomparequite well with those
reportedby other researchersusing alternateformulations [9,18].

4.2. Supersonicflow past a cylinder

Mach 2 flow pasta circular cylinder is computedfor two cases.In the first casethe original compressibleflow
equationsare employedby setting z = 1 in Eqs. (19)-(22), i.e. the pressureis determinedby the equationof
statefor a perfectgas.The secondcaseis computedwith the unified formulation wherez is definedby Eq. (23).
The meshemployed,consistsof 5120 quadrilateralelementsand 5264 nodes.The Reynoldsnumberbasedon
the diameterof the cylinder and the free-streamvaluesof the velocity and kinematic viscosity is 2000, and the
Prandtl Number is 0.72. The cylinder wall is assumedto be adiabaticand the no-slip condition is specifiedfor
the velocity on the surfaceof the cylinder. All the variablesare specifiedat the upstreamboundary.At the upper
and lower boundaries,normal componentsof the velocity and heat flux are set to zero together with the
tangentialcomponentof the stressvector. At the downstreamboundary,we specify a Neumann-typeboundary
condition for the velocity and energythat is consistentwith the variational formulation given by Eq. (33). The
computationsare initiated with free-streamconditions in the entire domain and continue till the steady-state
norm of the solution falls below a certain desired value. Shown in Fig. 4 are the density, temperatureand
pressurefields for the steady-statesolution computedand the augmentedconservationvaluesformulation with
238 S. Minai, T. Tezduyar I Comput. Methods Appl. Mech. Engrg. 161 (1998) 229-243

//'~

~ " I 0"'

ion ~rl1 ~#LJ ~

r"""" W~'.

":';/

'--'--~ ,""
"-'\, a.
,...". 1 ~
~\\
"\'-\-J
'-~
\ \1
\'

Fig. 5. Mach = 2, Re = 2000 flow past a cylinder computed with z defined by Eq. (23): density, temperature and pressure fields (and their
close-ups in the lower row) for the steady-state solution.

z = 1. One can observea strong bow shock upstreamof the cylinder and a weak tail shock in the wake. The
shock stand-off distancecomparesquite well with experimentalobservations[19]. It can also be observedthat
the shock has been capturedquite well within two to three elements.Solution to the sameproblem has been
computedby the conservationvariablesformulation on a much finer meshwith 16000 elementsand reportedin
[13]. On comparing the two solutions we observe that the conservation variables and the agumented
conservationvariablesformulations lead to quite comparableresults.
Fig. 5 shows the density, temperatureand pressurefields for the steady-statesolution computedwith the
unified formulation with z defined by Eq. (23). On comparing Figs. 5 and 4 we observethat they are quite
similar exceptthat the density and the temperaturefields exhibit somedifferencesclose to the cylinder wall. In
the caseof unified formulation, with z definedby Eq. (23), the divergence-freeconstraintkicks-in very closeto
the wall of cylinder where the Mach number is nearly zero. In this region, the mass balanceequation (19)
behaveslike an advectionequationfor density.As a result one observesfrom the contourplots, very closeto the
wall of the cylinder, that the flow attemptsto advect the density from the upstreamto downstreamlocations.
This is certainly not the casein Fig. 5 where the pressureis determinedby the stateequationfor a perfect gas
and not by the divergencefree constraint.It is quite interestingto note that the pressurefields computedby the
two formulations are almost identical. The drag coefficient computedby both the formulations is 1.48.
Thesetest problemsdemonstratethat the unified compressible-incompressible formulation results in correct
shock location and strengthfor both viscous and inviscid flows.

