UPWIND METHODS FOR FLOWS WITH NON-EQUILIBRIUM CHEMISTRY AND THERMODYNAMICS B. Grossman and P.

Cinnella Department of Aerospace and Ocean Engineering Virginia Polytechnic Institute and State University Blacksburg, Virginia 24061 USA ABSTRACT The numerical computation of gas flows with non-equilibrium thermodynamics and chemistry is considered. Several thermodynamic models are discussed, including an equilibrium model, a general non-equilibrium model and a simplified model based upon vibrational relaxation. The effects of the various models on the state equation and the homogeneity property of the Euler equations is described. Flux-splitting procedures are developed for the fully-coupled inviscid equations involving fluid dynamics, chemical production and internal energy relaxation processes. New forms of flux-vector split and flux-difference split algorithms valid for non-equilibrium flow, are embodied in a fully coupled, implicit, large-block structure. Several numerical examples in one space dimension are presented, including high-temperature nozzle flows with hydrogen-air chemistry. INTRODUCTION The numerical prediction of combustion in a high-speed flow environment remains a challenging computational problem. Flow fields such as those to be encountered in the SCRAM JET engine of future hypersonic vehicles may produce significant departures from equilibrium chemistry and thermodynamics. The design of these vehicles will require very accurate solutions to the entire three-dimensional non-equilibrium flow field. Accurate solutions of shock-wave dominated flows have been obtained using the class of algorithms referred to as upwind or flux split, (e.g., see the survey papers of Harten, Lax and Van Leer [1] and Roe [2]). These methods which include flux-vector splitting and flux-difference splitting, utilize difference procedures which are biased in the direction determined by the signs of the characteristic speeds. These approaches were originally developed for perfect gases and rely on the simplicity of the equation of state to develop algebraic relationships for the split fluxes and their associated Jacobians. In this paper, we seek to develop upwind methods for the computation of nonequilibrium flows. We will concentrate only on the convective algorithms and will not consider the other important features of this flow regime, such as viscous effects and radiative transfer. The algorithms will be developed for the one-dimensional Euler equations. Extensions to multidimensional flows may be accomplished through finitevolume methods using procedures such as those developed by Walters and Thomas [3]. We will also limit our discussion to the frequently-used flux-vector split algorithms of Steger and Warming [4] and Van Leer [5] and the flux-difference split algorithm of Roe

[6].

324

The computation of flows with non-equilibrium chemistry and thermodynamics is considerably more complex than the perfect gas situation. The number of dependent variables and partial differential equations increases, with production equations necessary for each species mass density (or mass fraction), and, for the case of non-equilibrium internal energy, a production equation for the portion of energy not in equilibrium, e.g., vibrational relaxation. Difficulties appear, due to the often disparate time scales associated with the fluid motion and the non-equilibrium chemistry and thermodynamics. One method to avoid this stiffness problem and the approach taken here, is to utilize a fully-coupled procedure, where the equations governing the fluid dynamics and non-equilibrium chemistry and thermodynamics are solved simultaneously with implicit numerical methods. This results in a very complex, large block structure for the solution algorithm which, however, fully accounts for all the non-equilibrium effects. In previous efforts, [7], [8], the present authors developed flux-split algorithms for a specific non-equilibrium energy model, which included a relaxing vibrational energy and negligible electronic excitation. This model, which we will call the simplified vibrational energy model, led to a simplified equation of state which allowed a non-iterative evaluation of the pressure and temperature from the conserved variables. We now extend this approach to more general energy models, which will include the previous simplified vibrational energy model, along with an equilibrium energy model as special cases. An important feature of this approach is that by performing the calculations with different energy models, we can assess the importance of non-equilibrium internal energy on high-temperature flows. In the next section, we discuss our chemical and thermodynamic models along with the corresponding equation of state. We derive the flux Jacobian for the Euler equations, and develop the corresponding eigenvalues and eigenvectors. We then develop nonequilibrium flow versions of Steger-Warming and Van Leer flux-vector split schemes and derive an approximate Riemann solver, which results in a flux-difference split scheme of the Roe type. We will present several elementary numerical examples of non-equilibrium flow computations in order to illustrate the accuracy and wave-capturing properties of our methodology. The test cases involve the steady combustion of a hydrogen-air mixture in a supersonic diffuser, including an embedded shock wave. The effects of nonequilibrium thermodynamics will be discussed for these cases. Also the beneficial effect of utilizing implicit computational procedures on the stiff set of governing equations will be discussed. Future research will be required to fully evaluate the accuracy of the algorithms and the efficiency of the fully-coupled approach in multi-dimensional flows. FORMULATION

