Sunteți pe pagina 1din 10

NUCLEAR SCIENCE AND ENGINEERING: 66, 93-102 (1978)

Characteristics, Stability, and Short-Wavelength


Phenomena in Two-Phase Flow
Equation Systems

John D. Ramshaw* and John A. Trapp+


Aerojet Nuclear Company,t 550 Second Street
Idaho Falls, Idaho 83401
Received January 25, 1977
Accepted October 27, 1977

The occurrence and significance of complex characteristics in two-phase flow equation


systems are clarified by a detailed analysis of separated two-phase flow between two parallel
plates. The basic system of one-dimensional two-phase flow equations for this problem pos-
sesses complex characteristics, exhibits unbounded instabilities in the short-wavelength limit,
and constitutes an improperly posed initial value problem. These difficulties have led some
workers to propose major modifications to the basic equation system. We show that the rela-
tively minor modification of introducing surface tension is sufficient to render the characteris-
tics real, stabilize short-wavelength disturbances, and produce a properly posed problem. For a
given value of the surface tension, the basic equation system thus modified is shown to correctly
predict the evolution of small-amplitude disturbances having wavelengths long compared to the
plate spacing. A formula is given for the artificial surface tension necessary to stabilize wave-
lengths on the order of the mesh spacing in a finite difference numerical calculation. A brief
discussion is given concerning the expected behavior of surface tension as compared to viscosity
in the nonlinear regime. The general relation between characteristics and stability is discussed
in an appendix.

I. INTRODUCTION equations being solved had complex characteristic


roots and therefore constituted an ill-posed ini-
In recent years, there has been much confu- tial-value problem. That is, the governing equa-
sion about the occurrence and significance of tions themselves, although apparently physically
complex characteristics in two-phase flow equa- well-founded, were inherently and catastrophically
tion systems. 1 This issue first received wide- unstable. Possible causes and cures of this
spread attention following early attempts to obtain problem became the subject of a spirited dialog
numerical solutions to two-phase flow problems in that continues to the present day.
nuclear reactor safety analysis. These attempts Although the complex characteristics issue is
encountered unexpected stability difficulties that relevant to two-phase flow in general, it has been
at first appeared to be numerical in nature. particularly troublesome in nuclear reactor safety
Subsequent analysis showed that the differential analysis because of the greatly increased re-
search activity in this area in recent years. A
*Present address: Los Alamos Scientific Laboratory, P.O. correct understanding of this issue is essential if
Box 1663, Los Alamos, New Mexico 87545. stable and physically meaningful reactor safety
+Now a consultant to EG&G Idaho, Inc., Idaho Falls, Idaho
83401.
calculations are to be performed. Unfortunately,
tNow EG&G Idaho, Inc. misunderstanding still persists. This situation is
1 aggravated by the lack of a single unified discus-
D. GIDASPOW, Chairman, "Modeling of Two-Phase Flow,"
Proc. Round Table Discussion RT-1-2 Fifth Int. Heat Transfer sion of the complex characteristics question in the
Conference, Tokyo, September 3-7, 1974, Heat Transfer I974, open literature. The present paper provides such
Vol. VII, p. 163 (1974). a discussion. It is based on an investigation we

0029-5639/78/0004-0093$02.00/0 © 1978 American Nuclear Society 93


94 RAMSHAW and TRAPP

carried out in 1974, which was previously docu- where o:;, p;, and u; are, respectively, the volume
mented in an Aerojet Nuclear Company report. 2 fraction, density, and velocity of phase i and
By considering in detail the simple special case of where p is the pressure; all these quantities are
separated flow between parallel plates, we were functions of the distance x and the time t. Equa-
able to isolate and clarify the essential features of tions (1), (2), and (3) are referred to as the basic
the problem and thereby to resolve the issue to equation system.
our satisfaction. To forestall any misinterpreta- Equations (1), (2), and (3) are now widely
tion, we caution that our analysis is not intended recognized to possess complex characteristics. 1
to be quantitatively relevant to two-phase flow (Our use of the term "characteristics" is in
problems of practical interest. Its purpose, rath- accord with standard mathematical terminology 4 ' 5
er, is to lead the reader to a qualitative insight and is explained in the Appendix.) Consequently,
into the relations between characteristics, stabil- the solutions are unstable at very short wave-
ity, and short-wavelength phenomena. The un- lengths, and the equations constitute an improp-
derstanding thus gained is then immediately erly posed in it i a 1-v a 1 u e problem. (Further
applicable to more complicated two-phase flow discussion is given in the Appendix.) However,
situations. opinions differ as to the correct interpretation of
The complex characteristics problem can be these facts and the proper course of action. 1
illustrated by considering the case of separated Some workers have intimated that complex
two-phase flow between two parallel plates in zero characteristics are necessary to the description
gravity (Fig. 1). For simplicity, the fluids are of physical instabilities. 1 We shall see that this
assumed to be inviscid and barotropic, with no belief is erroneous; it apparently stems from
mass exchange between them. The most straight- confusion about the relation between characteris-
forward way to describe this situation would be to tics and stability. This relation is discussed in
apply to each phase the well-established one- the Appendix, where characteristics and stability
dimensional equations for flow in a variable area are, in general, shown to be related only in the
duct. 3 This approach leads at once to the follow- limit of large frequency and short wavelength.
ing system of two-phase flow equations: One might instead adopt the related but less
extreme position that equations having complex
a
at
a
(o:;p;) +ax (o:;p;u;) = 0 , (i = 1, 2) characteristics may correctly describe physical
(1) instabilities. This view, when properly qualified,
a a 2 ap is correct, but the qualification is a rather
at (o:;p;ui) +ax (o:;p;u;) + o:; ax= o, (i = 1, 2) serious one: The wavelengths involved must be
(2) sufficiently long. Equations having complex char-
acteristics can indeed provide a correct descrip-
(3) tion of long-wavelength physical instabilities; Eqs.

