Sie sind auf Seite 1von 20

BOUNDARY CONDITION CALCULATION PROCEDURES

FOR INVISCID SUPERSONIC FLOW FIELDS*

Michael J. Abbett
Mgr. Fluid and Aerodynamics Department
Acurex Corporation/Aerotherm Division
Mountain View, California

Abstract considerations of calculation time com-


parisons are discussed, but detailed
Results are given of a comparative timing studies have not been made.
study of numerical procedures for comput-
ing solid wall and sharp shock boundary The four test problems used in the
points in supersonic inviscid flow calcula- evaluation are described in Section 2,
tions. Thirty different calculation pro- followed by a summary of the basic equations
cedures were tested on four sample problems and overall calculation procedure (Section
in two dimensional steady flow. A new 3 ) . The boundary condition calculation
simple but accurate calculation procedure procedures of interest are reviewed and
was developed for solid wall boundary categorized in Section 4, with comparative
point calculations. The merits and short- results and discussion thereof immediately
comings of the various procedures are following (Section 5 ) . The calculation
thoroughly discussed. of three dimensional steady flow fields is
discussed in Section 6. We conclude with
1. Introduction a summary of the most important findings
and recommendations. A list of symbols is
When inviscid flow field solutions are given at the end of the paper
obtained by finite difference techniques,
the finite difference mesh can be divided 2. Test Problems
into two regions, i) the interior points
and ii) boundary points. Boundaries in- The four test problems, illustrated in
clude solid and porous walls, shock waves, Figure 1, are:
contact discontinuities, etc. Though it
is desireable to employ a numerical tech- • I Simple isentropic compression
nique which utilizes no special computa-
tional procedures at boundary points, to • II Simple isentropic expansion
date that goal has been realized only in
very special situations. Accurate solu- • III 20° wedge/circular arc/30° wedge
tions of complex problems invariably re-
quire special treatment of boundary points. • IV 20° wedge/circular arc/horizonal
Though a wide variety of computational pro- plate
cedures have been used at boundary points,
often accompanied by strong assertions as The objective of the test problems is not
to their accuracy, little information has to be exhaustive, nor is it to test the
been available to aid the user in compar- various methods in situations favorable to
ing and evaluating the various procedures. certain ones. Rather, it is to test each
This paper presents the major results and case in situations which are not neces-
conclusions to date of a continuing study sarily favorable but which are representa-
to i) review and evaluate numerical tech- tive of what is required of a technique in
niques for handling solid wall and sharp every day working codes. Expansions and
shock boundary points in the numerical compressions are, of course, the heart of
solution of supersonic steady flow fields supersonic flow fields, and these sample
with finite difference procedures, and ii) problems serve as valid tests of the re-
to improve existing and develop new tech- lative merits of the various procedures.
niques which are simple to implement and Furthermore, although these flows are com-
accurate. Two and three-dimensional invis- paratively simple, computationally they
cid flows are considered. are quite challenging, as we shall soon
see. Finally, they enable us to make the
The studies are directed toward determin- comparison against a valid standard since
ing the characteristics of calculation we can obtain the exact solution along the
schemes generally useful to engineers and wall boundary for the first two problems
researchers. The approach is empirical and calculations with the full method of
with minimal reliance on formal accuracy characteristics serve as the standard for
analysis. Rather, the relative merits of the latter two test problems.
the various schemes are determined by quali-
tative comparisons of their distinguishing It has been noted* that these problems
characteristics and by comparing them one lack a certain cleanliness in that the
with another in test calculations of simple wall curvature is discontinuous. Moretti
but realistic flow fields. Some general has pointed out that it is not formally

This research was supported by NASA under Private communications from G. Moretti and
Contract NAS2-6341. C.P. Kentzer.

153
correct to differentiate across the result- equations are used at interior points.
ing interior region gradient discontinuity
any more than across a shock wave or Non Conservative Form
contact discontinuity.^ Unfortunately, the
user of computer.codes for calculating flow For steady isentropic flow,
fields of interest usually does not know
the data (surface geometry) with enough up^ + vpy = ^ (up^ + vp^) (6)
precision to be able to identify such
curvature discontinuities. Furthermore, with a^ = Yp/p. Combining (6) and (1)
if he did, they are often so numerous that gives the continuity equation in terms of
to treat them rigorously would result in derivatives of pressure and velocity com-
an unacceptably complicated and expensive ponents.
book keeping task as far as computer code
development and application are concerned. pu. + pv + ^ (up + vp ) = 0 (7)
Such a user is justified in requesting
calculations which ignore such effects, Transforming to (^^n) space, the equations
but he needs a data base on which to of motion become
evaluate the results he will obtain. The
results and discussions reported herein are
intended to serve as a small but useful P^<+ cP + Y(uv^ - vu^)
part of such a data base.

3. Basic Equations and + uvP /6 = 0 (8)


Overall Calculation Procedure n
The basic equations for the steady state 1
two-dimensional flow of an ideal gas u^ + /cu + uvu
ri 2 2
(isentropic flow, constant isentropic ex- ' u^ - a^
ponent) are
-a' fv + ^ P /6 = 0 (9)
continuity: Y n

(pu). (pv)^ = (1)


v^ ^ c + ^ W + ^ P /6 = 0 (10)
1st momentum: u / n yu n

puu^ + pvu^ + p^ = 0 (2) ^c M^ ^ I) V = ' (11)

2nd momentum: with c = n(B^ - T^) - B^ (12)


puv^ + pvv^ + p^ = 0 (3)
For later use, we note that
entropy:

uS + vS =0 (4)
X y = tane
state:
q^ = u^ + v^
S = Inp/p^ - ylnp/p^ = P - yR (5)
(13)
The actual calculations are performed in a
h = ^ , P
computational space, illustrated in Figure y - 1 p
2 for problem IV. For problems I and II
the upper boundary in the computational
space corresponds to a line y = constant in H = h. V
physical space, the constant chosen large
enough so that the upper boundary does not Conservative Form
influence the lower surface calculations
Equations 1) - 4) are rewritten in the
in the region of interest. The transfor-
mation from physical to computational space form
is given by
(pu)^ + (pv)^ = 0

(p + p u ^ ) ^ + (puv)^ = 0
_ y y - B (14)
T - B 6
(puv)^ + (p + p v ^ ) ^ = 0
where the lower and upper boundaries are
described by y = B(x) and y = T(x) respec-
tively and where 6(x) = T(x) - B(x). (Eu + p u ) ^ + (Ev + pv) = 0

Both non-conservative and conservative


forms of the governing partial differential

154
where index j=J, the problem is to obtain the
solution for all j , 2 _< j _< j , at station
E = + P u'' + v^ is the energy per i+1 assuming that the solution is known
unit volume. Transforming to the computa- for all j , 2 < j £ J, at station i. To
tional space while retaining the conserva- accomplish this, the mesh is divided into
tive form of the equations gives interior points (3 < j £ J-1), lower
boundary points (j='^) , and upper boundary
points (j=J). In test problems III and IV
(pu6) ^ + (puc + pv) = 0
j=J corresponds to the shock wave (leading
shock in case III).
[(p + p u M 6 ] ^ + [(p + pu2)c
At interior points, the equations are
integrated following the scheme of
+ puv] (15) MacCormack, a predictor corrector scheme
which is easily described by looking at
( p u v 6 ) ^ + (puvc + p + pv^) = 0 the equation

f^ + f (19)
[ (Eu + p u ) 6 ] ^ + [(Eu + p u ) c
Letting i be the E, index and j the n index,
to obtain the solution at a point ( i + 1 , j)
+ Ev + pv] 0
using data known at a preceding ^ station
(i, all j) we proceed:
These equations are written vector form as
predictor:
(F6), + G^ = 0 (16)
f(i + l,j) = f (i, j)
where

F = F^ = pu f(i, j + 1) ~ f(i.j)
(20a)
An
F9 = p + pu^
(17) corrector:
F^ = puv
1
f(i + l,j) 7 f(i, j) + f(i + l,j)
F^ = u(E + p)

and they are solved using MacCormack's f (i -f 1, j) - f(i + 1, j - 1)


second order accurate predictor/corrector A^
scheme (see below). To obtain p, u, v, p An
from the computed elements of the F array, (20b)
we have For further details about the method, see
the two papers by MacCormack.^^3
^3/^1
Of course, our interest is focused on
the boundary points, both lower wall and
F2 + ViF | . l - (P^ 2F^F4) shock wave. Thus, we evaluate the re-
(18) lative merits for boundary point calcula-
Y + 1 tion procedures when the interior points
are integrated with MacCormack's scheme.