Chemical and Thermodynamic Models

At high temperatures, chemical reactions will occur in gas flows resulting in changes in the amount of mass of each chemical species. We consider a system containing N species and assume that our chemical kinetics is limited to homogeneous reactions, neglecting any solid or liquid surface reactions and excluding all photochemical reactions. We assume that all the chemical reactions are known along with their corresponding

325

forward and backward reaction rates. For non-equilibrium chemistry, the rate of production of species i due to chemical reaction may be written as
dp~ = f ( P l , P~, . . . , PN, T ) , ~bi =-- dt

(1)

where pi is the species mass density. At high temperatures, imperfect gas effects are due to chemical changes in the a m o u n t of mass of each species and to the activation of internal energy modes which behave non-linearly with temperature. As long as the pressure is sufficiently low, away from the gas triple point, then it has been found t h a t each species of the gas mixture will behave as a thermally perfect gas. T h a t is, ei = ei(Ti) and pi = p i R i T i , where Ti is the translational temperature, ei is the internal energy per unit mass, Pi, the partial pressure, Pi the density and Ri the gas constant for species i. T h e gas constant for species i may be expressed as R i --- ]~/J~i where )~ is the universal gas constant (per mole) and ~ i is the mass per mole of species i. T h e t h e r m o d y n a m i c model that will be considered here is to assume that a portion of the internal energy of each species is in t h e r m o d y n a m i c equilibrium, and that the remaining portion is in a non-equilibrium state. T h e non-equilibribrium part of the energy is assumed to be modeled by appropriate production rates, c.f., Herzberg [9], and Jaffe [10]. We will also make the simplifying assumption that the mass of each species is approximately the same, whereby the translational t e m p e r a t u r e of all species will assumed to be the same, or Ti -- T . Thus, the analysis will not be applicable to flows with free electrons. We begin by defining the energy per unit mass of species i as
e, = +

where ei is the equilibrium portion of the energy. It is convenient to express ~i in terms of a specific heat as
ei -~ ~,, (T) d T + hy,,

(3)

T,-,I

where the specific heat at constant volume, ev, = d ~ i / d T , and hy~ is the heat of formation of species i. We consider a gas mixture composed of N species, with the first M species assumed to contain a non-equilibrium portion of their internal energy. T h e internal energy per unit mass of the mixture may then be written as
N e= --ei=~+ i=1 p i=1 M -., (4)

where we have introduced the definition of reduced specific internal energy

N
=

i = 1 fl

326

It is convenient to define the reduced specific heats of the mixture as

N N

cv ~ E
i:1

Pi r p e~ , ,

cP=--E
cP =-- E
i=l
p

Pl ~ --~epl

(6a, b)

We m a y also write the mixture gas constant as

N

(7)
i=1

and it is convenient to define

Equation of State
For the conditions specified here, each individual species behaves as a thermally perfect gas. Then, the pressure, according to Dalton's law, is
N

P= E
i=1

p,R,T = pRT,

(9)

where the mass density of the mixture is

N

p :
i=1

p,

(10)

T h e state relationship of the pressure to the specific internal energy occurs implicitly through the temperature. For a given chemical composition, internal energy and non-equilibrium energy, the t e m p e r a t u r e must be evaluated from

e= E
i:l f

~v,(T) dT + hf, +
i:l

Pl
pgni"

(11)

Iterative procedures m a y be used to solve for T. Once T is found, the pressure is determined from (9). T h e appropriate speed of sound for the models prescribed here is the frozen speed of sound, defined as _:
S, pi l P, e . i
=

+
Pi l P, c.. i

(G)
P 8, Pi l P, e,,.i

(12)

where s is the entropy per unit mass. Utilizing the First Law of T h e r m o d y n a m i c s , along with (9) and (11), we find the interesting result that a 2 = @/~T = ~ ( P ) . (13)

The above result is not approximate, but corresponds to the frozen speed of sound for this chemically reacting, non-equilibrium flow.