4.3. Subsonicflow past a cylinder

The main motivation to develop the unified compressible-incompressibleformulation is to improve the

performanceof compressibleflow algorithm at low Mach numbers.In this sectionwe presentour solutionsfor
 S. Mittal, T. Tezduyar / Comput. Methods Appl. Mech. Engrg. 161 (1998) 229-243 239

flow past a circular cylinder at Re 100 and low Mach numbers.The Prandtl Number is 0.72. The cylinder
residesin a rectangularcomputationaldomain whoseupstreamand downstreamboundariesare locatedat 15 and
35 cylinder radii, respectively,from the cylinder's center.The upper and lower boundariesare placedat 16 radii
from the centerof the cylinder. The finite elementmeshconsistsof 4688 quadrilateralspace-timeelementsand
4826 nodes.At eachtime step,4741.2nonlinearequationsare solved iteratively to computethe flow field. The
cylinder surfaceis assumedto be adiabaticand the no slip condition is specifiedfor the velocity on the cylinder
wall. At the upstreamboundary,density, velocity and temperatureare assignedto their free-streamvalues.At
the downstreamboundary,we specify a Neumann-typeboundarycondition for the velocity and energy that is
consistentwith the variational formulation given by Eq. (33). At the upper and lower computationalboundaries,
normal componentsof the velocity and heat flux are set to zero togetherwith the tangentialcomponentof the
stressvector. We first presentour results for Mach 0.2 flow computedwith z = 1. Fig. 6 showsthe vorticity,
pressureand temperaturefields correspondingto the peak value of lift coefficient. Fig. 7 shows the time
histories of the lift and drag coefficientsfor that part of the simulation when the periodic solution is achieved.
The Strouhalnumbercorrespondingto the variation of lift coefficient for this caseis 0.164. Figs. 8 and 9 show
the solution for Mach 0.2 flow computedwith the unified formulation with z definedby Eq. (23). The Strouhal
numbercorrespondingto the variation of lift coefficientfor this caseis 0.169. As expected,the two solutionsare
quite similar. On comparingFigs. 6 and 8 we observethat thereare somedifferencesin the temperaturefields of
the two cases.As has beenexplainedin the previous section,in the caseof unified formulation, the flow very
close to the wall is modeledby incompressibleflow equationsand therefore,the temperaturechangestake place
only becauseof viscouseffects. On the other hand,in the caseof computationswith z = 1, density,temperature
and pressurechangestake place in accordancewith the equation of state for perfect gas. Therefore, the
contribution to temperaturechangescome from, both, the viscous and compressibleeffects.
The formulation basedpurely on the compressibleflow equations,i.e. with z = 1.,fails to yield an acceptable
unsteadysolution at Mach 0.05 with the presentmesh.However,if the meshis refined it is possibleto compute

r++

\ r

~~ I I.
... 1\\
Fig. 6. Mach = 2, Re = 100 flow past a cylinder computed with z = 1: vorticity, pressure and temperature fields (and their close-ups on the
right) at the peak value of the lift coefficient.

1.50
~ ~
1.25

"C
1.00
0.75 ~~~J
U.0.50
<3
f"\ f\
0.25
0.00
i\ i\ f\ f\ /'\

-0.25
.0.50
\; \ j \/ v \J \ I v
0 20 40 60 80 100
t
Fig. 7. Mach = 0.2, Re = 100 flow past a cylinder computedwith z = 1: time histories of the lift and drag coefficients.
240 S. Minai, T. Tezduyar I Comput. Methods Appl. Mech. Engrg. 161 (1998) 229-243

the unsteadyflow at Mach 0.05. It hasbeenour experiencethat the low Mach numberflows are very sensitiveto
the spatialrefinement.For a given meshthere exists a certainMach numberbelow which the compressibleflow
formulation breaksdown. On the other hand, with the unified compressible-incompressibleflow formulation,
one is able to computeflows at any Mach number.Figs. 10 and 11 show the solution for unsteadyflow past a
cylinder at Mach 0.05 computedwith z definedby Eq. (23). The Strouhalnumbercorrespondingto the variation

Fig. 8. Mach = 2, Re = 100 flow past a cylinder computed with z defined by Eq. (23): vorticity, pressure and temperature fields (and their
close-ups on the right) at the peak value of the lift coefficient.