Along the surface streamline there is a


P = F^/u certain degree of flexibility since the
entropy and total enthalpy are known. In
almost all the calculations we use that
p = F<^ - pu information, as well as the known body
slope, in order to solve as few partial
Outline of the Calculation Scheme differential equations as possible. Thus,
usually we solve only one differential
The computation space is covered by a
equation along the wall, and that one will
finite difference grid of spacing A^, Anf
be an equation for pressure. In other
with C = C + iA5 and ^1 = ^ + j An, where
instances (e.g., Thomas' scheme) the basic
(i,j) are the indices associated with the
scheme requires that more than one dif-
abscissa and ordinate respectively. The
ferential equation be integrated along the
general flow direction, and the direction
wall. In such instances we can expect
in which the equations are integrated,
errors in surface entropy and/or total
corresponds to increasing ^ (increasing i ) .
enthalpy to develop. We proceed in this
Thus, denoting the lower boundary by the
manner because of our two-fold purpose:
index j = 2* and the upper boundary by the
i) to compare the relative merits of
various computation procedures currently
in use, and ii) to evaluate the best re-
The line j=l is reserved for implementa- sults that can be achieved within a par-
tion of reflection type calculation pro- ticular class of schemes. For instance,
cedures along the lower wall. take simple explicit integration procedures.

155
We could integrate all four differential Solid Wall Calculation Procedures
equations: continuity, x and y momentiim,
and entropy. Or we could integrate one The schemes tested are categorized into
equation, say for pressure, and use the the following groups
three relations for entropy, total enthalpy,
and wall slope to calculate the density • Reflection
and two velocity components. If the dif-
ferential equation to be integrated is • Explicit - One sided derivatives
chosen optimally, we expect the second
approach to yield better results since we • Standard Predictor/Corrector - one
are solving three of the four equations sided derivatives
exactly.
• Implicit - One sided derivatives
Full Method of Characteristic Calculations
• Extrapolation
For problems III and IV, Figure 1, and
occassionally for problem II, solutions • Method of characteristics (at bound-
were generated using the full method of ary point only)
characteristics. Those solutions serve as
a standard against which the calculations • Kentzer's scheme
using finite differences at interior points
and various boundary point calculation • Combined Equations
procedures at upper and lower boundary
points are evaluated. The computer code • Euler predictor/Simple Wave Corrector
used to generate the full method of charac-
teristics solutions is the BOWSHK version In this Section we outline the basic
of Aerotherm's two-dimensional/axisym- approach used in each category, and we
metric Method of Characteristics (MOC) flow focus attention on how schemes in each
field code. That code is a derivative of category account for the basic gas dynamic
the code described in Reference 4 and interactions which occur at solid wall
commonly referred to as Frankenstein. It boundaries.
has the following characteristics and
capabilities. In many of the schemes, we simply in-
tegrate Equation 8 to obtain the change
• Calculation of complete two-dimen- in surface pressure between stations i and
sional or axisymmetric supersonic i + 1. In such cases, we have
internal or external flow fields by
the method of characteristics Pi+1,2 = Pi,2 + (^S) (21)
• Ideal gas
where RHS is some representation of the
• Uniform or nonuniform initial condi- term
tions

• Intersection and reflection of waves RHS = - /CP y(uv^ - vu^)


of same and opposite families - no
limit to number
+ uvP /6 (22)
• Formation of shocks at compression
corners It is assumed that the reader is familiar
enough with standard notation to know that,
• Formation of shocks by coalesence of for instance, in a simple explicit scheme
waves from smooth walls with two-point one-sided n derivatives, we
would have at point (i, 2)
• Formation of contact discontinuities
P.1,2^
• Reflection of waves from walls i,3
cP = (Vi,2 (23)
An
4. Summary of Boundary Condition
Calculation Procedures Tested V. ^
1/3 ^i,2
1,2 An
More than 25 procedures for calculating n
solid wall points and 15 procedures for
calculating sharp shock points were tested. etc. In order to calculate the surface
Many are slight variations of others, and boundary point at (i + 1,2) from known
the number of distinctly different data at (i,j), j >^ 2, we assume the follow-
approaches is considerably smaller. In ing to be known:
this paper we concentrate on the most im-
portant results. Additional results and i) The complete solution at every
details of most of the calculation pro- mesh point at station i, [i.e., for
cedures are given in References 5 and 6. all (i,j) , j > 2]

and, if necessary.

156
ii) The complete solution of the pre- flow field about the solid wall. All
dictor and corrector steps at all quantities are reflected evenly, except
interior points at station i + 1 the normal velocity component, which suf-
[i.e., for all (i + 1,j) , j >_ 3] fers odd reflection in order to impose the
requirement that it vanish at the solid
Reflection - The use of this procedure is wall.
basically physically motivated. The wall
boundary of an inviscid flow is essentially To carry this idea further, to flows
a slip stream, and in this sense there is with curved walls or to three-dimensional
no basic difference between a streamline flows, leads to a real quagmire. Though
along a solid wall and the same streamline there can be no argument in any particular
which is located within a fluid on one case about the concept of an extended flow
side of a slip stream. On this basis, one field, except in special cases, the
reasons that, for computational convenience, difficulty of constructing one which will
he can replace his original problem with yield an accurate solution to the posed
its solid boundary with another problem problem is at least as difficult as it is
having an extended flow field which to obtain more directly a solution of com-
(hopefully) will include a slip stream parable accuracy.
where there was originally a solid wall
(sketch). It is evident from this discussion that
the biggest shortcoming of reflection pro-
cedures is that their accuracy depends so
strongly on the local characteristics of
the flow and on the coordinate system used
in the computation, and no one has utilized
procedures reflecting this fact with enough
slip stream precision. However, it should be recognized
wall that the use of body oriented coordinate
systems is an attempt to accomplish this.