Mr== a 2H
(1), (2), and (3) themselves are of this type.
However, in all such cases, the results will be-
come increasingly erroneous as the wavelength is
decreased. Eventually, at short enough wave-
lengths, equations having complex characteristics
will predict unphysical instabilities.
Probably the majority of investigators believe
·
H
_u, that in two-phase flow, as in other areas of
physics and engineering science, physical phe-
nomena should find their mathematical expression
in terms of properly posed problems. 5 ' 6 It then
follows that Eqs. (1), (2), and (3) must be incom-
plete or incorrect in some essential physical
respect, and one may expect that when their
Fig. I. Geometry of separated two-phase flow between physical deficiencies are corrected, their char-
two parallel plates.
acteristics will become real. We share this view
2 4
J. D. RAMSHAW and J. A. TRAPP, "Characteristics, P. R. GARABEDIAN, Partial Differential Equations,
Stability, and Short-Wavelength Phenomena in Two-Phase Flow John Wiley and Sons, Inc., New York (1964).
5
Equation Systems," ANCR-12 72, Aerojet Nuclear Company R. COURANT and D. HILBERT, Methods of Mathematical
(1976). Physics. Vol. II, Wiley-Interscience, Inc., New York (1962).
3 6
H. W. LIEPMANN and A. ROSHKO, Elements of Gas- J. HADAMARD, Lectures on Cauchy's Problem in Linear
dynamics, pp. 40 and 48, John Wiley and Sons, Inc., New York Partial Differential Equations, Yale University Press, New Haven,
(1957). Connecticut ( 1923).
TWO-PHASE FLOW EQUATIONS 95

but differ from some of its other adherents in the of the Helmholtz problem between two parallel
type of modifications to Eqs. (1), (2), and (3) we plates including surface tension can readily be
recommend. We find that the complex character- obtained by standard methods and used for com-
istics and their attendant difficulties can be re- parison with the results of the one-dimensional
moved simply by including known physical effects two-phase equations.
that, while for the most part negligible, become The remainder of the paper is organized as
important at short wavelengths (and/or at high follows. In Sec. II, the basic equation system is
frequencies). The inclusion of such effects, of generalized to allow for surface tension, and the
course, does not alter the behavior of the equation characteristics are then shown to become real. In
system at long wavelengths. This point is impor- Sec. III, a linear stability analysis of the equation
tant because, as will be seen, Eqs. (1), (2), and (3) system including surface tension is performed. If
as they stand predict correct long-wavelength the surface tension a is zero, then all wavelengths
behavior for the case of small perturbations about are found to be unstable and the problem is ill
a uniform steady-state solution. Other authors, posed, whereas if a > 0, short wavelengths are
disregarding this fact, have proposed modifica- stable and the problem is well posed. In Sec. IV,
tions to Eqs. (1), (2), and (3) that alter their the problem of separated two-phase flow is con-
behavior at both short and long wavelengths. The sidered as a classical Helmholtz instability oc-
most common examples of such modifications are curring in a bounded region. The results of an
(a) modifying the pressure gradient term in Eq. exact two-dimensional stability analysis, including
(2) to read a(a;p)jax (Refs. 1 and 7) and (b) ad- the effect of surface tension, are summarized and
dition of first-order differential terms, nominally compared with the results of Sec. III. The results
representing various interface forces, to the right of Sees. III and IV are in agreement for an
side of Eq. (2) (Ref. 8). arbitrary value of a(a ~ 0), provided that the
The particular short-wavelength physical effect wavelength is long compared to the plate spacing.
considered in this paper is that of surface tension. A brief discussion of numerical considerations is
This effect was chosen primarily for reasons of given in Sec. V. Section VI contains a few con-
mathematical simplicity. Other short-wavelength cluding remarks, including comments about the
effects, viscosity in particular, give rise to simi- expected behavior of surface tension as compared
lar behavior but are much more difficult to treat to viscosity in the nonlinear regime. The Appen-
rigorously in connection with the exact two- dix discusses the connection between character-
dimensional analysis of Sec. IV. istics and stability for a general system of
When surface tension is introduced into Eqs. first-order quasi-linear partial differential equa-
(1), (2), and (3), the characteristics are rendered tions in the independent variables x and t.
real, short-wavelength disturbances are stabi-
lized, and the initial-value problem becomes prop-
erly posed. Since the effect of surface tension is II. ADDITION OF SURFACE TENSION TO
THE BASIC EQUATION SYSTEM
confined to short wavelengths, the resulting equa-
tion system predicts that long wavelengths remain
unstable, and it allows such instabilities to be The effect of surface tension can easily be
calculated. This long-wavelength instability is incorporated into Eqs. (1), (2), and (3) by regard-
physical in nature; it is simply the familiar ing the two phases as having slightly different
classical Helmholtz instability, occurring in a pressures, the two pressures being coupled by the
bounded region. In fact, much of our motivation usual surface tension interface condition. Thus,
and development exploits this correspondence. a a
The conventional Helmholtz problem is known to at (a;p;) +ax (a;p;u;) =0 ' (i = 1, 2)
be ill posed in the absence of surface tension and
9
to be well posed when surface tension is included, (4)
which in the one-dimensional two-phase descrip-
tion corresponds to complex and real character-
a( ) a z
at a;p;U; +ax (a;p;u;) + a;
aP; = 0
h , (i = 1, 2)
istics, respectively. In addition, the exact solution
(5)

