Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

3D Relativistic Hydrodynamics

Download as pdf or txt
Download as pdf or txt
You are on page 1of 30

3D Relativistic Hydrodynamics

Miguel A. Aloy1 and José M. Martı́2


1
Max-Planck-Institut für Astrophysik,
Karl-Schwarzschild-Str. 1, 85748 Garching, Germany
2
Departamento de Astronomı́a y Astrofı́sica, Universidad de Valencia,
C/ Doctor Moliner 50, 46100 Burjassot, Spain

Abstract. We review the evolution of the numerical techniques applied in relativistic


hydrodynamics since the sixties until today. We focus our attention on the state–of–
the–art high–resolution shock–capturing methods and the astrophysical applications
involving three–dimensional simulations.

1 Introduction

A relativistic description of fluid dynamics is necessary in situations where the


local velocity of the flow is close to the light speed in vacuum (c) or where the
local internal energy density is comparable (or larger) than the local rest–mass
density of the fluid. Alternatively, a relativistic description should be used when-
ever matter is influenced by large gravitational potentials (≈ O(c2 ) ≈ 9 1020 erg
g−1 ), where a description in terms of the Einstein field theory of gravity is neces-
sary. Relativistic flows are present in numerous astrophysical phenomena, from
stellar to galactic (and even cosmological –early Universe, galaxy formation–)
scales. Among these phenomena are core collapse supernovae, X-ray binaries,
pulsars, coalescing neutron stars and black holes, micro-quasars, active galactic
nuclei, superluminal jets, gamma-ray bursts and, in general, any astrophysical
scenario involving compact objects.
More than thirty years ago, the pioneering work of May and White [103],
studying the process of stellar core collapse in spherical symmetry, triggered
the use of relativistic numerical simulations as a tool to get insight into the
above mentioned phenomena, complementing theoretical models and observa-
tions. However, the description of the ultrarelativistic regime has only been pos-
sible after the development, in the last ten years, of new numerical algorithms
(the so called high–resolution shock–capturing, HRSC, techniques).
This work is aimed to summarize the basic theory of the numerical techniques
applied to solve multidimensional (specifically three-dimensional, 3D) problems
in the frame of relativistic hydrodynamics (RHD). Recent works reviewing this
topic are [102] (in Special Relativity) and [45] (in General Relativity). In addi-
tion, we discusse the state of art of the current numerical 3D RHD codes. The
work is organized as follows. The equations of General Relativistic Hydrody-
namics (GRHD) are introduced in §2. Some important mathematical properties
of the GRHD equations (conservative character, hyperbolicity) are discussed in
Sections 2.1 and 2.2. Section 2.3 points out briefly the main differences between
2 M.A. Aloy and J.M. Martı́

classical and relativistic equations. Sections 3, 4 and 5 form the body of the
review. Section 3 is devoted to discuss different approaches for the integration of
the RHD eqs. paying special attention to the most recent numerical algorithms.
Present numerical 3D RHD codes are reviewed in §4 as well as some compu-
tational issues relevant for three dimensional simulations. Several astrophysics
applications are discussed in §5. We finish this work with a summary (6).

2 The equations of General Relativistic Hydrodynamics

The equations that describe the evolution of a relativistic fluid can be written
as covariant divergences,

∇·J = 0, (1a)
∇·T = 0, (1b)

representing the conservation of rest-mass and energy–momentum in the space–


time M, described by a metric g. In the previous equations, J is the current of
rest mass and T is the energy-momentum tensor. For perfect fluids (i.e., those
without shear or heat conduction) and using a system of natural units (G = c =
1; G is the gravitational constant), the components of J and T on a coordinate
basis are

J = ρuµ , µ, ν = 0, . . . , 3 (2a)
µν µ µ µν
T = ρhu u + pg , (2b)

