STABILITY OF THE FORWARD/REVERSE-SHOCK SYSTEM FORMED BY THE IMPACT OF A

RELATIVISTIC FIREBALL ON AN AMBIENT MEDIUM
Xiaohu Wang and Abraham Loeb
Astronomy Department, Harvard University, 60 Garden Street, Cambridge, MA02138; xwang@cfa.harvard.edu, aloeb@cfa.harvard.edu
and
Eli Waxman
Department of Condensed Matter Physics, Weizmann Institute, Rehovot 76100, Israel; waxman@wicc.weizmann.ac.il
Received 2001 September 14; accepted 2001 December 10
ABSTRACT
We analyze the stability of a relativistic double (forward/reverse)-shock system that forms when the fire-
ball of a gamma-ray burst (GRB) impacts on the surrounding medium. We find this shock system to be stable
to linear global perturbations for either a uniform or a wind (r
À2
) density profile of the ambient medium. For
the wind case, we calculate analytically the frequencies of the normal modes that could modulate the early
short-term variability of GRB afterglows. We find that perturbations in the double-shock system could
induce damped oscillatory fluctuations in the observed flux on short timescales during the early phase of an
afterglow.
Subject headings: gamma rays: bursts — shock waves
1. INTRODUCTION
Gamma-ray bursts (GRBs) and their afterglows are most
naturally described by the relativistic ‘‘ fireball ’’ model (see,
e.g., Paczyn´ ski & Rhoads 1993; Katz 1994; Me´sza´ros &
Rees 1993, 1997; Waxman 1997a, 1997b; Sari, Piran, &Nar-
ayan 1998). In this model, a compact source releases a large
amount of energy over a short time and produces a fireball
that expands relativistically as a thin shell. When the shell
encounters the circumburst medium, two shocks are
formed: a forward shock that propagates into the circum-
burst medium and accelerates it, and a reverse shock that
propagates into the relativistic shell and decelerates it (Rees
& Me´sza´ros 1992; Katz 1994; Me´sza´ros, Rees, & Papatha-
nassiou 1994; Sari & Piran 1995; Sari, Narayan, & Piran
1996). Later on, after a significant mass of circumburst
medium is accumulated, the shell approaches a self-similar
behavior, as originally described by Blandford & McKee
(1976), in which there is only one forward shock propagat-
ing into the circumburst medium. The circumburst medium
could be either the interstellar medium (ISM) or a progeni-
tor wind.
The stability of the Blandford-McKee (1976) solution has
been demonstrated recently by Gruzinov (2000). Here we
analyze the stability of a forward/reverse relativistic shock
system. This double-shock system exists during an impor-
tant phase in the evolution of GRBs, and its stability has
observational consequences. In particular, oscillations or
instabilities could translate to specific patterns of temporal
variability in the light curves of GRB afterglows.
In our linear perturbation analysis, we generalize the
‘‘ thin-shell ’’ method first introduced by Vishniac (1983) in
the nonrelativistic regime. This method simplifies the equa-
tions describing the stability of a spherical shock when the
wavelength of the perturbation is much larger than the
thickness of the shocked shell. In our relativistic treatment,
we focus on global perturbations for which the wavelength
is much larger than the thickness of the forward/reverse-
shock system. We consider the regime of GRB parameters
in which the reverse shock is relativistic (although in reality
it may also be nonrelativistic). In x 2 we derive the perturba-
tion equations for the forward/reverse-shock system. In x 3
we show the analytical results for the wind case and the
numerical results for both the wind and ISM cases. Finally,
in x 4 we summarize our main conclusions.
2. LINEAR PERTURBATION EQUATIONS
The interaction between a relativistically expanding shell
and the circumburst medium results in a double-shock sys-
tem, as shown in Figure 1. The system includes four distinct
regions: the circumburst medium (region 1) and the shocked
circumburst medium (region 2) are separated by the for-
ward shock (shock 1), while the shocked material in the shell
(region 3) and the unshocked material in the shell (region 4)
are separated by the reverse shock (shock 2). Regions 2 and
3 are separated by a contact discontinuity. Our analysis is
done in the GRB source frame, where the circumburst
medium is at rest. We use a spherical coordinate system
whose origin is located at the center of the explosion. The
radii of the contact discontinuity, shock 1, and shock 2 are
denoted by R
0
, R
1
, and R
2
, respectively. We refer to the
combination of regions 2 and 3 as the layer whose stability
we consider. Similarly to Vishniac (1983), we make the thin-
shell approximation, i.e., we assume that
R
1
ÀR
2
R
0
< kðR
1
ÀR
2
Þ51 . ð1Þ
where k is the wavenumber of the perturbations. Note that
although shock 1 is relativistic, shock 2 could be either rela-
tivistic or nonrelativistic. In this paper, we only consider the
situation in which shock 2 is relativistic.
The equations of motion for an ideal relativistic fluid are
0
0t
ð¸,Þ þ
D
x ð¸,uÞ ¼ 0 . ð2Þ
¸
2
c
2
ðe þpÞ
0u
0t
þðu x
D
Þu
_ _
þ
D
p þ
u
c
2
0p
0t
_ _
¼ 0 . ð3Þ
The Astrophysical Journal, 568:830–844, 2002 April 1
#2002. The American Astronomical Society. All rights reserved. Printed in U.S.A.
830
where ,, u, e, p, and ¸ are the fluid density, velocity, energy
density, pressure, and Lorentz factor, respectively. We
define the surface density o, bulk radial velocity V
r
, and
average tangential velocity V
T
of region 2 as
oð0. cÞ ¼ R
À2
0
_
R
1
R
0
¸,r
2
dr . ð4Þ
V
r
ð0. cÞ ¼ ðoR
2
0
Þ
À1
_
R
1
R
0
¸,u
r
r
2
dr . ð5Þ
V
T
ð0. cÞ ¼ ðoR
2
0
Þ
À1
_
R
1
R
0
¸,u
T
r
2
dr . ð6Þ
The time evolution of these variables can be obtained by
integrating equations (2) and (3) across region 2, using the
boundary conditions at shock 1 and at the contact disconti-
nuity, and neglecting terms of higher order in ðR
1
ÀR
2
Þ,R
0
and kðR
1
ÀR
2
Þ. We get
0
t
o ¼ À2
V
r
R
0
o þ,
1
c Ào
D
T
x V
T
. ð7Þ
0
t
¸ðR
0
Þ ¼ À2o
À1
¸ðR
0
Þ,
1
c þ
3
3,4
2
1,2
o
À1
p
3,4
ðR
0
Þ,
1,4
1
¸
1,2
ðR
0
Þc
1,2
þ
¸ðR
0
Þ
2c
V
r
D
T
x V
T
. ð8Þ
0
t
V
T
¼
1
3
o
À1
c
2
,
1
À3
3,4
2
1,2
p
3,4
ðR
0
Þ,
1,4
1
¸
3,2
ðR
0
Þc
3,2
_ _
D
T
R
0
Ào
À1
,
1
cV
T
À
V
r
R
0
V
T
Ào
À1
c
2

2
ðR
0
Þ
D
T
o
þ
c
2

3
ðR
0
Þ
D
T
¸ðR
0
Þ . ð9Þ
where ,
1
is the density of the unshocked circumburst
medium just in front of shock 1, ¸ðR
0
Þ and pðR
0
Þ are the
Lorentz factor and pressure at the contact discontinuity,
and
D
T
denotes the tangential derivatives. In deriving the
above equations, we also assumed that the radial velocities
are dominated by the bulk motion of regions 2 and 3
(denoted hereafter as the ‘‘ shock layer ’’), so that
_
RR
0
% V
r
.
The full derivation of the above equations is given in the
Appendix.
From equations (7) and (8), we obtain the unperturbed
equations
0
t
o
ð0Þ
¼ À2
v
c
R
ð0Þ
0
o
ð0Þ
þ,
1
c . ð10Þ
0
t
¸
c
¼ À2 o
ð0Þ
_ _
À1
¸
c
,
1
c þ
3
3,4
2
1,2
o
ð0Þ
_ _
À1
p
3,4
c
,
1,4
1
¸
1,2
c
c
1,2
. ð11Þ
where v
c
¼
_
RR
ð0Þ
0
¼ V
ð0Þ
r
and ¸
c
¼ ¸
ð0Þ
ðR
0
Þ ¼ 1, 1 Àv
2
c
, ð
c
2
Þ
1,2
are the velocity and Lorentz factor of the unperturbed
contact discontinuity, p
c
¼ p
ð0Þ
ðR
0
Þ is the unperturbed pres-
sure at the contact discontinuity, and we use a superscript
(0) to denote unperturbed values.
For the shocked shell in region 3, we define the surface
density, bulk radial velocity, and average tangential velocity
to be
o
3
ð0. cÞ ¼ R
À2
0
_
R
0
R
2
¸,r
2
dr . ð12Þ
V
r3
ð0. cÞ ¼ ðo
3
R
2
0
Þ
À1
_
R
0
R
2
¸,u
r
r
2
dr . ð13Þ
V
T3
ð0. cÞ ¼ ðo
3
R
2
0
Þ
À1
_
R
0
R
2
¸,u
T
r
2
dr . ð14Þ
where a subscript 3 denotes quantities in region 3. The time
derivatives of the above variables can be derived similarly to
those in region 2,
0
t
o
3
¼ À2
V
r
R
0
o
3
þ
¸
4
¸
2
ðR
0
Þ
,
4
c Ào
3
D
T
x V
T3
. ð15Þ
0
t
¸ðR
0
Þ ¼o
À1
3
¸
4
¸ðR
0
Þ
,
4
c À3
3,4
o
À1
3
p
3,4
ðR
0
Þ,
1,4
4
¸
1,2
ðR
0
Þ
¸
1,2
4
c
1,2
þ
¸ðR
0
Þ
2c
V
r
D
T
x V
T3
. ð16Þ
0
t
V
T3
¼
1
3
o
À1
3
c
2
_
2 3
3,4
_ _
p
3,4
ðR
0
Þ,
1,4
4
¸
1,2
4
¸
1,2
ðR
0
Þc
3,2
À
¸
4

