You are on page 1of 36

197 3ApJ. . .181. .

903M

The Astrophysical Journal, 181:903-938, 1973 May 1


© 1973. The American Astronomical Society. All rights reserved. Printed in U.S.A.

RELATIVISTIC SHOCK HYDRODYNAMICS


Chester R. McKee and Stirling A. Colgate
New Mexico Institute of Mining and Technology, Socorro
Received 1972 May 23; revised 1972 November 6
ABSTRACT
The special-relativistic equations for a perfect fluid are differenced in characteristic form. Since
information concerning the state of a fluid is propagated along the characteristic directions, the
differencing procedure preserves a fundamental property of the hydrodynamic equations. Shocks
are calculated using an energy Lagrange frame where energy fractions become a coordinate. Since
numerical zones are attached to surfaces of constant energy, they move so as to conserve the total
energy outside a surface and hence strong shocks appear stationary in this frame. The jump con-
ditions across the shock can be enforced directly, so the need for an artificial viscosity to spread the
shock over several mass Lagrange zones is eliminated. The result is an order-of-magnitude reduc-
tion in computation for equivalent accuracy. On the other hand, the differencing of the equations
in characteristic form using a mass Lagrange frame is particularly well suited to the problem of
expansion of a relativistic fluid into vacuum. Examples are given of calculations of shocks and
expansions that agree in detail with analytical solutions.
Subject headings: hydrodynamics — relativity — shock waves

I. INTRODUCTION

Relativistic hydrodynamics has several applications to cosmic phenomena that


demand a more detailed knowledge of fluid flow than can readily be obtained from
similarity solutions or other analytical forms. The shock acceleration and subsequent
expansion acceleration of the outer layers of a supernova star is one direct application
of the current numerical work, but the methods are applicable to black-hole formation
(May and White 1967) as well as to other relativistic problems.
The shock acceleration of the outer layers of a supernova was suggested by Colgate
and Johnson (1960) as a possible origin of cosmic rays and a similarity solution was
suggested that was obtained by the method of characteristics (Riemann invariants)
that has been published in detail by Johnson and (Chris) McKee (1971).
In nonrelativistic fluid flow, the method of characteristics is particularly applicable
to the vacuum expansion of a fluid initially static and with uniform energy and mass
density. The same problem in the extreme relativistic limit was shown by Johnson
and McKee to give a relatively simple power law for the increase in energy of the
leading edge of the fluid. They furthermore showed that the expansion of such a fluid
into a region of decreasing rest mass density resulted in a shock whose strength
increased as the density ahead decreased. They gave a significant justification that
such a shock ran away from, or was only weakly perturbed by, disturbances originating
from a region of nonuniform energy distribution. Since the shock itself produces a
very weakly varying energy distribution, the conclusion of a rapidly asymptotic
solution was fortified. The same could not be justified for partially relativistic con-
ditions yft < (rs — 1) < 10, rs = l/VO — ßs2)> cßs = shock velocity.
Postshock expansion was treated in the plane-parallel limit and led to a solution that
corresponded to the behavior of the leading edge of a vacuum expansion—again, with
negligible perturbation from nonuniform energy density deeper within the expanding
fluid.
903

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

904 CHESTER McKEE AND STIRLING COLGATE Vol. 181

In the case of a relativistic shock in the envelope of a supernova star, we are con-
cerned with results that duplicate many of these conditions (Colgate and White 1966),
particularly those where the shock wave becomes relativistic in the polytropic stellar
envelope and where the scale height is small compared with the radius (~10-3) so
that plane-parallel solutions are applicable. On the other hand, there are regions
outside the applicability of the asymptotic solutions where the shock makes the
transition from nonrelativistic to the relativistic limit and where the expansion is
initially plane-parallel geometry, but then becomes predominantly spherical. In the
latter case, Elgroth (1971) has shown that the asymptotic expansions are significantly
different depending upon the geometry of expansion.
For these reasons, as well as future considerations of radiation diffusion, a numeri-
cal relativistic computation is required. We have done so with an energy Lagrange
and shock-following program for calculating shocks and a mass Lagrange program
for postshock expansion. We have duplicated numerically the more direct asymptotic
solutions and confirmed the assertion that the shock outruns any initial perturbations.
We have performed a calculation that stimulates the shock acceleration in a supernova
envelope where the transition from nonrelativistic to relativistic shock strength is
demonstrated, as well as the integration of the expansion from planar to spherical
geometry.
The principal numerical work that preceded this is that of May and White (1967)
where the major emphasis was on general-relativistic solutions to the problem of
black-hole formation, but where, in addition, a calculation was made of a shock in
a density gradient. A formulation with both mass Lagrange zoning and artificial
viscosity was used. The effects of the artificial viscosity were investigated at length,
and the results were significantly insensitive to major changes in the details of the form
of the viscosity used. Nevertheless, several zones at the shock front showed a possibly
significant departure (up to 30 percent in energy density for Fs ^ 100) from the exact
form of the relativistic Hugoniot relations, and some doubt existed as to how this
might affect the shock strengthening solution. We confirm that these calculations gave
the right answer and apparently the shock was merely spread over a few too many
zones (~3) relative to the local scale height (~5) to develop the steady-state shock
conditions. The expansion of the postshock fluid could not be calculated because of
boundary-value problems, which have since been solved (Woods 1971) by synchroniz-
ing the time coordinate at r = 0, and by using finer zoning. Instead, we have found it
expedient to directly enforce the jump conditions at the shock front for greater accuracy
with an-order-of-magnitude fewer zones.

II. FUNDAMENTAL EQUATIONS


a) Derivation from Energy-Momentum Tensor
The energy-momentum tensor for a perfect fluid in the absence of force fields is
(Landau and Lifshitz 1959)

TßV = (p + e + Pc
2
)ußuv - gßVp (ILa.l)

where p = pressure, e = internal energy per unit volume, p = amount of matter


(excluding pairs) per unit volume, c = velocity of light, uß = four velocity measured
in chosen frame of reference, and gßV = metric tensor, and the thermodynamic
variables p, e, />, are measured in the proper frame. The equations of motion are
obtained by equating the divergence of equation (Il.a.l) to zero:

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 905

and by requiring that the four-divergence of the mass flux vanish (Landau and Lifshitz
1959):

1
(pwV-g) = o, (II.a.3)
V-g

which expresses the fact that the total amount of matter excluding pairs is conserved.
Choosing our metric to be

ds2 = (Jx4)2 - (dx1)2 + g22(dx2)2 + gw(dx*)2 , (II.a.4)

with g44 = 1, gn = — 1, and defining the components of the four-velocity as

dx1 ß
= u1 = w2 = w3 = 0
ds V(1 - ß2)
dx* 1 a dxl
= w4 = ß = dP (II.a.5)
ds V(1 - ß2)

we obtain the following expression for the energy-momentum tensor:

p + ß2(e + pc2) p + e + pc
0 0 ß
1 - ß2 1 - ß2

J'ßV _
0 -g22P o 0
(II.a.6)
0 -g33p 0
p + e + pc2 e + pc2 + ß2p
ß o
1 -ß'- 1 ~ß-

Using the appropriate values for the metric coefficients and Christoffel symbols, we
may write equation (II.a.2) and (II.a.3) in the general one-dimensional form

d ßP+J_± pc^ + p + ß2(e + pc2)


a (II.a.7)
3t i -ß'- r dr [' 1 -ß'- r

d e + pc2 + ß2p p + e + pc'-


+ 11 [ r°ß = 0, (II.a.8)
dr 1 - ^ r° dr [ 1 - ß2

P = 0, (II.a.9)
8t V(1 ) +1 ll\r°ß
ra dr I r V(1 ” ß2)

where r = ct and a = 0 denotes plane symmetry, ct = 1 denotes cylindrical symmetry,


and o- = 2 denotes spherical symmetry.

b) Principal Forms of the Equations


There are many forms in which the above equations may be written to serve as a
basis for deriving a difference method, each form displaying some particular advantage.
We will be interested in differencing the equations in characteristic form, but will note
other possibilities in our derivation of the characteristic form.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

906 CHESTER McKEE AND STIRLING COLGATE Vol. 181


i) Relativistic Euler's Form
Multiplying equation (II.a.8) by ß and subtracting from (II.a.7) yields

dß , a 8ß - 1 - ß2
(II.b.1)
d-r + ßTr- p + e + pc2 (4M)

while multiplying equation (II.a.7) by ß and subtracting from (II.a.8) gives

^ (e + pc2) + ßjr(e + pc2)

p + e + pc2 (Rdß dß\ o


2 + + e + pc2) . (II.b.2)
1 - ß {ßd-r T r J--rß^

Performing the indicated differentiation in equation (II.a.9) results in

8
p 4. a dP - (II.b.3)
Tr + ßTr-

From a computational point of view equation (II.b.2) is not very useful since if
p ~ e « pc2, then for all practical purposes equation (II.b.2) is identical to (II.b.3).
A more useful form is obtained by multiplying equation (II.b.3) by c2 and subtracting
it from (II.b.2), giving

de
4. ñde p + e
+ ß =
d-T dr 1 - ß2

and then combining this with equation (II.b.3) to produce

1 de de ßp + ß8-p = 0. (II.b.5)
p + e dr dr B dr^ P dr

Equations (II.b.1) and (II.b.3) are the relativistic Euler equations, while (II.b.5) is a
form of the first law of thermodynamics for fluids.