ρ, p, h, ε and uµ being the rest–mass density, the pressure, the specific enthalpy
(h = 1 + ε + p/ρ), the specific internal energy and the four–velocity of the
fluid, respectively. The system of equations (1a, 1b) is closed making use of the
normalization condition of the four velocity (uµ uµ = −1, where summation is
extended over repeated indices) and an equation of state (EoS), usually of the
form p = p(ρ, ε).
A consistent numerical simulation of the flow evolution within the framework
of Relativity requires, in principle, the solution of the RHD equations (relativis-
tic counterparts of the Euler or Navier–Stockes equations) coupled to the full
set of Einstein field equations (that control the evolution of the space–time).
However, the problem is so complex (especially in multidimensions), and the
variety of astrophysical phenomena is so wide that for practical purposes it is
worthy to consider different physical approximations. The easiest one is to con-
sider that the gravitational field is unimportant and, therefore, the space–time
is flat or, in other words, the metric of the space–time is the Minkowski metric.
This is the approach of Special Relativistic hydrodynamic (SRHD) simulations
and has been successfully applied to, e.g., extragalactic jets (e.g., [101]), after-
glows of gamma–ray bursts (GRBs) [122], and also in other fields of physics,
like e.g., relativistic heavy–ion collisions [150]. Another possibility is to assume
that the simulated flow is a test fluid evolving in an external static field created
3D Relativistic Hydrodynamics 3

by a massive object. This is the approximation followed in most GRHD prob-


lems. The GRHD approximation is fruitfully used in simulations of accretion of
matter onto compact objects [63] and also in the context of the formation of
GRBs [7] (see § 5.2). However, when strong and varying gravitational fields are
encountered, the evolution of the metric has to be obtained by solving the Ein-
stein field equations (EFE). This situation holds, e.g., in the case of coalescing
neutron stars (NSs) [86] or in the collapse of a massive star to a black hole (BH)
– e.g., [103] –.

2.1 Equations of GRHD in conservation form


Following the {3+1} formalism [9], and given a general coordinate system {xµ } ≡
(t, xi ), the space–time M can be foliated into a set of spacelike hypersurfaces
Σt (for each constant value of t) such that
∂t = αn + β i ∂i , n · ∂i = 0, i = 1, 2, 3,
where {∂t , ∂i } define a coordinate basis and n is the unit timelike vector field
normal to Σt .
The line element on M may be written in terms of the scalar lapse, α, the
shift vector, β i , and the three–metric of each hypersurface Σt , γij ,
ds2 = −(α2 − βi β i )dt2 + 2βi dxi dt + γij dxi dxj .
In coordinates {xµ }, equations (1a, 1b) are written (in conservation form) as
· √ √
∂ γU(W) ∂ −gF(i) (W)
¸
1
√ + = S(W), (3)
−g ∂t ∂xi
where g stands for the determinant of the four–metric of M and γ for the
determinant of γij , and where U ≡ (D, Sj , τ ) is the vector of unknowns, which
are the rest-mass, momentum and energy densities, respectively, as measured
by Eulerian observers (i.e., those having n as four–velocity). In the previous
equations F(i) are the vectors of fluxes
βi βi βi
· µ ¶ µ ¶ µ ¶ ¸
F(i) (W) = D v i − , Sj v i − + pδji , τ v i − + pv i
α α α
and S is the vector of sources
· µ ¶ µ ¶¸
µν ∂gνj δ µ0 ∂ ln α µν 0
S(W) = 0, T − Γνµ gδj , α T − T Γνµ .
∂xµ ∂xµ
Conserved variables are related with physical or primitive variables, W ≡
(ρ, ε, v i ), i.e., the rest-mass density and specific internal energy in the fluid rest
frame and the fluid three–velocity measured by Eulerian observers, respectively,
through
D = ρΓ (4a)
2
Sj = ρhΓ vj (j = 1, 2, 3) (4b)
τ = ρhΓ 2 − p − ρΓ, (4c)
4 M.A. Aloy and J.M. Martı́

where
u · ∂i
vi =
−u · n
ui βi
v i = γ ij vj = + ,
αut α
and Γ ≡ −u · n = (1 − γij v i v j )−1/2 is the Lorentz factor.