2
ðR
0
Þ
,
4
_
D
T
R
0
Ào
À1
3
¸
4
¸
2
ðR
0
Þ
,
4
cV
T3
À
V
r
R
0
V
T3
À
1

2
ðR
0
Þ
0
t
,
4
,
4
_ _
V
T3
Ào
À1
3
c
2

2
ðR
0
Þ
D
T
o
3
þ
c
2

3
ðR
0
Þ
D
T
¸ðR
0
Þ . ð17Þ
where ,
4
and ¸
4
are the density and Lorentz factor of the
unshocked shell (region 4) just in front of shock 2. We have
also used the relation V
r3
% V
r
, as appropriate in the thin-
shell approximation.
The unperturbed equations for region 3 are
0
t
o
ð0Þ
3
¼ À2
v
c
R
ð0Þ
0
o
ð0Þ
3
þ
¸
4
¸
2
c
,
4
c . ð18Þ
R R R
region 2 region 3 region 4
shocked
shell
shocked shell
circumburst
medium
contact
shock 1 shock 2 discontinuity
2 0 1
circumburst
medium
region 1
Fig. 1.—Structure of the forward/reverse-shock system
SHOCK STABILITY FROM IMPACT OF FIREBALL 831
0
t
¸
c
¼ o
ð0Þ
3
_ _
À1
¸
4
¸
c
,
4
c À3
3,4
o
ð0Þ
3
_ _
À1
p
3,4
c
,
1,4
4
¸
1,2
c
¸
1,2
4
c
1,2
. ð19Þ
Equations (7), (8), (9), (15), (16), and (17) make a com-
plete set of equations for the evolution of the forward/
reverse-shock system. The perturbation equations depend
on the density profile of the circumburst medium. In the fol-
lowing subsections, we derive the perturbation equations
for a uniform medium (such as the ISM) and for a progeni-
tor wind.
2.1. Uniform Medium (ISM)
For a uniform circumburst medium, ,
1
¼ const. We
define the perturbation variables c o,o
ð0Þ
À1, c
3

o
3
,o
ð0Þ
3
À1, DR R
0
ÀR
ð0Þ
0
, and D
p
pðR
0
Þ,p
ð0Þ
ðR
0
Þ À1.
From equations (7), (8), and (9), we obtain the following
perturbation equations:
0
t
c ¼ À
2
R
ð0Þ
0
0
t
DR þ2
v
c
R
ð0Þ
0
_ _
2
DR À o
ð0Þ
_ _
À1
,
1
cc
À
D
T
x V
T
. ð20Þ
0
2
t
DR ¼ 2 o
ð0Þ
_ _
À1
,
1
c
2
¸
2
c
À
3
3,4
2
1,2
o
ð0Þ
_ _
À1
p
3,4
c
,
1,4
1
c
1,2
¸
7,2
c
_ _
c
þ
_
4 o
ð0Þ
_ _
À1
,
1
c
À
7 3
3,4
ð Þ
2
3,2
o
ð0Þ
_ _
À1
p
3,4
c
,
1,4
1
¸
3,2
c
c
1,2
_
0
t
DR
þ
3
7,4
2
5,2
o
ð0Þ
_ _
À1
p
3,4
c
,
1,4
1
c
1,2
¸
7,2
c
D
p
þ
v
c

2
c
D
T
x V
T
.
ð21Þ
0
t
V
T
¼
1
3
o
ð0Þ
_ _
À1
c
2
,
1
À3
3,4
2
1,2
_ _
p
3,4
c
,
1,4
1
¸
3,2
c
c
3,2
_ _
D
T
DR
À o
ð0Þ
_ _
À1
,
1
cV
T
À
v
c
R
ð0Þ
0
V
T
À
c
2

2
c
D
T
c
þ
c
3
D
T
ð0
t
DRÞ . ð22Þ
Assuming that ¸
4
is a constant and that there is no shell
spreading, we get ,
4
/ R
À2
. Using this scaling, we derive the
following perturbation equations from equations (15), (16),
and (17):
0
t
c
3
¼ À2 o
ð0Þ
3
_ _
À1
¸
4
,
4
0
t
DR
þ 2
v
c
R
ð0Þ
0
_ _
2
À2 o
ð0Þ
3
_ _
À1
¸
4
¸
2
c
,
4
c
R
ð0Þ
0
_
¸
_
_
¸
_DR
À o
ð0Þ
3
_ _
À1
¸
4
¸
2
c
,
4
cc
3
À
D
T
x V
T3
. ð23Þ
0
2
t
DR ¼
_
À o
ð0Þ
3
_ _
À1
¸
4
¸
4
c
,
4
c
2
þ3
3,4
o
ð0Þ
3
_ _
À1
p
3,4
c
,
1,4
4
c
1,2
¸
1,2
4
¸
5,2
c
_
c
3
þ
_
À4 o
ð0Þ
3
_ _
À1
¸
4
¸
2
c
,
4
c
þ
5 3
3,4
_ _
2
o
ð0Þ
3
_ _
À1
p
3,4
c
,
1,4
4
¸
1,2
4
¸
1,2
c
c
1,2
_
0
t
DR
þ
_
À2 o
ð0Þ
3
_ _
À1
¸
4
¸
4
c
,
4
c
2
R
ð0Þ
0
þ
3
3,4
2
o
ð0Þ
3
_ _
À1
p
3,4
c
,
1,4
4
c
1,2
¸
1,2
4
¸
5,2
c
1
R
ð0Þ
0
_
DR
À
3
7,4
4
o
ð0Þ
3
_ _
À1
p
3,4
c
,
1,4
4
c
1,2
¸
1,2
4
¸
5,2
c
D
p
þ
v
c

2
c
D
T
x V
T3
.
ð24Þ
0
t
V
T3
¼
1
3
o
ð0Þ
3
_ _
À1
c
2
 2 3
3,4
_ _
p
3,4
c
,
1,4
4
¸
1,2
4
¸
1,2
c
c
3,2
À
¸
4

2
c
,
4
_ _
D
T
DR
À o
ð0Þ
3
_ _
À1
¸
4
¸
2
c
,
4
cV
T3
À
v
c
R
ð0Þ
0
V
T3
À
1

2
c
0
t
,
4
,
4
_ _
V
T3
À
c
2

2
c
D
T
c
3
þ
c
3
D
T
ð0
t
DRÞ .
ð25Þ
In total, we have six perturbation equations (20)–(25)in
six variables: c, c
3
, DR, V
T
, V
T3
, and D
p
. In order to solve
these equations, we first need to find the unperturbed
values o
ð0Þ
, o
ð0Þ
3
, ¸
c
, and p
c
from equations (10), (11), (18),
and (19). The time dependence of ¸
c
has been derived by
Sari & Piran (1995) and Sari et al. (1996). When both the
forward shock and the reverse shock are ultrarelativistic
and strong,
¸
c
/ ¸
1,2
4
f
1,4
. ð26Þ
where f ¼ ,
4
,,
1
. For ,
1
¼ const, ¸
4
¼ const, and ,
4
/ R
À2
,
we get
¸
c
/ R
À1,2
/ t
À1,2
. ð27Þ
With v
c
% c and R
ð0Þ
0
% ct, equation (10) yields
o
ð0Þ
%
1
3
,
1
ct . ð28Þ
For ,
4
/ t
À2
and ¸
c
/ t
À1,2
, equation (18) gives
o
ð0Þ
3
%
1
2
¸
4
¸
2
c
,
4
ct ¼ const . ð29Þ
By substituting equation (28) into equation (11), we find
p
c
%
11
4,3
3
7,3
2
2,3
¸
2
c
,
1
c
2
¼ 1.187¸
2
c
,
1
c
2
<
4
3
¸
2
c
,
1
c
2
. ð30Þ
832 WANG, LOEB, & WAXMAN Vol. 568
where ð4,3Þ¸
2
c
,
1
c
2
is the pressure just behind the forward
shock. Similarly, by substituting equation (29) into equation
(19), we get
p
c
%
5
4,3
4
4,3
3
¸
2
4
¸
2
c
,
4
c
2
¼ 0.449
¸
2
4
¸
2
c
,
4
c
2