ii) Matrix Form


To derive this form we choose the equation of state

P = Pie). (II.b.6)

Other more general expressions could be chosen, notably

P = PÍP, e). (II.b.7)

where € = energy per unit mass, but (II.b.6) will be adequate for the purposes of this
article. Using equation (II.b.6), derivatives of p may be replaced by derivatives of e
multiplied by dp¡de. Choosing any three independent equations from the previous
section and solving for the time derivatives of ß, e, and p results in three equations
which may be written in the matrix form

dW
^ tA^dW - r (II.b.8)
+ G
dr ^ dr '

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 907


where W and G are the vectors

ß (i - ß2)mr
W= e G = Z ~ß(P + e) (II.b.9)
rT my
LPJ -Pß
and [A] is the matrix

ßa _ o 2, (1 - ndp/de 0
R Pa)
p + e + pc2
1
[A] = p + e ß(l - ßa)2 0 (II.b.10)
i - my
Pß(l - ß2) dp
p
p + e de ß[l -Xßßa)2]

with

R — / p + e dpX1!2 (n.b.ll)

The characteristics of the system (II.b.8) are

dx\ ß ± ßa
= A± (II.b.l2)
dr) ± l±ßßa9

dx\
— K — ß9 (II.b.13)
Tt)0

which are the eigenvalues of the matrix A (Courant and Hilbert 1962). Equation
(II.b.12), recognized as the relativistic formula for addition of velocities, states that
small disturbances (sound waves) are propagated relative to the fluid with speed ßa,
so that ßa is the adiabatic sound speed of the fluid.

iii) Characteristic Form


A particular equation in the characteristic form contains total derivatives in only
one of the three characteristic directions (II.b.12) or (II.b.13). This is accomplished by
multiplying equation (II.b.8) by a matrix [71] to obtain

dW dW
[T}°-g + miA}0-^ = {T]G (II.b.14)

We then require that

[T][A] = [Aim, (II.b.15)

where

p+ 0 0 ■
[A] = 0 A_ 0 (II.b.16)
0 0 A0_

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

908 CHESTER McKEE AND STIRLING COLGATE Vol. 181

is a diagonal matrix consisting of the eigenvalues of [^4]. Using the condition (II.b.15),
the elements of |T] are found to be

1 - ¿82
1 ßa 0
p + e
1 - ff2
[T] = 1 — ßa 0 • (II.b.17)
p + e

0
p + e

Equations (II.b.14) may now be written as the three equations

I (Jl , ß_±_ßa g \o , ßa (Jl , ß_±_ßa g \ x^ ßßa


2 —
l — ß \Sr 1 + ßßa dr) — p + e \ör — 1 ± ßßa dr) r 1 + ßßa
(II.b.18)
and
1
(II.b.19)
p + e

which are in the characteristic form, and will be used to derive the difference method
in § V.
We will now derive the Riemann and adiabatic invariants which will prove useful
in checking the accuracy of the numerical solution in certain problems. Assuming ßa
and p to be functions of e only yields the partial integrations

l (II.b.20)
s), (i rH ± frh *) - ^rfk
and

(ILb.21)

where
ß±_ßgd
A + J (II.b.22)
dr)± dr 1 ± ßßa Sr
and

-) (II.b.23)
dT Jo

are the total differentials among the forward (+), backward (—), and fluid (0)
characteristics. The forward and backward characteristics represent the path of sound
waves in the (r, r)-plane moving respectively in the direction of larger or smaller r.
The fluid characteristic is the path of a fluid element as it moves about the (r, r)-plane.
If the equation of state p = (y — l)e is used in equation (ILb.21), the result is

(ILb.24)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 909

Hence e//>y is the adiabatic invariant along a fluid characteristic. Equation (II.b.20)
can be integrated in two special cases if o- = 0. The first way is to substitute the
relativistic limit for the sound speed, ßa — (y — l)1/2> into equation (II.b.20), yielding
the approximate Riemann invariants
1 2
!±i exp [ , 2(y - l) ' ' = 0. (II.b.25)
i-ß

The second way of integrating the equation (II.b.20) occurs when the adiabatic
invariant is everywhere constant (isentropic flow). If and pt are the initial conditions,
then substituting
P — Pi{eledlly (II.b.26)

in equation (Il.b. 11) reduces the sound speed ßa to a function of e only. Evaluation of
the integral in (II.b.20) is simplified by the substitution

(n b 27)
HfM““’' ' '

giving the Riemann invariants (Johnson and McKee 1971)

{[yifJL l)]m + [y(/x 1} +


± - " 1]1/2}±4M,'-1) = 0 . (II.b.28)

Equation (II.b.28) can also be written as

d\ l + ß 1/2 ±4/V(y-l)
= 0 (II.b.29)
drjtl- ß ++
■)

It now remains to choose a frame of reference in which to perform the calculations.


We will be guided in this choice by considerations involving the velocity of a strong
shock derived in the next section.

III. SHOCK RELATIONS


In this section we use the equations

S pP + e + pc2 Id \ p + ß\e + pc2)] °P _ n


(II.a.7)
8tP l - ß2 +
ra8r [ 1 - ß2 J r

d_ e + ß2p + pc2[l - VO - ß2)]


8r 1 - ß2
+ e +
i ^ pP Pc2^ — _ o (iiLi)
r°d?\r ß —r=j2

g p , i g = 0, (II.a.9)
St V(1 - ß2) r° dr V(1 - /32)]

where equation (III.l) was obtained by multiplying (II.a.9) by c2 and subtracting it


from equation (II.a.8). The above equations are easily seen to reduce to the classical
conservation laws.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

910 CHESTER McKEE AND STIRLING COLGATE Vol. 181

The essential features of these equations are contained in

¿O-M)+ !(/•'£) +C = 0. (III.2)

To derive the shock relations, equation (IIL2) will be converted to the integral form,
since this form is valid across a discontinuity. Choosing a simply connected region in
the (r, r)-plane to integrate over gives

M) + I (r'i?) + drdr = 0. (III.3)

Applying Green’s theorem in the plane to the above equation yields

cdrdr = 0, (IIL4)

where the first integral is over the closed path surrounding the region. Following
Harlow and Amsden (1971), we choose a small rectangle drawn about a portion of the
shock world line (see fig. 1). The segments of the rectangle normal to the curve will be
constructed small compared to those parallel to the curve. The contribution of the
normal segments to the line integral in equation (III.4) will then be negligible. If all
variables are finite, and the rectangle is very small, then the area integral in equation
(III.4) is a higher-order quantity than the line integral and can be neglected. Further-
more, r is continuous across the shock and can therefore be taken as approximately
constant over the rectangle. Equation (III.4) when specialized to this small rectangular
region reduces to

<J> {Bdr - Adr) = 0. (III.5)

In the notation of figure 1, this line integral is approximately

■d+(r2 - fl) - A ~(/*2 - Tl) = B + (t2 - Tj) - B~(t2 - Tj) . (III.6)

Fig. 1.—Shock world line. Above the line is unshocked material, and below is material processed
by the shock. Integration of the relativistic equations over the rectangle leads to the shock jump
conditions in the laboratory frame.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 911


T

Fig. 2.—The characteristic paths for the forward ( + ), fluid (0), and backward ( —) character-
istics are shown on a computational grid in which the space coordinate / is the external energy
fraction. Information is propagated to point j at time level « + 1, from the base of the character-
istics y + , y0, j- at time n. The energy frame moves faster than the fluid, which has the effect of
shifting the characteristics to the right as observed in this frame.