Explicit-One-Sided Derivatives - It has


Original Flow Field been implied above that it would be very
convenient to solve boundary points ex-
Fictitiously Extended actly as interior points, and this is one
Flow Field of the driving forces behind the common
utilization of reflection procedures. The
problem is that virtually all satisfactory
difference schemes for interior points re-
quire some form of centered differencing.*
However, there is no reason to require
SKETCH. THE CONCEPT OF REFLECTION that the idea of central differences be
PROCEDURES carried over to boundary points. In fact
it violates what we know to be the physical/
Knowing the solution, at a given initial mathematical structure of the problem. At
line, somewhat into the extended flow*, a solid wall, the solution along the wall
it is possible to continue the solution is determined solely by the interaction of
downstream accurately by treating the the wall geometry and the interior flow.
slip stream as a regular part of the in- Therefore, though it is not quite as con-
terior flow. So the problem of computing venient, it is much more natural to think
the solution at the solid boundary has in terms of one-sided derivatives at the
disappeared. Unfortunately, it has been wall if one is embarked on a course of
replaced by the problem of determining the solving the partial differential equations
fictitiously extended flow field below the along the wall.
solid wall.
With this in mind, one can easily con-
In looking at two-dimensional problems struct procedures for integrating one or
with straight walls, it is not difficult more of the differential equations along
to construct an extended flow field which the wall with derivatives normal to the
will effectively simulate the solid bound- wall in the computational space being com-
ary, and there the idea of reflection en- puted by one-sided differences. The
ters naturally. In this case, at the wall actual procedures employed can vary some-
the normal derivative of pressure vanishes what, particularly with respect to how
along with the normal component of velocity. many and which equations are to be
Therefore, the extended flow field can be integrated in this manner.
constructed by simply reflecting the actual
Standard Predictor-Corrector-One Sided
Derivatives - A logical extension to the
simplest explicit scheme with one-sided
Note that this extended solution is not
uniquely determined until one requires that
the solution be continuous across the slip-
stream. That assumption is usually It need not be symmetric, and multi-step
implicit. procedures are included in this rather
broad interpretation of centered differ-
encing.
157
derivatives is to consider a predictor- characteristics has been the m o s t accurate
corrector scheme w i t h derivatives in both procedure for numerically integrating the
the predictor and corrector steps computed governing differential e q u a t i o n s . It is
as one-sided differences on the same side. especially convenient at boundaries where
Such an approach is a logical extension other methods require partial derivatives
of a predictor-corrector scheme at in- normal to the boundary. It's biggest
terior points to boundary p o i n t s . Though shortcoming is its complexity, w h i c h r e -
w e can expect to have potentially more sults from two sources. F i r s t , to employ
accurate calculations than with a simple the method only at boundaries while u t i l i -
explicit calculation, accuracy w i l l be zing other differencing procedures at in-
limited by the fact that no direct coup- terior points leads to a number of i n -
ling is m a d e between the partial differ- terpolations that m u s t be m a d e in order to
ential equations and the w a l l geometry. integrate the equations along character-
istics. Second, the equations m u s t be
Implicit-One-Sided Derivatives - Implicit combined in the form of characteristic
schemes, and mixed implicit/explicit compatibility e q u a t i o n s , and this is i n -
schemes, are the same as corresponding convenient at best and time consuming* at
completely explicit schemes except that the worst. The result is that boundary points
derivatives "normal" to the wall are com- may require a disproportionate amount of
puted with the values at the next x station analytical/coding effort from the a n a l y s t /
(i.e., X = Xo + A x ) . T h u s , the solution programmer.
depends on the "normal" derivatives w h i c h ,
in turn, depend on the solution, and the These disadvantages are offset by the
solution m u s t be determined iteratively. fact that the use of characteristic
equations at the wall undoubtedly r e s u l t s ,
These are two obvious objections to in g e n e r a l , in as accurate solutions as
implicit schemes for our p u r p o s e s : i) since it is possible to achieve. That they are
the interior points are computed completely not more commonly used reflects the fact
explicitly, it is desirable to compute that the inconvenience to the analyst
the boundary points explicitly a l s o , and and computer coder is substantial and r e -
ii) computing normal derivatives impli- sults in considerable increase in code
citely "violates" the laws of the t r a n s - development cost. H o w e v e r , if large
mission of signals which are known for the numbers of production runs are to be m a d e ,
hyperbolic equations under consideration. these additional costs can be m o r e than
The latter is an objection concerned w i t h offset by the decreased cost per run for
accuracy, not stability, since purely im- given accuracy that can be anticipated
plicit schemes can be unconditionally when characteristic procedures are used at
stable. Keeping these objections in m i n d , boundaries.
for completeness w e include implicit
schemes in the comparison. The approach in its simplest form is
best illustrated by looking at our basic
Extrapolation - Probably the m o s t simple example, supersonic, inviscid, two-dimen-
procedure is simple extrapolation w h e r e sional flow. The governing differential
the solution at the boundary at x = X Q + Ax equations for the momentum equations in
is obtained by extrapolating the solution characteristic form are
at interior p o i n t s . ^Vhile it is possible
to construct problems in which this p r o - dr
cedure is adequate, in general it w i l l pu^ = 0 (24)
+ dx
yield completely unsatisfactory r e s u l t s , /M2 - 1 dx
especially when the solution changes along the characteristic directions
rapidly near the w a l l .

Method of Characteristics uv ± a/u^ + v^ - a^


(25)
dx ]2 - ^2
Gas dynamicists and numerical analysts
are familiar w i t h the method of charac-
teristics for^numerically solving the with T = v/u being the local streamline
quasi-linear hyperbolic equations for slope. Here
supersonic, inviscid flow. This method
is based on special properties of the d__ 9 ^ dy 9
governing partial differential equations (26)
dx ^ 3x "^y 93F ^ ^ ^
which permit them to be reduced to ordin-
ary differential equations (two indepen- is a total derivative with respect to the
dent variables) w i t h variable coefficients. X along the characteristic. The standard
These ordinary differential equations procedure for implementing this charac-
hold only in special d i r e c t i o n s , called teristic compatibility equation at a
characteristic d i r e c t i o n s , and they r e - solid boundary is indicated in the sketch.
flect the facts that: (i) there are p r e - The equation is written along the charac-
ferred directions for the propagation of teristic between points D and C , where D
signals in supersonic inviscid flow and
that (ii) these signals suffer no disper-
sion or dissipation (at least theoreti- ,
cally). In m o s t instances, the method of We are thinking, in particular, of three
or more independent v a r i a b l e s .

158
is somewhere between the mesh points A and The hope is that by combining the equations
B. Since the bounding geometry is speci- in this special way but still retaining their
fied, TQ is known. With T Q and p^ partial differential character, the solu-
obtained by interpolating between points tion will be obtained with accuracies
A and B, we have typical of method of characteristics cal-
culations while retaining the simplicity
Pc = PD + ^("^c - ^D> of standard explicit partial derivative
calculations.

Combined Equations* - Another approach,


which has been used by Thomas•'-^ and
. dy/dx = X Thomasll, et al., is to combine one or more
of Euler's equations with the surface
boundary condition in differential form.
The object is to obtain a direct coupling
of the surface boundary condition with the
partial differential equation(s) to be
integrated.
SKETCH - METHOD OF CHARACTERISTICS AT
BOUNDARY POINT Thomas replaced one momentum equation
with the boundary condition of zero velo-
with b an average value of pu^/ZM^ - 1 city component normal to the lower wall.
between D and C. Moretti has pioneered This is done by replacing the (x,y) velo-
the use of this approach, demonstrating city components, (u,v), with components
its accuracy on a wide variety of problems.; (U,V) parallel to and normal to the wall.
Since V is everywhere normal to the wall,
Kentzer's Scheme - A rather clever V E 0. Also
approach, proposed by Kentzer^, is based
on the method of characteristics. The U = u cos + V sm (30)
basic motivation seems to be to get most
of the advantages of utilizing character-
istics compatibility equations without U^ = U - S- U (31)
some of the disadvantages. We have seen K X 6 n
that there are basically two disadvantages
of characteristic methods: (i) the added u cos 0 + V sin 0
analytical work required to put the par-
tial differential equations into charac-
teristic form, and (ii) the additional - (u sin 6 - V cos 6) (
special coding required, including a number
of interpolations to get data at specific where
points on a characteristic. Kentzer's
approach eliminates the second disadvantage u^ = u 5 on
while retaining the first and, hopefully, (32)
the accuracy of a complete characteristic
procedure. Referring to the sketch above, V = v^ - T- V
X ^ 6 n
Kentzer's idea is to combine the com-
patability equations with the surface Equation (31) for U is integrated along with
boundary condition in partial differential Equation (8) for pressure. 6^^ = d6/dx is
form along Xc. Thus, we have known from tjie specified velocity dis-
tribution. V E 0, and p = p(p,S) complete
the integration along the lower wall, since
dr
+ dx dx (H * * I?) U cos e - V sin
(33)
>=(M-'l7)=« (27)
U sin 6 + V cos

The analysis used by Thomas employs the


specified variation of 6(x) in computing

t-*' 3p
3y + b (ij*»w) (28) the velocity changes by integrating the
streamwise momentum equation. No use is
made of the known variation of 0(x) in in-
with tegrating the equation for pressure, nor
/^ is there any direct coupling between the
A = (29) continuity equation and one or both of the
a2
momentum equations. We know that the pri-
mary physical force determining surface
and the rest of the right-hand evaluated
pressure variation is momentum exchange
at A. The interesting point is that,
while Equation (28) holds generally, in the
method of characteristics the derivatives
in Equation (27) are treated as directional Actually, Kentzer's scheme is one of this
total derivatives in the directions having group, but we consider it somewhat separa-
tely because of its unique features.