2.2 GRHD equations as a hyperbolic system of conservation laws


Following, e.g., [157], a system of q equations written in conservation form, like
(3), is said to be hyperbolic at a point (t, xi ) if the Jacobian matrices of F(i)
(B (i) = ∂F(i) /∂U) have p real eigenvalues λ1 , . . . , λq and corresponding sets of
p linearly independent right eigenvectors r(1) , . . . , r(q) .
The characteristic curves associated to the system are (in the one–dimensional
case) the integral curves of
dx
= λk (U(x, t)), k = 1, . . . q (5)
dt
where we can notice that the eigenvalues of B(i) are the velocities of propaga-
tion of the characteristic fields (characteristic speeds). Along each characteristic
curve, there is one characteristic variable, i.e., a component of U (i) = L(i) U (L(i)
is the matrix of left eigenvectors of B(i) ), which remains constant. This fact
allows one to construct formally solutions at advanced times by propagating
the values of characteristic variables at previous times along the characteristic
curves, i.e., by solving initial value problems (IVPs).
Anile [8] has shown that system (3) is hyperbolic for causal EoS, i.e., for those
where the local sound speed, cs , defined by
∂p ∂p
hc2s = + (p/ρ2 ) , (6)
∂ρ ∂ǫ
satisfies cs < 1.
In the particular case of SRHD the set of eigenvalues associated to the Jaco-
bian matrix in the i–direction are [43,35,3]

λ0 = v i , (7a)
1 ³
i 2
p ´
λ± = v (1 − cs ) ± cs (1 − v 2 )[1 − v 2 c2s − v i v i (1 − c2s )] . (7b)
1 − v 2 c2s
Let us notice that: (i) there exists a strong coupling between the components
of velocity along the different spatial directions through the modulus of the ve-
locity, v; (ii) in the one–dimensional case (1D) the expressions of λ± (associated
to the acoustic waves) reduce to the Lorentz addition of the flow velocity and
the local sound speed
v ± cs
λ± = ; (8)
1 ± vcs
3D Relativistic Hydrodynamics 5

(iii) λ± → v ± cs in the Newtonian limit (v, cs → 0), and λ± → 1 in the


ultrarelativistic limit (v → 1).
For the sake of completeness let us say that the eigenvalues and right eigen-
vectors corresponding to the Jacobian matrices of fluxes in system (3) appear
explicitly in ref. [12]. The explicit expression of the left eigenvectors can be found
in ref. [3].
An important property of hyperbolic systems of conservation laws is that they
admit discontinuous solutions (shocks). These discontinuous solutions satisfy the
Rankine-Hugoniot (RH) jump conditions which establish the continuity of mass,
momentum and energy fluxes across shocks. In the case of SRHD these conditions
[154] read:
[ρuµ ]rµ = 0 , (9)
[T µν ]rµ = 0 , (10)
rµ being the unit normal to the hypersurface of the space–time, σ, containing
the discontinuity. The square brackets define the jump of a given variable across
σ ([F ] = Fa − Fb ; where Fa , Fb are the values of F on the two sides of σ).
Considering σ as normal to the x-axis, a suitable choice of rν is rν =
Γs (Vs , 1, 0, 0), where Vs is interpreted as the coordinate velocity of the hyper-
surface that defines the position of the shock wave and Γs = (1 − Vs2 )−1/2 is the
Lorentz factor associated to the shock.
RH conditions (9), (10) can be written in terms of the conserved and primitive
quantities and the invariant mass flux across the shock, j (j ≡ Γs Da {Vs − vax } =
Γs Db {Vs − vbx }), · ¸
x j 1
[v ] = − , (11)
Γs D
j Sx
· ¸
[p] = , (12)
Γs D
· y,z ¸
S
= 0, or [hΓ v y,z ] = 0, (13)
D
j hτ i
[v x p] = . (14)
Γs D
Equation (13) implies that the orientation of the tangential velocity does not
change across shocks. This also holds for selfsimilar expansions or rarefaction
waves. However, the values of the components of the tangential velocity, (v y , v z ),
may jump across the shock (contrary to Newtonian hydrodynamics). In the
case of contact discontinuities (no mass flux across the discontinuity, j = 0)
arbitrary jumps in the tangential velocity are allowed although pressure and
normal velocity should be continuous.
6 M.A. Aloy and J.M. Martı́

2.3 Newtonian versus Relativistic Hydrodynamics

The classical Euler equations are easily recovered from the RHD equations in
the limit c → ∞. The limits of the conserved variables and flux vectors lead to
the corresponding quantities in the classical case:
1
U = (D, S j , τ ) → (ρ, ρv j , ρv2 + ρε)
2
1
F(i) = (Dv i , S j v i + pδ ji , S i − Dv i ) → (ρv i , ρv j v i + pδ ji , v i ( ρv2 + ρε + p))
2
i, j = 1, 2, 3