Dividing by (r2 — Ti) and letting the shock velocity be

ra ri
ft = lim ~ , (III.7)
r2”>ri,T2->Ti ^2

and introducing the notation

8A = A+-A-, 8B = B+-B-, (III.8)

we obtain the jump condition

ßs8A = 8B . (III.9)

The result is that shock relations derived for plane symmetry are equally valid for
cylindrical and spherical symmetry.
Applying equation (III.9) to the first three equations of this section, assuming plane
symmetry, gives the jump conditions
2
,sa(£^!) = a p + ftO? + pc ) (momentum) (III. 10)
1 - ft

e + ß2p + pcV - (1 - m]
fts{ 1 - ft /

p + e + pc2[\ - (1 - ft)1'2])
= S (energy), (III. 11)
1 - ft I

pß (mass). (III. 12)


V(i - ft) V(1 - ft)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


912 CHESTER McKEE AND STIRLING COLGATE Vol. 181

For our purposes it will be sufficient to consider the material ahead of the shock to be
at rest with the values e0, Po, and p0, which will in general be functions of r. The above
equations therefore become
p + e + pc2 p + ß\e + pc2)
-Po (III. 13)
1 - ß2 1 - ft
e + ß2p + pc2[\ - Vd - ß2)]
ft - e0
1 - ß2

p + e+ pc2[\ - V(1 - ft2)]


= ß (III. 14)
1 - ft

pß (III. 15)
ft 22 /5
(va
V(1 - ß
ft ) °) va - ft2)
which may be rearranged to give

= p
-±± o + PoC2) (III. 16)
P
1 - ftft
ß(l - ß2)(p + e0)
ßs — ß — ß2p + e + c2[l - V(1 - ß2)] - e (l - ß2) (III. 17)
P 0

ftva - ß2) (III. 18)


P = Po
ßs — ß
where equation (III. 16) is the result of subtracting equation (III. 13) from ßc2 times the
sum of equations (III. 14) and (III. 15).
The above equations may be manipulated into a form closely resembling the classical
jump conditions. Multiplying equation (III. 13) by ß and subtracting it from equation
(III. 14) yields

e = pC
pC2( ! + fte° ±_ßPo . mi ¡g)
\V(1 ~ß2) ) ßs-ß ^ ’

Solving equation (III. 17) for the combination ft/(ft — ft) and substituting into equation
(111.18) yields

p + e pc
2
(III.20)
po va - ft ) P + ^0 p + e0 (v(i - ft2) 0
Assuming a strong shock (e0 = p0 ~ 0), equation (III. 19) becomes
2
e = Pc [(l -ß2)-"2- 1], (III.21)

while equation (III.20) has the form

p + e
Ü = i + e (III.22)
Po P va - ft2)
For an ideal gas, equation (III.22), in the extreme relativistic limit, becomes

1
= y , i (III.23)
Po y - 1 va - ft2) y - 1 ’

© American Astronomical Society • Provided by the NASA Astrophysics Data System


No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 913

and in the classical limit gives the well-known compression ratio across a shock:
p_ = y + 1 _
(III.24)
Po y - 1
In the classical limit, equation (III.21) becomes

e = ±Pv2 , (III.25)

expressing the fact that immediately behind a strong shock the internal energy is equal
to the kinetic energy. Also, in this limit, the shock velocity (III. 17) is

». - »(> + ■ <111.26)

In the extreme relativistic limit we can relate, in a rather simple way, the relativistic
energy factor of the shock wave to that of the shocked fluid by introducing the
notation

& = 1 -t/s, ß = \ — u. (IIL27)


Then the relativistic energy factors are
1,2
rs = [ns(2 - Ms)] - , T = [n(2 - u)] ~1/2 . (III.28)

Introducing this notation into equation (III. 17) gives the relation
r
rs (IIL29)
V[1 - 2(y ~ l)/y] ’
which is valid in the extreme relativistic limit (ß-> l, e » >c2), for an ideal gas.
We shall discover in the next section that the shock velocity obtained from equation
(III. 14) with e0 = 0 (strong shock),

*
Ps
R P + e + PC2V - V(l - £2)] (III.30)
~Pß2p + e + pc2[l - V(1 - ß2)] ’

is the velocity of energy transport in the fluid, and will provide us with a method for
solving problems involving a strong shock without introducing an artificial viscosity.

IV. FRAMES OF REFERENCE


A coordinate system constructed such that the distance between two arbitrary
points in space does not change with time is said to be fixed or stationary. Such a
frame in hydrodynamics is called an Eulerian frame. If we require that the points of
the coordinate system remain fixed in the fluid, then each point moves with the local
fluid velocity, and the system is called a mass Lagrangian frame. One can also attach
the points of the coordinate system to surfaces of constant energy, in which case the
system is termed an energy Lagrangian frame.

a) Conservation Laws in Moving Frames


Conservation laws in one dimension have the form

(III.2)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

914 CHESTER McKEE AND STIRLING COLGATE Vol. 181

We will use this equation to find the velocity with which points of a coordinate system
must move in order to guarantee local conservation of a quantity Q. One can define a
Lagrangian frame for any quantity which is continuous and always greater than 0 or
always less than 0. The fraction of a quantity r external to a radius r is given by

f= J ^Adr, where M = j ¿FAdr, (IV.a.l)

and A = \ denotes plane symmetry, A — lirr denotes cylindrical symmetry, and


A = 477r2 denotes spherical symmetry. The radius r may therefore be considered as a
function of the Lagrange coordinate / and the time t = ct:

r = r(/, r). (IV.a.2)

We may also assume that equation (IV.a.2) can be solved for / in terms of r and r,
giving

f = f(r,T). (IV.a.3)

Using equations (IV.a.2) and (IV.a.3), we obtain the following expression for the
Jacobian of the transformation:

(IV.a.4)
df idf/dr) *

Evaluating dfjdr from equation (IV.a.l) gives

J = —MI&A . (IV.a.5)

Consider an arbitrary volume element in a Lagrange frame containing a quantity g.


The walls of the volume element are assumed to move with the velocity r¡(f, r).
Integrating g over this volume element and examining its time rate of change leads to

d_ /»r(/ + A/,t) r(/ + A/,T) r(/ + A/,T)


QAdr = f (IV.a.6)
dr J rr(/,r) (f )/*+ r(/,T)
where Reynolds’s transport theorem (Aris 1962) has been used, and the total derivative
is

Substituting for (dQldr)r from equation (III.2) into (IV.a.6), and recalling the diver-
gence theorem, one obtains

d rr(/+A/'T) r(f + A/,t)


QAdr = (gT? - S)A (IV.a.8)

Using the Jacobian of the transformation (IV.a.5), this integral may be evaluated in
terms of the Lagrange coordinate /:
f + Af
QM df = (Qr, - S)A (IV.a.9)
f

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 915

In particular, if we take the limit as A/ -> 0, we obtain the differential equation

showing that the conservation form is also preserved in the Lagrange frame.
We will now use equation (IV.a.9) to find the velocity with which the walls of an
arbitrary volume element must move in order that the time rate of change of the total
mass or energy within the volume element be zero. A select number of these arbitrary
volume elements will later be used as the zones in the finite difference method with the
computational points (mesh points) located at the boundaries of the volume elements.

b) Mass Lagrangian Frame


We begin with equation (II.a.9):

¿ P
+ 1^ r°ß = 0; (II.a.9)
dr V(1 - ß2) r° dr V(1 - ß2)
and identifying Q and S' with

s = pß (IV.b.l)
Q = 2
VO - ß ) ’ V(1 - ß2) ’
we have the correspondence to the conservation form (III.2) so that the results of
§ IVa are applicable.
Choosing = Q, equations (IV.a.l) and (IV.a.5) become

Adr,
VO - ß2)Adr ’ M
i VO - ß2)

df 1 pA
(IV.b.2)
dr J MVO - ß2)
Hence Mis the total relativistic mass of the system and/is the external mass fraction.
Equation (IV.a.9) then becomes
f + Af
(v - ß)A (IV.b.3)
VO - ß2) f
Here ( — MAf) is just the amount of mass contained between r(/ r) and r(f + A/ r),
and will remain constant if we choose the velocity of our zones to be Then the
total derivative is just the material derivative of hydrodynamics :

/_a\ = ( d\ > oí d\
\drjf dr \c WJr \dr]
which may be solved for

(IV.b.5)
(¿H¿)
Substituting this equation into equations (II.b.22) and (II.b.23) gives

, Ml - fi2) /g \ (IV.b.6)
- l±ßßa \dr/, ’

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

916 CHESTER McKEE AND STIRLING COLGATE Vol. 181

and

(IV.b.7)
Hi ‘ (¿),

for the differential operators along the characteristics. Transforming (d/dr)% to the
Lagrange frame by using (dfldr)z from (IV.b.2) yields

which is now an expression for the total derivatives along the forward and backward
characteristics in terms of the Lagrange variables (/, r).
For completeness we include the conservation equations in the mass Lagrangian
frame for plane symmetry. Use of equations (II.a.7) and (III.l) in the Lagrange
conservation form (IV.a.10) gives

d_ ßp + e + pc2 J_ dp
(IV.b.9)
dr P VO - ß2) Mdf'
d (e + ß2p + Pc2[l - VO - ß2)] l_dßp
(IV.b.10)
dr\ PVO - ß2) M df

The third equation is obtained by appealing to the equations which define the Lagrange
frame

(IV.b. 11)

Equating the cross derivatives obtained from (IV.b. 11) leads to

d_ V(1 - ß2) l_dß


(IV.b. 12)
dr M df

Equations (IV.b.9), (IV.b.10), and (IV.b. 12) easily reduce to the nonrelativistic
conservation equations for a mass Lagrangian frame.

c) Energy Lagrangian Frame


We begin with equation (III.l):

d e + ß2p + Pc
2
[\ -Wiß - ß2)]
dr 1 - ß2

+ > I , 0_ (m^

since it is this equation, and not equation (II.a.8), which reduces to the classical
expression for conservation of energy. Using our previous notation, let

e
qr - n =
^-0 + ß2p + Pc2V - V(1 - ^2)]’

P + e + Pc2[l - V(1 - ß2)]


S = ß (IV.c.l)
1 - ß2

American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 917

The energy coordinate and Jacobian are then defined by

e + ß2p + Pc
2
[l - V(1 - ß2)]
Adr,
1 - ß2

e_±lpjr pc2[l - V(1 - ff2)]