7
S. I. PAl, Z Flugwiss .. 19,353 (1971). a1 + az = 1 (6)
8
1. BOURE eta!., "Choking Flows and Propagation of Small
Disturbances," Paper Fl, European Two-Phase Flow Group a
Mtg., Brussels (1973 ).
P1 - /J2 = R , (7)
9
R. D. RICHTMYER and K. W. MORTON, Difference Meth-
ods for Initial- Value Problems. 2nd ed., pp. 351-360, Wiley- where Pi is the pressure of phase i, a is the
lnterscience, Inc., New York ( 1967). surface tension, R is the radius of curvature of
96 RAMSHAW and TRAPP

the interface, given by a1u1


2 2 312 2 0 GtPt 0 PtUl 0
-1 = - H() al
-2 -[ 1 +
r.2 (aal) ]-
n- (8) c1
R ox ox a2u2
0 0 a2P2 -p2U2 0
and H is the plate spacing. The assumption is c~
made that H 2 (ilat/ilx) 2 « 1, since if such is not B= Gt 0 GtPtUt 0 0 0
the case, the one-dimensional variable-area -duct
equations on which Eqs. (1), (2), and (3) were 0 a2 0 G2P2U2 0 0
based become unrealistic anyway. By combining 0 0 0 0 0 aH
Eqs. (7) and (8) under this assumption, the follow-
ing is obtained: 0 0 0 0 1 0

il al
2 (16)
Pt - P2 = -aH ox2 (9) 112
in which c; denotes the sound speed (dp)dp;) of
Equations (4), (5), (6), and (9) are referred to as phase i, to be evaluated from Eq. (12), and c is a
the sigma equation system. "source" column vector,
To determine the characteristics of the sigma c = (O,O,O,O,Pt - P2, -{3f (17)
equation system, the system must be converted to
first-order form. This conversion can be done by The characteristics 1-l are defined by the equa-
means of the substitution {3 = oa1/ox, so that tion
Eq. (9) is replaced by det(A- ~J.B) = 0 (18)
af3 Equation (18) has six roots IJ., which are readily
Pt - P2 + aH ax= 0 (10)
obtained by virtue of the fact that the equation can
and be analytically factored. The roots are zero
aal (twice), (ul ± c1r\ and (u2 ± c2) - 1. The preceding
a--x-{3=0 (11)
analysis applies, of course, only if a 0. Thus, *
The two phases are assumed to be barotropic, so
the sigma equation system, with a 0, has real *
characteristics.
p; = /; (p;) , (i = 1, 2) (12) The basic equation system (a 0) requires a
separate characteristic analysis. This analysis
where the functions /; may be considered known. can be carried out analytically only for the incom-
Algebraic relations (6) and (12) can be used to pressible case (p 1 = constant, P2 = constant); the
eliminate a2, p1, and P2 as dependent variables. compressible case must be solved numerically.
When these variables have been eliminated, the In both cases, the characteristics are complex. 1
equation system can be written in matrix form as Thus, the introduction of surface tension into the
follows: basic equation system has the effect of changing
ov ov the characteristics from complex to real.
A·-+B·-+c=O (13)
at ax
Ill. LINEAR STABILITY ANALYSIS OF THE
where v is a column vector of dependent var-
SIGMA EQUATION SYSTEM
iables,
tJ = (pt,h,Ut,U2,at,{3)T (14) A linear stability analysis of the sigma equa-
tion system can be performed analytically for the
the superscript T denoting the matrix transpose. case of incompressible flow (pt = constant, P2 =
Here, A and B are matrices: constant) and an unperturbed solution having space
and time derivatives of zero (that is, a uniform
a1
2 0 0 0 Pt 0 steady state). For incompressible flow the sigma
C1 equation system becomes
a2
0 2
0 0 -p2 0 oa; oa; ou;
C2 - + u; - + a; - = 0 (i=1,2) (19)
A= at ax ax
0 0 G1P1 0 0 0
au; au; ap;
0 0 0 G2P2 0 0 0
at+u;--a;+--a;=' (i=1,2) (20)