The equations keep their conservative and hyperbolic characters but there are
several factors that make RHD more complex to solve numerically than classical
hydrodynamics: (i) the RHD equations are tightly coupled through Γ and h and,
therefore, they display a larger non-linearity; (ii) there is no explicit relation be-
tween W and U (except for particular EoS), i.e., obtaining the primitive from
the conserved variables needs an iterative numerical method; (iii) the tangential
flow velocity can change across discontinuities (see § 2.2) and, in addition, the
characteristic speeds may suffer from aberration (see the coupling between dif-
ferent directions in expression (7b)); (iv) in the ultrarelativistic limit (v → 1),
the eigenfields are degenerate (λ0 → λ± → 1) which triggers the appearance
of very thin structures in the flow (like, e.g., in the case of relativistic blast
waves) that may become a source of numerical errors; and (v) relativistic strong
shocks can display unbounded jumps in physical variables; e.g., for an ideal gas
(i.e., with an EoS: p = (γ − 1)ρε), the compression ratio between the post– and
pre–shocked densities of a relativistic strong shock is such that (see, e.g., [155])
ρb /ρa ≤ (γΓa + 1)/(γ − 1) which tends to infinity in the ultrarelativistic limit
(va → 1). This should be compared with the compression ratio in a Newtonian
strong shock: ρb /ρa ≤ (γ + 1)/(γ − 1) which is ∼ 4 − 7 for typical values of γ.

3 Numerical integration of the RHD equations


The first attempts to solve numerically the equations of RHD started in the late
60’s [103]. Since then and up to the 80’s the field has been dominated by the so
called artificial viscosity (AV) techniques. This was the result of the application of
von Neumann & Richtmyer’s [162] ideas (in the framework of classical hydrody-
namics) to RHD. Basically, the algorithms consisted of standard finite difference
techniques together with viscous terms added to the equations to damp spuri-
ous oscillations near shocks. The AV terms were non–consistent because they
were not based on the energy–momentum tensor of a viscous fluid. However,
they provide an artificial dissipative mechanism that makes the shock transition
smooth, i.e., extended over several numerical zones. The approach needs a large
dissipation (the amount of viscosity being problem dependent) to handle strong
relativistic shocks and, therefore, AV schemes may be very diffusive (smear-
ing out every discontinuity in the flow). In addition, the algorithms were not
3D Relativistic Hydrodynamics 7

conservative and, hence, did not guarantee the propagation of discontinuities


at the correct (physical) velocities. Due to all these problems, traditional AV
methods are not optimal both in the mildly and in the ultrarelativistic regimes
(i.e., Γ ≥ 2).
Over and above all these considerations, it remains true that AV meth-
ods have been widely used in many fields of astrophysics and we summarize
briefly some of their key contributions. The first 1D, Lagrangian, full GRHD
code (i.e., also evolving the EFE) corresponds to May & White [103] and it was
applied to stellar core collapse. However, due to the Lagrangian character of
the code, it was impossible to extend it to multidimensions. Another impor-
tant improvement was made by Wilson during the 70’s [164,165]. In ref. [164]
he presented the first multidimensional (2D), Eulerian, full GRHD numerical
code. This code (and some different versions of it) has been successfully applied
to axisymmetric stellar collapse (e.g., [127,151,148,114,111,40]), accretion onto
compact objects (e.g., [63,125]) and numerical cosmology (e.g., [24]).
As an example of the performance of the methods, we reproduce in Fig. 1b the
plot given by Norman & Winkler [118] showing the evolution of the relative errors
of Centrella & Wilson’s algorithm [24] for the case of the reflection of a mildly
relativistic shock against a wall (see Fig. 1a for a schematic representation). The
solution develops a shock that moves away from the wall. In the postshock state
the gas is at rest and much hotter than the gas in the preshocked medium. It is
noticeable from Fig. 1b that errors are larger than 5% even for values of Γ as
small as ≃ 2.3.