M = Í Adr, (IV.c.2)
Jo 1 - ß2

df i e + ß2p + Pc
2
[l - Vil - ß2)] A
(IV.c.3)
ôr~ J 1 - ß2 M

With these definitions equation (IV.a.9) becomes

1 (-MA/) = |e + ßjp ± - V(1 - ^)] ^

_ pj? + e + pc2[I - V(1 -jS2)] , f + àf


ß l _ g2 >A (IV.c.4)
f

In this case ( — MAf) is the total energy of a parcel of fluid contained between r(/, r)
and r(f + A/, r), and will remain constant if we choose the velocity of the boundaries
to be given by

p + e + pc2[l - V(1 - jS2)]


V = (IV.c.5)
ß2p + e + Pc2[l - V(1 - ß2)]

This formula is exactly the same as equation (III.30), so that the velocity for energy
transport in the fluid is the same as the velocity for a strong shock. In the extreme
relativistic limit (e » pc2) this equation becomes


rj = (IV.c.6)
3 + ß‘

for an ideal gas with y = j. Performing a Lorentz transformation on equation (IV.c.5)


to find the relative velocity, denoted by r)ß, of r¡ with respect to the fluid gives

V - ß (IV.c.7)
Iß I -r,ß = ß~ e + pc2[\ - V(1 - j82)]

which agrees with equation (III.26) in the nonrelativistic limit (ß2 -> 1). In the limit
(e » pc2), for an ideal gas, we obtain

ve = (r - i)£ (IV.c.8)

from the above equation. This result was first derived by Johnson and McKee (1971)
for y = f, and ß = \.
The differential operator which follows surfaces of constant energy is

(IV.c.9)

American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

918 CHESTER McKEE AND STIRLING COLGATE Vol. 181

Solving for (S/St)„ and using equation (IV.c.3) yields

-ßp + ßa{e + pc2[l - V(1 - j82)]} A (IV.c.lO)


1 ± ßßa M a)

(IV.c.ll)
(¿)o= +

for the differential operators along the characteristics in the energy Lagrange frame.
The conservation equations in this frame are (cr = 0)

p + e
8t I ß p + e + pc2[\ — V(1 ~ ß )]
2

pV(l - ß2){e + pc2[l - Vd - ß2)])' (IV.c.12)


Mdf ß2P + e + Pc2[l - V(1 - ß2)]

i. 2 pW - ß2)
ôt \ß p + e + Pc2[l — V(1 —
ß2)]

pßpva - ß2)
= -11!. 2 (IV.C.13)
M df \ß p + e + pc2[l - i/(l - ß2)]

1-ß'-
/-
dr \ß2p + e + pc2[l - V(1 - ß2)]

= 1 g f p + e + pc2[l - va - j82)] (IV.C.14)


M df\ß2p + e + pc2[l - V(1 - ß2)]

We conclude this section by noting that we have proved that the velocity of the
energy Lagrange frame is the same as that of a strong shock. This property will
considerably simplify the solution to problems of strong shock waves, since a moving
boundary is replaced by one that is stationary with respect to the coordinates in the
energy frame. This unique feature of the energy frame was noted by Brownell (1970) in
connection with the equations of classical hydrodynamics, and as we have seen, also
holds true in the relativistic case.

V. DIFFERENCE METHOD
The characteristic equations (II.b.18) and (II.b.19) may be written in the form

1 (d\ R , _ß^(d\ p = Tg ßßa (V.l)


1 - ß2 ydr) ± p + e ydr) ± +
r 1 ± ßßa

(V.2)
p + e yar/o p \dTj o

where

(V.3)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 919

is taken along either the forward ( + ), backward ( —), or fluid (0) characteristics. The
velocities along the characteristics (ß+>0,_) will have different expressions depending
on the frame of reference.

ß = 1 — u, l - ß2 = u(2-u). (V.4)

Equation (V.l) then becomes

/ d\ ßß.
— M ± e = (V 5)
w(2 - Ü) (^) ± 7+^ \Tt) ± +7 1 ± Wa ' '

We will therefore solve equations (V.2), and (V.3) together with the ideal gas equation
of state p = (y — l)e, and appropriate boundary conditions.
In the Lagrange frame we take r, u, e, p as dependent variables with (/, r) as the
independent variables. The first step in the finite-difference method is to discretize
these variables by assigning the integer subscript j to number the points in space and
the superscript n to represent the number of time steps taken. For instance, at the
time Tn given by

r" = 2 At*”1'2 = r"-1 + At»-1'2 , t° = 0 , (V.6)


k=l
and at the point /y we have a value of w/, ef, p/1, and
i
r n
i = + 2 Arfc_1/2n , with Arí_1/2n = o” - . (V.7)
k=2
The subscript j has the value 1 when / = 1 (inner boundary), and its maximum value
is / when / = 0 (outer boundary). It will also be necessary to define the average values
of a function. In such a case we will calculate the function of the averages and not the
average of the functions. For example, if the function is G = G(e), then Gj^1i2n will
mean

Gj-1121 = G(^_1/2n) , where ey_1/2n = Ke/1 + ^-in) (V.8)

and not

Gj-U2n = MG^/1) + Gfo-i11)] • (V.9)


Both approximations have second order error when expanded in a Taylor series about
ry-i/a71. However, in experiments with numerically integrating several types of func-
tions it was found that mean values based on equation (V.8) gave more accurate
results when step sizes were large.
If r;0, Wy°, e/, pj0 are given as initial conditions, then /• may be found by using the
second-order-accurate trapezoidal rule on the transformation (IV.a.l) to obtain

¿ = ¿
1V1 i (^^-1/2°, (V.10)
fc = i + l
where

M= , (V.ll)
k—2

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

920 CHESTER McKEE AND STIRLING COLGATE Vol. 181

^/-i/2n = 1 denotes plane symmetry, A^^11 = + r^i1) denotes cylindrical


symmetry, and = ^[(r/)2 + + (ry.!71)2] denotes spherical symmetry.
The Lagrange coordinate/y and M are now fixed and need not be recalculated.
To solve the differential equations, we integrate along a characteristic from a point
where the characteristic originates at time rn to a mesh point at time rn + 1. The
accuracy of the method therefore depends on the manner in which we approximate the
dependent variables between the mesh points at a given time. The lowest-order
accuracy is obtained when one represents the dependent variables between the mesh
points by straight line segments. Higher-order methods are obtained by representing
the dependent variables between mesh points by successively higher-order poly-
nomials. The method was discovered by Hartree (1958), and differs from the usual
method of characteristics in that the locations of the mesh points at the next time
level are defined before one proceeds to integrate along the characteristics.
Figure 2 shows the three characteristics as they might appear in the energy Lagrange
frame. The characteristics in the mass Lagrangian frame have a slightly different
appearance. There the fluid characteristic ß0 is always a vertical line joining the points
(y, ri) and (y, n + 1), while the ß+ and characteristics are symmetric about the ß0
characteristic.
The next step is to define a way to locate the origin of the characteristics j+,j0>j-
at time rn. This is accomplished by using the equations for the characteristics

(V.12)

From equations (IV.b.7) and (IV.b.8) we have

pW - ß2)ßa A_
ßo — 0 , ß± — i (V.13)
1 ±ßßa M

in the mass Lagrangian frame, and equations (IV.c.lO) and (IV.c.ll) give

-ßp± ßa{e + pc2ß2/[l + a/(1 - iS2)]} A


]
o = ßPjf ß± = (V.14)
1 ± ßßa M

for the velocities along the characteristics in the energy Lagrangian frame. Second-
order-accurate approximations to these equations are

fu-fi=-ß+n+1,2^n+1,2> (V.15)
n+1,2
-ßo ^+u*9 (V.16)

-fi = -ß-n+ll2kTn+1i2 . (Y.17)

The velocity along the characteristics is evaluated as the function of the averages
between (y+, n) and (y, « + 1) as previously discussed, except that the area factor is
evaluated in the manner indicated below equation (V.ll). For example, A+n + 112 in
spherical symmetry would be
^+n + i/2 _ ^[(^+1)2 + rj
n+1
rj+n + (rj+n)2]. (V.18)

To find the dependent variables at the base of the characteristics, we must now
choose an interpolating function. Parabolic arcs were used so that the difference
equation will be accurate to second order in space. If steep gradients develop in the
dependent variables, then linear interpolation should be used in this region to introduce
a weak artificial viscosity.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 921

The parabola passing through three unequally spaced points, and centered about / is

e\f) = c/ + j ■ [(/, y" ~ f + (fj+1 -ft)


Jj+i — Jj-i L Jj+i — Jj Jj ~ Jj-i J

Jj+i ~ Jj-i L Jj+i ~ Jj Jj ~ Jj-i J


with similar expressions for u (f), p (f), rn(f). Using equations (V.15), (V.16), and
n n