0 0 0 0 0 0 a1 + a2 = 1 (21)
0 0 0 0 2
0 0 A A 0 al
PtPt - P2P2 = -aH - - (22)
(15) ax 2
TWO-PHASE FLOW EQUATIONS 97

where P; = p;/p;. To perform the stability anal- ical wave number, kc, is defined by the equality
ysis, each dependent variable v is replaced by sign in Eq. (30):
v + v' exp[i(kx - wt)], and the resulting equations p 1p 2 ]1/2
are linearized with respect to the primed quanti- kc = lu1 - u2J [ (31)
aH(o:2P1 + o:1p2)
ties. The assumption of a uniform steady-state
unperturbed solution means that v is independent Wavelengths 1 on g e r than 27T /kc are unstable,
of x and t. The result of this replacement is whereas wavelengths shorter than or equal to
21T /kc are stable. As is verified in the next
( -iw + iku1)o:{ + iko:1ui = 0 (23) section, the instability at long wavelengths is
-(-iw + ikll2)o:{ + iko:2u~ = 0 (24) simply the well-known Helmholtz instability.
Inspection of Eq. (29) shows explicitly that the
( -iw + iku 1)u{ + ikPI = 0 (25) basic equation system and the sigma equation
( -iw + ikll2)~ + ikp~ = 0 (26) system exhibit virtually identical behavior when
k « kCl since in this case the first term in
(27) brackets becomes negligible compared to the
Equations (23) through (27) are a linear homo- second and can be omitted, just as though a were
geneous system in the variables PI, p~, u{, u~, and zero. Therefore, for wavelengths long compared
o:{. For a nontrivial solution to exist, the deter- to 21T/kc, the physical Helmholtz instability can be
minant of coefficients must vanish. This condition correctly calculated with either equation system.
leads to the equation Thus, an equation system having complex charac-
2 2 4
teristics can correctly predict long-wavelength
O:lP2 (ku2 - w ) + 0:2P1 (ku1 - w ) - o: 1o: 2aHk = 0 physical instabilities.
(28) At shorter wavelengths, however, the disper-
sion relation given by Eq. (29), for a given value
which determines the dispersion relations w(k). of a > 0, begins to differ appreciably from that for
The solutions of Eq. (28) are a = 0. Eventually, at wavelengths shorter than
w= ko:1o:2 {p1u1 + p2u2 21T/ko the two equation systems predict totally
0:2P1 + 0:1P2 0:1 0:2 different behavior, the sigma system predicting
stability and the basic system predicting in-
stability. Now the sigma equation system is
clearly physically the more realistic, because
surface tension is a real physical effect that
(29) should be included in a complete physical de-
The criterion for stability is Im(w) s 0 for all scription. We are, therefore, justified in the
roots w. In the present case, complex roots occur inference that the basic equation system exhibits
in conjugate pairs, so that w must be purely real incorrect and unphysical behavior at short wave-
for stability. Equation (29) shows that instability lengths, in particular, unphysical instability for
results if the quantity in brackets is negative, k > kc. This unphysical behavior, like the com-
whereas stability results if it is positive or zero. plex characteristics, results from the unwar-
The condition for stability is therefore ranted neg 1 e c t of surface tension (or other
2 2 short-wavelength effects, such as viscosity).
aHk (0:2P1 + 0:1P2) 2: P1P2(u1 - u2) (30) The growth factor corresponding to Eq. (29) is
Unlike the characteristic analysis of the pre-
ceding section, the stability analysis is valid for
G(k,t) =lexp[ -iw(k)t]l = exp[Im(w)t] (32)

an arbitrary value of a, including a = 0. As a When w is complex, the positive sign in Eq. (29) is
consequence, this analysis yields valuable infor- the sign that corresponds to growth. Setting a = 0
mation about both the sigma equation system and in Eq. (29), we find that
the basic equation system. klu 1 - u2l
The trivial case u 1 = u2 is disposed of at the lm (w) = (o:1p1o:2p2) 112 (33)
0:2P1 + 0:1P2
outset. In this case, stability is obtained for all
wavelengths and all values of a 2: 0. Henceforth, for the basic equation system. For the sigma
we assume that u 1 u2.* equation system, a > 0, and the corresponding
Setting a = 0 in Eq. (30), we see that all wave- result is
lengths are unstable in the basic equation system.

l
o if k 2: kc
For the sigma equation system, a > 0, and
Eq. (30) shows that very short wavelengths (large Im(w) = k[ a 1 a2 aH
(k~ - k2 )
]112 if k < kc
k) are stable, whereas very long wavelengths 0:2P 1 + 0:1P2
(small k) are unstable. For convenience, a crit- (34)
98 RAMSHAW and TRAPP