159
away from the wall, and it is natural to the continuity equation are combined, and
consider combining the pressure equation the specified variation of 6(x) is used to
(Equation 8} with one or more of the simplify the resulting equation and to give
momentum equations and with the specified the direct coupling between changes in wall
variation of 6(x). This is, of course, inclination and the streamwise pressure
what is done in the method of character- gradient. Equation 7 is multiplied by -u
istics and Kentzer's method. We have in- and added Equation 2 to get
vestigated two such additional combinations
of the differential equations and surface
boundary conditions. Inspection of 2 uv 1 - ^ (38)
Equations (3) and (7) leads to one obvious •y y^ a.^
a2 I ^x
possibility. Multiply (7) by -v and add
to (3) to get Equations 38 and 36 are added to yield

p(uv - vu ) + I 1 -
3L^ r-y
( 1 u^ + uv \ ^ .A v^ + u v \

uv ^ + pu^(T^ - Ty) = 0 (39)

Noting that the streamline slope is given Converting to (^,n) and P we get
by

(34)
6^ n a2
and + uv
+
1 -
(35) (40)

we get 1 (U + UV)C + V -i- uv


2 -
a2
pu^T + 1 - :^ + uv
X a2
which is a well behaved, simple partial
^ Px (36) differential equation which can be very
easily integrated.
This becomes, after converting to (C,n)
and P, Euler Predictor/Simple Wave Corrector
(EP-SWC)
Second order (and higher) accurate nu-
=I ? 0 n
merical procedures for integrating the
T equations at interior points obviously in-
corporate the mathematical/physical signal
propagation phenomena so well characterized
I ' - •U 2 T + T (37)
by the theory of characteristics. The
predictor/corrector procedures (c.f.
where x^ can be computed directly from the MacCormack's Scheme) can be thought of as,
specified wall geometry. A quick inspec- in essence, computing the solutions for
tion of Equation 37 gives considerable two simple waves, the solutions of which
cause for worry, since x appears twice in are summed to yield the complete solution.
the denominator, and x can be zero. This Referring to the sketch,
reflects the fact that Equation 37 embodies
only the continuity equation and the normal
momentum equation for determining the in-
duced streamwise pressure gradient. The 0) (hj + 1
most important equation for determining
that pressure gradient, the streamwise >predictor
momentum equation, was not used in obtain-
ing Equation 37. Nevertheless, a calcula- CD
tion was made for the simple expansion
problem utilizing Equation 37. The re- ycorrector
sults were quite poor, as was expected,
the pressure being low by more than an
order of magnitude in the first step. O (1)3 - 1
That large error is directly attributable
to numerical inaccuracies resulting from i + 1
the X in the denominator.

Another form of this approach is


obtained if both momentum equations and

160
in calculating the solution at (i+1,j), derivatives, along the shock, of dependent
if the derivatives in the predictor are variables. In the case of two-dimensional
evaluated between (i, j+1) and (i, j) steady flow, consider the jump conditions
and in the corrector between (i+1, j) in the form^^
and (i+1, j-1)/ then one can think of the
predictor as contributing the effect of
the signals in the interval (i, j ) , 2Y
(i, j-1) and the interaction of the sig- P - y + 1 Mrsin^e - Y + 1 = 0 (41)
nals from the two intervals. At the
wall, it is not possible to do a regular
predictor-corrector, since there is no [2 + M^ (Y + 1 - 2 sin^e)]T
flow on one side of the wall. The wall
boundary condition replaces "the in-
formation feeding in from the other - 2 cot eCM^sin^e - 1) = 0 (42)
side."
This qualitative view of MacCormack's where p = p^/Pnf T is the streamline slope
scheme leads to a predictor/corrector behind the shock, and t-^ = streamline slope
calculation procedure for solid boundaries. upstream of the shock H 0. Differentiating
The predictor step, the standard Equations (41) and (42) along the shock
MacCormack predictor with two point one we have
sided derivatives, conveys the effect of
the flow near the wall on the solution at
the wall boundary point. In general, the cos 0 p + sm 6 p
X y
surface tangency condition will not be
satisfied by the results of the predictor
calculation. The corrector must account —iXr-M?sin 9 cos 9de^ = 0 (43)
Y + 1 1 OS
for the influence of changes in the sur-
face slope and their interaction with the
interior flow. Locally those changes can
be considered to be a sequence of in- [2 + M^(Y + 1 - 2 sin^9)](COS9T^ + sin9T )
finitesimal simple expansion or compression
waves. Thus, for the corrector we take +[2csc^0(M^sin^9 - 1)
a simple wave, expansion or compression,
which is used to modify the predictor
solution to satisfy the surface tangency - 4M?(T sin 9 cos 9 + cos^9)] 1 ^ = 0
1 ds
condition. (44)
Sharp Shock Calculation Procedures where d/ds means differentiation along the
Basic calculation procedures at shock shock. The compatibility equation can be
boundary points are often similar to tech- written as
niques used at solid walls except that the
surface tangency boundary condition is re-
placed by the jump conditions at the shock. p + Xp + b(T + XT ) = 0 (45)
^x ^y 'X y'
The main difference is that, in advancing
from station i to i+1, the shock inclina- Converting Equations (43), (44), and (45)
tion at i+1 is unknown, whereas the surface to (^,ri) space yields a system of three
slope at that station is known. Sharp linear algebraic equations for pr, Xr,
Shock calculation schemes tested include: d9/ds in terms of data known on an initial
data line (station i) , including such r\
• Method of characteristics derivatives as p^ and T . The result is

• Kentzer's scheme
d9 ^ CD cos^6
• Barnwell's scheme (46)
ai" " Bb - AC
• Thomas scheme where

• Victoria's scheme A = 4 —^^—^Jsin 9 cos


Y + 1 1
• Combined equations
B = 2[M^(1 - 2 cos^9 - 2 T sin 9 cos 9)
Method of Characteristics - The calculation
is similar to that described for a solid - csc^9]
wall, except an iteration is involved to
determine the shock slope at station i+1 C = 2 + M^(Y + 1 - 2 sin^9)
and the up running characteristic is used.

Kentzer's Scheme - The jump conditions are D = -[(X - tan 9)p + b(X - tan e)T ]
differentiated along the shock, and the
resulting partial differential equations Barnwell's Scheme 12 - As in the method of
are combined with the characteristic characteristics calculation, the compati-
compatability equation, resulting in a bility equation is solved simultaneously
system of linear algebraic equations for with the usual form of the jump conditions.

161
However, the characteristic compatibility Victoria's Scheme - K. Victoria has sug-
equation is written in difference form gested an approach, the basic idea of which
along the shock wave, not along a charac- he credits to T. Kubota*. Again, the jump
teristic. Thus, one obtains a system of equations are differentiated along the
simultaneous nonlinear algebraic equations shock. Each differentiated jump equation
for the unknown dependent variables, in- is then combined with the corresponding
cluding shock angle, at point (i+l,J). field equation, eliminating the ^ deriva-
tives of PfU, and v to form a system of
Thomas' Scheme ' - Thomas' idea is to three inhomogeneous linear algebraic equa-
combine the jximp conditions with one of tions of the form
the partial differential equations from
the interior. Thus, Thomas' method is de_
similar to Barnwell's except that the ^1 ds ^ ^ 1 % + d.
compatibility equation used by Barnwell is
replaced by one of the equations in the de
basic set of governing partial differential d^ «3-- = bo P„ + c^v^ + d^u (49)
equations (Equations (1)). The advantage 2 as 2 ^n 2 n 2n
of Thomas' approach is that one can mani-
pulate the equations so that they can be a-, 3— = b^ p + c-,v . n
solved sequentially rather than simul- 3 ds 3 ^n 3 n + d^u
taneously. For example, the jump condi- where, of course, p ,v , and u are known
tions can be put in the following form at station i. Thes8 e§uations^can be com-
bined to form the three equations:
Y + 1 2. + Y - (47) de
2 BTP
Pi Y + 1 ^1 a^ i^n

tan e = (48) H = B,v^ (50)


