Hypersonic Aerodynamics
Edited by
T.K.S. Murthy
T.K.S. Murthy
Associate Director of Extension Programmes
Wessex Institute of Technology
Ashurst Lodge
Ashurst
Southampton S04 2AA
U.K.
Published by Kluwer Academic Publishers,
P.O. Box 17, 3300 AA Dordrecht, The Netherlands.
In copublication with Computational Mechanics Publications
Ashurst Lodge, Ashurst, Southampton, U.K.
Sold and distributed in the U.s.A. and Canada
by Kluwer Academic Publishers Group,
101 Philip Drive, Norwell, MA 02061, U.S.A.
In all other countries, sold and distributed
by Kluwer Academic Publishers Group,
P.O. Box 322,3300 AH Dordrecht, The Netherlands.
British Library CataloguinginPublication Data
A Catalogue record for this book is available
from the British Library
{1SBNXl792316738 Kluwer Academic Publishers Dordrecht/Boston/London
'ls13N 1853121568 Computational Mechanics Publications, Southampton
ISBN 1562520830 Computational Mechanics Publications, Boston, USA
Library of Congress Catalog Card Number 9177003
All rights reserved. No part of the material protected by this copyright notice may be reproduced
or utilized in any form or by any means, electronic or mechanical, including photocopying, recording
or by any information storage and retrieval system, without the written permission of the publisher.
@Kluwer Academic Publishers, 1991
@Computational Mechanics Publications, 1991
@See also p. 151
Acknowledgement is made to P. Gnoffo for the use of Figure 8 on p. 140 which appears on the front
cover of this book.
Printed and bound in Great Britain
The use of registered names, trademarks etc., in this publication does not imply, even in the absence
of a specific statement, that names are exempt from the relevant protective laws and regulations and
therefore free for general use.
CMP
.,
~
R. MOREAU
MADYLAM
Ecole Nationale Superieure d' Hydrau/ique de Grenoble
Bofte Postale 95
38402 Saint Martin d'Heres Cedex, France
The median level of presentation is the first year graduate student. Some texts are
monographs defining the current state of a field; others are accessible to final year
undergraduates; but essentially the emphasis is on readability and clarity.
CONTENTS
Preface
VB
29
81
115
153
203
233
265
293
341
387
447
Index
491
PREFACE
This book contains chapters written by some eminent scientists and researchers on
Computational Methods in Hypersonic Aerodynamics and is a natural sequel to the
earlier books on Computational Methods in Potential Aerodynamics (1986) and Computational Methods in Viscous Aerodynamics (1990) published by Computational
Mechanics Publications in their Aerodynamics Series.
In this book, the earlier attempts at the solution of the highly nonlinear NavierStokes equations are extended to the aerothermodynamics of flow in the hypersonic
regime. It includes the effects of viscosity on the physical and chemical processes
of hightemperature nonequilibrium flow at very high speeds, such as vibrational
excitation, dissociation and recombination, ionization and radiation, as well as real
gas effects and the effects of high temperature and low density.
The opening chapter by Grundmann provides an introduction to hypersonic flow
with a description of the physical (and chemical) aspects of the flow. As an example, the author describes these processes involved in the flow along the flight path
of an orbital vehicle with a basic consideration of fluxes and transport properties.
Grundmann also provides a specimen of the nonequilibrium boundary layer along a
flat plate. The chapter concludes with a useful set of definitions for a reacting gas
mixture in hightemperature flow.
Computational methods for the progressive solution of the NavierStokes equations
are presented in the second chapter by Hanel. The governing equations and relations
for equilibrium flow are first described with an outline of their numerical formulation,
including conservative discretization and estimation of fluxes. The effect of numerical
damping on the solution is demonstrated for the steadystate solution. The chapter
concludes with a consideration of the numerical problems of real viscous, hypersonic
flows.
In Chapter 3, Shang continues the numerical simulation of hypersonic flow. This
author explains that the current methods of solution deal with continuum flow and
attempt the solution of the NavierStokes equations in conjunction with chemical
physics equations, including the nonequilibrium chemical reactions and internal degrees of excitation. In the rarefied gas domain, continuum concepts cannot be relied
upon and particle simulation is necessary. Shang discusses this briefly together with
the added mechanism of the energy transport process in radiation.
Chapter 4 by Gnoffo contains an upwind biased, pointimplicit relaxation algorithm using the finite volume formulation for the numerical solution of the governing
equations for three dimensional, viscous, hypersonic flow in chemical and thermal
nonequilibrium. The author also includes an overview of physical models employed
for thermochemical nonequilibrium with several test cases and comparison with experimental data for hypersonic flow over blunt bodies.
Cinella in Chapter 5 reviews several numerical techniques for the simulation of fluid
flows spanning the range of reactive regimes, from local chemical equilibrium to full
thermochemical nonequilibrium. In particular, characteristicbased algorithms of the
flux vector and the fluxdifference type are considered together with a discussion of
the problems associated with the modelling of thermochemical behaviour of reactive
for flux linearisation. The authors conclude with a section on grid generation and
provide some examples of advanced visualisation of hypersonic flow.
The book has been prepared as a valuable contribution to the stateoftheart on
computational methods in hypersonic aerodynamics. All the chapters have been written by eminent scientists and researchers well known for their work in this field. The
Editor wishes to express his special thanks to them for their cooperation during the
various stages leading up to actual publication.
The Editor
Southampton, January 1992
Chapter 1:
Introduction to the Physical Aspects of
Hypersonic Aerodynamics
R. Grundmann
Von Karman Institute for Fluid Dynamics,
B1640 RhodeStGenese, Belgium
1. INTRODUCTORY REMARKS
Any obstacle, like meteorites, satellites or manned space vehicles, coming from space
and approaching the earth's atmosphere is subjected to physical flow phenomena.
These change its state of motion and also the state of its surface.
The surface temperature for example will be influenced. The friction caused by the
increasing number of air molecules the closer the obstacle approaches the ground heats
the surface by conduction, convection, radiation and other heat transfer mechanisms.
All this coincides with changes in the surrounding air becoming denser and denser. A
bow shock will appear in front of the object. A region with very high temperatures
around the body's nose will develop which has the energy to split even molecules
like oxygen or nitrogen into their atoms. This gives the opportunity to form new
combinations like nitricoxygen. Single electrons may even leave the structure of an
atom or a molecule and the air becomes a partially ionized plasma.
One of the most important topics of research in hypersonic aerodynamics is to
find a reasonable way of calculating the temperature which is felt by a surface. Any
blunt nosed reentry body experiences at a velocity of about 11.2 km/s. It is the
orbital escape velocity. The temperature behind the bow shock belonging to this
velocity range is about 11 000 K. For comparison the temperature of the sun's surface
is about 5 000 K.
A temperature of 11 000 K will have an important feed back on the chemical
constitution of the air. Physical and chemical processes like vibration, dissociation,
recombination, ionization, radiation and many others arise. They need to be formulated to describe the phenomena involved in flows of these very high speeds and
temperatures. Simplified methods of calculating the temperature in the nose region
of the reentry vehicle will fail.
Perfect gas in inviscid flow
To demonstrate this, the total temperature To first will be calculated for a calorically
perfect gas in an inviscid flow. This total temperature is computed from the integral
energy equation assuming that the kinetic energy u~/2 of the flow is decelerated to
zero and completely transferred into a total temperature rise To. The total enthalpy
Ho is then written:
Ho = cpTo = cpToo
u2
+ :;
For a Mach number of 31.6 which corresponds to the velocity of 8910 m/sec and an
altitude of 78 km above the surface of the earth the total temperature To will read:
To
with
,1
Too = 197K
Moo = 31.6
, = 1.4
h = 78km
where Too is the static temperature of the atmosphere at the altitude h of 78 km. The
ratio of the specific heats, for a calorically perfect gas is constant.
The calculated result of approximately 40.000 K is much too high. This is because
some physical effects in high velocity flows were neglected. The existence of a bow
shock in front of the blunt body was not considered in the calculation. Behind a shock
the velocity is less than in front of it and therefore the kinetic energy is less which is
expressed in the energy equation.
The application of shock relations again for a calorically perfect gas at very high
Mach numbers will be used now for the static temperature Tshock behind a normal
shock. The shock is assumed to be strong, namely normal, which corresponds to a
shock angle f3 of 900 :
Tshock
 1) 2 2 f3
= Too 2,(,
(
,+1 )2 Moo sm = 38250K
This temperature is already smaller by a considerable amount (1290 K), but still far
from being correct. The total temperature TO,shock has the same value as the previously
calculated total temperature To in front of the shock, since in an adiabatic perfect
gas the total enthalpy is constant across the normal shock. Hence, these formulas for
inviscid compressible flow computations are not applicable to these types of hypersonic flows. One of the reasons is that the above computed temperatures are purely
translational temperatures since the air particles only perform translational motions.
At high temperatures other energy modes may be excited like the rotational and vibrational modes for molecules. The molecules also perform rotations and vibrational
motions. These modes will reduce the calculated values for the pure translational
mode.
Another reason is that the viscous effects are missing which make the temperature
vary within the boundary layer. Its maximum is not necessarily located at the surface
of the body but close to it.
Furthermore, energy is transported in different modes for which exist different
energy equations. Energy is consumed for dissociation processes and many other
reasons are responsible for measuring a much lower temperature than calculated with
simple formulas given above. Additionally, the gas was assumed to be calorically
perfect which is not true at very high temperatures.
A thermally perfect gas gives an even better result, since there the specific heats
and their ratio, are now functions of the temperature.
It must be imagined that chemically reacting flows need even more complicated
definitions of these properties, as will be seen later. The ratio of the specific heats,
will additionally depend on the pressure p.
, = ,(T,p)
From these considerations it is obvious that a very careful modeling of the physics
of hypersonic flows has to be performed. Otherwise the expected good results of numerical approaches are too far away from reality. The following sections are meant to
give an introduction to hypersonic aerodynamics. The physical phenomena of thermochemical nonequilibrium flows in this high temperature environment are described.
The glide path of an orbital vehicle leading through this environment of various energy
excitation zones is discussed. These physical effects have to be modeled in the governing set of the fluid mechanical equations. For this purpose some basic considerations
on fluxes and transport properties in hypersonic flows are given. Sample computations of chemical nonequilibrium boundary layers along a flat plate will show the
general behavior of high temperature effects. Finally, a set of definitions for reacting
gas mixtures in high temperature flows will be provided to lead into the kernel of this
book on computational methods in hypersonic aerodynamics.
2. DESCRIPTION OF THE FLOW PHENOMENA
Between subsonic and supersonic flows there is a decisive border which presents itself
physically by a shock, namely, a jump in pressure. This shock is clearly audible when
a flight object accelerates from subsonic to supersonic speeds.
The nondimensional number describing the ratio of the local velocity u to the
local speed of sound a of a calorically perfect gas is called the Mach number M. It is
unity in case of the transient from subsonic to supersonic velocities.
u
M=a
where
, = V,RT
a
R
T
<=>
<=>
<=>
<=>
A distribution of the temperature T and the local speed of sound for a calorically
perfect gaS is shown in Figure 1.
90
80
altitude h
(kml
62
52
47
J119sosphere ...
32
20
11
I....
..
,,
stratosphere
aircraft
troposphere
clouds
0
0
50
100
150
200
250
300
350
T [K 1. a [m/s 1
Figure 1: Atmospheric temperature and speed of sound distribution versus the altitude
The temperature versus altitude distribution is based on the U.S. Standard Atmosphere from 1962 given in [1]. All free stream conditions in the following figures
and calculations are referring to these tables.
The earth's atmosphere consists of two main layers as explained in Fig. 1. The
first one is called homosphere and its extension is from the sea level to 90 km. Beyond
this the layer is called heterosphere. In the homosphere the composition of the air
is almost constant. There is about 78%N2 present and 21 %0 2, the remaining 1%
consists of other gases like argon and carbon dioxide. Beyond the homosphere the
composition of the air changes accordingly with the very high temperatures existing
there. Dissociation and ionization effects occur, which change the composition of the
air by creating for example nitricoxygen and other products.
The homosphere is again split into three major parts which are the troposphere
extending from 0 to 11 km, the stratosphere extending from 11 to 20 km and the
mesosphere from 20 to 90 km. The troposphere contains 75% of the mass of the
atmosphere. A further 20% of the mass is in the stratosphere. That is, 95% of the
mass of the atmosphere is located within a layer of 20 km extension above the surface
of the earth.
In the eight shells in the homosphere it is known that the temperature changes
nearly linearly with altitude, see Fig. 1. In the heterosphere the temperature increases
linearly as well, but the gas no longer remains a continuum because the distance
between single molecules becomes larger and larger.
The curves in Fig. 1 are given for the atmospheric static temperature T and the
speed of sound a at zero motion of the airflow. In hypersonic flows the speed of sound
cannot be calculated applying the calorically perfect gas assumption anymore. Due
to the very high temperatures, the ratio of the specific heats now becomes a function
= J(8p/8p).
where p is the pressure and p is the density. The suffix s stands for isentropic flow
conditions, meaning that within a sound wave the flow conditions are reversible and
adiabatic. All thermodynamic properties of chemically reacting flows are functions of
the temperature and the pressure. The partial derivative of the pressure with respect
to the density may be replaced, such that the result for the equilibrium local speed
of sound becomes:
RT[l+(~)(fvh]
2_
(Oh)
a  I
[1  P op T]
with
I
= cCp = f(T,p)
v
where e denotes the inner energy, v the volume and h the enthalpy for an equilibrium
chemically reacting gas. These thermodynamic quantities are combined through the
first and second law of thermodynamics and describe an isentropic process of a sound
wave. If the gas is calorically perfect, the formula reduces to the earlier one as expected. In that case the flow is frozen meaning the chemical state of the fluid does
not change with temperature.
Looking for a criterion to describe the transient from supersonic to hypersonic
velocities is not obvious. There is no visible or audible border as for the transient
from subsonic to supersonic flows.
A convention says that the border is placed around the Mach number:
M
This border is not fixed to that certain number, but around M ~ 5 some physical
phenomena become more and more important and therefore characterize hypersonic
flows. In the following text some of these conceptions, frequently mentioned in the
literature, are presented and discussed, see also ANDERSON [2,3].
Due to the very high speeds, the kinetic energy of the flow is considerable. When
this flow is decelerated by friction effects close to body surfaces an energy transfer to
thermal energies is observed. These effects are usually called strong viscous effects.
The boundary layer becomes thicker now depending also on the Mach number of the
flow.
Re=
Jl
V
L Fnonnal = mRearth
 = L 2
where
{:} velocity
{:} lift
W cos ()
90%
10%
10
11
12
13
14
15
velocity [ km/s I
10
11
12
13
velocity [ km/s I
14
15
()=o
:>
:>
weight
incidence
In this formula EFnonnal lists all the forces directed normal to the flight path.
If the angle of incidence () is zero which may be assumed for a gradually horizontal
glide path, then the above mentioned formula solves for the density distribution p for
variable velocity V.
(~ _ ~)
p(h) = 2W
SCI
V2
Vc2
where W is the weight, S is the planform and CI is the lift coefficient of a shuttle type
aircraft. The velocity Vc is related to the gravitation g being also a function of the
altitude and the radius Rearth of the earth by the following formula:
v:,2 =
g Rearth
Since the density p is only a function of the altitude h the flight corridor results
in a band of the above mentioned constant load factor 2W/(Scl). This is a parameter
which lies between too high or too hot values.
In the same Fig. 3 constant free stream unit Reynolds numbers are also given.
Unit Reynolds number means that the Reynolds number is divided by a reference
length x of the vehicle. The formula describing this number is:
Rex
; =
p(h)V
J1(T)
This formula again may be solved for the density p which simply is a function of
the altitude h. The same yields for the viscosity J1 in the standard atmosphere.
p( h) = Rex J1(T)
x V
These constant unit Reynolds number curves indicate that the glide path leads from
pure laminar flow at high altitudes where the density is still low through transitional
to fully turbulent flow regions close to the surface of the earth. The horizontal lines
show the Reynolds number starting in the turbulent region at values of 3 . 108 down
to 3 10 in the laminar part.
The last set of curves in Fig. 3 are the free stream Mach numbers. They are based
on the assumption of a calorically perfect gas where the ratio of the specific heats
I is a constant, I = 1.4. The Mach number lines are shown from 8 to 48. They
are not constant with increasing altitude because they depend on the temperature
distribution, see Fig. 1.
These glide paths, also called reentry trajectories, lead from free molecular flow
to continuum flow. On the way they pass the physical phenomena of vibrational
excitation, dissociation of oxygen and nitrogen and even ionization. In Figure 4 the
current glide path is plotted including the various energy excitation zones for the
stagnation point.
10%
10"10 90%
90%
10%
90%
10%
10
11
12
13
14
15
velocity [ km/s I
From this diagram it is seen that the region is very small where the air obeys the
ideal gas concept.
p=pRT
The remaining regions where the molecules are vibration ally excited, the oxygen
and the nitrogen begin to dissociate, and where ionization takes place have to be
described by the additional knowledge of the microscopic states of the various energy
levels. The gas considered in these cases is assumed to stay a perfect gas but now
consisting of different species. Now, it is a mixture of perfect gases.
~
A region where the flow is in vibrational excitation starts very early in the plot.
The excitation begins at temperatures of 800 K and is terminated at about 2000 K.
Here the dissociation of oxygen begins. At around 4000 K it is completed and the
first nitrogen molecules split into their atoms. This process is finished close to the
9000 K temperature line.
The phenomenon of ionization, where atoms set electrons free and a plasma develops, begins at the same temperature border of about 9000 K.
All these regions are given in a band width from 10% to 90%. From the plot
it is to imagine that these regions partially overlap. Thus, a complete numerical or
experimental consideration of the flow field along the reentry trajectory of a space
shuttle must include all these flow phenomena. The complexity can be demonstrated
driving
force
fluxes
Concentration
Gradient
Mass
(first ord.
tensor=
vector)
Fick's
law
Energ~
Oufour
effect
OT
(first ord.
tensor=
vector)
Teml2erature
Gradient
Gradient
Soret
effect
OT
i j
Velocit~
Fourier's
law
k
Newton's
law
Momentum
(second ord.
tensor)
/1
chanical driving forces, namely on the ordinary, pressure, and forced diffusion, and
on the temperature gradient, also called thermaldiffusion or Soret effect. The energy
flux depends both on the temperature gradient causing heat conduction and on the
concentration gradient which is either called diffusionthermo or Dufour effect. The
momentum flux is dependent only on the velocity gradient.
In order to describe the Soret and the Dufour effect an additional transport property to the existing ones like viscosity /l, thermal conductivity k and diffusivity Dij
has to be added. It is the thermal diffusion or the Soret coefficient Dr.
From the first two rows of the table it turns out that the total mass and energy
fluxes are sums of two elementary fluxes each. This is true for the considered case.
There are even more of these fluxes due to other mechanical driving forces not listed
in the table, like the pressure gradient or external force differences. Provisionally,
these existing mass fluxes are summed up.
The index s stands for the individual species s of the gas mixture. Since the mass
flux j. is a vector, another index should be added to precise its spatial extensions.
The socalled ordinary diffusion j~rdinary is the portion of the mass flux which originally depends on the gradient of the concentration of each species. The distribution
of the concentration of the species will be the result of the mass transport equations
qconduction
The internal summations on s in that equation count the energy fluxes due to the
diffusionthermo effect and interdiffusion of each single species s. The heat flux q also
is a vector and therefore it is dependent on the spatial conditions.
The conductive energy flux qconduction is the dominant one for lower temperature
ranges. It is comparable to the ordinary diffusion term in the considerations on the
mass transport, since it is only dependent on the temperature gradient.
Due to the concentration of species, a heat transfer q~ufour is created which is
normally of negligible size compared to the others. It is also called Dufour or thermodiffusion effect and for completeness it is given in the second term.
The third heat flux term q~nterdiffusion is caused by the mass diffusion of the species
themselves. Therefore, its name is interdiffusion. Due to the different species mass
fluxes a transport of energy by the species is performed. Each species carries its own
enthalpy including its energy of formation which might be very different in chemically
reacting gases.
Last but not least, there is the radiative flux ifadiation The threshold temperature
at which air radiates is 10 000 K. Therefore, this term is of minor order in general.
However, the heat transfer by radiation should not be neglected when the thermal
load of a surface must be estimated correctly.
5. SAMPLES OF CHEMICAL NONEQUILIBRIUM BOUNDARY LAYERS
ALONG FLAT PLATES
To illustrate some of the different fluxes given in the table above a Couetteflow is
considered in Figure 5. This is a laminar flow between two flat plates involving no
downstream pressure gradient. The upper plate moves from the left to the right with
U=v
T=Tw
,,
flow direction
,,
P=P w
,
\
U=O
"
CA = CAmax
= Twad
:
:
:
:
P=Pwad
I
1
I
,
CA=
the velocity Voo. By doing that it forces the flow to move producing a linear velocity
profile as shown in the left part of Fig. 5.
Due to friction, a temperature and a density profile of the base flow develops. The
temperature shows a normal tangent at the nonmoving wall which defines the state of
an adiabatic wall. No heat transfer by conduction is possible through this wall. The
wall takes the temperature T of the flow adjacent to it. Since there exists no normal
pressure gradient in ydirection, the density profile p follows the equation of state and
presents an opposite shape to the temperature profile. At high temperatures is the
density low. A specialty of the upper wall is that it is slightly soluble meaning that
a very little amount of it will diffuse into the fluid flow. The amount is so small that
the mass balance is not disturbed. Therefore the mass fraction of the wall material PA
has its maximum at the upper moving wall while the concentration decreases slowly
in the negative ydirection approaching the value zero.
For the friction in the flow stands the transport of momentum expressed by Newton's law.
au
Txy
= /1 ay
The shear stress Txy is the only momentum flux in the equation of change for the
momentum. It can be influenced by other transport mechanisms only via the viscosity
/1, since this is a function of the temperature. The momentum transport takes place
always in the direction of decreasing velocity.
The energy transport by conduction describes a heat flux. The temperature gradient is the means to transport energy in the direction of decreasing temperature.
Fourier's law of heat transfer by conduction is:
qy =k{)y
The material of the soluble wall will be transported in the negative direction to
the wall again in the direction of decreasing mass fraction PA. The law to model this
movement was given by Fick. It is the socalled law of diffusion.
JA
= 
D AB{)PA
{)y
qAy
= ( JAy
.T )
+ JAy
PA
(1
'RTI