(a)
ρ e ρ = 1, e=0 (b)
2 2 1 1 10.0

Γ=5/3
Γ=4/3
8.0
Relative error (%)

shocked cold inflowing gas


6.0
gas
wall
4.0

v=0 |v|<c
2 1
2.0
1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4
W

Fig. 1. (a) Scheme of the relativistic shock reflection test. (b) Relative errors of the
relativistic shock reflection test as a function of the Lorentz factor (W in the plot) of
the inflowing gas for two different values of the adiabatic index (Γ in the plot legends)
using the explicit Eulerian techniques of [24]. Data from Centrella and Wilson [24].
Plot reproduced from Norman and Winkler [118].

In the mid eighties, Norman & Winkler [118] proposed a reformulation of


the difference equations with artificial viscosity consistent with the relativistic
8 M.A. Aloy and J.M. Martı́

dynamics of non-perfect fluids. Accurate results across strong relativistic shocks


with large Lorentz factors were obtained in combination with adaptive mesh
techniques. However, the strong coupling introduced in the equations by the
presence of the viscous terms in the definition of relativistic momentum and
total energy densities required an implicit treatment of the difference equations
and prevented the extension of the algorithm to multidimensions.
During the 1990’s, a major break-through in the simulation of ultrarela-
tivistic flows was accomplished when high–resolution shock–capturing (HRSC)
methods, specially designed to solve hyperbolic systems of conservation laws,
were applied to solve the RHD equations. Their application has caused a revo-
lution in numerical RHD because (i) writing the equations in conservation form,
guarantees convergence to the physically correct solution; (ii) HRSC methods
exploit the hyperbolic character of the RHD equations and, therefore, the so-
lution automatically satisfies RH conditions (this is the reason why they are
called shock–capturing); and (iii) they provide high resolution in the sense that
they have high order of accuracy in smooth regions of the flow while keeping
discontinuities stable and sharp.

3.1 Basic procedure of HRSC methods


As in any finite difference or finite volume scheme, the first step consists in
discretizing the equations on a finite numerical grid (tn , xj )
xj = (j − 1/2)∆x, j = 1, 2, . . . , (15)
tn = n∆t, n = 0, 1, 2, . . . , (16)
∆t, ∆x being the time step and the zone size, respectively. The discretization
of the system is such that the time evolution of zone averaged state vectors,
Unj , is governed by functions called numerical fluxes, F̂j±1/2 , evaluated at zone
interfaces:

dUnj 1 ³
=− F̂j+1/2 (Unj−r , Unj−r+1 , . . . , Unj+q )−
dt ∆x ´
F̂j−1/2 (Unj−r−1 , Unj−r , . . . , Unj+q−1 ) .

where p and q are positive integers. Usually, Unj is an approximation to the zone
average of U defined by
Z xj+1/2
1
Unj = U(tn , x)dx , (17)
∆x xj−1/2
which is consistent with the integral form of the system of conservation laws
and, F̂j±1/2 are the time–averaged fluxes across the interfaces xj±1/2 (xj±1/2 =
(xj + xj±1 )/2) :
Z tn+1
1
F̂j±1/2 ≈ F(U(t, xj±1/2 ))dt . (18)
∆t tn
3D Relativistic Hydrodynamics 9

Computation of numerical fluxes. Numerical fluxes must be consistent with


the actual fluxes, i.e., F̂(U, U, . . . , U) = F(U). In (18) the only unknown is the
value of the solution at every cell interface, U(t, xj±1/2 ) and its dependence
on time. There are several ways of calculating (18), that give rise to different
HRSC schemes. One of the most fruitful subset of methods are called Godunov–
type methods (because they were first used in Godunov [51]). The discretization
procedure approximates the solution by a piecewise continuous function at each
time step (see Fig. 2 bottom panel) and, therefore, at every cell interface a
jump between adjacent zone averaged values may occur. The situation is such
that locally (at each interface) at every time level, tn , one has an IVP with
piecewise constant initial data for the RHD equations, i.e., a Riemann problem
(RP; see, e.g., [157]). In Godunov-type methods, U(t, xj±1/2 ) is calculated by
solving these RPs at every zone interface. The exact solution of the RP for the
Euler equations is known (e.g., [30] and references therein) and consists of a
set of constant states separated by centered rarefactions, contact discontinuities
and/or shocks connecting the piecewise constant left and right states. The
top panel of Fig. 2 shows a space–time diagram of the evolution of the solution
at several adjacent cell interfaces. The time step between two consecutive time
levels, tn , tn+1 must be limited in order to prevent the interaction of the solution
of adjacent interfaces. This restriction on dt is known as Courant–Friedrich–Lewi
(CFL) condition.