1.50
1.25
1.00
0.75 ~~~J
C3
-: 0.50
U
0.25
~!\ i\ f\ I '\ ('\ i\ I '\

v
0.00
-0.25
-0.50
\J \ I \/ '\ I vi \/ '\ I
0 20 40 60 80 100
t
Fig. 9. Mach = 0.2, Re = 100 flow past a cylinder computedwith z defined by Eq. (23): time histories of the lift and drag coefficients.
 S. Mitral, T. Tezduyar I Comput. Methods Appl. Mech. Engrg. 161 (1998) 229-243 241

1.50
1.25
1.00

"C
0.75 [JJ~~J
U
..: 0.50
U
(\ (\
I
0.25
0.00 ~(\ (\ (\ (\ ( '\

-0.25

-0.50
1

0
'V '\J
20
'V 'V '\ / v '\ / \ i
60 40 80 100
t
Fig. 11. Mach = 0.05, Re = 100 flow past a cylinder computedwith z definedby Eq. (23): time histories of the lift and drag coefficients.

of lift coefficient for this case is 0.170.We observethat the solutions at Mach 0.05 and Mach 0.2 are quite
similar except for certain differencesin the temperaturefields that are due to the compressibilityeffects. The
demonstratethe robustnessof the unified formulation we computethe unsteadyflow past a cylinder at Mach
0.001 and in the incompressiblelimit. The solution at Mach 0.001 is shown in Figs. 12 and 13 while the one in
the incompressiblelimit is shownin Figs. 14 and 15.The incompressibleflow caseis computedby settingz = O.
The Strouhal number for vortex sheddingin both the casesis 0.170.We observethat the solutions for Mach
numbers0.05, 0.001 and in the incompressiblelimit are almost indistinguishableand agree quite well with
results from alternateformulations for incompressibleflows [20].

:~
,I -- ~
Fig. 12. Mach = 0.001, Re = 100 flow past a cylinder computed with z defined by Eq. (23): vorticity, pressure and temperature fields (and
their close-ups on the right) at the peak value of the lift coefficient.

.- -- -- .--
t
Fig. 13. Mach = 0.001, Re = 100 flow past a cylinder computedwith z definedby Eq. (23): time historiesof the lift and drag coefficients.
242 S. Mittal, T. Tezduyar I Comput. Methods Appl. Mech. Engrg. 161 (1998) 229-243

Fig. 14. Incompressible, Re = 100 flow past a cylinder computed withz= 0: vorticity, pressure and temperature fields (and their close-ups
on the right) at the peak value of the lift coefficient.

1.50
1.25
1.00

"0
0.75 ~~~]
O- 0.50
U
0.25
"(\ (\ f\ f\
f\ f\ f"\ f\
v v v
0.00
-0.25
-0.50
\/ \; \/ \J \/
20 40 60 80 100

Fig. 15. Incompressible, Re = 100 flow past a cylinder computed with z = 0: time histories of the lift and drag coefficients.

5. Conclusions

A unified formulation for compressibleand incompressibleflows in terms of the augmentedconservation

variableshasbeenproposed.In caseof compressibleflows the equationof statedeterminesthe pressurewhereas
it is the divergence-freeconstraint on velocity field that sets the pressurefor incompressibleflows. The
appropriategoverning equationsare chosen locally based on the local Mach number. The formulation was
successfullyapplied to various numerical testsinvolving steadyand unsteadyflows over a range of Mach and
Reynoldsnumbers.

Acknowledgement

This work was sponsoredby ARPA and by the Army High PerformanceComputing ResearchCenterunder
the auspicesof the Department of the Army, Army ResearchLaboratory cooperative agreementnumber
DAAH04-95-2-0003/ contract number DAHH04-95-0008.The content does not necessarilyreflect the position
or the policy of the government,and no official endorsementshouldbe inferred. The CRAY time was provided,
in part, by the University of Minnesota SupercomputerInstitute.