Now the essence of a well-posed problem is that, Eq. (28) with Eq. (35). Since x coth x ~ 1 for
for a given time t, G(k,t) be bounded with respect 0 < x « 1, Eq. (35) reduces to Eq. (28) in the limit
to the wavenumber k in the interval 0 :s k < oo kH « 1, that is, when the wavelength is long com-
(Ref. 9). This in turn requires that Im(w) be pared to the plate spacing. In this limit, there-
bounded with respect to k, which is true for fore, the sigma equation system exhibits correct
Eq. (34) but not for Eq. (33). Thus, the basic physical behavior, at least with regard to small
equation system is shown to represent an im- perturbations about a uniform steady-state solu-
properly posed problem, whereas the sigma equa- tion. Setting a= 0 in Eqs. (28) and (35), we also
tion system represents a well-posed problem. see that the basic equation system exhibits cor-
The fact that the basic equation system constitutes rect physical behavior in the hypothetical case
an improperly posed problem may indeed be when the surface tension is identically zero. In
regarded as due to a physical deficiency in the this latter case, the problem is well known to be
system: the failure to account for surface tension improperly posed, and the basic equation system
(or other short-wavelength effects). appropriately reflects this behavior. In the case
a > 0, the problem is well posed [as can explicitly
IV. EXACT TWO-DIMENSIONAL be verified by solving Eq. (35) for Im(w)], and the
STABILITY ANALYSIS sigma equation system reflects this behavior,
even though it becomes quantitatively incorrect at
The conventional equations of multidimensional
short wavelengths.
fluid dynamics are used in this section to analyze
The failure of the sigma equation system at
the behavior of separated two-phase flow between
wavelengths comparable to the plate spacing was
parallel plates, including surface tension. The
anticipated in the remark following Eq. (8). It
physical validity of the basic equations is well
occurs because at such wavelengths, the flow
established, so that the results of the analysis can
becomes essentially two-dimensional in character
confidently be regarded as correct and can, there-
and can no longer be described by a one-dimen-
fore, be used to check the predictions of the one-
sional equation system.
dimensional two-phase equations.
The fact that the sigma equation system ex-
As in the preceding section, the fluids are
hibits correct physical behavior when kH « 1
assumed to be incompressible and the motion of
places severe restrictions on the form of any
small perturbations about a uniform steady-state
further modifications to the system. In particular,
solution is examined. The necessary mathemat-
any additional "interface force" terms added to
ical development differs only slightly from that
10 the momentum equations must be such that they
given in Article 232 by Lamb and hence is not
yield a vanishing contribution, except possibly at
given in detail. Instead, only the modifications to
short wavelengths, in the problem of small per-
Lamb's development that are necessary to treat
turbations about a uniform steady-state solution.
the case of present interest are indicated.
However, a possible exception to this restric-
The pressure interface condition is modified by
2 tion should be mentioned. Such a restriction would
adding the term aa TJ/ax 2 to the left side of Lamb's
clearly not apply if the objective were to obtain an
Eq. (4) to allow for the presence of surface
equation system that does not attempt to describe
tension. The second modification is to allow for
small-amplitude disturbances correctly, but seeks
the finite depth of the two fluids by replacing the
instead to describe the average behavior of a
factors exp(ky) and exp(-ky) in Lamb's Eq. (5) by
"fully developed" two-phase flow in which ap-
the factors cosh[k(y + a1H)] and cosh[k(y - a2H)],
parently random motions reminiscent of turbu-
respectively. The third modification is to set
lence appear to be superimposed on a slowly
g = 0. The equation that then results in place of
varying deterministic mean motion. This objec-
Lamb's Eq. (9) is, in the notation of the present
tive is indeed worthy but also uncommonly diffi-
paper,
2
cult, probably as difficult as that of constructing a
a 1p2[a2kH coth(a2kH)](ku 2 - w) complete theory of turbulence directly from the
2 Navier-Stokes equations. In the present paper, we
+ a2p1[ a1kH coth(alkH)](ku 1 - w)
4
have implicitly restricted attention to the problem
- a1a2aHk = 0 (35) of obtaining a satisfactory system of equations to
Equation (3 5) determines the exact dispersion describe the instantaneous motions in separated
relations w (k) for the physical problem under two-phase flow.
consideration. The validity of the sigma equation
system can now be investigated by comparing V. NUMERICAL CONSIDERATIONS

10
H. LAMB, Hydrodynamics, 6th ed., Dover Press, Inc., Suppose that it is desired to use the sigma
New York (1945). equation system to calculate a long-wavelength
TWO-PHASE FLOW EQUATIONS 99