/^— -Nf
where N, = velocity component normal to de
the shock. The solution then proceeds as B^u
follows. A predictor value of Pi.-i -r is 3 n
obtained from the regular predictor' step Presumably, identical results would be
used at interior points. This gives obtained with each of Equations 50.
(Po/P-i) i-i-1 J which, with Equations (47)
ana (487/ ^ives the shock angle, ^±j^\f at Jump Equations and Simple Finite Difference
the station i+1. Then the corrector Equations - In Barnwell's scheme, the
finite difference step is used to obtain characteristic compatibility equation
the final value of Pj_4.2 J' which then written along the shock is the additional
leads to a final value 6f Oi+i. Note that equation which, when combined with the
the n derivatives must be taken from the jump conditions, yields a complete system
same side for both the predictor and the of simultaneous nonlinear equations for
corrector steps. Also note that Thomas' the shock solution. Obviously, any other
scheme appears to have a rather loose equation which is not a linear combination
coupling between the jump conditions and of the jump equations, and which does not
the partial differential equation used introduce additional variables can be used
from the interior. Compare it, for in- in place of the characteristic compatibility
stance, with the intimate coupling of the equation. In particular, the differential
full MOC approach and Barnwell's scheme, form of the continuity equation or one of
both of which require simultaneous solu- the momentum equations would suffice, the
tion of the system of nonlinear equations results being more or less accurate. This
obtained when the jump conditions are approach has been used by Li-^.
combined with the compatibility equation.
Also compare the coupling in Thomas' First a guess is made of the shock angle,
scheme with that in Kentzer's scheme where e-JL+i, at station i+1. Then the conditions
the "interior" and shock equations are behind the shock are calculated from the
also solved simultaneously. This lack jump conditions (see Equations). Then
of close coupling between the interior either p, u, or v is selected as the pri-
flow field and the boundary conditions re- mary variable, and the corresponding field
sulted in quite poor results at a solid equation is solved. The prediction of the
wall, but the ability of the scheme to jump equation is compared to that of the
describe sharp shocks is much better. The field equatioix, and ej^^^ is adjusted until
reason is quite simple. In the sharp they agree.
shock version of Thomas' scheme, all the
shock jump equations are used fully, while Calculations were done with the field
the surface boundary condition was weakly equations solved at the shock point in
incorporated in the differential equations both conservative and nonconservative
solved on the surfacel/5. Thus, in
Thomas's scheme, the shock boundary condi-
tion is incorporated far stronger than is
the surface boundary condition. Due to a misinterpretation of some notes,in
Ref. 6 a similar but less accurate procedure
was attributed to Victoria.
162
form. The standard predictor-corrector These results, which are representative of
scheme of MacCormack was used to solve the more extensive results reported in
the field equations, except that the n Reference 5/demonstrate that many
derivatives were calculated with two point procedures in common use are quite in-.
one sided "backward" differences in both accurate. The best results are achieved
the predictor and corrector steps. with the method of characteristics and the
new Euler predictor/simple wave corrector
5. Results and Discussion (EP/SWC). The maximum error for those two
schemes is 0.28 and 0.10 percent for the
As already noted, a large number of compression and 0.58 and 0.43 percent for
calculations were performed during the the expansion respectively. These results
course of this study, and space limita- are achieved in calculations where the sur-
tions prohibit presenting most of the re- face pressure changes by as much as 26 per-
sults. Therefore, in this section we pre- cent in one integration stepl Kentzer*s
sent and discuss the most important re- scheme gives excellent results on the
sults. The results and discussion are pre- compression, the maximum error being 0.56
dominately concerned with the calculation percent. It is not quite as good as MOC
of surface pressure, with the sharp shock and EPSWC on the expansion, the maximum
calculation procedures playing a relatively error being about 10 percent, though it is
minor role. considerably better than the remaining
schemes. The maximum error for each scheme
Surface Boundary Points - The results for is shown in Table 2 where it is clear that
the simple compression and simple expan- most of the schemes are subject to very
sion (test problems I and II) are presented large errors.
in Figures 3 and 4 respectively. The
boundary condition calculation procedures
are summarized in Table 1. All these Maximum % Error
Scheme
calculations were done with interior points Compression Expansion
computed in non-conservative form. With
the exception of Thomas' scheme, in all Reflection 26.5 27.1
calculatins shown on Figures 3 and 4 Explicit 60.6 27.1
use was made of the known surface entropy, Implicit 17.9 10.6
total enthalpy, and slope to obtain p, u, 1 Predictor-Corrector 35.7 17.0
V from the calculated p. The step size Method of Characteristics 0.28 0.58
1 Kentzer's Method 0.56 10.0
in each run is approximately Ax/R = 0.05.
Thomas' Method 62.5 26.2
Combined Equations 13.4 7.8
Procedure Brief Description EP/SWC 0.10 0.43

Reflection ^i+1 2 csLlculated from Eqs. 8,20 as at


interior points. Data at i,l obtained TABLE 2. r^AXIMUM ERROR FOR EACH SCHEME-
by reflection in physical plane.
PROBLEMS I AND II (FIGURE 1)
- e.g. u.^^ = u.^3(cos^9.^2 " =^"'^1,2^

+ 2v^ 3(sin 0^ 2^°s ®i 2^


Moretti has suggested an alternate ver-
sion of Kentzer's scheme in which the co-
Explicit Eq. 8 integrated explicitly for Pi+i 2 efficients of the derivative terms in Eq.
using two-point one-sided n derivatives 28 are averaged over the integration step.
evaluated at station i
With this averaging the maximum error on
Implicit Same as Explicit except n derivatives the simple expansion was reduced from 10
evaluated at station i+1. percent to about 4 percent.
Predictor- Eq. 8 integrated as at interior points
Corrector except both predictor and corrector Additional comparisons reported in Ref.
derivatives evaluated between j=3 and
j=2 1 5 demonstrate that
Method of See Discussion in Section 4 • Predictions are usually more accurate
Characteris-
tics if full use is made of the known sur-
face entropy, total enthalpy, and
Kentzer's
Method
Eq. 28 integrated along wall with y (i.e
(i.e., n) derivatives calculated with
surface slope.
two-point one-sided differences
• Greater errors result from the use
Thomas Eqs. 31 and 8 integrated using two-point of three-point one-sided derivatives
one-sided n derivatives evaluated at |
station i than from two-point derivatives.
Combined Eq. 40 is integrated using two-point one- It is a common misconception that three-
Equations sided n derivatives evaluated at station i
point, second-order-accurate, one-sided
Euler Predictor value of p, p, u, v at i+1,2 derivatives should yield more accurate
Predictor/ calculated with eq. 20a. Corrector value solutions than two-point one-sided de-
Simple Wave of p obtained from simple wave to turn the
Corrector flow parallel to wall. rivatives. As a general rule, the reverse
(EP/SWC) is probably true. Three-point derivatives
violate the laws of signal propagation
TABLE 1. SUMMARY OF PRIMARY SOLID WALL and often lead to numerical instability^,10.
BOUNDARY CONDITION CALCULATION PROCEDURES
TESTED