(V.17) one can obtain en(fj+) = eun and the other dependent variables at the base of
the characteristics. The differential equations (V.5) and (V.2) are therefore
p + e n+l/2
e/l+1 - eun - (ur1 - %+n)
ßau(2 - u)

y-rrrnlT1^"''2- (V 20)
-
n+l/2
,n + 1 _ e » + [ P + 6 (Ujn + 1 - Uj_n)
+
^ [ßau{2- u)
a ß(p + e) n+l/2 Am+1/2, (Y21)
r l - ßßa
(D 4- p\n+l¡2
^Li)o (p/+1-p/„n) = o. (V.22)

Equations (V.20) and (V.21) may be solved as two simultaneous linear equations for
ejn + 1 and t/yn + 1. Using the value obtained for ^7l+1 in equation (Y.22) gives the
density pjn + 1.
The radius at (y, « + 1) is obtained by using a finite-difference analog to the first
integral in equations (IV.b.2). In plane symmetry

\r
^j-112 n +1 —
_ <zr^A/}-l/2
n +1 . (V.23)

In cylindrical symmetry we have the quadratic equation

+ A/v_1/2)2 - ri_12]n+1 + = 0 (V.24)


j —1/2
for Ary_1/2n + ;L. The solution to this equation is
1/2
L\y
r n+1 ■^A/}-l/2 /f„ n + 1 i i(r n + U2 ^A/}-l/2 1 (V.25)
^ j-H2 — 7rr n+1 0-1
y-i/2
In spherical symmetry we must solve the cubic equation

MOV-in+1 + Ari_1/2n+1)3 - (/v^1)3] + ^A/y;+i = 0 (V.26)


</-l/2
for A^_1/2n + 1. Its solution is

A« n+1 Í W + 1A3 _ 3AfA/}-i/+1 ]1/3 (V.27)


^i-l/2 - [(0-1 J 4^_1/2" 2 J 'v-1 n + 1 •

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

922 CHESTER McKEE AND STIRLING COLGATE Vol. 181

Unfortunately this formula requires too many significant figures to give an accurate
answer when Ary_1/2n + 1 « r?-_1n + 1.
Multiplying and dividing equation (V.27) by

(p3 - q)213 + (p3 - q)ll3p + p2 , (V.28)


where

*. u + 1 3MA/y_1/2
P = fj-l and q
^-mn+1 ’
yields the solution

Ary-1/2 n + 1 q (V.29)
Q?3 - q)m + (/?3 - q)ll3p + p2

The radius is then found by the recursion relation


«
r n +1 n + 1 + Ar,_ n + 1 (V.30)
i 1/2
The procedure outlined above is a point iterative scheme. A first guess to the
dependent variables at (j, n + 1) is provided by a linear extrapolation. For example,
Arn + l/2
n + 1 _ e? + o/* - e}n~1), (V.31)
Ar71-112

with similar expressions for ujn+1, Pjn + 1, r/l + 1. One then begins with equation (V.15),
and solves for ejn + 1, ujn + 1, pjn + 1, rjn + 1 using the above algorithm. These values are
now better estimates to the solution of the difference equations. The iteration is
continued until the (k + l)th iterate is within a specified tolerance to the fcth iterate.
The dependent variable e was found to be satisfactory for the tolerance check
K^n + iyc + l _ (ejn + 1f\
< 5 x lO"5 (V.32)
(^n + T

for 1 < y < /. There is a danger in this approach in that the true error may be an
unknown multiple of the relative error given by the above equation. However, when
check problems were solved and compared with analytic solutions (see §VI), no
problems arose in this respect.
After convergence is assured for all mesh points, the time step for the next cycle is
chosen as

At* + 3'2 = Min [Arc, 1.2At” + 1/2, Atp, Are, Aru, At,] , (V.33)

where the Courant time step is

Arc = Min ÍO.95 ^ " J = 2,j] (V.34)


[ P- j-112 J
and

at Mi (V.35)
‘ - ° [|P/0;°2-*P/i ~ i-f]

with similar expressions for Are, Arw, Arr. In practice the time steps which always
controlled the problem run were either Are or At,. Using these time-step controls,
convergence generally occurred after two or three iterations per mesh point.

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 923

The Courant time step in equation (V.34) is a stability condition with 0.95 being a
safety factor. This condition is the result of the fact that the base of the characteristics
must remain within the bounds of the interpolating function (/y_i < / < /y+i). A
more rigorous proof using the Fourier method is given in the Appendix.
Along with the time-step controls, one should also ensure that no dependent
variable changes by more than a factor exp(l) over 5 mesh points. This “rule of
thumb” always seems to give good results when check problems are compared with
analytic solutions. The practical way to find the required zoning is by trial and error
or to use some theoretical estimate appropriate to the problem.
The boundary condition at y = 1 in all problems was ß1n + 1 = 0. This gives u1n+1
= 1 from equation (V.4) and r1n+1 = constant. One then uses the interpolating
functions for the interval fx < f < /3, and the equations for the fluid (0) and backward
( —) characteristics to solve for e1n + 1 and p1n + 1.
The boundary conditions at j = J (outer boundary, / = 0) require more careful
consideration. Here the adiabatic blowoff and shock boundary conditions initially
impose a large discontinuity, which makes interpolation for the dependent variables
near such a discontinuity highly inaccurate. For the adiabatic blowoff these boundary
conditions should be
p(0, r) = 0 , (V.36)

e(0, r) = 0 ; (V.37)

in which case equations (V.13) and (V.14) yield

ßo = ß+= 0 (V.38)
at / = 0. Hence the forward (+) and adiabatic (0) invariants lie along the same
characteristic. These boundary conditions imply, from equation (II.b.29), that a gas
initially at rest with initial energy ^ and at / = 0 will have its leading edge im-
mediately expand into a vacuum with velocity

(V.39)

where
4/V(r-l)
= (V.40)
(^r+ i

Contrasting with this the finite difference method must integrate to the final velocity
along the forward characteristic in a finite length of time. The implication is that the
vacuum blowoff conditions are not applicable to the finite-difference equations;
however, we must still find a way to simulate such problems.
Our compromise was to begin the problem with an initial time step of 0.01 Arc and
initial boundary conditions

P,0 = Pi, (V.41)

ej° (V.42)

with the corresponding adiabatic invariant

(V.43)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

924 CHESTER McKEE AND STIRLING COLGATE Vol. 181

After the initial time step, successive time steps are determined from equation (V.31).
The density at / = 0 is taken as
n+1
Pj = Max (0.985/)/, O.lpjS*1) . (V.44)
The first condition in this equation controls the density change early in time and effects
a smooth integration of the forward characteristic at j = J. The second factor limits
the size of the discontinuity in density near the boundary. This factor was chosen from
numerical experiments as the largest discontinuity which could be permitted at the
boundary and still preserve the forward and adiabatic invariants to within a few
percent. The numerical experiments were performed in both the relativistic and classical
regions with the material initially at rest and having uniform initial density and energy.
In such problems the adiabatic invariant is always preserved, while the forward
invariant remains constant until the reflected wave from / = 1 overtakes the point in
question.
Once the density is given, the energy at / = 0 is determined from
n+l (/>/ + 1)M/ (V.45)
ej
which preserves the adiabatic invariant at the boundary. The quantity w/+1 is
obtained by using the difference equation for the forward invariant and the inter-
polating functions for the interval/j_2 </<// = 0.
Physically, the boundary condition applied to the difference equations can be
thought of as a piston whose mass continually decreases with time. The error therefore
appears as a small energy loss owing to the pdv work done against the piston. In
numerical experiments it was found that the energy lost in such a process is always
much less than the error in energy conservation due to the discretization error inherent
in the difference method.
While the adiabatic blowoff can be calculated in either the mass or energy Lagrange
frame, the strong-shock problem will only be calculated in the energy frame. The
reasons for this choice were discussed in § IV.
The shock is generated by the conditions
u = u(r, 0) , e = e(r, 0) , p = p(r, 0) , (V.46)
for rr < r < S(0), and
u = 1 , e = 0, p = po(r ), (V.47)
for r > 0, and r > S(t), where S(t) locates the position of the shock front.
In the energy Lagrange frame, mesh points are required only in the region described
by equations (V.46). The reason for this is that/is now the external energy fraction;
and since the region ahead of the shock contains no energy, it requires no mesh
points.
To solve for the unknowns at / = 0, we use the jump conditions (III.16)-(III.18),
the definitions (III.27), and the ideal gas equation. (Note that eq. [HI. 19] should be
used instead of [III. 16] when the gas is nonideal.) The jump conditions now become

¿o + [ßßJ(y - l)](gp + ppc2)} (V.48)


us + ußs

PoftVM/ - »)] )
(V.49)
u - us

ß(2 - u)(p + c0)


us = i/jj^l - (V.50)
ß2p + e + pc2ß2/{l + VM2 - «)]} - e0u(2 -«)]

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 925

We now have three equations in the four unknowns w, ws, e, p. The incomplete informa-
tion provided by the jump conditions must be supplemented by additional information
propagated to the shock at Tn + 1 from the fluid behind it. The only characteristic that
is in a position to do this is the forward characteristic, which represents a sound wave
moving to the right. The difference equation (V.20) for the forward characteristic has
the form