instability by a finite difference numerical pro- Thus, even if we start out using the sigma eq ua-
cedure. If the wavelength of interest is long tion system, the true surface tension can be
compared to 2rr /ko whether the sigma equation neglected in comparison with the artificial one.
system or the basic equation system is used is The preceding considerations suggest that the
immaterial, because both exhibit the same be- numerical calculations be performed using the
havior for k « kc. However, this statement sigma equation system with a replaced by a*,
implicitly assumes that care is taken not to regardless of whether the sigma equation system
introduce perturbations at other wavelengths hav- or the basic equation system was originally ad-
ing growth rates larger than that of the wavelength vocated. From this point of view, which system
9
of interest. As Richtmyer and Morton point out, was originally preferred does not seem to matter.
the finite difference representation itself effec- It does matter, however, from a more fundamental
tively introduces perturbations with wavelengths point of view. If use of the basic equation system
on the order of the mesh spacing ~x. Thus, having complex characteristics had been advo-
instabilities at wavelengths of order ~x cannot be cated, we would effectively be switching to a
tolerated in a finite difference calculation. This different equation system having real characteris-
operational numerical constraint must be ob- tics at the point when a* was introduced, regard-
served regardless of the equation system being less of the fact that it is introduced for a purely
used. Since all wavelengths, including those of numerical reason.
order ~x, are unstable for the basic equation Of course, the use of a* is not a unique way of
system, we would not expect to be able to suc- achieving the desired stabilization of wavelengths
cessfully calculate even a long-wavelength in- of order ~x. Other artificial terms based on
stability using this system without introducing physical effects (for example, artificial viscosity
some artificial stabilizing effect at wavelengths of or mass diffusion) lead to similar results. In
order ~x. addition, certain methods of differencing (in par-
In the sigma equation system, wavelengths of ticular, "donor cell" or "upwind" differencing)
order ~x may or may not be unstable, depending themselves are well known for introducing an
on the magnitudes of ~x and a and the values of the artificial stabilizing effect, u, 12 even though no
flow variables [Eq. (30)]. In practice, for typical additional stabilizing term has explicitly been put
values of ~x and the flow variables and typical in. If such a differencing method is used, one may
physical surface tensions, wavelengths of order indeed be able to perform calculations in which
~x are still unstable; the stability condition, the governing equations are apparently those of
Eq. (30), is not satisfied until wavelengths con- the basic equation system, which have complex
siderably shorter than ~x are reached. The ideal characteristics. However, an analysis of the
solution to this problem would be to reduce ~x differencing truncation errors would reveal the
until it becomes comparable to the (very small) effective presence of a stabilizing term that, if it
value 2rr /kc. Unfortunately, reducing ~x to this were included in the characteristic and stability
extent will not ordinarily be economically feasi- analyses, would render the characteristics real
ble. Thus, even when the sigma equation system and stabilize short-wavelength disturbances.
is used, introduction of an artificial stabilizing Finally, it should be noted that in particular
effect for wavelengths of order ~x will ordinarily special cases, calculations may be performed
be necessary. using the basic equation system with no artificial
The most straightforward way of introducing an stabilizing effects of any kind, provided the entire
artificial stabilizing effect would be to define an calculation is finished before appreciable error
artificial surface tension a*, as suggested by growth has a chance to occur. This approach is
Teller (quoted by Richtmyer and Morton 9 ). The clearly unsuitable for general use, because it will
magnitude of a* would be chosen in accordance only work for calculations over ''short'' intervals
with Eq. (30) to be just sufficient to stabilize in problem time, where the precise meaning of
wavelengths of order M. Setting k '== rr /~x (the '' short'' is not known a priori.
smallest wavelength that can be represented in the
mesh is 2~x), we obtain the following formula for VI. CONCLUDING REMARKS
a* from Eq. (30):
In seeking to clarify the interrelationships of
(36) characteristics, stability, and short-wavelength
phenomena in two-phase flow, we have limited our
where C is a dimensionless constant "safety 11
F. H. HARLOW and A. A. AMSDEN, J. Comput. Phys., 8,
factor'' that should probably be set slightly larger 197 (1971 ).
than unity. As implied, a* is typically much 12
K. W. MORTON, Proc. Roy. Soc. London A, 323, 237
larger than physical values of the surface tension. (1971 ).
100 RAMSHAW and TRAPP

consideration to a detailed analysis of the effect ical than the basic system, appears likely to
of surface tension. This particular short-wave- exhibit unphysical behavior in the nonlinear re-
length effect was chosen primarily for conve- gime.
nience in the theoretical development. Physically, In the case of viscosity, however, the short
to include surface tension without also including wavelengths are actually damped, so that once
viscosity is probably unreasonable, although it kinetic energy is transferred into them, it is
might be reasonable to include viscosity and dissipated as heat and effectively removed from
neglect surface tension. However, if an attempt the system. Thus, a sink exists in the spectrum
had been made to analyze the effect of viscosity, at short wavelengths whereby kinetic energy can
we would not have had an exact two-dimensional leave the system. One can then picture a situa-
solution with which to compare. Although the tion, reminiscent of turbulence, in which energy is
analysis of Sees. II and III would not have been continually being extracted from the mean motion,
hindered and would have led to very similar transferred in a cascade process to shorter and
results, the exact two-dimensional results of shorter wavelengths, and ultimately dissipated in
Sec. IV could not have been obtained because of the very short wavelengths to heat. No wavelength
the inapplicability of potential flow theory to in the spectrum will then grow indefinitely; its
viscous fluids. (In addition, the presence of amplitude will reach an equilibrium in which the
viscosity would cause the velocity to vary much rate at which it acquires energy directly, as a
more over a cross-sectional plane perpendicular result of its instability, is equal to the net rate at
to the flow direction, which would make the which it transfers energy to other wavelengths.
variable-area equations on which our basic equa- Although speculative in nature, the preceding
tions were based somewhat less satisfactory.) remarks suggest that the primary usefulness of
Nevertheless, for the purpose of obtaining a the sigma equation system is more likely to be
two-phase flow equation system suitable for actual theoretical, as in the present paper, than prac-
numerical calculations, there is reason to believe tical. One suspects a priori that the equation
that viscosity is a more appropriate effect to system obtained by incorporating viscosity into
include than is surface tension because viscosity the basic equation system would give a better
is a dissipative effect whereas surface tension is representation of physical reality in the non-
not. 13 Although surface tension exerts a stabiliz- linear regime and hence would be more suitable
ing effect, it exerts no damping, and intuitively for practical applications. The preceding qualita-
damping seems necessary for the equation system tive remarks also suggest that methods and con-
to be stable in some nonlinear sense. Imagine cepts from turbulence theory may prove useful in
what happens when a small perturbation of a given formulating a theory of the average behavior of
wavelength is introduced. The perturbation grows fully developed two-phase flow, as discussed at
in accordance with the linear stability theory until the end of Sec. IV.
its amplitude becomes so large that nonlinear Finally, we remark that much of the concern
terms become important. One thinks of these over complex characteristics in two-phase flow
nonlinear terms as having the effect of transfer- appears to have been due to the fact that charac-
ring energy from the original wavelength into teristics and stability, although not unrelated, as
other wavelengths. Eventually, the entire wave- shown in the Appendix, are in reality two separate
length spectrum becomes excited, and energy concepts that should not be confused. Indeed, in a
finds its way into the small stable wavelengths. In practical sense, characteristics are largely ir-
the case of surface tension, these wavelengths are relevant; it is stability about which one should be
stable but they are not damped. Their amplitudes most concerned when analyzing the behavior of
will therefore grow, not because of instability but equation systems. As discussed at the end of
because energy is being transferred into them Sec. III, it is stability rather than characteristics
from other wavelengths. When their amplitudes that is directly related to the question of well
become large enough, the nonlinear terms will posedness. Theorems relating characteristics to
again come into play and will begin to transfer well posedness typically restrict attention to the
energy back out of the stable short wavelengths case of constant coefficients (for example, see
into the unstable longer wavelengths. Thus, in the Lax 14 ), in which stability and characteristics are
nonlinear regime, all wavelengths would be ex- trivially equivalent, as shown in the Appendix. If
pected to continue growing indefinitely, which attention had been focused at the outset on stabil-
is physically unacceptable because real physical ity rather than on characteristics in two-phase
systems are dissipative. Thus, the sigma equa- flow, it might have been realized more quickly
tion system, although unquestionably more phys- that the single most essential modification to the
13 14
J. M. DELHA YE,!nt. J. Multiphase Flow, 1, 395 (1974). P. D. LAX, Comm. Pure Appl. Math.,ll, 175 (1958).
TWO-PHASE FLOW EQUATIONS 101