High order of accuracy. In order to understand the meaning of high resolu-


tion we need to introduce some mathematical concepts (see, e.g., [87] for details).
We say that the solution converges under grid refinement when the global error
||E∆x ||, defined as X
||E∆x || = ∆x |Ūnj − Unj |, (19)
j

tends to zero as ∆x → 0 (Ūnj is the average of the true solution in the cell j).
The idea behind definition (19) is that the finer the grid is, the better the numer-
ical solution. Lax equivalence theorem [133] asserts that stability is a necessary
condition to guarantee convergence. A measure of the stability of a solution can
be its total variation at t = tn , TV(Un ), defined as
+∞
X
TV(Un ) = |Unj+1 − Unj | . (20)
j=0

A numerical scheme is said to be TV–stable, if TV(Un+1 ) ≤ TV(Un ), ∀n for


any initial data. For non–linear, scalar conservation laws, a numerical scheme
converges if it is written in conservation form with consistent numerical flux
functions, and if it is TV–stable [87].
High–order accuracy in smooth parts of the flow is achieved by using con-
servative monotonic polynomial functions to interpolate the approximate solu-
tion within numerical cells. The reason to choose monotonic functions is that
10 M.A. Aloy and J.M. Martı́

rarefaction shock contact discontinuity


n+1
t

n
u j-1 u jn u nj+1 n
u j+2 n
t
x j-1 x j-1/2 xj xj+1/2 x j+1 x j+3/2 xj+2

continuous solution
discrete solution

x j-1 xj x j+1 x j+2

Fig. 2. Godunov’s scheme: local solutions of Riemann problems. At every interface,


xj− 1 , xj+ 1 and xj+ 3 , a local Riemann problem is set up as a result of the discretiza-
2 2 2
tion process (bottom panel), when approximating the numerical solution by piecewise
constant data. At time tn these discontinuities decay into three elementary waves which
propagate the solution forward to the next time level tn+1 (top panel). The time step of
the numerical scheme must satisfy the Courant-Friedrichs-Lewy condition, being small
enough to prevent the waves from advancing more than ∆x/2 in ∆t.

they lead to a decrease of the total variation of the solution (total–variation–


diminishing schemes, TVD; [61]), ensuring stability. This interpolation or cell
reconstruction provides more accurate left and right states for the Riemann
problem by substituting the mean values Unj (that produces only first–order ac-
curacy) by better representations of the true flow near the interfaces, thereupon,
new RPs are setup at each cell interface. Different choices of the interpolation
polynomial lead to TVD schemes of different order of accuracy. Piecewise con-
stant functions provide first order accuracy (this was the original method of
Godunov [51]). Second order of accuracy is obtained if piecewise linear functions
are used (e.g., monotonic upstream scheme for conservation laws – MUSCL –
method [158]) and, piecewise parabolic functions provide third order accuracy
( piecewise piecewise method – PPM – [28]). Albeit it is possible to construct
higher order polynomials to obtain better accuracy in smooth parts of the flow, it
should be remarked that TVD methods degenerate to first order accuracy at ex-
trema [119]. Therefore, other types of reconstruction (not being TVD) have been
developed like, e.g., total –variation–bounded (TVB) schemes [146], the essen-
3D Relativistic Hydrodynamics 11

tially non–oscillatory (ENO) schemes [62] and the piecewise–hyperbolic method


(PHM) [93]. In other approaches, like the FCT algorithm (see, e.g., [20]), a high
order of accuracy is obtained by adding an anti–diffusive flux to a first order
accurate numerical flux.

Time advance. There are two main procedures to advance the conserved vari-
ables in time. One possibility is a standard discretization of the time derivative
in system (17). In this case, the accuracy of the method depends on the order
of accuracy up to which the numerical fluxes have been computed. The sec-
ond alternative (known as the method of lines) regards (17) as a semidiscrete
(i.e., spatially discretized) system of ordinary differential equations to which any
standard ordinary differential equation solver (e.g., a Runge Kutta solver) can
be applied.