PA
D~
1MA )D
AB
5.,,
x = 0.0 m
0.1 m
4 1Ii... .. .........t=1:.=0mi
X
x=
x = B.O m
y/6
4 ~4~~
,~
y/6
31\~f__1
.,~
~.~
\~
~~~~
.....\
\'.. \
2 r.~~.~~..~~~
'\:...., ,
1f':+1':'~1
~ "
~~~~~~
25
50
Figure 6: Temperature and concentration profiles along an adiabatic flat plate (Ma
= 15)
steps. These steps are performed by particle collisions. It takes several collisions to
change the level in the different energy modes. The number of collisions needed for
thermal equilibrium, about 20 000, is ten times less than for chemical equilibrium. If
now the time scale of thermal equilibrium is less than the one of chemical equilibrium,
then the energy mode of the vibrational energy is already in equilibrium before the
chemical process is finished.
The gas under consideration is a diatomic gas like oxygen or nitrogen. It only
consists of a concentration of atoms and molecules. A mixture of oxygen and nitrogen
is not considered here.
Vibrational energy modes, electron translation energies, ionization and radiation
are excluded. Because a boundary layer flow is considered here, the effect of catalycity
at the wall on the heat transfer can be discussed. The reactions with the wall material
are called heterogeneous reactions while inside the flow homogeneous reactions take
place. This difference will be pointed out by the sample calculations.
The geometry of the flow problem is simply a flat plate which excludes surface
curvature effects and the downstream pressure gradient is assumed to be zero. The
velocity profiles in a boundary layer along a flat plate are similar. They do not change
in shape but in size with increasing boundary layer thickness. Therefore, they will
not be shown here.
Examples of this twodimensional chemically reacting nonequilibrium boundary
layer flow for a diatomic gas were given by PETERS [6]. Temperature and concentration profiles of atomic oxygen are shown for increasing downstream distance in Figure
6.
. +'T _
or
catalytic
The first term on the left hand side describes the ordinary diffusion at the wall and
the second one involves the mass flux based on a temperature gradient also known as
the thermal diffusion or Soret effect.
The right hand side is simplified. The knowledge about the reaction rate of the
wall material with the gas mixture is subsumized in the empirical rate coefficient kw.
The density PA of the oxygen atoms can be transferred to the concentration CA
by simple transformations and definitions which will be reported later.
This boundary condition of the socalled third kind is one of the very rare cases. It
involves the value of the concentration CA and its gradient in normal direction 8CA/ 8y,
prescribing together a Dirichlet and a von Neumann boundary boundary condition.
Additionally, the temperature gradient at the wall finds a place in this complex mass
flux boundary condition.
From this boundary condition of the third kind it results that a normal gradient of
the concentration profile at the wall can be generated by two reasons: the wall behaves
adiabatic and the thermal diffusion or Soret coefficient D~ is set to zero because it is
negligible. In both cases the wall catalycity kw must have the value zero.
Figure 7 shows another set of temperature and atom concentration plots for this
second case of mass flux boundary conditions. Prescribing a constant wall temperature
of about 660 K changes the boundary condition for the concentration also. The
catalycity of the wall and the thermal diffusion coefficient were assumed to be zero.
The maximum of the temperature profiles reduces while going in downstream
direction, since the energy is consumed by dissociation. The boundary condition
for the concentration prescribes a normal gradient of zero. No atoms are forced to
kw=
4
'W=
0 m/sec
x= Om
x= 2m
x= 10m
x=200m
y/6
3
'.~
'. ..~
, '. ........
'
0
0
15
TfTe
30
..
."\
y/6
0 m/sec
,,
"
....... .....
,,
.....
"
"
"
......
]
,;
:."
/.
,""
0
0.10
."
e Ie 0.20
'A Ae
Figure 7: Temperature and concentration profiles at constant wall temperature along a flat plate
(Ma = 24)
recombine due to the presence of a catalytic wall material. The concentration increases
at first in downstream direction. The temperature decreases far downstream because
of the growing boundary layer thickness. Thus, recombination takes place close to the
wall and by this the concentration of atoms reduces.
Walls can have different surfaces. They can be made of metals or glasslike materials. This involves different reaction rates or time scales in the heterogeneous reaction
at the wall. Some values for the catalycity coefficient kw are printed below.
kw = 00
kw = 1 to 10m/sec
kw = 0.01 to O.lm/sec
kw = 0
{:}
{:}
{:}
{:}
only recombination
metal oxides
glass like material
no recombination
In case of a catalytic wall Figure 8 shows the concentration of oxygen atoms for
different values of the reaction rate kw at the wall.
Clearly the effect of a sink for atoms is visible with increasing catalycity. This
means that the recombination is the dominant reaction at the wall. Oxygen atoms
are forced to recombine and form oxygen molecules again.
In Figure 9 a graph of the downstream development of the wall concentration CAw
and the heat transfer to the wall qw is illustrated for different values of the catalycity
coefficient kw.
For small values of the catalytic wall reaction rate a steeper gradient of the wall
concentration is seen than for values up to infinity. In the latter case equilibrium
is achieved already very early. The other diagram shows heat transfer for zero and
infinitely high catalycity. The difference is in the double logarithmic scale 15% at
maximum.
~.~
x=2m
4
y/8
3
~r4
r~~~d~
Omfsec
5m1sec
50mfsec
~~~i~~loooo~~mf~se~c
0.05
c./c
p,
0.10
Ae
Figure 8: Concentration profiles of oxygen atoms for different catalytic wall reactions (Ma
= 24)
0.10
Omfsec
0.05
....... ........
....
..............
...........

0.00
0.0
0.5
1.0
X/L
1.5
2.0
10 6
kw=
Omfsec
N/(m sec)
10 5 ~~~~~~~_+~
10 4
10 1
X/L
Figure 9: Wall concentration and heat transfer at the wall along a flat plate (Ma
= 24)
x=1m
x=1m
4
Pr = 1.0
Pr = 0.7
y/o
y/O
1
0
Figure 10:
(Ma
= 24)
15
TITe
30
0.05
e Ie
'A
0.10
Ae
Temperature and concentration profiles for variable Prandtl and Lewis numbers
x=2m
4
Vl6
3
2
Figure 11: Concentration profiles of oxygen atoms including the thermal diffusion effect (Ma
0.50
cAwIe Awe
0.25
'W= 0 mlse~
t

f
I
I
I
I
I
I
I
I
K"v= 50 m/s
= 24)
I
I
0.00
0.0
0.1
0.2
0.50
cAw Ie Awe
0.25
0.00
I
I
I
I
I
I
I
I
0.4
X/L
0.5
0.6
'V,,= Om/se
I
I
I
I
I
I
1"'0.0
0.3
0.1
0.2
0.3
0.4
X/L
0.5
0.6
Figure 12: Wall concentration of oxygen atoms along a flat plate with variable catalycity (Ma
= 24)
,,
4.0
kw= Om/se<
'\At 106
N/(m sec)
2.0
"'
~w= som/sep
1\
r
,,
0.00
0.1
0.0
0.50
~ 106
N/(m sec)
0.25
0.2
~w=
0.3
0.4
X/L
0.5
0.6
,,
SOm/se:;
kw= Om/se
,
,
"

,,
,,
,
L
,
0.00
0.0
0.1
0.2
0.3
0.4
X/L
0.5
Figure 13: Heat transfer at the wall along a flat plate with variable catalycity (Ma
p=pRT
0.6
= 24)
R = 287,1 s21<
This equation of state sets the pressure p in relation to the density p and the
temperature T.
The definition of a perfect gas contains the assumption that no intermolecular
forces are acting. These forces are described for example by the LENNARDJONES
potential CPm which is a function of the distance r from the gravitating body.
where d is the characteristic diameter of the molecule and t: is the characteristic energy
of interaction between molecules. This potential is related to the intermolecular forces
by differentiating it with respect to the distance r.
F: __ dCPm
m dr
M.
where M. is defined also as the mass of a species s per mole of this species which is
measured in kg/kgmole. The unity kgmole is one single expression which cannot be
split into parts. The universal gas constant has the value:
R = 8.3144 . 103
m 2 kg
I {:} universal gas constant
g  mo e
2 I< k
k = 1.38 . 1O23::~Z.
Now, in case of a gas mixture consisting of several species the equation of state
IS:
p=
E p.R.T = E.~, RT
where P. and Ms are the species density and the molecular weight. The latter also is
referred to as the mass per mole of species s. Finally, from this state equation it turns
p=
L:P.
The molecular weight of the gas composition M can be obtained by simply reintroducing the perfect gas equations for the mixture as well as for the species.
R=
L: P. R.
P
M=L.. it
M.
N.
v M
 y.
V
.Y.t..
M.
JV
Ar
JV. 
&
M.
V=~
VS
x...
N.
There are different ways of describing the concentrations of the various species in
a multicomponent gas mixture. Obviously, they differ from reference to reference.
The most common ones are presented only.
Ms
Ps= V
cs =~
Ms
Ps
cs = 
ps
=
_ Cs
Xs 
Ps
p

Due to the given definitions they can be rewritten in terms of the partial pressure
divided by the pressure of the gas mixture.
A frequently appearing ratio is the socalled number density which expresses the
number of particles of the species s in a unit volume of the gas composition.
Ns
ns= V
Two expressions are often used to define a mean velocity of the gas composition
usually appearing in the momentum or global continuity equations. One supports a
mass averaged velocity Ui and the other a molar averaged velocity Ui. The index i
stands for the three velocity components in the three spatial directions.
{:} mass average velocity
The species velocity is shown by Usi . This velocity is meant to be the sum on
all the velocities of the molecules of species s inside of a very small volume in the gas
composition, divided by the number of these molecules. Thus, the species velocity
and the average velocity are measured with respect to a stationary coordinate system.
The mass averaged velocity is the most frequently applied in fluid mechanical
literature. The molar way of averaging will be omitted here, although all the further
relations can also be performed in this way.
The difference between the species and mean flow velocities is called the diffusion
species velocity indicated by Vai. This means that each species moves in the gas mixture
flow with its individual speed. The diffusion velocity is measured with respect to the
mass averaged velocity. The index i again shows that there are three spatial directions
involved.
Vai
= Usi  Ui
The diffusion velocity represents a measure of the relative motion between the
gas mixture in total and its single species. Multiplying this mass averaged diffusion
velocity by the species density Ps defines the mass flux of the species s. It is the
diffusion flux of this individual species in the gas composition of different species.
jsi
= paVsi = Pa(Uai  Ui) {:} species mass flux with mass averaged velocity
This species mass flux jai is a vector quantity. It is the mass of the species s
passing through a unit area per unit time, or a partial mass flux per unit volume Pa
with the corresponding velocity Vai.
Ljs;
= 0
This section on the definitions of properties in reacting gas mixture flows closes this
chapter on some introductory remarks to hypersonic aerodynamics.
REFERENCES
1. U.S. Standard Atmosphere 1962. NASA, USAF, USWB, US Government, Printing Office, Washington DC, 1962.
2. Anderson, J.D., Jr. Modern Compressible Flow, McGrawHill, Second Edition,
1990.
3. Anderson, J.D., Jr. Hypersonic and High Temperature Gas Dynamics, McGrawHill, First Edition, 1989.
4. Anderson, J.D., Jr. Introduction to Flight, McGrawHill, Third Edition, 1989.
5. Hankey, W.L. Some Design Aspects of Hypersonic Vehicles, AGARDLS 42,
Aerodynamic Problems of Hypersonic Vehicles, 1972.
6. Peters, N. Losung der Grenzschichtgleichungen fur chemisch reagierende Gase
mit einem Mehrstellenverfahren, DLRFB 7258, 1972.
Chapter 2:
Computational Methods for Viscous Hypersonic
Flows
D. Hanel
Department of Combustion and Gasdynamics,
UniversitiitGHDuisburg, D4100 Duisburg, Germany
ABSTRACT
This paper is concerned with computational methods for the solution of the NavierStokes equations for hypersonic flows. The governing equations and relations for
equilibrium flow are described in the first part. Their numerical formulation is outlined
including a brief consideration of the conservative discretization, the evolution of the
numerical fluxes and of typical methods of solution. As a further topic the influence
of the numerical damping on the solution is discussed demonstrated for steadystate
solutions of the NavierStokes equations. The special properties of hypersonic, viscous
flows and the resulting numerical problems are considered finally.
INTRODUCTION
The computational fluid dynamics is an essential tool for the design of spacecrafts
and hypersonic airplanes. The physical and numerical problems to be solved cover
a wide range of the fluid mechanics, which are ranging from rarefied gasdynamics,
chemical nonequilibrium to problems of nonreactive, continuum flow. Rarefaction
and chemical effects are of great importance, in particular for reentry vehicles. Large
portions of the flight path are governed by continuum flow, in particular for hypersonic
planes in the Mach number range of five to ten, but also for reentry bodies in the
lower atmosphere.
The present paper is concerned with the computation of viscous, hypersonic flow in
the continuum flow regime. The viscous flow is of great importance for the prediction
of the flight properties, in particular for the prediction of aerodynamical forces and
heat flux rates. The vehicles considered are usually blunt bodies or planes with
a low aspdct ratio, thus the flow is threedimensional and strong viscous/inviscid
interaCtions can appear. The typical flow around blunt bodies is characterized by
strong shock waves and expansions, embedded subsonic regions, and shockboundary
(1)
A differential form (divergence form) can be obtained with the integral theorem of
Gauss:
(2)
Herein Q = (p, pv, pEf is the vector of the conservative variables. The generalized
flux H can be split into a vector for inviscid flow Hinv and a vector H vi8C describing
the contribution viscosity and heat conduction on the flow. The terms read:
Hinv
and
HVisc
(0, a, a . v+ iff
(3)
where is the stress tensor, and ifis the heat flux vector.
For the sake of simplicity in the following discussion the conservation equations
and their approximations are written for a twodimensional Cartesian coordinate system (x,y,t).
Full NavierStokes Equations The most complete description of continuum flow is
given by the N avierStokes equations. This system of equations is formed by the laws
of conservation of mass, momentum and energy for viscous, heat conducting fluids.
In a Cartesian frame the flux vectors are split up in Cartesian components, i.e.
H inv = (F, G) T and Hvi.
= (5, R)T
. The'mtegral form now reads
(4)
and the corresponding divergence form gives
(5)
Herein is
where 84 = qx + UTxx + VTxy and r4 = qy + UTxy + VTyy . With the Stokes assumption
/1v = 2/3/1 the stress terms and the components of the heat flux vector are:
aT
ay
qy =A
The Reynolds averaged NavierStokes equations for turbulent flows show the same
structure and therefore are treated in the same way.
Thin Layer Approximation The Thin Layer Approximation of the NavierStokes equations is a widely used approximation for the computation of viscous flows at high
Reynolds numbers. Similar as in the boundary layer theory all the viscous terms with
streamwise derivatives are neglected for the Thin Layer Approximation. However, in
contrast to the boundary layer theory, the Thin Layer Approximation retains all the
terms of the Euler equations, and (,e time derivatives as well. Consequently, the Thin
Layer Approximation preserves all the properties of the inviscid flow, in particular the
information transport along characteristics and discontinuous solutions.
In 2D Cartesian coordinates, assuming the xcoordinate as the nearly streamwise
direction, the Thin Layer Approximation would read:
(6)
where Q, F, and G have the meaning as in the full equations, but the viscous term
R contains only those stress terms which have derivatives in normal (y) direction.
There is no rigorous theory for the derivation of this approximation, but its range
of validity can be considered approximately the same as that of the higher order
boundary layer theory. It means that local flow separation and small normal pressure
gradients in viscous layers are covered by the approximation.
A further motivation for using this approximation is given by the fact that very
different scale lengths exist in a boundary layer, and therefore the step sizes in streamwise direction are much larger than in normal direction. By this, even when the full
Fx
+ (G 
R)y = 0
(7)
where the viscous terms R contain only terms with derivatives normal to the main
flow, F, and G are the complete Euler fluxes.
Efficient space marching methods are the motivation for using this approximation.
Space marching is well suited for stationary, supersonic flow, where all information
is transported downstream within the Mach cone. But in the subsonic regions this
assumption fails, since the stationary equations become elliptic and upstream influence
occurs in main flow direction. Therefore the term "parabolic" is somewhat misleading.
To preserve the "parabolic" behaviour, space marching in main flow direction must be
enforced numerically by onesided differences for all derivatives in main flow direction.
This numerical manipulation can be justified by the "parabolic" nature of attached
boundary layers. However neglecting the upstream influence in subsonic regions,
numerical instabilities can arise, which must be suppressed. This can be done with
different strategies.
In common marching procedures for external flows either the pressure gradient
normal to the wall is assumed to be zero across the subsonic layer, Schiff and Steger
[2], or the contribution of the streamwise pressure gradient is decreased in the subsonic
layer as a function of the Mach number based on a stability analysis, as proposed by
Vigneron [3].
Space marching methods can also be constructed for the timedependent Thin
Layer equations, Eq.(6), using an time marching technique. The solution converges in
time in an iterationlike manner for each cross flow plane separately. A usual upwind
scheme can be used, where all of the variables needed downstream of the actual cross
section are extrapolated from the upstream sections without further assumption for
the pressure, as reported by Menne [4] for computations of external hypersonic flows.
An advantage of this strategy is that the same timemarching code can be used in
"parabolic" and "elliptic" regimes.
Since only a single space marching sweep is employed in fully parabolic approximation procedures, the computations become very efficient with respect to the computation time.
+ Fx + G y = 0
Qtdr + Fdy Gdx =
Qt
(8)
0
(9)
(10)
The characteristic variables Ware defined by dW = T 1 dQ and the diagonal matrix
is given by the eigenvalues, A = diag( u + a, u, U  a) . The characteristic form is
the basis of the method of characteristics, but also the basis for constructing upwind
shock capturing schemes.
The weak solution, describing the jump conditions over a discontinuity (e.g. a
shock wave or a slip line), can only be derived from the conservative integral form
of the Euler equations. Therefore, if embedded discontinuities are considered, only
the conservative form guarantees the correct jump conditions. For a discontinuity C
moving with the velocity c, the application of the integral conservation laws results in
the jump conditions which read in the general form with the definition [Jl = 12  11:
= 0
(11)
By means of this jump condition the RankineHugoniot relations can be derived, and
computationally shockfitting procedures can be constructed.
Thermal and calorical relations
The solution of the conservation laws requires additional closure relations to express
the thermal and calorical state, and the transport quantities in the flux Ii as function
of the conservative variables Q.
Different situations have to be considered for the formulation of thermal and
calorical closure relations. The situations are equilibrium, frozen and nonequilibrium
flows. For the present consideration the gas is assumed to be in thermodynamical and
chemical equilibrium.
Assuming equilibrium flow, the equations of state can be expressed as algebraic
closure relations for the thermodynamical and calorical state as function of conservative flow variables Q. The basic input quantities for these relations usually are the
density p, and the internal energy c:, which can be calculated from the total energy
pE = p(c: + ij2/2).
The caloric equation of state expresses the internal energy c: with two thermodynamical variables, e.g. with p and T.
c: = c:(p, T)
Mi
For a thermally perfect gas, e.g. air without dissociation, it yields P = pRT. In
addition for a calorically perfect gas it is P = h  1) pc.
With the equations of state known, and with the thermodynamical laws all the
other quantities can be derived. In general the equations of state for a real, nonperfect gas cannot be formulated in a closed form. The calculation of the state has
to be carried out e.g. by an evaluation of the thermodynamical partition functions,
from which Mollier diagrams can be constructed.
Computationally it results in a system of nonlinear, algebraic equations, which has
to be solved for each state (grid) point. For computational purposes the precalculated
variables of state can be subdivided in regions in the p, c plane and expressed by curve
fits in those regions (e.g. for air by Tannehill [7]). For improving the vectorization
the equation of state can also be used in form of interpolation tables with equidistant
or stretched stepsizes in the p, c plane (e.g. Vinokur [8]).
Transport Quantities
The computation of viscous, heat conducting flow requires additional relations for the
transport coefficients of momentum, (viscosity), of energy (thermal conductivity), and
in mixtures of gases for the diffusion of species masses (diffusivity). For the situation
of chemical equilibrium, as considered here, the diffusion of species is usually assumed
to be negligible.
Laminar flow The transport quantities in laminar flow are functions of the molecular
properties, depending on the local thermodynamical state. In general the values and
the dependence are well known.
At moderate densities the coefficients of viscosity and heat conduction of a single
perfect gas are functions of the temperature only. Then simple models can be used
for the calculation of the viscosity of a single component gas, as e.g. the Sutherland
formula or potential laws. The corresponding thermal conductivity can be calculated
from the viscosity using the Eucken relation (Vincenti, Kruger [1]).
For a mixture of gas at equilibrium, where each component is assumed to be a
perfect gas, the composite viscosity can be determined from the semiempirical mixing
on
In the transition regime (not too far from the continuum) the flow is still sufficiently
represented by NavierStokes equations. The influence of rarefaction effects can be
restricted to the boundary conditions at the wall by using the socalled slip conditions (or incomplete accommodation of momentum and energy). Then the boundary
conditions for the temperature and the tangential velocity form a system of coupled
equations for the value, and for the gradient at the wall. They can be written in a
generalized form:
ali
AV+B=C
an
with
(14)
~Vol
ut
+ L:(F~y 
G~x)
(15)
k=l
Herein Q is the volumeaveraged value, (F~y  G~x) are the normal fluxes summed
up over the single cell interfaces. The advantage of the "FiniteVolume" approach
is the direct application to the physical (x,y) space, and the easy interpretation in
curvilinear meshes. This approach can be used for structured and unstructured meshes
as well.
The "FiniteDifference" approach, based on the divergence form, Eq. (2), is
slightly more expensive in its evolution, since first the differential equations have
to be transformed to the curvilinear coordinate system and afterwards the equations
are discretized in the transformed plane. The advantage of this approach is given for
the more complex terms like the viscous terms in the full NavierStokes equations.
For the 2D example the transformed equations in a curvilinear coordinate system
(~,1J) read:
(16)
with Q
QJ and F
FY7J  Gx7)' Here J is the metric Jacobian, which can be
interpreted as the volume, and x 7J ' etc. are parts of the surface normal vector. The
details can be found elsewhere. After the discretization in the transformed plane with
the corresponding (properly discretized) metric formulations, the resulting discretized
equations agree with discretization derived by the "Finite Volume" method.
In the "Finite Element" approach, the domain 0 is subdivided into finite elements and the solution vector Q is interpolated using (linear) shape functions. With
a weighted residual statement and by use of the Gauss theorem the conservation
equation can be cast in an approximate form, e.g. as in [20], [21].
~Q=
M k J
~t
JF:dOJF:nNdf
ax'
aNj
(17)
with the finite element matrix M jk For further details see the related literature.
Inspecting the three formulations in more detail, it can be seen that the Finite Element
formulations for conservation laws agree exactly or approximately with those from the
Finite Volume or Finite Difference discretizations, see e.g. [21]. It means the different
ways of conservative discretization result in equivalent formulations. Remarkable
differences for spatial discretization can occur by different arrangements of the control
volume, by the different numerical flux formulations, and by the different updating of
the cellinterface fluxes.
(18)
Herein ~Q / ~t is the discretized time derivative, defined later, and Fi 1/2 and Si1/2
are the numerical flux functions for the corresponding inviscid, and the viscous fluxes
at the cell interfaces.
The viscous fluxes are updated by means of central differences O(~X2), which
correspond to the elliptic nature of the viscous effects. Then a viscous term of the
form S = I1Ux may be written as
(19)
Cross derivatives can be treated in a similar way.
The numerical formulation of the inviscid fluxes has a great influence on the
properties of the solution method, since they contain the essential information of wave
spreading. These 'terms are strongly nonlinear and are mathematically more difficult
than the viscous terms. Therefore most schemes for the NavierStokes equations differ
by the formulation of the inviscid flux terms (Euler solver).
The numerical formulation can be divided into two parts: the evolution of the
numerical flux function at cell faces, which should be a consistent approximation of
the physical fluxes, and the projection of the volumeaveraged variables to the cell
faces for updating the fluxes.
In the "projectionevolution" approach, sometimes called "MUSCL" approach,
(van Leer [22]), the basic variables, e.g. Q are extrapolated from both sides to the
cell interface. With these values a common numerical flux formulation is formulated.
In the "evolutionprojection" approach, the "nonMUSCL" approach, the flux
functions are formed on the grid points, where the variables are stored, thereafter the
fluxes are projected to the cell interface. This approach is used e.g. for the "cellvertex" schemes or in the modified flux approach [23]. Also mixed forms of both may
be possible. The numerical flux functions become different for the two approaches, in
particular for higher order accurate schemes. However, which of the approaches is to
be prefered cannot be decided uniquely, discussions about that are made e.g. by Yee
[24].
Evolution of numerical fluxes
The evolution of numerical fluxes can be considered as the solution of a Riemann
problem at the cell interface. Within each cell there exists an averaged value of the
43
conservative variables. The values at the interface to the neighbouring cell result in
a jump when they are extrapolated from the left and the right cellaveraged values.
According to the theory of the nonlinear hyperbolic Euler equations this jump of
the values generates a local Riemann problem, whereby information is transported
forward and backward by the different gasdynamical waves and shocks. The solution
of the local, exact Riemann problem results in Euler solvers of the Godunowtype [25],
[26], which describe very accurately the wave phenomena, but these methods are very
expensive. If the jump at the interface is considered to be weak, the Riemann problem
can be solved approximately using the characteristic solution of the Euler equations.
A large number of numerical flux formulations, called approximate Riemann solvers
can be derived with this assumption. Widely used approximate Riemann solvers are
the schemes of van Leer [27], Roe [28], Osher, Chakravarthy [29], Steger and Warming
[30], and many others.
The derivation of the approximate Riemann solvers starts from the 1D conservative form
(20)
The characteristic form of the Euler equations is derived by diagonalization of the
flux Jacobian A =
which results in an uncoupled system of characteristic Euler
equations. These equations describe the information transport along the characteristic
directions.
(21)
M,
(22)
An upwind difference formulation of this equation for the cell "i" with the faces i 1/2
IS:
~W
~t
A+
+ ~x (Wi~I/2 
A
Wi::. 1 / 2 )
+ ~x (Wi+1/ 2 
Wi=I/2)
=0
(23)
Herein W are the backward or forward interpolated variables according to the sign
of the eigenvalues. In the simplest (first order accurate) case they are Wi~I/2 = Wi
W;+1/2 = Wi+l .
The desired conservative difference form can be constructed from the characteristic form by assuming a locally constant characteristic field. This is achieved by
multiplying the characteristic system, Eq. (23) with the eigenvector matrix T, which
results in a conservative form:
and
(24)
The flux conservation requires that F = F+ + F. The values Q are the cell
face values extrapolated from left and right. In the literature there are different
formulations for fluxvector splitting.
The fluxvector splitting by Steger, Warming [30] uses the eigenvalue splitting
A = 1/2 (A IAI). Then the numerical fluxes are defined by:
(26)
This split flux results in a discontinuous numerical eigenvalue whenever the corresponding physical eigenvalue vanishes, van Leer et al. [31]. Therefore it causes a
finite dissipation for steady waves, favorably for strong shocks, but too dissipative for
viscous computations. This property is utilized for a robust hypersonic code by Eberle
et al. [32] using a blended formulation of Steger Warming fluxes and the characteristic flux approach. Furthermore the eigenvalues of Steger Warming fluxes change
discontinuously near the sonic point, which reduces the rate of convergence, as found
in test calculations.
The fluxvector splitting by van Leer [27] avoids the latter drawback of the Steger,Warming fluxes by defining the split fluxes as polynomials of the Mach number,
such that they are smooth near the sonic point. Since one eigenvalue vanishes over
a steady shock, its representation becomes very sharp. Furthermore this splitting, if
applied in implicit schemes, results in very efficient, diagonaldominant solution methods, e.g. Thomas and Walters [33], Schroder and Hanel [34]. However, this splitting
formulation also shows a remaining splitting error for steady tangential discontinuities.
This drawback influences strongly the accuracy in viscous layers, as shown by Hanel,
Schwane [35]. Therefore a number of modifications was developed and reported e.g.
by van Leer [36] and by Hanel, Schwane [35J. Using these modifications the accuracy
in viscous flows could essentially be increased, as shown by Hanel et al. [35], [37].
The fluxdifference splitting method Roe [28], can be derived from Eq. (23) by using
the eigenvalue splitting A = HA IAI). Rearranging Eq. (23) for A and IAI and
applying the conservative back transformation, the numerical flux component Fi+1/2
.
_!
F'+l/ 2  2 (F(Q;+1/2)
 T1 (Qi+l/2
+
+ F(Q;+l/2))
+ !2T IAI

Qi+l/2)
(27)
The matrices f' and A are evaluated with Roe averaged values Q for the cell interface,
which shows an improved shock representation.
The eigenvalues are correct represented by this scheme, even if one eigenvalue
vanishes in A. Just this violates the entropy condition in form of expansion shocks
and reduces the robustness of schemes. Following Harten [23] the eigenvalues are
bounded by a so called entropy correction 8, which is a small number of order of
0(10 1 ), scaled by a typical velocity. This correction can be used in different forms,
e.g.:
(28)
Effects of the entropy correction 8 in viscous flow solutions are studied e.g. in [37].
Central fluxes The central schemes can be considered as a special class of approximate Riemann solvers. Then the characteristic range of influence is equally weighted,
independently of the Mach number by assuming A+ = A = A/2. The numerical flux
at the cell interface results in:
(29)
The central formulation includes the correct eigenvalues of the flux Jacobian A, and
the spatial discretization error is of second order. However since this numerical approach ignores the characteristic range of influence, the numerical solution shows a
poor shock capturing capability. Furthermore the central formulation does not include
dissipative parts (even derivatives) thus any highfrequency error components of the
solution cannot be damped. Therefore artificial damping terms for highfrequency
damping and shock capturing are added [38].
In general, the damping formulations consist of a linear fourth order term d(4) and
of a nonlinear term d(2). Then the numerical flux reads:
Fi+l/2 =
21 ( Fi+1 + Fi ) + di(4)+l/ 2 
(2)
di +l/ 2
(30)
The high frequency damping term d(4), necessary to smooth errors of short wave
lengths, (e.g. roundoff errors) has the form:
d(4) =
(.::(4)
.6.x) .6.x4 . Q
.6.t
xxx
(31)
(32)
+ I\:)~ Q + (1 
I\:)~ +Q)i+l
1
+ Fi ) + 2"T <I>i+1/2
(34)
The second term corresponds to the modified Roes upwind term in Eq. (27), T is the
eigenvector matrix and <I> contains cell face differences of the characteristic variables
controlled by flux limiters. Examples for this flux splitting are e.g. the "symmetric"
and "upwind" TVD schemes of Vee and Harten. A detailed report about the different
formulations is given by Vee in [24].
The preceding consideration of different Euler solvers covers only a small part of
existing methods, which are in their details so manifold as the authors are. Therefore
critical studies and comparison of the different flux formulations are necessary, in
particular for application to viscous flows. Much information can be found in Vee's
paper [24]. Analysis of different flux algorithms were carried out e.g. by Grossman
and Walters [48], by Montagne, Vee and Vinokur [49] and by Kroll et al. [50]. The
behaviour in viscous solutions were studied e.g. by van Leer et al. [31] and Hanel et
al. in [51], [35], [52], [37].
METHODS OF SOLUTION
Commonly the timedependent NavierStokes equations are used for solutions of unsteady, and stationary flow problems as well. The advantage of using the timedependent equations for steadystate computations is that the initialboundary value
problem remains parabolic or hyperbolic in the timespace plane, independent from
the Mach number range. Therefore one and the same numerical method can cover
small and the time accurate computation becomes expensive if additional smaller scale
lengths have to be resolved, like in viscous flows at high Reynolds numbers or in stiff
chemical rate equations. For this reason time accurate computations with an explicit
scheme may be still efficient for inviscid flows and viscous flows at moderate Reynolds
numbers, but for high Reynolds numbers some other ways, like implicit schemes or
multigrid methods should be taken into consideration.
For steadystate computations the transient solutions have no meaning, as long as
the steadystate solution will not be influenced by them. Therefore consistency and
accuracy in time is not required, the method of solution (in time) can be chosen e.g.
for an optimal rate of convergence. This results in classes of pseudo timedependent
methods which correspond to iteration schemes. Herein different acceleration strategies for improving the convergence can be applied, like e.g. GaussSeidel relaxation,
LowerUpper matrix decomposition, local time stepping or the multigrid methods.
The methods of solutions and their variants are numerous, therefore in the following
section only some of typical methods of solution for the NavierStokes equations will
be discussed. For the discussion the twodimensional NavierStokes equations are
considered. The treatment of the 3D equations is analogous. The equations in
conservative form, written for a curvilinear coordinate system, let's say (~, T/, t), read:
(35)
Details of the transformed equations can be found elsewhere. With a conservative
discretization in a way as described above the approximated NavierStokes equations
(35) read:
!:lQ
!:It
+ Res(Q) =
(36)
where !:lQ /!:It is the discrete time derivative, and the residual Res (Q) corresponds
to the discretized steadystate operator:
(37)
The fluxes F, S, etc., are the numerical fluxes, upwind or centrally discretized, and
the difference operators be , bTl mean:
Linear multistep methods for the discretized conservation equations were investigated
by Beam and Warming [55]. Practical applications are restricted usually to implicit
one and twostep methods. A general method of solution, based on two time levels,
may be written as:
Qn+!
Qn
~~
!:It
= 0
(38)
For = 0 the scheme is explicit of O(~t), with = 1/2 the implicit CrankNicholson
scheme, O(~t2), and with e = 1 the implicit Eulerbackward scheme is achieved.
Implicit schemes
The formulation of an implicit scheme requires time linearization of the unknown
fluxes in Res(Q)n+l. This yields:
Res(Qt+ 1 =
Res(Qt+~t.aa Res(Qt+ =
t
The time linearization of the components of Res( Q)n+l result in the implicit scheme:
I
[~t
+ e {8eA + 81)B 
Re (8 eC
A, B, 6, b
~
+ 81)D)}]n
~Qn =
Res(Qn)
(40)
F,e,s and T.
The discrete steadystate operator on the right hand side (RHS) is usually discretized higher order accurate with a numerical flux formulation, as described before.
Thus the RHS determines the spatial accuracy of the solution.
For timeaccurate solutions, the implicit operator on the left hand side (LHS) has
to be developed consistently to the RHS to achieve time accuracy. Then the LHS
results in a pentadiagonal block matrix system for each direction.
For steadystate calculations, i.e. RH S t 0, consistency in time is not required,
and therefore the LHS can be manipulated to achieve faster convergence. To reduce the computational work, and to increase the diagonaldominance of the solution
matrix three point stencils are used for the LHS. Thus only systems of tridiagonal
block matrices have to be inverted. Such approximations can be achieved in upwind
methods by using first order upwinding for the LHS, partially with additional simplifications (e.g. spectral radius instead of complete eigenvalues [56]). This idea is also
applicable to central schemes as shown e.g. in [56] or special damping terms are used
there.
Even with such approximations for the LHS, the implicit scheme, Eq. (40) requires
the inversion of a large system of difference equations. In general the direct inversion
of the solution matrix for more than one dimension is too expensive, therefore in
most cases the matrix is solved in simpler, approximate steps, using factorization or
relaxation methods.
Approximate Factorization method (AF) The Approximate Factorization method
(AF) by Beam and Warming [55], was the most used implicit scheme in the seventies. The solution matrix is split in onedimensional operators, which are sequentially
inverted by a Gauss elimination method. For the present example, Eq. (40) with
= 0, the solution steps of the AF method are:
[1+~t8e(A ~e6)] ~Q
[I
+ ~t 81)(B  ~e
b)] ~Qn
(41)
0;/
Qn+I
Q(O)
~t
Res( Q(II))
+ ~Q(I)
( 44)
= QN
A second acceleration technique is the implicit residual smoothing [67], for which
with appropriate smoothing coefficients the CFL number can be increased by a factor
of two to three.
A further, now widely used acceleration technique, is the multigrid method, as
proposed by Jameson [67] for the RungeKutta scheme. More details about multigrid
are given in a later section.
Applications of the RungeKutta method to the 3D NavierStokes equations are
reported e.g. by Vatsa and Wedan [40], Cima and Yokota [68], RadesPiel' Rossow
and Swanson [41] on structured grids, and by Mavriplis [21] on unstructured grids.
Timeaccurate calculations, using the RungeKutta method are given e.g. by Meinke
and Hanel [69].
Methods on unstructured grids
Finite Element as well Finite Volume methods on unstructured triangulated grids
enable geometrical flexibility and adaptive meshing for a high degree. Such methods
have proved valuable for solutions of the Euler equations in complex 2 D and 3D flows,
e.g. [70], [71], [72]. The same concept applied to the compressible N avierStokes
equations for high Reynolds numbers has been not so far developed as for inviscid
flows. The major reason is the presence of very different scalings in viscous flows.
This has the consequences of higher grid resolution, the directionality of gradients
in viscous layers and the corresponding deformation of grid cells. Furthermore the
implementation of frequently used algebraic turbulence models is more difficult than
in structured grids. About the latter item an approach can be found in [73]. For
these reasons attempts are made to combine structured meshes in viscous zones with
unstructured meshes outside. Computations on such hybrid meshes are reported e.g.
in [74], [75]. Another approach is to use structured meshes in one direction normal
to the wall, where viscous terms are important, and in the other directions of a 3D
grid unstructured organization [76] is used.
Despite of the difficulties mentioned above (and because of the advantages against
structured meshes) the number of computations for the NavierStokes equations on
fully unstructured meshes shows an increasing tendency. NavierStokes computations
on unstructured grids are shown e.g. in [21], [77], [78]. A large number of 2D and 3D
hypersonic applications of finiteelement methods for reacting and nonreacting flows
were presented by Periaux [79] and Mallet [80], summarizing the works of INRIA and
AMD BA in France.
For the solution of the NavierStokes equations on unstructured grids nearly all
of the numerical flux approaches were used as known in structured grids. Typical
approaches are central RungeKutta scheme with multigrid, e.g. Mavriplis [21], central
twostep LaxWendroff scheme, e.g. [78], Roe's fluxdifference splitting, e.g. [81], van
Leer's fluxvector splitting, e.g. [82], TVD Osher scheme, e.g. [83], fluxcorrected
transport approach, e.g. [78]. The methods of solution are explicit as well as implicit
schemes, details can be found in the related papers.
Summarizing one can say that methods on unstructured grids for the N avierStokes equations have a great potential for geometrical complex flow problems and
(45)
Herein .6.Q;/.6.t is the discretized time derivative, defined by the method of solution,
and Fi 1/2 are the numerical fluxes at the cell interfaces i 1/2.
Central schemes The central schemes are widely used for approximating the Euler
equations. However central differences do not result in dissipative truncation errors,
therefore such artificial damping terms have to be added. As mentioned above, the
damping formulations consist of a linear fourth order term d(4) and of a nonlinear
term d(2). Then the numerical flux reads:
( 46)
The high frequency damping term d(4), necessary to smooth errors of short wave
lengths (e.g. roundoff errors) has the form as given in Eq. (31). The shock capturing
term d(2), Eq. (32), has to suppress oscillations from the nonlinear terms in particular
for shocks.
The effect of the damping terms becomes evident by considering the truncation
error. Applying the numerical formulation, Eq. (46), to the linear, hyperbolic model
equation Qt + Qx = 0, the truncation error in space for a central discretization with
damping terms is:
_ ((2).6.X)
.
_ ~ 2.
Ts c .6.t .6.x Qxx 6.6.x Qxxx
(c
(4).6.x
3.
.6.t).6.x Qxxxx+ ...
(47)
(48)
The variables Q are variables extrapolated from the left (+) and the right () to
the cell interface. For the higher order extension of upwind schemes, van Leer has
suggested the so called MUSCL approach [22], where the variables Q are extrapolated
with a higher order polynomial from both sides to the cell interface, Eq. (33).
The damping properties of such an upwind scheme are strongly influenced by
the extrapolation, Eq. (33), controlled by the discretization parameter K, and by
the limiter r.p. This can be demonstrated by means of the spatial truncation error,
analogous to Eq. (47) by applying the extrapolation (33) to the linear model equation
Qt + Qx = 0. The truncation error for Qx yields:
.6. x
Ts
.6. x 2
.6.x 3
The fluxdifference splitting concept, as proposed by Roe [28], Harten [23], Vee [24],
and others, results in a first order upwind scheme, which consists of a central flux
term and an upwind term taking into account the different directions of information.
To extend the fluxdifference splitting to second order accuracy the MUSCL principle,
Eq. (33), can be used, or an additional term 9i+1/2 can be added, according to the
modified flux approach, Eq. (34), by Harten [98].
Fi+1/ 2 =
1
2 (Fi+l + Fi ) + 2T ~i+1/2
(50)
The different formulations for the modified flux are summarized and discussed e.g.
in [24]. With a simplified form of this flux the spatial truncation error Ts of flux
difference splitting, applied to the linear model equation Qt + Qx = 0, can be written
as:
zone on the "canopy" , which could be resolved by the computations using the modified
fluxes in agreement with experiment.
Another significant influence of the numerical flux formulation in context with van
Leer's splitting was noticed in supersonic bluntbody calculations where significant
deviations of the computed wall temperature were found in regions of large Mach
number gradients [101]. Investigations have shown that this effect is mainly caused
by the nonpreservation of the total enthalpy H t using the original split energy flux.
This effect could be removed substantially by using an alternate split energy flux [101]:
(52)
where the enthalpy H t is transported as a whole by the split mass fluxes Ft It has
the advantage of being simpler and of being generally valid also in the case of real
gases. The improvement of the accuracy by this formulation is clearly demonstrated
in Fig. 9 for the wall temperature distribution over a 3D hemispherecylinder body.
The idea of this alternate split energy flux, Eq. (52), was extended to generalized
fluxvector formulations in [32]. Using this change of split energy flux a better preservation of total temperature could be stated.
APPLICATIONS TO HYPERSONIC VISCOUS FLOW
Typical properties of hypersonic flow
Viscous hypersonic flow in the continuum regime are characterized by some typical
features, which are not or not so strongly present in other flow regimes, like subsonic
and transonic flows. Therefore numerical methods have to satisfy some special requirements for simulating such flows. Typical properties of hypersonic flow are:
Strong shocks are appearing. The strongest shock, the bow shock, can either be captured or fitted by numerical methods. Imbedded shocks are weaker and are captured
in general. Therefore good shock capturing methods are required.
Strong expansions can occur, e.g. by transient shock movements e.g. in the initial
phase, or stationary from the stagnation point around the nose. Expansions are continuous solutions, but with a rapid decrease of temperature and density. The decrease
can result in very small values or even in negative values for temperature or density for
accelerated numerical methods during the nonphysical transient phase. Very small
or negative values mean loss in convergence or even breakdown for the numerical
method.
Hypersonic vehicles are blunt bodies or airplanes of low aspect ratio. Therefore the
flow is strongly threedimensional and thus powerful numerical methods in general coordinates are required.
The 3D, viscous flow is rather complex, even when the boundary layer theory is valid.
But in many cases this theory cannot be applied since strong viscousin viscid interactions and strongly separating flow for high angle of attack is occurring. To resolve the
viscous phenomena, fine grids and highly accurate methods of solution are required.
An example for using an explicit scheme for hypersonic flows is the "historical"
paper by Mac Cormack, 1969, [53], where the wellknown predictorcorrector method
was applied to the NavierStokes equations. Also nowadays this method is in use
as a reliable and cheap method of solution. A very complex application for that is
given by Shang, Scherr [54], where the hypersonic flow around a complete aircraft was
computed with an algebraic turbulence model included.
Hybrid explicitimplicit schemes are a compromise to avoid the severe stability restriction by an implicit procedure in direction normal to the wall. An example is
given by Riedelbauch, Kordulla [102] for the preceding test problem.
Implicit methods and relaxation schemes are more robust, if strong shocks and very
different characteristic scale lengths have to be resolved, as it is in high Reynolds
number flow or chemical reaction flow. Typical approaches for this type of method
are discussed in a preceding section.
Fully implicit relaxation schemes, based on higher order upwinding and diagonaldominant solution matrices, have shown very efficient convergence properties and
shock capturing capabilities in hypersonic applications. An example of that is the
3D relaxation method by Schwane, Hanel [35], [99], based on fluxvector splitting.
Results were presented above (e.g. Fig. 8).
A number of hypersonic viscous flow applications are presented in the paper by
Schmatz [103]. The method of solution, called NSFLEX, is based on a third order
accurate local characteristic flux extrapolation, blended with a fluxvector splitting of
StegerWarming type for capturing of strong shocks. The unfactored implicit scheme
is solved by a point GaussSeidel relaxation method. The fluxvector splitting was
reformulated in [32] in a generalized flux formulation, including the modified split
energy flux, as proposed in [101].
The application of fluxdifference splitting (Roetype) to hypersonic flows shows
an increasing tendency. The high accuracy of this splitting and sufficient robustness
can be combined by special implicit relaxation formulations for steadystate solutions. Implicit relaxation schemes for the 3D N avierStokes equations, based on
fluxdifference splitting, are described e.g. by Miiller [104] and by Riedelbauch and
Brenner [65]. In [65] comparison was made between the upwind method and a central
scheme including fitting of the bow shock [105]. Nearly the same performance and
accuracy was reported for both schemes, but with a superiority for the upwind scheme
if strong embedded shocks are present. The relaxation scheme in [104] also can be
used in a parabolic space marching manner.
Recently Schroder [66] has presented a 3D relaxation method for the NavierStokes equations, based on fluxdifference splitting, and demonstrated the capability
to deal with very complex hypersonic, laminar and turbulent flows. As an example
from [66], Fig. 10 presents the results for the laminar flow during the stage separation
of a twostage space vehicle at M a oo = 6. and Re/m = 4.2 x 10 5 A two block
mesh was used with 51 x 139 x 90 and 40 x 139 x 30 grid points. The Fig. 10 shows
the corresponding surface mesh and the computed lines of constant density for the
References
[1] Vincenti, W., Kruger, C.: Introduction to Physical Gasdynamics. Wiley and Sons, Inc., New
York, 1967.
[2] Schiff, L.B., Steger, J .L.: Numerical Simulation of Steady Supersonic Viscous Flow. AIAApaper, No. 790130, 1979.
[3] Vigneron, Y.C., Rakich, J .V., Tannehill, J .C.: Calculation of Supersonic Viscous Flow over
Delta Wings with Sharp Supersonic Leading Edges. AIAApaper, No. 781137, 1978.
[4] Menne, S, Weiland C.: SplitMatrix Marching Methods for ThreeDimensional Viscous and
Inviscid Hypersonic Flows. Notes on Numerical Fluid Mechanics, vol 29, Vieweg Verlag, 1990.
[5] Cline D. D., Carey G.F.: Shock Sensitivity in Parabolized NavierStokes Solution of High
AngleofAttack Supersonic Flow. AIAA Journal, vol. 28, pp. 406413, 1990.
[6] Power, G.D., Barber, T.J.: Analysis of complex hypersonic flows with strong viscous/inviscid
interaction. AIAAJ., vol. 26, No.7, pp 832840, 1988.
[7] Srinivasan, S., Tannehill, J .C., Weilmuenster, K.J.: Simplified curve fits for the thermodynamic properties of equilibrium air. NASA RP 1181, 1987.
[8] Liu Y., Vinokur M.: Equilibrium Gas Flow Computations. I Accurate and Efficient Calculation
of Equilibrium Gas Properties. AIAA paper No. 891736, 1989.
[9] Bird, R.B., Stewart, W.E., Lightfoot, E.N.: Transport Phenomena. Wiley and Sons, Inc., New
York, 1960.
[10] Srinivasan S., Tannehill J .C., Weilmuenster K.J.: Simplified Curve Fits for the Transport
Properties of Equilibrium Air. Iowa State Univ., ISUERIAmes 88405, 1987.
[11] Baldwin, B.; Lomax, H.: Thin Layer Approximation and Algebraic Model for Separated Thrbulent Flows. AIAApaper, No. 78257, 1978.
[12] Vandromme D.: Thrbulence Modelling and Implementation in NavierStokes Solvers. VKI
Lecture Series LS 198906, von Karman Institut for Fluid Dynamics, RhodeSaintGenese,
1989.
[13] Gokcen, T., Mac Cormack, R.W., Chapman, D.R.: Computational fluid dynamics near the
continuum limit. AIAA paper, No. 871115, 1987.
[14] Dutt. P.: Stable Boundary Conditions and Difference Schemes for NavierStokes Equations.
ICASE report No. 8537, 1985.
[15] Nordstrom, J.: Energy Absorbing Boundary Conditions for the NavierStokes Equations. In:
Lecture Notes in Physics, vol 264, Springer Verlag, 1986.
[16] Lax, P.D., Richtmeyer, R.D., Comm. on Pure and Appl. Math., vol 9, 1956.
[17] Jameson, A., Baker, T.J.: Euler Calculations for a Complete Aircraft. Lecture Notes in Physics,
vol. 264, SpringerVerlag Berlin, 1986.
[18] Rossow C.: Comparison of Cell Centered and Cell Vertex Finite Volume Schemes. In: Deville,
M. (Ed.), Notes on Numer. Fluid Mech., vol 20, pp. 327334, Vieweg Verlag 1988.
[19] Dortmann, K.: Computation of Viscous Unsteady Compressible Flow about Airfoils. In: Proc.
of the ICllNMFD, Lecture Notes in Physics, Springer Verlag, 1989.
68
[20] Lohner, R., Morgan, K., Zienkiewicz: The Solution of NonLinear Hyperbolic Equation Systems by the Finite Element Method. Int. J. of Numer. Methods in Fluids, vol. 4, pp. 10431063,
1984.
[21] Mavriplis D. J., Jameson A.: Multigrid Solution of the NavierStokes Equation on Triangular
Meshes. AIAA Journal, vol. 28, pp. 14151425, 1990.
[22] van Leer, B.: Towards the Ultimate Conservative Difference Scheme. A secondorder sequel to
Godunov's method. J. Compo Phys. vo1.32, pp.101136, 1979.
[23] Harten, A.: High Resolution Schemes for Hyperbolic Conservation Laws, J. Compo Phys., vol.
49, pp. 357393, 1983.
[24] Yee H. C.: A Class of HighResolution Explicit and Implicit ShockCapturing Methods. In:
VKI Lecture Series 198904, RhodeSaintGenese, 1989.
also: Yee H. C.: Upwind and Symmetric ShockCapturing Schemes. NASA TM89464, 1987.
[25] Godunov, S.K.: FiniteDifference Method for Computation of Discontinuous Solutions. (in
Russian), Math. Sbornik vol.47, pp 271, 1959.
[26] Collela, P., Glaz, H.M.: Efficient Solution Algorithms for the Rieman Problem for Real Gases.
J. Compo Phys., vol 59,pp 264, 1985.
[27] van Leer, B.: FluxVector Splitting for the Euler Equations. Lecture Notes in Physics vol. 170,
pp. 507512, 1982.
[28] Roe, P.L.: Approximate Riemann Solvers, Parameter Vectors and Difference Schemes. J. Compo
Phys., vol. 22, pp. 357, 1981.
[29] Osher, S., Chakravarthy, S.: High Resolution Schemes and Entropy Condition. SIAM J. Num.
Anal., vol 21 pp. 955984, 1984.
[30] Steger, J .L., Warming, R.F.: FluxVector Splitting of the Inviscid Gas Dynamic Equations
with Applications to FiniteDifference Methods. J. Compo Phys., vol 40, pp 263293, 1981.
[31] van Leer, B., Thomas, J. L., Roe, P. L., Newsome, R. W.: A Comparison of Numerical Flux
Formulas for the Euler and NavierStokes Equations. AIAA paper 871104 CP 1987.
[32] Eberle A., Schmatz, M., Bissinger, N.: Generalized Fluxvectors for Hypersonic Shock Capturing. AIAA paper No. 900390, 1990.
[33] Thomas, J.L., Walters, R.W.: Upwind Relaxation Algorithms for the NavierStokes Equations.
AIAApaper No. 851501, 1985.
[34] Schroder, W., Hanel, D.: An Unfactored Implicit Scheme with Multigrid Acceleration for the
Solution of the NavierStokes Equations. Compo &. Fluids, vol 15, pp. 313336, 1987.
[35] Schwane, R., Hanel, D.: An Implicit FluxVector Splitting Scheme for the Computation of
Viscous Hypersonic Flow. AIAApaper No. 890274, 1989.
[36] van Leer B.: Flux Vector Splitting for the 1990's. CFD Symp. on Aeropropulsion, NASA Lewis
Res. Center, Cleveland, 1990.
[37] Seider G., Hanel D.: Numerical Influence of Upwind TVD Schemes on Transonic Airfoil Drag
Prediction. AIAA paper 910636, 29th Aerospace Sciences Meeting, Reno (USA), Jan. 1991,
1991.
[38] Jameson, A., Schmidt, W., Turkel, E.: Numerical Solution of the Euler Equations by FiniteVolume Metaods Using RungeKutta TimeStepping Schemes. AIAApaper 811959, 1981.
72
[96] Gerolymos G. A.: Implicit Multiple Grid Solution of the Compressible NavierStokes Equatios
Using k Turbulence Closure. AIAA Journal, vol. 28, pp. 17071714, 1990.
[97] Meinke M., Hanel D.: Application of the Multigrid Method in Solutions of the Compressible
NavierStokes Equations. Proc. of the Fourth Copper Mountain Conf. on Multigrid Methods.
(Mandel J. et al, Editors) SIAM, Philadelphia, 1989.
[98] Harten A.: On a Class of High Resolution TotalVariationStable FiniteDifference Schemes.
SIAM J. Num. Anal., vol. 21, pp 123" 1984.
[99] R. Schwane, D. Hanel: An Upwind Relaxation Method for Hypersonic Flows. Computer &
Fluids, vol. 19, 1991.
[100] Workshop on Hypersonic Flows for Reentry Problems. Organized by INRIA and GAMNISMAI, Jan. 1990, Antibes, France, 1990.
[101] Schwane, R., Hanel, D.: Computation of Viscous Supersonic Flow around Blunt Bodies. 7th
GAMM Conf. on Num. Math. in Fluid Mech., LouvainLa Neuve, 1987.
[102] Riedelbauch, S., Miiller, B., Rues, D.: 3D Viscous Hypersonic Flow over Blunt Bodies. DVLRFB 8909, 1989.
[103] Schmatz, M.: Hypersonic Threedimensional NavierStokes Calculations for Equilibrium Gas.
AIAA paper No. 892183, 1989.
[104] Miiller, B.: Development of an Upwind Relaxation Method to Solve the 3D Euler and NavierStokes Equations for Hypersonic Flows. GAMM Wissenschaftliche Jahrestagung, Hannover,
1990.
[105] Riedelbauch, S.,Miiller, B., Kordulla, W.: SemiImplicit Finite Difference Simulation of Laminar
Flow over Blunt Bodies. Proc. of 11th ICNMFD, Williamsburg, 1988.
[106] Prabhu, D.K., Tannehill, J.C., Marvin, J.G.: A New PNS Code for Chemical Nonequilibrium
Flows. AIAAJ .,vol 26, No.7, pp. 808815, 1988.


100~
______________~
,/,;;.;;~:~> .
.,.,.
_ 
,.
C'"
~~ = 0
o
" 0
"o:;';'COOoMO
Noslip
Moo = 20,
0,7
Generalized slip
~~ = 0
Maxwell slip
0,7
la'
Monte Carlo
~~ = 0
10'
Generalized slip + T.  Tw
10'
Re oo
10'
10'
Oi,j , i'i,j
node centered
Oi,j, ri,j
cellvertex
x Oi,j , ri,j
cell centered
0.08
(1)
(2)
(3)
(I.)
3
I.
(5)
')(. = 1
')(. = 113
')(.=0
')(. =1
Ijl = 1
Ijl = 1
Ijl = 1
Ijl = 1
Ijl=O
5
Fig. 3: Relative error between numerical and exact solution of the Burgers equation.
Influence of different discretizations, using the MUSCL approach.
2
UU.,x.
u.,x
0.08
1
2
(1 )
'l<.=
a }
3
0) IP = 1
b) IP = IPv.AlI!Al:lA
I.
Fig. 4: Relative error between numerical and exact solution of the Burgers equation.
Influence of the flux limiter, using the MUSCL approach.
  no limiter (cp = 1)
    with limiter (cp =
 
CPv.Albada)

u:
0
U1
Cp
0
C
U1
0
I
Experiment
Cook et aI., 1979
FDS solution
minmodlimiter with w = 2
. minmodlimitcr with w = I
0
~
I
U1
~
1.0
x/c
Fig. 5: Computed pressure distribution for transonic, turbulent flow around a
RAE 2822 airfoil [37]. (Re = 6.5 x 106 , Ma oo = .73, 0: = 2.79).
Influence of different minmod flux limiters:
k'l}
'P ,. = minmod{2wr.
l+r'l+r'
= 2:
W = 1:
W
   
Cl
Cl
to.
= .823 , Cd = .0187
= .820 , Cd = .0216
05
Blasius
Cf
o Van Leer
6.
0.()I.
Vee/Harten
17<33
Rea:> :. lOS
Ma~
0.03
= .5
0.02
0,01
0.5
X/L _
1.0
Fig. 6: Solution of the 2D NavierStokes equations for laminar flow over a flat plate.
Influence of different upwind formulations on the values of skin friction:
fluxvector splitting, van Leer [27],
fluxdifference splitting, Vee [24],
O(~2)
O(~2)
FluxVectorSplLtting
0
6.
Blasius
Van Leer
present
0,03
o
0.02
0.01
0.1
0.2 0.3
0.8 0.9
1.0
x/L
Fig. 7: Solution of the 2D NavierStokes equations for laminar flow over a flat plate.
Influence of different formulations for fluxvector splitting on the values of
skin friction:
original fluxvector splitting by van Leer [27]
modified splitting by Hanel, Schwane [35]
a)
Ma.,. = 2.91.
Re.,. = 2.2)(105
1,1
Pr
=.72
1,0
0.9
2._
Fig. 10: NavierStokes solution for the hypersonic, laminar flow during the stage separation of a twostage space vehicle. (M a oo = 6., Re/m = 4.2 x 10 5 , 0: = 0).
a) The two surface grids
b) Lines of constant density in the symmetry plane
Fig. 11: Computation of the inviscid and viscous flow around a space shuttle, [56].
(Ma = 8,Re/m = 106 ,0: = 30).
Isomach lines in the cross section x = 12.7m for viscous (left) and inviscid
flow (right).
t.Orrr,
CfX 10
NS
  PNS
   RNS
HOLDEN AND
MOSELLE
0.5
OL_ _ _ _ _
______
XlL
Fig. 12: Comparison of the skin friction over a compression ramp calculated
using the NS, PNS, and RNS equations. (Results taken from [6]).
b)
Fig. 13: Computation of hypersonic flow around a twostage vehicle with a space
marching method [4]. (M a = 6.8, Re oo = 8 x 106 ,0: = 80 )
a) Isobars of inviscid flow on the upper surface
b) Isobars of inviscid flow in crosssection z
= 4.2
Chapter 3:
Numerical Simulation of Hypersonic Flows
J. S. Shang
Wright Laboratory, WrightPatterson Air Force Base,
Ohio 454336553, USA
ABSTRACT
Numerical simulation of aerodynamic and chemical kinetic phenomena encountered
in the hypersonic flow regime is outlined. In the continuum flow region, the current
numerical methods focus on solving the Navierstokes equations in conjunction with
chemical physics equations, including the nonequilibrium chemical reactions and internal degrees of excitation. In the rarefied gas domain, the stateoftheart progress
using particle simulation which does not rely on continuum concepts, such as the Direct Simulation Monte Carlo (DSMC) method, has gained confidence but is still in
the embryonic state, thus will only be briefly delineated. Similarly, the added mechanism of energy transport process in radiation is presented in the same line of thought.
UNIQUE FEATURES OF HYPERSONIC FLOWS
Rarefied Gasdynamics
In practical applications, the flight envelope of hypersonic vehicles encompasses the
entire domain of gas dynamics from free molecule to continuum regime. In these two
relatively well defined realms, the interaction of a flight vehicle and its surrounding
transits from an intermolecular collisions dominated environment to a circumstance,
in which particle collisions with the vehicle is critical. The Knudsen number, defined
by the ratio of the mean free path and a pertaining length scale of the flowfield, is
a measure of the degree of rarefaction [1,2]. The governing equations of these two
regions are the NavierStokes equations and the Boltzmann equation, respectively.
The two systems of equations are closely related through the kinetic theory. The
hierarchy of the pertaining governing equations in aerodynamic applications is given
in Figure 1. Although the shock wave structure usually dominates the hypersonic
flowfield, the interaction with a viscous layer is intensified as the Mach number increases. Separate treatment of the invisicid and viscous phenomena, as if they were
independent from each other, becomes increasingly inaccurate. Therefore, the inviscid
and the boundarylayer approximations have a rather limited range of applicability in
hypersonic flows. To define hypersonic flows by the ratio of the flow velocity relative
(1)
where Pi is the density of species i. The relative diffusion velocity of a chemical species
(Ui) is defined by the difference of the massaveraged value and the species diffusion
velocity with respect to a stationary frame of reference.
Ui
= Vi 
(2)
By definition, the sum of mass diffusion velocities of all species is identical to zero,
= O.
For a multicomponent chemically reacting system, the concentrations of various species are defined as: mass concentration, Pi, mass fraction, 0i = p;/ p, molar
concentration, Xi = p;/Mi' Mi denoting molecular weight, and the mole fraction,
f3i = X;/ I:i Xi. The fractions have the following properties, I: 0i = 1 and I: f3i = 1.
The relationship between mass and mole fraction is
I: PiUi
(3)
The diffusive mass flux is generated by three mechanical driving forces and a contribution by temperature gradient [13,25]. The forced diffusion of a species can be induced
by an external force exerted on electrically charged particles, and is important only
in an ionized gas system where the electric field is present.
In general, the mass flux vector is expressed by the firstorder spatial derivatives
of various physical properties and massaveraged velocity. Although the diffusive flux
vectors were originally derived for monatomic gases at low density condition, they have
been applied to polyatomic gases with little error [25]. According to Bird, Stewart,
and Lightfoot [13], the diffusive mass fluxes are given as:
Ordinary diffusive mass flux
(4)
",here c, Rand T denote the speed of sound, the gas constant and temperature
respectively, and with P denoting pressure and S denoting entropy, G i is the Gibbs
free energy, G; = E; + PV  TS.
Pressure diffusive mass flux
(5)
88
(6)
Thermal diffusive mass flux
(7)
=  DrlnT
PiUi
where Dij and DT are multicomponent diffusion and thermal diffusion coefficients
respectively. The Ii on the right hand side of the force diffusion equation is the
external body force, a field force.
In aerodynamic application, the composition of air mixture has small differences
in molecular structure and weight between Nitrogen and Oxygen. In the dissociated
state, the weight difference between molecular and atomic species are also similar.
Therefore, the limiting case of a twocomponent system is frequently used in place of
a general diffusion relation. The binary mixture approximation reduces the diffusion
mass flux to the following [13]
PiUi
PjUj
[8
(GMi
i
)
T,p
\l j3j  Pj
P
1) ]
I  Ii ) + ( Vi  Mi
\l P
(8)
If the forced and pressure diffusions together with thermal diffusion (Soret effect)
are neglected, then only the ordinary diffusion will contribute to the mass flux. For
hypersonic flow, the most important contribution to the mass flux is due to the concentration gradient. The relationship between diffusive mass flux and gradient of
molar concentration is known as Fick's first law of diffusion
P."U."

cD
tJ
'\7
V
13
t
(9)
In essence, the species i moves relative to the mixture in the direction of decreasing
mole fraction of this species. The physical unit of mass diffusivity is the same as that
of the kinematic viscosity. In hypersonic applications, the transport properties of air
mixture are usually determined by a group of dimensionless parameters such as the
LewisSeminov, Prandtl, and Schmidt numbers [11,13,14,25]. Therefore, if anyone of
the coefficients of the transport properties can be found, the rest are determined by
an approximated value of these parameters.
In a multicomponent system, the energy fluxes also are driven by the identical mechanisms of mass and momentum fluxes. In fact, there is a unique coupling
between mass and energy transport. In case the average velocity of any species differs from the massaveraged velocity of the gas mixture, the individual species will
carry specific enthalpy, equal to Piujhi across the control surface where h denotes the
enthalpy. This additional energy flux is contributed by the interdiffusion process.
Onsager's reciprocal relation of irreversible thermodynamic processes implies that if
temperature gives rise to the Soret effect in diffusion, the concentration gradient must
produce a heat flux. The coupling effect, known as the Dufour energy flux, provides
an additional contribution to the energy transfer. The Dufour effect is quite complex
(10)
where k is the thermal conductivity and qr denotes radiative heat transfer.
Internal Degree of Excitation of Gas Medium
From the thermodynamic point of view, the macroscopic state of a system must
be definable by a small set of state variables. The relationship between statistical
mechanics and thermodynamics is always discussed in terms of the partition function
(Z). Therefore, the partition function of a microscopic particle is fundamentally
important in relating atomic/molecular structure, which influence the energy level,
and thermodynamic behavior [11,14,25]. With k representing the Boltzmann constant,
(11)
Physically, it is a measure of the fraction of the total number of particles in the
system which possess an energy level /OJ. It describes how the energy is distributed
among particles of a system. Thus, the partition function has also been defined as the
normalizing constant for the quantum mechanical canonical distribution function [25].
/OJ is the energy level which quantum mechanics assigns to an atom/molecule on the
basis of its structure. An energy level may consist of all energy states having the same
values /OJ. Any energy level containing more than one state is said to be degenerate
and the number of states of degeneracy is denoted gi. Degeneracies are explicitly
related to the internal structure of the particles. Since large numbers of degrees
of freedom exist for a system, the accurate calculation of the partition function is
impractical. In applications, modeling of physics becomes necessary. Usually, the
coupling of internal degrees of freedom of gas medium is neglected, this assumption
leads to the factorization property of partition functions [11,14].
(12)
In accordance with commonly accepted convention, the energy of the ground level is
set to zero. At each level of internal degree of excitation, the quantity /O;f k has the
dimension of temperature, and is used to define the characteristic temperatures of the
internal degree of freedom. For air mixture, the composition changes according to
temperature encountered, and has increasing degrees of excitation as the temperature
rises. The energy of different internal degree of excitations is additive, the total
internal energy is then found by summing the translational, rotational, vibrational,
and electronic modes.
(13)
All the thermodynamic state variables can now be derived from the partition function.
P= NkT (~aZ)
zav
E = NkT2 (~ aZ)
ZaT v
(14)
S = Nk {In
(~) + 1 + ~ (~~) v}
(15)
(16)
From quantum mechanics, the partition functions of translational and rotational motion are well known. The translational energy levels are very closely spaced, the
distribution of energies can be considered to be continuous. Here, h denotes Planck's
constant.
27rmkT) ~
(17)
Ztrs = ( h 2
V
The partition function of rotational degree of excitation of a diatomic gas, based on
rigid dumbbell model, can be given as
T
Zrot = () ,
a
(18)
1 exp ~
(19)
Ev =
RBv
:(,...,)cexp ~  1
(20)
DEv
Dt
(21)
where E~ denotes the equilibrium value, and T is called the relaxation time, a function
of pressure and temperature. In theory, the LandauTeller formulation is valid even
for large departure from equilibrium, but in practice the range of validity is limited.
A part of the limitation is imposed by the harmonic oscillator model which is only
accurate in the lower vibrational quantum levels. Again, the basic premise of this
formulation rests on the vibrational ladder climbing mechanism. Energy transition
is permitted only to the nearest energy level. This model has exhibited discrepancy
when compared with experimental facts [15,17]. Research efforts by Park et at. [18,29]
have attempted to bridge the prediction shortfalls by introducing a two temperature
model.
The contribution to thermodynamic properties from the excited atomic and electronic states is probably negligible for temperatures below 10,000K. This assertion is
supported by the fact that the next electronic energy level lies far above the ground
state [14]. The electronic partition function is naturally grouped into degenerate
levels [11].
Ze/
= 90
~
+ 79;
exp (
Be;)
T
(22)
Again, Be,; denotes the corresponding characteristic temperature of electronic excitation. The only electron internal energy is associated with its spin, and contains
a degeneracy of two. The equilibrium composition of an ionized air mixture can be
analyzed by the the law of mass action, like that of dissociated air. The Saha equation
has been derived to yield the degree of ionization [11,14,30].
,,","
(23)
J1i X i = L J1i X;
,
where the J1~, and J1~ are the stoichiometric coefficients for reactants and products,
respectively. These coefficients are integers to denote the number of molecules that
partake in the reaction. Xi is the mole concentration of the reacting species.
The Law of mass action is actually an empirical formulation confirmed by numerous experimental observations. The law states that the rate of appearance or
disappearance of a chemical species is proportional to the products of the concentrations to a power equal to the corresponding stoichiometric coefficient, IIi
,
Rj,i = /{j,i (T) II (Xit
(24)
Rb,i
(25)
where Kj,i and Kb,i are the socalled reaction rate constants which are functions of
temperature only and independent of the concentrations of the reacting species. The
net rate of change in concentration of any species Xi is the combined result of the
individual, elementary processes of single, simplestep reactions. In a complex overall
reaction consisting of a number of elementary reactions, the rate of reaction is often
determined by the slower rate of an elementary process. In this case, stoichiometric
coefficients may not necessarily govern the overall reaction rate.
dX' ) = ("
f ) Kj)'II(X;)V;,)
f
vv
dt.
t,l
t,l
(I
+ (fvv.
Kb'II(X)V;,)
't,l'}
II
II
t,}
(26)
where the subscript j, denotes the jth forward or backward reaction. In order to
apply the law of mass action, the rate of reactions is assumed to be independent
of whether or not the system is in equilibrium [11,14). This commonly accepted
assumption is consistent with the local equilibrium approach. The rate constant is
II
given as Kc,j = Kj,j/ Kb,j = II (X;)"i,j /II(X i )";,) which is considered to be a function
of temperature but not of the nonequilibrium variables. The relationship is equally
applicable to the dissociationrecombination reaction, as well as any simple, singlestep reaction (11). Only when the gas mixture is in chemical equilibrium, the forward
and reverse reactions are in dynamic balance, and the net rate of change in compostion
vanishes.
dXi) =
( dt.
)
(V~f.
_ v' .) K j
',)
I,)
.
,)
(27)
CJ
K c,j = C exp ( 
I;~
(28)
This is the wellknown Arrhenius equation, and the constant fa is often refered to as
the Arrhenius activation energy (11). The exponential term is also frequently denoted
as the Boltzmann factor. The Arrhenius Law simply states that only those molecules
that possess energy greater than a certain amount of activation energy will lead to
reaction. This equation has been modified to better fit to measurements as
KC,T = CTTI exp ( 
I;~)
(29)
Sometimes CTTI is called the collision frequency to reflect its root in statistical quantum mechanics. Unfortunately, the reaction rate constants are known with a large,
varying degree of uncertainty. At present, it is an intensive area of research. The
works of Park [29,31) are used as the bearer of standards for high temperature air.
aXi
at
E,A,S
(30)
where the photon number density and distribution function are denoted by nand
j, the velocity vector by Ci . E, A and S refer to the contributions by emission,
absorption and scattering respectively. To understand mechanisms of emission and
absorption of radiation in a gas requires knowledge of molecular and atomic spectroscopy. Atomic radiation arises from the boundbound, boundfree, and freefree
transitions. Molecular radiation can trace their origins from the shifting of energy
levels between internal degrees of freedom. Since the radiant energy flux is directional and contains a range of frequencies, the energy transfer must be found in a
detailed description of the radiation field. In all, radiative intensity in one direction
is independent from any other direction, and is different from frequency to frequency.
These areas of special expertise are considered to be outside the scope of the present
discussion. However, several classic simplifications have been developed to evaluate
the radiation heat transfer and might be useful. In the Planck limit, emission is dominant, the optically thin gas approximation reduces the problem of just evaluating the
Planck absorption coefficient [11].
(31 )
where a denotes the Planck coefficient and a the StefanBoltzmann constant. Under
conditions of small departure from radiative equilibrium, the heat flux depends on the
first spatial derivative of the local temperature. The radiant heat flux can be obtained
by the Rosseland Diffusion approximation.
16aT3
qr =    VT
3ar
(32)
where a r denotes the Rosseland mean free path for radiation. Finally, the Greygas
approximation is also used to estimate the radiant energy flux. Since this approximation is not a rigorous average over the range of frequencies, accurate results cannot
be anticipated [11,18,32].
Several approaches have been developed to calculate absorption and emission under nonequilibrium conditions. Two of the most frequently adopted methods are
the correction factor and the detailed linebyline calculation over all frequency spectra [18,32,33). The correction factor method utilizes a simple equilibrium step model
for absorption, the effects of nonequilibrium population density are appended into
the calculation by a correction factor [33). On the other hand, the linebyline calculation of spectra is capable of fully accounting for all absorption phenomena. Park et
al. [18) have found that under earth reentry conditions, the radiation emission from
nonequilibrium region is less intense than that from the equilibrium domain.
Macroscopic Conservation Equations
The complete conservation equations of the multicomponent reacting system for hypersonic flows can now be written as:
The species conservation equations:
apai
at
+ \1. ( pai u + paiui )=.ri
(33)
where r denotes the rate of production. For an N number species system, only
N  1 species conservation equations are required. Anyone of the N equations may
be replaced by the global equation of continuity for the mixture. In practice, all
species conservative equations are usually computed and verified by the balance of
the total number of atoms in the reacting system. The global continuity equation
can be obtained by summing over all species conservation equations and by noting
EPiUi = 0, and Eri = O.
(34)
which is the most fundamental principle of Newtonian mechanics; matter can be
neither created nor destroyed.
The conservation of momentum equations:
apu
(35)
= p L..fi
The conservation of momentum equation is the only vector equation in the system,
and it is the Newton's second law of motion. The nonlinear convective transport is
represented by the dyadic, the socalled inviscid term of the system. The shear stress
term (1') is a second rank tensor and can be given as
(36)
where A and Jl are the bulk and molecular viscosities respectively, and j is the identity
matrix.
The conservation of energy equation:
pe
aa
t
p Ladi (u
.
I
+ Ui)
(37)
k \1 T
p Ladi. (u
+ Ui)
(38)
where e is the total energy and ~hi is the standard heat of formation of species i.
The system of equations will be formally closed when all the auxiliary equations for
transport properties, state variables and chemical reaction rate are prescribed. The
description of the differential system is completed only by incorporating all appropriate initial and boundary conditions.
In numerical simulations of hypersonic flows, this system of conservative equations
is usually written in strong conservation form.
(39)
au aF aG _ 0
at + ax + ay where
u={
{~ }
pe
={
PuS
(pe
+ P) u
}G
={
(40)
pi:
(pe
(41)
+ P) v
It is well known, that the flux vector of the hyperbolic system of equations is a
homogeneous function of degree one in the elements of U [36]. By means of chainrule
98
au
au
au
at + A 8x + B 8y
= 0
(42)
u~'\ u~'\ ~
o
o
1~P ]=0
u'\
0
0
u  ,\
,p
(43)
10]
1/C
1/c2
5 =
1I pc 1 I pc 0 0
0
0 1
[ o
1
(44)
0 0
If the lefthand inverse of the similar matrix 5 can be found, then the diagonalized
matrix A or B is realizable.
(45)
1
10
5 =
[ oo
pc/2
0
pc/2
(46)
99
The relationship between the diagonalized coefficient matrix and the characteristic
variables is derived in the following
aU
aU
at + AaX =
aU
at
A=
SAS1 aU
aX
(47)
=0
(48)
~ u~c ~
[
o
o
0
0
u
0
Sl aU
at
AS 1 aU =
aX
0
(49)
Assume Sl is a slowly varying local value function which allows the matrix to be
brought inside of the differential operator. This is the linearized approximation usually
needed in the differencing formulation to yield
Ow
ow
 + A  =0
at
ax
'
(50)
~ (p at
P) + u~
(p  P) = 0
ax
c~
c~
(51 )
(52)
(53)
(54)
The subscript 0 indicates that the variables are evaluated at local conditions. Similarly, the system of characteristic equations for the other directions can be derived.
These equations simply show that along trajectories of dx = (u + c)dt, dx = udt, and
dx = (u  c)dt, the associated characteristic variables ws, are invariant. Equations
(50) through (53) are decoupled from each other, and are scalar equations. In this
form and with appropriate boundary conditions, it is the Reimann problem. The
purpose here is to demonstrate that all split flux schemes are based on characteristic
theory.
Since the decoupled system of equations in each spatial dimension is not necessarily easy to solve numerically and requires additional addressable memory for
characteristic variables in every spatial direction, a wide variation of flux vector splitting or the flux difference splitting methods were developed [39,43,44,46,48]. However,
a unique feature is shared among all flux vector splitting and flux difference splitting
(55)
to separate the positive and the negative eigenvalues. Since the coefficient matrix
A is related to the flux vector by the homogeneous of order one relationship to the
dependent variables, F = AU. The coefficient matrix A, can be split to account
for the correct orientation of signal propagation. For the flux contribution from the
characteristic which has a positive slope (positive eigenvalue), backward discretization
is used. Accordingly, forward differencing is used for the flux with a negative slope or
negative eigenvalue.
(56)
(57)
which is the sum of the total flux calculation F(U) = F+(U) + F(U), and is the
numerical equivalent of requiring only the physically correct information exchange
between the adjacent control volumes.
The flux vector split algorithm may be extended to secondorder by appropriate
extrapolation of information to the control surface. When the split fluxes are extrapolated, the method is called flux differencing. A better method is to use the MUSCL
approach [39]. In this procedure, the flow variables, either in conserved or primitive
form, are obtained on both sides of an interface by extrapolation. The extrapolation
process is then followed by the splitting of the fluxes.
8F..X
,
F+(U:+ 1 )
'2
F+(U__ 1 )
'2
+ F(U:+ 1 ) '2
b.x
F(U __ 1 )
'2
(58)
where,
U  U. ",Uj
j+t  ,+ '/'.
Uj 
(59)
(60)
The secondorder approximation of the flux splitting procedure requires some form of
a 'limiter' (<p) to prevent the occurence of under and overshoots of numerical results
immediately adjacent to discontinuous surface, and to reduce the solution accuracy
to first order.
Two of the more popular flux vector split approaches are originated by Steger et
al. [43] and Van Leer [46]. The former approach contains a nondifferentiable flux
at sonic and stagnation points, where the eigenvalues either vanish or change sign.
101
au al _
at + ax 
TV(u)
L: I~:I
00
(62)
dx;
00
For this initial value problem, the total variation of consecutive solutions shall never
increase: TV(u n + 1 ) < TV(u n ). Therefore, solutions of TVD schemes have the unique
feature of not containing spurious oscillations when applied to onedimensional nonlinear scalar hyperbolic conservation laws and constant coefficient hyperbolic systems [45,47). Since the applications of TVD schemes are mostly concentrated on the
shocks and contact discontinuities, weak solutions must be admitted. The piecewisesmooth weak solution of the conservation equation only satisfies the governing equation in two domains which are separated by the RankineHugoniot jump. However,
weak solutions of the conservation law are not uniquely determined by the associated
initial data [47,55). The physically relevant solution is ensured by implementing a
numerical flux function as an entropy inequality criterion. The flux function is sometimes referred to as the numerical viscosity term, but in the modifiedflux approach
Ut+l
+ .6..6.tx (pnV.
_ F,:,+;1
.) + .6.t
,)
>+2"')
>2"')
.6.y
(0':+11 >')+2"
G~+l1)
= ut).'
>')2"
(63)
Fi+~,j
Fi_},j =
Gi,j+~
Gi,j_~
(64)
where A and B are the matrices of right eigenvectors of the Jacobian matrices 8Fj8U
and 8Gj8U, respectively. The matrices A and B are evaluated using Roeaveraged
variables. <I> is a vector that contains the firstorder diffusion and secondorder correction terms [55]. A typical element of <I> is
<I>i = 'Ij; (Ci  Qi)
(65)
Ci is the local characteristic speed of the considered characteristic variable, and 'Ij;
provides the entropy correction to enforce the selection of the physical meaningful
solution. In general, TVD schemes always converge, and their limit is the physical
weak solution [47,55].
All TVD schemes [45,47,54,55] employing a flux limiter satisfy the TVD property
for the hyperbolic scalar conservation laws. This feature ensures that the integration procedure is monotonicity preserving. However, these methods are not globally
secondorder accurate [47,48]. The mechanism that eliminates spurious oscillations
in the solution also reduces the numerical accuracy locally to firstorder at extrema
103
and discontinuities. In any event, the TVD, flux vector split, and flux difference
split schemes have been very successfully applied to hypersonic problems. In solving the Euler equation, the improvement in shock wave definition over that of the
conventional methods is clearly demonstrated for both the single bow shock structure, Figure 4, and the complex shockonshock simulations [56]. Along the line of
symmetry, these higherorder resolution schemes uniformly capture the bow shock
waves between two adjacent cells which represent the best achievable by both split
flux and TVD schemes. For the more complex shockon shock problems, these numerical results exhibit equally impressive improvements. However, quantification of
these results is difficult to obtain, and thus is not included in here.
CURRENT METHODOLOGY
Rarefied Gas Regime
The Burnett equations are the natural extension of the NavierStokes equations for a
rarefied gas from the development of Boltzmann solutions [12,24,57,58]. Near translational equilibrium, the Burnett equations are a higher order approximation beyond the
ChapmanEnskog expansion [12,58]. Another formulation for rarefied gasdynamics,
the thirteenmoment equations, was derived by Grad [59] through a truncated expansion for distribution function identifiable with that of Chapman and Enskog [57].
These two systems of equations for rarefied gas are formally equivalent to a Maxwell
gas [12,57], and in a state of small departure from equilibrium, the thirteen moment
equations can be reduced to the Burnett equations [57]. Cheng et al. [24,57] have
successfully developed a shocklayer theory based on the thirteenmoment equations
for rarefied hypersonic flows and achieved comparable results to that of the DSMC
method for hypersonic flow over a flat plate at high angle of attack.
The basic approach of the gas kinetic theory is based on the premise that if
the nature of collision process were known, the motion of gas could be described
by the Boltzmann equation with known initial conditions via a general concept in
statistics [11,12J. Numerical solutions to a statistical problem almost always rely on
the Monte Carlo methods. Here, as usual, any method that employs random numbers
may be referred to as a Monte Carlo method. The DSMC is gradually replacing the
Molecular Dynamics method and the Test Particle method to become the main stay
for rarefied gas simulation. In the DSMC method, the intermolecular collisions are
treated on a probabilistic rather than a deterministic basis and are required to assume
molecular chaos of dilute gas flows. The result is equivalent to a numerical solution
of the Boltzmann equations [60].
The DSMC method is a computationally intense particle simulation technique,
thus has very close ties with the advent of advanced computer architectures. In the
current state of development, validity and capability of DSMC methods have been
demonstrated by comparison with experiments and known numerical solutions [24,57,
60,61]. Future applications will be strongly influenced by the progress of computing
technology.
Continuum Flow Domain
The NavierStokes equations have a limited extension into the Knudsen number tran
OU; = b; 
Lifj
a;joUj
a;i
(66)
where a;/s are the elements of the coefficient matrix, b;'s are the vector components
of the inhomogeneous terms. The difference between the Jacobi and GaussSeidel
method is that, unlike the former, the latter uses the most recently available value
of OU; to update solution without waiting for the complete set of new OU;. Applied
to elliptic differential equation, the GaussSeidel (Liebmann) scheme converges faster
than the Jacobi method (a special case of the Richardson scheme). The advantage
of the gain in numerical efficiency unfortunately also incurs an undesirable numerical
sweep bias [35,64]. Therefore, the Jacobi scheme is frequently adopted for the crossflow plane computation, and the GaussSeidel iteration is used in the streamwise
sweep to mimic the natural behavior of the flowfield development [64]. Convergence
of the solution is guaranteed if the coefficient matrix A can achieve the irreducible
form
A
[~~ ~3]
where Al is a square matrix, and the diagonal elements of matrix A must be dominant
over other entries.
(68)
(69)
(70)
Again the inviscid terms are linearized:
F~ + (~~): (U n+ un)
(71)
F~ +A~bU
(72)
(73)
F~
+ A~bU
(74)
Identical procedures are also applicable to the inviscid flux terms in two other spatial
orientations. The viscous terms may be similarly linearized:
Fn+
Fn+ bFn
v
(75)
Separating the inviscid and viscous terms, the above equations are expressible as
{I + ~t
~t
[! (A
+ A) +
ax + (ac)n
ay + (aH)n]
az
(76)
[axa (bF:;) + aya (bG~) + aza (bH:) ] = ~t [(aF)n
Note that the Right hand side represents the residual and may be evaluated by any
desired method. The treatment of the viscous terms requires some additional attention. First, the energy fluxes contributed by viscous effects are functions both of the
(77)
where, the coefficient matrix defines the relationship between the viscous flux and the
primitive variables [35,50,64]. The grouping of the conservative dependent variables
from the primitive variables is related by the following transformation Jacobian
.{ )~
p
pu
RV
pw
pe
1
u
v
w
0
P
0
0
pu
pv
p
pw
0
0
0
0
p
u
v
w
ei
)~ N'l )
p
u
v
w
(7S)
ei
The typical flux component contributed by the viscous terms now can be given as
8Fv =
Mxfj0 (N8U)
(79)
Xi
OXi
(SO)
It may be interesting to note that although the lefthandside of the implicit differencing of the viscous terms can be the thinlayer approximation to the NavierStokes
equations, all the viscous terms of the NavierStokes equations, as the driving force of
the implicit numerical formulation [35,50], must remain intact on the Righthandside
of equation (SO). The converged numerical solution to a steady asymptote will then
satisfy the exact macroscopic equation of motion.
According to MacCormack's algorithm [35,50] the time advancement of a numerical solution is given by the implicitly predicted change of the conservative variables
8U. This implicit increment is obtained by satisfying the explicit change of the NavierStokes equations,
!:lU.
{I+!:It [!..ox (A+ + A  Mx~N) +!..oy (B+ + B  M~N) +
!..ax (c+ + c  M~N)]} 8U = !:lU .
OXi
Ui~t1 = Ui~j,k
8Xi
+ 8Ui ,j,k
't,),
OXi
t,),
(S1)
(S2)
This linear algebraic system is factored in delta form and solved by combined GaussSeidel and Jacobi iterative sweeps [64]. In each spatial direction, the factored one
       _ .__.
107
Since the alternative backward and forward numerical sweeps are standard procedures
for a relaxation calculation, the detailed initiation and initial/boundary condition
implementations will not be repeated here [35,50,64].
In essence, MacCormack's explicit/implicit finite volume numerical method [35],
including a flux splitting option, has been demonstrated to be very effective for solving threedimensional, NavierStokes equations. At the present time, it is one of the
most popular algorithms for solving hypersonic flow problems. The spatially secondorder accurate results, and the alleviated numerical directional bias by a sweep lagging
process have been achieved with a minimum degradation of numerical efficiency. As
a matter of fact, nearly an order of magnitude of reduction of computational resource was realized in comparison with earlier numerical procedures [40,64]. More
importantly, comparable numerical results have been achieved by this method with
much lower numerical resolution requirement [42,64]. The same numerical procedure
is equally effective to simulate the high temperature hypersonic flow, in which the
added complications of chemical and thermal nonequilibrium phenomena have introduced additional equations of chemical physics into the system [32,42,52,62].
To summarize, for hypersonic flows in the continuum regime, numerical simulations require a solving procedure for equations of motion and chemical physics. From
statistical quantum mechanics, the atomic and molecular structure and its macroscopic behavior in high temperature in equilibrium state are well understood [11,14].
Transport properties of gas media and radiative heat transfer in limiting conditions
are also known [11,13,25]. However, in hypersonic applications, the chemical and thermodynamic equilibrium state is rarely attainable. The basic physical phenomena are
evolving around the energy transport by both elastic and inelastic collision processes,
and by the shifting of elementary particles to different energy levels through radiation.
All these mechanisms require a wide range of time and length scales to proceed to
an asymptote, if one exists. At present, the nonequilibrium problem is studied on
the framework of the concept of local equilibrium. Little assurance can be assessed
as to what extent the concept of a small departure from equilibrium is still valid for
numerical simulation. Therefore, the pacing items in hypersonic research are hinged
on the progress in chemical kinetics and the statistical quantum mechanics for the
nonequilibrium phenomena.
In the continuum regime, the governing equations of threedimensional hypersonic flows include five scalar components of the compressible NavierStokes for the
conservation of the mass, momentum, and the energy. Excluding the heterogeneous
ablating chemical reactions and the trace of noble gases, the dissociating and ionizing
Nt,
air may have eleven individual reacting species (N2 , O 2 , N, 0, NO, 0+, N+,
NO+, e+). In a state of thermal nonequilibrium, additional equations to describe energy exchange between internal degrees of freedom and a system of integrodifferential
equations for radiation are also required. In the rarefied gas domain, the N avierStokes
equations may be applied with modifications of the surface boundary conditions to
reflect the transition from dominance of molecular collisions to interaction with solid
ot,
References
[1] Tsien, H.S., Superaerodynamics, Mechanics of Rarefied Gases, J. Aero. Science,
Vol 13, 1946, pp 653664.
[2] Hays, W.D. and Probstein, R.F., HYPERSONIC FLOW THEORY, First ed.
Academic Press, New York, 1959.
[3] Van Dyke, M.D., Application of Hypersonic Smalldisturbance Theory, J. Aero.
Science, Vol. 21, 1954, pp 179186.
[4] Hays, W.D., On Hypersonic Similitude, Quart. Applied Math. Vol 5, 1947, pp
105106.
[5] Il'yusin, A.A., The Law of Plane Sections in the Aerodynamics of High Supersonic
Speeds, Prikl, Mat. Mekh. Vol. 20, 1956, pp 733755.
[6] Taylor, G.I., The Formation of a Blast Wave by a very Intense Explosion, Proc.
Roy. Soc. London. Ser. A., Vol. 201, 1950, pp 159186.
[7J Sedov, L.I., On Unsteady Motions of a Compressible Fluid, Dokl. Akad. Nauk
SSSR, Vol 47, 1945, pp 9193.
[8] Stewartson, K. and Williams, P.G., Selfinduced Separation, Proc. Roy. Soc.
Series A, Vol 312, 1969, pp 181206.
[9] Gokcen, T. MacCormack, R.W. and Chapman, D.R., Computational Fluid Dynamics Near the Continuum Limit", AIAA 871115CP, 1987.
[10] Cheng, H.K., Hypersonic ShockLayer theory of the Stagnation Region at Low
Reynolds Number, Proc. Heat Transfer and Fluid Mech. Inst. Stanford Univ.
Press, 1961, pp 161175.
[44] Roe, P.L., Approximate Riemann Solvers, Parameter Vectors, and Difference
Schemes, J. Compo Physics, Vol 43, 1981, pp 357372.
[45] Harten, A., High Resolution Schemes for Hyperbolic Conservation Laws, J.
Compo Physics, Vol. 49, 1983, pp 357393.
[46] Van Leer, B., Fluxvector Splitting for the Euler Equations, Lee. Notes Phys. Vol
170, 1982, pp 507512.
[47] Yee, H.C., Construction of Implicit and Explicit Symmetric TVD Schemes and
Their Applications, J. Compo Physics, Vol 68, 1987, pp 151179.
[48] Roe, P.L., Characteristicbased Schemes for the Euler Equations, Ann. Rev. Fluid
Mech. Vol 18, 1986, pp 337365.
[49] Lax, P., Hyperbolic Systems of Conservative Laws and the Mathematical Theory
of Shock Wave, SIAM Regional Series on Applied Math., Volll, 1973.
[50] MacCormack, RW. and Candler, G.V., The Solution of the NavierStokes Equations Using GaussSeidel Line Relaxation, Computers & Fluids, Vol. 17, No.1,
1989, p135150.
[51] Collela, P. and Glaz, H.M., Efficient Solution Algorithm for the Reimann Problem
for Real Gases, J. Compo Physics, Vol 59, 1985, pp 264289.
[52] Candler, G.V., The Computation of Weakly Ionized Hypersonic Flows in Thermochemical Nonequilibrium, Ph.D. Thesis, Stanford University, Calif. 1988.
[53] Grossman, B. and Cinnella, P., Fluxsplit Algorithms for Flows with Nonequilibrium Chemistry and Vibrational Relaxation, J. Compo Physics, Vol 88,
1990, pp 131168.
[54] Harten, A., On a Class of High Resolution TotalVariation Stable Finite Difference Schemes, Siam J. Num. Analy., Vol 21, 1984, pp 123.
[55] Yee, H.C. Warming, R.F., and Harten, A., Implicit Total Variation Diminishing
(TVD) Schemes for SteadyState Calculations, J. Compo Physics Vol 57, 1985,
pp 327360.
[56] Kroll, N., Gaitonde, D., and Aftosmis, M., A Systematic Comparative Study of
Several High Resolution Schemes for Complex Problems in High Speed Flows,
AIAA 910636, 1991.
[57] Cheng, H.K., Lee, C.J., Wong, KY, and Yang, H.T., Hypersonic Slip Flows and
Issues on Extending Continuum Model Beyond the NavierStokes Level, AIAA
891663, 1989.
[58] Burnett, D., The Distribution of Molecular Velocities and the Mean Motion in a
NonUniform Gas, Proc. Lond. Math. Soc., Ser 2, Vol 40, 1936, pp 382397.
[59] Grad, H., On the Kinetic Theory of Rarefied Gas, Comm. of Pure Appl. Math.,
Vol 2, No.4, 1949, pp 331407.
112
[60] Bird, G.A., Monte Carlo Simulation in an Engineering Context, Prog. In Astro.
and Aero.: Rarefied Gas Dynamics, Vol. 74, Part 1, Edited S. Fisher, 1981, pp
239255.
[61] Moss, J.N., Bird, G.A., and Dogra, V.K., Nonequilibrium Thermal Radiation for
an Aeroassist Flight Experimental Vehicle, AIAA 880081, 1988.
[62] Candler, G.V. and MacCormack, R.W., The computation of Hypersonic Ionized
flows in Chemical and Thermal Nonequilibrium, AIAA 880511, 1988.
[63] Newsome, R.W., Walters, R.W, and Thomas, J.L., An Efficient Iterative Strategy
for Upwind/Relaxation Solutions to the Thin Layer NavierStokes Equations,
AIAA 87113, 1987.
[64] McMaster, D.L., Shang, J.S. and Gaitonde, D., A Vectorized GaussSeidel Line
Relaxation Scheme for Solving 3D NavierStokes Equations, AIAA 891958CP,
1989.
MACROSCOPIC DESCRIPTION
(CONTINUUM MECHANICS)
MICROSCOPIC DESCRIPTION
(FREEIIOLECULE DYNAMICS)
CHAPIIANENSKOG EXPANSION
BURNETT EQUATIONS
THIRTEENMOMENT EQUATIONS
STEADY
RELATIVELY SIMPLE
CONAOURATION
LIMITED DOMAIN
01
CONSEQUENCE
INERTIA FORCE
DOMINANT
SMALL DISTURBANCE
Moo =12
AlT = 30,480 m
.... . . . . . . . DISSOCI.AT.I
.
...
NGGAS
Cv,tr
Cv,V
D
D.
eV
eV,s
l
g
h
h
hs
hv,s
H
I
kb,r
kj,r
lx, ly, lz
L/D
m x , my,
mz
ML
ML,INV
ML,vIS
ML,SRC
Ms
n
n
n x , ny, n z
n
Nr
Ns
p
q
Qrad
rjINv
rjvIs
drag coefficient
heat transfer coefficient
reference heat transfer coefficient
lift coefficient
specific heat for translationalrotational energy
specific heat for vibrationalelectronic energy
base diameter
effective diffusion coefficient for species s
energy per unit mass of species s
mixture vibrationalelectronic energy per unit mass
vibrationalelectronic energy per unit mass of species s
total energ~ per unit mass of mixture
flux vector in Cartesian space
inviscid component of flux vector relative to cell face
viscous component of flux vector relative to cell face
enthalpy or altitude
enthalpy per unit mass of species s
vibrationalelectronic enthalpy per unit mass of species s
total enthalpy per unit mass of mixture
identity matrix
backward reaction rate coefficient for reaction r
forward reaction rate coefficient for reaction r
unit vector tangent to computational cell wall
components of
in X,y,z directions, respectively
lift to drag ratio
unit vector tangent to computational cell wall
components of m in x,y,z directions, respectively
pointimplicit Jacobian of flux terms
pointimplicit Jacobian of inviscid terms
pointimplicit Jacobian of viscous terms
pointimplicit Jacobian of source terms
molecular weight of species s
iteration level index
unit vector normal to computational cell wall
components of n in x,y,z directions, respectively
number density
number of reactions
number of species
pressure
vector of conserved variables
radiative energy transfer rate
righthandside residual vector
relaxation factor used with inviscid Jacobian matrices
relaxation factor used with viscous Jacobian matrices
Tv
u,v,w
U, V, W
X,Y,Z
Ys
a
/3slr
ir
10, to
eN
'TJ
7]v
Oe
OR
A
~
A
/l
/lsj
~,'TJ, (
p
p.
(j
1>
X
w
Wv
Subscripts
e
i,j, k
I,J,K
1
L
r
s
V
00
electron
indices of cell walls in the ~, 'f/, ( directions
indices of cell centers in the ~,'f/, ( directions
dummy index for a cell wall
dummy index for a cell center
reaction r or species r
speCIes s
vibrationalelectronic
free stream
INTRODUCTION
The National Aeronautics and Space Administration's interest in viscous, hypersonic
flow field simulation has grown in recent years in anticipation of the design needs
for space transportation and exploration over the next three decades, e.g. Walberg
[1]. Aeroassisted space transfer vehicles will use the upper layers of planetary atmospheres in hypersonic aerobraking maneuvers. The National AeroSpace Plane,
with its supersonic combustion ramjet engines, is being designed to achieve hypersonicspeeds through the Earth's atmosphere in its climb to orbit. Various concepts
for a secondgeneration Space Shuttle are now being considered, as are concepts for
a space station crew emergency return vehicle. The external flowfield surrounding
such vehicles, as well as the internal flowfield through the scramjet engine, may be
significantly influenced by thermochemical nonequilibrium processes in the flow. Accurate simulations of these phenomena would provide designers valuable information
concerning the aerodynamic and aerothermodynamic character of these vehicles.
There are two major challenges to the simulation of flowfields in thermochemical
nonequilibrium around vehicles travelling at hypersonic velocities through the atmosphere. First, these simulations require modelling of the nonequilibrium processes in
the flow, frequently at energies where the models currently lack sufficient experimental or analytic validation. Nonequilibrium processes occur in a flow when the time
required for a process to accommodate itself to local conditions within some region
is of the same order as the transit time across the region. The equations and models
used in this chapter for nonequilibrium flow have been documented in Reference [2]
and were substantially derived from the work of Park [3, 4] and Lee [5]. The conservation equations are formulated within the context of the LAURA algorithm (Langley
Aerothermodynamic Upwind Relaxation Algorithm) [7]. Calibration and validation
of the physical models intrinsic to this code, which must deal with nonequilibrium
chemical and thermal processes, curve fits of thermodynamic and transport properties
at high temperatures, and radiative energy transfer, was first discussed in Reference
[8]. Other code development and calibration programs (e.g. Candler [9, 10], Park
and Yoon [11], Netterfield [12] Coquel et al. [13]) are now in progress within the area
of viscous, hypersonic, reacting gas flowfield simulations. Still, these calibration and
validation studies do not fully validate the flowfield simulations at velocities characteristic of vehicles returning to earth from the Moon or Mars. These simulations
should be used to help design groundbased and flight experiments for obtaining the
necessary data by performing parametric studies which identify critical aspects of the
physical models and locations in the flowfield which are most sensitive to unknowns
in these physical models.
Second, because of the large number of unknowns associated with chemical
species and energy modes and because of disparate time scales within the flowfield,
these simulations require algorithmic innovations to maintain numerical stability and
fully exploit supercomputer resources. Numerical stability is maintained through an
implicit treatment of the governing equations. A great variety of implicit treatments
is possible. For problems in which only the steady state solution is required, one is
free to evaluate any element of the difference stencil at any iteration (pseudo time)
level which facilitates the relaxation process. In the most rigorous implicit treatment,
all variables in all cells are simultaneously solved at an advanced iteration level, requiring the solution of a linearized equation set involving nI J J{ equations where n is
the number of unknowns at a point and I, J, and J{ are the number of computational
points in the three respective coordinate directions. The various forms of factored implicit schemes and line relaxation methods sequentially solve equation sets involving
nI, nJ, and/or nJ{ variables. The pointimplicit schemes, as utilized in the present
work, sequentially solve equation sets involving n simultaneous, linearized equations.
Further simplification is possible in chemical kinetic problems by linearizing contributions to the residual from only the source terms to alleviate problems of disparate
chemical time scales, resulting in methods which involve no matrix operations.
The essence of the pointimplicit strategy is to treat the variables at the cell center
of interest implicitly at the advanced iteration level and to use the latest available
data from neighbor cells in defining the "lefthandside" numerics. The success of this
approach is made possible by the robust stability characteristics of the underlying
upwind difference scheme. The basic algorithm requires only a single level of storage,
and numerical experiments show excellent stability characteristics, even when working
directly with the steady state equations (i.e., Newton iteration). Furthermore, the
algorithm is particularly amenable to generating solutions on unstructured grids [14]
and is efficiently implemented on vector or parallel processors [15]. Details of the
relaxation algorithm for the case of a gas in thermal and chemical nonequilibrium
are presented herein. Sample applications related to hypersonic flows over aero brakes
follow.
FINITEVOLUME FUNDAMENTALS
The integral form of the conservation laws applied to a single cell in the computational
domain is written
JJJ~7
dD
JJf .
JJJw
ii da =
dD
(1)
In Equation 1 the first term describes the time rate of change of conserved quantity
q in the control volume; the second term describes convective and dissipative flux f
through the cell walls; and the third term accounts for sources or sinks of conserved
quantities within the control volume. The third term is identically zero for perfectgas
flows but is required for flows in chemical or thermal nonequilibrium.
The finitevolume approximation to Equation 1 for a general, unstructured grid
is written
I:
[ ~]
Of L +
IIh
""'
L.J fm
 m
n",a
= [.W H0] L
(2)
m=l
nm
oq D
[8t h,J,l(
+ [ ;+1
ni+1 a i+1
~+1
nj+I a j+1
+ [ fk+I
nk+1 a k+1
iJ
fk
niaj ]J,K
(3)
na']IY
J J
, \
i"hak
]l,J = [w D]l,J,K
A shorthand notation for Equation 3 that will be used throughout this paper follows:
; .
11/a/]
= [w D]L
(4)
/=i,j,k
Note in Equations 24 that uppercase integer variables I, J, J{, and L denote computational coordinates at cell centers, and lowercase integer variables i, j, I.:, t, and m
denote cell faces or cell corners. For example, ai,J,h" refers to the cell wall corresponding to indices 11/2, J, J{. (See Figure 1.) In the shorthand notation of Equation 4,
the integer variable I is used as a generic index for i, j, or k. This notation is convenient because most of the formulations for quantities at cell faces are independent
of the coordinate direction. The geometric quantities D, a, and 11 are easily derived
given the Cartesian coordinates of the cell corners. Details may be found in Appendix
A of Reference [6J.
The formulations that follow are based on a rectangularly ordered, structured
grid. A firstorderaccurate formulation of the inviscid equations on a structured grid
MICROSCOPIC DESCRIPTION
(FREEIIOLECULE DYNAMICS)
.....CROSCOPIC DESCRIPTION
(CONTINUUM MECHANICS)
CHAPIIANENSKOG EXPANSION
BURNETT EQUATIONS
THIRTEENIIOIIENT EQUATIONS
INTERACTIVE BOUNDARYLAYER THEORY
THINLAYER N.S. EQUATIONS
STEADY
RELA TlVEL Y SIMPLE
CONAOURATION
LIMITED DOMAIN
01
CONSEQUENCE
INERTIA FORCE
DOMINANT
SMALL DISTURBANCE
= 12
AlT = 30,480 m
Mm
.........
_..
DISSOCIATING GAS
o.g
D
0.8
EXPERIMENTAL DATA
 NUMERICAL RESULT
0.7
0.6
0.5
0.4
~
03
02
0.1
0.0
0.1
02
0.3
0.6
0.4
0.2
0.0
0.2
0.4
0.6
1 ,
2
Moo = 16.33
........
~HH~_,45.0
30.0
37.5
ci+~~~~~~~~~
0.0
7.5
15.0
22.5
Grid Pointa
Figure 4: Mach number distribution along the line of symmetry of a blunt body [56].
Chapter 4:
PointImplicit Relaxation Strategies for Viscous,
Hypersonic Flows
P. A. Gnoffo
Aerothermodynamics Branch, NASA Langley Research
Center, Hampton, Virginia 23665, USA
ABSTRACT
An upwindbiased, pointimplicit relaxation algorithm for obtaining the numerical
solution to the governing equations for threedimensional, viscous, hypersonic flows
in chemical and thermal nonequilibrium is described. The algorithm is derived using
a finitevolume formulation in which the inviscid components of flux across cell walls
are described with a modified Roe's averaging and Harten's entropy fix with secondorder corrections based on Vee's Symmetric Total Variation Diminishing scheme.
Newton relaxation of the fully coupled equation set is employed on a cell to cell basis. Underrelaxation of the inviscid and overrelaxation of the viscous contributions
to the residual are implemented. Computational work is easily partitioned among
many processors in an asynchronous, dynamic mode for convergence acceleration.
An overview of the physical models employed herein for thermochemical none quilibrium is included. Several test cases and comparisons with experimental data are
presented involving hypersonic flow over blunt bodies which illustrate the qualitative
and quantitative capabilities of this approach.
NOMENCLATURE
Symbols
Boldface, lowercase symbols refer to vectors in parameter space. Boldface, uppercase
symbols refer to matrices in parameter space. An arrow over a lowercase symbol
refers to vectors in physical space, with x, y, and z coordinates.
a
A
B
Cv,tr
Gv,v
D.
es
eV
eV,s
f
g
h
h
hs
hv,s
H
I
kb,r
kj,r
m x , my, m z
ML
ML,INV
ML,vIS
ML,SRC
Ms
n
ii
n x , ny, n z
n
Nr
Ns
p
q
Qrad
rjINV
rjvIs
drag coefficient
heat transfer coefficient
reference heat transfer coefficient
lift coefficient
specific heat for translationalrotational energy
specific heat for vibrationalelectronic energy
base diameter
effective diffusion coefficient for species s
energy per unit mass of species s
mixture vibrationalelectronic energy per unit mass
vibrationalelectronic energy per unit mass of species s
total energ~ per unit mass of mixture
flux vector in Cartesian space
inviscid component of flux vector relative to cell face
viscous component of flux vector relative to cell face
enthalpy or altitude
enthalpy per unit mass of species s
vibrationalelectronic enthalpy per unit mass of species s
total enthalpy per unit mass of mixture
identity matrix
backward reaction rate coefficient for reaction T'
forward reaction rate coefficient for reaction r
unit vector tangent to computational cell wall
in X,y,z directions, respectively
components of
lift to drag ratio
unit vector tangent to computational cell wall
components of m in x,y,z directions, respectively
pointimplicit Jacobian of flux terms
pointimplicit Jacobian of inviscid terms
pointimplicit Jacobian of viscous terms
pointimplicit Jacobian of source terms
molecular weight of species s
iteration level index
unit vector normal to computational cell wall
components of ii in x,y,z directions, respectively
number density
number of reactions
number of species
pressure
vector of conserved variables
radiative energy transfer rate
righthandside residual vector
relaxation factor used with inviscid Jacobian matrices
relaxation factor used with viscous Jacobian matrices
R
Rb,r
R"r
RN
Rs
R
s
t
T
Tv
u,v,w
U, V, W
X,Y,Z
Ys
a
8p/8pE
stoichiometric coefficient for products in reaction r
8p/8Pr
eN
Ti
Tiv
()
(}e
(}R
A
~
A
/l
/lsi
~, Ti, (
P
Ps
a
T
8p/8p ev
w
Wv
Subscripts
e
i,j, k
I,J,/{
I
L
r
V
00
electron
indices of cell walls in the ~, TJ, ( directions
indices of cell centers in the ~,TJ, ( directions
dummy index for a cell wall
dummy index for a cell center
reaction r or species r
speCles s
vibrationalelectronic
free stream
INTRODUCTION
The National Aeronautics and Space Administration's interest in viscous, hypersonic
flowfield simulation has grown in recent years in anticipation of the design needs
for space transportation and exploration over the next three decades, e.g. Walberg
[1]. Aeroassisted space transfer vehicles will use the upper layers of planetary atmospheres in hypersonic aerobraking maneuvers. The National AeroSpace Plane,
with its supersonic combustion ramjet engines, is being designed to achieve hypersonic speeds through the Earth's atmosphere in its climb to orbit. Various concepts
for a secondgeneration Space Shuttle are now being considered, as are concepts for
a space station crew emergency return vehicle. The external flowfield surrounding
such vehicles, as well as the internal flowfield through the scramjet engine, may be
significantly influenced by thermochemical nonequilibrium processes in the flow. Accurate simulations of these phenomena would provide designers valuable information
concerning the aerodynamic and aerothermodynamic character of these vehicles.
There are two major challenges to the simulation of flowfields in thermochemical
nonequilibrium around vehicles travelling at hypersonic velocities through the atmosphere. First, these simulations require modelling of the nonequilibrium processes in
the flow, frequently at energies where the models currently lack sufficient experimental or analytic validation. Nonequilibrium processes occur in a flow when the time
required for a process to accommodate itself to local conditions within some region
is of the same order as the transit time across the region. The equations and models
used in this chapter for nonequilibrium flow have been documented in Reference [2]
and were substantially derived from the work of Park [3,4] and Lee [5]. The conservation equations are formulated within the context of the LAURA algorithm (Langley
Aerothermodynamic Upwind Relaxation Algorithm) [7]. Calibration and validation
of the physical models intrinsic to this code, which must deal with nonequilibrium
FINITEVOLUME FUNDAMENTALS
The integral form of the conservation laws applied to a single cell in the computational
domain is written
(1)
In Equation 1 the first term describes the time rate of change of conserved quantity
q in the control volume; the second term describes convective and dissipative flux
through the cell walls; and the third term accounts for sources or sinks of conserved
quantities within the control volume. The third term is identically zero for perfectgas
flows but is required for flows in chemical or thermal nonequilibrium.
The finitevolume approximation to Equation 1 for a general, unstructured grid
is written
[ ~]
{)t L +
IIh
~f~
L...J m
n",(J"m
[.W H0] L
(2)
m=l
where {)q = qn+I  qn, {)t = t n+ 1  tn, Jlh is the number of faces of cell L
having volume n, and subscript m refers to cell face m with surface area (J"m' The
quantity nm is a unit vector normal to cell face m in a direction facing away from
the cell center. The dependent variable q is defined at cell centers. The independent
variables x, y, and z are defined at cell corners.
The finitevolume approximation to Equation 1 for a rectangularly ordered, structured grid is written
{)q n
[;5t ]J,J,J(
+ [ ;+1
+ [ ~+I
+ [ k+I
ni+1 (J"i+1
n'(J"']J
r
"'
t
, \
nj+I(J"j+I
nk+I (J"k+1
n(J"'
J J
nk(J"k
]1 ,\Y
(3)
]I,J = [w nh,J,I(
A shorthand notation for Equation 3 that will be used throughout this paper follows:
2:
[it+1
ii/+I (J"/+1
it . ii/(J"/]
[w n]L
(4)
/=i,j,k
Note in Equations 24 that uppercase integer variables I, J, f{, and L denote computational coordinates at cell centers, and lowercase integer variables i, j, k, 1, and m
denote cell faces or cell corners. For example, (J"i,J,I\ refers to the cell wall corresponding to indices 11/2, J, f{. (See Figure 1.) In the shorthand notation of Equation 4,
the integer variable I is used as a generic index for i, j, or k. This notation is convenient because most of the formulations for quantities at cell faces are independent
of the coordinate direction. The geometric quantities n, (J", and ii are easily derived
given the Cartesian coordinates of the cell corners. Details may be found in Appendix
A of Reference [6].
The formulations that follow are based on a rectangularly ordered, structured
grid. A firstorderaccurate formulation of the inviscid equations on a structured grid
j=3
(I,j) or (i,J)
1
1
1
1
1
1
cell corner
.9~I~fM1te[. _
1
j=2
1
1
1
1
1
__
cell wall
1
1
1
____ ..l ____
1
1
1
    1 ___
1
I
I
I
I
I
I
I
ceiliwall
I
I
I
I
1
1
I
I
I
I
rI
1
1
121
~alL_~
____


 
1
1
1
1
1
1
j=1
Figure 1: Cell indexing system with cell corners defined by lower case letters and cell
centers defined by upper case letters.
is identical to the formulation on an unstructured grid. The modifications required
to achieve secondorder accuracy on an unstructured grid are not addressed in this
paper. However, it should be noted that the formulations for obtaining secondorder
accuracy only involve modifications to the righthandside residual vector. The pointimplicit relaxation procedure that will be defined by the formulation of the lefthandside matrix is independent of grid structure. Consequently, much of the development
which follows will carryover to unstructured grid formulations, as in the paper by
Thareja et al. [14].
CONSERVATION EQUATIONS
The inviscid, viscous, and source term contributions to the complete conservation
laws are considered separately for convenience. Let
(5)
where gl defines the inviscid terms and hi defines the viscous terms. The finitevolume
formulation of the conservation laws is now expressed as
L
I=i,j,k
[hl+ 1 0"1+1
h[O"l]
(6)
Ps
pu
pv
pw
pE
pev
(7)
PsU
pUu + pn,.
pUv + pny
pUw + pnz
pUH
pUev
(8)
 pDs OSn
~Y'
h

UTnx
VTny
WTnz
Tl
!tIY..
'IVas n
Tnx
Tny
Tnz
aT
!tIY..
"I aS n
"Iv aS n
I D ~
P ",N,
L....s=l ~ V,s s aS n
(9)
I D ~
P ",N,
L....s=l ~s s aS n
Ws
0
0
0
(10)
Qrad
Wv
The first element of the vectors defined in Equations 710 describes the species
conservation; the next three elements describe :1:, y, and z momentum conservation;
the fourth element describes total energy conservation; and the fifth element describes
vibrationalelectronic energy conservation. The present model considers Ns = 11
species. Consequently, the vectors defined in Equations 710 are composed of a total
of 16 elements. Species 1 through 5 are the neutral components of air consisting of
N, 0, N 2 , O 2 , and NO. Species 6 through 10 are the ions corresponding to species 1
through 5, in which one electron has been removed. Species 11 are the free electrons.
Implicit in the use of a vibrationalelectronic energy equation is the assumption that
the partition of energy in the vibrational, bound electronic, and free electron modes
among all species can be described by a single temperature, Tv. This approximation is based on rapid equilibration of vibra.tional and electronic energy and electron
translational modes [4, 16]. The translational and rotational energy modes of heavy
particles are assumed to be fully excitecl and described by temperature T.
The thermochemical nonequilibrium model is described in detail in Reference
[2]. Some specifics on their formulation is given in a later section but a brief overview
123
is given below. The reactive source terms for the species conservation equations
are denoted by
The radiative energy transport term Qrad may be treated as
a source term in the energy equation though its effects are not considered in the
present analysis. Finally, the vibrationalelectronic energy source term Wv accounts
for the mechanisms by which vibrationalelectronic energy is lost or gained due to
collisions among particles in the cell. These mechanisms include the energy exchange
(relaxation) between vibrational and translational modes due to collisions within the
cell; the vibrational energy lost or gained due to molecular depletion (dissociation) or
production (recombination) in the cell; the electronictranslational energy exchange
due to elastic collisions between electrons and heavy particles; the energy loss due to
electron impact ionization; the rate of energy loss due to radiation caused by electronic transitions; and a term related to the work done on electrons by an electric field
induced by the electron pressure gradient minus the flow work due to electron pressure. The electron pressure flow work is normally considered as part of the electronic
enthalpy in the inviscid (convective) portion of the flux balance. Moving the electron
pressure from the convective term to the source term simplifies the expressions for
eigenvalues and eigenvectors of the Jacobian of the inviscid flux vector.
wS'
(11 )
where gL,1 = [fL . ndlNv.
The first term in braces is a secondorderaccurate base approximation for gl.
The second term in braces provides the upwindbiased numerical dissipation. It is
a firstorder dissipation when 0 equals zero. It is a secondorder dissipation when ()
equals one.
The vector SI is defined
(12)
The matrix Ri 1 in Equation 12 and the matrices R/ and Al in Equation 11 are
related to the Jacobian of the inviscid flux vector g with respect to q in the following
manner:
A =
= R A R 1
(13)
og
oq
= {
(14)
The eigenvalue limiter ( was first used by Harten [17] to prevent the formation of
expansion shocks across a sonic line where one eigenvalue equals zero. Its application
is critical in blunt body flows to prevent instabilities, often in the form of reversed
flows at the stagnation point, which occur in the stagnation region where the repeated
eigenvalue U is near zero. The magnitude of (, which is nondimensionalized by
the freestream velocity, is problem dependent. Vee [18] has suggested a functional
dependence of ( on the local values of sound speed and velocity. This relation has
been adapted for use in the present work as follows:
(15)
where (0 is a constant which generally varies from 0.01 to 0.4. Larger values of (0
are required for flows with extensive stagnation regions, as in the case of blunt body
flows.
The term (Xn)1 is a shorthand notation for Vx . f'i l and may be thought of as
the inverse of the projected distance between cell centers Land L  1 in a direction
normal to cell face t. It is defined in Equations 7981 in the Appendix. The variable
X is a generic computational coordinate running in the direction of increasing generic
index L.
The limiters used in the present work were introduced by Vee [19, 20]. They
involve symmetric functions of gradients in the neighborhood of the cell face, and
algorithms based on these limiters are refcITcd to as Symmetric Total Variation Diminishing (STVD) schemes. STVD schemes involve little extra programming work
over simple firstorder algorithms because most of the quantities required in their
implementation are already available. These antidiffusive corrections are defined as
(16)
where the min mod function returns the argument of smallest absolute magnitude
when all the arguments are of the same sign or returns 0 if the arguments are of
opposite sign. Equation 16 is evaluated element by clement of vector s.
The STVD limiter does not yield a strictly upwind biasing on the formulation
of the flux vector. Also, the scheme reduces to firstorder at cell faces where there is
a sign change in the arguments of the min mod function. The present formulation
differs from that of Reference [20J in that the differences Ri 1 (qL  qLd have been
scaled by (Xn k This scaling, which is eventually removed by the leading factor of
the numerical dissipation term in braces, reduces the effects of grid stretching on the
argument returned by the min mod function. The present treatment has been found
to be well suited to the range of numerical test problems considered herein.
125
where
lAd
RI
lAd
I=i,j,k
[(AL,I+l
+ IA1+11)al+l
I=i,j,k
I=i,j,k
I=i,j,k
l==i,j,k
l=i,J,k
where
(20)
Tnn
Tnl
Tnm
au
Al ( au + av + aw) + 2J..LIaS n
aS n aS I aS m
( aS
av +
III
n ~~
aw+au)
J..LI ( aS
aSm
(21 )
(22)
(23)
Sn, S[,
J. L
Sm
A
1n
(24)
Substitution of Equations 2123 into Equation 24, collecting terms, and simplifying (see Reference [6]) yields the following relation for shear stress in the S direction:
=
as
as
as
au a~ au a1] au ac)
a1]1]n + ac Cn + a~ as + a1] as + ac as
J..LI(a~~n +
w . '\l~
~
w ~
w ~
AI(+  . '\l1] +  . '\lOn.
a~
a1]
ac
(25)
Tns = IlI(
as 1 au
a
+ a
nshn
X
3 X
(26)
where n refers to the direction normal to a constant X surface, and the prime superscript refers to the thinlayer approximation. The use of Stokes relation,
=
and geometric identities have also been used in the simplification of Equation 25. The
viscous terms on the other two coordinate surfaces are also neglected in the thinlayer
approximation because their contribution to overall momentum and energy balance
is small. These approximations are valid so long as the boundary layer is relatively
thin and the X direction is approximately normal to the high gradient region.
Mass diffusion and energy conduction contributions to the viscous terms are
functions of gradients normal to the cell face. For example, the gradient of T in the
A 2J..L/3,
127
aT
,
a~ ",n
i'i
aT
a1] 1]11
aT
rn
(27)
a('
aS
aT
x
ox ~ n
(28)
Derivatives in the X direction are evaluated to secondorder accuracy in computational space as follows:
au)*
( aX
I,L
UL
+ bttL 
au)*
*
n+l _
*
( OX
1+1,L = UL+1  UL
 UL+ 1
Ur._l
ttL 
_
ttL 
(au)
OX 1+1
b
UL
*
41 ( UI,J+l,K

*
UI,Jl,/\
*
+ U!+l,J+1,K

*
)
U/+ 1 ,Jl,K
The derivatives in the directions along the face (i.e., those derivatives neglected in the
thinlayer approximation) have no functional dependence on the cell center. Therefore
the pointimplicit treatment of the full NavierStokes equations is identical to the
thinlayer N avierStokes equations.
Now, define hi as a function of differences evaluated using currently available
data, for example (~~)I' and define hi,L as a function of differences using predicted
values at cell center L, for example (aaU)i
x ,L' These definitions permit the linearization
of the viscous terms to be expressed as follows:
hi,L
hi 
hi+l,L
h l +1
BI,L bqL
B I+ 1 ,L bqL
BI,L
_ ahi,L
aqL
_ Ol;i,L
B I+ 1 ,L
ahi+l,L
OqL
Ol;i+l,L
OqL
(29)
where
aqL
(30)
The pointimplicit implementation of the viscous terms follows the example set
in the previous section on the inviscid terms.
I=i,j,k
I=i,j,k
[B/+ 1,LO"I+1
+ Bl,LO"d
(32)
l=i,j,k
In the case of the thinlayer NavierStokes equations, the summation would only include one of the i, j, or k directions, depending on the orientation of the computational
coordinates with the body.
Ws = Ms L((3s,r  os,r)[Rf,r 
(33)
Rb,r]
,'=1
where N r is the number of reactions; Os,r and (3s,, are the stoichiometric coefficients
for reactants and products in the T' reaction, respectively; and Rf,r and Rb,r are the
forward and backward reaction rates for the T' reaction, respectively. These rates are
defined by
Rf,T
Rb,T
(34)
[ kf,T g(Ps/Ms)"s,r ]
=
[kb,r
(Ps / Ms ){3"r ]
where kf,T and kb,T are the forward and backward reaction rate coefficients, respectively, defined in [2], and Ns is the number of chemical species.
The reaction rate coefficients are explicit functions of T and Tv. Consequently,
the Jacobian of Ws with respect to q can be explicitly evaluated as follows:
(35)
where the differential relations between T and q and between Tv and q can be
expressed
N,
L(e s 
eV,s)dps
s=1
udpu
vdpv 
wdpw
dpE  dpev
(36)
N,
pCv,vdTv
dpev
L eV,sdps
s=1
(37)
129
Wv
(Ri,r  Rb,r)t }
r:elec . mp.
L ps (ey,s  ev,s) +
{ s:mol.
< Ts >
{Qrad  Pe 9
3peR(T  Tv)
L ~~ }
ste
(38)
.u }
The first set, the reactive source terms in the first pair of braces of Equation 38,
are composed of terms that are proportional to either Ws or to (Ri,r  Rb,r)' In the
first case, the proportionality factor, D., represents the average vibrational energy per
unit mass created or destroyed through recombination or dissociation of molecules.
In the simplest approximation, it is set equal to the average vibrationalelectronic
energy, ev, though more comprehensive treatments which model preferential dissociation of vibrationally excited molecules may be employed. In the second case, the
proportionality factor, r , represents the average translational energy per mole lost
by a free electron in freeing another electron from a neutral heavy particle in reaction r through the process of electron impact ionization. It is approximated by the
ionization energy from an excited state of the target particle. Further detail on these
points is available in Reference [2] and the book by Park [16J. The pointimplicit formulation of these terms treats the proportionality factor explicitly and the reaction
rates implicitly according to Equations 35:37.
The second set, the relaxation terms in the second pair of braces in Equation 38,
model the energy exchange between heavy particle translationalrotational modes
and vibrationalelectronic and electron translational modes. The first term in these
braces, which models the exchange between vibrational and heavy particle translational modes, can be approximated by
s:mol.
Ps
(ey,s  e~T,s)
< Ts >
T  Tv
pCv,v TV
(39)
~."
nJexp
Tsp =
< Ts >
(  ~)1
f7 sCsn
= Tt1W
18 . 4?~ ]
(41 )
(42)
T;
( 43)
Values of As in Equation 41 are 220 for N 2 , 129 for O 2 and 168 for NO and p is in units
of atmospheres. The twotemperature model should also have a corresponding term
relating the energy exchange of translational and electronic energy. This transfer has
not yet been formally included in the present work; however, the driving potential
is already based on both the vibrational and electronic energies and the relaxation
times are expected to be on the order of < Ts > so that the net effect on the present
model should be small. The second term in these braces in Equation 38 models the
direct exchange of translational energy between electrons and heavy particles. This
exchange rate is generally much slower than the previous term. Both terms in this
set are now proportional to the difference between the translational and vibrational
temperatures, T  Tv. Here again, the pointimplicit formulation of these terms
treats the proportionality factor explicitly and the driving potential T  Tv implicitly
according to Equations 3637.
The third set, the fielddependent terms in the third pair of braces of Equation 38,
are functions of properties at the cell center and at neighboring cells. (The first
two sets are functions only of properties at the cell center.) These terms include
radiative energy transport, work done by the electric field on electrons and electron
pressure flow work, combined into a single term. Radiative energy transport is treated
explicitly as described before. The other contribution to the field dependent terms
is also treated explicitly. In fact, in the cases tested to date with maximum electron
number densities approximately 4 percent of the total number density, omission of
this term has little effect on the flowfield.
131
OWL
ML,SRC
ML,SRC
= OqL
(45)
RELAXATION ALGORITHM
The governing relaxation equation is obtained by combining the results of Equations 6,20,32, and 45 and taking the limit as time step 8t goes to infinity. Thus,
(46)
where
ML
= 
[(gIH
+ hl+dal+l  (gl +
hl)ad
wLfk
(48)
I=i,j,k
Relaxation factors are used to control stability and convergence. Numerical tests in
Reference [15] indicate that underrelaxation is appropriate for the inviscid contribution to the residual, with r flNV > 1.5. Overrelaxation is appropriate for the viscous
contribution to the residual with I'fv IS > 0.5 provided relaxation sweeps are across
the boundary layer; otherwise, r fv IS ~ 1. The lower limits yield the fastest convergence rates but may lead to instabilities if the solution is far from convergence or if the
pointimplicit Jacobian is "frozen", as discussed below, for too long. It is sometimes
necessary to chose r flNV ~ 3 and r fv IS ~ 2 to get past some difficult transients
in the early stages of the relaxation process that defy linear analysis. \Vhen these
transients pass it is then advisable to switch to the lower limits of these parameters
to get the best convergence rate. Convergence may eventually sta.ll at some point due
to limit cycles associated with the min mod function in Equation 16. This stalling
may be alleviated by again increasing the relaxation factors.
The solution vector rL and the Jacobian ML are evaluated using the latest available data. Consequently, the algorithm requires only a single level of storage. One
can solve for 8qL using Gauss elimination. Numerical experiments have shown that
pivoting is not required, and so the algorithm is easily vectorized. However, it is more
efficient to calculate and save Mr: 1 for large blocks of iterations (typically 20) and
solve for 8qL directly using
( 49)
0.0
1.0
t:,.
task 2
task 3
task 4
task 5
task 6
2.0
133
109,o(Error)
3.0
        _________s~n9Ie task
4.0
5.0
6.0~~~~~~~~~~~~~~~~~~~~~~J
100
300
200
400
500
600
CPU time, s
Figure 2: Convergence histories for singletask and sixtask, adaptive partitioned algorithms applied to problem of nonequilibrium, hypersonic flow over a blunt, axisymmetric body.
cessor Cray YMP in Reference [15]. In these tests, the flow domains were subdivided
into partitions with a single task assigned to each partition. Partition boundaries are
dynamically adjusted to concentrate relaxation sweeps in the regions that are slowest
to converge. Because no synchronization is required, all tasks (processors) may execute throughout the computation without interruption. A comparison of convergence
histories for the solution of hypersonic flow in thermochemical nonequilibrium over an
axisymmetric body is shown for a single task and a sixtask, adaptive partition test
in Figure 2. The symbols show the error norm for each individual task of the sixtask
run. The solid line shows the total error norm for the sixtask run and the dotted
line shows the error norm for the single task run. Adaptive partitioning has allowed
the sixtask case to converge to a lower error norm than the single task case for the
same amount of CPU time. Furthermore, the actual elapsed time for the sixtask case
would be a factor of six smaller than for the single task case on a dedicated machine.
Figure 3: Mach number contours in the plane of symmetry of the Aeroassist Flight
Experiment (AFE) model at Mach 10.
Mach number contours over the AFE model including the sting for Mach 10
flow and a = _5 are shown in Figure 3. The Reynolds number for this case is
159000 based on a model diameter of 9.322 em and laminar flow is assumed. Grid
adaption is used over the forebody to align the grid with the captured bow shock.
The contours clearly illustrate the sharp, captured bow shock, the captured shock on
the wind side of the cylindrical sting, and the freeshear layers emanating behind the
circular shoulder of the aerobrake.
Comparisons with experimental data of Micol [26] for pressure and heat transfer
are shown in Figure 4 on the forebody. Differences between numerical predictions
and experiments are generally very small and within experimental accuracy, except
for the heat transfer in the stagnation region. These differences are believed to be
caused by both the varying truncation error behavior as the coordinate singularity
Experiment} C
 LAURA
P
2.0
nOn
QY~O
1.6
C
'Q....Q
0 Experiment} CH
  LAURA
CH, FR
f19<l~ ~
\~
CH,FR
.8
.4
o~~~~~~
.3
.3
.6
.9
s/L
Figure 4: Comparisons between prediction and experimental data for pressure and
heating over the AFE at Mach 10 and a = 0.
o
o
.8
q
q ref
88;
 O}E
.
180
xpenment
LAURA
.6
.4
.2
0
.5
1.0
1.5
s/L
Figure 5: Comparison between prediction (laminar) and experimental data for heating
on the sting supporting the AFE model at Mach 10 and a = 0.
F =
t:.pL(l +
4160'\
CN)
(51)
where ,\ is the wavelength and L is the geometrical path, taken here as the cylinder
LAU RA, Park Model 1
M 00
Experiment, Hornung
=6.14
=12000
Re oo
T00 =1833K
Figure 6: Calculated and experimental fringe patterns over cylinder in partially dissociated nitrogen obtained in shock tunnel using chemical kinetic model of Park.
length equal to 0.1524 m (6 in). No attempt was made to account for changes in
the geometric path through the shock envelope due to threedimensional effects at
the edge of the cylinder. There is reasonahly good agreement between the calculated
137
and experimental fringe patterns in Figure 6, in which the chemical kinetic model is
due to Park as described in Reference [2]. A revision of Park's kinetic model was also
tested for this case in [8]. Some shifting in the fringe pattern was observed in the
calculation but no conclusion could be drawn as to which model best fit the data.
The calculated shock shape is in excellent agreement with the experimental data.
A thermal equilibrium, chemical nonequilibrium case was run by reducing the characteristic time for vibrational relaxation by a factor of 105 . The change in shock
standoff distance at the stagnation streamline was less than 5% as compared to the
thermochemical nonequilibrium case.
FIRE II Stagnation Point Heating  Flight Data
FIRE II flight data [30] consist of measured total heating rates (convective plus absorbed radiative), radiative intensity in the 0.2 to 6.2 eV range, and spectral radiation
in the 2.0 to 4.0 eV range. Only the stagnation point measurements are examined
in the present set of results. The vehicle was constructed with a layered heatshield,
consisting of three beryllium layers. The vehicle had a truncated, hemispherical shape
with a small corner radius. The calculations were performed at three points during
the early data period on the first heatshield, which is characterized by low radiation and significant thermochemical nonequilibrium. A fully catalytic wall boundary
condition is assumed. An equivalent sphere geometry was defined as in References
[31,32,33] with a nose radius of 0.747 m. The freestream velocity equals 11.3 km/s,
and the altitude range covered by this early data period was between 84.6 and 67.05
km.
.
01.0
Measured total
Measured total minus measured radiative
LAURA, Park model
.8
.6
q ref
.4
.2
0
1630
Time, sec
Parameters
Rs, m Oc
0.3861 60.
0.3048 70.
0.6096 70.
0.3048 70.
OR
eN
73.
90.
90.
90.
2.0
l.0
l.0
l.0

~
Table III
case
afe1
afe1
afe4
afe4
afe4
lunara
lunara
lunarb
lunarc
 Aerodynamic Parameters
Q'
CD LID
CL
17. 0.366 1.37 0.267
22. 0.443 1.27 0.349
12. 0.302 1.43 0.211
17. 0.376 1.33 0.283
22. 0.429 1.22 0.352
O. 0.000 1.65 0.000
10. 0.232 1.53 0.152
10. 0.206 1.45 0.142
10. 0.239 1.58 0.151
Noslip and finiterate wall catalytic boundary conditions [35] are used throughout. Aerobrake surface temperatures are held constant at 1750 K. Base surface
temperatures are held constant at 1500. Radiative energy transport is not included
in the computations. (Radiative heating accounts for about 5% of the total, maximum
heating for case "afe4" and about 80% of the total heating for case "lunarc".)
A new approach has been taken in the definition of the lunar aerobrake that
overcomes the problems associated with the axis singularity in the stagnation region.
This singularity may induce erratic behavior in the calculated heat transfer distribution. These problems were specifically addressed in [27] where it was noted that a
localized grid restructuring does much to overcome the problem. This concept was
extended to create a more uniform, singularity free surface mesh as shown in Figure 8.
A grid generation package for general blunt bodies with circular shoulders derived
from circular or elliptic cones has been coded to simplify the definition of singularity
free grids. As an added benefit, the new grid structure provides enhanced circumferential distribution of cells which feed directly into the wake. All of the lunar aerobrake
flows were generated with this new grid structure. Grid lines are constructed normal to the body. An algebraic stretching function is used to control distribution
of points across the shock layer. Mesh spacing at the wall is defined to enforce a
cell Reynolds number pallsl fL equal one. An adaptive grid routine aligns the outer
boundary with the bow shock so that the bow shock stands at 80% distance between
the wall boundary and the outer boundary of the computational domain.
Convective heating contours for case "lunara" at Q' = 10 are also presented in
Figure 8. Heating distributions for cases "lunara", "lunarb", and "lunarc" along
the plane of symmetry at Q' = 10 are compared in Figure 9. These distributions show
none of the erratic behavior evidenced in earlier work associated with the coordinate
singularity. The peak heating point for the "lunara" configuration occurs on the
shoulder for Q' = 10. The larger shoulder radius of the "lunarb" configuration
reduces the peak shoulder heating (the reduction factor is slightly less than the square
root of shoulder radii ratio) at the expense of lower values for C L , CD, and LID.
The larger base radius of the "lunarc" configuration raises the effective radius of
curvature of the stagnation point on this sonic corner body, thus reducing stagnation
point heating relative to the two smaller bodies, even though they all have the same
nose radius.
Figure 8: Surface mesh and heating contours over aerobrake for case "lunara" at
a = 100.
60
lunara
lunarb
lunarc
50
40
'"E
(.)
0
30
'\
I, ' ' ,
,,
,
I
,
k
20
,;,
/"
10
0
20
10
x, ft
10
20
III
141
near wake
The outer boundaries of the wake computational domain in the symmetry plane
are straight lines extrapolated from th~ last two points of the outer boundaries of
the forebody grid. The forebody grid has been aligned with the captured bow shock.
An exit plane location is specified at a fixed z location several base diameters behind
the base. The midpoint of the exit plane is calculated and a circle is defined with
circumferential discretization equivalent to the forebody exit plane. The outer shell
of the wake boundary is defined by straight lines between corresponding points on
the exit plane of the forebody and the circular exit plane of the wake. Analytically
defined stretching functions are used to control mesh size extending from the exit
plane of the fore body flow domain to the exit plane of the wake.
An inner control shell which defines the boundary between wake zone 2 (extending from the outflow boundary of zone lover the forebody) and wake zone 3
(extending from the aerobrake base) must also be defined. The intersection of this
inner control shell with the exit plane forms a circle with a radius approximating the
radius of the wake core (one quarter of the base diameter). The radial distribution of
points in zone 2 blends smoothly with the radial distribution of points in zone 1. The
radial distribution of points at the exit plane of zones 2 and 3 are algebraically defined
to enhance resolution of the wake core, where the free shear layers come together. The
radial distribution of points along the base of the aerobrake in zone 3 is algebraically
defined to help resolve the free shear layer in the vicinity of the boundaries of zones
2 and 3.
(a) pressure
(b) streamlines
Figure 11: Pressure contours and streamlines in symmetry plane in near wake behind
lunar aerobrake "lunara" at a = 100. Contour levels defined by pi Poo F~ varying
between 0 and 0.04 with 81 levels, inclusive.
An initial flowfield is calculated to better define the positions of the free shear
layer and wake core. Grid control shells are then moved to enhance resolution of these
structures. The wake neck region, where the free shear layers come together and the
flow compresses and turns into the wake core through the action of the recompression
shock, may also be identified in the initial solution. The inner control shell is then
reconfigured into a funnel like shape in which the converging surface of the funnel
extends from the aerobrake shoulder to the wake neck and the tubular extension of
the funnel follows the wake core to the exit plane.
A representative grid is shown in Figure 10. Pressure contours in Figure 11 (a)
clearly show the recompression shocks surrounding the wake core, as defined by the
streamlines in Figure 11 (b) for case "lunara" at a = 100. The recompression shocks
are diffused somewhat because they are oblique to the grid aligned with the wake
core. Better resolution requires application of a solution adaptive grid strategy as in
reference [36). The translational and vibrational temperature contours, the subsonic
mach number contours and electron number densities in the near wake are recorded
in Figure 12. The subsonic flow domain in the wake is one measure of a relative
quiescent zone for payload packaging.
Caution must be excercised in the interpretation of near wake flow results for
continuum calculations at high a.ltitudes. For example, the computed Knudsen numbers in the base flow (i.e., the ratio of mean free path to local characteristic length)
143
~~
Figure 12: Contours in symmetry plane in near wake behind lunar aerobrake "lunara" at a = 10. Mach number contour levels vary between 0 and 1.0 with 11 levels,
inclusive.
for case "afe4" are as large as 1.0 in the lowdensity region immediately behind the
shoulder. Knudsen numbers in the highgradient bow shock are of order 0.10. The
local characteristic length is equal to Ipj\! pi and gives an approximation of the length
scale over which significant changes in the flowfield are accommodated. The continuum approximation breaks down when the Knudsen number is of order 1 or larger.
A calculation at a higher altitude earlier in the trajectory (case "afeI") revealed
order 1 Knudsen numbers in the captured shock and a larger region of order 1 values
behind the corners of the aerobrake. Simulation errors due to the limitations of the
present equation set in these isolated areas of large Knudsen number require more
study, but may be more amenable to treatment by direct simulation methods [37] or
a modified Burnett equation analysis [38].
CONCLUDING REMARKS
An upwindbiased, pointimplicit relaxation algorithm for obtaining the numerical solution to the governing equations for threedimensional, viscous, hypersonic
flows in chemical and thermal nonequilibrium is described. The present thermochemical nonequilibrium model includes 11 species and two temperatures. The algorithm is derived using a finitevolume formulation in which the inviscid components
REFERENCES
[1] Walberg, Gerald D.: "A Survey of Aeroassisted Orbit Transfer" ,Journal of Spacecraft and Rockets, Vol. 22, No.1, JanuaryFebruary 1985, pp 318.
[2] Gnoffo, Peter A.; Gupta, Roop N.; and Shinn, Judy: "Conservation Equations and Physical Models for Hypersonic Air Flows in Thermal and Chemical
Nonequilibrium," NASA TP 2867 (1989).
[3] Park, Chul: "Problems of Rate Chemistry in the Flight Regimes of Aeroassisted
Orbital Transfer Vehicles", Thermal Design of Aeroassisted Orbital Transfer Vehicles, edited by H. F. Nelson, Progress in Astronautics and Aeronautics, Vol.
96, AIAA, New York, NY, 1985.
[4] Park, Chul: "Assessment of TwoTemperature Kinetic Model for Ionizing Air",
AIAA Paper 871574, June 1987.
[5] Lee, JongHun: "Basic Governing Equations for the Flight Regimes of Aeroassisted Orbital Transfer Vehicles", Thermal Design of Aeroassisted Orbital Transfer Vehicles, edited by H. F. Nelson, Progress in Astronautics and Aeronautics,
Vol. 96, AIAA, New York, NY, 1985, pp 353.
[6J Gnoffo, Peter A.: "An UpwindBiased, PointImplicit Relaxation Algorithm for
Viscous, Compressible PerfectGas Flows," NASA TP 2953, February 1990.
[7J Gnoffo, Peter A., "UpwindBiased, PointImplicit Relaxation Strategies for Viscous, Hypersonic Flows," AIAA Paper 891972CP (1989).
[8J Gnoffo, Peter A., "Code Calibration Program in Support of the Aeroassist Flight
Experiment," Journal of Spacecraft and Rockets, Vol. 27, No.2, pp 131142,
MarchApril 1990.
[9J Candler, G.: "On the Computation of Shock Shapes in Nonequilibrium Hypersonic Flows," AIAA Paper 890312, January 1989.
[10] Candler, G. V. and MacCormack, R. 'vV.: "The Computation of Hypersonic
Ionized Flows in Chemical and Thermal Nonequilibrium", AIAA Paper 880511,
January 1988.
[11] Park, Chul and Yoon, Seokkwan: "A FullyCoupled Implicit Method for ThermoChemical Nonequilibrium Air at SubOrbital Flight Speeds," AIAA Paper 891974, June 1989.
[26] Micol, John R.: "Experimental and Predicted Pressure and Heating Distributions for an Aeroassist Flight Experiment Vehicle in Air at Mach 10," AIAA
Paper 891731,1989.
[27] Grasso, Francesco and Gnoffo, Peter A.: "A Numerical Study of Hypersonic Stagnation Heat Transfer Predictions," Proceedings of the Eighth GAMMConference
on Numerical Methods in Fluid Mechanics, NNFM. Vol. 29, Vieweg, Braunschweig 1990, pp 179188.
[28] Wells, William 1.: "Surface Flow and Heating Distributions on a Cylinder in
Near Wake of Aeroassist Flight Experiment (AFE) Configuration at Incidence
in Mach 10 Air," NASA Technical Paper 2954, 1990.
[29] Hornung, H. G.: "Nonequilibrium Dissociating Nitrogen Flow Over Spheres and
Circular Cylinders," J. Fluid Mechanics, Vol. 53, part 1 (1972), 149176.
[30] Cauchon, D. 1.: "Radiative Heating Results from the Fire II Flight Experiment
at a Reentry Velocity of 11.4 Kilometers per Second," NASA TM X1402 (1966).
[31] Sutton, K.: "Air Radiation Revisited," AIAA Paper 841733 1984.
[32] Zoby, E. V. and Sullivan, E. M.: "Effects of Corner Radius on StagnationPoint
Velocity Gradients on Blunt, Axisymmetric Bodies," NASA TM X1067 1965.
[33] Gupta, R. N.: "NavierStokes and Viscous ShockLayer Solutions for Radiating
Hypersonic Flows," AIAA Paper 871576 1987.
[34] Gnoffo, Peter A., Price, Joseph M., and Braun, Robert D.: "On the Computation
of Near Wake, Aerobrake Flowfields," AIAA Paper 911371, June 1991.
[35] Stewart, David A.; Leiser, Daniel B.; Kolodziej, Paul; and Smith, Marnell:
"Thermal Respone of Integral Multicomponent Composite Thermal Protection
Systems," AIAA851056 1985.
[36] Davies, C. and Venkatapathy, E.: "A Simplified SelfAdaptive Grid Method,
SAGE," NASA TM 102198, October 1989.
[37] Moss, James N.; Bird, Graeme A.; and Dogra, Virendra K.: "Nonequilibrium
Thermal Radiation for an Aeroassist Flight Experiment Vehicle", AlA A Paper
880081 1988.
[38] Moreau, S., Chapman, D. R., and MacCormack R. W.: "Effect of Rotational
Relaxation and Approximate Burnett Terms on Hypersonic FlowField Radiation
at High Altitudes," AIAA Paper 911702, June 1991.
[39] Roe, P. L.: "Approximate Riemann Solvers, Parameter Vectors, and Difference
Schemes," Journal of Computational Physics, Vol. 43 (1981), 357372.
APPENDIX
Definition of A, R, R 1, and A
The Jacobian of g with respect to q is expressed
U(o"  c.)
irnx  Uu
irny  Uv
1'rnzUw
irU  UH
Uev
c&n X
un x (1  (3) + U
f3uny + vnx
f3un z + wnx
f3uU + Hnx
evnx
c.ny
f3vn x + un y
vny(l  (3) + U
f3vn z + wny
f3vU + Hny
evny
c$n Z
f3wn x + un z
f3wny + vn z
wnz(l (3) + U
f3wU + Hnz
evnz
0
f3nx
f3ny
f3nz
f3U +U
0
0
</Jnx
</J n y
</Jnz
</JU
U
(52)
bsr / a 2
u/a 2
v/a 2
w/a 2
0
Ix
Iy
Iz
iJ(u 2 +v 2 +w 2 )'Yr
iJa 2
a 2bsr  c/Yr
V
R 1
w
ir  Ua
ir + Ua
evir
0
cs /2a 2
mx (u + an x )/2a 2
my (v + any)/2a 2
m z (w + an z )/2a 2
W (H + aU)/2a 2
0
ev/ 2a2
f3uc s
Ix
mx
f3u + anx
f3u  anx
f3 ue v
f3vc s
Iy
my
f3v + any
f3v  any
f3 vev
0
cs /2a 2
(u  an x )/2a 2
0
(v  any)/2a 2
0
(w  an z )/2a 2
0
(H  aU)/2a 2 / f3a 2
ev /2a 2
1/a 2
f3wc s
lz
mz
f3w + an z
f3w  an z
f3 we v
f3c s
0
0
f3
f3
f3e v
(53)
c s
0
0
a 2  ev
(54)
U 0 0
0
0
0
0 U 0
0
0
0
0
0
0
0 0 U
0
0 0 0 U +a
0
U a 0
0 0 0
0
0
0 0 0
0
U
(55)
149
c. = P./p
ir
(57)
Pe
= R 
(58)
8p _
8pE 8p
8p ev
R
pr
pCv,tr r=l,re Mr
pCv,v Me
8p
RTq
u2 + v2
+
(3
2
8ps
M.
The variable a is the frozen speed of sound.
is =
+ w2
N.
a2
I>.is + (3[H  u
v2
w2]
ev
(3e.  ev,s
(59)
= (1 + (3)'E..
=1
(60)
This definition of a 2 comes from the evaluation of the eigenvalues of A. The variable
R is the universal gas constant, and M. is the molecular weight of species s.
The variables n x , ny, and n z are the x, y, and z components of a unit vector
normal to a computational cell face, and U is the normal component of velocity
through the cell face, defined by
U = un x
vny
(61 )
wn z
The two unit vectors I and are defined such that ii, ~ and are mutually orthogonal
(i.e., nil; = nimi = limi = 0). The velocity components in the land
directions,
tangent to the cell face, are then defined by
(62)
W = um x
vmy
W1n z
(63)
In the matrices defined above, the first row and column correspond to the N.
species continuity equations. Subscript s refers to row s and species s, and subscript
r refers to column r and species r where both sand r vary from 1 to 11 in the present
model. Note in Equation 59 above that Tq equals Tv when s is an electron; otherwise,
Tq equals T. Further details of the derivations may be found in Reference [2].
The variables at cell faces are evaluated as follows:
C1 ai
C1
is,/
RT. +
= '!1
Ms
+ aLl
+ 1
2+ 2+ 2
(31 U I
v21
wI
(64)
(65)
(66)
(67)
tv, C s ,
C1 =
( TL )
TL 1
C2 =
( PL )
PLl
(68)
(69)
This averaging procedure is a departure from the Roe's method [39] used in
the perfect gas version of this algorithm [6]. It evolved over the early code development period using a series of numerical tests which revealed a sensitive dependence
of the TVD (nonoscillatory) shock capturing capabilities on the definition of al and
1's,/. The weighting on sound speed and temperature is motivated by the observation that the thermodynamic state of the gas on the freestream (low temperature)
side of a strong shock is of negligible importance in the consideration of momentum
and energy balance across the shock. The weighting on the other variables follows
the example of Roe. The present formulation neither enforces Roe's property U on
the pressure jump between two adjacent cells nor does it provide an average sound
speed al which exactly satisfies Equation 60 as a function of cell averaged variables.
However, it has provided robust convergence characteristics in the vicinity of strong
shocks (Moo > 30) without significant smearing of the shock transition zone using
a very simple formulation. Vinokur [40], Liu and Vinokur [41], and Grossman and
Cinnella [42] have proposed averaging schemes which enforce Roe's property U. These
formulations have not been investigated here.
Geometrical Relations
The recommended, secondorderaccurate formulations for facecentered metrics, which
are required in the evaluation of the viscous dissipation terms across cell faces, are
presented below.
V~i,J,J( =
n I ,J,Idifi  1 ,J,I';
(70)
4n l ,J,J(n l  1 ,J,I(
nl,J,J(( ifi ,Jl,J(
(71)
4fh,J,J(nl,Jl,K
n I ,J,K(ifi ,J,Kl
nl,J,K( if ll,i,K
(72)
(73)
DI,J,K(ihjl,K
'9TJ[,j,K
151
(74)
4DI,J,[{D[,Jl,K
DI,J,K(ih,j,Kl
'9 TJ[,J,k
(75)
4D[,J,KD[,J,Kl
'9 (i,J,K
D[,J,K(8[1,J,k
(76)
4D[,J,KD[1,J,K
D[,J,K(8[,Jl,k
'9([,j,K
(77)
4D I ,J,I,D[,Jl,K
'9 (I,J,k
D I ,J,K(8[,J,kl
(78)
4D[,J,I{D[,J,K 1
In the case of the thinlayer NavierStokes equations, only the vectors defined
by Equations 70,74,78 are required, depending on the orientation of the coordinate
system, The dot product of these vectors with the corresponding unit normal to the
cell face (recall, xn = '9X . ii) can be approximated as follows,
~ni,J,K
TJn[,j,K
DI,J,[{( i7il,J,K
i7i,J,K)
D[I,J,g(i7j,J,K
.i7 i+1,J,K)
(79)
4D[,J,KD I  l ,J,K
DI,J,K (i7I,jl,K
i7[,j,K)
D[,Jl,K( i7 [,j,K
i7[,j+1,K)
(80)
4D[,J,KD I ,Jl,K
(nI,J,k
D[,J,K(i7I,J,kl
i7[,J,k)
D[,J,K 1 (i7I,J,k
i7[,J,k+1)
(81)
4Dl,J,KD[,J,Kl
( ~: )
( ~~ )
(
~~)
(~ii ns)i,J,K
(82)
i,J,K
(83)
I,j,K
(84)
I,J,k
@No copyright is asserted in the United States under Title 17, U.S. Code. The U.S. Government
has a royaltyfree license to exercise all rights under the copyright claimed herein for Governmental
purposes. All other rights are reserved by the copyright owner.
Chapter 5:
FluxSplit Algorithms for Hypersonic Flows
P. Cinnella (*), B. Grossman (**)
(*) ERC for Computational Field Simulation, Mississippi
State University, P.O. Box 6176, Mississippi State,
MS 39762, USA
(**) Department of Aerospace and Ocean Engineering,
Virginia Polytechnic Institute and State University,
215 Randolph Hall, Blacksburg, VA 24061, USA
ABSTRACT
This study reviews numerical techniques for the simulation of fluid flows spanning the range of reactive regimes, from local chemical equilibrium to full thermochemical nonequilibrium. In particular, characteristicbased algorithms are considered, of the fluxvector and the fluxdifference type, which are becoming ever more
popular due to their accurate rendition of physically complicated, shockwave dominated flows. Consideration is given to the problems associated with modeling the
thermochemical behaviour of reactive mixtures of thermally perfect gases, and a
few topics of current research are outlined.
1. INTRODUCTION
In recent years, a renewed interest in the field of hypersonic aerodynamics has
been registered among the scientific community. Vehicles like the Space Shuttle
or the European Hermes are fully operational or in an advanced design stage, respectively. Their conception would not have been possible without some knowledge
of chemical and thermal nonequilibrium phenomena and how they affect the performance of engines, lifting surfaces and materials, to name a few examples. The
Aeroassisted Orbital Transfer Vehicle (AOTV), conceived as an essential tool for
possible exploratory missions on Mars and for efficient satellite management in high
earthcentered orbits!, is anticipated to make use of aerobraking in conditions that
are not reproducible on earth in existing experimental facilities, nor will be in the
near future. Other aircraft, like the National AeroSpace Plane (NASP) or the European Hypersonic Transport Vehicle, are becoming national priorities and subjects
for commercial competition. In this context, the availability of accurate, robust
and efficient simulation tools for the study of the complex physical phenomena that
occur in the hypersonic regime becomes an essential priority for the scientist and
the designer2. To this end, several computer codes have been recently developed 3  8
which are capable of representing geometrically complex threedimensional flowfields
with nonequilibrium chemistry and thermodynamics. Most of them use upwind
technology in the discretization of the nonlinear convective terms that appear III
the governing equations.
The essential core of this work is the analysis of the numerical techniques that
are at the center of the current, upwindbased simulation capabilities. In particular,
some classes of fluxsplit algorithms will be discussed, with the intent of showing
how these methodologies are applicable to the study of reactive flows. Originally developed for perfect gases (e.g., see the survey papers of Harten, Lax and Van Leer 9 ,
and Roe IO ), these techniques have produced accurate solutions of shockwave dominated flows. Their essential feature is the treatment of the convective terms in the
NavierStokes equations using a characteristicbased differencing. In particular, the
three most popular types of approaches will be discussed, namely the fluxvector
splittings of Steger and Warming ll and Van Leer l2 , and the fluxdifference splitting
of Roe l3 . Extensions of these algorithms to flows in local chemical equilibrium l4  2o
and to finiterate chemistry flowS 21  27 have appeared in the literature, including in
some instances consideration of thermodynamic nonequilibrium effects.
The subject matter of this work will be organized in the following way: the influence of hightemperature, highvelocity effects on the thermodynamic behaviour
of mixtures of gaseous species will be examined in 2, and the influence on their
chemical behaviour will be discussed in 3. The governing differential equations for
viscous, heatconducting flows which are out of chemical and thermal equilibrium
will be presented in 4. The reduced form valid for flows in local chemical equilibrium will also appear. Nonequilibrium thermodynamic and transport phenomena
will be briefly summarized in 5. Fluxsplit algorithms for the treatment of chemically and thermally active flows will be detailed in 6, and efficient techniques for
timeaccurate or steadystate calculations will be mentioned in 7. Finally, some
conclusions will be drawn in 8.
2. THERMODYNAMIC MODELS
Hypersonic flow is characterized by regions of high temperatures, induced by
the strong shocks associated with the high velocities involved. In this context,
deviations from the ideal gas behaviour are warranted, and manifest themselves in
the activation of internal energy modes which behave nonlinearly with temperature,
in the onset of chemical reactions, and ultimately in the establishment of local
thermodynamic nonequilibrium.
155
(1)
In the above, eAT) is the contribution due to translation and internal modes that
can be assumed to be in equilibrium at the heavyparticle translational temperature T. Moreover, en, (Tn,) is the contribution due to internal modes that are in
thermodynamic nonequilibrium, meaning that they are not in equilibrium at the
translational temperature T, but may be assumed to satisfy a Boltzmann distribution at a different temperature Tn,. Finally, ee, (Te) is the contribution from
the electronic states in a particle, or from free electrons when present, where it
has been recognized that the "electronic cloud" will reach internal equilibrium very
rapidly24, hence only one electronelectronic temperature Te has been used for all
of the species. Cases when the electronic contributions are considered coupled with
the vibrational and rotational states can be handled by this representation if the
nonequilibrium temperature Tn, corresponds to these coupled modes, and the term
ee, is dropped.
Equation 1 is valid for a very broad range of physical conditions, and is susceptible to simplification and reduction to simpler models when the flow regime is less
"extreme". Typically, the contribution in equilibrium at the translational temperature es will represent translational and rotational modes, the nonequilibrium term
en, will coincide with vibrational energy, and the electron/electronic contributions
will be either neglected 29 , or considered to be in equilibrium at the electronic temperature Te 28. Some authors suggest that partitioning vibrational and rotational
modes is not appropriate for hypersonic flow conditions 24 ,3o, and in that case en,
would represent the combined rotational/vibrational contribution. Other practical
It is convenient to express
volume
es , en"
ee, =
Te
cVe,,(T)dT,
(2a, b, c)
where h f, is the heat of formation of species s at the reference temperature T ref .
The mixture value for the internal energy per unit mass, e, may be written using a
standard massfraction averaged summation, as follows
N
s=1
s=1
s=1
s=1
(3)
157
where the species mass fraction C s has been introduced, defined as the ratio of
species density Ps to mixture density p = L:~ Ps. Similarly, the frozen specific
heats at constant volume, cv , and at constant pressure, cp , may be obtained by
means of the mixture rule, as follows
N*
N*
Cv =
2::::: csc v ,
N*
cp =
s=l
s=l
(4a, b)
8=1
R=
2::::: C
R8 =
p 
Cv
(5)
8=1
(6a, b)
P=
(7)
s=l
where the mixture gas constant R has been defined in Equation 5. The Simplified NonEquilibrium Thermodynamic Model and the Equilibrium Thermodynamic
Model will utilize a similar Thermal Equation of State, but the contribution from
the electron/electronic temperature Te will be dropped, due to the fact that free
electrons, when present, will have the same translational temperature as the heavy
particles.
Unlike the ideal gas case, the state relationship of the pressure to the specific internal energy cannot be written directly in general, but occurs implicitly
through the translational temperatures. For the Full NonEquilibrium Thermodynamic Model, Caloric Equations of State such as Equations 2c and 3 will relate
internal energy, or portions thereof, to temperature. Moreover, when the knowledge of the nonequilibrium temperatures Tn, is required, additional relationships
such as Equation 2b will have to be utilized. Iterative procedures are necessary for
the determination of the relevant temperatures from known values of the internal
energy contributions, due to the inherently nonlinear character of these equations,
but they are found to be numerically efficient in most circumstances 27 . Once T and
Te are found, the pressure is evaluated from Equation 7.
The Simplified NonEquilibrium Thermodynamic Model and the Equilibrium
Thermodynamic Model will still necessitate a caloric equation of state such as
Equation 3, which in general will be nonlinear in the translational temperature
T. Practical models have been proposed 27 where the nonlinearity is removed, and
the temperature can be eliminated between Equations 3 and 7 to yield an Equation
of State linking pressure to internal energy directly. However, barring special cases
a ==
(8ap )
P
= 'Y_RT = 'Y_(p)
8,p,/p,e n ,
(8)
where s is the entropy, and l' is the ratio of frozen specific heats l' = cp/c v . The
Equilibrium Thermodynamic Model will feature the same result, however the nonequilibrium internal energy contributions en, will formally disappear from the previous expression. The interestingly simple final formula for the frozen speed of sound
was described by Clarke and McChesney36.
The special sub case of the Equilibrium Thermodynamic Model that features
flows in local chemical equilibrium will have an equilibrium speed of sound defined
as
a
==
Cap7P )
= rRT = r(~),
P
where the isentropic index r = r(p ,T) is given by the following
s
(9)
s=l
,_ 
ah/ aT  1 +
ae/aT 
(lOa, b)
="="'''C
.
In the above, l' is the ratio of the partial derivative of enthalpy with respect to
temperature at constant density to the partial derivative of the internal energy with
respect to temperature, where the latter is the specific heat at constant volume C v .
The use of the isentropic index preserves the simplicity of the formula for the sound
speed, but the functional dependence of this index from density and temperature is
relatively involved.
There are at least five different "gammas" that can be defined for a gas in
chemical equilibrium: the isentropic index r, the ratio of specific heats, = cp/c v ,
the ratio of frozen specific heats l' = cp/c v , the ratio of enthalpy and internal energy
derivatives 1', and the ratio of enthalpy to internal energy
= h/ e = 1 + p/ pe. For
a frozen flow, the first four of these quantities will reduce to the same value, but for
general flow conditions their values will remain different.
,*
es =nsRsT + h~
3
eN =2RNTe,
(lla,b)
    _ .__. _   
161
where the reference temperature has been set to the absolute zero, and h~ is the
corresponding species heat of formation, which is zero for free electrons. The coefficient ns will be equal to 3/2 for a monatomic gas, it will be 5/2 for a diatomic or a
linear polyatomic molecule, and will equal 6/2 for a nonlinear polyatomic molecule.
The relationships such as Equations 2b and 2c that link vibrational and electronic
contributions to their respective temperatures are usually specialized using simple
harmonic oscillator formulas 35 for the vibrational modes and only the first few electronic states
e
 R
n, 
Ne"
'\""
s ~
i=l
8.
V,S,!
(12a,b)
where Ne,s is the number of characteristic vibrational temperatures 8 v ,s,i for species
s, and the sum over electron energy levels whose degeneracy factors are gi and
characteristic temperatures are 8i is typically truncated after the first or second
contribution. In the context of the Simplified NonEquilibrium Thermodynamic
Model, the previous relationships will have to be modified to make Te disappear
(13)
where the nonequilibrium contribution en, will contain both vibrational and electron terms
and free electrons, when present, will be in equilibrium at the (only) translational
temperature T.
The second practical nonequilibrium model is somewhat simpler than the first
one. It assumes that the vibrational and electronic modes will exchange energy very
efficiently, and consequently they will reach a common vibrational/electron temperature Tn valid for all species that possess an equilibrium contribution. There has
been some indication that the previous assumption is valid for air during atmospheric reentry37. This model is less computationally expensive than the previous
one, because only one nonequilibrium energy equation will be necessary instead of
either M or M + 1. Moreover, it may be used again in both Full and Simplified
contexts, provided that the previous Equations 1114 are modified by substituting
Tn for every entry that read Te or Tn,.
The first practical equilibrium model is a direct simplification of the nonequilibrium models considered, in that it considers translational and rotational contributions to the species internal energy at their fully excited values, whereas the
vibrational modes are included by means of a simple harmonic oscillator formula
TFD
~TCH,j'
~ TCH,j,
(16)
~ TCH,j'
In the first case, reaction j has no time to occur, and a frozen flow can be assumed with respect to that specific reaction. The second case is the general case of
finiterate chemistry whereby the actual kinetics of the reaction must be taken into
consideration. In the third case, reaction j has a virtually infinite time to evolve,
and consequently it will reach its equilibrium value given by the Law of Mass Action.
163
In practice, calculations are performed where all of the chemical reactions are
supposed to be in equilibrium, frozen, or evolving with a finite rate. Perfect gas
results are a particular class of frozen chemistry simulations. Moreover, from the
above discussion it should be clear that frozen and equilibrium models are important limiting cases of the reallife finiterate chemistry situation, which requires
simulating the actual kinetic behaviour of the chemical system.
3.1 FiniteRate Chemistry
A general simulation of chemical effects can be achieved for a system containing
N species where J reactions take place
' ,]X 1
VI'
'X>."X
fIX N
' 2 + ... + VN'
+ V'2 X
N ~ VI' 1 + V2/IX
' 2 + ... + VN'
,]
,]
,],J
j = 1, ... ,J,
,}'
(17)
where the v:,j and the V~,j are the stoichiometric coefficients of the reactants and
products of species Xs in the jth reaction, respectively. Then, the rate of production
of the sth species, w s , may be written as a summation over the reactions, as
follows 35
ws
==
dPs
dt = Ms
g(
",
[N
 Vs,j) kt,j
~(VS,j
PI )
MI
v; .
,J
 kb,j
gN (
v!'.]
PI )
MI
,J
'
(18)
where M s is the species molecular mass. For reaction j the forward and backward
reaction rates, k t,j and kb,j, are assumed to be known functions of temperature,
and are related by thermodynamics
(19)
where Ke,j is the equilibrium constant, which is a known function of the thermodynamic state. Using Equation 19, Equation 18 can be rewritten as follows
J
_ dps
",
Ws = dt = Ms '2)Vs,j  Vs,j)kb,j
)=1
[N
_
PI
ViI,J.
PI
(MI)
vOl
I,J
1=1
(20)
where the term in the brackets goes to zero when equilibrium compositions are
reached for ps, because it reduces to a formulation of the Law of Mass Action
j = 1, ... ,J,
valid for equilibrium concentrations.
(21)
The finiterate chemistry model described remains valid in conjunction with all
of the thermodynamic models discussed in the previous section. However, the presence of thermodynamic nonequilibrium in the flowfield is likely to exercise some
effect upon the reaction rates, especially when diatomic or polyatomic molecules
are involved, whose vibrational modes may be excited. Several attempts to model
the interaction between chemical and thermodynamic nonequilibrium have been
recorded 42  44 . In most cases, the forward and backward reaction rates are redefined so that they are functions of both translational and nonequilibrium temperature(s). The simplest model employs an "effective" temperature defined as the
geometric average of translational and vibrational temperatures, to be used for the
molecular dissociation reactions 37 . More refined models try to model "preferential
dissociation" of the molecules with excited vibrational modes 43 , although there is
only controversial evidence that this is a firstorder effect for most flows of interest 45 .
Park 46 suggests that the relationship between forward and backward rates through
the equilibrium constant, Equation 19, may be incorrect when effective temperatures enter the picture, which would invalidate the standard procedure of measuring
and/ or calculating only one of the two sets of reaction rates and using the equilibrium
constant to recover the other set. The presence of an electron/electronic temperature Te further complicates the picture, due to the appearance of more effective
temperatures 46 .
The previous chemistry model is valid for gaseous systems and involves homogeneous reactions only, therefore excluding solid or liquid surface as well as photochemical reactions. At the present time, wall effects in general and wall catalyticity
in particular are still not entirely understood, let alone modeled. Wall treatment in
the hypersonic regime is typically accomplished by assuming either a fully catalytic
solid boundary, and consequently chemical equilibrium at the local temperature and
density, or a frozen wall, with corresponding zero gradients of mass fractions. A few
attempts of modeling finiterate catalyticity in terms of absorption and desorption
phenomena have been initiated 47 , but an accurate simulation of solid boundaries
for highspeed flows has yet to come.
3.2 Chemical Equilibrium
There are many problems that render a finiterate computation into a nontrivial task. The most important one is the uncertainty in the reaction paths and
in the reaction rates 2 ,35. Although the composition of a given mixture in chemical
equilibrium as a function of thermodynamic variables can be determined quite accurately by theoretical and experimental means, there is very scarce knowledge of
the actual mechanism with which a given reaction occurs. Many paths, leading to
intermediate products or highly unstable excited states, are available for virtually
every reaction. It is a delicate theoretical work to rule out some of the possibilities
with respect to others, and it is an impressive experimental work to confirm these
results in order to obtain kinetic rate data for specific reaction paths. Moreover, the
numerical simulation of flows in chemical nonequilibrium requires the inclusion of
165
N species continuity equations into the algorithms instead of only one global mass
conservation equation, as will be discussed in 4. Also, the finiterate equations
become extremely stiff when the flow is close to chemical equilibrium48 . All of
the above reasons render a numerical simulation based upon chemical equilibrium
very appealing from a computational standpoint, when compared with finiterate
calculations. In addition, for a large number of flows of practical interest results of
considerable accuracy can be obtained using this simplifying assumption.
From Equation 20, it is easy to recognize the limiting cases of frozen and equilibrium flows. When the backward rate kb,j  t for any j, then the species rates of
production go to zero, and it can be shown that the species mass fractions are unchanged if no diffusion occurs. On the other hand, when the backward rate kb,j  t 00
for any j, then the terms in brackets will tend to zero, because the equilibrium composition is maintained, and the species rates of production reach a limit value that
will balance convection, diffusion and unsteady terms. Consequently, local chemical
equilibrium corresponds to the limiting value of reaction rates going to infinity at
every point in the flowfield 34 . This limiting value will be reached while keeping two
state variables constant, which will allow for a unique description of the equilibrium
state. A close examination of the species continuity equation and of the combined
First and Second Laws of Thermodynamics indicates that local chemical equilibrium
calculation should be performed at constant density and internal energy49. At this
stage, the unknowns to be determined are species densities and temperature.
Consequently, aside from problems involving the fundamental base air mixture, the
simulation of flowfields in local chemical equilibrium for a general gas mixture has
to rely on the efficient coupling of chemical composition solvers, based upon the
previously mentioned governing equations, with flowfield solvers, for the "standard"
Euler or NavierStokes equations.
3.3 Thermodynamic Properties for Flows in Local Chemical Equilibrium
After the local equilibrium composition at a given density and internal energy
has been determined, the different thermodynamic state variables can be evaluated.
Among them, several have a significant practical interest, for example speed of sound
a, given by Equation 9, isentropic index r, given by Equation lOa, and pressure, as
well as transport properties such as viscosity coefficient and thermal conductivity.
The evaluation of all of these quantities is made possible by the knowledge of species
densities and temperature at equilibrium, but it also requires the values of the
partial derivatives of the mass fractions with respect to temperature and density, as
it appears from Equations 10. In general, these derivatives are determined from the
governing equations by means of the Implicit Function Theorem 34 , due to the fact
that Laws of Mass Action and Mass Constraints are the relationships that implicitly
define the behaviour of mass fractions with changing temperature and density.
For the specific important example of a standard air mixture, tabulations and
curve fits are available for a broad range of density and temperature for virtually all
of the thermodynamic and transport properties of interest 50 ,51. Again, for a generic
mixture of thermally perfect gases either specific tabulations have to be created
prior to the solution of a flow problem, or the chemical equilibrium solver has to be
coupled with the flow solver.
3.4 Practical Chemistry Models
It was previously mentioned that a detailed knowledge of the kinetic behaviour
of a chemically reacting gas mixture is necessary when finiterate chemistry investigations are deemed necessary. In particular, the specific reaction paths leading to
dissociation, ionization, and atomic rearrangement are required. In recent years, a
lot of attention to the kinetics of air and of hydrogen/ air mixtures has been registered in the scientific community, spurred by the drive towards hypersonic flight and
the necessary propulsive tools to achieve it 2. A compilation of the most recent data
for the kinetics of air may be found in Park 37 , where attention is given to thermodynamic nonequilibrium and its influence upon the reaction rates through effective
temperature, as previously discussed. Detailed studies of hydrogen/air mixtures
have been published recently52.
When local chemical equilibrium calculations are performed, the actual reaction
paths are no longer necessary. Chemical equilibrium is dictated by thermodynamics,
namely the entropy has to be a maximum for a system at constant density and
internal energy35. Even when chemical reactions are actually necessary in order to
 

167
formulate the Laws of Mass Action, any set of independent reactions can be used
to obtain the same final result. The treatment of chemical reactions for equilibrium
chemistry becomes a bookkeeping device.
4. GOVERNING EQUATIONS
The governing equations for flows in thermochemical nonequilibrium represent the conservation and/or time variation of physical quantities such as mass of
a species in the mixture, momentum, nonequilibrium energy contributions, and
energy, established either for a control volume of arbitrary size (finite or infinitesimal), or for a generic point in the flowfield. The former approach results into an
integrodifferential form,
, the latter into a nonlinear differential system.
4.1 Integral FormThermoChemical NonEquilibrium
The governing equations written in integral form are valid both in regions of
smooth flows and across discontinuities 53 , where they can be shown to reduce to
the RankineHugoniot jump conditions in the limit of infinitesimal size. The control volume employed for the derivation is allowed to move and deform with time,
although many important applications take advantage of the simpler case of fixed
control volumes. For an arbitrary volume V, closed by a boundary S, the governing
equations in integral form read
Sv)' ndS =
JJfv
WdV,
(22)
where Q is the vector of conserved variables, W is the vector of source terms, and
Sand Sv are the inviscid and viscous flux vectors, respectively. The unit vector n
is normal to the infinitesimal area dS and points outwards. In conjunction with the
Simplified NonEquilibrium Thermodynamic Model, and utilizing a Cartesian frame
of reference (Xl, X2, X3) whose unit vectors are i, j, and k, respectively, the vectors
Q and W read
PI
P2
Q=
PN
PUI
PU2
PU3
Ple nl
PMe nM
pea
WI
W2
WN
W=
2::. P.g'
2::. P.g'
2::. P.g'
l
2
3
Ql
QM
(2::. P.g.) . u
(23a, b)
S=
Pl(UUV)
pz(u  uv)
PI V I
PZ V2
PN(UUV)
PNVN
Tlli + TIZj + T13k
T21 i + TZ2j + T23 k
T31 i + T3zi + T33k
P1enl VI  qnl
pu 1 (u  U V) + pi
pU2(UUV)+pj
pU3(U  uv) + pk
P1enl(u  uv)
(24a,b)
PMenM(u  UV)
peo(u  UV) + pu
where
q  qR 
j=1
s=1
(25)
In the previous formulas, the massaveraged velocity U for the mixture has been
utilized, defined in terms of species velocities Us as follows
N
pu =
2..= psu s .
(26)
s=1
The difference between species and massaveraged velocities IS a specIes diffusion
velocity Vs
.5 = 1, ... ,N.
(27)
Vs=usu,
The velocity Uv that appears in the inviscid flux vector is the control surface velocity,
which is identically zero when the control volume is fixed with time. The first N
equations are species continuity equations, relating the time rate of change of species
densities Ps to convective and diffusive transport and to the creation/destruction of
the species due to chemical reactions. The species rate of production Ws has been
defined by Equation 18, when finiterate chemistry models were discussed. Following
species continuity are the three components of the momentum equation. The source
term :Z::::s Psgs represents body forces, e.g. gravity and electric fields, and the viscous
fluxes involve the components of the shear stress tensor, Tij. After momentum, M
nonequilibrium energy equations are written, describing the creation and evolution
in time and space of the nonequilibrium components of the internal energy. The
viscous flux associated with these equations describes the heat transfer due to the
transport of nonequilibrium energy by diffusion, pse n , Vs, as well as the conductive
169
heat transfer, qns. The source terms for the nonequilibrium energy equations, Qs,
will be described in more detail in 5. Finally, the global energy equation describes
the time evolution of the total internal energy per unit volume pea = pe + pu6/2.
The viscous flux in this case represents viscous dissipation, convective heat transfer
due to both equilibrium and nonequilibrium portions of the internal energy, q
and 2:: j qnj respectively, radiative heat transfer, qR, and diffusive heat transfer,
2::s Pshsvs, where hs is the species enthalpy, hs = e s + RsT.
In addition to the equations presented, thermal and caloric equations of state
will relate pressure and temperatures to equilibrium and nonequilibrium internal
energy components, as discussed in 2.2, and the viscous flux vector entries will
be expressed in terms of the entries in the vector of conserved variables, as will be
described in 4.4 and 5.5.
The vectors that appear in the governing equations for cases when the Full N onEquilibrium Thermodynamic Model is employed are very similar to those already
discussed, but they contain one additional equation, namely a nonequilibrium energy equation for free electrons 33 . At the present time, numerous assumptions and
approximations are used in the derivation and several simplifications are made to
this equation 28 . The reader is referred to the work by Park 37 , and Grossman, Cinnella, and Eppard 32 for some examples of the use of this equation in hypersonic
flows. The whole problem of multiple translational temperatures and their effect
upon the gasdynamics of nonequilibrium flows is still an object of current research.
Simplifications in the thermophysical model employed for the analysis of nonequilibriumflows will bring about corresponding changes in the governing equations.
When only one common vibrational/electronic temperature appears in the thermodynamic model, only one nonequilibrium energy equation is solved instead of M,
which can be written by summing up the contributions from each nonequilibrium
component of the internal energy. In this instance, if free electrons are present,
their kinetic energy is usually neglected 21 , and the complications arising from the
presence of two translational temperatures are ignored 37 . Similarly, the Equilibrium
Thermodynamic Model can be handled by simply dropping all of the nonequilibrium
energy equations, along with the nonequilibrium contributions to the global energy
equation. The whole process is equivalent to setting M equal to zero. Inviscid flows
are modeled by dropping the viscous flux vector, in conjunction with any of the
thermodynamic models discussed.
4.2 Integral FormLocal Chemical Equilibrium
The previous forms of the governing equations are written for finiterate chemistry problems. When the important special sub case of the Equilibrium Thermodynamic Model that deals with local chemical equilibrium is considered, the vectors
that appear in Equation 22 are further simplified. The global continuity replaces the
N species continuity equations, the nonequilibrium energy contributions disappear,
PUp 1 )
PU 2
(28a, b)
( PU 3
peo
where only body forces are present in the vector of source terms W. Inviscid and
viscous fluxes read
s=
p(u  uv)
)
PU1(U  uv) + pi
pU2(UUV)+pj ,
( PU3(U  uv) + pk
peo(u  uv) + pu
(29a, b)
where
(T13U1
+ T23U2 + T33U3)k  q  qR 
L Pshsvs .
(30)
s=l
It may be noticed that the thermal effect of species diffusion is still present in the
global energy equation. However, when a local chemical equilibrium calculation
based upon curvefit thermodynamic properties is performed, there is usually not
enough information available on the local composition to warrant an evaluation
of diffusion, which has to be neglected. This is not the case when the chemical
composition and flow simulation problems are coupled and solved simultaneously.
The closure to the mathematical problem of the governing equations for flows in
local chemical equilibrium will be provided again by thermal and caloric equations
of state, as shown in 2.2. With temperature and species composition known from
the equilibrium solver, the previous equations are tantamount to a general equation
of state relating pressure to density and internal energy
p = pep, e).
(31)
t),
TJ =TJ(Xl, X2, X3, t),
(=((Xl, X2, X3, t),
(33)
W=J'
and
(35a,b)
G  Gv
IV'TJI[ _
=J
V'TJ (8 
IV( I
8v)
_
+ 1JtQ] ,
H  Hv =j[V(. (8  8 v ) + (tQ].
(36a,b,c)
(37)
'\It,
and
'\lit, '\l( are the unit vectors in the directions normal to the coordinate
surfaces
(38a,b,c)
Moreover, tt, itt, and (t are the normalized time derivatives of the transformation,
that is, the time derivatives divided by the magnitude of the gradient in their respective direction.
Denoting the components of the velocity in generalized coordinates, also known
as the contravariant components, as follows
(39)
F and Fv
become
PN U 1
PNVN,
+ tXIP
PU1 U2 + {X2P
PU1 U3 + ~X3P
PU1 U1
P1en , u1
PMe nM U1
pe OU1
+ pur
qnl,1 
PIe n, VI,
qnM,1  PMenMVM,
, (40a,b)
173
where
In the above formulas, Vs stands for the species diffusion velocity written in generalized coordinates, whose components are obtained by a formula similar to Equation 39, but without the time derivative contributions. Similarly, the first component
of the generalized heat transfer vectors q, qR, and q~ are obtained by means of the
expresslOn
(42)
valid for the generic vector
of reference.
f,
Similar results apply for the generalized inviscid and viscous flux vectors in
the other two space directions, when is replaced by r, and ( respectively, and the
proper components of it, V., q, qR, and in are selected. The expressions that are
valid for local chemical equilibrium can be obtained in an entirely similar way. The
final result for inviscid and viscous vectors F and Fv reads
, (43a, b)
where
e=
iiI  iff 
(44)
s=l
The other two space directions are similar, once the proper components of velocity
and heatflux vectors are selected.
A very interesting analogy can be drawn between the governing equations written in generalized coordinates and those written in integral form for an infinitesimal
control volume 55 . Inspection of Equations 3438 reveals that the inverse of the J acobian of the transformation between Cartesian and generalized coordinates, 1/ J,
is geometrically equivalent to an infinitesimal volume. Moreover, the term /V k 1/ J,
where k stands for the generic generalized coordinate (~, 'T], or 0, is geometrically
equivalent to an infinitesimal area in the direction k. Consequently, the three space
partial derivatives can be interpreted as the firstorder difference of fluxes in the ~, 'T],
and ( direction, respectively, which would account for the flux across the six faces of
an infinitesimal prism. Moreover, the normalized time derivatives of the generalized
coordinates, et, r,t, and (t respectively, are equivalent to components of the boundary velocity. Overall, the differential equations written in generalized coordinates
.. _
T'J 
J.L
(Guo.
GXj
GUO j
GXi
~ (GUO l
3 J.L Gxl
GUo2
GX2
GU03)8
GX3 'J
i,j=1,2,3,
(45)
where J.L is the viscosity coefficient and the symbol 8ij stands for the Kroneker delta.
For mixture of gases in thermal equilibrium, the heat flux vector q is modeled
by means of the product of a thermal conductivity coefficient k times the temperature gradient (Fourier's Law). The extension of this approach to flows in thermal
nonequilibrium is usually done by considering similar contributions from the nonequilibrium temperatures 21 ,37,39. The resulting expressions read
= kVT,
s = 1, ... ,M.
(46a, b)
However, it should be pointed out that the validity of Fourier's Law for nonequilibrium conditions has not been fully assessed to date. The extension of this
model to Full NonEquilibrium is usually done by introducing more contributions
from the electron/electronic temperature 21 ,28. More details are given by Lee l .
175
The radiative heatflux vector qR is modeled taking into account emission, absorption, and induced emission by means of a quasiequilibrium approximation 2
A quasionedimensional "slab" theory is used to simplify the analysis of the complicated integrals that ensue 59 . Modeling radiative heat transfer phenomena in
the context of thermochemical nonequilibrium is very much a subject of current
research 60 , especially because flows at Mach number in excess of 25 in general, and
atmospheric reentry conditions in particular, are such that convection and diffusion
are secondary forms of heat transfer when compared with radiation 2
The previous expressions involved quantities that are expressed in Cartesian
coordinates. When generalized coordinates are used, the space derivatives 8()/8x;,
for i = 1,2,3, need to be expanded according to the chain rule
(47a, b, c)
For many practical applications, one can make the fairly usual approximation of neglecting viscous contributions except in one space direction, normal to a
solid surface. The resulting equations are known as the ThinLayer NavierStokes
equations 53 . The viscous flux vectors of Equations 24b and 2gb simplify because
only one component of each vector entry is nonzero. Moreover, when generalized
coordinates are used, the previous derivative expansions, Equations 47, reduce to
the following
where ~ has been considered to be the direction normal to the solid surface. Most
viscous computations performed in conjunction with thermal nonequilibrium are
actually ThinLayer calculations, due to the excessive fine computational meshes
that would be necessary to actually resolve the crossderivative terms that are neglected in the ThinLayer approach.
More details on the modeling of viscous terms, and in particular on the treatment of diffusion velocities, will be discussed in 5.5.
177
when performing some of the simplifications, because in a few instances the governing equation might lose its invariance to an arbitrary Gallilean transformation 32 .
For cases where only one common vibrational/electronic temperature is employed, the treatment of electronic nonequilibrium reverts back essentially to what
has been briefly described in the previous subsection, and only one nonequilibrium
energy equation is present in the system given in Equation 22. The source terms
will then include nonelastic phenomena and vibration/electronic relaxation to the
equilibrium translational temperature. Kinetic energy and partial pressure of free
electrons are usually neglected 28 .
5.3 Rotational NonEquilibrium
Many practical applications of hypersonic theory involve reentry flow problems.
Depending upon vehicle size and speed, there is an altitude regime in the atmosphere
where the NavierStokes equations cease to be accurate and the Burnett equations
become necessary for the simulation of thick shock waves 40 . In these cases, large
deviations from rotational equilibrium can occur 32 . Several investigators in recent
years have tried to model rotational nonequilibrium in the context of the continuum
hypothesis 41 ,62. Eppard and Grossman 41 present a simplified model for rotational
nonequilibrium, based upon the introduction of bulk viscosity, but show that overall
better results are obtained when the structure of shock waves is investigated by
means of a rotational nonequilibrium model. It may be useful to point out that
their nonequilibrium model can be represented as a special case of the Simplified
NonEquilibrium Thermodynamic Model, and consequently the algorithms described
in the present study will be applicable to cases with rotational nonequilibrium.
5.4 Body Forces
Contributions due to body forces 2::s Ps9s appear in both momentum and energy equations, as shown in the vectors of source terms W of Equations 23b and 28b.
However, gravity is negligible for hypersonic applications, and the only physical
phenomena that are sometimes modeled are the forces due to both selfinduced
and externally applied electric and magnetic fields 1 . These contributions become
important for ionized flowfield, as is the case in reentry applications. The electric
field is usually related to the gradient of the free electrons 37 . A few studies of the
interaction of magnetic fields with flowfields appear in the literature 63 . In general,
however, the consideration of electric and magnetic fields as part of the gasdynamic
source terms is complicated by the necessity of solving the Maxwell equations for
their determination. Again, this is subject of ongoing research efforts.
5.5 Transport Properties
The vector of viscous fluxes has been introduced in Equations 24b and 29b,
complemented by the expressions for the stress tensor components and the heatflux vectors, Equations 45 and 46. It makes use of several transport coefficients
that have yet to be determined, namely mixture viscosity coefficient fl, mixture
179
the gasdynamic equations. The resulting algorithm has been obtained for a perfect
gas, however it is a direct extension of a Roetype scheme of the kind that will be
discussed in the next section, and seems to be readily applicable to nonequilibrium,
reacting flows.
6. NUMERICAL FORMULATION
In order for a computer simulation to be possible, the governing equations presented in 4 must be discreiized, that is reduced to a set of algebraic equations to be
solved. A very convenient discretization approach is the finitevolume technique 55 ,
whereby the integral form of the governing equations is solved for the unknown volume averages of conserved variables in some small, but finite, control volume. The
advantage of this method is its use of the integral form of the equations, which allows
a correct treatment of discontinuities and also consistently treats general grid topologies. Other discretizations are also possible, like spectral techniques 68 , which have
been successfully applied to flow simulations in a chemically active environment 69 ,
and the more "standard" finitedifference approach 53 . Both of these alternative
discretizations are based upon the differential form of the equations, and special attention has to be paid to the proper treatment of discontinuities to avoid numerical
problems.
The governing equations in differential form have been written in conservation
form. Mathematically, this means that the coefficients of the derivative terms are
constant or, if variable, their derivatives do not appear in the equations. Moreover,
the divergence of some physical quantity can usually be identified in the equations.
This formulation generally leads to fewer numerical problems in the discretized version for flows with discontinuities. Related to the conservation form of the governing equations is the conservative property of a discrete algorithm 53 . Physically, this
means that the conservation principle that was used for the derivation of the equations is enforced exactly over volumes of arbitrary size within the domain of interest.
The finitevolume approach has the conservative property. Discretizing the integral
form of the governing equations with a conservative scheme yields the possibility of
capturing shocks and other discontinuities. Shockfitting techniques have also been
attempted for flows out of chemical equilibrium 7072, but their implementation is
at a relatively initial stage for general thermochemical nonequilibrium situations.
6.1 FiniteVolume Technique
The finitevolume technique takes advantage of the integral form of the governing equations to achieve a discretized algebraic approximation that can be handled
by a digital computer. The physical domain of interest is partitioned in a (large)
set of subdomains. In the following, the nodecentered approach will be described,
whereby the volume and surface integrals that appear in the governing equations
are approximated in terms of the unknown volumeaveraged values of each computational cell. This technique can be applied to arbitrarily shaped volumes, but
and the Jacobian J of the coordinate transformation was defined in Equation 37.
Substituting into the governing equations in integral form, Equation 22, yields
8( <~; V) +
is
(51)
For the case of a structured mesh, the computational cell will be defined by two
= constant, two T/ = constant, and two ( = constant surfaces, respectively. Associating indices i, j, k to the ~, T/, ( coordinates, respectively, the arbitrary volume
under consideration will be described by the indices ij k and the surface integral will
be split into six contributions, one from each delimiting surface. Then areaaveraged
values of a generic quantity f can be introduced
~
(52)
with similar definitions for the other indices, with the area Si+1/2 given by
(53)
a( <Q> V)
at
181
+F  FVi+l/2Si+l/2  F  FVi_l/2Sil/2+
+G  GVj+l/ZSj+l/Z  G  GVj_l/ZSjl/Z+
+H  HVk+l/2Sk+1/2  H  HVk_l/2Skl/2
=<w> V,
(54)
where the orientation of the surfaces has been taken into account for the six contributions to the surface integral. The vectors F, Fv , have been defined in Equations 40
and 43, and their counterparts in the other directions are similar. In the above, the
components of the boundary velocity Uv in the ~direction appears in the expression
for F, Equations 40a and 43a, as [to It is important to point out that Equation 54
is still exact, because volume and areaaveraged values have not been approximated.
However, at this stage it is necessary to relate the areaaveraged values to the unknown volumeaveraged values. The resulting set of algebraic equations will be the
discretized form of the original problem that is actually solved by means of a digital
computer.
An interesting analogy 55 between the finitevolume formulation discussed so far
and a more standard finitedifference technique can be obtained when the volume V
is written in terms of an average Jacobian and average increments in the coordinate
directions
(55)
Similarly, the areas can be written as follows
(56)
with corresponding results for the other terms.
Substituting Equations 55 and 56 into Equation 54, dividing by
dropping the average notation yields
aat (Q)
J
1 [ IV~I)
I ~~
(J
i+l/2(F  F v )i+l/2 
IV~I
( J )i_l/2(F 
b.U~l.''1b.(
and
F v )il/2 +
1 [ IV1]I)
 IV1]1
 ]
+ b.1]
( J j+l/2(G  G V)j+l/2  (J) j_l/2(G  G v )jl/2 +
1 [ IV'(I
 IV'(I
 ] w
+ b.( (J) k+l/2(H  H v h+l/2  (J) kl/2(H  H v hl/2 = J.
(57)
Equation 57 is the discretized form of the governing differential equations written
in generalized coordinates, Equation 34, that would have been obtained using a
(58)
with similar results for the other indices. In the previous formulas, I is the identity
operator, \7, t:. are backward and forward difference operators, respectively, and
q, q+ are left and right extrapolations. The bars in the difference operators
indicate the presence of a limiter for the higherorder corrections. A review of
commonly used limiters and their properties is given by Sweby 73. The parameter
K, determines the extrapolation scheme used. A value K, = 1/3 yields a thirdorder
upwindbiased technique, a value K, = 1 produces the secondorder upwind method,
whereas K, = 1 gives the classic secondorder centered scheme.
Two different techniques have emerged in recent years for the treatment of the
inviscid flux vector. The first one identifies q with the flux vector itself. Consequently, a vector of inviscid fluxes is created from volumeaveraged values of the
variables in a cell, and subsequently face values are created by extrapolation. The
second method, called MUSCL extrapolation, identifies q with the vector of conserved variables. The extrapolation produces face values for the conserved variables,
then the fluxes are constructed from the functional relationships
(59)
The latter technique is claimed to be more accurate than the former, at least for
perfectgas flows 74 . It is useful to point out that for many applications the extrapolation of primitive or even characteristic variables is preferred to the extrapolation
of conserved variables, Q. Numerical experiments have shown that with this procedure more robust codes are obtained for highspeed flows in the case of both perfect
gas and reacting flows 8 . Specific techniques for the discretization of the inviscid flux
vectors will be discussed in 7.
The viscous vectors are usually discretized using standard secondorder accurate
central differences for the second derivatives, whereas crossderivative terms may be
treated in a diagonaldominancepreserving fashion 75. Using central differences for
the inviscid vectors (K, = 1) yields unstable schemes, unless artificial dissipation is
added.
7. FLUXSPLIT ALGORITHMS
In recent years, fluxsplit techniques have obtained wide acceptance as an accurate means of discretizing the inviscid fluxes. Originally developed for a perfect
gas 9  13 , they have been extended to flows in chemical equilibrium 14  2o , and to
mixtures out of chemical and thermal equilibrium 21  27 . They are found to be very
accurate and robust when used for transonic, supersonic and hypersonic flows 74 , and
are fully compatible with conservative finitevolume, shockcapturing approaches.
Probably the two most popular schemes in the fluxvector splitting category
are the ones due to Steger and Warming l l and to Van Leer12. The basic idea is
to split the inviscid flux vector in one space dimension in two parts, containing the
information that propagates downstream and upstream, respectively. Then the two
parts are constructed using the extrapolation formulas of Equation 58 consistently
with the direction of propagation. When the MUSCL approach is used the result is
(60a,b,c)
where the downstream propagating flux P+ uses the left extrapolation Q and
the upstream propagating flux P the right extrapolation Q+. Alternatively, positive and negative fluxes are extrapolated directly using the negative and positive
extrapolation formulas, respectively.
Another very popular scheme, less robust for hypersonic flows 74 but more accurate, is the fluxdifference splitting technique due to Roe 13 . It consists of an
approximate Riemann solver, where an arbitrary discontinuity is supposed to exist at the cell surface between the left and the right state, and an approximate
solution for this situation is written in terms of waves propagating upstream and
downstream. In this instance, the inviscid flux vector is not split, but reconstructed
from the upstream and downstream contributions, that constitute the left and the
right states in the Riemann problem.
Extensions to more space directions are usually made by superimposing pseudoonedimensional problems, where the extrapolation formulas to get left and right
states keep track of the relevant direction only55. In recent times, a lot of effort
has been spent towards the design of "truly" multidimensional algorithms, which
would be at least approximately independent of the grid orientation 81 ,82. It is not
clear yet what the computational cost of the possible increase in accuracy would be,
and the topic is still subject of current research. It should be mentioned, however,
The algorithms to be presented do not exhaust the category of fluxsplit techniques. Oshertype formulations exist for perfect gases, and their extension to reacting flows has been attempted 84 . More recently, new kinds of flux splitting are
emerging85 , whereby the treatment of the pressure gradient plays a central role, and
they seem to show promise of improvement over the "older" schemes. Again, the
original development is for perfect gases.
F =
AQ =
F+
+ F ,
(61a, b)
with similar results for the other space dimensions. The final result for the positive
187
PI
1'1
=_AA
PN
PUI
PU2
PU3
Ple n,
PMe nM
ph o pa 2/(i'l)
PI
PN
PI
A
p(u2+tx2 a)
B
+ 21' p(u3+txa a)
ple n,
PN
p(Ul~Xla)
A
c p(u2tx2 a)
+ 21' p(u3txa a)
Ple n,
PMe nM
p(h o + u~a)
PMe nM
p(ho u~a)
p(Ul+~Xla)
(62)
where the nonnegative and nonpositive eigenvalues of the Jacobian matrix are
given by the following
.
(63a, b, c)
It is interesting to point out that simpler forms of the algorithm, including thermodynamic equilibrium and perfectgas versions, can be recovered by simplifying the
equations in a fashion consistent with that discussed in 2.
C
(64)
where the isentropic index f plays a central role in the splitting. The nonnegative
and nonpositive eigenvalues are given by Equations 63, where the speed of sound
is the equilibrium value, defined in Equation 9, and f3 is equal to
(f  a)
f3 = f(fl) ,
(65)
where the parameter C\' is arbitrary. The value C\' = 0 allows this scheme to reduce
nicely to the perfectgas form, and to be formally similar to its nonequilibrium
counterpart when the ratio of frozen specific heats l' is substituted by the isentropic
index r.
Other formulations have appeared in the scientific literature. Grossman and
Walters20 utilize the ratio of enthalpy to internal energy "(* to split the flux, and
approximate the speed of sound by substituting the same ratio for the isentropic
index, yielding (a*? = "(*p/ p. The final algorithm is formally identical to the one
shown in Equation 64 when C\' = 0 and "(* is substituted for r everywhere. Liou,
Van Leer, and Shuen 16 split only a portion of the fluxes, and add to both positive
and negative contributions one half of the unsplit portion. Again, their splitting
is not based upon the "true" sound speed. It may be useful to point out that no
conclusive evidence has yet been presented as to the superiority of one approach to
any other.
p = IV'~I f
J
PN/p
Plenl/p
(67)
189
where
f e
 h0
t (iiI _ 2a)
c.,t
m (Ul
a )2 ,
(68)
"(
and m is an arbitrary parameter. The value m = 1/(1'+1) allows the present scheme
to reduce nicely to the "classic" perfectgas version, but numerical experiments
do not show any significant effect on the value of m. In the supersonic range,
p = P, p'f' == 0, when Ml > 1 and Ml < 1, respectively. Different authors
have developed very similar schemes 22  24 . The other space dimensions are treated
in a similar fashion.
7.4 Van LeerType AlgorithmLocal Chemical Equilibrium
The derivation of a Van leertype algorithm for flows in local chemical equilibrium has been successfully performed by several authors 15 ,16,19,20. The final result
given by Liou, Van Leer, and Shuen16 reads
(69)
where I;:' and I!' are given by Equations 66 and 68 respectively. The results presented by other authors differ in the choice of the parameter m in Equation 68,
although no strong argument has been provided in favour of any specific value.
Grossman and Walters20 develop a similar splitting, but again the enthalpy to iTlternal energy ratio "(* has to be substituted for the isentropic index r in all of the
expressions given, including the definition of an "approximate" speed of sound.
(70)
ai,
such that
N+M+4
[Q]
N+M+4
aiEi,
[F] =
;=1
ai ~i Ei ,
(71a,b)
i=l
for cell interface states which are not necessarily close to each other, so that [Q]
is arbitrary. In the above, a Cartesian component of the flux vector has been
considered, along with the eigensystem associated with its Jacobian matrix. The
results obtained will be extended to generalized coordinates by standard means 55 .
For this case, it is possible to define the eigenvalues
i = 1, ... ,N+M+2,
i = N+M+3,
(72)
i=N+M+4,
and the eigenvectors
Pi
o
o
o
o
Pi U1
Pi U2
Pi U3
, i = 1, ... , N,
Ej+N =
, j = 1, ... ,M,
(u~ ~i)Pi
(73a, b, c, d, e)
0
ih
P2
E N +M+1=
0
0
1
0
0
0
0
0
1
0
EN+M+2 =
U2
U3
PN
U1
U2
U3
en!
, k
= 3,4,
Cl'i
Cl'j+N
Pi 
[P]
0,2 '
[P]
[pje nj ]
A
e nj
~,
i = 1, ... ,N,
(74a)
j = 1, ... ,M,
(74b)
O:N+M+l = p[U2],
(74c)
O:N+M+2 = p[U3] ,
(74d)
k = 3,4.
(74e)
Pi =
p'
(;:)
e ni =
(P;;:'i)
P ,
(75a,b)
have been used, and 'ljJi represents pressure derivatives taken with respect to the ith
species density and keeping the other conservative variables constant, according to
the formula
(76)
It is noteworthy that these averages are not unique. As pointed out by Abgra1l 25 ,
the algebraic problem posed in Equations 71 has multiple solutions, and different
values have been published in the literature for some of the Roetype averages 22  27
The major differences between the different approaches is in the evaluation of averages of the pressure derivatives, or equivalently of ,(J;i, whereas the same results
are obtained for mass fractions and nonequilibrium contributions, if present. No
numerical evidence has been presented of the superiority of one scheme over the
others.
Following the derivation of Grossman and Cinnella27, and using the notation
(77)
for the arithmetic average, the necessary averages are determined to be
p =<.JP> 2  ~[.JPt
(78a)
= .jPrPl,
U 
(78b)
~~~~~~
= 1, ... ,N,
(78c)
j = 1, ... ,M,
(7Sd)
(7Se)
(7Sf)
(7Sg)
T = _<_T,v'P_p_>
(79a)
<yIp> '
N
(79b, c)
i = 1, ... , N,
(79d)
(7ge)
(79f)
In Equation 79c, the integral averages of the species frozen heats at constant volume
C~i have been introduced, which in general will have to be numerically determined
with any of the available quadrature formulas. For cases where the simple trapezoidal rule is employed, it is easy to recognize that C~i reduces to the arithmetic
average of right and left values. Moreover, it may be noticed that all of the above
definitions reduce to the "usual" Roeaverages for a perfect gas model.
Rewriting Equation 71 b by grouping the repeated eigenvalues, simplifying the
result and transforming it to generalized coordinates 55 yields
(SO)
The [F]A term arises from the first N +M +2 terms of the sum in Equation 71b,
[pN I p]
[Ul]  txJUl]
[U2]  tX2 [Ul]
[U3]  tX3[Ul]
[PI en! I p]
e
lt~  a2 /erI)
nM
(81 )
where
and Ul, Ul are trivially defined starting from the averages of the Cartesian components Ul, U2, {h
"*
'"
""
Ul =~XI Ul + ~X2 U2
"
. . . * ""
Ul =U l + ~t.
'"
+ ~X3 U3 ,
(83a,b)
The [Fh and [F]c contributions arise from the last two terms in the summation
of Equation 71b, corresponding to the eigenvalues iiI +a and iiI a, and are found
to be
PN
Ul
tXI a
U2 tX2a
U3 tX3a
en!
(84)
(85)
,*.
a2
(op)
op
+h (
pe
op )
ape
(87)
and the definition of the Roeaveraged speed of sound will be affected by the choice
made.
Following Vinokur 18 , and using Cartesian coordinates, the starting point for
the construction of the approximate Riemann solver is again the determination of
195
(88)
eigenvectors
[p] 
[~] ,
(90a)
P[U2] ,
= P[U3] ,
(90b)
(90c)
0,2 =
0,3
o,k =
2~2
([P] pa[ud) ,
k = 4,5,
(90d)
where the partial derivatives of pressure with respect to internal energy per unit
volume, 8pj8pe == K, and with respect to density, 8pj8p == X, have been introduced.
The solution of the approximate Riemann problem involves determining algebraic averages p, it, ho, X, r;" a such that Equations 71 are satisfied. The first three
averages are formally unchanged from the previous development, with density being
averaged geometrically and velocity and total enthalpy being defined by means of
"standard" Roeaverages. Vinokur1B suggests relatively involved formulas for the
pressure derivative averages, and determines the average speed of sound as follows
a2 = X+ (ho 
'2
~O)r;,.
(91)
Rewriting Equation 71b by grouping the repeated eigenvalues, simplifying the result
and transforming it to generalized coordinates55 yields Equation 80 again, where the
[F]A term arises from the first three terms of the sum in Equation 71b, corresponding to the repeated eigenvalue Ai = 5: 1 , and may be written as
(93)
and 5. 1 , 5.; have been defined in Equations 83. The [Ph and [P]c contributions
arise from the last two terms in the summation of Equation 71 b, corresponding to
the eigenvalues 5. 1 +a and 5. 1 a, and are found to be
1
{Xl a
U2 (X2 a
U3 ~X3a
,
,*
ho Ul a
Ul
(94)
_(8P)