basic equation system is the introduction of such ov = OVo exp[i(kx - wt)] (A.4)
short-wave-length effects as surface tension or is assumed. Equation (A.3) then becomes
viscosity. In retrospect, this modification is
clearly suggested by the well-known fact that such
effects are sufficient to stabilize short wave-
-iwA(v) ·OVo + ikB(v) ·ovo + (ovo ·~:) ·aa~
lengths and produce properly posed problems in
other contexts. 9 + (OVo. av
aB) . av ac
ax + 01Jo. av = 0 (A.5)

Equation (A.5) is a homogeneous linear system in


APPENDIX
the components of ov0 • For a nontrivial solution
RELATIONSHIP BETWEEN CHARACTERISTICS to exist, the determinant of the coefficient matrix
AND STABILITY must vanish. The dispersion relations w (k) are,
therefore, determined by
In this Appendix, the relation between charac- det(-iwA + ikB +D) = 0 (A.6)
teristics and stability for a system of n first-
order quasi -linear partial differential equations where
of the form
0 =(a A. av)r + (~- av)r + (ac)r
av at av ax av
(A.7)
A(v) · ~~ + B(v) · ~: + c(v) =0 (A.l)
the superscript T denoting the matrix transpose.
is clarified. In Eq. (A.l), v is an n-dimensional If w(k) is assumed nonzero, Eq. (A.6) can be
column vector of dependent variables, A and B are rewritten as
n x n square matrices, and c is an n-dimensional
source vector. The elements of A, B, and c are det(A - ~ B +to) = o (A.8)
unspecified functions of the components of v.
The characteristic roots, J.l, of system (A.1) Because of the presence of i in Eq. (A.8), the
(which for brevity we refer to as characteristics complex roots w(k) corresponding to a given real
throughout the paper) are defined by the equation value of k need not occur in conjugate pairs. In-
spection of Eq. (A.4) shows that the condition for
det(A - J.LB) =0 (A.2) stability is that Im{w) be :s:Q for all roots w.
Comparison of Eqs. (A.2) and (A.8) shows that
(The source vector c is by definition not included the characteristics and stability questions differ
in the characteristic analysis.) because of the presence of D in Eq. (A.8). If D
The characteristics J.l defined by Eq. (A.2) are were zero, the dispersion relations could be ob-
the reciprocals of those customarily used. 4 ' 5 The tained from the characteristics simply by equating
reason for this choice is that if one of the (k/w) to J.l. Equation (A.7) shows that D equals
matrices is singular, it will usually be A rather zero trivially when A, B, and c are constants
than B. If A is singular and B is non singular, independent of v. When A or B depend on v, a
Eq. (A.2) still possesses n roots. Some of these nonzero D will result when the unperturbed solu-
roots may be zero. If so, the corresponding tion is not constant in time or space. A nonzero
reciprocal roots ,\ = 1/J.l are infinite and would not D will also result if the source vector c is non-
be obtained by solving the equation zero and depends on v.
det(XA - B) = 0 Thus, if the equation system has real charac-
teristics and no source terms, then a uniform
Next, the local linear stability behavior of
steady-state solution is necessarily stable with
Eq. (A.l) is examined. To do so, v is replaced by
respect to small disturbances. This stability is
v + ov, and the result is linearized with respect to
neutral; small disturbances neither grow nor
ov, resulting in
decay with time. However, when D *
0, real
A(v) · ~ (ov) + B{v) · i_ (ov) + (ov ·aavA)· avat characteristics are not sufficient to ensure sta-
at ax bility; this fact explains how systems with real
characteristics can exhibit physical instabilities
+ (ov. aavB) .ax
av + ov. ac = 0
av
(A.3) and clearly indicates that complex characteristics
are not nee essary to describe such instabilities.
Equation (A.3) describes small perturbations ov Although stability and characteristics are not
about an unperturbed solution with instantaneous related in general, they become essentially
values of v, av/at, and av/ax. These instantaneous equivalent (with an exception to be noted) in the
values are treated as constants, and a solution of limit of very high frequencies (large w). This can
the form be seen intuitively by inspection of Eq. (A.8),
102 RAMSHA W and TRAPP