0 — e+) + ö!(w — w+) = a2. (V.51)

We will solve the above four equations by the Newton-Raphson method. To this end
we define the functions

G\u, us, e) = (us + ußs)e ^-r (e0 + p0c2) - e0 , (V.52)


y — i

G2(u, us, p) = (u- us)p - V[w(2 -u)], (V.53)

G3(u, us, e, p) = (u - us)^ß2p + e + j + - u)] ~ e°u(2 ~

- ßu(2 - u){p + e0), (V.54)

G4(w, e) = e + a-ji — {a2 + + CI-LU+) . (V.55)

When G1 = G2 = G3 = G4 = 0, the desired solution

u = w/ + 1 = w* , us = (ws)/+1 = ws* ,

^ = ejn + 1 = ^* , p = p/+1 = />* , (V.56)

will have been attained. Expanding the G functions to first order about the approximate
solution uk, usk, ek, pk yields

GHt/*, ws*, e*) = (G1)^ + Aw(Gw1)fc + ^{G^f + = 0, (V.57)

G2(w*, t/s*, p*) = (G2)k + Aw(Gw2)fc + Aws(GWs2)fc + kP{Gp2)k = 0, (V.58)

G3(w*, ws*, *?*, /o*) = (G3)fc + Aw(Gw3)fc + Aus(GUs3)k + A^(Ge3)fc + AP(Gp3)k

= 0, (V.59)

G4(w*, e*) = (G4)fe + Aî/(Gw4)fc + Ae(Ge4)fc = 0 , (V.60)

where G^1 = {^G1!du)UstetP etc. A first guess to the solution (k = 1) is obtained from
a linear extrapolation of the form (V.31) using values from the two previous time
steps. Successive iterates are obtained by solving the above system of equations for the
increments Aw, Aws, Ae, Ap, using Cramer’s rule, and advancing the variables according
to the relations

uk + 1 = uk + Au 9 usk+1 = usk + Aus,

ek+i = eK + &e 9 pK+i = pk + Ap. (V.61)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

926 CHESTER McKEE AND STIRLING COLGATE Vol. 181

The iteration process is stopped when the following conditions,


Aw Aus
€ 6
UF+T — ’ "TTÏ
u — ’

Ae (V.62)

Jc + l — » k+l — " >

where e = 5 x 10"5, are all satisfied.


We have deviated slightly from the standard Newton-Raphson procedure by
assuming al9 a2, u+, e+, as constant, which considerably simplifies the algebra since
their partial derivatives are not present in equation (V.60). These values are updated
after each iteration, and convergence generally occurs after approximately five
iterations.
For shock problems the initial time step is 0.01 rc, as in the adiabatic blowoff. The
value e0 was retained in the jump conditions in order to spread the contact discontinuity
over a few zones. This is accomplished by taking

e00 = e}° (V.63)

with values at successive time steps given by

e0n+1 = 0.975eon. (V.63)

After approximately 130 time steps e0130 and successive values are set to zero. The
extra energy swept in was always less than 0.05 percent of the total energy in the
problem, so that the resulting error is very small.

VI. CHECK PROBLEMS


All problems in this section will have y = f and initial conditions of the form

Pi = p(/> 0) = 1 , = e(f, 0) = fjLC2 O = constant) ,


ui = u(f90)=l=>ß(f90) = 09 0 < r(f9 0) < 1 , (VI.1)

with the exception of the spherical shock in which case r will have the limits

5<r(/,0)<6, (VI.2)

where 0 < / < 1 for all the above dependent variables. The Ar’s were varied by a
fixed ratio so that the zone spacing becomes finer as / = 0 is approached. This is
accomplished by choosing a ratio
Ar'V+1/2 0 _
ia
0 (VI. 3)

for successive zone spacing and then using the following formula:

1 — a
V = 1 - ArQ¡2° — rj, Ary + i/20 — aAry-.^0 (VI.4)
“1) ’

derived from a geometric series to insure that all the Ar’s sum to one.
To obtain high accuracy, one can use double precision and a large number of mesh
points. This approach would rarely be used in everyday problem solving because of the
prohibitive investment in computer time. We therefore present single-precision results

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 927

which use 50 to 60 mesh points and require about 30 to 60 minutes per problem on an
IBM system 360 model 44 computer. Such results will have the advantage of giving
the investigator a reasonable idea of the error he is likely to encounter in practice.
Under these conditions we will critically examine the two problems of interest to us,
namely, the vacuum blowoff, and shock propagation into a density gradient.
In classical hydrodynamics one of the most difficult problems to compute accurately
is the vacuum blowoff. This is due to the discontinuity introduced at / = 0, which
instantaneously accelerates the leading edge of the blowoff to its limiting velocity of
2ßa°/(y — 1) (classical expression), where ßa° is the dimensionless sound speed at
f = T = 0 (r = ct). We will first examine this problem in the nonrelativistic region
where we possess analytic solutions. These solutions for expansion in the direction of
positive r in plane symmetry (a = 0) have the form (Harlow and Amsden 1970)

£ = (!-«) = (£ + ßa°) , (VI. 5)


y + I

y - (VI-6)
ßa
~ y +

P = Piißalßa0)21«-» (VI.7)

where
£ = (r- l)/r (VI-8)

and

ßa° = Vlyir - i)m] » M« 1 • (VI.9)

These expressions apply when £ lies in the interval

2ßa° (VI. 10)


— ßa <> Ï ^ 1

Since the numerical solution uses the independent variables (/, r), we will obtain |
as a function of (/, r). In our problems the total mass is 1 so that the external mass
fraction is
2/?a0/(y-l)
pdr. (vi.ii)

which may be integrated to yield


/ 2 \(y+i)/(y-i>/ v — 1 \ (y+D/(y-i)
(VI. 12)

This equation now gives £(/, r) :

2ßa° 1 - y + i (X)—].
HI, r) = y - 1

If t > 0 and /—> 0, then the substitution of the above equation in (VI.5) gives

2ßa° as/-»O (VI. 14)


ß y - 1

American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

928 CHESTER McKEE AND STIRLING COLGATE Vol. 181

The implication is that only an infinitesimal quantity of mass immediately attains the
limiting velocity. In the numerical method the mass is discretized so that a mesh
point j has a mass

associated with it. In particular we can associate a mass

Am, = ifj-uzM (VI. 15)


with the outermost zone. The velocity of the outer zone then cannot be given by
equation (VI. 14); instead, if this zone is to have the correct momentum, its velocity
should be given by

ßJ=\° ßdf/f df
* fJ ——1/2
1/2 ' Jfj-112
1 2ßa° r r,_r+ju_\ <y-i)/(r+i) df. (VI. 16)
~fj-n2 y' -1 L 2 Wrj
Performing the integration gives the result
! if, .„Vr-D/ir+Dl
2ßf
ßj = y - 1 1 - (y + i)
4y (fe?) J ' <VU7>
If the gas were semi-infinite, ßj would attain the limiting velocity as r oo. However,
our problem has a rigid boundary at / = 1. The first sound wave generated by the
blowoif hits the “wall” at/= 0 in a time r0 = l/ßa0 and is reflected as a forward
characteristic. This infinitesimal disturbance propagates back into the fluid and
indicates where the numerical solution should begin to deviate from the analytic
solution. The location of this characteristic is obtained by integrating the equation for
the forward characteristic (Landau and Lifschitz 1959)
d
(r — 1) = (VI. 18)
dr y-fl y+1
If the integrating factor T~(3~Y)Kv+1) is used, one obtains
(3-y)/(y+l)
(r _ 1) = r + A (VI. 19)
y - 1 (i)
Requiring that r = 0 when r = r0 determines A and yields
r — 2(y-l)/(y+1)
(VI.20)

Setting equation (VI.18) equal to (VI.13) with/ = fj-^ and r0 = l/ßa° gives
(VI.21)
as the time for which begins to deviate from equation (VI. 17). The conclusion is
that we cannot expect the numerical method to attain the limiting velocity. If we use
the time furnished by equation (VI.21) in (VI. 17) as giving a rough estimate of ß7,
we obtain
2ßa° 11 __ Y + 1 Jj
f 112 2(y —l)/(y + l) (VI.22)
y - 1 4y ~