8p
a2 
_
hhl)
,
pe
8p )
K,
== ( 8pe
_
= 'Y  1.
(95a, b)
ACKNOWLEDG EMENTS
This work was supported in part by the US Air Force, Eglin AFB, Florida, under
Contract Number F0863589C0208, in part by the National Science Foundation,
which funds the Engineering Research Center for Computational Field Simulation,
and in part by NASA Langley Research Center, under Grant NAGI776.
REFERENCES
1. Lee, J .H. Basic Governing Equations for the Flight Regimes of Aeroassisted
Orbital Transfer Vehicles, Progress in Astronautics and Aeronautics, Vo1.96,
pp. 353, 1985.
2. Anderson, J.D.Jr. Hypersonic and High Temperature Gas Dynamics, McGrawHill, New York, 1989.
3. Palaniswamy, S., Chakravarthy, S.R. and Ota, D.K. Finite Rate Chemistry for
USASeries Codes: Formulation and Applications, AIAA Paper No. 890200,
1989.
4. Hoffman, J.J. Development of an Algorithm for the ThreeDimensional FullyCoupled Navier Stokes Equations with Finite Rate Chemistry, AIAA Paper
No. 890670, 1989.
5. Drummond, J.P., Rogers, R.C. and Hussaini, M.Y. A Detailed Numerical Model
of a Supersonic Reacting Mixing Layer, AIAA Paper No. 861427, 1986.
6. Molvik, G.A. and Merkle, C.L. A Set of Strongly Coupled, Upwind Algorithms
for Computing Flows in Chemical Nonequilibrium, AIAA Paper No. 890199,
1989.
7. Gnoffo, P.A. A Code Calibration Program in Support of the Aeroassist Flight
Experiment, AlA A Paper No. 891673, 1989.
8. Walters, R.W., Cinnella, P., Slack, D.C. and Halt, D. Characteristic Based
Algorithms for Flows in ThermoChemical Nonequilibrium, AIAA Paper No.
900393, 1990.
9. Harten, A., Lax, P.D. and Van Leer, B. On Upstream Differencing and GodunovType Schemes for Hyperbolic Conservation Laws, SIAM Review, Vo1.25, No.1,
pp. 3561, 1983.
10. Roe, P.L. CharacteristicBased Schemes for the Euler Equations, Annual Review of Fluid Mechanics, VoLl8, pp. 337365, 1986.
11. Steger, J.L. and Warming, R.F. Flux Vector Splitting of the Inviscid Gasdynamic Equations with Applications to FiniteDifference Methods, Journal of
Computational Physics, Vo1.40, pp. 263293, 1981.
12. Van Leer, B. FluxVector Splitting for the Euler Equations. Lecture Notes in
Physics, VoLl70, pp. 507512, SpringerVerlag, Berlin and New York, 1982.
13. Roe, P.L. Approximate Riemann Solvers, Parameter Vectors, and Difference
Schemes, Journal of Computational Physics, Vo1.43, pp. 357372, 1981.
14. Colella, P. and Glaz, H.M. Efficient Solution Algorithms for the Riemann Problem for Real Gases, Journal of Computational Physics, Vo1.59, pp. 264289,
1985.
199
1989.
32. Grossman, B., Cinnella, P. and Eppard, W.M. Algorithms for NonEquilibrium
Hypersonic Flows, in Fourth International Symposium on Computational Fluid
Dynamics, A Collection of Technical Papers, VoLl, pp. 443448, University of
California, Davis, 1991.
33. Cinnella, P. and Grossman, B. Upwind Techniques for Flows with Multiple
Translational Temperatures, AIAA Paper No. 901660, 1990.
34. Cinnella, P. and Cox, C.F. An Efficient "Black Box" Solver for the Equilibrium
Composition of Real Gases, AIAA Paper No. 913322, 1991.
35. Vincenti, W.G. and Kruger, C.H.Jr. Introduction to Physical Gas Dynamics,
Robert E. Krieger, Malabar, 1986.
36. Clarke, J.F. and McChesney, M. The Dynamics of Real Gases, Butterworth
& Co. Ltd., London and Boston, 1964 (second edition under the title The
Dynamics of Relaxing Gases, 1975).
37. Park, C. Nonequilibrium Hypersonic Aerothermodynamics, John Wiley & Sons,
New York, 1991.
38. McBride, B.J., Heimel, S., Ehlers, J.G. and Gordon, S. Thermodynamic Properties to 6000 K for 210 Substances Involving the First 18 Elements, NASA
SP3001, 1963.
39. Gupta, RN., Yos, J.M., Thompson, RA. and Lee, K.P. A Review of Reaction
Rates and Thermodynamic and Transport Properties for an IISpecies Air
Model for Chemical and Thermal Nonequilibrium Calculations to 30000 K,
NASA RP1232, 1990.
40. Fiscko, K.A. and Chapman, D.R Hypersonic Shock Structure with Burnett
Terms in the Viscous Stress and Heat Flux, AIAA Paper No. 882733, 1988.
41. Eppard, W.M. and Grossman, B. Calculation of Hypersonic Shock Structure
Using FluxSplit Algorithms, ICAM Report No. 910704, Virginia Polytechnic
Institute and State University, Blacksburg, 1991.
42. Treanor, C.E. and Marrone, P.V. Effect of Dissociation on the Rate of Vibrational Relaxation, The Physics of Fluids, Vo1.5, No.9, pp. 10221026,1962.
43. Marrone, P.V. and Treanor, C.E. Chemical Relaxation with Preferential Dissociation from Excited Vibrational Levels, The Physics of Fluids, Vo1.6, No.9,
pp. 12151221, 1963.
44. Candler, G. TranslationVibrationDissociation Coupling in Nonequilibrium Hypersonic Flows, AIAA Paper No. 891739, 1989.
45. Sharma, S.P. and Schwenke, D.W. The Rate Parameters for Coupled RotationVibrationDissociation Phenomenon in H 2 , AIAA Paper No. 891738, 1989.
46. Park, C. A Review of Reaction Rates in High Temperature Air, AIAA Paper
No. 891740, 1989.
47. Bruno, C. Real Gas Effects, Hypersonics (Eds. Bertin, J.J., Glowinski, Rand
Periaux, J.), VoLl, pp. 303354, Birkhiiuser, Boston, 1989.
48. Oran, E.S. and Boris, J.P. Numerical Simulation of Reactive Flow, Elsevier,
Amsterdam, 1987.
66. Baysal, 0., Engelund, W.C. and Tatum, K.E. NavierStokes Calculations of
ScramjetAfterbody Flowfields, presented at Symposium on Advances and Applications in Computational Fluid Dynamics, Chicago, 1988.
67. Morrison, J. Flux Difference Split Scheme For Turbulent Transport Equations,
AIAA Paper No. 905251, 1990.
68. Drummond, J.P., Hussaini, M.J. and Zang, T.A., Spectral Methods for Modeling Supersonic Chemically Reacting Flowfields, AlA A Journal, Vo1.24, No.9,
pp. 14611467,1986.
69. Eklund, D.R., Drummond, J.P. and Hassan, H.A. The Efficient Calculation of
Chemically Reacting Flow, AIAA Paper No. 860563, 1986.
70. Aftosmis, M.J. and Baron, J.R. Adaptive Grid Embedding in Nonequilibrium
Hypersonic Flow, AIAA Paper No. 891652, 1989.
71. Botta, N., Pandolfi, M. and Germano, M. Nonequilibrium Reacting Hypersonic
Flow about Blunt Bodies: Numerical Predictions, AIAA Paper No. 880514,
1988.
72. Valorani, M., Onofri, M. and Favini, B. On the Solution of Nonequilibrium
Hypersonic Inviscid Steady Flows, AIAA Paper No. 890671, 1989.
73. Sweby, P.K. High Resolution Schemes Using Flux Limiters for Hyperbolic Conservation Laws, SIAM Journal of Numerical Analysis, Vo1.21, No.5, pp. 9951011, 1984.
74. Van Leer, B., Thomas, J.L., Roe, P.L. and Newsome, R.W. A Comparison of
Numerical Flux Formulas for the Euler and NavierStokes Equations, AIAA
Paper No. 87ll04CP, 1987.
75. Chakravarthy, S.R., Szema, K.Y., Goldberg, U.C., Gorski, J ..L and Osher, S.
Application of a New Class of High Accuracy TVD Schemes to the NavierStokes Equations, AIAA Paper No. 850165, 1985.
76. Fletcher, C.A.J. Computational Techniques for Fluid Dynamics, VoLl, SpringerVerlag, Berlin and New York, 1988.
77. Bussing, T.R.A. and Murman, E.M. A FiniteVolume Method for the Calculation of Compressible Chemically Reacting Flows, AIAA Paper No. 850331,
1985.
78. Iannelli, G.S. and Baker, A.J. A StifflyStable Implicit RungeKutta Algorithm
for CFD Applications, AIAA Paper No. 880416, 1988.
79. Barth, T.J. Analysis of Implicit Local Linearization Techniques for Upwind and
TVD Algorithms, AIAA Paper No. 870595, 1987.
80. Liou, M.S. and Van Leer, B. Choice of Implicit and Explicit Operators for the
Upwind Difference Method, NASA TM 100857, 1988.
81. Hirsch, C. and Lacor, C. Upwind Algorithms Based on a Diagonalization of the
Multidimensional Euler Equations, AIAA Paper No. 891958, 1989.
82. Dadone A. and Grossman, B. A Rotated Upwind Scheme for the Euler Equations, AIAA Paper No. 910635, 1991.
83. Wilson, G.J. and MacCormack, R.W. Modeling Supersonic Combustion Using
a FullyImplicit Numerical Method, AlA A Paper No. 902307.
  