327

Governing Equations T h e governing equations for an inviscid, non-heat-conducting one-dimensional flow with non-equilibrium chemistry and non-equilibrium internal energy may be written in vector conservation form as aU OF
at + = w,

(14)

where U is the vector of conserved variables, F is the flux vector and W is the vector of production rates given by
Pl Pl u p2u
"d)1

P2

tb2

U =

PN

pu
plenl

F =

pu2+p
pl enl lt

pNU

tbN 0

(15a, b, c)

pMenM peo

pMenMu

pMen~-enMWM

puho

Equation (14) represents N + M + 2 conservation equations, with the first N corresponding to species continuity, followed by m o m e n t u m conservation, M non-equilibrium energy conservation equations and the total energy conservation equation. In the above equations, u is the velocity in the x direction, p is the global mass density, p is the pressure, e0 is the total energy per unit mass, h0 is the stagnation enthalpy per unit mass, pi is the mass density of species i and en~ is the non-equilibrium portion of the internal energy per unit mass of species i. T h e chemical production terms are tbi are assumed to be a given function of t e m p e r a t u r e and species densities, and the non-equilibrium energy p r o d u c t i o n terms are denoted ~,~ and are assumed to be a given function of pressure, t e m p e r a t u r e and non-equilibrium energy T h e system is completed by the density definition, (10) and the equation of state, defined implicitly through (9) and (11), giving the pressure in terms of conserved variables. T h e stagnation enthalpy, is defined by h0 = eo + pip. Flux Jaeobian Matrix F r o m the above description, the flux vector F may be considered a function of the vector of conserved variables U, or F = F(U). A quantity which plays a major role in flux-vector split algorithms and in any implicit formulation is the Jacobian matrix

328

aF/aU.

We have evaluated A, in general, as

-pA=
(pl-p) p2u u plu (p2--p)u : : pNu al pNU

. . . . . . . . . . . . : : . . . . . .
aM

piu p2u : : : : (pN--p)u :

--Pl --P2 : : --[N aN+l --plen~ :

u --pMenM

0 0 : : 0
p~

... .'. : : '''

0 0 : : 0
p~

0 0 : : 0
--p~

Pl eni u

plenlu :
pMenMu

aN plen~u :
pMenM

--pu :
0

:
. . . .

0 :
pU

0 :
0

:
flMenM u

bl

"" p(u 2 - Op/Opi),

bM

"'" bi = p u ( h o

bN - Op/Opi),

bN+l

pu~

...

pu~

--~pu

(16)
where ai = for i = 1 , . . . , N along with
aN+l =

-p,~(3 - ~), bN+l = p(u2~ - ho) and ~ - (~-1). It can be shown that for this set of equations, the Jacobian matrix A has the homogeneity p r o p e r t y /~ A U. (17)
=

Of course, this p r o p e r t y was noted for thermally perfect gases in [4]. The N + M + 2 eigenvalues of the Jacobian m a t r i x A are found to be
u
Ai = u + a u-a

i=1,

....

N + M + I

i = N + M

(18)

i = N + M + 2

The corresponding set of N + M + 2 linearly independent right eigenvectors have been evaluated as:

o
Pi/P

o
0

pllp
P2/P "

Ei =

0 piu/p 0

Ej+N

Ek+N+M

0
"
pj e,~j / p

PN/P u + a plenl/P

pMer~M/p

(u ~ - ~i)pi/p,
i=l,...,N

pje,~/p J
j=I,...,M k=1,2.

ho ua

(19a, b, c)

329
In (19a), we have introduced the definition from (9)-(11) of ~i~ ~ ------ 1 Opi ~- 1

Op

RiT

~i(T)+--. 2

u2

(20)

Flux- Vector Splitting

Utilizing the homogeneity property of the governing equations as determined in (17) we m a y develop a flux-vector splitting along the lines of Steger and Warming, [4], for perfect gases. T h e essential feature of this approach is to diagonalize the Jacobian m a t r i x A given in (16). We can write

A=

S-' AS,

(21)

where A is a diagonal m a t r i x whose diagonal elements correspond to the eigenvalues of A and the rows of m a t r i x S are composed of the right eigenvectors of A. A flux-splitting may be developed by splitting A into a non-negative m a t r i x h + and a non-positive m a t r i x A - , where the diagonal elements of h a= are ~a= = ( I I),])/2. Utilizing the homogeneity property, (17),
y = A u = s (^+ + ^ - ) s - 1 u = F + + e - . (22)