4
3
" ¸
3
¸
3
2
,
4
c
2
. ð31Þ
where " ¸
3
¸
3
% ¸
4
,ð2¸
c
Þ is the Lorentz factor of the shocked
shell (region 3) with respect to the unshocked shell (region
4) and ð4,3Þ " ¸
3
¸
3
2
,
4
c
2
is the pressure just behind the reverse
shock. The pressure difference between the two sides of the
layer causes it to decelerate. By combining equations (30)
and (31), we find
¸
c
% 0.784¸
1,2
4
ð,
4
,,
1
Þ
1,4
. ð32Þ
Substitution of the values of o
ð0Þ
, o
ð0Þ
3
, and p
c
into the per-
turbation equations (20)–(25) yields
0
t
c ¼ À
2
ct
0
t
DR þ
2
ct
2
DR À
3
t
c À
D
T
x V
T
. ð33Þ
0
2
t
DR ¼
1
2
c
¸
2
c
t
c À
29
4
1
t
0
t
DR þ
33
8
c
¸
2
c
t
D
p
þ
c

2
c
D
T
x V
T
.
ð34Þ
0
t
V
T
¼ À
8
3
c
t
D
T
DR À
4
t
V
T
À
c
2

2
c
D
T
c þ
c
3
D
T
ð0
t
DRÞ .
ð35Þ
0
t
c
3
¼ À4
¸
2
c
ct
0
t
DR À
2
ct
2
DR À
2
t
c
3
À
D
T
x V
T3
. ð36Þ
0
2
t
DR ¼
1
2
c
¸
2
c
t
c
3
À
7
4
1
t
0
t
DR À
11
4
1
¸
2
c
t
2
DR
À
15
8
c
¸
2
c
t
D
p
þ
c

2
c
D
T
x V
T3
. ð37Þ
0
t
V
T3
¼
4
3
c
t
D
T
DR À
3
t
V
T3
À
c
2

2
c
D
T
c
3
þ
c
3
D
T
ð0
t
DRÞ .
ð38Þ
Combining equations (34) and (37) and eliminating D
p
, we
get
0
2
t
DR ¼
5
32
c
¸
2
c
t
c þ
11
32
c
¸
2
c
t
c
3
À
111
32
1
t
0
t
DR À
121
64
1
¸
2
c
t
2
DR
þ
5
32
c
¸
2
c
D
T
x V
T
þ
11
32
c
¸
2
c
D
T
x V
T3
. ð39Þ
Next we expand the spatial dependence of the perturba-
tion variables in spherical harmonics. We choose to normal-
ize these variables so as to make them dimensionless and of
a similar magnitude through the definitions
DR ¼

l.m
DRðl. m. tÞ R
ð0Þ
0

2
c
_ _
Y
lm
ð0. cÞ . ð40Þ
c ¼

l.m
cðl. m. tÞY
lm
ð0. cÞ . ð41Þ
c
3
¼

l.m
c
3
ðl. m. tÞY
lm
ð0. cÞ . ð42Þ
V
T
¼

l.m
V
T
ðl. m. tÞ cR
ð0Þ
0

c
_ _
D
T
Y
lm
ð0. cÞ
l
. ð43Þ
V
T3
¼

l.m
V
T3
ðl. m. tÞ cR
ð0Þ
0

c
_ _
D
T
Y
lm
ð0. cÞ
l
. ð44Þ
Equations (33), (35), (36), (38), and (39) can be rewritten
as
d
t
DR ¼ F . ð45Þ
d
t
c ¼ À
2
¸
2
c
F À
2
¸
2
c
t
DR À
3
t
c þ
ðl þ1Þ
¸
c
t
V
T
. ð46Þ
d
t
V
T
¼
l

c
F À
2l
¸
c
t
DR À
l

c
t
c À
9
2t
V
T
. ð47Þ
d
t
c
3
¼ À4F À
8
t
DR À
2
t
c
3
þ
ðl þ1Þ
¸
c
t
V
T3
. ð48Þ
d
t
V
T3
¼
l

c
F þ
2l
¸
c
t
DR À
l

c
t
c
3
À
7
2t
V
T3
. ð49Þ
d
t
F ¼ À
239
32t
F À
143
16t
2
DR þ
5
32t
2
c þ
11
32t
2
c
3
À
5
32
ðl þ1Þ
¸
c
t
2
V
T
À
11
32
ðl þ1Þ
¸
c
t
2
V
T3
. ð50Þ
The perturbation variables in the above six equations are
dimensionless and only functions of time. We have added
the variable F so that all the equations have the form of
first-order differential equations.
2.2. Wind Medium
If the circumburst medium is a progenitor wind,
,
1
/ R
À2
. Accordingly, the perturbation equations (20) and
(21) need to be changed to
0
t
c ¼ À
2
R
ð0Þ
0
0
t
DR þ 2
v
c
R
ð0Þ
0
_ _
2
À2 o
ð0Þ
_ _
À1
,
1
c
R
ð0Þ
0
_
¸
_
_
¸
_DR
À o
ð0Þ
_ _
À1
,
1
cc À
D
T
x V
T
. ð51Þ
0
2
t
DR ¼ 2 o
ð0Þ
_ _
À1
,
1
c
2
¸
2
c
À
3
3,4
2
1,2
o
ð0Þ
_ _
À1
p
3,4
c
,
1,4
1
c
1,2
¸
7,2
c
_ _
c
þ
_
4 o
ð0Þ
_ _
À1
,
1
c
À
7 3
3,4
ð Þ
2
3,2
o
ð0Þ
_ _
À1
p
3,4
c
,
1,4
1
¸
3,2
c
c
1,2
_
0
t
DR
þ
_
4 o
ð0Þ
_ _
À1
,
1
c
2
¸
2
c
R
ð0Þ
0
À
3
3,4
2
3,2
o
ð0Þ
_ _
À1
p
3,4
c
,
1,4
1
c
1,2
¸
7,2
c
R
ð0Þ
0
_
DR
þ
3
7,4
2
5,2
o
ð0Þ
_ _
À1
p
3,4
c
,
1,4
1
c
1,2
¸
7,2
c
D
p
þ
v
c

2
c
D
T
x V
T
.
ð52Þ
No. 2, 2002 SHOCK STABILITY FROM IMPACT OF FIREBALL 833
Equations (22)–(25) remain the same as in the uniform
medium case.
Since ,
1
/ R
À2
and ,
4
/ R
À2
in the wind case, equation
(26) implies that ¸
c
is constant over time. Equations (10),
(11), (18), and (19) then yield the unperturbed parameters
o
ð0Þ
% ,
1
ct . ð53Þ
o
ð0Þ
3
%
¸
4
¸
2
c
,
4
ct . ð54Þ
p
c
%
4
3
¸
2
c
,
1
c
2
%
1
3
¸
2
4
¸
2
c
,
4
c
2
. ð55Þ
¸
c
%
1
ffiffiffi
2
p ¸
1,2
4
,
4
,
1
_ _
1,4
. ð56Þ
Substitution of equations (53)–(55) into the perturbation
equations (51), (52), and (22)–(25) yields
0
t
c ¼ À
2
ct
0
t
DR À
1
t
c À
D
T
x V
T
. ð57Þ
0
2
t
DR ¼ À
3
t
0
t
DR þ
3
¸
2
c
t
2
DR þ
3
2
c
¸
2
c
t
D
p
þ
c

2
c
D
T
x V
T
.
ð58Þ
0
t
V
T
¼ À
c
t
D
T
DR À
2
t
V
T
À
c
2

2
c
D
T
c þ
c
3
D
T
ð0
t
DRÞ . ð59Þ
0
t
c
3
¼ À2
¸
2
c
ct
0
t
DR À
1
t
c
3
À
D
T
x V
T3
. ð60Þ
0
2
t
DR ¼ À
3
2
1
t
0
t
DR À
3
2
1
¸
2
c
t
2
DR À
3
4
c
¸
2
c
t
D
p
þ
c