Chapter 6:
Efficient Multigrid Computation of Steady
Hypersonic Flows
B. Koren, P. W. Hemker
Center for Mathematics and Computer Science,
P.O. Box 4079, 1009 AB Amsterdam, The Netherlands
Note: This work was supported by the European Space Agency (ESA), through Avions Marcel Dassault  Breguet Aviation (AMDBA).
ABSTRACT
In steady hypersonic flow computations, Newton iteration as a local relaxation procedure and nonlinear multigrid iteration as an acceleration procedure may both easily fail. In the present chapter,
some remedies are presented for overcoming these problems. The equations considered are the
steady, twodimensional NavierStokes equations. The equations are discretized by an upwind finite
volume method.
Collective point GaussSeidel relaxation is applied as the standard smoothing technique. In
hypersonics this technique easily diverges. First, collective line GaussSeidel relaxation is applied as
an alternative smoothing technique. Though promising, it also fails in hypersonics. Next, collective
point GaussSeidel relaxation is reconsidered and improved; a divergence monitor is introduced and
in case of divergence a switch is made to a local explicit time stepping technique. Satisfactory singlegrid convergence results are shown for the computation of a hypersonic reentry flow around a blunt
forebody with canopy.
Unfortunately, with this improved smoothing technique, standard nonlinear multigrid iteration
still fails in hypersonics. The robustness improvements made therefore to the standard nonlinear multigrid method are a local damping of the restricted defect, a global upwind prolongation of the correction and a global upwind restriction of the defect. Satisfactory multigrid convergence results are
shown for the computation of a hypersonic launch and reentry flow around a blunt forebody with
canopy. For the test cases considered, it appears that the improved multigrid method performs
significantly better than a standard nonlinear multigrid method. For the test cases considered it
appears that the most significant improvement comes from the upwind prolongation, rather than from
the upwind restriction and the defect damping.
204
1. INTRODUCTION
pu
a
ax
pu 2 +p
puv
+~
ay
pu (e +..)
p
Re
aX
Tn
pv
pvu
pv 2 +p
pv(e+..)
p
Tn
=0,
(I a)
with
au 2 av
ax 3 ay'
au av
Tn= ay +~,
4 av
2 au
.=3 ay
3 ax'
4
3
T=xx
(Ib)
II
e = _ _ .L.... + (u 2 +v 2 ).
yI p
2
( Ic)
For a detailed description of the various other quantities used, assumptions made and so on, we
refer to any standard textbook. Suffice it to say here that these are the full, steady, 20, compressible
NavierStokes equations with as main assumptions made: zero bulk viscosity and constant diffusion
coefficients. (So, the flow is assumed to be laminar and its diffusion coefficients are assumed to be
temperatureindependent.) For 1/ Re =0, diffusion has vanished and the remaining equations are the
Euler equations.
So far, real gas effects are not taken into account. The specific heat ratio y of the diatomic gas
considered is assumed to be constant and determined by fully excited translational and rotational
energies only. (Though it could easily be replaced by a function ranging from zero up to the full
equipartition value, the vibrational energy is assumed to be zero.)
1.2. Discretization method
For a description of the basic computational method which is taken as a point of departure, we refer
to Hemker and Spekreijse [1,2,3] and Koren [4,5,6]. Here we give a concise overview of the main
characteristics only. For both the Euler and NavierStokes equations, we have considered first and
higherorder accurate discretizations. Since in solving all types of equations our multigrid solution
methods are applied to the firstorder discretized equations only (higherorder discretized equations
are solved in an outer, singlegrid defect correction iteration), here we likewise limit the description of
the discretization to the firstorder accurate one only.
r,
r,
'0
For / = 1, the coarse grid correction is skipped of course. For the restriction operator Ii I and
the prolongation operator Ii  1 we take
(r, I ),.; =
,
_ ;
,
,
(/'lq'lh,1.2;I(/'lq, lh,1.2;(/, Iq'lhr.2;I(/,