We obtain after some algebraic rearrangement 1 where +

+

1 . ~p),cFc,

(23)

Pl/P P2/P

p,/p P2/P

FA =

PN/P u
Plenl/P
PMenM/P

.F~ ~c

PN/P u~a #l enl /P pMen~/P ho ua

(24a, b, c)

ho

and ),A = u, I B = u + a and I c = u - a. This represents a Steger-Warming-type flux-vector splitting for a vibrationally-relaxing, chemically-reacting flow. An alternate flux-vector splitting has been developed for perfect gases by Van Leer, [5]. His formulation has continuously differentiable flux contributions and has been shown to result in smoother solutions near sonic points. For the non-equilibrium flow model considered here, the procedures of [5] may be directly utilized by considering

330

the flux vector as F = F(p, a, M, pi/p, e,~,), where M is the Mach number u/a. Then the mass flux, pu = p a M may be split as pu = f+m + fro, where

f ~ = ~pa(J=M + l 2

(25)

The remaining fluxes when written in the above functional form may be split to yield a Van Leer-type flux-vector splitting for a non-equilibrium flow:

Pl/ P P2/P

"~

F ~ = f~

PN/P [(~_l)u=i=2a]/Z/ plevl/P pMevM/P

'

(26)

where

f ~ = ho - r n ( - u ~ a) 2.

(27)

and m is an arbitrary parameter. For consistency with previous developments [8], rn has been assigned the value 1/(~ + 1). Numerical experiments have shown no sensitivity to the value of this parameter.

Flux-Difference Splitting The essential features of flux-difference split algorithms involve the solution of local Riemann problems arising from the consideration of discontinuous states at cell interfaces on an initial data line. The scheme developed for perfect gases, by Roe [6], falls into this category and has produced excellent results for both inviscid and viscous flow simulations (c.f., [11]). We now extend this approach for the present non-equilibrium thermodynamic and chemistry model. We follow the procedure outlined by Glaister [12] for real gases, which followed from the approach of Roe and Pike [13] for perfect gases. In their approach, the first step corresponds to obtaining a solution for the linearized, approximate Riemann problem, valid for small jumps in the interface states. The second step involves the determination of the appropriate averaging procedures, (often referred to as Roe-averages), in order to render the solution valid for arbitrary jumps in the cell interface values. We may extend this procedure directly to our non-equilibrium flow model, following the details outlined in [8] for the simplified vibrational energy model. The resulting algorithm is now summarized. We initially seek a solution to the approximate Riemann problem for two states, Dr, Ue, which represent the conserved variables evaluated at the cell interface using data from the right and left, respectively. This involves the determination of the relationship

331

between the j u m p in U, denoted as ~UI ~ Ur - U~, to average eigenvectors Ei of the Jaeobian matrix, and average wave strengths &i, i = 1 , . . . , N + M + 2, such that
N+M.4-2

~u~ =

~_.
i=l

a, ~,.

(281

The jump in the flux vector F is related similarly as

N+M+2

~F~ =

~
4=1

a,

k,i~,.

(291

The eigenvalues ),i and the eigenvectors Ei are found to be those given in (18) and (19), with averages denoted by a caret over each dependent variable. We also adopt the notation
h

en,

=--

(30a, b)

The corresponding wave strengths &i are found to be:

P--7-- a-~-'

i= 1,...,N,
j --- 1 , . . . , M , k = 1,2.

(31a)
(31b) (31c)

~pJc.3
~j-}-N -^

~p~

enj

a2 ,

1 &k+N+U = 2--~ (~p~ ~: ~a~Ul),

The solution of the approximate Riemann problem involves the algebraic determination of the averages ~, fi, h, 15, h0, ~i, e,~,, ~i, such that (28) and (29) are satisfied. We define the Roe-average of a quantity f to be ~ ( f ) ~ (frv/-fi7 + f ~ V / ~ ) / ( X / ~ + x/~)- Most of the averages turn out to be Roe-averages. Namely, fi = ~(u), h0 = ~(h0), Pi = ~(Pi/P), ~,tj = ~(Pienj/p), /~ = ~(/~), :~ = ~(T) and ei = ~(ei). The remaining quantities needed are the usual density average, ~ = ~ and
N

~* = F_, pl ~* ~ ~,,
i=1

(32a)

where c~, =- IT ~ We also find ~ , dT.