163
Accuracy of the Solution at Interior Points Further downstream, there will be more
Simple Compression — The Envelope Shock - points between the shock and the lower
For the compression problem it is important wall (see Figure 5 ) .
to consider the influence that the envelope
shock which will form may have on the re- In a study of one-dimensional time-
sults. No provision was taken to treat dependent flows with shock waves computed
that shock as a sharp discontinuity, so it by shock smearing, Gary^^ compared shock
is possible that, particularly with those speeds predicted by Lax-Wendroff differen-
schemes using three-point one-sided dif- cing of the Euler equations in divergence
ferences, the solution at the lower wall is form with predictions using the equations
is affected by numerical inaccuracies re- in non-divergence form (as done in this
sulting from the shock smearing. study). Gary obtained much greater errors
in computed shock speed utilizing non-
As a first step to evaluate this effect, divergence form than when the divergence
a full method of characteristics calcula- form was used. Kutler anticipated*
tion was generated with a code which com- similar behavior in the present case, ex-
putes shock waves as sharp discontinuities. pecting the smeared shock location to dif-
The code (referred to as SUPER), detects f^.r significantly from the correct value.
envelope shocks by the crossing of charac- To investigate this behavior as well as to
teristic slopes^ and coefficients in the study the effect of the shock smearing on
equations are averaged in a solution which the present results, the computed pressure
iterates each step. In order to obtain profiles at two sections are plotted for
mesh point spacing on the lower wall at Case EP-SWC in Figure 6. In Figure 6 we
about the same interval as in the "finite see that when the shock is just beginning
difference" calculation, the mesh points to form (x/R ~ .25) the region of strong
on the initial data line were spaced at pressure gradients as predicted by the
distance Ay/R = 0.01 apart. The calcula- finite difference grid somewhat lags that
tion proceeds from mesh points on the ini- predicted by the full method of character-
tial data line (x = 0.0) along right-run- istics calculation. The finite difference
ning characteristics, and it was continued solution overshoots by about 9 percent the
until the code halted because the number pressure profile on the compression side.
of points on the characteristic line ex-
ceeded 50, the maximum allowed. The last Because of the mesh point limitation,
computed right-running characteristic we have no complete method of character-
originated at x/R = 0.0, y/R = 1.28. The istics solution for all y beyond x/R = 0.37.
surface pressure computed with the full However, since the computed shock angle is
method of characteristics calculation very close to the asymptotic value, we can
agrees very well with the exact isentropic extrapolate it to obtain the shock loca-
compression solution. The resulting flow tion at another abscissa, for example for
field and some interesting additional in- x/R = 0.5264 (Figure 6 ) . Then it is easy
formation are shown in Figure 5. The shock to construct the exact pressure profile
was first detected at x/R = 0.254, at which and to compare it with the prediction using
the finite difference formulation with shock
point a sharp shock solution yielded a
smearing (Figure 6 ) . Again, we see the
pressure ratio of p> = P2/Pi = 1.129. In displaced shock location and pressure over-
order to determine as closely as possible shoot (27 percent) predicted with shock
the actual abscissa of the beginning of the smearing.
shock, the pressure ratio across the shock
was plotted as a function of x/R and the
plot was extrapolated back to the abscissa Now, what is the importance of these
where p" = P2/Pi = 1-0 yielding the initia- results vis a vis this study? One might
tion of the shock at x/R = 0.24. The right remark, for instance, that if the envelope
running characteristic passing through the shock were treated as a sharp discontinuity,
that point hits the lower wall at x/R = the problem of inaccuracies due to displaced
0.38. On the last computed right running shocks and pressure overshoots would dis-
characteristic, the shock was located at appear. However, that is misleading, for
x/R = 0.379, y/R = 1.146, and the shock the inaccuracies in boundary point calcula-
slope has alm.ost reached its asymptotic tions do not primarily originate with the
value (see Figure 5 ) . Also shown on poor shock description. Also, no matter
Figure 5 are the initial finite difference how the shock is treated, we must always
grid for the calculations reported herein, live with some errors which will result
the finite difference grid at x/R = 0.255,* in the interior calculation even if the
the left running Mach Line which originates boundary solutions are exact. Furthermore,
at the beginning of the compression turn, we must face the situation depicted in
and the asymptotic shock angle. Note that, Figure 6, where the shock, as a shock, con-
including the lower wall point, there are tributes a negligible part of the compres-
four mesh points between the lower wall sion, but where there still is the usual
and the shock in the region x/R = 0.25. pressure overshoot which occurs even when
the equations are cast in divergence form.
Thus, it is just as erroneous to replace

At a given x, there are twenty equal in-


tervals between the lower wall and the or-
dinate y/R = 1.4. Private communication.

164
that smooth compression with a shock as interior equations in both conservative and
it is to represent the compression through non-conservative form, with the method of
a shock with shock smearing. Even if we characteristics and EP/SWC schemes at the
attempt to devise a code in which shocks lower wall, and with MOC, Thomas, and
are essentially always treated as sharp Kentzer's schemes at the sharp shock. The
discontinuities, we must realize that if calculations are summarized in Table 3.
the flow is at all complicated, weak
secondary compressions and shocks may exist
which are not accounted for within the code Run Lower
Shock Interior
logic. Our task is to insure that in such No. Wall
situations relatively minor inaccuracies
in the interior do not lead to large in- 1 MOC MOC Conservative
accuracies at boundaries. 2 EP/SWC MOC Conservative
3 EP/SWC Thomas Conservative
Finally the purpose of this study is 4 EP/SWC Thomas Non-conservative
to evaluate various computation procedures 5 EP/SWC Kentzer Conservative
in situations they are expected to handle +
as a matter of course. All the c3omputa-
tions are identical except for the boundary
point computations, and the tests have been TABLE 3. PROBLEM III CALCULATION MATRIX
of situations typical of those faced by
everyday working codes. As a standard, the flow field was calculated
with the full method of characteristics
Simple Expansion ~ To further clarify some code. The calculated flow field is shown
aspects of the interaction between solu- in Figure 8 and the calculated surface pres-
tions at boundary points with those at sure distributions in Figures 9 and 10 (Run
interior points, we consider the simple 5 gave results identical with Run 3, both
expansion problem. Since the solution is agreeing quite closely with Run 2 ) . In
just a simple Prandtl-Meyer turn, we can Runs 1-5, there were 15 intervals between
calculate the exact solution at any point the lower wall and the bow shock. There
in the flow field. First we consider one were 12 such intervals on the initial data
of the better schemes for computing bound- line of the full method of characteristics
ary points. calculation. Examination of Figures 8-10
indicates that in this case:
The results of the EP-SWC calculation
are compared with the exact solution at • EP/SWC is as accurate as MOC on solid
x/R = 0.5095, in Figure 7. walls.
Note that the agreement in the expansion
region is quite good, but that there is a • There are negligible differences bet-
tendency to develop small wiggles near the ween MOC, Kentzer, and Thomas at the
discontinuity in profile slope. However, shock.
as was demonstrated above this is a scheme
for computing boundary points which is in- • '^^ all shocks are not treated as
sensitive to these small wiggles, and the sharp shocks, conservative differencing
wiggles do not have a deleterious affect should be used at interior points to
on the boundary calculations. minimize errors in calculated surface
pressure.
It is also interesting to compare the
pressure profile for one of the poorer • The largest source of error in Runs
schemes with the exact solution. The re- 1-3, 5 is the truncation error at
sults for Thomas' scheme at x/R = .43 are interior points which tends to smooth
shown in Figure 13. Note that the large the pressure profile where it should
errors in computed surface pressure do not have a discontinuous slope.
really have much effect on the interior
flow, at least in this case. However, it Shock Boundary Points - The basic studies of
is important to remember that surface sharp shock calculation procedures were per-
pressure is generally the single most im- formed with problem IV, Figure 1. Again,
portant result of these calculations! space limitations prohibit presentation
and discussion of all the results, most of
The first two test problems contain the which can be found in Ref. 6. Full method
most important ingredients of flow field of characteristics calculations were genera-
calculations, expansions and compressions. ted with three different mesh sizes to
However, their simplicity might possibly establish the standard. Those calculations
result in misleading conclusions, par- had 7, 13, and 25 points on the initial data
ticularly with respect to the EP/SWC scheme .line between the shock and the lower surface.
Problem III, the 20° v/edge/circular arc/30° Calculations with finite differences at
wedge was selected to further test that interior points (conservative differencing),
scheme. In addition, since only the lead- MOC at the shock, and EP/SWC at the lower
ing shock is calculated as a sharp shock, surface with 6, 11, and 26 points between
this problem affords us with a means to shock and lower surface were generated. The
further evaluate the effect on surface pres- calculated flow fields and pressure distri-
sure calculations of smearing very weak butions behind the shock and on the lower
shocks. Calculations were done with the surface for these six calculations are given