2
c
D
T
x V
T3
.
ð61Þ
0
t
V
T3
¼
1
2
c
t
D
T
DR À
2
t
V
T3
À
c
2

2
c
D
T
c
3
þ
c
3
D
T
ð0
t
DRÞ .
ð62Þ
Combining equations (58) and (61) and eliminating D
p
, we
get
0
2
t
DR ¼ À
2
t
0
t
DR þ
c

2
c
D
T
x V
T
þ
c

2
c
D
T
x V
T3
. ð63Þ
Using the same normalized perturbation variables as
defined in equations (40)–(44), we can rewrite equations
(57), (59), (60), (62), and (63) as
d
t
DR ¼ F . ð64Þ
d
t
c ¼ À
2
¸
2
c
F À
2
¸
2
c
t
DR À
1
t
c þ
ðl þ1Þ
¸
c
t
V
T
. ð65Þ
d
t
V
T
¼
l

c
F À
2l

c
t
DR À
l

c
t
c À
2
t
V
T
. ð66Þ
d
t
c
3
¼ À2F À
2
t
DR À
1
t
c
3
þ
ðl þ1Þ
¸
c
t
V
T3
. ð67Þ
d
t
V
T3
¼
l

c
F þ
5l

c
t
DR À
l

c
t
c
3
À
2
t
V
T3
. ð68Þ
d
t
F ¼ À
4
t
F À
2
t
2
DR À
ðl þ1Þ

c
t
2
V
T
À
ðl þ1Þ

c
t
2
V
T3
. ð69Þ
These six first-order differential equations are the final per-
turbation equations for the wind case.
3. SOLUTIONS OF THE PERTURBATION EQUATIONS
For the wind case, ¸
c
¼ const, and we can solve the per-
turbation equations analytically. If we define F ¼ F
0
,t, then
equations (64)–(69) can be rewritten in a matrix form:
d
t
F
0
DR
c
V
T
c
3
V
T3
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
¼
1
t
Â
À3 À2 0 À
ðl þ1Þ

c
0 À
ðl þ1Þ

c
1 0 0 0 0 0
À
2
¸
2
c
À
2
¸
2
c
À1
ðl þ1Þ
¸
c
0 0
l

c
À
2l

c
À
l

c
À2 0 0
À2 À2 0 0 À1
ðl þ1Þ
¸
c
l

c
5l

c
0 0 À
l

c
À2
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
Â
F
0
DR
c
V
T
c
3
V
T3
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
. ð70Þ
In matrix notation, the above equation is
d
t
y ¼
1
t
Ay . ð71Þ
where y is a vector and A is a 6 Â6 time-independent
matrix.
The matrix A can be diagonalized through the transfor-
mation
X
À1
AX ¼ diagð`
1
Á Á Á `
6
Þ ¼ D . ð72Þ
where `
1
, . . ., `
6
are the eigenvalues of the matrix A, and X
is the matrix formed by columns from the eigenvectors (i.e.,
the kth column of X is the eigenvector corresponding to the
eigenvalue `
k
). Equation (71) can then be transformed to
d
t
y ¼
1
t
XDX
À1
_ _
y¼)d
t
X
À1
y
_ _
¼
1
t
D X
À1
y
_ _
. ð73Þ
By defining a new vector y
0
¼ X
À1
y, we get the equation
d
t
y
0
¼
1
t
Dy
0
. ð74Þ
which has six components:
d
t
y
0
k
¼
1
t
`
k
y
0
k
. k ¼ 1. . . . . 6 . ð75Þ
834 WANG, LOEB, & WAXMAN Vol. 568
Since `
k
can be a complex number, we write `
k
¼ a
k
þib
k
.
The solution to equation (75) is then
y
0
k
¼ c
k
t
a
k
þib
k
¼ c
k
t
a
k
e
ib
k
ln t
. ð76Þ
where c
k
is a constant dictated by the initial conditions.
Each y
0
k
defines a mode of the shock system. There are six
modes in total corresponding to six eigenvalues of the
matrix A. The real part of each eigenvalue dictates the over-
all temporal behavior of the mode, while the imaginary part
determines its oscillation frequency. The vector y can be
derived fromthe relation
y ¼ Xy
0
. ð77Þ
Hence, each component of the vector y is a linear combina-
tion of the six different modes. The vector c whose compo-
nents are c
k
can be obtained from the initial conditions
c ¼ y
0
ðt ¼ 1Þ ¼ X
À1
yðt ¼ 1Þ . ð78Þ
namely,
c
1
c
2
c
3
c
4
c
5
c
6
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
¼ X
À1
F
0
ðt ¼ 1Þ
DRðt ¼ 1Þ
cðt ¼ 1Þ
V
T
ðt ¼ 1Þ
c
3
ðt ¼ 1Þ
V
T3
ðt ¼ 1Þ
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
. ð79Þ
The eigenvalues of the matrix A can be calculated for dif-
ferent values of ¸
c
and l. Figure 2 shows the real and imagi-
nary parts of the six eigenvalues as functions of l for
¸
c
¼ 500. Each row in the figure contains two panels, corre-
sponding to the real and imaginary parts of a particular
eigenvalue. The results show that for l 320, all six eigen-
values are real numbers, and so there are no oscillations.
For 320 < l 432, two eigenvalues are complex numbers
and they are a pair of complex conjugates, implying that
two modes are oscillating with the same frequency. For
l 432, there are two pairs of complex conjugates. The
transition from real eigenvalues to complex eigenvalues
occurs when l $ ¸
c
, as expected from the fact that oscilla-
tions are possible only when causality allows communica-
tion across the scale of a wavelength for modes with l &¸
c
.
For the thin-shell approximation to be valid, we require
that the wavelength of the perturbation be much larger than
the thickness of the forward/reverse-shock system in the
shock frame. The thickness of the shock systemis .2R
0
,¸¸
2
c
in the observer frame, and thus .2R
0
,¸¸
c
in the shock
frame. Here ¸ is a constant that ranges between $4 and $12
for the wind and ISM profiles, respectively. The wavelength
of the perturbation is $2¬R
0
,l in the shock frame. There-
fore, we enforce an upper limit on l of $10¸
c
. Figure 2 shows
that for all values of l, the real parts of the six eigenvalues
are À1. This implies that all modes are decaying faster
than or are proportional to t
À1
. Since each perturbation
variable is a linear combination of the six modes, we con-
clude that all perturbation variables should also decay faster
than or be proportional to t
À1
. Thus, the system is stable.
Note that for large ¸
c
, we expect the results to depend only
on l,¸
c
(see eq. [70]), and so our particular choice of
¸
c
¼ 500 can be scaled appropriately to other values of ¸
c
.
For l4¸
c
, the eigenvalues admit the following analytical
solutions,
`
1
¼ À1 . `
2
¼ À
3
2
À
1
ffiffiffi
3
p
l
¸
c
i . `
3
¼ À
3
2
þ
1
ffiffiffi
3
p
l
¸
c
i .
ð80Þ
`
4
¼ À
19
9
. `
5
¼ À
13
9
þ
1
ffiffiffi
2
p
l
¸
c
i . `
6
¼ À
13
9
À
1
ffiffiffi
2
p
l
¸
c
i .
ð81Þ
while in the limit of l5¸
c
,
`
1
¼ À1 . `
2
¼ À2 þ
1
3
l
2
¸
2
c
. `
3
¼ À1 À
1
3
l
2
¸
2
c
. ð82Þ
`
4
¼ À1 À
1
2
l
2
¸
2
c
. `
5
¼ À2 À
1
3
ffiffiffi
2
p
l
¸
c
. `
6
¼ À2 þ
1
3
ffiffiffi
2
p
l
¸
c
.
ð83Þ
We have calculated the corresponding eigenvectors
numerically as shown in Figure 3 (for ¸
c
¼ 500). Each row
in the figure contains two panels that show the real and
imaginary parts of one of the six components of the eigen-
vectors as functions of l. Different line types correspond to
the six different eigenvectors. The complex eigenvectors are
all scaled to have a unit magnitude. Since each eigenvector
corresponds to a mode, the relative values of the six compo-
nents of the eigenvector measure the physical significance of
perturbations in different physical parameters for that
mode. For example, the mode corresponding to the eigen-
value `
1
¼ À1 has the temporal behavior of t
À1
; the eigen-
vector for this mode is (À0.275, 0.275, À0.824, 0, 0.412, 0),
implying that this mode does not involve V
T
and V
T3
pertur-
bations. Also note that this mode does not depend on l,
while all other modes change with l.
Equations (64)–(69) can also be solved numerically. We
normalize all initial values of the perturbation variables to
unity. The temporal interval of the calculation is from t ¼ 1
to 15, and ¸
c
is chosen to be 500. Here t ¼ 1 marks the time
when the perturbations are added to the forward/reverse-
shock system. To a distant observer, this corresponds to
T
t¼1
% R
t¼1
,2¸
2
c
c,
1
where R
t¼1
is the radius of the double-
shock system at t ¼ 1. Figure 4 shows our results. The six
panels show the time evolution of DR, d
t
DR, c, V
T
, c
3
, and
V
T3
. We show results for four different l values, namely
l ¼ 5, 50, 500, and 5 Â10
3
. These plots indicate that all per-
turbation variables decay quickly with time. For small val-
ues of l (e.g., l ¼ 5 and 50), there are no oscillations. For
large values of l (e.g., l ¼ 5 Â10
3
), the oscillations exist but
damp away quickly. These results are consistent with our
analytical derivations. For l ¼ 500, the oscillations start to
appear, although they are not apparent in the plot because
of their low frequency. For l4¸
c
, we can calculate the fre-
quencies of the oscillations using the eigenvalues listed in
equations (80) and (81). The oscillations have the form
exp i 1,
ffiffiffi
3
p _ _
l,¸
c
ð Þ ln t
_ ¸
or exp i 1,
ffiffiffi
2
p _ _
l,¸
c
ð Þ ln t
_ ¸
so that
the oscillation period increases with time. For l ¼ 10¸
c
, the
1
Note that for a GRB located at a cosmological redshift z, one should
use T
t¼1
% ð1 þzÞR
t¼1
,2¸
2
c
c instead. We omit the cosmological redshift
factor in the text in order to keep the expressions simpler.
No. 2, 2002 SHOCK STABILITY FROM IMPACT OF FIREBALL 835
shortest period is $2.4, corresponding to 2.4T
t¼1
to an
observer. Hence, the oscillations could produce fluctuations
in the observed flux on timescales as short as a few times
T
t¼1
. Because the decay of the modes is not exponentially
fast but rather moderate, the observed flux might show
measurable fluctuations if the initial perturbations are suffi-
ciently large.
In the ISM case, ¸
c
is not constant, and so we can not
write equations (45)–(50) in a matrix form that admits an
analytic solution. Instead, we have to solve these equations
numerically. In order to test the validity of the perturbation
equations and the numerical code, we compared the numeri-
cal results for a spherical perturbation with l ¼ 0 with the
analytic solution derived by directly perturbing the radial
equations of motion, and found an excellent agreement
between the two calculations. Similarly to the wind case, we
normalized all initial values of the perturbation variables to
unity and chose an initial ¸
c
¼ 500. Our numerical results
are shown in Figure 5 and qualitatively resemble the wind
case. Overall, the perturbations decay rapidly with time,
and oscillations appear only for large values of l. Similarly
to the wind case, for l ¼ 10¸
c
the shortest period of the oscil-
0 200 400 600 800 1000
−2
−1.5
−1
−0.5
0
l
R
e
[
λ
1
]
0 200 400 600 800 1000
−1
−0.5
0
0.5
1
l
I
m
[
λ
1
]
0 200 400 600 800 1000
−2
−1.8
−1.6
−1.4
l
R
e
[
λ
2
]
0 200 400 600 800 1000
−1
−0.5
0
l
I
m
[
λ
2
]
0 200 400 600 800 1000
−1.6
−1.4
−1.2
−1
l
R
e
[
λ
3
]
0 200 400 600 800 1000
0
0.5
1
l
I
m
[
λ
3
]
0 200 400 600 800 1000
−2
−1.5
−1
l
R
e
[
λ
4
]
0 200 400 600 800 1000
−0.4
−0.3
−0.2
−0.1
0
0.1
l
I
m
[
λ
4
]
0 200 400 600 800 1000
−2.2
−2
−1.8
−1.6
−1.4
l
R
e
[
λ
5
]
0 200 400 600 800 1000
−0.5
0
0.5
1
1.5
l
I
m
[
λ
5
]
0 200 400 600 800 1000
−2
−1.8
−1.6
−1.4
l
R
e
[
λ
6
]
0 200 400 600 800 1000
−1.5
−1
−0.5
0
0.5
l
I
m
[
λ
6
]
Fig. 2.—Real and imaginary parts of the six eigenvalues in the wind case as functions of l, with ¸
c
¼ 500
836 WANG, LOEB, & WAXMAN Vol. 568
lations is $2, corresponding to $2T
t¼1
to an observer.
Again, the damped fluctuations in the flux are detectable if
the initial perturbations are sufficiently large.
For the double-shock system we are considering, Sari &
Piran (1995) have defined four critical radii, R
N
¼