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 929

The velocity calculated by the difference method is always within a few percent of this
value. The finer we zone the region, the closer we come to attaining the limiting
velocity. However, the Courant condition is inversely proportional to A/, so that
unless there is interest in the small mass fractions, it is not advisable to use very fine
zoning because of the longer computing time required.
Figures 3 and 4 show the comparison between the analytic and numerical solutions
in a mass and energy Lagrange frame for a vacuum blowoff with /x = 0.01. The
sound speed is 2 x 109 cm s_1, and the curves at i cï 5 x 10"10 s are when the first
sound wave generated at / = 0 just begins to impinge on the rigid boundary at/ = 1.
Hence the similarity solution is still valid for the whole region, and the agreement
between the numerical and analytic solution is quite good. The velocity ßj calculated
from equation (VI. 17) agrees with the numerical solution to within 1 percent. The
curves at t ~ 10"7 s are after the sound wave has reflected from / = 1, with the
vertical line on the curves indicating its position. Behind the vertical line the numerical
solution deviates from the analytic solution as expected. In figure 4 at ¿ = 7.6 x 10"8 s,
the density in the energy Lagrange frame has developed an oscillation near / = 0.
This oscillation is most likely due to an inconsistency in the treatment of the boundary
condition. If we let p = (y — l)e and e = />€, where e is the specific internal energy,
then in the limit (e 0) we obtain rj = ß from equation (IV.c.5). The boundary
point would therefore move with the fluid as it should. However, as discussed in § V,
we cannot use the vacuum boundary condition, so rj > ß and the outer boundary will
try to separate from the fluid. This difficulty does not arise in the shock problem, since
the boundary conditions are treated correctly. We therefore conclude that blowoffs
should be calculated in a mass Lagrange frame for best results.
In the above case we had an analytic solution to provide an accuracy check. In
problems where no closely related analytic solution or approximate results are avail-
able to guide us, we must rely upon our intuition and the internal consistency of the
calculation. One such self-consistent check is to observe that the difference equations
in a particular frame can automatically conserve either energy or mass but not both
simultaneously. Then depending on the frame of reference, the error made in con-
servation of mass or energy furnishes a measure of the accuracy of the numerical
solution. For example, in the previous problem the energy was conserved to within
0.2 percent in the mass Lagrange frame; and until the oscillation developed in the
energy Lagrange frame, mass was conserved to within 0.17 percent. When the oscilla-
tion began, mass conservation deteriorated until an error of 7 percent was attained,
at which point the calculation was terminated. If poor conservation of a quantity is
obtained in a calculation, then the obvious remedy is to add more mesh points until
the desired conservation is reached. However, if a boundary condition is inconsistent
with a given problem, then no amount of refining of the mesh will remove the basic
difficulty. The adiabatic and forward invariants can also be checked in certain problems.
In the previous example the adiabatic invariant should always remain constant while
the forward invariant remains constant until the reflected wave from / = 1 overtakes
the point in question. In the mass Lagrange frame the forward invariant was constant
to within 1 percent while the error in the adiabatic invariant was 0.008 percent. In the
following examples the appropriate quantities are conserved within 1 percent.
Figures 5 and 6 show the results of moderate and extreme relativistic blowoffs. A
surprising result is obtained in that the density distribution is no longer monotonie.
Furthermore, dp/dr > 0 at / = 1, implying a pressure gradient tending to retard the
expansion. However, we shall now show that these effects are due entirely to viewing
the proper density p in the laboratory frame.
To understand this, let us consider the metric of special relativity for plane-parallel
motion
ds2 = dr2 - dr2 . (VI.23)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

930

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

«"8 cÇ.S^^
^ G
Il ^ °^as
H G'U G
- âi
* G, ö
Ö OÙ >>
O 4> ^ > G.G
VOÆ W O 'G §
Il G ÜÂ G 5
fH C« Jü "O
§.22 o~
oC I
^•2 tsH
il11 ° æ 3 >, «
G
8 O ^ O.G S
<D O Æ ^ 52 Ci
S8 G§ i^.'G'O »h
j-i ü
(U S
5 ts
>.S a a
uZo oS O ^ G Ai
o 0) Æ 4^
taij 7: æ ^ ^
•G ^ ü.G
!.s •- d ^ G
tl G
S2 -1 ||rS
S-2-1
2^ ^ G £\ti
2« ^G ^S g2 § Ö d
.ë.2
<D ^ Il S
al 3 á :£ëê
ce -i-> 5P
¿¡.52^ SË <M ’ Sfl
GG
S? ^ o «^5.2 2
çS ’S ’>
|S2 s> I 8
CÖ ^ Oh 2-§' a
sSlí/
<«¿5 ¿5 5b^ o o -S
G OCL G ^ M
J
h 2x g
Sc- O ^
^ en 8
G« *_» m O
•Son QQ-S
, <l) o G
G I ^ 11
G
S^o§ -
|5l £ O
¿ ’w —-( x)
_5QQ_ S ’G 3
Gh ^ 'ien 22 on ÖG3
G .24 rj çh
G H G? ^_ C> en G
’G il G ^ G
>> .G H 13 V .2
G euö i! «2 aG
T3
'G’ô ’ i £ x S
T3 G2 ^3 Ils® S
GG i.
G ^S il ^2
O (L>^ 8 II 22 2
O TJ G G<ü qIg on *Ë G
"7 «_ô W
§ 1
gill's
111 W •^11 .2Hî3
«r Ö on *Ë
. » (3D ^ O^’G
^
0^-G 2 8 O 2
Ph II G II dJ II G
„*4-1
Q- O ^ ^ ^ «FH

931

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

932 CHESTER McKEE AND STIRLING COLGATE Vol. 181

Following the approach of Taub (1957), and using the defining relations for a mass
Lagrange frame, (dr/dT)f — ß, dfjdr = — Fp/M, together with (dr/df)z = ll(df/dr)z
and equation (VI.23), leads to

2MßT df
1 + (VI.24)
p dr

If we hold / constant (follow particles), and integrate, then equation (VI.24) becomes

(VI.25)

where ds = dr0 has been identified as the invariant interval of proper time. Equation
(VI.25) was integrated for several zones, and the results appear in figure 6. The
density distribution now appears monotonie in terms of proper time, and furthermore
dp/dr < 0. Therefore, it is only in the proper frame that our intuition based on non-
relativistic hydrodynamics is applicable. The “density hump” is only an artifact of the
relativistic time transformation to the laboratory frame. A similarity solution for an
extreme relativistic blowoff can be derived. Following Mathews (1971), we introduce
the similarity variable £ = (r — l)/r into equations (Il.b.l), (II.b.3), and (II.b.5) with
a = 0. Our approach is slightly different in that we use the equation of state p =
(y — l)e with y a constant. We then obtain

(VI.26)
1
where

y - 1 (VI.27)
gV) =
1 + 0’
and

V(1 + 0) - 1 4- = /I + ftWcy-D/s
(VL28)
a/(1 + 0) + 1 0o \1 — ß)
Equation (VI.28) is the forward Riemann invariant, in which we have used the
approximation 0O « 1 for the initial condition. Using p = />o(0o/0)1/(y"1) in equation
(IV.b.2) together with the above relations in the limit ß-> l gives

MÍ. — 0 Cl + V(y - i)]/(y -1>2 - 2/V(y - D [(1 + 0)1'2 + - g2) + g26]


gd&.
PT ° i (y _ 1)(1 _ +
•JQ
(VL29)

In the limit 0 « 1, we obtain

[1 - V(y - l)
(VI.30)
2-\/(y - 1) Po^J

whereas in the limit 0 » 1, we have


22/V(y"1)“1 ñ 1/2 - 1/V(y -1) Mi 2/V(y-D
P = Po (VI.31)
V(y-1) 0 Por

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 933


which can be obtained from equation (VI.28). Figure 6 shows the comparison of the
analytic solution with the numerical in the region that is uneifected by the reflection
from the wall, and the agreement is seen to be excellent.
We now turn to the question of shock propagation in a density gradient of the form

Po(r) = exp[-(r-ro)], (VI.32)

where we have used r0 = 1 in plane symmetry and r0 = 6 in the spherical shock wave.
To propagate a shock into the small mass fraction (large r) requires that the mass
fraction of the zones near the front of the shock to be of the order e~\ The energy
near the shock front is also of the order e~r (Grover and Hardy 1966), so that the
same fine zoning in energy fraction is also required behind the shock front. The
required zoning can be achieved either by initial fine zoning near the shock front or by
inserting the zones when needed as the calculation proceeds. Initially, zoning the
region very finely means a small Courant time step early in time, which can waste a
considerable amount of computer time. Since these zones are only needed later in time
when r becomes large, we chose to insert the zones by dividing the interval J — \ <
j < J in half (with respect to energy fraction) whenever

^ 0.9 , (VI.33)
Pj-1
thus adding another mesh point. The radius of the new zone is determined using
equation (V.23) in the energy frame, with the remaining dependent variables obtained
from parabolic interpolation using the values at the mesh point 7 — 2, / — 1, J
before zone division.
All the shock examples begin with 50 mesh points and finish with approximately
70 points. The ratio of successive Ar^i^’s was a = 0.94, except for the case /x =
1000c2 where a = 0.9 was used.
In the nonrelativistic case the similarity solution of Grover and Hardy predicts
(for y = t)

^ = A exp (0.1765) (VI.34)

where S(t) is the position of the shock front measured from r0.
In spherical geometry this law is modified to

^ exp (0.1765) (VI.35)

(Grover and Hardy), where B and v must be determined from the numerical solution.
For the extreme relativistic case {e » pc2, ß -> 1) the appropriate formula can be
derived by using the approximate Riemann invariant (+) and the strong-shock
relations (III.21) and (III.23) to obtain (Johnson and McKee 1971)

= D exp VCr - l) (VI.36)


V(1 - |82) y + Wiv - !)
For y = 435

V(r - i) 31;2 (VI.37)


y + 2V(y - 1) 2(2 + 31/2) '

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

934 CHESTER McKEE AND STIRLING COLGATE Vol. 181

The results for ^ = 0.01, 1, and 1000 are shown in figure 7, which displays the
excellent agreement between the analytic and numerical solutions. For /z = 0.01, the
plane shock begins nonrelativistically with the slope 0.177 and becomes relativistic
with the correct slope. In the spherical shock the values B and v are determined at
S = 13.63 as

£ = 1.9 x 109 , v = -0.2483 (VI.38)

from the numerical solution. The same law appears to also hold in the relativistic
region, if exp (0.232S) is used in equation (VI.35). The value of B and v determined at
*S = 34.51 gives

B = 8.63 x 108 , v = -0.3973 (VI.39)

for the relativistic portion. The vertical bar in figure 7 indicates where the two
solutions match. As the spherical shock propagates to larger radius the power should
asymptotically approach the plane-parallel value of 0.232. At = 36, the value is
0.223 which is within 4 percent of the plane-parallel value.
Figure 8 shows the agreement between the asymptotic solution of Grover and
Hardy and the numerical solution behind the shock as a function of the distance

ç = R-S = r-r0-S (VI.40)

behind the shock.