(2a)
Iq,
Ihr.2;(q'I),,j"
(2b)
If not mentioned otherwise, for n a' n pre and n pm' we use in the basic multigrid method at each
level I: n a = I and n pre = n P'b' = I; i.e. as nonlinear multigrid cycles we use Vcycles with a single preand postrelaxation sweep per level.
The prolongation operator for obtaining the first approximation on each next finer grid may be
the piecewise constant operator (2b) or  preferably  a more accurate operator (for instance a 'bilinear
operator).
2.3. Numerical results
To give a quick impression of the performance of our basic multigrid method, both outside and inside
the hypersonic flow regime, we consider the following Euler flows: (i) the NACA0012airfoil at
Moo =0.63, a=2 (smooth subsonic flow) and Moo =0.85, a= I (nonsmooth transonic flow), and
(ii) a blunt forebody with canopy at Moo =8.15. a=30 (nonsmooth hypersonic flow).
2.3.1. Subsonic and transonic airfoil flow. As finest finestgrid for the NACA0012airfoil we consider
the 128 X 32 Otype grid given in Figure 1. In all corresponding multigrid cases, as coarsest grid we
consider the corresponding 8 X 2 Otype grid. (Hence for the 128 X 32grid we have L = 5.)
For the present two airfoil flows the multigrid convergence histories are given in Figures 2a and
2b.
In both graphs, the residual ratio along the vertical axis is the ratio
Lt = II(N1Jq2k liLt = II(Ndq Lk I. L = 3.4. 5 versus the number of cycles performed; (i) one multigrid
cycle being a Vcycle with npre=npost=l, 't;fl, and (ii) for Q 5 only, one singlegrid cycle being the
equivalent number of finestgrid relaxation sweeps. I(NL(q1Jh I denotes the summation  over all
volumes at Q L  of the absolute values of the kth component in the firstorder Euler defects, with q2
denoting the solution at Q L after the nth multi or singlegrid cycle. Considering the corresponding
singlegrid convergence histories, for both nonhypersonic cases. the effectiveness of the multigrid
method appears to be good.
a. In full.
b. In detail.
10
cycleS
a, At Moo =0.63, a=2.
Figure 2. Convergence histories NACAOO 12airfoil at 1/ Re = 0
(  : singlegrid,   : multigrid).
2.3.2. Hypersonic blunt body flow. However, for the hypersonic blunt body flow the results appear to
be different. The forebody is composed out of two ellipse segments (Figure 3), given by
l l
~
0.06
_x_
[ 0.035
2+
1
L. 12  I
0.015
x<O,
2+ [L. J2 ~ I
0.025
J
(3a)
(3b)
0";;;;x";;;;0.016.
As finest finestgrid, we consider here the 64 X 32 Ctype grid given in Figure 4. As coarsest grid, the
corresponding 4 X 2grid is applied. (Hence, for this 64 X 32grid we also have L = 5.)
0,025  .  .    . 
'1
o
0,0\5  
T . ==._+:
i
I
0.06
0.016
I<l
~+..~
0.25
0.15
O.OS
0.05
x
Figure 4. 64X32grid (rl s) blunt forebody with canopy.
With this gridding, with the basic computational method described before and with an initial solution q I which is equal to the upstream farfield boundary conditions, not any flow solution could be
obtained. Already in the first relaxation sweep on the coarsest grid, the solution process broke down!
A new research topic was found: extension to hypersonics of the basic method's applicability.
10
rl
I
Ml
+
a. Viscous sublayer.
0'
ea';::l"'
.~
"0
'",
1]5
1]4
'l'
<1.l
=OIl
1]5
1]4
1]3
1]2
1],
.2
10
1]]
........~I];
~
+~''~6'~lDl],
10
work units
work units
work units
a. Point relaxation.
+~,
2.5
7.5
time (sec)
12.5
JD
20