3,2
,D
1,2
j
2
, where the reverse shock becomes relativistic;
R
D
¼ ‘
3,4
D
1,4
, where the reverse shock crosses the shell;
R
¸
¼ ‘,j
2,3
, where the forward shock sweeps up a mass
M
0
,j; and R
s
¼ Dj
2
, where the shell begins to spread if the
initial Lorentz factor varies by the order of j. Here
‘ ðE,n
1
m
p
c
2
Þ
1,3
is the Sedov length, D is the width of the
shell, E is the equivalent isotropic energy of the fireball, M
0
is the mass of the initial baryonic load of the fireball, j is the
initial thermal Lorentz factor of the fireball, and n
1
is the
number density of the ISM. These four radii are simply
related as follows: R
N
,¸ ¼ R
¸
¼ ¸
1,2
R
D
¼ ¸
2
R
s
, with
the dimensionless quantity ¸ ðl,DÞ
1,2
j
À4,3
. When ¸ 1,
we have the Newtonian case, with R
s
< R
D
< R
¸
< R
N
; the
0 200 400 600 800 1000
−0.4
−0.2
0
0.2
0.4
l
R
e
[
v
1
]
0 200 400 600 800 1000
−0.02
−0.01
0
0.01
0.02
l
I
m
[
v
1
]
0 200 400 600 800 1000
−0.2
0
0.2
0.4
l
R
e
[
v
2
]
0 200 400 600 800 1000
−0.05
0
0.05
l
I
m
[
v
2
]
0 200 400 600 800 1000
−1
−0.5
0
0.5
1
l
R
e
[
v
3
]
0 200 400 600 800 1000
−0.5
0
0.5
l
I
m
[
v
3
]
0 200 400 600 800 1000
−1
−0.5
0
l
R
e
[
v
4
]
0 200 400 600 800 1000
−0.2
−0.1
0
0.1
0.2
l
I
m
[
v
4
]
0 200 400 600 800 1000
−1
−0.5
0
0.5
l
R
e
[
v
5
]
0 200 400 600 800 1000
−0.5
0
0.5
l
I
m
[
v
5
]
0 200 400 600 800 1000
0
0.2
0.4
0.6
l
R
e
[
v
6
]
0 200 400 600 800 1000
−1
−0.5
0
0.5
1
l
I
m
[
v
6
]
Fig. 3.—Real and imaginary parts of the six components of different eigenvectors (modes) in the wind case, with ¸
c
¼ 500. Each row corresponds to one
component of the eigenvector. Different line types correspond to six different eigenvectors: the thick solid line refers to the eigenvector of `
1
, the thick dashed
line to `
2
, the thin solid line to `
3
, the thin dashed line to `
4
, the dotted line to `
5
, and the dash-dotted line to `
6
.
No. 2, 2002 SHOCK STABILITY FROM IMPACT OF FIREBALL 837
reverse shock is still Newtonian when it crosses the shell.
When ¸ < 1, we have the relativistic case, with
R
N
< R
¸
< R
D
< R
s
; the reverse shock becomes relativistic
before it crosses the shell. In this paper, we only consider
the latter case. For typical GRB parameters, E ¼ 10
52
ergs,
j ¼ 10
3
, n ¼ 1 cm
À3
, and D ¼ 3 Â10
11
cm, we have
‘ % 2 Â10
18
cm, ¸ % 0.25, R
N
% 5 Â10
15
cm, and
R
D
% 4 Â10
16
cm. For our calculation, the perturbations
can be added to the system between R
N
and R
D
. Thus t ¼ 1
corresponds to the time between R
N
,c and R
D
,c. To an
observer, this corresponds to the time T
t¼1
between
R
N
,2¸
2
c
c and R
D
,2¸
2
c
c, or $0.3–2.5 s. Thus, the timescales
of the fluctuations in the observed flux could be as short as a
fewseconds.
4. DISCUSSION
We have solved the perturbation equations describing the
double (forward/reverse)-shock system that forms during
the impact of a highly relativistic fireball on a surrounding
0 5 10 15
0
0.2
0.4
0.6
0.8
1
1.2
t