~2

(32b)

& = R'i~ - ~' + 7 '

and
k - 1 = ^-:,
Cv

(33)

(34)

332
and

f2 :

(~ _

1) [~t0

~22

N i=1

M j:l

We have not proven that the average sound speed, defined above, will be positive for arbitrary left and right states. However, numerical experiments with very strong discontinuities confirm its utility We have now defined the averages so that (29) constitutes an approximate Riemann solver. We may rewrite (29) by grouping the repeated eigenvalues and upon algebraic simplification we obtain where the ~F~A term arises from the first N + M terms of the sum in (29), corresponding to the repeated eigenvalue Ai = fi and may be written as

~2

/"

[r~h : (I~.II\
err 1 errM Izo - &2/(~/-1)
^ ^

+~

IPN/pl

Ipl~,~,/pl
,

~pMe,~/p~

(37a)

Similarly the ~F~B and ~F]]C arise from the last two terms in the summation of (29), corresponding to the eigenvalues ~2+ a and fi - a, and are found to be

lf~,o

1 (~e~~aI,~g)(aa) = 2--g

aia ~N
er~l ^ ertM [ /

"

(37b, c)

T h e approximate Riemann solver is implemented in a numerical integration of the governing equations (14), by computing the cell interface flux as a summation over wave

333

speeds, [2], as
1 N+M+2
i=1

Numerical Formulation
The implementation of the non-equilibrium flux-split algorithms follows procedures similar to those for perfect gases, e.g., [3]. Since we anticipate stiffness problems in the numerical solution of the governing equations, we seek to compute solutions using an Euler-implicit time integration. The implementation of the implicit algorithm in a flux-vector split scheme involves the splitting of the flux, F = F + + F - , and the splitting of the flux Jacobian, A = A + + A - . The spatial derivatives of these terms are evaluated using the usual upwind differencing, (in MUSCL form), appropriate for flux split codes The details m a y be found, for example, in [3]. The split flux Jacobians, A ~: = OFi/OU, have been analytically determined from the flux-vector splittings given in (23), (24) for the Steger-Warming-type scheme and given in (25)-(27) for the Van Leer-type scheme The Jacobian of the chemical and thermodynamic source terms, OW/OU, are needed for the computation and depend upon the specific chemical and thermodynamic models Some examples are given in [8]. In addition, for a quasi-one-dimensional flow computation, the effect of area variation on the governing equations (14) may be included by adding to the source vector W, (15c), a vector W~, where
Pl

PN

Wa-

u dA A

pu

dx

plenl

(39)

pMen~

pho
and A(x) is the cross-sectional area distribution. The Jacobian bWa/OU must also be included in an implicit formulation. NUMERICAL SOLUTIONS

We include several numerical examples of flow computations with non-equilibrium chemistry and thermodynamics, in order to illustrate the accuracy and wave-capturing properties of our methodology. The test cases involve the steady combustion in a supersonic diffuser, including an embedded shock wave. In all the test cases presented here, we will utilize the simplified vibrational model. Thus the internal energy of each (non-monatomic) species will contain an equilibrium portion which is linear in T and a non-equilibrium vibrational energy whose production rates come from a Landau-Teller model, [14].

334

T h e flow entering the diffuser is composed of hydrogen mixed with air. For the hydrogen-air chemistry, we utilized the same model used in [151, the simple, two-reaction hydrogen-oxygen model of Rogers and Chinitz [16]. It consists of the two basic reactions:

H2 + 02 ,~- 2011 2 0 t l + H2 ,~- 2Hg9

(41)