30
time (sec)
so
80
120
time (sec)
c. 1]5'
J60
200
..~~
10
10
10
work units
work units
work units
a. Point relaxation.
Figure 10. Multigrid convergence histories for three types of GaussSeidel relaxation,
supersonic flat plate flow at M'l.) =2, Re=2.96 105.
An objection that can be made against the use of crosswise line relaxation throughout the computation is that though it is welladapted to the strong coupling in the viscous sub layer with its high
aspect ratio cells, it is not welladapted to the opposite coupling in the outer flow. (There, streamwise
lines are to be preferred.) The switch in direction of strong coupling suggests an adaptive local line
relaxation to be optimal (Figure 11).
In conclusion: for the supersonic Euler flow considered, streamwise line relaxation appears to be
most efficient. Already with a relatively slow solver for the large linear system, streamwise line relaxation may be more efficient than point relaxation. For NavierStokes flow computations with a practically relevant resolution of the viscous layers, crosswise line relaxation is to be preferred. Its advantage clearly is its greater robustness. It is less sensitive to a strong local illconditioning of the flow
equations.
3.1.3. Numerical results for hypersonic blunt body flow. Unfortunately, for the hypersonic blunt
body flow already considered in section 2.3.2, line GaussSeidel relaxation fails. In the next section we
return to point GaussSeidel relaxation and introduce a robustness improvement for it: a switch to
local, explicit time stepping.
_._


~~j
t~l.'~
x
Figure 11. Locally adapted relaxation lines.
J,
(4)
where T(</ denotes the matrix for rotation to a local coordinate system, F(ql, qr) the numerical flux
function (with its left and right cell face states q' and qr) and 1 the length of a finite volume wall. (For
further details we refer to Hemker and Spekreijse [1].)
3.2.1. Failing Newton iteration. As a nonfailing Newton iteration to solve q,., from Equation (4) we
.
define: a Newton iteration for which: (i)
1("J,./q7./
I)h I
iii'
/I
1
",;;,
IUi)qi)kl
k = 1,2,3,4,
'!;I"J'
(5)
for any nth Newton iterate (n =0, I, ... ,N) and each of the four residual components, and for which
I is physically correct, with physical correctness defined in the following way.
(ii) each iterate
Considering the local iterate q7./ =(u?j'
c?/, z;:)T and the corresponding hypersonic, upstream state
vector q 00 =(u Xl' v"" CXl' Z ",,)T, we know that the flow speed may not exceed the value corresponding
with adiabatic expansion to vacuum, starting from upstream conditions:
qU
<J'
'!;I,.}"
(6)
Further, we know that after this expansion, the speed of sound equals zero, its minimally allowable
value:
(7)
The maximally allowable value of the speed of sound is that corresponding with the stagnation temperature (which is the same for both isentropic and nonisentropic compression). For adiabatic flows
we can write:
/I C.
Ci'J~
2
ex
+ 2y I (U 2oo + V 2
OO
'
(8)
For the lower limit of Z;'j we can directly write with the entropy condition:
'!;I'f
(9)
For the upper limit of z?/ we have to consider the state q2 at the downstream side of a normal shock
wave which has at its upstream side a state q I which has expanded to vacuum, departing from
upstream conditions. Given the gas dynamics relations
P2=
2yMT (y I)
y+1
PI,
(y+ I)MT
P2= (yI)MT+2PI,
(lOa)
(lOb)
214
it is clear that
(11 )
Summarizing, we see that in adiabatic flows both the flow speed and the speed of sound have a physical lower and upper limit. The entropy only has a lower limit.
In the algorithm, the relations (5)(9) are checked after each update in each local Newton iteration. As soon as one or more of these five requirements are not satisfied, that local Newton iteration
is said to have failed and the corresponding local correction found is rejected.
3.2.2. Evolution technique. As the alternative for a failing Newton iteration, we apply next one or
eventually two explicit time stepping schemes to the local, semidiscrete system
aqi.; +_I_Gi' (
)0
A. Ji,j qi,i  .
at
( 12)
I';
with for qi,j here the conservative state vector (pl,j'(pu)"j'(PV)i.j,(pe)i,j)T and for A"j the area of finite
volume Qii'
As time stepping scheme to be applied first, we take the following version of the explicit, twostep
rational RungeKutta scheme of Wambecq [IOJ:
11+1_
qi.;
11
~Jfj(q7j)
T?j
"
I,'
( 13)
<,
with W a possible damping factor for which we initially take w= I, and with
the local time step for
which we safely take the one which is maximally allowed for the forward Euler scheme:
A",h ,.j
T;: i = [,.':d<ff;)q;:j)
2.''"''0;1 '
(14)
sup
dq',i
with h,) a characteristic local mesh size. With our upwind discretization, Equation (\4) may be rewritten by good approximation as
hi.;
T;:, = ,,
1(=u="=)=2=+::::::(V=I='=)2=+c"V
/./
/./
( IS)
1./
For the evaluation of the denominator in scheme (\ 3) we use the Samelson inverse of a vector:
\11
1~12~,
(\6)
being a vector, whereas for the norm of a vector, we simply use the Cartesian inner product. As ini
tial solution we take the same q7. i that just failed for the Newton iteration. The motivation for applying this scheme is its good stability as demonstrated in Wambecq [IOJ for a stiff and coupled system
of four equations, which is precisely what we have here in hypersonics. However, a potential danger
of scheme (\ 3) is that there is no guarantee for the denominator to be nonzero.
To protect Wambecq's scheme against a possibly too large time step and against a (nearly) zero
denominator, in each time step we require both the predictor and corrector to satisfy the conditions
(6)(9). As soon as a physically unrealistic value occurs, the time stepping is stopped immediately,
rejecting any update made. Then, at first we assume that the unphysical result is due to a too large
time step. Therefore, as a remedy, we take w= lJ:! and restart the time stepping with Wambecq's
scheme, using the same q))1' In case of reoCCurrence of something unphysical, we assume that the
denominator was the problem. Therefore, as a new remedy, we restart with an explicit time stepping
scheme which is safe in this sense; the simple forward Euler scheme


n  0 , I ,.,., N .
(17)
1./
For T7J in scheme (17) we also apply that according to relation (14). Further, for w we continue with
w =!/;? and for q?,j we also take the same as before. When a physically unrealistic value (according to
the conditions (6)(9 occurs again, w is halved for the second time and the time stepping with forward Euler is restarted, still using the same q~j' In case of something unphysical once more, the time
stepping is stopped and the finite volume visited is left without any update being made. (Notice that
for both time stepping schemes, we do not require condition (5) to be satisfied.)
With the present switchedrelaxationevolution approach we expect that in those volumes where
Newton iteration fails, the local evolution technique will finally bring the solution into the attraction
domain of the Newton iteration (for the next sweep) and so make itself quickly superfluous.
3.2.3. Numerical results for hypersonic blunt body flow. To illustrate the benefits of the switchedrelaxationevolution approach we consider again the hypersonic blunt body flow at M'l) = 8.15,
IX = 30.
In Figure 13, for the 16 X 8, 32 X 16, and 64 X 32grids shown in Figure 12, we give the
corresponding convergence behaviors obtained by the switchedrelaxationevolution technique. The
residual ratio along the left vertical axes is again the ratio Lt = II(Nh(qh, I/LL II(Nh(q2, I. Here,
the solution q2 is the uniformly constant initial solution, which is equal to the hypersonic upstream
boundary conditions. (Notice that these are singlegrid results.) The quantity along the right vertical
axes is the permillage of volumes in the total number of finite volumes visited during one cycle (one
cycle being defined as two diagonally opposite, symmetric switchedrelaxationevolution sweeps), In
which a switch to the evolution approach is made.
The robustness of the switchedrelaxationevolution technique is clear. For none of the cases considered is there an abortion of the solution process due to overflow or such. We even have convergence for all three cases. Further, from Figures 13b and 13c it appears that the evolution technique
makes itself superfluous indeed in the course of the iteration process. The extension to NavierStokes
can be quickly made. Only the time step needs to be reconsidered for diffusion. For this we refer to
e.g. Hindmarsh et al. [II].
~r
?+~r~
0."
o.IS
a. 16 X 8grid.
D.~
b. 32X 16grid.
Figure 12. Grids blunt forebody with canopy.
x
c. 64 X 32grid.
0.05
cycles
a. 16 X Sgrid.
10
cycles
cycles
b. 32 X 16grid.
c. 64 X 32grid.
Notice that the convergence slow down with decreasing mesh size is expected for a plain relaxation method and is supposed to vanish by application of a suitable multigrid technique. Unfortunately, the switchedrelaxationevolution technique combined with the basic multigrid method as
described in section 2, does not lead to satisfactory results; see Figure 14, with here for q~ the
approximate solution obtained by the nested iteration. Standard changes to the multigrid algorithm,
such as for instance the replacement of Vcycles (n a = I) by W cycles (n a = 2) do not help. It appears
that when applying multigrid to hypersonic test cases, in the standard way as described in section 2,
local coarsetofine grid corrections may be transferred which sweep the corresponding fine grid
iterates out of the attraction domain of the pure relaxation technique and even out of that of the
switchedrelaxationevolution technique. The cause of these problems may be either the coarse grid
corrections themselves, or the prolongation operator, or the combination of both. Therefore, in the
next section, to avoid possibly bad coarse grid corrections, we present a local damping technique for
the restricted defects and hence  implicitly  a local damping technique for the coarse grid corrections. To avoid a possibly bad correction transfer, in the next following section we also present an
alternative prolongation: a directiondependent prolongation. The improvements have already been
published in Koren and Hemker [12]. For reasons of simplicity. here we keep ourselves restricted to
the Euler equations.
.,,
o
,+,,r,.~
10
cycles
Figure 14. Convergence histories blunt fore body with canopy at Moo =S.15. 0:=0
(  : singlegrid.   : multigrid).
(n
( 19)
with S, ~ 1 denoting the operator for the defect damping, with ql' +', denoting the fine grid iterate as
obtained after the (fine grid) prerelaxation, and with ql'~1 and ql'!i denoting the coarse grid iterates
before the (coarse grid) prerelaxation and after the (coarse grid) postrelaxation, respectively. By
linearization (neglecting higherorder terms). from Equation (19) we derive the relation
11+1_ n
q'~1
q'~1

dN
(11 ) ~ 1
dN ( 11 + ")
'~I q'~1
S
I'~I
,q,
(11+"_ ')
d
'~I
d
q,
q"
q'~1
q,
01
N
n=, , ... , ,
(20)
in which qi denotes the fully converged solution of Equation (18). Considering for the coarse grid
correction q/' !II  ql' ~ 1 also the relation
j)~ 1 (q/,!/ ql'~ d=(q/' +, q/' + "),
(21)
with ql' ++, denoting the fine grid iterate before the (fine grid) postrelaxation, it follows the twogrid
convergence result
n+ o' , _ '  [11'
q,
q,,
'~I
ldN'~I(ql'~I)]IS
j'_ldNMI'+")]( 11+"_ ')
d
'~I ,
d
q,
q"
q'~1
q,
(22)
in which I, denotes the identity operator on Q,. From Equation (22) it is clear that in case the property
(23)
is satisfied, it holds
I ,, ~ 1
dN( n+!,)
dN( 11+")
, q,
( n + "  ' )  (I  S )1' I , q,
(11 +
d
q,
q,  '~I
'~I'
d
q,
, '
')
q, '
(24)
which indicates that for optimal twogrid convergence no damping should be applied (S, ~ 1 = I, ~ I)'
[I
II(N;I )iJII
[
] ,
max II(N;h l.2j III, II(N;hi l.2tll, II(N;h,. 2j III, II(N;h. 2j II
(25)
with N;_I dNtl(qid1dqtl, with N; dNt<qi+I')ldqt and with 1111 some matrix norm. Notice
that the local damping factor (25) is more or less the 2D equivalent of the I D damping introduced by
De Zeeuw [13]. To see if some additional gain can be obtained by also allowing local defect
amplification, in a numerical experiment we will also consider
n+1
II(N;di)1
(StI)i.j
[,
max II (Nth
,
l.2j III,
,],
(26)
At convergence of the solution, the defect multiplication will also have converged, both in case of
multiplication (25) and in case of multiplication (26). However, as opposed to the correction damping
proposed by Reusken [14], the present defect multiplication will probably not have vanished at convergence, neither in case of (25), nor in case of (26).
3.3.2. Numerical results for hypersonic blunt body flow. We proceed with evaluating the two defect
multiplication techniques proposed. As the starting point for improvement we consider the multigrid
results given in Figure IS. The results have been obtained for the blunt forebody with canopy at
Moo =8.15, a=O, without any multigrid improvement and without nested iteration, but with the
switchedrelaxationevolution approach. The initial solutions q?, I = 1,2, ... , L are taken uniformly constant and equal to the hypersonic upstream boundary conditions. In this way we obtain a poorer initial approximation, but we have a more discriminating test problem and we are ensured of an unambiguous evaluation, since q?, I = 1,2, ... , L will be the same for the different multigrid improvements to
be considered.
er
0
':;1
"a
='
:g
''
"',
.so ':'
~
10
cycles
Figure IS. Multigrid convergence histories blunt forebody with canopy at M ox; =8.15, a=O,
without nested iteration and without any multigrid improvement.
'(44
'2
]"
II(N,h.2J11=2 L.k>l.k,=I(N,h,.k,
'
(27)
2,.2)
the factor 2' simply accounting for the fact that in our case N,(q,) is a line integral form. Corresponding with the results in Figures 16a and 16b, in Figures 17a and 17b we show distributions of the multiplication operator SL I,L = 5 (the multiplication factor distribution on [24), as applied in the last
(i.e. the IOth) multigrid cycle. In both cases the local damping is confined to only the close neighborhood of the blunt body. Locally, in Figure 17b the damping appears to be a little bit stronger than
that in Figure 17a. However, globally this is more or less compensated by the local amplifications.
Notice that the maximal amplification factor that was found to be applied in Figure 17b is still 0(1)
only. The minimal damping factors in Figures 17a and 17b are much larger than those found by De
Zeeuw [13] for his specific nonlinear test case. A second difference is the good improvements found
by De Zeeuw [13] for his basic multi grid method's performance and the present modest improvements. Both differences suggest that (at least) for the present test case, in order to significantly
improve the results presented in Figure 15. defect multiplication is not needed as much as improved
grid transfer operators; the topic of the next section.
04
Os
,
Os
04
...
0'
.g
.~
~ ,
'il
"0
'@ "',
,go
::s
03
''
''
"',
03
,go
o
o
,+~,.,.~
10
,+,.,,4
1
cycles
cycles
10
Figure 16. Multigrid convergence histories blunt forebody with canopy at Moo =8.15. a=O,
with defect multiplication.
,;
N
C
,;
~~
..... 1l
0
0.9
\~,1.
?+.,~~~
O.OB
0.06
0.04
0.02
?+r.~_r~
O.OB
0.00
0.06
0.04
0.02
0.00
Figure 17. Distribution multiplication factors applied on Q 4 in the 10th multigrid cycle.
blunt forebody with canopy at M % =S.15. 0:=0.
(q?ew l2, 
1.2)  1
(q?ew h,  1.2i
= (q)'ld hi  1.2)
+(tlqlIl,.,.
(q?ewb. 2, 1
= (q)'ld h,. 2) 
+(tlq'I)i.1'
(q?ewl2,.2,
= (q)'ld b . 2i
(2Sa)
+(tlq'I)i.,.
with
(2Sb)
The directiondependent correction prolongation that we propose now can be illustrated as in Figure
19. In mathematical terms  analogous to formulae (2S)  solution correction by means of this
directiondependent prolongation is written as
(q?ewl2,. 2) (
q,new) 2,.2)
'A,].
+,1
(29a)
with the four fine grid cell center corrections (Figure 19b) defined as central averages of the coarse
grid cell face corrections (Figure 19a). The coarse grid cell face corrections are defined by
( uq,
 I )i
)
+ 'h,j  ( q,new)
 I i + 'h,j  ( qIold
 I ,+ ,",),
(~q'I )i.j 
'h
= (q7,,WI ),.j 
'h
(29b)
where the cell face states are computed in an upwind manner. We notice that instead of the present
procedure of computing coarse grid cell face states in an upwind manner and from that, in a central
manner, fine grid cell center corrections, might as well have been the reverse: computing the coarse
grid cell face states in a central manner and from that the fine grid cell center corrections in an
upwind manner. However, a drawback of the latter approach is that it requires additional geometrical
data for the upwind computation of the fine grid cell center corrections.
I
I
(~/l)i,j
I
I
I
(~/l).,j
rlI
I
I
(~/l)i,j
(~/l)i,j
I
I
Figure 18. Coarse grid finite volume with corresponding nextfiner grid volumes
and corresponding piecewise constant corrections.
;
;
+
~
I
,":
I
~
(~/l)i,jl;
Figure 19. Coarse grid finite volume with corresponding nextfiner grid volumes
and corresponding directiondependent corrections.
qright),
where j(q) and F(qleft,qright) denote the exact and numerical Euler flux function, respectively. A
drawback of the Euler equations is that obtaining a primitive state vector like e.g. q=(p,U,V,p)T from
j (q)=(pu,pu 2 +p,puv,pu(e +p / p)l requires the solution of a quadratic algebraic equation. Fortunately, with the Pvariant of Osher's scheme (see Hemker and Spekreijse [I]), for most Riemannproblem cases arising in aeronautics F(qleft,qright)= j(q.), q. being a welldefined, single state vector
on the wave path connecting qleft and qright in state space. Hence, with the Pvariant, in most cases without evaluating F(qleft,qright)  we can directly identify qface as qface=q . For the Ovariant of
Osher's scheme (see Hemker and Spekreijse [I]), in almost all Riemannproblem cases arising in
aeronautics F(qleft,qright) is found to be the sum of three different fluxes j(q). In these cases, because
of j(q)'s nonlinearity, the previous simple procedure is not possible. In the (rare) cases where the Pvariant also leads to a sum of fluxes, we solve the quadratic equation and in case of a positive
discriminant and one zero being physically irrelevant (negative p and/or p), we take the zero which is
physically relevant (positive p and p). In all other cases, we simply take qrace = 'l2(qleft + qright).
Because of the consistency of Osher's scheme at boundaries, there the present upwind prolongation
can also be applied in a consistent way. Notice that the upwind prolongation may lead to cell face
states which are local extrema in state space. In conclusion, we emphasize that by replacing the
piecewise constant prolongation operator by the present upwind prolongation operator, the complete
numerical method has become more consistent. Both the discrete Euler operator and the correction
prolongation operator are upwind now, both being based on the same upwind scheme: the Pvariant
of Osher's scheme.
3.4.2. Restriction. A provable consequence of the upwind prolongation is that no restriction operator
I can be made for which the coarse grid finite volume discretization is a formal Galerkin approximation of the fine grid finite volume discretization. The possibly most effective restriction operator
that can be really made is the exact adjoint of the nonlinear prolongation operator. Unfortunatelyas opposed to the upwind prolongation  the exactly adjoint restriction operator will certainly lead to
a significant increase of the computational overhead. More suitable seems to be a linear approximation of the exact (nonlinear) adjoint. For this we write the latest obtained coarse grid cell face states
as linear combinations of the corresponding left and right states:
II
k = 1,2,3,4,
ak=
(qrace)k/(qright)k  I +12
(qleft)k/(qright)k  I
(3Ia)
(3Ib)
1,
where q is the conservative state vector, q=(p,pu,pv,pe)T, and a small parameter which guarantees
that (qface)k is a central average in case (qleft)k =(qrighth =(qraceh. With next the central computation
of the fine grid cell center states, we then have
(q/hi  I.2j I.k =(i! I q/I hi  1.2;l.k
'l2(a/I)i.j~,.k(q/di.)I,ko
k = 1,2,3,4,
(32a)
'e,k] (q/I)i,j.k +
k =1,2,3,4,
(32b)
=(j~Iq'I h"
2j 
U:
l.j.k
2j,h =
(32c)
k = 1,2,3,4,
k = 1,2,3,4,
(32d)
With for q the conservative state vector q =(p,pu,pv,pe l, the linear relations (32a)(32d) display how
the upwind prolongation distributes mass, momentum and energy from a coarse grid to the overlying
finer grid, For the approximately adjoint restriction operator (i.e, the approximation of the exact nonlinear adjoint), we can then write
(r,_ d',j.k
=(I~I r,),.j,k =
21
2,k] +
[(r,h,
1,21
k = 1,2,3,4.
(33)
Of course, the weak spot in the approach (31 )(32) is the linear approximation of the nonlinear prolongation. In case (qrace)k is a local extremum, i.e. does not lie in between (qleft)k and (qrighth, we
have a negative coefficient in relation (3Ia) (either ak or Iak) and hence also in relation (33). We
do not accept this situation. If occurring, locally and for that kth component only, we neglect how
the upwind prolongation really was and simply consider (qracelk = 0qleft)k + (qrightlk).
3.4.3. Numerical results for hypersonic blunt body flow, We now proceed with evaluating the
directiondependent grid transfer techniques, As starting point for improvement we consider again the
multigrid results as obtained for the blunt forebody with canopy at Moo =8,15, a=O without any
multigrid improvement, but with the switchedrelaxationevolution approach, (See Figure 20, which is
the same as Figure 15.) The multigrid behavior obtained after having replaced both the basic correction prolongation operator and the basic defect restriction operator by the directiondependent operators, is given in Figure 21a. (We remark that defect damping is not applied.) With upwind grid
transfers only, the improvement with respect to Figure 20 is significant indeed.
Replacing in the basic multi grid algorithm only the standard correction prolongation (by the
upwind prolongation), we obtain the multi grid performance given in Figure 21 b, These results are
only a little bit less good than those in Figure 21a and hence make us conclude that the previous,
rather cumbersome efforts in also upwinding the defect restriction, do not payoff enough. Therefore,
in the following we refrain from applying the upwind restriction,
224
10
Figure 20. Multigrid convergence histories blunt forebody with canopy at M'l.) = 8.15, 0: =0,
without nested iteration and without any multigrid improvement.
,.
.g
,. ':'
0
.~
tIS
....
<a::s r

<a::s
D5
D4
~ 'f'