R
l=5
l=50
l=500
l=5000
0 5 10 15
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
1.2
t
d
t

R
0 5 10 15
−4
−3
−2
−1
0
1
2
t
δ
0 5 10 15
−1.5
−1
−0.5
0
0.5
1
t
V
T
0 5 10 15
−1
0
1
2
3
4
t
δ
3
0 5 10 15
−0.5
0
0.5
1
1.5
t
V
T
3
Fig. 4.—Evolution of the perturbation variables in the wind case, with ¸
c
¼ 500. Four different line types correspond to l ¼ 5, 50, 500, and 5 Â10
3
, as
marked.
838 WANG, LOEB, & WAXMAN Vol. 568
medium. For both a uniform and a wind ð1,r
2
Þ density pro-
file of the ambient medium, we have found the shock system
to be stable to global perturbations. We therefore do not
expect the shock to fragment. Our results are limited to rela-
tivistic reverse shocks, and appear to differ qualitatively
from previous results in the nonrelativistic regime (Vishniac
1983).
Our results also apply to collimated outflows as long as
the double-shock system is formed at a time when the Lor-
entz factor of the outflow is larger than the collimation
angle.
We derived the frequencies of the normal modes that
could modulate the short-term variability at the early phase
of GRB afterglows. The results imply that perturbations in
the double-shock system could produce fluctuations in the
observed flux on timescales as short as a few seconds for
¸
c
$ 500 in the ISM case. These damped short-term fluctua-
tions are detectable if the initial perturbations are suffi-
0 2 4 6 8 10
−0.2
0
0.2
0.4
0.6
0.8
1
1.2
t

R
l=5
l=50
l=500
l=5000
0 2 4 6 8 10
−1
−0.5
0
0.5
1
t
d
t

R
0 2 4 6 8 10
−6
−5
−4
−3
−2
−1
0
1
t
δ
0 2 4 6 8 10
−3
−2.5
−2
−1.5
−1
−0.5
0
0.5
1
1.5
t
V
T
0 2 4 6 8 10
−1
0
1
2
3
4
5
t
δ
3
0 2 4 6 8 10
−0.5
0
0.5
1
1.5
2
2.5
t
V
T
3
Fig. 5.—Evolution of the perturbation variables in the ISMcase, with ¸
c
¼ 500
No. 2, 2002 SHOCK STABILITY FROM IMPACT OF FIREBALL 839
ciently large. The fluctuations could be supplemented by
variability on much longer timescales due to density inho-
mogeneities in the ISM; such inhomogeneities can lead to
variability on timescales of tens of minutes in the optical
band and days in the radio (Wang &Loeb 2000).
This work was supported in part by grants from the
Israel-US Binational Science Foundation (BSF 98-00343),
NSF (AST 99-00877; AST 00-71019), and NASA (NAG 5-
7039; NAG5-7768).
APPENDIX
Here we provide full details for the derivation of equations (7)–(9) in x 2. We start by listing the equations of motion for a rel-
ativistic fluid in spherical coordinates. The continuity equation reads
0
0t
ð¸,Þ þ
1
r
2
0
0r
ðr
2
¸,u
r
Þ þ
1
r sin 0
0
00
ðsin 0¸,u
0
Þ þ
1
r sin 0
0
0c
ð¸,u
c
Þ ¼ 0 . ðA1Þ
and the three components of the momentumequation are
¸
2
c
2
ðe þpÞ
0u
r
0t
þu
r
0u
r
0r
þu
0
1
r
0u
r
00
þu
c
1
r sin 0
0u
r
0c
À
1
r
ðu
2
0
þu
2
c
Þ
_ _
þ
0p
0r
þ
u
r
c
2
0p
0t
¼ 0 . ðA2Þ
¸
2
c
2
ðe þpÞ
0u
0
0t
þu
r
0u
0
0r
þu
0
1
r
0u
0
00
þu
c
1
r sin 0
0u
0
0c
þ
1
r
u
r
u
0
þcot 0u
2
c
_ _
_ _
þ
1
r
0p
00
þ
u
0
c
2
0p
0t
¼ 0 . ðA3Þ
¸
2
c
2
ðe þpÞ
0u
c
0t
þu
r
0u
c
0r
þu
0
1
r
0u
c
00
þu
c
1
r sin 0
0u
c
0c
þ
1
r
ðu
r
u
c
þcot 0u
0
u
c
Þ
_ _
þ
1
r sin 0
0p
0c
þ
u
c
c
2
0p
0t
¼ 0 . ðA4Þ
where ,, e, p, and ¸ are the fluid density, energy density, pressure, and Lorentz factor, respectively, and u
r
, u
h
, and u
c
are the
three components of the fluid velocity.
For the forward/reverse-shock system under consideration (see Fig. 1), we define the following shell–averaged variables for
region 2,
oð0. cÞ ¼ R
À2
0
_
R
1
R
0
¸,r
2
dr . ðA5Þ
V
r
ð0. cÞ ¼ ðoR
2
0
Þ
À1
_
R
1
R
0
¸,u
r
r
2
dr . ðA6Þ
V
T
ð0. cÞ ¼ ðoR
2
0
Þ
À1
_
R
1
R
0
¸,u
T
r
2
dr . ðA7Þ
where u
T
is the tangential velocity vector.
Since the shocked material is relativistic, we adopt the relativistic equation of state, p ¼ e,3, in region 2. Equation (A1)
yields the evolution of the surface density
0
t
o ¼ À2
_
RR
0
R
0
o þ
R
1
R
0
_ _
2
¸ðR
1
Þ,ðR
1
Þ
_
RR
1
À¸ðR
0
Þ,ðR
0
Þ
_
RR
0
þR
À2
0
_
R
1
R
0
0
0t
ð¸,Þr
2
dr
¼ À2
_
RR
0
R
0
o þ
R
1
R
0
_ _
2
¸ðR
1
Þ,ðR
1
Þ½
_
RR
1
Àu
r
ðR
1
ފ þ¸ðR
0
Þ,ðR
0
Þ u
r
ðR
0
Þ À
_
RR
0
_ ¸
ÀR
À2
0
_
R
1
R
0
D
x ð¸,u
T
Þ ½ Šr
2
dr . ðA8Þ
Since r ¼ R
0
defines the contact discontinuity between the shocked shell and the shocked circumburst medium and there is no
mass flow across the contact discontinuity, we get u
r
ðR
0
Þ ¼
_
RR
0
. Because the forward shock is a strong relativistic shock, we
have the following shock jump conditions at shock 1,
¸
2
ðR
1
Þ ¼ ¸
2
s1
,2 . ðA9Þ
,ðR
1
Þ,,
1
¼ 4¸ðR
1
Þ . ðA10Þ
where ¸ðR
1
Þ and ,ðR
1
Þ are the Lorentz factor and density of the fluid just behind the shock front, ¸
s1
is the Lorentz factor of
the shock front, and ,
1
is the density of the unshocked circumburst medium just in front of the shock front. In the highly rela-
tivistic regime,
_
RR
1
% c 1 À
1

2
s1
_ _
. ðA11Þ
u
r
ðR
1
Þ % c 1 À
1

2
ðR
1
Þ
_ _
. ðA12Þ
840 WANG, LOEB, & WAXMAN Vol. 568
From equations (A9)–(A12), we obtain
¸ðR
1
Þ,ðR
1
Þ½
_
RR
1
Àu
r
ðR
1
ފ % ,
1
c . ðA13Þ
Thus, equation (A8) can be rewritten as
0
t
o ¼ À2
_
RR
0
R
0
o þ
R
1
R
0
_ _
2
,
1
c ÀR
À2
0
_
R
1
R
0
½
D
x ð¸,u
T
ފr
2
dr . ðA14Þ
To linear order, the last integration termin the above equation can be approximated by
ÀR
À2
0
_
R
1
R
0
½
D
x ð¸,u
T
ފr
2
dr ¼ Ào
D
T
x V
T
þR
À2
0
_
R
1
R
0
½
D
x ð¸,u
T
ފ
r
R
0
À1
_ _
r
2
dr . ðA15Þ
where the operator
D
T
ð
^
hh,R
0
Þð0,00Þ þð
^
}},R
0
Þð0,0cÞ acts as follows on a scalar Éand a vector f :
D
T
É ¼
1
R
0