165
in Figures 11, 12, and 13. It is interes- for the sharp shock calculation. The pro-
ting that, except for the knee of the curve blem is one of violating the law of for-
for pressure behind the shock, the dif- bidden signals when the three-point one-
ference between different mesh sizes is sided derivative is used (see the Ref. 5
much less for the finite difference than for further results in this vein).
for the full MOC calculation^ and the
agreement between the rough mesh FD-MOC 6. Extension of the EP/SWC Scheme
calculation and the fine mesh full MOC to the Three Dimensions*
calculation is significantly better for
large x/R than it was for the rough mesh In Section 4 we related the components
full MOC calculation. (2.3 percent vs 6.5 of the Euler Predictor/Simple Wave Corrector
percent and 2.0 percent vs 4.5 percent res- scheme to the gas dynamic interactions
pectively behind the shock and on the occuring near and at solid walls. The dis-
lower wall). At first this seems rather cussion depended heavily on two dimensional
unusual since one generally considers considerations. Results to date indicate
method of characteristics calculations to that this scheme is quite powerful, yield-
be more accurate than finite difference ing predictions of accuracy comparable to
calculations for a given mesh size. How- the method of characteristics. Because
ever, for the rough characteristic mesh, MOC is quite inconvenient in three
along the right running characteristic dirensions, we are interested in extending
originating on the shock at x = 1.107, y = the EP/SWC scheme to three dimensional
1.774 and terminating on the lower wall at steady flow. An attempt to do so was re-
X = 4.939, the coefficient b in the com ported earlier,^ and comparisons with full
patibility equation (b = pu^//M^ - 1) 3-D method of characteristics calculations
varies by 70 percent between mesh points are quite faborable.15 However, that
in the center of the expansion! Since initial extension to three dimensions
the finite difference equations are possesses a certain degree of arbitrariness
written in conservative form, there is no which is undesireable, and an improved
coefficient multiplying the partial deri- procedure has recently been constructed,
vative terms. It is probable that the though it has not yet been tested.
apparent superiority, at least in this
problem, of the finite difference cal- The improved procedure rests heavily on
culation at interior points for rough mesh the concept of MacCormack's predictor-
can be explained in this manner. corrector scheme outlined in Section 4 in
the discussion of the EP/SWC scheme, namely
The results of the prediction using that the predictor conveys information frpm
Kentzer's scheme are comparable to those one side and the corrector conveys informa-
obtained with the method of characteristics tion from the other side as well as accoun-
at the shock (compare Figures 13 and 14). ting for the interaction between the signals
Similar results were obtained with from the two sides. The extension to three
Barnwell's, Thomas', and Victoria's schemes dimensions can be briefly outlined using
as well as with the approach using the MacCormack's operator notation. Consider
shock jump equations solved simultaneously a cartesian coordinate system (x, y, z)
with a field equation for pressure. with indices (i, j , k) respectively and
with the flow in the general x direction.
All the calculations used two-point one- Then for the equation
sided backward derivatives at the shock.
As already noted, a common misconception
is that three-point, second-order-accurate, "x + ^y + <^z = 0 (51)
one-sided derivatives should lead to more
accurate solutions. To investigate this
matter a little further, two additional MacCormack's- scheme can be written as"
calculations were done using Thomas' scheme.
In the first (referred to as Thomas 3-2),
three-point one-sided derivatives were used
in the predictor step [e.g. , pr, = (-3p-L j _j.
^i:i = L L L L Ut ,
y z z y 3,k (52)

^Pi J-1 ••" Pi J-2)/2An] and two-point one- where L^ and L are two dimensional operators
sided derivatives were used in the cor- of the nature of Equations 20, each con-
rector. In the second case (Thomas 3-3), sisting of two components, a predictor and
three-point derivatives were used in both a corrector. Applying the EP/SWC technique
predictor and corrector steps. The results in three dimensions involves replacing the
are shown in Figure 15 along with the corrector component of the Ly operator with
standard calculation using two-point de- a two dimensional simple wave. In so doing,
rivatives in both the predictor and cor- note that the solution of the predictor
rector steps (Thomas 2-2). Both schemes component of Ly involves all three velocity
Thomas 3-2 and Thomas 3-3, gave unaccep- components. Therefore, for this approach
table results; the latter calculation to yield accuracy comparable to that
"blew up" after x/R = 2.4. Thomas earlier achieved in two dimensions, it is probable
alluded to instability problems with three- that the coordinate system should be
point one-sided derivatives.H It should
be noted that these results are not
characteristic of Thomas' scheme. That
basic approach appears to give good results It is assumed that the reader of this Sec-
tion is familiar with MacCormack's work,
particularly Ref. 3.
166
oriented so that the y derivative term Sharp Shock Boundary Point Calculations
in the z momentum equation is small com-
pared to the z derivative terms, for that • Truncation error at interior points
equation is omitted entirely from the is greater than the differences bet-
simple wave corrector. ween shock calculation procedures.

7. Concluding Remarks • Surface pressure is relatively in-


^nd Recommendations"" sensitive to local errors in shock
calculations.
The most important result of this study
is the demonstration that many calcula- • All the schemes tested resulted in
tion schemes in common use are too in- comparable accuracy.
accurate for the purposes to which they
are applied, for we are considering tech- • If weak shocks are smeared, the con-
niques used in codes intended to give exact servative form of the field equations
solutions as the mesh size vanishes and should be used at interior p o i n t s .
very accurate solutions for operational
mesh sizes. There is little justification Therefore, for sharp shock c a l c u l a t i o n s , it
in using an expensive computer code cal- appears that the simpler schemes can be
culation to obtain results whose used with little loss in accuracy at the
accuracy is no better than what can be surface.
obtained by relatively simple approximate
theories and correlations. Acknowledgment

The main conclusions are: Many researchers responded helpfully to


my request for results and suggestions to
Surface Boundary Point Calculations make this study as complete as possible.
• Any purely explicit integration G. M o r e t t i , C. K e n t z e r , R. MacCormack, and
procedure not incorporating ex- P. Kutler have each transmitted a number
plicitly the wall geometry changes of morsels for thought during the past
possesses an inherent lag in that two years. I regret that limited time has
it has no mechanism for introducing not yet permitted m e to satisfactorily
into the solution for one step address every suggestion. I am particularly
changes in wall geometry occuring indebted to G. Moretti for the many years
duirng that step. of tutelage and encouragement I enjoyed at
his side learning a little of this exciting
field.
• Higher order one-sided differences
do not generally yield better re-
sults. They tend to be more quickly Symbols
responsive to changes in surface
slope, but they also show strong E energy per unit volume
tendencies to large, poorly damped h static enthalpy
oscillations.
H total enthalpy
• Three of the calculation procedures if J finite difference mesh indices
give markedly better results than
any of the others. The best pro- M Mach number
cedures are the method of character- P pressure
istics, the simple explicit predic-
tor/simple wave corrector procedure, P In p/p^
and Kentzer's method. Though very R In p/p^
competitive on the compression,
Kentzer's method did not do as well s distance along shock
comparatively on the expansion pro- S entropy, S = P - yR
blem.
u,v velocity components
• In short, accuracy depends critically x,y cartesian coordinates
on how the field equations are
coupled with the surface boundary ^rr\ transformed coordinates
conditions. Y specific heat ratio
For best accuracy, the method of character- X characteristic slope
istics and the EP/SWC scheme are recom- P density
mended first, though the recommendation of
the EP/SWC scheme in three dimensions is T v/u
qualified because insufficient comparisons e flow angle, shock angle
have been made. Because of its comparative
simplicity, Kentzer's scheme is a good sub- Subscripts
stitute for the full MOC if moderate local o stagnation conditions
errors can be tolerated.
w wall
oo free stream

167
References 11. Thomas, P. D., "On the Computation of
Boundary Conditions in Finite Differ-
1. Moretti, Gino, Complicated One-Dimen- ence Solutions for Multidimensional
sional Flows, Polytechnic Institute Inviscid Flow Fields," Lockheed Palo
of Brooklyn, Farmingdale, New York, Alto Research Laboratory, Palo Alto,
PIBAL Report No. 71-25, September, California, LMSC 6-82-71-3, March 2,
1971. 1971, Revision A, September 13, 1971.