S= r-r0 (cm)
Fig. 7.—Similarity profiles of relativistic and nonrelativistic shock waves into an exponential
density distribution, with = numerical solution and ... = spherical solution of Grover
and Hardy. The slope calculated by the’numerical solutions is correct to three significant figures.
The top three curves are plane-parallel shocks, while the bottom curve is a spherical shock. The
vertical bar is a match point for the corrected Grover-Hardy solution in the relativistic region with
the interpretation

J? = ßi(i - .

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 935

Fig. 8.—Numerical solution compared to the asymptotic solution of Grover and Hardy for the
density, pressure, and velocity distributions behind a strong classical shock into a density gradient
The parameters are ^ = 0.01c2, = 1 with • • • = numerical solution and = analytic
solution. The good agreement supports the zone-dividing procedure used in the energy Lagrange
frame.

APPENDIX

STABILITY ANALYSIS

The system of equations (V.20)-(V.22) together with the interpolating function for
equal spacing have the form

<?/l+1 - eun + a1{ujn+1 - uun) = a2, (Al)

e/1*1 - ej.n + ¿>1(w/,+1 - u,_n) = b2, (A2)

er1 - elon + ¿(pr1 - P/0n) = 0 » (A3)

en(f) = e/1 + (/-/,) ej±


^2^^ + (L
- ^— 2
a"2 + C/
—> (A4)
A/:
Uj+1n - UJ.1n (f-fj)2 u, + 1n - lu/1 + Uj-in
un(f) = + (f-fi) + (A5)
2A/ A/:

Pn(f) = P/ + (/-/,) + {I
^^ - 2
lp +
— ’ (A6)

where fj < f < fj+1. Also required are the equations (V.15)-(V.17) for the character-
istics

//+ ~ fj — —ß+kT 9 (A7)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

936 CHESTER McKEE AND STIRLING COLGATE Vol. 181

-ß-Ar, (A8)

-ßoAr. (A9)
If a roundoff error is introduced, the above system of equations may amplify this
error so as to eventually swamp the desired solution. The difference method would
then be termed unstable. To demonstrate that the above system of equations is con-
ditionally stable, we will use the Fourier method of von Neumann. To perform the
analysis we must assume al9 a2, bl9 b29 c, ß+, ß-, ß0 as constant. This linearizes the
system of equations. The objection may be raised that the equations are inherently
nonlinear, and consequently such an analysis will not really settle the question of
stability. However, one may argue in a heuristic manner that the conditions required
by the linear analysis, while not sufficient, are necessary for stability, and indeed the
practical experience of many investigators supports this assertion.
If the numerical solution vector is
Ui
Nn(j) = = S«(j) + En(j), (A10)

LPjnA
where S(j, n) is the true solution without roundoff error, and E{j, ri) is the error
vector with components
* n'

En(J) = E" (All)


71
Ep ^
then the substitution of equation (A10) into the system of equations (A1)-(A9) gives

Een+1(j) - Een(j+) + a[Eun+1(j) - Eun(j+)] = 0, (A12)

Een+Hj) - Een(j-) + b[Eun+Kj) - £/(;_)] = 0 , (A13)

Een+1(j) - Ee%j0) + c[£/+1(j) - E0\j0)] = 0, (A14)


where a = au b = bx, and a2, b2 have dropped out of the equations along with
S(j, n), since together they exactly satisfy the difference equation. The interpolating
functions, with the substitution (A7)-(A9), now become

En(j+,-,o) = E%j) - iCT+,_,0[£"(y + 1) - En(j - 1)]

+ ia+,_,02[£"0' + 1) - 2£nO) + E\j - 1)], (A15)


where
a+>_>0 = ^+>.>0(Ar/A/). (A16)

We will assume periodic boundary conditions at the endpoints of the interval


0 < y < /, where JAf = Irr and that the error can be expressed as a Fourier series.
Since the equations are linear, we need only examine a single component of the form

£»(;•) = V\k) exp lik(jlXf)) = exp [ik(jt±f)\. (A17)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

No. 3, 1973 RELATIVISTIC SHOCK HYDRODYNAMICS 937

The essence of the stability analysis is to determine whether the Fourier coefficients
amplify, remain constant, or decay with time.
Substituting equation (A 17) into (A 15) gives

EnU+,-,o) = exp [ik(jAf)], (A18)

where

^+,-,0 = 1- icr+,-,o sin kAf- <7+>_,o2(1 - cosÆA/). (A19)

Equations (A12)-(A14) may be solved by Cramer’s rule to give


ß n+n
n+1 1
c(a — b)
.v+1.

acEunU+) - bcEun(j.) + cEen(J+) - cEen(j^)


—abcEun(j+) + abcEun(j_) - bcEen{j+) + acEen(j_)
x
abEun(j+) - abEun(j_) + abEun(j+) - aEen(j_) + (a - b)Een(j0)
+ c(a - b)EpnUo).
(A20)
Substitution of equation (A17) into (A20) yields
rVn + l-l
u
1
ven+1
c(a — b)
L'yp n + 1 J
n
c(af¿+ - bii-) c(fM+ - /i_) 0 v
ru '
— abc(¡jb+ — /x_) —c(bfjb+ — a¡iJ) 0 vs
_aè(/x+ - /i_) è/x+ - + (a - ô)/x0 c(a — ô)/i0J
(A21)

The eigenvalues of the above matrix satisfy the equation

\cifjb+ — hfl- byb+ — ap. 0¿+ - IX-J


Oo - A
) — ab = 0, (A22)
L a — b a — b a — b

which has the solution

K = /^+ » Ao = P ^3 — H'O (A23)

If the Fourier coefficients are expressed in terms of their eigenvectors WS, W2n, fV3n,
one obtains an equation of the form
M7
rr n + 1 AmIFmn(no sum on m) . (A24)
m
In order that the eigenvectors not amplify with time we require

w < i (A25)

© American Astronomical Society • Provided by the NASA Astrophysics Data System


197 3ApJ. . .181. .903M

938 CHESTER McKEE AND STIRLING COLGATE

which from equation (A 19) gives

Max(|or+j, |<7_|, |(j0|) < 1 , (A26)

as the condition for stability, in agreement with equation (V.34) which was obtained
by an intuitive argument.
REFERENCES
Aris, R. 1962, Vectors, Tensors, and the Basic Equations of Fluid Mechanics (Englewood Cliffs,
N.J.: Prentice-Hall).
Brownell, D. H. 1970, Report No. 3SR-147, Systems, Science, and Software, P.O. Box 1620,
La Jolla, CA, 92037.
Colgate, S. A., and Johnson, M. H. 1960, Phys. Rev. Letters, 5, 235.
Colgate, S. A., and White, R. H. 1966, Ap. J., 143, 626.
Courant, R., and Hilbert, D. 1962, Methods of Mathematical Physics, Vol. 2 (Wiley Interscience,
New York).
Eltgroth, P. G. 191 \, Phys. Fluids, 14, No. 12, 2631.
Grover, R., and Hardy, J. W. 1966, Ap. J., 143, 48.
Harlow, H. F., and Amsden, A. A. 1970, Fluid Dynamics: An Introductory Text (Springfield, Va.:
Clearing House for Federal Scientific and Technical Information).
Hartree, D. R. 1958, Numerical Analysis (2d ed.: London and New York: Oxford University
Press).
Johnson, M. H., and McKee, C. F. 1971, Phys. Rev. D, 3, 858.
Landau, L. D., and Lifschitz, E. M. 1959, Fluid Mechanics (Reading, Mass.: Addison-Wesley
Publishing Co.).
Mathews, W. G. 1971, Ap. J., 169, 147.
May, M., and White, R. H. 1967, Methods in Computational Physics, Vol. 1 (New York: Academic
Press)
Taub, A.’ H. 1957, Phys. Rev., 107, No. 3, 884-900.
Woods, L. C. 1971, private communication.

© American Astronomical Society • Provided by the NASA Astrophysics Data System

You might also like