00
^
hh þ
1
R
0
sin 0

0c
^
}} . ðA16Þ
D
T
x f ¼
1
R
0
sin 0
0
00
ðsin 0f
0
Þ þ
1
R
0
sin 0
0f
c
0c
. ðA17Þ
Note that the second termon the right-hand side of equation (A15) is of higher order in ðR
1
ÀR
0
Þ,R
0
than the preceding term,
and hence can be ignored in the thin-shell approximation. Thus, equation (A14) can be rewritten as
0
t
o ¼ À2
_
RR
0
R
0
o þ
R
1
R
0
_ _
2
,
1
c Ào
D
T
x V
T
. ðA18Þ
Similarly to the above derivation, we obtain for the bulk radial velocity
0
t
V
r
ð0. cÞ ¼ À
0
t
o
o
V
r
À2
_
RR
0
R
0
V
r
þo
À1
R
1
R
0
_ _
2
¸ðR
1
Þ,ðR
1
Þu
r
ðR
1
Þ½
_
RR
1
Àu
r
ðR
1
ފ þo
À1
¸ðR
0
Þ,ðR
0
Þu
r
ðR
0
Þ½u
r
ðR
0
Þ À
_
RR
0
Š
ÀðoR
2
0
Þ
À1
_
R
1
R
0
½
D
x ð¸,u
T
ފu
r
r
2
dr ÀðoR
2
0
Þ
À1
_
R
1
R
0
,c
2
4¸p
0p
0r
þ
u
r
c
2
0p
0t
_ _
r
2
dr
ÀðoR
2
0
Þ
À1
_
R
1
R
0
¸, u
T
x
D
u
r
ð Þr
2
dr þ oR
2
0
_ _
À1
_
R
1
R
0
¸,u
2
T
r dr . ðA19Þ
The last two terms on the right-hand side of the above equation are nonlinear. By substituting equations (A13) and (A14) into
the above equation and keeping terms to the linear order, we get
0
t
V
r
ð0. cÞ ¼ o
À1
R
1
R
0
_ _
2
,
1
c u
r
ðR
1
Þ ÀV
r
½ Š þ oR
2
0
_ _
À1
V
r
_
R
1
R
0
½
D
x ð¸,u
T
ފr
2
dr À oR
2
0
_ _
À1
_
R
1
R
0
u
r
½
D
x ð¸,u
T
ފr
2
dr
_ _
À oR
2
0
_ _
À1
_
R
1
R
0
,c
2
4¸p
0p
0r
þ
u
r
c
2
0p
0t
_ _
r
2
dr . ðA20Þ
In order to evaluate the integral in the last term of the above equation, we need the relation between p and , inside region 2.
Since entropy is conserved in this region,
d
dt
p
,
4,3
_ _
¼ 0 . ðA21Þ
implying that p,,
4,3
remains constant for a given fluid particle. Hence, a fluid layer that is at a distance xðtÞ from the contact
discontinuity (r ¼ R
0
) inside region 2 maintains a constant p,,
4,3
over time, and its value is decided by the Lorentz factor of
shock 1 at the time when this layer first crosses shock 1. However, at a particular time, different layers across region 2 have dif-
ferent values of p,,
4,3
. Assuming that region 2 is decelerating with ¸ / r
À1,2
(for the uniform ISMcase), we get
pðxÞ
,
4,3
ðxÞ
¼
c
2
3 4
1,3
ð Þ,
1,3
1
¸
a
R
1,2
a
ð8¸
2
a
R
a
x þR
2
a
Þ
1,4
_ _
2,3
. ðA22Þ
where ¸
a
and R
a
are the Lorentz factor and radius of region 2 at the initial time. Apparently, the dependence of pðxÞ,,
4,3
ðxÞ
on x is very weak, and so within the context of the thin-shell approximation, we simply assume that p,,
4,3
is constant across
region 2 at any given time. This assumption is indeed satisfied in the numerical simulations performed by Kobayashi, Piran, &
Sari (1999). In equation (A22), the term 8¸
2
a
R
a
x can be at most comparable to R
2
a
(this happens in the very last stage of the
evolution when the reverse shock crosses the shell), and so the error introduced by our approximation is small. For the wind
case, ¸ % const, and we can also treat p,,
4,3
as a constant across region 2.
No. 2, 2002 SHOCK STABILITY FROM IMPACT OF FIREBALL 841
We can now calculate the integral
_
R
1
R
0
,c
2
4¸p
0p
0r
r
2
dr ¼
_
R
1
R
0
c
2

0,
0r
r
2
dr ¼
c
2
3¸ðR
0
Þ
_
R
1
R
0
0,
0r
r
2
dr %
c
2
3¸ðR
0
Þ
,ðR
1
ÞR
2
1
À,ðR
0
ÞR
2
0
À
_
R
1
R
0
2r,dr
_ _
. ðA23Þ
In the thin-shell approximation, all radial velocities are dominated by the overall radial motion of the shock layer. Hence, ¸ is
treated as a constant and is taken out of the integral. For the uniform ISM case, the density difference between the two edges
of region 2 is not small; hence, the last term inside the square brackets is of order ðR
1
ÀR
0
Þ,R
0
times the difference between
the previous two terms, and so it can be neglected. For the wind case, there is no pressure gradient across region 2, and so the
above integral vanishes. However, even in this case we can ignore the last term and keep only the first two terms, because later
on we replace R
1
with R
0
, and so the difference between the first two terms vanishes.
Another relevant integral is
_
R
1
R
0
,c
2
4¸p
u
r
c
2
0p
0t
r
2
dr ¼
_
R
1
R
0
c
2

u
r
c
2
0,
0t
r
2
dr ¼
V
r
3¸ðR
0
Þ
_
R
1
R
0
0,
0t
r
2
dr ¼
V
r
3¸ðR
0
Þ
0
0t
_
R
1
R
0
,r
2
dr À
_
RR
1
,ðR
1
ÞR
2
1
þ
_
RR
0
,ðR
0
ÞR
2
0
_ _
¼
V
r
3¸ðR
0
Þ
o
¸ðR
0
Þ
2R
0
_
RR
0
þ
R
2
0
¸ðR
0
Þ
0
t
o À
R
2
0
o
¸
2
ðR
0
Þ
0
t
¸ðR
0
Þ À
_
RR
1
,ðR
1
ÞR
2
1
þ
_
RR
0
,ðR
0
ÞR
2
0
_ _
. ðA24Þ
In the above derivation, we pulled u
r
out of the integration assuming that it equals V
r
, as appropriate in the thin-shell approxi-
mation. By substituting equation (A18) into equation (A24) and making use of the following two relations,
c
2
ÀV
r
_
RR
0
%
c
2
¸
2
ðR
0
Þ
. ðA25Þ
c
2
ÀV
r
_
RR
1
%
3c
2

2
ðR
0
Þ
. ðA26Þ
we get
_
R
1
R
0
,c
2
4¸p
0p
0r
þ
u
r
c
2
0p
0t
_ _
r
2
dr ¼
4R
2
0

2
ðR
0
Þ
,
1
c
2
À
R
2
0

3
ðR
0
Þ
,ðR
0
Þc
2
Ào
R
2
0

2
ðR
0
Þ
V
r
D
T
x V
T
Ào
R
2
0

3
ðR
0
Þ
V
r
0
t
¸ðR
0
Þ . ðA27Þ
Thus, equation (A20) is nowchanged to
0
t
V
r
¼ o
À1
R
1
R
0
_ _
2
,
1
c½u
r
ðR
1
Þ ÀV
r
Š þ oR
2
0
_ _
À1
V
r
_
R
1
R
0
½
D
x ð¸,u
T
ފr
2
dr À oR
2
0
_ _
À1
_
R
1
R
0
u
r
½
D
x ð¸,u
T
ފr
2
dr
_ _
Ào
À1
4

2
ðR
0
Þ
,
1
c
2
þo
À1
1

3
ðR
0
Þ
,ðR
0
Þc
2
þ
1

2
ðR
0
Þ
V
r
D
T
x V
T
þ
1

3
ðR
0
Þ
V
r
0
t
¸ðR
0
Þ . ðA28Þ
In order to close the final equations, we can only have one free variable for the radial velocities. We use the approximation that
u
r
is constant across region 2 with V
r
¼ u
r
ðR
1
Þ, as appropriate under the thin-shell approximation. Hence, the first two terms
in equation (A28) both vanish, and we end up with the following equation:
0
t
V
r
¼ Ào
À1
4