References

[1] E. Turkel, Review of preconditioning methods for fluid dynamics, Technical Report 92-47, Institute for Computer Applications in
Science and Engineering, NASA Langley Research Center, September 1992.
 S. Mittai, T. Tezduyar I Comput. Methods Appi. Mech. Engrg. 161 (1998) 229-243 243

[2] G. Hauke and T.J.R. Hughes, A unified approach to compressible and incompressible flows, Comput. Methods Appl. Mech. Engrg. 113
(1994) 389-395.
[3] G. Hauke, A unified approach to compressible and incompressible flows and a new entropy-consistent formulation of the K-epsilon
model, Ph.D. Thesis, Department of Mechanical Engineering, Stanford University, 1995.
[4] J.M. Weiss and W.A. Smith, Preconditioningapplied to variable and constantdensity flows, AIAA J. 33(11) (1995) 2050-2057.
[5] S.M.H. Karimian and G.E. Schneider, Pressure-based control-volume finite element method for flow at all speeds, AIAA J. 33(9)
(1995) 1611-1618.
[6] T.E. Tezduyar and T.J.R. Hughes, Development of time-accurate finite element techniques for first-order hyperbolic systems with
particular emphasis on the compressible Euler equations, Report prepared under NASA-Ames University Consortium Interchange, No.
NCA2-0R745-104, 1982.
[7] T .E. Tezduyar and T.J .R. Hughes, Finite element formulations for convection dominated flows with particular emphasis on the
compressible Euler equations, in: Proc. AIAA 21st Aerospace Sciences Meeting, AIAA Paper 83-0125, Reno, Nevada, 1983.
[8] G.J. Le Beau, The finite element computation of compressible flows, Master's Thesis, Aerospace Engineering, University of
Minnesota, 1990.
[9] G.J. Le Beau and T.E. Tezduyar, Finite element computation of compressible flows with the SUPG formulation, in: M.N. Dhaubhadel,
M.S. Engelman and J.N. Reddy, eds., Advances in Finite Element Analysis in Fluid Dynamics, FED-Vol. 123 (ASME, New York,
1991) 21-27.
[10] G.J. Le Beau, S.E. Ray, S.K. Aliabadi and T.E. Tezduyar, SUPG finite element computation of compressible flows with the entropy
and conservation variables formulations, Comput. Methods Appl. Mech. Engrg., 104 (1993) 27-42.
[II] S.K. Aliabadi and T.E. Tezduyar, Space-time finite element computation of compressible flows involving moving boundaries and
interfaces, Comput. Methods Appl. Mech. Engrg. 107(1-2) (1993) 209-224.
[12] S.K. Aliabadi, S.E. Ray and T.E. Tezduyar, SUPG finite element computation of compressible flows with the entropy and conservation
variables formulations, Comput. Mech. II (1993) 300-312.
[13] S. Mittai, Finite element computation of unsteady viscous compressible flows, Comput. Methods Appl. Mech. Engrg. (1997), to
appear.
[14] R.L. Panton, Incompressible Flows (John Wiley and Sons, New York, 1984).
[15] T.J.R. Hughes and A.N. Brooks, A multi-dimensional upwind scheme with no crosswind diffusion, in: T.J.R. Hughes, ed., Finite
Element Methods for Convection Dominated Flows, AMD-Vol. 34 (ASME, New York, 1979) 19-35.
[16] T.J.R. Hughes and T.E. Tezduyar, Finite element methods for first-order hyperbolic systems with particular emphasis on the
compressible Euler equations, Comput. Methods Appl. Mech. Engrg. 45 (1984) 217-284.
[17] Y. Saad and M. Schultz, GMRES: A generalized minimal residual algorithm for solving nonsyrnrnetric linear systems, SIAM J. Scient.
Statist. Comput. 7 (1986) 856-869.
[18] F. Shakib, Finite element analysis of the compressible Euler and Navier-Stokes equations, Ph.D. Thesis, Department of Mechanical
Engineering, Stanford University, 1988.
[19] H.W. Liepmann and A. Roshko, Elements of Gas Dynamics (John Wiley and Sons, Inc., New York, 1957).
[20] M. Behr, D. Hastreiter, S. Mittal and T.E. Tezduyar, Incompressible flow past a circular cylinder: Dependence of the computed flow
field on the location of the lateral boundaries, Comput. Methods Appl. Mech. Engrg. 123 (1995) 309-316.