2. MacCormack, R. W., "Numerical Solu- 12. Barnwell, Richard, "Three Dimensional


tion of the Interaction of a Shock Flow Around Blunt Bodies with Sharp
Wave with a Laminar Boundary Layer," Corners," AIAA Paper 71-56, presented
presented at the Second International at Ninth Aerospace Science Meeting,
Conference on Numerical Methods in New York, N. Y., January 25-27, 1971.
Fluid Dynamics, University of
California, Berkeley, California, 13. Li, C. P., "Time Dependent Solutions
September 15-19, 1970. of Non-Equilibrium Airflow Past a
Blunt Body, AIAA paper 71-595, pre-
3. MacCormack, R. W., and Paullay, A. J., sented at the Fourth Fluid and Plasma
"Computational Efficiency Achieved Dynamics Conference, Palo Alto,
by Time Splitting of Finite Difference California, June 1971.
Operators," AIAA paper 72-154,
presented at the Tenth Aerospace 14. Ames Research Staff, "Equations,
Sciences Meeting, San Diego, California. Tables, and Charts for Compressible
Flow," NACA Report 1135, 1953.
4. Johnson, J., ed., "Investigation of
the Low Speed Fixed Geometry Scram- 15. Rakich, John, and Kutler, Paul,
jet. Part I - Inlet Design Practice "Application of Shock Capturing and
Manual," General Applied Science Semi-Characteristics Methods to
Labs, Westbury, New York, TR-667, Shuttle Flow Fields," presented
September 1967 (also AFAPL-TR-68-7, at 10th Aerospace Science Meeting,
February 19 68). AIAA, San Diego, California, January
1972.
5. Abbett, Michael J., Boundary Condition
Computational Procedures for Inviscid, 16. Gary, John, "On Certain Finite
Supersonic Steady Flow Field Calcula- Difference Schemes for Hyperbolic
tions, Aerotherm Corporation, Mountain Systems," Math, of Computation, 18,
View, California, Report 71-41, Nov. January 1964, pp. 1-18.
30, 1971.
I. II.
6. Abbett, Michael J., Sharp Shock Com- Y = 1.4
Y = 1.4
putational Procedures for Inviscid, Uniform Uniform
Supersonic, Steady Flow Field Cal-
culations, Aerotherm Div., Acurex
Corp., Mountain View, California, simple compression simple expansio'n
Report 72-50, June 30, 1972.

7. Bohachevsky, I. and Rubin, E., "A


Direct Method for Calculation of the
Flow About an Axismmetric Blunt Body III.
at Angle of Attack," AIAA Journal, £,
5, May 1966, 776-782.

8. Bohachevsky, I. and Kostoff, R. N.,


"Hypersonic Flow Over Cones with
Attached and Detached Shock Waves,"
AIAA Paper No. 71-55, presented at 20" Wedge/Circular Arc/aO" Wedge
the Ninth Aero. Science Meeting, New
York, N. Y., January 25-27, 1971.

9. Kentzer, Czeslaw P., "Discretization


of Boundary Conditions on Moving
Discontinuities," presented at Second
International Conference on Numerical
Methods in Fluid Dynamics, University
of California, Berkeley, California,
September 15-19, 1970.

10. Thomas, P. D., et al., "Numerical Solu-


tion for.the Three Dimensional Hyper-
sonic Flow Field of a Blunt Delta
Body," AIAA paper 71-596, June 1971.

FIGURE 1. TEST PROBLEMS

168
Lower Surface
i—>-?
FIGURE 2 . Physical and Computational Spaces

.10 .>0|
.(»
tA-
.081

.07 .01 7:
^5 ^^arr
P/l=^
M
P/Po
.Of
v-/
.03 eePLecTioN .03
IMPLICIT
EXPCICIT peeDtcToe-coiZKecTop.
THOMA?>

.OZj

.-2
X/t2 y/(2

F»GUEE. 3 COMPA;2(SOM OF WALL BOUNiDAl^Y COMDlTlOM CALCULATIOM


P^OCEQUee.S WITH EXACT SOLUT/OM - SIMPLE COMPf^eSSlOM

.040,

P/Po P/Po

FI6Ui2E 4- COMPAKl-SONi OF \AJ^LL BOUMPA^Y CONiCPlTlOM CALCULATIOM PI^OCEPUiSES


W\TH EXACT SOLUTION - -SIMPLE E:XPANiSlOM

169
-2-7.3°- ASYMPTOTIC
y/s

E N i P OF ^ u J e v / E ^ E 6 T l

Fi^u;2e 5 (rlOMpQTED FLbW F\ELD F^OCS "THE ^OMP^^e-^-^lOM


PeoBLGM

.©o . 1 1 1_
EXACT -^OuUTiON
<.oi-u~riON .Uo - Q THOMAS-Z
O E P - -SNAJC

o .40
o
^ O ^—^wocjc i-CiC-fcnohi
^^TlMATETD FeOKA
Wf^ <i:i^^eA^lx£ ei«^-riii<. .7.D
<::AL<iuLA,T(<:iNj
Z LOWER •^HOWM I N F^l^. ©
WM-I_
P/Po
. lO
UPPER - /
WALL 08

I
\.'Z

04

j
_.__] •" I n
«CHAP^cTee\6.Tic-e>
5") -SCHEME -THoMA-S-2
C-AL<CULA,TlCM 1

Q EP-'SWd \(sjTe(2ioe
FLQNAJ FIFLO 1
.1? l.U
--.OLun~ I O M ; ^ — i—
- EXACT -SOLUTION

.4f
- LOWE P W^UL
Flfl
,oe\—
L/ n I

\
\
\ V

,oe

1.04- 1.0 b 1,06 i.io 11.4


Y/e
^) X/ e= .-243
fj) "x|e= ©.so-^s
FI6U(2E: 6 ^oMPAKi^oN OF P e B ^ s u p e peoFiLE^
Fi<:Suee 7 SIMPLE exPAN-sioN- COMPAPI-SOM
< 1 0 M p u T E O 3 Y P"ULL M E T H O O O F OF PE-SUUT^ OF e'P-'S.wc -scweMC
^ M A P A d T E e i - S T l C S Ct^LCULA~riOH WITH W\TH EXACT -SOLUTION FOP
T H 0 4 £ OF F i M i T e P i P p e e E N C E S O L U T I O N P'ee'S'SURe PROFILES
U ^ I N ^ B O O N O A e V <:10NiC:>\T(OM ^<l.v^eMB
e^P-5.UgC ( C O M P K E ^ ^ i O M P J ^ Q B L E / V l )

170
t 1 1 1 1 1
-4
. ^

5^ 2
(jl^
- ^
/ y / / / J,
/ / / / / / / / / / / / / /
\

/=^/30!eS^ II C^/^/='/P/^/S<^A/^^T^^^^A/y^C?C/PA//P/='/?-Af(^C

FlcaUKH a E F F E C T O F MOMCOM6ERVOTIVE. DlFFE^eNCJt.


W I T H E M B E P P £ D SHOCK'S OM ^SHAEP 5 H 0 C I ^ C A L C l
LAT\ONi'S.

3?

F - P i N r e E l O R , MOC AT ^ H O C K AMD ^ ^ U L
fw/P^. F-c> i N ) T £ E i o e T N\oc AT ^ H o c k : , epZ-s^Nii
AT NAJALL
c o M 6 > e e v A T i v ; £ Q\FpeeesiCiN<2» AT
- INSTe!2lO(2 PO(MTt>

Fl^iUeB S C0MPAKI50M OF ACCUK^CY OF NAOC AND 6P(5\\


6(CHEMes OM S O L I P WALL - 20''WeCG^e/ARC/30°
NAJEPcSB

Ife

1
14 \\
K
12
^1 " 1 1
-FULL METMOP OF ^ H A2/\OT£El'STl(lf>
FiN\TE . P i F F . coNiseevjATive F O f ^ M - r w o M A s
Kj/P« AT -&M OCK, EPSWC AT 'SUeFACe .
— F i N i r e C?lFF. NON dCN'^.EKvATlVe F O ^ M -
THOM ^6 AvT ^MOCK , EPSNAJC AT 'SU'^P^ACE.

Fl^iUee. 10 EFFECT OM K\CM CCNSG^VATIVE PiFFEeEMClM^j


NAJITH E M i ' - E P P E P :5HOCl<^S CM SUk?~AC£ peESlbOSe
CALCULATION - 20^^ w e P 6 E / A S ' C / ^ 0 ° WEP6c

171
/f.) /^^€'SS4//?^€^ ^£///A/0 SA^iPc/er

8^ ^£>f^/'£•/^• A<^^c /='i^^:sso'/ee' ^:>/:^7~y^/^i/r/<pAy.


/7S^^£/^ (:c>/^^/9/e/s<:p/]/ ^/=-/(r£/^r^<E-/^'s SCA/<£A^£ i^/r/i

/ / Av/A/rs aA/ /A//r//>c


rAy<pA^/fS3-z(<i,^a/A/r£) '
a^sX 6 ^^/A/rs (PA/ /AA/r/AC rA/aA^/9S 3 - 3 (6 <»<?/A/rs)
£>A)r/9 ^/AAS-
\

^ /.a

^^//=7=^£/^s^/vr CPA/^'-S/^E-O ^//^/=^s^eAyd:^


sc/ze/^es.
172

Das könnte Ihnen auch gefallen