2
ðR
0
Þ
,
1
c
2
þo
À1
1

3
ðR
0
Þ
,ðR
0
Þc
2
þ
1

2
ðR
0
Þ
V
r
D
T
x V
T
þ
1

3
ðR
0
Þ
V
r
0
t
¸ðR
0
Þ . ðA29Þ
Since
0
t
¸ðR
0
Þ
¸
3
ðR
0
Þ
¼
V
r
c
2
0
t
V
r
. ðA30Þ
equation (A29) can be rewritten as
0
t
¸ðR
0
Þ ¼ À2o
À1
¸ðR
0
Þ,
1
c þ
1
2
o
À1
,ðR
0
Þc þ
¸ðR
0
Þ
2c
V
r
D
T
x V
T
. ðA31Þ
Because p,,
4,3
is a constant across region 2, we obtain the following relation:
,ðR
0
Þ ¼ 3
3,4
2
1,2
_ _
p
3,4
ðR
0
Þ,
1,4
1
¸
1,2
ðR
0
Þc
3,2
. ðA32Þ
Using this result, we can rewrite equation (A31) as
0
t
¸ðR
0
Þ ¼ À2o
À1
¸ðR
0
Þ,
1
c þ
3
3,4
2
1,2
o
À1
p
3,4
ðR
0
Þ,
1,4
1
¸
1,2
ðR
0
Þc
1,2
þ
¸ðR
0
Þ
2c
V
r
D
T
x V
T
. ðA33Þ
842 WANG, LOEB, & WAXMAN Vol. 568
For the tangential velocity, we have
0
t
V
T
ð0. cÞ ¼ À
0
t
o
o
V
T
À2
_
RR
0
R
0
V
T
þo
À1
R
1
R
0
_ _
2
¸ðR
1
Þ,ðR
1
Þu
T
ðR
1
Þ
_
RR
1
Àu
r
ðR
1
Þ
_ ¸
þo
À1
¸ðR
0
Þ,ðR
0
Þu
T
ðR
0
Þ u
r
ðR
0
Þ À
_
RR
0
_ ¸
À oR
2
0
_ _
À1
_
R
1
R
0
¸,
u
r
u
T
r
r
2
dr À oR
2
0
_ _
À1
_
R
1
R
0
,c
2
4¸p
D
T
p þ
u
T
c
2
0p
0t
_ _
r
2
dr
À oR
2
0
_ _
À1
_
R
1
R
0
¸,ðu
T
x
D
u
T
Þr
2
dr À oR
2
0
_ _
À1
_
R
1
R
0
u
T
D
x ð¸,u
T
Þr
2
dr . ðA34Þ
Apparently, the last two terms in the above equation are nonlinear and can be ignored. By substituting equations (A13) and
(A18) into the above equation, we get
0
t
V
T
¼ o
À1
R
1
R
0
_ _
2
,
1
c½u
T
ðR
1
Þ ÀV
T
Š þV
T
D
T
x V
T
À oR
2
0
_ _
À1
_
R
1
R
0
¸,
u
r
u
T
r
r
2
dr À oR
2
0
_ _
À1
_
R
1
R
0
,c
2
4¸p
D
T
p þ
u
T
c
2
0p
0t
_ _
r
2
dr .
ðA35Þ
The second term on the right-hand side of equation (A35) is nonlinear and can be neglected. In the thin-shell approximation,
the third term on the right-hand side of equation (A35) can be approximated as ÀV
r
V
T
,R
0
. Using the shock jump conditions
at shock 1 and making use of the fact that the tangential velocities must be continuous across the shock front, we obtain
u
T
ðR
1
Þ ¼ Àu
r
ðR
1
Þð
D
T
R
1
Þ . ðA36Þ
Based on these considerations, equation (A35) can be rewritten as
0
t
V
T
¼ o
À1
R
1
R
0
_ _
2
,
1
c½Àu
r
ðR
1
Þð
D
T
R
1
Þ ÀV
T
Š À
V
r
R
0
V
T
À oR
2
0
_ _
À1
_
R
1
R
0
,c
2
4¸p
D
T
p þ
u
T
c
2
0p
0t
_ _
r
2
dr . ðA37Þ
Next we consider the integration termin the above equation, which includes
_
R
1
R
0
,c
2
4¸p
ð
D
T
pÞr
2
dr ¼
_
R
1
R
0
c
2

ð
D
T
,Þr
2
dr ¼
_
R
1
R
0
c
2

2
½
D
T
ð¸,ފr
2
dr À
_
R
1
R
0
,c
2

2
ð
D
T
¸Þr
2
dr
¼
c
2

2
ðR
0
Þ
D
T
_
R
1
R
0
¸,r
2
dr À¸ðR
1
Þ,ðR
1
ÞR
2
1
ð
D
T
R
1
Þ þ¸ðR
0
Þ,ðR
0
ÞR
2
0
ð
D
T
R
0
Þ
_ _
À
c
2
D
T
¸ðR
0
Þ

3
ðR
0
Þ
_
R
1
R
0
¸,r
2
dr
¼
c
2

2
ðR
0
Þ
R
2
0
D
T
o þ2oR
0
ð
D
T
R
0
Þ À¸ðR
1
Þ,ðR
1
ÞR
2
1
ð
D
T
R
1
Þ þ¸ðR
0
Þ,ðR
0
ÞR
2
0
ð
D
T
R
0
Þ
_ ¸
À
c
2

3
ðR
0
Þ
R
2
0
o
D
T
¸ðR
0
Þ . ðA38Þ
In the last pair of square brackets of equation (A38), the second term is much smaller than the fourth term, and so it can be
neglected. Another integration termis
_
R
1
R
0
,c
2
4¸p
u
T
c
2
0p
0t
r
2
dr ¼
_
R
1
R
0
1

2
ð0
t
ln p޸,u
T
r
2
dr %
1

2
ðR
0
Þ
½0
t
ln pðR
1
ފ
_
R
1
R
0
¸,u
T
r
2
dr
¼
0
t
¸ðR
0
Þ

3
ðR
0
Þ
oR
2
0
V
T
þ
1

2
ðR
0
Þ
0
t
,
1
,
1
_ _
oR
2
0
V
T
. ðA39Þ
Because 0
t
ln p does not change much across region 2, we took it out of the integration in the above derivation.
By substituting equation (A31) into equation (A39), we get
_
R
1
R
0
,c
2
4¸p
u
T
c
2
0p
0t
r
2
dr % À
R
2
0
¸
2
ðR
0
Þ
,
1
cV
T
þ
R
2
0

3
ðR
0
Þ
,ðR
0
ÞcV
T
þ
1

2
ðR
0
Þ
0
t
,
1
,
1
_ _
oR
2
0
V
T
. ðA40Þ
Now, by substituting equations (A38) and (A40) into equation (A37), we get
0
t
V
T
¼
1
3
o
À1
c
2
,
1
À
,ðR
0
Þ
¸ðR
0
Þ
_ _
D
T
R
0
Ào
À1
,
1
cV
T
À
V
r
R
0
V
T
Ào
À1
c
2

2
ðR
0
Þ
D
T
o þ
c
2

3
ðR
0
Þ
D
T
¸ðR
0
Þ À
1

2
ðR
0
Þ
0
t
,
1
,
1
_ _
V
T
.
ðA41Þ
In deriving the above equation, we made the assumption that surface irregularities due to variations in the thickness of the
No. 2, 2002 SHOCK STABILITY FROM IMPACT OF FIREBALL 843
shock layer are of higher order than irregularities due to the bulk displacement of the shock layer (regions 2 and 3 in Fig. 1), so
D
T
R
0
¼
D
T
R
1
¼
D
T
R
2
. This is appropriate under the thin-shell approximation. Also, the last term in the above equation is
much smaller than the third term for the wind case and is equal to zero for the ISM case, and so can be neglected. If we now
substitute equation (A32) into the above equation, we get equation (9) in x 2.
The derivation of the perturbation equations of region 3 is very similar to that of region 2.
REFERENCES
Blandford, R. D., &McKee, C. F. 1976, Phys. Fluids, 19, 1130
Gruzinov, A. 2000, preprint (astro-ph/0012364)
Katz, J. I. 1994, ApJ, 422, 248
Kobayashi, S., Piran, T., &Sari, R. 1999, ApJ, 513, 669
Me´sza´ros, P., &Rees, M. J. 1993, ApJ, 405, 278
———. 1997, ApJ, 476, 232
Me´sza´ros, P., Rees, M. J., &Papathanassiou, H. 1994, ApJ, 432, 181
Paczyn´ ski, B., & Rhoads, J. E. 1993, ApJ, 418, L5
Rees, M. J., &Me´sza´ros, P. 1992, MNRAS, 258, 41P
Sari, R., Narayan, R., &Piran, T. 1996, ApJ, 473, 204
Sari, R., &Piran, T. 1995, ApJ, 455, L143
Sari, R., Piran, T., &Narayan, R. 1998, ApJ, 497, L17
Vishniac, E. T. 1983, ApJ, 274, 152
Wang, X., &Loeb, A. 2000, ApJ, 535, 788
Waxman, E. 1997a, ApJ, 485, L5
———. 1997b, ApJ, 489, L33
844 WANG, LOEB, & WAXMAN