where D is seen to be multiplied by 1/w. This But k/lwl is assumed to approach a finite limit,
intuitive expectation can be made rigorous by which may be positive or zero, as k -+ <Xl, Thus,
means of a theorem due to Ostrowski (see Ref. lw F/k -.+G() as k -. G().Since the complex charac-
15). We first assume that k/lw(k)i approaches a teristics occur in conjugate pairs, il1 may be
finite limit as k -. as is invariably the case.
G(), considered negative. Equation (A.10) then implies
Ostrowski's theorem then implies that that Im(w) -+ +<Xl as k -+ <Xl, Thus, an instability re-
sults at very short wavelengths; moreover, the
liml!l
k~oo
-~~
W
= 0 (A.9) growth rate is unbounded as k -+ so that the
G(),

problem is improperly posed.


where !l is the characteristic root corresponding The preceding analysis applies only when the
to the stability root k/w. Equation (A.9) shows derivatives av /at and av /ax of the unperturbed
that the characteristic and stability roots become solution are finite; otherwise, D/w in Eq. (A.8)
arbitrarily close together in the limit of large k. cannot be made arbitrarily small by increasing k.
First, the case for which !l is real is considered. Unbounded derivatives appear to be the reason the
Equation (A.9) then implies that as k -. the G(),
usual equations of inviscid incompressible single-
imaginary part of k/w approaches zero, whereas phase fluid dynamics, whose characteristics are
the real part approaches ll· Thus, provided !l 0, * real, can predict instability as k -. in the clas-
G()
the imaginary part of w becomes small compared sical Helmholtz problem with zero surface ten-
to the real part as k -. so that if unstable
G(),
sion: The density and velocity are discontinuous
growth occurs at all, it occurs at worst on a time at the interface, so that the unperturbed solution
scale much slower (in the limit, infinitely slower) has infinite space derivatives.
than the period of the harmonic oscillations. In The development presented in this Appendix
this sense, the motion can be regarded as almost has been based on a type of stability analysis that
stable; thus, if all the characteristic roots !l are we refer to as local linear stability analysis. The
real and nonzero, then the motion is either stable term "local" refers to the fact that the coeffi-
or almost stable at very short wavelengths. cients in the linearized Eq. (A.3) were treated as
However, if !l = 0 for one or more roots, this constants to admit solutions of the form given in
conclusion does not follow, and no information is Eq. (A.4). In a rigorous linear stability analysis,
available about the growth rate of any residual these coefficients would be treated as given func-
instabilities that may remain as k -. regardless G(),
tions of x and t determined by the unperturbed
of the fact that all the !l are real. The case !l = 0 solution v(x ,t). The coefficients are then variable,
is thus the exception to the relation between and ov must be solved for by numerical proce-
characteristics and stability that was referred to dures. The local linear stability analysis is less
previously. As seen in Sec. II, the sigma equation rigorous but very useful, as we have seen, in
system is of this type. One might be tempted to obtaining a substantially correct understanding
conjecture, on the basis of the results of Sec. III, with a minimum of computational labor. This
that stability at very short wavelengths results method of analysis appears first to have been
when all the !l are real even if some of them are used by von Neumann and Richtmyer. 16 The
zero. Such a conjecture cannot be correct without stability analyses performed in Sees. III and IV,
qualification, however; a counter example is pro- however, were rigorous linear stability analyses,
vided by considering a to be negative. If a < 0, which could be analytically performed because the
the characteristic analysis of Sec. II is unaltered, coefficients were constant for the particular un-
but an unbounded instability is introduced as k -. G()
perturbed solution under consideration.
(Eq. (29)].
In the case for which !l is complex (iJ. = ilR +
ill, ill* 0), Eq. (A.9) implies that the imaginary
ACKNOWLEDGMENT
part of k/w approaches ill as k -. that is, G(),

This work was performed under contract with the U.S. En-
-Im(w)k] (A. 10)
i~~ [ lw 12 = ill
ergy Research and Development Administration.

lSJ. H. WILKINSON, The Algebraic Eigenvalue Problem,


16J. VON NEUMANN and R. D. RICHTMYER, J. Appl.

p. 63, Oxford University Press (1965 ). Phys., 21,232 (1950).

S-ar putea să vă placă și