OIl
D3
OIl
D4
~ ,
~
..9
....
D5
..9
'l'
D3
'l'
c
cycles
10
b. With upwind
cycles
cor~ection
10
prolongation only.
Figure 21. Multigrid convergence histories blunt forebody with canopy at Moo =8.15,0:=0.
';"
'0'
'.::1
...
~
~
'0
'i...3
""
.,
Os
114
'",
.9 ,
03
'"
0
,
6
10
cycles
Figure 22. Multigrid convergence histories,
blunt forebody with canopy at Moo =8.15, a=O,
with both defect damping and upwind correction prolongation.
~
~
;..,~
Ei
o.jj)~~
,.
Ei
.;
...,~
(S4)min ROO. 15
~
~
Ei
p"'
{::
~ '(s.).;.~o."
~O'_
0.8
0.9
'0.7
C;
C;
? 0.08
.;
~9
0.06
0.01
0.02
0.00
? 0.08
0.06
0.01
0.02
0.00
3.5.2. Combination of defect damping, upwind correction prolongation and nested iteration. Comparing the basic multi grid method's results as presented in Figures 14 and 15  results obtained with and
without nested iteration, respectively  clearly shows the natural beneficial influence of nested iteration.
For the improved multi grid method, the method with defect damping and upwind correction prolongation, the benefit of nested iteration (consistently, with upwind solution prolongation) is observed by
counting the number of finite volumes in which  locally  the switch is made from the relaxation
technique to the evolution technique (Figure 24). In both Figure 24a and Figure 24b (without and
with nested iteration, respectively), the quantity along the vertical axis is a scaled number of switches
made during the nth multigrid cycle (n = 1,2, ... ,10), the scaling factor being the total number of
volumes visited during one nonlinear multigrid cycle; a Vcycle with npre=npost=1 and with symmetric relaxation sweeps. (Notice that the scaling factor increases when going from Q3 to Qd The
nonzero percentage at n = 0 in Figure 24b indicates the total amount of switches made during the
nested iteration. For all four grids considered, the expected positive influence of the nested iteration
appears to be significant.
In Figure 25 we give the multigrid convergence behavior corresponding with the latest favorite
strategy, the strategy with defect damping, upwind prolongation and nested iteration. In this figure,
for Q6, a comparison is also made with the corresponding singlegrid convergence behavior. In Figure
26 we show the corresponding converged damping factor and Mach number distributions. Notice that
the smallest damping factors are mainly concentrated along the bow shock, in particular there where
the jumps across the shock are largest. Finally, we show results again for the more interesting reentry
case Moo = 8.15, 0: = 30. Also for this test case, the convergence results (Figure 27) show the
beneficial influence of the changes in the basic multigrid method. Given the very low convergence
rate of the singlegrid computation (Figure 27a) and given the absolute failure of the basic multigrid
method (Figures 27a and 27b), the multigrid improvements do not just appear to be a nice luxury,
but a real necessity. Analogous to Figure 26, in Figure 28 we still show the converged damping factor
and Mach number distributions. Notice that  like in Figure 26a  the smallest damping factors in
Figure 28a are located at the most pronounced part of the bow shock.
  0..
10
cycles
cycles
10
.'0'
.,
':'
......:~06
e ..,
:9
e'"
........ '
.2
':'
0
o.
10
cycles
Figure 25. Convergence histories blunt forebody with canopy at Moo =8.15, a=O,
with both defect damping, upwind prolongation and nested iteration,
(  : singlegrid.   : multigrid).
~'~7r~n
~.~
(!;
?+~..~~
0.08
0.06
0.01
0.02
.oe
0.00
Q s.
Q6'
Figure 26. Converged results blunt forebody with canopy at M YO =8.15, a=O,
with both defect damping, upwind prolongation and nested iteration.
\
"
"
"
"
,,
,, ,,
: ::
,,
f :;
"
...
,+..r~
10
15
cycles
a. Convergence histories.
20
8..
'
~',, _____________________ _
10
15
20
cycles
b. Amount of volumes with
switch from relaxation to evolution,
percentage of volumes visited
during one multi/singlegrid cycle.
Figure 27. Iteration histories blunt forebody with canopy at M'l) =8.15, a=30,
with both defect damping, upwind prolongation and nested iteration, Q 6 only,
(  : singlegrid, ........ : basic multigrid,   : improved multi grid).
;;
,;
N
0
,;
;..,~
;..,~
;;
;;
?+~~~~~~~
0.06
0.06
O.Oi
0.02
0.00
?+...0.06
O.Oi
0.02
0.00
Figure 28. Converged results blunt forebody with canopy at Moo =8.15, a=30,
with both defect damping, upwind prolongation and nested iteration.
Concerning the efficiency of the improved multigrid method, one may find the paradoxical result
that one multigrid cycle with both defect damping and upwind correction prolongation is still cheaper
than one multigrid cycle without both. The cause of this simply is that the computations with the
improved multigrid method may result in a significantly smaller number of switches from the local
relaxation to a local evolution during the smoothing phases and hence in a lower computational cost.
Concerning the efficiency of the upwind computation of cell face states (as applied in the upwind prolongation), for the test cases considered it appears that with the Pvariant, at almost all cell faces it
holds that F(qleft,qright)= f(q.). (For both 06multigrid cases considered in this section, solving a
quadratic equation for q face appears to be necessary at about I % of all cell faces only.)
230
To finish, we summarize the improved multigrid algorithm, the improvements being indicated in
bold.
Nested iteration:
 Choose ql.
 Improve q 1 by a single nonlinear multigrid cycle.
 Transfer the improved approximation q 1 to rl2' by applying the upwind prolongation operator.
 Improve q2 by a single nonlinear multi grid cycle.
 Continue the previous process until an initial estimate for qL has been obtained by upwind prolongation of qL ~I'
Nonlinear multigrid iteration:
Apply npre prerelaxation sweeps to N/(q/)=r/.
Compute the defect d/=N/(q/)r/ and restrict it: d/~I =I~~ld/.
Compute the local damping factors (S/ ~ 1 )',j and damp the restricted defect: d/  1 : = S/ ~ 1 d/ ~ I'
Compute the righthand side r/ ~ 1 = N/ ~ 1(q/ ~ I)  d/ ~ 1 .
Approximate the solution of N/~I(q/~d=r/~I by the application of no nonlinear multigrid cycles.
Correct the current solution, by applying the upwind prolongation operator.
Apply npost postrelaxation sweeps to N/(q/)=r/.
4. CONCLUSIONS
For the hypersonic test cases considered in this paper, the essential element for robustness of the
smoother is the continuous monitoring of both the local relaxation and the local evolution. The
essential element for convergence appears to be the combination of Newton iteration, Wambecq's
explicit, twostep rational RungeKutta scheme and the explicit Euler scheme. For steady NavierStokes flow computations at a finite Reynolds number, the proposed checks on physical correctness
can be maintained as long as the flow remains adiabatic. Only the time step needs to be reconsidered
for diffusion.
A satisfactory remedy against divergence of nonlinear multigrid appears to be the combination of
a (local) damping of the restricted defect and a (global) upwind prolongation of the correction.
Besides a positive influence on the robustness of the algorithm, the combination of upwind prolongation and defect damping also has a positive influence on the computational efficiency. Applicationin addition  of an (approximately adjoint) upwind restriction operator does not really payoff. For
the test cases considered, the best improvement is obtained by the application of the upwind prolongation operator. With this operator we have achieved a greater upwind consistency throughout the
complete numerical method. For sake of clearness. we remark that the separate merits as observed
here for the multi grid improvements, may well be different for other test cases. Just as for the
switchedrelaxationevolution approach, the multigrid improvements are not restricted to the Euler
equations, but can be carried over as well to the NavierStokes equations.
Finally, we remark that the new techniques are such that the improved algorithm, just like the
basic algorithm, do not require any tuning of parameters.
Hemker, P.W. and Spekreijse, S.P. Multiple Grid and Osher's Scheme for the Efficient Solution
of the Steady Euler Equations, Applied Numerical Mathematics, VoI.2, pp. 475493, 1986.
2.
Hemker, P.W. Defect Correction and Higher Order Schemes for the Multi Grid Solution of the
Steady Euler Equations, in Lecture Notes in Mathematics (Ed. Hackbusch, W. and Trottenberg,
U.), VoI.1228, pp. 149165, Proceedings of the Second European Conference on Multigrid
Methods, Cologne, 1985. SpringerVerlag, Berlin and New York, 1986.
3.
4.
Koren, B. Defect Correction and Multigrid for an Efficient and Accurate Computation of Airfoil
Flows, Journal of Computational Physics, VoL77, pp. 183206, 1988.
5.
Koren, B. Multigrid and Defect Correction for the Steady NavierStokes Equations, Journal of
Computational Physics, VoL87, pp. 2546, 1990.
6.
7.
Godunov, S.K. Finite Difference Method for Numerical Computation of Discontinuous Solutions of the Equations of Fluid Dynamics (Cornell Aeronautical Lab. TransL from the Russian),
Matematicheskii Sbornik, VoL47, pp. 271306, 1959.
8.
Osher, S. and Solomon, F. Upwind Difference Schemes for Hyperbolic Systems of Conservation
Laws, Mathematics of Computation, VoL38, pp. 339374, 1982.
9.
Hakkinen, R.J., Greber, I., Trilling, L. and Abarbanel, S.S. The Interaction of an Oblique Shock
Wave with a Laminar Boundary Layer, Memo 21859 W, NASA, Washington, 1959.
10. Wambecq, A, Rational RungeKutta Methods for Solving Systems of Ordinary Differential
Equations, Computing, VoL20, pp. 333342, 1978.
II. Hindmarsh, A,c., Gresho, P.M. and Griffiths, D.F. The Stability of Explicit Euler TimeIntegration for Certain Finite Difference Approximations of the MultiDimensional AdvectionDiffusion Equation, International Journal for Numerical Methods in Fluids, VolA, pp. 853897,
1984.
12. Koren, B. and Hemker, P.W. Damped, DirectionDependent Multigrid for Hypersonic Flow
Computations, Applied Numerical Mathematics, Vo1.7, pp. 309328, 1991.
13. De Zeeuw, P.M. Nonlinear Multigrid Applied to a OneDimensional Stationary Semiconductor
Model, SIAM Journal on Scientific and Statistical Computing (to appear).
14. Reusken, A.A. Convergence Analysis of Nonlinear Multigrid Methods, doctoral thesis, State
University of Utrecht, Utrecht, 1988.
Chapter 7:
LaminarTurbulent Transition
D. Arnal
ONERA/CERT, Aerothermodynamics Department,
2 avenue E. Belin, 31055 Toulouse Cedex, France
ABSTRACT
The chapter is devoted to a description of theoretical, numerical and experimental problems related with
boundary layer transition at high speeds. On the theoretical point of view, emphasis is given on linear
stability theory which is able to take into account the effects of some parameters acting on transition. In
order to predict the transition location, the application of the socalled "en method" is discussed. Another
problem is to model the transition region. This can be done by using intermittency methods; some
comparisons with experimental results are presented. Other transition mechanisms are also described, in
twodimensional (large roughness elements) as well as in threedimensional flows (leading edge
contamination).
1  INTRODUCTION
Since the classical experiments performed by Osborne Reynolds (1883), the instability of laminar flow and
the transition to turbulence have maintained a constant interest in fluid mechanics problems. This interest
results from the fact that transition controls important aerodynamic quantities such as drag or heat transfer.
For example, the heating rates generated by a turbulent boundary layer may be several times higher than
those for a laminar boundary layer, so that the prediction of transition is of great importance for hypersonic
reentry spacecraft, because the thickness of the thermal protection is strongly dependent upon the altitude
where transition occurs. Drag reduction for supersonic aircraft also requires a good knowledge of the
transition mechanisms.
When a laminar flow develops along a given body, it is strongly affected by various types of
disturbances generated by the model itself (roughness, vibrations ...) or existing in the freestream
(turbulence, noise, temperature fluctuations). These disturbances are the source of complex mechanisms
which ultimately lead to turbulence. Two kinds of transition processes are usually considered:
a) If the amplitude of the forced disturbances is small, one can observe at first exponentially growing
instabilities, the development of which is governed by ~ equations (Toll mienSchlichting waves,
GlirtIer vortices, ... ). Threedimensional and non linear effects occur subsequently, inducing secondary
instabilities and then transition. In these cases of socalled "natural transitions", the transition Reynolds
numbers can be very large.
b) If the amplitude of the forced disturbances is large (high freestream turbulence level, large roughness
elements, ...), non linear phenomena are immediately observed and transition occurs a short distance
downstream of the leading edge of the body. This mechanism is called a "bypass", in the sense that the
linear stages of the transition process are ignored (bypassed).
~z
 rot)]
(1)
00
00
~z
 Olrt)]
(2)
Oli represents the temporal amplification rate. Depending on its sign, the disturbances are amplified
(Oli > 0), neutral (Oli = 0) or damped (Oli < 0). Olr is a circular frequency. a and ~ are the components of
the wavenumber vector in the x and z directions, respectively. It will be shown that the wavenumber
direction 'If = tanl(~/a) is a very important parameter.
in the spatial theory,
00
is real, a and
~iZ)]
exp [i (arx +
~rz
 Olt)]
(3)
We may defme a real wavenumber1 with an amplitude Ikl and a direction 'If :
Ikl=(a; +
~;r2
(~r/ar)
(a~
(4)
~ ~1 ) 1/2
and direction
'If = tan 1 ( ~i/ ai) If ~i is set equal to zero, the waves can be amplified (ai < 0), neutral (ai = 0)
or damped (ai> 0).
As a general rule, Gaster's relation makes it possible to convert a temporal to a spatial amplification rate
by using the group velocity concept [20].
Introducing (1) into the linearized equations leads to a system of six ordinary differential equations in y
for the amplitude functions r(y). Due to the homogeneous boundary conditions (all fluctuations must vanish
at the wall and in the free stream, except the pressure fluctuations which have a non zero amplitude at the
wall), the problem is an eigenvalue one: when the mean flow is specified, a non zero solution exists only
for peculiar combinations of the four parameters a, ~, 00 and Re, where Re is the Reynolds number. In
fact, there are five real parameters in temporal theory (a, ~, Olr, Oli, Re) and six in spatial theory (ar , ai,
~
r, ~i,
00,
00,
Re).
From a numerical point of view, the input data are the boundary layer mean velocity and temperature
profiles, the free stream Mach number, the wall temperature, the characteristics of the fluid and all but two
parameters. The computation gives the two remaining parameters (eigenvalues) as well as the disturbances
amplitude profiles (eigenfunctions).
II
dy
[dU]
p
= 0
(5)
dy ys
where p and u are the mean density and the mean velocity. It was demonstrated by Lees and Lin [10] that the
presence of such a point is a sufficient condition for the appearance of unstable disturbances; it is also a
necessary and sufficient condition for the existence of a neutral subsonic wave (the definition of a subsonic
disturbance will be given below) ; the phase velocity of this wave is the mean velocity at Ys. However,
these results are valid only if:
(6)
Let us consider the flat plate case. When the Mach number increases, the generalized inflection point
goes from the wall towards the outer edge of the boundary layer, and the numerical results indicate that the
range of unstable frequencies is enlarged at high Reynolds numbers. It is clear that this "inflectional
instability" plays a crucial role for instability and transition in hypersonics.
A
" = Mc/a
(7)
M is the local Mach number; for classical boundary layer profiles, it increases from 0 to Me between
the wall and the free stream. c is the phase velocity of the waves; it does not depend on y : for twodimensional waves (~ =0), c = wr/a in temporal theory, and c = w/a r in spatial theory. a represents the
local speed of sound, which depends on the mean temperature distribution; obviously, it takes a non zero
"
"
cia
= I, supersonic if M2
"
> I.
t
1
"
0.4
ar
a,
0.3
t
0.2
0.1
_
0
510'
a) Me
ar
R61
0
10'
=0
R61
510'
b)
10'
= 1.3
Me
ar
_101ai
0.2
0.4
0.2
0.1
0.2
_
0
510'
a) Me
a,
R61
_
0
10'
= 2.2
510'
d) Me
ar
R61
10'
=3
0.5
0~______~===;~
__~~~R:61~
10'
e) Me = 4.5
ar
ar
R61
0~10~'~210'
210'
f) Me
= 4.8
r_~10~'a;~=ro'~=:==~=C~====~
4
<:;6
<6
__ R61
o~~~~=~~~~~
210'
410'
g) Me = 5.8
hJ Me = 7
ar
t6
, no =
_10
00~'2'10'7"'.J104
4
i) Me = 10
238
In the stability computations, the spatial theory is used, and it is assumed that the amplification vector
is aligned with the mean flow direction, i.e. ~i = 0 or 'II = 0 [21].
a) Twodimensional waves. 'I'  0
Figure 2 shows examples of stability diagrams computed for twodimensional waves ('II = 0) and for
Mach numbers ranging from 0 to 10. For the sake of clarity, only some isoamplification curves
( ai > 0) are plotted in the (Reynolds number, wavenumber) plane. The Reynolds number is computed
from the displacement thickness BI, the free stream velocity Ile and the kinematic viscosity Ve of the outer
flow. a r and a i are made dimensionless with this displacement thickness. For each Mach number, there is
a critical Reynolds number (RBIcr) below which all disturbances are damped.
The first observation is that the stability diagrams do not change verJ much when the Mach number
increases from 0 to 1.3 (figures 2ab). This can be explained by the fact that the generalized inflection point
remains very close to the wall in this Mach number range, and the boundary layer is unstable essentially
through the action of viscosity: this phenomenon, whereby the maximum amplification rate increases
with decreasing Reynolds number at a fixed value of ro or a r is called viscous instability; stability
diagrams where viscous instability is present exhibit a local maximum of  ai at fmite Reynolds numbers,
and the curves of equal amplification rate are closed around this peak. Another important observation is that
the amplification factors of the unstable waves are clearly smaller in the transonic range than in
incompressible flow, even if the critical Reynolds number remains of the same order.
At higher Mach numbers, the altitude of the generalized inflection point increases, and the inflectional
instability plays a more and more important role: the in viscid theory shows that there is a range of
unstable Clr or ro at infinite Reynolds number and this range is enlarged with increasing Mach numbers; as
a consequence, at large but finite Reynolds numbers, the curves of equal amplification rates tend to be
parallel with the RBIaxis.
For Me = 2.2 (figure 2c), the stability diagram clearly shows the viscous instability (below
RBI = 7 (00) and the inflectional instability (beyond RBI = 7 0(0). The latter becomes predominant for
Me = 3 (figure 2d).
It has been noticed previously that higher modes could be computed as soon as the free stream Mach
number becomes larger than 2.2, but they are at first associated with very large wavenumbers. For this
reason, only the first unstable mode is ploued in figures 2c and 2d for Me = 2.2 and 3. However, when the
Mach number is increased, the second unstable mode appears at lower and lower wavenumbers. This is
illustrated in figure 2e (Me = 4.5), where the first and the second modes are close together. Let us observe
that the second mode is more unstable than the first one: the ratio of the maximum amplification rates is
of the order of four of five. At Me = 4.8 (figure 2f), the two unstable regions join each other and there is
only one neutral curve in hypersonic conditions (figures 2g to 2i).
b) ObliQue waves. 'I' "# 0
The previous results were obtained for twodimensional waves. However, an important aspect of
instability in compressible boundary layer is the effect of the wavenumber direction", on the amplification
rates. In incompressible flow, only twodimensional waves need to be considered, because it can be
demonstrated that they are the most unstable ones (Squire's theorem). This is no longer the case in
compressible flow, even at moderate Mach numbers. In supersonic and hypersonic conditions, the situation
is particularly complex, as it can be seen in figure 3, where two stability diagrams are represented for
Me = 4.5 and two values of",: 0 and 60. Changing the wave orientation stabilizes the second mode,
but increases the instability of the first one, so that it becomes difficult to define "the most unstable
direction".
t2
t2
0.5
lit' =0'1
oL~=====10~4==~~2~10~
_ R51
Figure 3 Effect of the wave orientation on the amplification rate (Me = 4.5)
c) Real gas effects
To date, most of the compressible stability analyses have assumed ideal gas behavior. Real gas effects
were studied quite recently by Malik [22], [23], Gasperas [24], Stuckert and Reed [25]. The latter study
demonstrates that the amplification rates associated with the second mode do not change very much from the
equilibrium to non equilibrium and ideal gas models, but that there is a shift in the range of the unstable
frequencies. The first mode is usually more stable than the second one, although it is strongly affected by
real gas effects.
240
stream disturbances; this constitutes one of the possible reasons of the "unit Reynolds number effect",
which illustrates the importance of the receptivity process.
4  TRANSITION PREDICTION
4.1. The en method
For a given melin flow, it is possible to compute a stability diagram (figure 4) showing the range of
unstable frequencies f as a function of the stream wise distance x. Let us consider now a wave which
propagates downstream with a fixed frequency f. This wave passes at first through the stable region ; it is
damped up to xO, then amplified up to x I, and it is damped again downstream of x I. At a given station x,
the total amplification rate of a spatially growing wave can be defined as :
x
In(NAQ) =
f
(l
i dx
(8)
xO
~~~.x
n = Max [In(NAQ)]
f
(9)
In incompressible flow, with a low disturbance environment, it is assumed that transition occurs as
soon as the n factor reaches a critical value in the range 710, Le. when the locally most unstable frequency
is amplified by a factor e 7 to elO [33, 34]. The problem is to know if the en method remains valid for
compressible boundary layers.
4.2. First application: flat plate with adiabatic wall
Envelope curves are plotted in figure 5 for four supersonic Mach numbers. We considered twodimensional
waves (~r = 0), as well as oblique waves (~r 0) associated with the most unstable", angle (strictly
speaking, this angle depends on the Reynolds number for each Mach number, but its variation is rather
weak). Up to Me'" 5, the oblique waves are the most unstable ones; however, for hypersonic flows, the
twodimensional waves become the most amplified again.
'*
Figure 6 shows an application of the en method for adiabatic flat plates. The stability diagrams were
used for computing the theoretical stream wise Reynolds numbers which correspond to different values of the
n factor, as a function of the freestream Mach number. The dotted lines represent theoretical results given by
Mack [16] for n =4.6 and 6, and the crosses are computations performed by Chen and Malik for Me = 3.5
and n = 2, 4, 6, 8 and 10 [35]. If it is assumed that transition occurs for a fixed value of the n factor, each
curve represents the evolution of the transition Reynolds number when Me increases. It can be observed that
compressibility has a strong stabilizing effect in transonic flow; it becomes destabilizing for Mach
numbers between 2 and 3.5, and stabilizing again in hypersonic conditions.
It has been noticed previously that the value of n at transition onset is usually between 7 and 10 in
"clean" subsonic wind tunnels. The problem is obviously to know if similar values of n are observed at
high speeds. The solid symbols in figure 6 represent experimental data obtained at ONERA some years ago
[36] : they correspond to low values of the n factor, between 2 and 4. Numerous examples could be given
for illustrating the fact that, in conventional hypersonic wind tunnels, the measured transition Reynolds
numbers are rather low, say between 2 and 3 millions. Possible origins of this behaviour are discussed
below.
4.3. Wind tunnel simulation  Empirical criteria
It is well known that transition on a smooth surface can be triggered by the disturbances which are present
in the free stream: velocity fluctuations u' (free stream turbulence), pressure fluctuations p' (acoustic
disturbances), temperature fluctuations T. The problem is very complex, because the effect of these various
disturbances depends on the Mach number range (Pate, [5]).
At low speed, the transition Reynolds number is very sensitive to the free stream turbulence level Tu ;
however, if Tu becomes very low, pressure fluctuations (fan noise for instance) can be of major importance
for inducing transition.
In transonic flow, acoustical phenomena linked with slotted or perforated walls give rise to strong
pressure disturbances, which cause early transitions. Better results are obtained in transonic wind tunnels
with solid walls.
In supersonic and hypersonic flows, the main factor affecting transition is also the noise, the origin of
which lies in the pressure disturbances radiated by the turbulent boundary layers developing along the wind
tunnel walls. The effects of u' and T' have not been firmly established for Me > 3.5.
The strong effect of the wind tunnel noise on the transition Reynolds number is illustrated in figure 7
(Harvey, [37]), where values of RXT, measured on cones and flat plates, are given as a function of
n
10
/'
0
/'
t IMe =2.21
IMe = 1.1 ]
5
n
10
/'
If! = 40 /
,//0 0
~
10
t IMe =
0
00
10
..10 3 R~1
10
n
10
4.81
IMe = 5.81
o~~~~
20
20
40
10 6 Rx
t
15
IhJ:.w.
n= 10 /
./
8//
Experiments
luillen
"quiet tunnel"
./
10

Mack
Chen and Malik
Present computations
./
5
Figure 6 Application of the en method
(flat plate, adiabatic Wall)
O~~~2~~4~~6~~
Me
243
lrfl
RXT
8
6
4
,
Cones,
flight
107
8
6
\r
,
M,z5
lri'
8
6
(unflagged symbolsl
, "
(flag~ed
Sharp cones
li:
ata fairing
p~/qoo, for 4 < Moo < 23 (qoo is the mean dynamic pressure in the free stream and the  denotes a rootmeansquare value). There are two separate mean curves for cones and flat plates, but the effect of Mach
number disappears completely in this representation: transition is governed by
Pate also analyzed available wind tunnels data and developed an empirical criterion for "natural"
transitions measured in supersonic and hypersonic facilities. This criterion is a correlation between the
transition Reynolds number RXT and the parameters acting on the noise intensity: the tunnel test section
circumference P and two characteristic parameters of the turbulent boundary layer on the nozzle walls (mean
skin friction coefficient CF and displacement thickness SO [5]. If the free stream Mach number and the wall
temperature ratio Twffaware constant, the criterion reduces to :
(10)
g increases as the test section circumference increases. This shows that the transition Reynolds number
increases with increasing unit Reynolds number UelVe and increasing wind tunnel size.
Another correlation, which was used during wind tunnel tests of a smooth model of COLUMBIA space
shuttle, is expressed by the simple relationship:
(R9/M e )T = constant, of the order of 200
(11)
R9 is the momentum thickness Reynolds number of the laminar boundary layer which develops on the
model. The values of R9T deduced from (11) are too low when compared with free flight data on smooth
models.
Since the radiated noise is inherent in the presence of walls around the model, there is little doubt that
wind tunnel conditions cannot simulate properly flight conditions. In order to reduce the noise level, it is
necessary to delay transition on the nozzle walls, because a laminar boundary layer is less noisy than a
turbulent one. This was done in the "quiet tunnel" built in NASA LANGLEY with a free stream Mach
number Moo = 3.5. A detailed description of the wind tunnel was given by Beckwith et al [38]. The nozzle
is twodimensional and differs in design substantially from more conventional ones. Three notable features
are the use of a rapid expansion contour, a wide separation of the two planar sidewalls and the provision of
boundary layer bleed slots shortly upstream of the nozzle throat. For unit Reynolds numbers between
2.5 105 and 15 105 per inch, the observed values of P~oo were found to vary from 0.03 102 up to about
244
0.8 102 , depending upon the unit Reynolds number, the axial location and the bleed slot flow (bleed valve
open or closed). The minimum value is one or two orders of magnitude below that which is measured in
conventional wind tunnels; it corresponds to laminar boundary layers on the nozzle walls. The higher
noise levels are caused by radiations from transitional or turbulent wall boundary layers; they are
comparable with the levels reported for other tunnels at the same Mach number. Design and operational
details of new lowdisturbance wind tunnels (up to Mach 18) are described in [39].
4.4. further applications of the en method
Transition experiments were carried out in the quiet tunnel: on a flat plate, transition Reynolds numbers as
high as 12 106 were measured. This value is nearly an order of magnitude larger than those obtained in
conventional, noisy facilities [35]. What is interesting to notice is that it corresponds to the theoretical RXT
given by the en method for Me = 3.5 and n = 10 (figure 6).
The flow on supersonic sharp cones constitutes a second case where the free stream Mach number is
constant in the stream wise direction. Measurements performed in the quiet tunnel on a 50 half angle sharp
cone indicated values of RXT close to 7 or 8 106 , two or three times the values obtained in conventional
wind tunnels. The situation is summarized in figure 8 ; it shows flight transition data which were collected
for sharp cones by Beckwith [40]. The transition Reynolds numbers are plotted as a function of the Mach
number. The figure also contains a correlation for wind tunnel transition data, which lies much below the
flight experiments. The range of results obtained in the "quiet tunnel" are reported for comparison.
10
8
64,
RXT
t
10
t.
li"t>t>
t>t>
~t>"t>t>
t.
"tit.
ti
~t.
t.
I At.
%>
"A
t.
<>0
"quiet tunnel"
t.
10
<>
Me
0
4
8
12
16
From a theoretical point of view, it can be assumed that the stability equations are the same on cones
and on flat plates, the evolutions of the mean flow properties in both cases being related by Mangler's
transformation. For cones, the predicted transition Reynolds number computed for n = 10 in the quiet
tunnel conditions is 8 106, in very good agreement with experimental data.
The first conclusion is that the en method, with n '" 10, can be applied for predicting transition in
supersonic flow provided the background disturbance level is low enough. Bushnell et al [41] stated that the
applicability of this method is much wider than previously conjectured: n factors of the order of 9 to 11
correlate experiments at low and high speeds and include the effects of TollmienSchlichting, Gllrtler and
cross flow instabilities. But what does "low background disturbance level" mean? Is a low turbulence wind
tunnel representative of flight conditions? In this respect, the flight experiments on the AEDC cone
(Fisher and Dougherty, [42]) provided us with some very interesting information. This cone was mounted
on the nose of an F15 aircraft and flown at Mach numbers from 0.5 to 2 and at altitudes from 1500 m to
15000 m ; the measured transition locations were well correlated with n factors ranging from 9 to 11
(Malik, [43]). However, as it was pointed out by Bushnell et al. [41], in the absence of engine
noise/vibration, the n factors could be of the order of 15, or greater!
The en method was also used by Malik et al [44] for the rather complex reentryF experiment. The
reentryF flight vehicle [45] consisted of a 50 semivertex cone with an initial nose radius of 2.54 mm.
Stability calculations were performed for an altitude of 30.48 km, where the freestream Mach number was
rJRx:
layer. For instance, when the wall is cooled, the shape factor H decreases, as well as the ratio R5 1
As
far as the influence of T wrr aw on transition onset is concerned, figure 9 shows experimental results
collected by Potter [46] : the ratio RXT/RxTO is plotted as a function of T wrraw for Mach numbers between
1.4 and 11. RXTO is the transition Reynolds number measured under adiabatic conditions. Although there is
some scatter in the data, it appears that cooling the wall delays transition onset. This effect is rather strong
in the transonic range, but it is greatly reduced when the Mach number increases. Another interesting feature
which can be observed in figure 9 is the appearance of "transition reversals" and "transition rereversals"
(Me = 3.54 and 8.2). The origin of this behaviour has not been clearly established.
4
_ Cone
 Flat plate
RXr
RXro
Me = 3.54
8.2
, ' I,\\
/I
"
6.8
~ \ "
6.0,___
4.9
......6.8'::..,..... _
~ ......
_
......
......
""::...':::::::~~==
11.1
0.2
0.4
==:4.7
0.6
....... Tw/Taw
0.8
1.0
The stabili ty of laminar boundary layers on cooled walls was studied by Boehman and Mariscalco [47] in
transonic flow, by Mack [15], Wazzan et al [48], Malik in supersonic flow [49]. Systematic computations
were performed at ONERNCERT by Vignau [50].
246
Ur
t
4
Ur
_ '0
cr;
0
O~5
(b)
2
=3 4
_ 10
(C 3
cr;
0
~e=
C _
    _ _ REi,
o
80000
40000
40~
REi,
800CD
Two examples of stability diagrams [50] are given in figure 10 for Me = 7, 'V = 0 (twodimensional
waves), Twrraw = 1 and 0.3. ar and ai are made dimensionless with the displacement thickness ~1, and
R~1 = Ue~l/Ve. A noticeable feature is that increasing wall cooling tends to separate the second mode
from the first one: for T wrr aw = 0.3, two distinct neutral curves are observed, as it was the case on
adiabatic walls at lower Mach numbers.
Rb1=
Max (O(i)
30000
26000
17500
0.005
0.05
rnx=150C/
0(1
11)
made dimensionless
with 01
R~1
between
16700 and
33000 )
b)
TwlTaw
Tw/Taw
1~~O~.5~~O
for Me = 7, '" = 0
As T wtraw decreases, the instability of the first mode decreases and vanishes for sufficiently low values
of the cooling mte. As far as the second mode is concerned, figure lla shows the evolution of its
maximum amplification mte as a function of Twtraw for three values of the Reynolds number. It is clear
that the effect of wall cooling is rather small. However, if another reference length is chosen, the
conclusions seem to be different. For instance, the maximum amplification rate which is plotted on
figure lIb was made dimensionless with L = (Vex/Ue)l/2. In this case, it can be said that cooling is
destabilizing. Obviously, when the n factor is computed, the results do not depend on the reference length.
It is interesting to note that the en method (with a fixed value of the n factor) can reproduce the
experimental trends shown on figure 9 : wall cooling has a strong stabilizing effect at low Mach numbers.
This effect is less and less apparent at higher speeds.

a,..
a"
7.
_10 (Xi
=0
0.3
0.9
7.
 10 (Xi
=0
==B
la)
I b)
0
0
10000
20000
a
0
10000
20000
lei
10000
20000
Vignau [50] analyzed the influence of positive pressure gradients for Me varying linearly with x. The
stability diagrams presented on figure 12 were obtained for the same value of the local freestream Mach
number (Me = 5.8) and three pressure gradients, which are characterized by the value of the incompressible
shape factor Hi : zero pressure gradient (Hi =2.99), mild positive pressure gradient (Hi =3.46) and strong
positive pressure gradient (close to separation, Hi = 4.04). The wall is adiabatic and twodimensional
waves are considered.
When Hi increases, the first mode is stabilized, but the instability of the second mode becomes more and
more important. A striking feature of the stability diagram associated with the strong pressure gradient is
the appearance of a third unstable mode. Similar evolutions were also computed for Me = 5 and 7 [50].
5.3. Nose bluntness
The effect of nose bluntness on transition has been studied experimentally by many investigators due to its
importance for many hypersonic configurations. Stetson et ai, for instance [52], studied the transition
mechanisms on a cone which could be equipped with interchangeable spherically blunted noses of various
radii. The free stream Mach number was 8 and the halfangle of the cone 7. The main result was that
transition location moved downstream for small values of the nose radius, in agreement with other
experiments (for instance the experiments of Potter and Whitfield [53] on flat plates).
On the other hand, the destabilizing effect of centrifugal forces on concave walls leads to the formation
of pairs of counterrotating vortices, the axes of which are parallel to the principal flow direction
(figure 13). This instability, which was first treated by GOrtier [56], acts in two ways on the boundary layer
development. At first, the GOrtler vortices affect indirectly the transition process by modifying the
development of unstable waves and the breakdown to turbulence occurs earlier than on a flat plate. The
second aspect is essentially important in compressible flows: the vortices induce span wise variations of
the wall heat flux, which result in strong nonhomogeneities of the wall temperature in the laminar regime.
In fact, the appearance of GOrtler instability is not necessarily linked with the presence of a concave
surface; arrays of longitudinal vortices can be observed on flat plates as soon as the model geometry
=w
(12)
(y) sin(az) exp(crx)
where a is the wavenumber in the spanwise direction and cr the spatial growth rate in the xdirection along
the curved wall. a and cr are usually made dimensionless with the characteristic length L = (vooX/uoo)1/2.
As for the TollmienSchlichting waves problem, relations (12) are substituted into the linearized NavierStokes equations. Keeping the leading term leads to an eighth order system of homogeneous, linear,
ordinary differential equations with homogeneous boundary conditions. This forms an eigenvalue problem
for the three real parameters a, cr and G. G is the Gortler number, defined as :
G = uooL
Voo
(1.)
1/2
R
(13)
100
100
0=
0= 10
5
20
10
10
0.1
(al
0.1
0.01
0,1
(b 1
Q
01 0.01
Figure 14Stability diagrams
Q
0.1
250
Two stability diagrams are presented in figure 14 for an adiabatic wall and two Mach numbers (Moo = 0
and 3). Curves of constant amplification rate are plotted in the G()rtier numberwavenumber plane. The
shape of these curves does not change very much when the Mach number increases; as for the TollmienSchlichting waves, there is a critical G()rtler number, G cr , below which the flow is stable for any
disturbance wavenumber. It can be observed that compressibility has a stabilizing effect: as Moo increases,
the maximum growth rate at a given G()rtler number decreases; however, the influence of compressibility
on reducing cr becomes very small at high values of G. Computations by EI Hady and Verma [62] showed
also that G()rtier instability is more difficult to influence and control by suction or cooling than TollmienSchlichting instability. Malik [23] demonstrated that real gas has very little effect on the G()rtier vortices.
It is then possible to calculate the total amplification rate InNAo by integrating the local growth rate in
the stream wise direction. For the TollmienSchlichting waves, this integration is carried out for each
frequency; for the G()rtler vortices, it is performed for each spanwise wavelength. This was done by EI
Hady and Verma, who concluded that compressibility reduces the maximum amplitude ratio by 20 % as
Mach number increases from 0 to 5. Beckwith et al [63] used the en method in order to predict transition
location along the nozzle walls of the "quiet tunnel" ; they found that TollmienSchlichting instability is
weak, and that transition results in the growth of G()rtier vortices along the concave walls. Nevertheless,
calibration of the en method for the G()rtier instability remains an open question.
Quite recently, a new approach has been developed to study the linear stability problems for TS waves
(Herbert [64]) as well as for G()rtier vortices (Spall and Malik [65]). In this approach, it is now assumed
that the amplitude functions of the disturbances (which depend on y only in the framework of the classical
linear theories) depend also on the streamwise distance x. For the G()rtier vortices, the term exp(crx)
disappears, and the disturbances are written as :
r' = r (x, y) cos(az) with r' = u', v', p', or T'
(14)
w' =w (x, y) sin(az)
Substituting the previous expression into the stability equations leads to a system of partial differential
equations of parabolic type which can be solved using a marching procedure in x (pSE : Parabolic Stability
Equations). This approach makes it possible to take into account the non parallel effects.
Spall and Malik [65] numerically studied the development of G()rtler vortices in supersonic and
hypersonic boundary layers by solving the parabolic partial differential equations. They determined the effect
of various initial conditions on G()rtler instability and computed growth rates which were somewhat
different from those given by a normal mode solution, although the trends were identical.
5.5. Bypass by large roughness elements
It has been shown in the previous paragraphs that the linear stability theory, associated with the en method,
can describe some of the essential features of the transition process when "natural" instability waves are
present However, these waves no longer appear if the amplitude of the forced disturbances is too large, so
that the problem becomes fully non linear: it is a "bypass" (Morkovin [66]), in the sense that the linear
mechanisms are completely ignored. As there is no general bypass theory, it is necessary to develop
different criteria for each type of bypass (the word criterion must be interpreted as an empirical correlation
between boundary layer and flow parameters at transition onset).
Boundary layer tripping by large roughness elements constitutes a typical example of bypass. When
isolated tripping devices (spheres, small cylinders normal to the wall, ... ) are introduced into a nominally
twodimensional boundary layer developing on a flat plate, wall visualisations show the existence of
horseshoe vortices wrapped around the roughness element, see figure 15. At a distance L from the tripping
device, the "legs" of these vortices break down and create a turbulent wedge. When the height k of the
protuberance increases, L decreases up to a minimum value which is reached for an "effective" roughness
size, kerr. In other words, L remains constant for k > kerr.

vortices
251
"tur bulent"
wedge
t
_x
Figure ISExample of wall visualization using thermosensitive paint
From a practical point of view, the value of kerr is of great importance. Van Driest and Blumer ([67] to
[69]) performed a series of experiments in order to deduce empirical correlations between Rkeff, RXk and
the flow parameters; Rkeff and RXk are the Reynolds numbers formed with the effective roughness height
and the roughness location, respectively. The measurements were made on cones and correlated by the
following relationship:
Rkeff = 33.4
[1 + Y
(15)
Van Driest and Blumer assumed that this expression was also applicable to flat plate flows by using
Mangler's transformation, which simply consists in replacing the coefficient 33.4 by 33.4 (3)1/4 = 44.
Vignau [50] demonstrated that the modified correlation largely underestimates the effective roughness height
for flat plate flows, because Mangler's transformation is valid for mean flow properties, but it cannot be
used for stability and transition problems.
R~x
= (60 + 4.86
M~92) RX~
(16)
Me =
8
Figure 16Extent of the transition region
which reflects the experimental trend. Parameters such as wall temperature, unit Reynolds number, ... , are
not taken into account in these correlations. It was noted by Morkovin [73] that "the lateral or transverse
growth of a turbulent spot decreases from about 11 0 semiangle at low speeds to about half the angle at
hypersonic speeds". The same observation can be made for the spreading of a turbulent wedge behind an
isolated roughness element. Both phenomena involve the same physical mechanism which is called
"transverse contamination".
6.3. Transition region modellin!1;
From a practical point of view, the modelling of the transition region becomes a more and more important
problem when the Mach number increases; this is due to the fact that, for supersonic and hypersonic
flows, the evolution from the laminar to the turbulent state occurs along a stream wise distance which can
be much more important than the laminar region extent which precedes it.
253
Incompressl~le flows [74] and then extended to high speed conditions: it is assumed that the turbulent
(17)
Ilt is an eddy viscosity coefficient, which is computed by using a classical turbulence model. At first
sight, the ~oefficient E represents the intermittency factor y, in the sense that it increases from 0 in laminar
flow to 1 In turbulent flow.
1.5 r,.=~
1.01/;:::::".1
Experiments: Juillen
Fla t
pia te
relations
Calculations
OL~~
____
0.2
X(m)
 L_ _ _ _~_ _ _ _ _ __
0.4
254
10 Re/cm
o
~
} Experiments
Calcula tions
(w/(m~~)
5.5
2.76
1.46
____________________________ .
Ti " 1070 K
Tw = 295 K
8
6
4
2
O~LL~~
~x(m)
0.6
0.4
0.2
0.90
0.85
,)"""m",'
\'I/'!.":j!..j~~!..!.'..!!..'.!.'!.!
X(m)
0,80
0,15
0,20
0.25
0,15
0,20
0,25
X(m)
wall
streamline
256
disturbances; this leads to the formation of stationary, corotating vortices aligned in the local streamwise
direction. In the experiments, crossflow vortices are observed as regularly spaced streaks.
Experimental and numerical results on crossflow instability are not numerous for supersonic and
hypersonic flows. Balakumar and Reed [79] performed stability computations for a supersonic boundary
layer developing on a rotating cone at zero angle of attack. Transition on a cone at angle of attack was
investigated by Marcillat at Moo = 5 [80] and by Stetson et aI at Moo = 8 [81] ; Creel et aI [82] reported
measurements on circular cylinders at sweep angles of 45 and 60 ; the data were obtained in the Mach 3.5
"quiet tunnel" at NASA LANGLEY. In the following paragraph, the problem of crossflow instability is
illustrated by experiments performed in the supersonic R3Ch wind tunnel of the ONERA CHALAIS
MEUOON Center [83].
7.2. Example of theoretical and experimental results
The R3Ch wind tunnel is a free jet, blowdown wind tunnel with a circular nozzle, the exit diameter of
which is 34 cm. The freestream Mach number Moo is 10, for a stagnation temperature and a stagnation
pressure equal to 1 100 K and 120 bars, respectively (unit Reynolds number close to 10 106 m l ).
Static pressure
+ Thermocouples
Z(m)
The model is a swept circular cylinder (diameter D = 6 cm) equipped with an hemispherical nose, see
figure 22. In the present experiments, the sweep angle cp is equal to 40. X and Z represent the directions
normal and parallel to the attachment line (and to the cylinder axis), respectively. Let us recall that the
attachment line is the line where the static pressure is maximum on an infinite swept body of constant
chord. As X is measured along the surface of the model, the azimuthal angle 9 is equal to 2X/D. The
attachment line corresponds to X =0, i.e. 9 =0. In fact two models were used: one is made of silastene,
in order to perform wall visualizations by using a thermosensitive paint; the second one is made of steel
and it is equipped with a row of thermocouples and a row of static pressure taps distributed in the span wise
direction Z, see figure 22.
Figure 23 shows a wall visualization obtained with the first model. For 9 = 40, where the flow is
strongly accelerated, one can observe streaks which are practically aligned with the external streamline.
These streaks constitute the signature of the cross flow instability, which amplifies zero frequency,
stationary waves. The linear stability theory was used in order to see if it was possible to predict the
wavelength of the streaks. The laminar boundary layer was first computed from the measured pressure
distribution, which closely agreed with the newtonian law. The mean velocity and mean temperature
profiles obtained at 9 = 40 were then introduced as basic profiles in the stability code. It was assumed that
the disturbances were expressed by :
r' = r(y) exp( aiX) exp [i(arX + PrZ)]
(18)
A(mm)::2n/(a~ +f3~)1!2
2
Ej
Theory
ljJ(O)
~~~~~~~~~
500
o~~
...... Tw{K)
______
~______~
300
400
500
=WeTlIv
(19)
258
1\
1\
[dU]
~ X::::O
U e and We are the freestream velocity components in the X and Z directions. For R > 245, leading
edge contamination occurs: the turbulent structures generated by a source of gross disturbances become
selfsustaining and develop in the spanwise direction. They are damped and disappear for R < 245. Poll
[86] extended this criterion to compressible flows by introducing a transfonned Reynolds number R
which has the same definition as R , except that the kinematic viscosity v is replaced by v* which is
computed at a reference temperature T*.
O.
O~.
D=4cm
D=6cm
400
Tripping
00 No tripping
e Not clear
(intermittent?)
..
200
POLL
0 0 0
Leading edge contamination experiments were carried out by placing roughness elements on the
attachment line of swept cylinders in the R3Ch wind tunnel (paragraph 7.2.). As an example of result,
figure 26 shows a threedimensional picture of the distribution of the wall heat flux coefficient h for a
cylinder of diameter D = 6 cm and an angle of sweep cp = 40 ; this corresponds to R * = 326. The
spanwise position is given by the thennocouple number, see figure 22. The roughness element is a small
cylinder nonnal to the wall; its height k is equal to its diameter; in the present case, k = 0.6 mm, about
two times the boundary layer thickness. An important increase in h is observed around the tripping device,
which constitutes the apex of a turbulent wedge developing with a spreading half angle close to 15. At the
last thennocouple position, the boundary layer is fully turbulent.
Similar experiments were carried out for several values of D, cp and k. The results are summarized on

1\
figure 27 in the ( R *, k/11 *) plane, with 11 * = (v* / k) 1/2. For large values of
tripping becomes effective as soon as R * exceeds a critical value close to 250, in agreement with Poll's
criterion (see also Da Costa and Alziary [87]).
Additional measurements were perfonned by placing the roughness element at non zero values of Elk: (9k
denotes the azimuthal angle of the device location). It was found that the minimum roughness height
On the theoretical point of view, the linear stability theory constitutes a very efficient tool to understand
the fundamental mechanisms leading to transition and sophisticated experimental studies gradually confmn
these theoretical elements. The linear theory can also explain, at least qualitatively, the influence of more or
less controlled parameters (bluntness, wall cooling for instance), but the key problem lies in the
understanding of the receptivity mechanisms and the exact relation between instability and transition is not
very well known. Despite these facts, the extension of the en method in supersonic and hypersonic flow
could provide a reliable prediction of transition; for a low disturbance environment, the n factor has
roughly the same value as in incompressible flow.
As far as the transition region is concerned, we believe that simple methods, such as the intermittency
method, are able to give right predictions in very different situations. The inaccuracies of these techniques
are certainly small as compared with large errors which can arise in the prediction of the transition onset.
As examples of bypassinduced transitions, the effects of isolated, threedimensional roughness elements
have been studied. For twodimensional mean flows, the first effect of these tripping devices is to generate
turbulent wedges, the apexes of which move upstream when the roughness size increases. As this movement
cannot be described by the linear stability theory, purely empirical criteria need to be developed. The second
effect is the appearance of horseshoe vortices around the roughness element; these vortices are responsible
for very large values of the wall heat flux, as it is also observed for Gl>rtler vortices as well as for crossflow
vortices; in fact, the problem of minimizing the heat transfer rates in hypersonic boundary layers requires
not only to delay transition, but also to avoid stream wise vortices in the laminar boundary layer. For threedimensional mean flows, it has been demonstrated that Poll's criterion is valid at high speeds, but that its
validity is restricted to the attachment line, which is the location where the boundary layer is the most
sensitive to roughness elements. At the present time unfortunately, there is no general criterion which could
be able to predict boundary layer tripping in a broad range of practical conditions.
REFERENCES
1.
Morkovin, M.V. Critical evaluation of transition from laminar to turbulent shear layers with
emphasis on hypersonically travelling bodies, Report AFFDLTR68149, WrightPatterson Air Force
Base, Ohio, 1968
2.
Morkovin, M.V. Transition at hypersonic speeds, ICASE NASA Contractor Report 178 315,1987
3.
Reshotko, E. Boundary layer stability and transition, Annual Review in Fluid Mechanics, Vol. 8,
1976
4.
Mack, L.M. Boundary layer linear stability theory, AGARD Report nO 709,1984
5.
Pate, S.R. Effects of wind tunnel disturbances on boundary layer transition with emphasis on radiated
noise: a review, AIAA Paper 800431, Colorado Springs, 1980
6.
Amal, D. Laminarturbulent transition problems in supersonic and hypersonic flows, AGARD Report
nO 761, 1988
7.
Stetson, K.F. Hypersonic boundary layer transition, 3rd Joint Europe/US Short Course in
Hypersonics, Aachen, 1990
Schubauer, G.B. and Skramstad, H.K. Laminar boundary layer oscillations and transition on a flat
plate, Report 909 NACA, 1948
9.
Schlichting, H. Boundary layer theory, 6th ed. McGraw Hill, New York, 1968
10. Lees, L. and Lin, C.C. Investigation of the stability of the laminar boundary layer in a compressible
fluid, NACA TN W 1115, 1946
11. Lees, L. The stability of the laminar boundary layer in a compressible fluid, NACA TN W 876, 1947
12. Dunn, D.W. and Lin, C.C. On the stability of the laminar boundary layer in a compressible fluid, J.
aero. Sci., Vol. 22, pp. 455477, 1955
13. Lees, L. and Reshotko, E. Stability of the compressible laminar boundary layer, J.F.M., Vol. 12, Part
4,pp.555590, 1962
14. Mack, L.M. Stability of the compressible laminar boundary layer according to a direct numerical
solution, AGARDograph 97, Part I, pp. 329362, 1965
15. Mack, L.M. Boundary layer stability theory (2 volumes), Jet Propulsion Laboratory, California Inst.
of Techn., Pasadena, California, November 1969
16. Mack, L.M. Linear stability and the problem of supersonic boundary layer transition, AIAA J., Vol.
13, N 3, pp. 278289, 1975
17. Mack, L.M. Transition prediction and linear stability theory, AGARD Conf. Proc. nO 224, Paris,
1977
18. Mack, L.M. Compressible boundary layer stability calculations for sweptback wings with suction,
AIAA J., Vol. 20, pp. 363369, 1981
19. Mack, L.M. Remarks on disputed numerical results in compressible boundary layer stability theory,
Phys. Fluids 27(2), February 1984
20. Gaster, M. A note on the relation between temporally increasing and spatially increasing disturbances
in hydrodynamic stability, J.F.M., Vol. 14, pp. 222224, 1962
21. Arnal, D. Stabilite et transition des couches limites laminaires bidimensionnelles en ecoulement
compressible, sur paroi athermane, La Recherche Aerospatiale nO 19884, 1988
22. Malik, M.R. Transition in hypersonic boundary layers, Numerical aJld Physical Aspects of
Aerodynamic Flows IV, Springer Verlag, T. Cebeci ed., 1990
23. Malik, M.R. Stability theory for chemically reacting flows, IUTAM Symp. "LaminarTurbulent
Transition", Springer Verlag, D. Amal and R. Michel ed., 1990
24. Gasperas, G. Stability of the laminar boundary layer for an imperfect gas, IUTAM Symp. "LaminarTurbulent Transition", Springer Verlag, D. Amal and R. Michel ed., 1990
25. Stuckert, G.K. and Reed, H.L. Stability of hypersonic, chemically reacting viscous flows, AIAA
Paper 901529,1990
26. Goldstein, M.E. The evolution of TS waves near a leading edge, J.F.M. Vol. 127, pp. 5981, 1983
27. Kerschen, E.J. Boundary layer receptivity, AIAA Paper 891109,1989
30. Normand, X. Transition la turbulence dans les ecoulements cisailles compressibles Iibres ou
parietaux, Grenoble University, Thesis. 1990
31. Thumm, A, Wolz, W. and Fasel, H. Numerical simulation of spatially growing threedimensional
disturbance waves in compressible boundary layers, IUTAM Symp. "LaminarTurbulent Transition",
Springer Verlag, D. Amal and R. Michel ed., 1990
32. Masad, I.A. and Nayfeh, A.H. Subharmonic instability of compressible boundary layers, Phys. Fluids
A, Vol. 2, nO 8, pp. 13801392, 1990
33. Smith, A.M.O. and Gamberoni, N. Transition, pressure gradient and stability theory, Douglas Aircraft
Co. Rept ES 26388, EI Segundo, California, 1956
34. Van Ingen, J.L. A suggested semiempirical method for the calculation of the boundary layer transition
region, Univ. of Techn., Dept. of Aero. Eng., Rept. UTH74 Delft, 1956
35. Chen, F. J. and Malik, M. R. Comparison of boundary layer transition on a cone and flat plate at
Mach 3.5, AIAA Paper 880411,1988
36. Juillen, J.C. Determination experimentale de la region de transition sur une plaque plane
et 7, ONERA Technical Report W 10/2334 AN, 1969
a M = 5,6
37. Harvey, W.D. Influence of free stream disturbances on boundary layer transition, NASA Technical
Memorandum 78635,1978
38. Beckwith, I. E., Creel Jr, T. R., Chen, F. J. and KendaIl 1. M. Freestream noise and transition
measurements on a cone in a Mach 3.5 pilot lowdisturbance tunnel, NASA TP 2180,1983
39. Beckwith, I.E., Chen, F., Wilkinson, S., Malik, M. and Tuttle, D. Design and operational features of
lowdisturbance wind tunnels at NASA Langley for Mach numbers from 3.5 to 18, AIAA Paper 901391, 1990
40. Beckwith, I.E. Development of a high Reynolds number quiet tunnel for transition research, AIAA
Journal, Vol. 13, W 3,1975
41. BushneIl, D.M., Malik, M.R. and Harvey W.D. Transition prediction in external flows via linear
stability theory, IUTAM Symp. Transsonicum III, G&tingen, Germany, May 2427,1988
42. Fisher, D.F. and Dougherty, N.S. Inflight transition measurements on a 10 cone at Mach numbers
from 0.5 to 2, NASA TP 1971, 1982
43. Malik, M.R. Instability and transition in supersonic boundary layers in Laminarturbulent boundary
layers (E.M. Uram and H.E. Weber Ed.)  Proc. of Energy Resources Technology Conference, New
Orleans, Louisiana, 1984
44. Malik, M.R. Boundary layer transition in hypersonic flows, AIAA Paper 905232,1990
45. Johnson, C.B., Stainback, P.C., Wiker, K.C. and Bony, L.R. Boundary layer edge conditions and
transition Reynolds number data for a flight test at Mach 20 (Reentry F), NASA TM X2584, 1972
Beckwith, J.E., Malik, M.R., Chen, FJ. and Bushnell, D.M. Effects of nozzle design parameters on
the extent of quiet test flow at Mach 3.5, IUTAM Symp. on LaminarTurbulent Transition,
Novosibirsk, Springer Verlag, Ed. V.V. Kozlov, 1984
Coleman, G.T., Elfstrom, G.M. and Stollery, J.L. Turbulent boundary layers at supersonic and
hypersonic speeds, AGARD Conference Proceedings N 93,1971
76
Poll, D.I.A. Transition description and prediction in threedimensional flows, AGARD Report N 709,
1984
77
78
Saric, W.S. and Reed, HL Threedimensional stability of boundary layers. Perspectives in turbulence
studies, Springer Verlag, 1987
79. Balakumar, P. and Reed, H.L. Threedimensional stability of boundary layers, Phys. Fluids A3(4),
1991
80
81
Stetson, K.F., Thompson, E.R., Donaldson, J.C. and Siler, L.G. Laminar boundary layer stability
experiments on a cone at Mach 8  Part 3 : sharp cone at angle of attack, AIAA Paper N 850492,
1985
82
Creel Jr., T.R., Beckwith, I.E. and Chen, FJ. Transition on swept leading edges at Mach 3.5, J.
Aircraft, Vol. 24, W 10, 1987
264
83
Amal, D. and Laburthe, F. Recent supersonic transition studies with emphasis on the swept cylinder
case, Conf. on Boundary Layer Transition and Control, Cambridge, 812 April, 1991
84
Malik, M. R. and Beckwith, I. E. Stability of a supersonic boundary layer along a swept leading edge,
AGARD CP N 438, 1988
85
Pfenninger, W. Flow phenomena at the leading edge of swept wings, AGARDograph 97, Part 4,1965
86
Poll, D. I. A. Boundary layer transition on the windward face of space shuttle during reentry, AIAA
Paper 850899,1985
87
88
Morrisette, E.L. Roughness induced transition criteria for space shuttletype vehicles, J. Spacecraft,
Vol. 13, W 2, 1976
Chapter 8:
SecondOrder Effects in Hypersonic Laminar
Boundary Layers
J. Ph. Brazier, B. Aupoix, J. Cousteix
ONERACERTDERAT, 2 avenue Edouard Belin,
B.P. 4025, 31055 Toulouse Cedex, France
ABSTRACT
In hypersonic flows, the bow shock wave in front of the body is the cause of a vortical inviscid
flow in the shock layer. The conventional boundary layer theory of Prandtl can not deal
with it. Van Dyke has derived an extended boundary layer theory using matched asymptotic
expansions where several second order effects as outer flow vorticity, displacement, or wall
curvature are accounted for. However, the matching between the boundarylayer flow and
the external inviscid flow is not perfect with second order expansions when the Reynolds
number is low and thus the boundary layer thick. To improve this and ensure a smooth
merging of the viscous flow into the inviscid one, a defect approach coupled with asymptotic
expansions has been proposed. Calculations on simple twodimensional hypersonic bodies
are performed using the two methods and compared to full NavierStokes solutions.
INTRODUCTION
An hypersonic flow past a blunt body is characterized by a bow shock wave, detached in
front of the nose. For an inviscid flow, the entropy remains constant along a streamline,
except through the shock wave, where the entropy jump depends on the local slope of the
shock, and thus varies from one streamline to another one because of the shock curvature.
An entropy gradient orthogonal to the streamlines is thus created in the shock layer. On
the rear part of the body, the shock wave tends to be straight and the entropy gradient
disappears. The entropy gradient is thus confined in the vicinity of the wall in a region
called "entropy layer". This layer represents the streamlines which crossed the curved shock
wave near the nose.
This entropy gradient is related to the vorticity of the flow field through Crocco equation :
c~;l iT 1\ iT =  grad H t + T grad S
This means that the streamwise component of the velocity varies through the entropy layer
along the normal to the body (fig. 1). Since the total enthalpy is constant in the whole flow,
the temperature varies too.
A reliable computational method for viscous flows must include the influence of the
vortical inviscid flow on the boundary layer. Solving the whole NavierStokes equations
266
Ro
p~
TO'
= U;'; ICp
fLo
= fL*(TO')
reference temperature
reference viscosity
~~
.,,*
= R*
longitudinal velocity: u
= ;:
00
normal velocity: v
v*
=
U;"
p*
density: p = 
p~
pressure: p
p*
=U2
P* *
00
temperature: T
tY : fL
VISCOSI
enthalpy: h
00
= T*
TO'
= fL*
fLo
=
h*
U*2
00
These variables remain bounded in the stagnation region when the Mach number tends
towards iniinity (Van Dyke [19]). For an ideal gas, we then get:
h= T
268
 (momentum:
 7]momentum :
 energy:
1 oh
pu hl o(
oh
op
op
+ pv 07] = hl o( + v 1}7]
1 I) [ I'
h3 I}T]
1 I} [I'
I}T]
+hlh31}( Pr Re hl I}( + hlh31}7] Pr Re hlh31}7]
I'
+Re
[2 (hI1I}u
V I}hl)2
2(OV)2 2( u I}h 3 V I} h 3)2
I)( + hl 1}7]
+ 1}7] + hlh3 I}( + h3 1}7]
1 I}v
+ ( hI I)( +
(U ))2
I)
11}7]
hI
2 ( 1 I}u
3" hI
 state:
I}(
I}v
+ 1}7] + hlh3
,I
p=   p T
 sound velocity:
f! = j
(r  I)T
= To
_ Reynolds number: Re
I'*C*
I}h3
I}(
V I}hl
1}7]
+ hl
V I}h3 )2]
1}7]
+ h3
= 1 + ~ = 1 + K.TJ
where R( {) is the longitudinal curvature radius of the body, and r( {) is the transverse
curvature radius; a({) is the angle of the tangent to the body with the axis of symmetry
(fig. 2). For a plane flow j = 0 whereas for an axisymmetric flow j = 1.
r(FJ
_ _ _ _ a...._
l/K(r,;)
4~
v=Q
p=l
P = ,M;"
wall :
v=O
T= Tw
where Tw is the wall temperature. Instead of fixing the wall temperature, one can as well
impose the wall heat flux, or a relation between the temperature and the flux. The velocity
slip and temperature jump effects happen at the wall only for rarefied gases and will not be
accounted for here. Van Dyke [19) showed them to be secondorder effects.
Asymptotic expansions
Two approximations of the NavierStokes solutions will be sought for: one far from the
wall, in the region where the viscous effects are weak, and another one near the wall, where
270
the viscous effects are important. These approximations are looked for as expansions in
powers of a small perturbation parameter :
1
e:= ffe=
p.(To)
1
p.(T00) VReoo
Van Dyke [19J showed e: to be the right similarity parameter in the hypersonic limit when
both the Mach and Reynolds numbers tend towards infinity.
The outer expansions read:
u((,.,,)
v((,.,,)
= VI ((,.,,) + e:V2((,.,,) +
p( ( , .,,)
= PI ( ( , .,,) + e: P2 ( ( , .,,) +
= R1 ((,.,,) + e: R2(e,.,,) +
= TI(e,.,,) + e:T2(e,.,,) +
p((,.,,)
T(e,.,,)
In the inner region, a stretched normal coordinate fi = .,,/e: is used according to the principle
of least degeneracy (Van Dyke [21, 19]). In this way, the normal coordinate is referred to a
quantity of the same order of magnitude as the boundary layer thickness, so that the new
normal coordinate is of order unity in the boundary layer. Similarly, a special expansion
must be written for the normal velocity because this quantity is of order e: in the boundary
layer, according to the continuity equation. Therefore, the inner expansions have the form:
u(c.,,)
v(e,.,,)
p(e,.,,) =Pl((,fi)
+ e:P2(e,fi) +
+ e:P2(e, fi) +
+ c:i2(e, fi) +
T((,.,,)
= t1(e, fi)
All the coefficients and their derivatives are assumed to be of order unity.
Outer and inner expansions
These expansions are brought in the dimensionless NavierStokes equations and the like
powers of e: are equated. This leads to the following equations:
(momentum:
7]momentum :
energy:
state:
P2
,1
= (RIT2 + R2Tl)
These are just the Euler equations linearized for small perturbations. No viscous terms
appear until third order in the outer region.
These are the usual Prandtl equations. Note that the curvature radius only appears in the
continuity equation.
 inner region, second order:
continuity:
(momentum:
TJ momentum:
energy:
state:

+
0, it yields:
r; +
00 :
(1)
The secondorder inner term is thus matched with the wall value of the secondorder outer
term, plus the slope at the wall of the firstorder outer term. The slope at the wall of the
secondorder term arises only in thirdorder expansion, with the wall value of the secondorder derivative of the firstorder term. The same argument is also valid with p, p and T.
The case of v is slightly different:
So:
or:
first order
second order
Figure 3: matching schemes
The other boundary conditions are given by the freestream conditions for the oncoming
flow, and the noslip condition and fixed temperature at the wall :
at infinity:
Pa = 0
OPl
or,
we get Pl(,7]) = Pl(,O). The pressure is thus constant across the boundary layer and
equal to the inviscid flow pressure at the wall.
A single composite expansion valid in the whole domain can be written as the sum of
the internal and external expansions minus the common part:
This development is equivalent to each expansion in the corresponding region. The slope at
the wall of the velocity and temperature profiles is now equal to the boundary layer slope
plus the inviscid flow slope.
Second order effects
The secondorder boundary layer equations are linear and thus the source terms and the
boundary conditions can be split in several parts corresponding to different second order
effects (Van Dyke [19]) :
longitudinal curvature (terms proportional to K)
transverse curvature (terms proportional to j cos Q)
displacement
entropy gradient
total enthalpy gradient
velocity slip
temperature jump
276
These effects can be solved separately, except the displacement and entropy gradient effects
which are coupled, since in the secondorder outer equations appears the term
which is the product of the displacement velocity by the shear of the first order inviscid
flow. Thus, for a given firstorder boundary layer, the secondorder flow will be different
depending on whether the external flow is sheared or not.
DEFECT APPROACH
Decomposition
As written above, the secondorder matching is not perfect when the external inviscid profiles
are not linear across the boundary layer. This may happen when the Reynolds number is
moderate and thus the boundary layer is thick. To ensure a smooth matching at any
order whatever the external flow, a defect approach has been used, coupled with asymptotic
expansions. In the inner region, the variables are no longer the physical variables, but the
difference of them with the external solution (Le Balleur [13]). So we write:
= PE + PD
= VE + VD
 VE(('O)
= PE + PD
T = TE + TD
P
where the subscript E stands for the external variables and the defect variables are labelled
D. The term VE(CO) has been added to keep the condition VD(('O) = O.
Expansions are then written using the same small parameter e as Van Dyke, with the
coordinates ((,71) for the external functions and the stretched ((, r;) coordinates for the
defect variables:
= Ul((,71) + eU2((,71) +
VE((,71) = Vl(~,71) + eV2(~,71) +
PE((,71) = Pl((,71) + eP2((,71) +
PE((,71) = Rl(~,71) + eR2(~,71) +
TE((,71) = T1 ((,71) + eT2(~,71) +
UD(~,71) = Ul((,r;) + eU2(~,r;) +
VD((,71) = eih((,r;) + e2v2(~,r;) +
PD((,71) = Pl((,r;) + ep2((,r;) +
PD((,71) = Pl((,r;) + ep2((,r;) +
TD(~,71) = tl(~,r;) + et 2((,r;) +
UE(C71)
VI(~,1/)  VI(~,O)
= eVi(~,1/)
V2(~,1/)  V2(~,O)
= eV2(~,1/)
The other coefficients are assumed to be of order unity, as well as their derivatives relative
to (~, 1/) for the external variables and (~, 7)) for the defect ones. This supposes that the
normal gradients in the inviscid flow are lower than those in the boundary layer, since they
do not have the same scale of magnitude.
Equations
In the outer region, the defect variables are null and the equations for the outer flow are
exactly the same as for Van Dyke's theory, i.e. Euler equations. Concerning the inner region,
one must first bring the above expansions into the NavierStokes equations, then substract
the external equations, and then equate same powers of e. For practical convenience, the
inner equations can then be rewritten in outer coordinates, replacing 7) by 1/, and with
 {momentum:
 1/momentum :
 energy:
 state:
PI
=' 
1 [PITl
+ (RI + pdtl]
As in Prandtl equations, the walJ curvature appears in the firstorder equations only through
the transverse curvature radius in the continuity equation. The secondorder equations are
smallperturbations of the above ones plus source terms due to curvature effects, like in
278
+:'"

V2(~' 0))]
[",JRe(jcosa+Krj)P1(V1+vd] =0
~momentum
~
()~ + (R1 + pl)(U1 + U1) 8~
[(R1 + P1)(U2 + U2) + (R2 + P2)(U1 + ud ] ~
8U1
[
] 8U2
+ [P1U2 + P2 U1 + (R1 + Pl) U2 + (R2 + P2)U1 ] 8~ + P1U1 + (R1 + pdU1 8~
+ [R1(V1JRe 
+ KU1)
V2(~'0) + V2)
= 8P2
+
8~
~~
U
8/11 (8U1
+1 /11j cos a 8 1 + 1 JRe r
8",
JRe 8", 8",  ",momentum:
(U1 + U1 ))
+ [p t U2
+(Rt
+ PI)(UI + ud] ~~
K"7VRe (PIUI
+ (RI + PI)U I )] ~t
2 [
] 8T2
+ pI)(Ut + uI) 8t
8{ + PtUt + (Rt + pI)Ut 8{
It"7VRe(RI
V2({,O)
8t2
+ V2) + (R2 + P2)(Vt + VI) ] 8tt
8"7 + (Rt + PI)(VI + vd 8"7
8P2
8{
+ (U2 
IT>'":) 8P
8P2
8
+(vtvRe
 V2({'O) 8"7 + (Vt + vda;; + 8"7
8 [
+8"7
IJ.t ] 8Tt
Pr y'Re' 8"7
+1J.2
PE
+
PE
+
TE
8P2
+ (Ut + ud8[
( . cos
)r
8
+ 8"7
[ 1J.2 8tl]
Pr Re 8"7
UD
+
VD
+
VE({,O)
PD
+
PD
+
TD
+
IJ.t
+ K + Pr y'Re'
Matching conditions
At the edge of the boundary layer, we can write:
+
8Pt
[ IJ.t 8t2]
Pr Re 8"7
 state:
. IT>'":
"7 lt v Reu t) 8{
y'Re'
8t t
8"7
+ ud But
8"7
=0
lim Pl = 0
ij+oo
lim tl = 0
1j+00
lim Pl = 0
11+00
lim Ul
11+00
ij+oo
=0
=0
_lim P2 = 0
'1+ 00
lim t2
tj+oo
hence:
Ul(~'O)
U2(~'O)
Vl(~'O)
tl(~'O)
t2(~'O)
= Ul(~'O)
= U2(~'O)
= V2(~,O) = 0
= Tw  Tl(~,O)
= T2(~'O)
Discussion
Like in Van Dyke's theory, the coupling between the boundary layer and the external flow
is done by blowing at the wall. The chaining of the calculations is thus identical.
Using the above conditions, the first order 1/momentum equation reduces to
Pl
=0
So, the pressure in the firstorder boundary layer is everywhere equal to the local inviscid
flow pressure, instead of its wall value like in Van Dyke's theory. The second order defect
pressure is null too only on a plane surface, or neglecting curvature effects.
The defect boundary layer equations are parabolic and can be solved by space marching
at a very low cost, like the standard Prandtl equations.
From a theoretical point of view, it can be shown that the defect expansions are consistent with Van Dyke's ones by the fact that at a given order they differ only by terms which
are higherorder in Van Dyke's theory. If the flow over a plane surface is considered, the
a(pu)
a(pv)
ar+ {f;1
au
pu ae
+ pv a1J 
au
aUE
(pv  PEVE) 01J
aT
pu a(
+ pv a1J
aT
aTE
 (pv  PEVE) a."
a [
+a."
p, aTD]
Pr Re a."
p
which reduce to Prandtl's equations only when the inviscid flow is constant along the wall
normals. Note that only the defect variables appear in the viscous terms.
EXAMPLES OF APPLICATIONS
To experiment the defect approach, the first tries have been performed for an incompressible
shear flow past a flat plate. Second order calculations have been made with interesting results
([6, 7, 4, 5]).
Concerning compressible flows, several cases have been selected for a blunt body in a
hypersonic flow of ideal gas. The general shape of the body is a plane or axisymmetric
hyperboloid, defined by the nose radius and the angle of the asymptots, at zero degree incidence. The numerical data are given by Shinn, Moss and Simmonds [17] for an hyperboloid
equivalent to the windward symmetry line of the U.S. space shuttle. Two points of the
reentry trajectory of the STS2 flight are considered here:
Reentry trajectory  Flight STS2
26,6
Mach Moo
time (s)
250
altitude (km)
85,74
1,322
nose radius Ro (m)
asymptotes halfangle (0)
41,7
0,3634
pressure poo (Pa)
199
temperature Too (K)
7530
velocity Uoo (m/s)
6,3487.10 6
density Poo (kg/m 3 )
reference tempera.ture To (K)
56321
Reynolds number Re
small parameter e
= PooUooRo
p,(To)
= Re 1 / 2
= PooUooRo
23,4
650
71,29
1,253
40,2
4,0165
205
6730
6,7979.10 5
44900
183,55
1865,65
0,074
0,023
4792
42374
P,oo
The Prandtl number is assumed constant and equal to 0.725. The ratio of specific
heats I is 1.4. The wall temperature is fixed a.nd equal to 1500 K. The viscosity law is
282
Sutherland's. No comparison with experimental data is possible since the real gas effects are
not yet included. So NavierStokes solutions [12] have been taken as reference, to compare
the two Euler + boundary layer methods. Euler calculations are made with a code from
ONERA [23]. Classical boundary layer solutions are obtained using a program developed in
DERAT ([3]). Only firstorder boundary layer are presented here since secondorder outer
:flow solutions are not yet available. Several secondorder calculations using Van Dyke's
theory have been made on a hypersonic blunt body [1, 8, 9,10,11,14,15, 16].
Axisymmetric hyperboloid
In this case, the entropy layer is characterized by a nonzero entropy gradient at the wall and
a decreasing thickness towards the rear. Boundary layer profiles are displayed on figures 4
to 7 for the case Mach 23.4 . Longitudinal velocity are plotted on figure 4 at a distance of
four nose radius from the stagnation point. One can see on this figure the important velocity
gradient at the wall in the inviscid :flow. This gradient diminishes distinctly between the wall
and the boundary layer edge. So even with a secondorder expansion, Van Dyke's method
could not give a good matching, since it considers only the wall value of the gradient. In
this case, it would widely overestimate the skin friction (Adams [1]). Due to the very low
wall temperature compared to the inviscid :flow one, the displacement effect is quasinull
and the NavierStokes solution recasts exactly the inviscid profile in the outer region. In
this case, the agreement is excellent with the firstorder defect boundary layer, better than
Van Dyke's composite expansion.
The corresponding profiles for the normal velocity and the temperature are shown on
figures 5 and 6. The very slight discrepancy observed between the Euler and NavierStokes
profiles confirms that the displacement effect is very low in this case. So the first order
defect approach gives very good results. Figure 7 shows the temperature profiles at nine
nose radius. The growing boundary layer has overlapped a larger part of the entropy layer.
The defect profile is in rather good agreement with the NavierStokes solution, but in this
case the composite expansion written with Van Dyke's first order solutions gives very bad
results and does not improve the inner solution. This is due to the negative slope at the wall
for the inviscid temperature. Because of the constant total enthalpy, the positive velocity
gradient at the wall induces a negative temperature gradient. In spite of this, the wall
heat :flux is increased by the vorticity as the skin friction, as can be seen on the figures 8
and 9. But the increase is far more important for the wall friction than for the flux. The
defect approach underestimates slightly these quantities but gives better predictions than
the classical boundary layer.
Figures 10 and 11 show the velocity and temperature profiles with an arbitrary temperature of ten times the temperature of the preceding case. The displacement effect is then far
more important and it is obvious on these figures that the N avierStokes solution is shifted
from the Euler solution in the outer zone. So the firstorder boundary layer methods give
poor results and a secondorder calculation seems to be necessary.
The velocity and temperature profiles at four nose radius abscissa in the Mach 26.6
case are presented on figures 12 and 13. Because of the lower density, the Reynolds number
is small and the boundary layer is about twice as thick as in the Mach 23.4 case. So a
large part of the entropy layer is overlapped by the boundary layer. The inviscid velocity
and temperature gradients at the edge of the boundary layer are far weaker than their wall
values. Due to the high value of the expansion parameter c, the second order effects are
more important and a slight displacement effect is visible between the Euler and Navier
Eulerhl ..der
0.12
0.10
defect hI order
0.10
\
\
0.08
0.08
\
\
\
\
'" 0.08
I
/
0.02
. " ,,"
0.1
0.3
,
\
0.02
0.5
0.4
0.6
O.OO+r..,..r_+_
.(l.016
0.020
0.000
0.18
.(l.006
Eulerl.l ...d ..
oompoolte
0.010
v
0.18
Eulerl.l ...der
0.10
'..
0.12
'.
0.04
""",,/
0.2
\
\
O.OO~:IU'><'Tr""'"""''"""T"""
0.0
,,
,,
I
"'0.06
0.04
NavlerSloke.
NavlerStak..
1., ...
der
NavI...Stak..
NavI ...Stak..
0.14
0.12
0.08
'" 0.08
I
0.04
/
/
./
,.,.''i!'''
0.02
"."
,/
,,
,,,
/./
0.1
0.2
T
0.08
0.06
/'
0.04
,."
0.02
o.oo+_.oo<"""'r..r'
0.0
",0.10
0.3
0.4
0.00
0.0
0.1
0.2
T
0.3
0.4
0.007
0.006
NovierStokes
\ 0
0.005
~ 0.004
00
\
000
,
:E
~ 0.003
000
000
"
00000
0.002
........ ........
 
 

0.001
0.0001+,r..o
5
10
15
20
distonce oIong the body
= 1500 K
0.006
NovierStokes
0.005
0.004
)(
c2
0.003

0.002
0.001
0.000
0
10
15
20
= 1500 K
0.14
.....
Euler 1 st order
0.12
NavierStokes
0.10
0.08
""
,
,
,,
,,"""
0.04
0.02
"
I
I
0.12
0
I
I
NavlerStokes
I
I
I
I
0.10
I
I
I
I
I
I
I
I
0.08
""
0.06
0.14
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
:.0
,
I
0.04
'0 :
. I
0/
,," "0
0.02
,,,"0
"
0.00
0.00
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.4
I
I
0.0
0.1
0
Van Dyke 151 order
I
I
I
I
I
NavierStokes
I
I
0.4
0.3
0.2
T
0.4
0.3
o
0.06
0.3
NavlerStokes
I
I
I
I
""0.2
I
I
I
I
I
I
I
I
I
I
I
,
"
/'
,,'
"
,,
0.1
,.."
0.1
/.
"
,," """
.
O.O.;IlCI~r'.'~r...0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.0+a'IP<>W:"'',..,,'0.0
0.1
0.3
0.4
0.2
T
0.025
,
\
0.020
\
\
NavierStokes
\
\
\
:S.50!
0.015
.e
" 0.010
.ii
,
" , ...........
....... ......
0.005
  

O.OOO+,r,,,,,,,o
2
4
6
8
10
12
14
16
distance ciong the body
= 1500 K
o
Van Dyke 1st order
\ 0
0.020
0
\
NovierStokes
0.015
,
"
00
0 0
....
'........
....
00
000
.............
......
0.010
...
000
 
  
0.005
O.OOO+r____,.....~_r_r____,....._
6
8
10
12
distance along the body
14
16
= 1500 K
288
0.30
0.30
. . . .. Euler I order
_. Von Dyke 11. order
0.26
defect I order
Navte,S\ok..
0.20
s:
0.26
0.20
s:
0.16
0.16
0.10
0.10
0.06
0.06
'T,.,......_._'
O.OO ..........
0.00 0.06
0.10
0.16
0.20
u
0.26
0.30
0.36
=,...,,.....:
0.00+...........
0.0
0.1
0.2
0.4
Navie,Stok..
0.3
T
defect I order
0.8
Navie,S\ok..
0.8
Navte,Stok..
0.6
0.8
0.4
0.4
0.2
0.2
0.0+,1oLI:".,..,._C.,0.2
0.3
0.0
0.1
0.4
u
o.o++';loO.<~.r.__"T'
0.0
0.1
0.3
0.2
T
0.4
0.010
NovierStokes
0.008
.~
:g 0.006
.10
0.004
0.002
O.OOO~rrrr
10
15
20
= 1500 K
0.012
NovierStokes
10
15
0.010
)(
0.008
oi!
0.006
0.004
0.002
0.000
0
20
Figure 21: Wall heat flux on the hyperbola  Mach 26.6, T",
= 1500 K
References
[lJ Adams J .C. Higherorder boundarylayer effects on analytic bodies of revolution, Arnold
Engineering Development Center, Report AEDCTR"6857 (1968)
[2J Anderson D.A., Tannehill J.C., Pletcher R.H. Computational fluid mechanics and heat
transfer, Mc GrawHill (1984)
[3J Aupoix B., Arnal D. CLIC: calcul des couches limites compressibles, Rapport technique
OA 23/5005 AYD (1988)
[4J Aupoix B., Brazier J.Ph., Cousteix J., Monnoyer F. Secondorder effects in hypersonic
boundary layers, Third Joint EuropeU.S. Short Course on Hypersonics, Aachen (FRG),
October 15 (1990)
[5J Aupoix B., Brazier J .Ph., Cousteix J. An asymptotic defect boundary layer theory applied to hypersonic flows AIAA 910026, 29th Aerospace Sciences Meeting, January 710, Reno, Nevada (1991)
[6J Brazier J .Ph., Aupoix B., Cousteix J. Etude asymptotique de La couche limite en formulation dejicitaire, CompteRendus de l'Academie des Sciences, t. 310, serie II, pp.15831588 (1990)
[7J Brazier J .Ph. Etude asymptotique des equations de couche limite en formulation
dejicitaire, PhD thesis, Ecole Nationale Superieure de l'Aeronautique et de l'Espace,
Toulouse, France (June 25 1990)
[8J Davis R.T., FliiggeLotz I. The laminar compressible boundarylayer in the stagnationpoint region of an axisymmetric blunt body including the secondorder effect of vorticity
interaction, International Journal of Heat and Mass Transfer, Vol. 7, pp. 341370 (1964)
[9J Davis R.T., FliiggeLotz I. Laminar compressible flow past axisymmetric blunt bodies,
Journal of Fluid Mechanics, Vol. 20, Part 4, pp. 593623 (1964)
[10J Fannel6p T.K., FliiggeLotz I. Twodimensional hypersonic stagnation flow at low
Reynolds numbers, Z. Flugwiss. Vol. 13, nO 8, pp. 282296 (1965)
[l1J Fannel6p T.K., FliiggeLotz I. Viscous hypersonic flow over simple blunt bodies; comparison of a secondorder theory with experimental results, Journal de Mecaruque, Vol.
5, nO 1, pp. 69100 (1966)
[12J Lafon A. Calcul d'ecoulements visqueux hypersoniques, Rapport technique DERAT
nO 32/5005.22 (1990)
[13J Le Balleur J .C. Calcul des ecoulements a forte interaction visqueuse au moyen de
methodes de couplage, AGARD  CP 291 (1980)
[14J Marchand E.O., Lewis C.H., Davis R.T. Secondorder boundarylayer effects on a slender blunt cone at hypersonic conditions, AIAA 6th Aerospace Sciences Meeting, AIAA
Paper nO 6854 (1968)
[15J Maslen S.H. Secondorder effects in laminar boundary layers, AIAA Journal Vol. 1,
nO 1, pp. 3340 (1963)
Chapter 9:
Real Gas Effects in Two and ThreeDimensional
Hypersonic, Laminar Boundary Layers
B. Aupoix, J. Cousteix
ONERA/CERT, Aerothermodynamics Department,
2 avenue E. Belin, B.P. 4025, 31055 Toulouse Cedex,
France
ABSTRACT
This paper is devoted to hypersonic laminar boundary layers as the maximum heat
loads during the reentry of a space shuttle occur in the laminar regime. The emphasis is placed upon the real gas effects which deeply affect the evolution of the
flow. As boundary layers are addressed, thermal nonequilibrium effects are neglected
and the attention is focused upon chemical nonequilibrium effects. A simplified
atom/molecule model is justified and more complex models are detailed.
The influences of the wall catalysis, of the scatter in chemical reaction rate data
and of the wall temperature are brought into evidence.
Self similar solutions, to determine the flow and the wall heat flux at two and
threedimensional stagnation points or along attachment lines are presented.
INTRODUCTION
The design of a space shuttle heavily relies upon CFD. One of the key problems is
the evaluation of the heat loads the vehicle will experience. An Euler plus boundary
layer approach is well suited to give such information for a reasonable computational
cost. This paper is devoted to the problem of real gas effects which occur in hypersonic flows as the hypervelocity flow upstream of the shock is transformed into
a hyperenthalpy flow in which dissociation and ionization phenomena occur. It is
restricted to laminar boundary layers as the maximum heat loads are encountered in
the laminar regime. The entropy layer swallowing phenomenon, which must be accounted for to correctly predict the heat loads, is addressed in a companion paper [14].
INTRODUCTION TO REAL GAS EFFECTS IN REENTRY FLOWS
The source of real gas effects
The gas stagnation enthalpy is conserved through a shock wave. For hypersonic flows,
a high velocity, low temperature flow upstream of the shock wave is transformed into a
~~!
hI
h
(2)
p=nM
hI =
h~ + loT Cp[dT
(3)
where C p [ is the specific heat of the species I at constant pressure per mass unit and
h~ the heat of formation of the species I which, by convention, is zero for molecular
oxygen and nitrogen.
For a gas mixture, assuming an ideal mixture, the mixture enthalpy is the sum of
the contribution of each species, i.e. as h is the enthalpy per mass unit
(4)
The specific heat of the mixture should read
(5)
I.e. from (4)
(6)
where the first term of the RHS is the frozen specific heat
CPf = LCICp [
(7)
i.e. the specific heat of a mixture of constant composition while the second term
usually does not make sense since the gas composition does not depend only upon the
gas temperature and pressure, except for gases at chemical equilibrium.
All energy levels of particles are quantified. It is therefore possible to describe how
energy is stored through an energy partition function which includes the contribution
of all energetic modes. Thermodynamic functions such as the internal energy, the
enthalpy or the entropy can then be related to this partition function. As the energy
levels can be deduced from spectroscopic measurements, they are known with great
accuracy.
R = 8.314 Jmol 1 K 1
PI = PI MI T
(8)
= pT
(9)
If Mo is the molar mass of air upstream of the shock i.e. before any chemical reactions,
the state equation can be expressed with the help of the compressibility factor Z as
P = ZpRoT
Z= Mo
M
(10)
N 2 +0 2 ;=:NO+NO
(13)
introduced in early models (see e.g. [36]) is now usually discarded as it can be obtained
as a combination of dissociation and shuffle reactions. This gives a system of seventeen
chemical reactions.
o +ot
N 2+N+
o +NO+
N 2+O+
N+NO+
02+ NO+
N+NO+
o +NO+
>.
.,
O 2 +0+
>.
N +N;
NO+O+
0 +N;
NO+N+
NO+Ot
0 +N;
O2 +N+
.,
>.
.,>.
.,
;=:
>.
.,>.
.,>.
.,
(14)
N+N
;=:
;=:
;=:
0+0
N+O
(15)
(16)
;=:
;=:
A+ B
+
AB
(17)
where WAB/MAB represents the variation per time unit of nAB due to chemical reactions. Since, from the atom conservation nA + nAB = constant and nB + nAB =
constant, we also have:
Arrhenius [2] proposed to express the temperature dependence of the reaction rate
constant by his celebrated law
(18)
Chapter 9:
Real Gas Effects in Two and ThreeDimensional
Hypersonic, Laminar Boundary Layers
B. Aupoix, J. Cousteix
ONERA/CERT, Aerothermodynamics Department,
2 avenue E. Belin, B.P. 4025, 31055 Toulouse Cedex,
France
ABSTRACT
This paper is devoted to hypersonic laminar boundary layers as the maximum heat
loads during the reentry of a space shuttle occur in the laminar regime. The emphasis is placed upon the real gas effects which deeply affect the evolution of the
flow. As boundary layers are addressed, thermal nonequilibrium effects are neglected
and the attention is focused upon chemical nonequilibrium effects. A simplified
atom/molecule model is justified and more complex models are detailed.
The influences of the wall catalysis, of the scatter in chemical reaction rate data
and of the wall temperature are brought into evidence.
Self similar solutions, to determine the flow and the wall heat flux at two and
threedimensional stagnation points or along attachment lines are presented.
INTRODUCTION
The design of a space shuttle heavily relies upon CFD. One of the key problems is
the evaluation of the heat loads the vehicle will experience. An Euler plus boundary
layer approach is well suited to give such information for a reasonable computational
cost. This paper is devoted to the problem of real gas effects which occur in hypersonic flows as the hypervelocity flow upstream of the shock is transformed into
a hyperenthalpy flow in which dissociation and ionization phenomena occur. It is
restricted to laminar boundary layers as the maximum heat loads are encountered in
the laminar regime. The entropy layer swallowing phenomenon, which must be accounted for to correctly predict the heat loads, is addressed in a companion paper [14].
INTRODUCTION TO REAL GAS EFFECTS IN REENTRY FLOWS
The source of real gas effects
The gas stagnation enthalpy is conserved through a shock wave. For hypersonic flows,
a high velocity, low temperature flow upstream of the shock wave is transformed into a
hI
h
(2)
p=nM
hI =
h~ + loT Gp]dT
(3)
where Gp] is the specific heat of the species I at constant pressure per mass unit and
h~ the heat of formation of the species I which, by convention, is zero for molecular
oxygen and nitrogen.
For a gas mixture, assuming an ideal mixture, the mixture enthalpy is the sum of
the contribution of each species, i.e. as h is the enthalpy per mass unit
(4)
The specific heat of the mixture should read
(5)
I.e. from (4)
"""
Gp = L..J GI
I
aaT
hI"",, aGI
+ L..J hI aT
I
"""
I
= L..J GIGp ]
aGI
+ """
L..J hI aT
(6)
where the first term of the RHS is the frozen specific heat
GpJ =
2: GIGp ]
(7)
i.e. the specific heat of a mixture of constant composition while the second term
usually does not make sense since the gas composition does not depend only upon the
gas temperature and pressure, except for gases at chemical equilibrium.
All energy levels of particles are quantified. It is therefore possible to describe how
energy is stored through an energy partition function which includes the contribution
of all energetic modes. Thermodynamic functions such as the internal energy, the
enthalpy or the entropy can then be related to this partition function. As the energy
levels can be deduced from spectroscopic measurements, they are known with great
accuracy.
Chapter 9:
Real Gas Effects in Two and ThreeDimensional
Hypersonic, Laminar Boundary Layers
B. Aupoix, J. Cousteix
ONERA/CERT, Aerothermodynamics Department,
2 avenue E. Belin, B.P. 4025, 31055 Toulouse Cedex,
France
ABSTRACT
This paper is devoted to hypersonic laminar boundary layers as the maximum heat
loads during the reentry of a space shuttle occur in the laminar regime. The emphasis is placed upon the real gas effects which deeply affect the evolution of the
flow. As boundary layers are addressed, thermal nonequilibrium effects are neglected
and the attention is focused upon chemical nonequilibrium effects. A simplified
atom/molecule model is justified and more complex models are detailed.
The influences of the wall catalysis, of the scatter in chemical reaction rate data
and of the wall temperature are brought into evidence.
Self similar solutions, to determine the flow and the wall heat flux at two and
threedimensional stagnation points or along attachment lines are presented.
INTRODUCTION
The design of a space shuttle heavily relies upon CFD. One of the key problems is
the evaluation of the heat loads the vehicle will experience. An Euler plus boundary
layer approach is well suited to give such information for a reasonable computational
cost. This paper is devoted to the problem of real gas effects which occur in hypersonic flows as the hypervelocity flow upstream of the shock is transformed into
a hyperenthalpy flow in which dissociation and ionization phenomena occur. It is
restricted to laminar boundary layers as the maximum heat loads are encountered in
the laminar regime. The entropy layer swallowing phenomenon, which must be accounted for to correctly predict the heat loads, is addressed in a companion paper [14].
INTRODUCTION TO REAL GAS EFFECTS IN REENTRY FLOWS
The source of real gas effects
The gas stagnation enthalpy is conserved through a shock wave. For hypersonic flows,
a high velocity, low temperature flow upstream of the shock wave is transformed into a
  
~

Thermal relaxation
Boundary layers will be addressed in the following. For the boundary layer concept
to be valid, an inviscid region must exist between the shock wave and the outer edge
of the boundary layer. For reentry flows, vibrational relaxation occurs downstream
of the shock wave and is achieved at the outer edge of the boundary layer. The flow
in the boundary layer can be assumed at thermal equilibrium.
It must however be pointed out that this thermal equilibrium assumption neglects
two physical phenomena:
When atoms recombine at the wall, the soformed molecules can be out of vibrational equilibrium. As this vibrational nonequilibrium quickly relaxes in the
wall region, this phenomenon can be neglected .
Free electrons due to ionization reactions hardly interact with atoms and molecules so that their translation energy is out of equilibrium with the atom and
molecule translation energy.
In the following, we shall restrict ourselves to flows at thermal equilibrium, which
are relevant to boundary layer flows.
Thermodynamics of a gas mixture
Characteristics of a gas mixture Before considering the thermodynamic properties
of a gas mixture of species (noted by capital indices I), let us define some quantities
we will need later:
V:
mI:
m
PI:
P
CI
nI:
n :
MI :
M
PI
P
hI
h
LPI
(1)
(2)
p=nM
hI =
h~ + loT Cp[dT
(3)
where Cp [ is the specific heat of the species I at constant pressure per mass unit and
h~ the heat of formation of the species I which, by convention, is zero for molecular
oxygen and nitrogen.
For a gas mixture, assuming an ideal mixture, the mixture enthalpy is the sum of
the contribution of each species, i.e. as h is the enthalpy per mass unit
(4)
The specific heat of the mixture should read
(5)
I.e. from (4)
Cp
" hI aC
" hI aC
= '"
L.J CI ah
aT + 'L.J
aT = '"
L.J C[C p[ + 'L.J
aT
[
(6)
where the first term of the RHS is the frozen specific heat
CPJ = LC[Cp[
(7)
i.e. the specific heat of a mixture of constant composition while the second term
usually does not make sense since the gas composition does not depend only upon the
gas temperature and pressure, except for gases at chemical equilibrium.
All energy levels of particles are quantified. It is therefore possible to describe how
energy is stored through an energy partition function which includes the contribution
of all energetic modes. Thermodynamic functions such as the internal energy, the
enthalpy or the entropy can then be related to this partition function. As the energy
levels can be deduced from spectroscopic measurements, they are known with great
accuracy.
= PI MIT
(8)
LPI
(9)
pT
If Mo is the molar mass of air upstream of the shock i.e. before any chemical reactions,
the state equation can be expressed with the help of the compressibility factor Z as
Z= Mo
M
P = ZpRoT
(10)
N z +M
Oz +M
NO+M
where M stands for any of the species .
Shuffle reactions
NO+O
+Nz
~
~
~
~
N+N+M
O+O+M
N+O+M
(11 )
N+O z
N+NO
(12)
It must be noticed that the direct formation reaction for the nitrogen monoxide
(13)
introduced in early models (see e.g. [36]) is now usually discarded as it can be obtained
as a combination of dissociation and shuffle reactions. This gives a system of seventeen
chemical reactions.
o +ot
N 2+N+
o +NO+
N 2+O+
N+NO+
02+NO+
N+NO+
o +NO+
"
"<""""""
"
"<""""""
"
"<""""""
"
"<""""""
"
"<""""""
"
"<""""""
"
"<""""""
"
"<""""""
O2 +0+
N +Ni
NO+O+
0 +Ni
NO+N+
NO+Ot
0 +Ni
O2 +N+
(14)
(15)
~
~
(16)
A+B
t
AB
(17)
where wABI MAB represents the variation per time unit of nAB due to chemical reactions. Since, from the atom conservation nA + nAB = constant and nB + nAB =
constant, we also have:
=
Arrhenius [2] proposed to express the temperature dependence of the reaction rate
constant by his celebrated law
(18)
where EA is the activation energy. This form has later been extended, to cover a
wider range of reactions, as
= ATE exp (  ~; )
(19)
where f and b indicate the forward and backward reactions, the A are the chemical
species and the v the stoechiometric coefficients. The difference VI = v'f  VI must
be introduced as it indicates whether the species AI is destroyed or produced by the
forward reaction. The evolution of the reactants is given by the following expressions
which are generalisations of equation (17)
Forward reaction
(20)
Backward reaction
,,"
nI=kbvI II (nJ)J
(21)
where k j and kb are respectively the forward and backward reaction constants. The
chemical production rate of species I is the balance between the two terms
(22)
NonArrhenius expressions for the reaction rate constants can be found in various
publications, either from reviews of data for combustion [7, 24, 51] or from papers
dealing with reentry problems [12, 16, 28, 29, 33, 36, 37, 38, 50]
The reaction rate constants are mainly determined by experimental techniques.
It must be pointed out that such experiments are very difficult to design and analyze.
Reaction rate data are thus known with little accuracy, especially at high temperatures. Various sets of reaction .rate constants, namely one used by Straub [47] and
due to Oertel [36], one due to Park [40, 37], one used by Shinn et al. and due to
Bortner [12], the one recommended by the National Bureau of Standard [51] i.e. the
one due to Baulch [7] and, at last, the review due to R. H. Hanson and S. Salimian
in Gardiner's textbook [24] are compared on figures 1 and 2. Figure 1 shows a very
large scatter for the oxygen dissociation reaction which is one of the most important
reactions. Even if Straub data are discarded, there is one order of magnitude discrepancy between Baulch and Park models. The ratio of reaction rate constants to the
value recommended by the National Bureau of Standard is plotted on figure 2 for a
,..Straub
<1'
'
_ Park
_Sninn
Baulch
("0
~105
10;~LO~0~2~0~0~O~3~00~O~~LO~0~O~5~0~O~0~~~8~O~OO~~
TI K I
+ N2
>
+ 0 + N2
J .0
,,
,,
'"
CD
'"o
"'2.5
..................
Straub
2.0
_ 
 Park
0.5
O.
1000
2000
3000
LOOO
5000
5000
1000
8000
TI K)
+0
>
+ O2
shuffle
shuffle reaction. The dotted lines indicate the evaluated uncertainty range [51]. The
scatter is larger than estimated, a ratio about three between extreme values can be
observed.
Further information about the forward and backward reaction rate constants can
be obtained by considering chemical equilibrium. Chemical equilibrium is obtained
for a reaction when the forward and backward reactions exactly balance. As the
production is zero, the reaction rate constants can be linked to the number of moles
of species per volume unit
"I
kf
(23)
nI eq
= k
II(
I
or, using the state equation PI = nIkT where k is the Boltzmann constant,
(24)
which is known as the mass action law. This law can also be deduced from thermodynamical arguments, as the equilibrium state corresponds to the minimum of the
Gibbs free energy of the system formed by the species involved in the chemical reaction. The equilibrium constant J<p and the mixture composition can be determined
from thermodynamical data. The ratio of the forward and backward reaction rate
constants can be obtained so; therefore it is accurately known from thermodynamical
data.
Chemical time scale According to the above formulae, the chemical time scale for
e.g. the forward reaction in equation (20) reads:
Tchem
rv
nI (kf
PI (kf
rv
(25)
1
PI = nIMI :
gn~~)
gP~~)
(26)
1
We define the reaction order as the sum of the stoechiometric coefficients of the
reactants, i.e. for the forward reaction
I
= ~vJ
(27)
rv
C(k P'"
I
1
gC;~)
1
(28)
i.e. the chemical time scale decreases when the chemical reaction rate or the density
increases.
Let us consider a reentry trajectory such as the STS2 flight reentry trajectory
the laminar regime part of which is given in table 1.
t
(8)
200
250
330
460
480
540
650
770
830
1000
1120
1215
Altitude
(km)
92.35
85.74
77.91
74.98
74.62
73.33
71.29
68.67
66.81
60.56
52.97
47.67
u
(km 8 1)
7.50
7.53
7.42
7.20
7.16
7.03
6.73
6.31
6.05
4.99
3.87
2.96
p
(kg m 3 )
2.18410 6
6.36510 6
2.33510 5
3.18510 5
4.05510 5
4.79410 5
6.82410 5
9.66910 5
1.21610 4
2.62110 4
6.76210 4
1.34410 3
p
(atm)
1.128 10 6
3.58710 6
1.316 10 5
2.14210 5
2.28010 5
2.83110 5
3.96510 5
5.99210 5
7.92510 5
1.87710 4
5.02510 4
9.90010 4
(I<)
324
199
199
198
198
200
205
219
230
253
262
260
Mach
number
27.90
26.60
26.30
25.50
25.40
24.80
23.40
21.30
19.90
15.70
11.90
9.15
Angle of
attack
40.4
41.0
40.2
40.0
40.3
40.4
39.4
38.5
41.4
42.0
38.3
34.8
(from [46])
As the shuttle goes deeper into the atmosphere, the density increases by roughly
three orders of magnitude. As the order of the reactions is usually two or three, this
implies a variation of three or six orders of magnitude of the pv' 1 term. At the
same time, the shuttle slows down so that the flow enthalpy and the temperature
level decrease. In the outer region of the boundary layer, near the stagnation point,
the temperature decreases from about 6000 K to about 3000 K. The variations of
reaction rate coefficients with temperature can be very steep. For example, for the
atomic oxygen dissociation reaction, the forward reaction rate coefficient decreases by
four order of magnitude between 6000 K and 3000 K while the backward reaction rate
coefficient hardly changes. So, the chemical time scale strongly depends upon both
the considered chemical reaction via the temperature dependence of the reaction rate
coefficient and the altitude via the density.
For boundary layer flows, the flow is highly dissociated in the stagnation point
region and recombines downstream as the flow acceleration leads to a decrease in the
temperature level. As the recombination reactions are third order reactions, the variation of the density is the leading term and the chemical time scale decreases with
the altitude. On the other hand, the mean motion time scale, i.e. the length of the
shuttle divided by the upstream velocity, varies weakly. This results in an important
variation of the chemical relaxation/mean motion time scale ratio during the reentry.
The flow is frozen from a chemical point of view at the beginning of the reentry
(high altitudes, low densities) and near equilibrium at the end of the laminar regime
(lower altitudes, higher densities). We shall focus our attention here on chemical nonequilibrium flows, which occur over quite all the laminar part of the shuttle reentry.
Gas/surface interface chemistry The surface material may act as a catalyst. The
heterogeneous catalysis phenomenon occurs in three basic steps.
where the subscript w indicates wall values and the reaction order m should be between
one and two. Scott [44J assumes the reaction order to be equal to one, the reaction
rate constant klw has thus the dimension of a velocity.
As pointed out previously, the heterogeneous catalysis first depends upon the
ability of the surface to adsorb gas species. These various behaviours are reflected by
the values of the reaction rates coefficients k1w . When the coefficients tend towards
zero, the species are not modified at the wall. The wall is thus said to be noncatalytic.
On the contrary, when the coefficients tend towards infinity, the atoms impinging the
wall recombine very quickly. The wall is said to be catalytic. One may also assume
that the wall has a catalytic effect on all possible chemical reactions so that chemical
equilibrium is achieved at the wall. As the wall temperature is "low" (i.e. up to 1500 K)
chemical equilibrium means almost complete recombination of atoms at the wall as
air hardly dissociates even at low pressure for these temperatures. Therefore, the
atom dissociation energy is transferred at the wall; wall catalytic efficiency increases
the wall heat flux. The noncatalytic and catalytic wall cases are extreme cases which
are not encountered practically. Real materials have a finite catalytic efficiency. Here
again, experimental determination of the wall catalytic efficient is very difficult [13,
22, 43, 44J and moreover the wall catalytic efficiency evolves with the aging of the
surface material.
Concerning ionized species, it is taken for granted that they are neutralized at the
wall, whatever the wall catalytic efficiency.
Transport properties of gas mixtures
Collisions are also responsible for momentum exchange between particles which leads
to transport of momentum or heat by the gas mixture. They also tend to homogeneize the mixture composition via a diffusion process. Gas kinetic theory yields the
~=
~diV11.~) + 1]div11.~
(30)
"
~P~
MIMJ D d
DTgradT
IJJ I T
J#
(31 )
QI
nI+ (nI
PI) gradp
gra d n
n
P
p
(32)
"
"" nJ
Dr (qI=  =qJ)
1
q = AgradT+ ~hN +RT ~~I
t
I
n MI DIJ
PI
(33)
pJ
where fi is the viscosity, 1] the bulk viscosity, DIJ the binary diffusion coefficient of
species I in species J, D IJ the polynary diffusion coefficient of species I in species J in
the thermal diffusion coefficient and the thermal
presence of the other species,
conductivity.
The thermal diffusion term in the mass diffusion flux (31) corresponds to the Soret
effect i.e. the fact that light species diffuse towards hot regions, heavy species towards
cold regions. The thermal diffusion term in the heat flux (33) corresponds to the
Dufour effect. The thermal diffusion term is a secondorder term in gas kinetic theory
so that these effects are small.
The various transport coefficients of the mixture can be expressed in terms of
collision integrals which are known with little accuracy at high temperatures.
Fick's law provides an approximate expression for the mass diffusion flux as
Dr
9.1 =  pDgradCI
(34)
Prandtl number
Lewis number
Schmidt number
conduction
mass diffusion
X conductio.D
S  P  L_momentum dlttusion
 C  pD mass diffusion
.c 
pD PI _
(35)
Гораздо больше, чем просто документы.
Откройте для себя все, что может предложить Scribd, включая книги и аудиокниги от крупных издательств.
Отменить можно в любой момент.