and five species ( N = 5) are utilized, N2, 02,H2, OH and H20, all of which are considered to be vibrationally relaxing ( M = 5). The reactions of nitrogen are neglected in this model. T h e particular geometry, an axisymmetric, rapid-expansion supersonic diffuser, has a radius defined as r / L = 0.12511 + sin(~rx/2n)] for 0 < x __< L, with a length L = 2.0 meters. The flow at the inlet to the diffuser (x = 0), has a velocity of 1245 m/sec., a t e m p e r a t u r e of 1884.3 K and a pressure of 0.8026105 N / m 2. The initial chemical composition of the flow consists of an equivalence ratio (I), the ratio of the mass fraction of H2 to the mass fraction of O~, normalized by the stoichiometric mass fraction ratio, of (~ = 0.29841. This composition causes very steep concentration gradients near the inlet of the duct. Computations were performed using three different energy models. The first one utilized the simple vibrational model, which did not require any iterations for the determination of pressure and temperature from the conserved variables. In the second case, the vibrational energy was considered to be in equilibrium, distributed according to an equilibrium harmonic oscillator. The resulting ei are the sum of linear and exponential contributions, and iterations are required to recover the value of the temperature. For a typical solution using Newton's method, two to three iterations were needed to produce values accurate to seven significant digits. For the third case, a curve-fit ei distribution using a quadratic t e m p e r a t u r e profile for each species was incorporated. T h e resulting model is the same one used by D r u m m o n d , Hussaini and Zang [15]. In this model, one can directly solve a quadratic relationship for the temperature from the conserved variables. All our computations were performed using 161 uniformly-spaced, axial grid points and the equations were solved iteratively, using our fully-coupled implicit method, with the Steger-Warming-type flux-vector splitting, given by (23) and (24). T h e computation was performed with a CFL number of 10 and convergence to a steady state, with a residual reduction of 10 -1 in 110 N 130 cycles. It may be noted that an explicit solution of the same problem would have required CFL numbers of the order of 10 -3 and at least one order of magnitude more computer time, due to the stiffness of the combustion model. In Fig. 1 we present the comparison of our computations with a linear c,~ model along with the spectral and finite-difference computations of D r u m m o n d , Hussaini and Zang [15]. It should be noted that their spectral calculation used 17 nodes, and their finite-difference calculation used 101 grid points (although only every second value is indicated on our plots). T h e three results for pressure (given in bars, 1 bar = 105 N / m 2 ) , and t e m p e r a t u r e are in excellent agreement. In Fig. 2 we present the comparison of our computations using the three energy models for this case. T h e three cases do not show any significant differences for the pressure distribution. The t e m p e r a t u r e distribution however, indicates that substantial

335

1.2 1.0

2600 2400 0 Spectral [15]

S o.s~
~D

%
CL

0.6 P

2OOO 1800' 1600

1400
I I q I 1 I I I i I I I I I I I i I I

-U~0.4 0.2 O.O @,0

0,4 0.8
x

1.2

1.6

2.0

0.0

0.4

O.B

1.2

1.6

2.0

(~)

(~1

Figure 1. Non-equilibrium flow of an H2-air mixture through a supersonic diffuser. Comparison with other numerical methods.
non-equilibrium energy effects are present under these conditions. The two equilibrium energy models are, as expected, in agreement. Plots of the vibrational temperatures for each species, (not presented here), show that the flow has not reached vibrational equilibrium. It appears that a non-equilibrium internal energy model is needed to obtain accurate temperature distributions for this case. Whether the simple vibrational energy model is adequate remains to be determined.
1.2F 2600 24O0 0
x -

Linear c v
Equil, evl b

8 0.8~ tn
o

~2200
>

Non eq. ev;b

m 0,6-

200O
1800 ~ 1600 ~ 1400!

- ~ 0,4 m 0,20.0 0,0

0.4 0.8 1.2 1.6 2.0

0.0

0.4

0.8

1.2

1.6

2.0

(m)

(m)

Figure 2. Non-equilibrium flow of an H2-air mixture through a supersonic diffuser. Comparison between three thermodynamic models.
Another computation was performed with this same supersonic diffuser, for conditions where a normal shock wave stands inside the duct. The exit pressure was raised to 10.02 times the perfectly expanded value. Using the identical inlet conditions and chemistry model, discussed previously, the computations were performed with an exit pressure specified to be 105 N / m 2. The conserved variables, pi,pien~ and pu were extrapolated to the exit boundary. The Van Leer-type flux-vector splitting, (25)-(27), with second-order upwind, spacial differencing and a MIN-MOD limiter, was used for this computation. We utilized a CFL number of 5 for the Euler-implicit time integration. The resulting pressure and temperature distributions, as plotted in Fig. 3, show the computation performed with the same three thermodynamic models and the same grid as in the previous example. The results for the two equilibrium energy models are

336

again in good agreement, and they both depart from the simplified vibrational model results in the lower temperature region. After the shock, the flow reaches vibrational equilibrium after a short relaxation zone. The slight discrepancy in the temperature profile between the exponential and the quadratic ei cases are likely due to the differences in the thermodynamic models.
1.2 1.0
u)

3200[

0.8c
0.6

~ 12812600 i24oo1
22ooif % .

// o
I

iooor cv

UJO.4 c~
0.2 0.0 0.0
I r I I I J I I r

~ Eqoi,. %,~

0.4

0.8

1.2

1.6

2.0

1600~0.0

, ""~

0.4

0.8
(~)

1.2

1.6

2.0

Figure 8. Non-equilibrium flow of an H2-air mixture through a supersonic diffuser with an embedded shock wave. Comparison between three thermodynamic models.
The computed mass fractions for both flows considered here, with solutions for the three thermodynamic models are shown in Fig. 4. The left hand figure corresponds to the shockless diffuser flow and the right hand figure is for the flow with an imbedded shock wave. The results for the linear cv, model are in very good agreement with those of [15].

025 f-

0"25 f 0 . 2 0 ~

~- 0

+5
o

:~ 0 ~

15'

-- ............

0.10 0.05 0.00 0.0

H20

0 x
~ 0.10 u H20 -

Linedr cv Equil. evi b

Non eq. evib

0.05 O.OO 0.0

I i J I i I I r I i

0.4

0.8
(m)

1.2

1.6

2.0

0.4

0.8
(m)

1.2

1.6

2.0

Figure 4. Computed mass .fractions ]or the non-equilibrium flow of an H2-air mixture through a supersonic diffuser.

337

Although we have been successful in obtaining efficient computations of steadystate solutions in one space dimension with an implicit algorithm, these results may not necessarily carry over to multi-dimensional flows. The penalty associated with solving very large block-structure tri-diagonal (or penta-diagonal) systems may overrun the benefits in CFL number for an implicit method. ACKNOWLEDGMENT This research was funded by the NASA Langley Research Center under grant NAG1-776. REFERENCES A. Harten, P. D. Lax and B. van Leer, SIAM Rev., 25, 35, 1983. P. L. Roe, Ann. Rev. Fluid Mech. 18,337, 1986. R. W. Walters and J. L. Thomas, "Advances in Upwind Relaxation Methods", in State-of-the-Art Surveys on Computational Mechanics, ed. A. K. Noor, (A.S.M.E. Publication, New York, 1988). J. L. Steger and R. F. Warming, J. Comp. Phys., 40,263, 1981. B. van Leer, in Lecture Notes in Physics, 170, (Springer-Verlag, Berlin, 1982), p. 507. P. L. Roe, J. Comp. Phys., 43, 357, 1981. B. Grossman and P. Cinnella, "The Development of Flux-Split Algorithms for Flows with Non-Equilibrium Thermodynamics and Chemical Reactions", AIAA Paper No. 88-3596, 1st National Fluid Dynamics Congress, July 1988. B. Grossman and P. Cinnella, "Flux-Split Algorithms for Flows with Non-equilibrium Chemistry and Vibrational Relaxation", ICAM Report 88-08-03, Virginia Polytechnic Institute and State University, Blacksburg VA, August 1988. G. Herzberg, Molecular Spectra and Molecular Structure, L Spectra of Diatomic Molecules, (D. van Nostrand, Inc., Princeton, N. J., 1950). R. L. Jaffe, "The Calculations of High Temperature Equilibrium and Nonequilibrium Specific Heat Data for N2, 02 and NO", AIAA Paper No. 87-1633, 1987. B. van Leer, J. L. Thomas, P. L. Roe and R. W. Newsome, "A Comparison of Numerical Flux Formulas for the Euler and Navier-Stokes Equations, AIAA Paper No. 87-1104-CP, 1987. P. Glaister, J. Comp. Phys., 74, 382, 1988. P. L. Roe and J. Pike, "Efficient Construction and Utilisation of Approximate Riemann Solutions", in Computing Methods in Applied Sciences and Engineering VI, edited by R. Glowinski and J.-L. Lions (North-Holland, Amsterdam, 1984), p. 499. W. G. Vincenti and C. J. Kruger Jr., Introduction to Physical Gas Dynamics, (John Wiley and Sons, Inc., New York, 1965), p. 135. J. P. Drummond, M. Y. Hussaini and T. A. Zang, AIAA J. 24, 1461 (1986). R. C. Rogers and W. Chinitz, AIAA J. 21,586 (1983).

1. 2. 3.

4. 5. 6. 7.

8.

9. I0. 11.

12. 13.

14. 15. 16.