3.2 Riemann Solvers


The key ingredient of Godunov–type methods is the solution of the RP at zone
interfaces. Nevertheless, there is no exact closed–form to the RP neither for
the RHD equations nor for the classical Euler equations, not even for ideal
gases. Iterative algorithms are devised whereby the solution can be computed
numerically up to any practical degree of accuracy. Numerical schemes aimed to
solve RPs are known as Riemann solvers.
The solution of the RP can be either exact or approximate and, therefore,
the Riemann solvers can be, at first instance, classified as exact or approximate
Riemann solvers. Although the exact solution to the RP is known, there are a
number of reasons to compute approximate solutions. The first is that the preci-
sion of any difference method is finite and, actually depends on the order of the
method. Hence, an approximate RP solution with an accuracy better than that
of the finite difference scheme might be acceptable. The second reason is that
exact Riemann solvers are computationally expensive algorithms while approxi-
mate ones are more efficient. That is particularly important in multidimensional
calculations because then numerical efficiency is a main issue. Finally, the so-
lution of the RP is averaged in time to compute the numerical fluxes (see 18)
and, therefore, the fine details of the solution are lost. Hence, it may pay off to
evaluate an approximate (but accurate enough) solution to the RP.
Among the class of exact Riemann solvers, Godunov [51] is credited with
the first exact Riemann solver for the Euler equations. Martı́ & Müller [98] ex-
tended this classical solution of the RP to the SRHD equations when the veloc-
ities parallel to the initial discontinuity (tangential velocities) are zero. Pons, et
al. [130] have given the general solution (for non–vanishing tangential velocities)
for the RP in SRHD. Wen, Panaitescu & Laguna [163] have extended Glimm’s
method or Random Choice Method [50,25] to 1D SRHD. Balsara [11] and Dai &
Woodward [31] have extended to SRHD the two–shock approximation method
of Colella [27] for classical fluid dynamics.
The strategy to construct approximate Riemann solvers (in both classical
and relativistic hydrodynamics) relies on the local linearization of the Jacobian
12 M.A. Aloy and J.M. Martı́

matrices of the flux vectors, B(i) , extending a procedure introduced by Roe [132]
and sometimes referred to as local characteristic approach (LCA). The idea of
the LCA is to use the spectral decomposition of B(i) to rewrite the original sys-
tem as a new one of uncoupled scalar equations, in terms of the characteristic
variables. The locally linear system can then be easily solved to obtain appropri-
ate numerical fluxes for the original system. Sometimes the linearization process
involves an averaged intermediate state at zone interfaces. This is the case of the
original Roe solver [132] and its relativistic extension [39], and Martı́ et al. [97]
or Falle & Komissarov [41] approaches. Donat & Marquina [36] have extended a
numerical flux formula which was first proposed by Shu & Osher [147] for scalar
equations to systems. In the scalar case and for characteristic wave speeds which
do not change sign zone interfaces, Marquina’s flux formula is identical to Roe’s
flux. Otherwise, the scheme switches to the more viscous, entropy satisfying lo-
cal Lax–Friedrichs scheme [147]. In the case of systems, the combination of Roe
and local–Lax–Friedrichs solvers is carried out in each characteristic field after
the local linearization and decoupling of the system of equations [36]. However,
contrary to Roe’s and the previously cited linearized methods, the extension of
Marquina’s method to systems does not require on any averaged intermediate
state. Marquina’s flux formula has been successfully used in the ultrarelativis-
tic regime and in 2D axisymmetric [100,101] and 3D SRHD [3] and even in 3D
GRHD [7,44].
Finally, a very simple approach is an extension of the Harten, Lax, van Leer
(HLL) solver to SRHD [139]. This method avoids the explicit calculation of the
eigenvalues and eigenvectors of the Jacobian matrices and is based on an approx-
imate solution of the original Riemann problems with a single intermediate state,
U∗ , determined by requiring consistency of the approximate Riemann solution
with the integral form of the conservation laws in a grid zone. The algorithm
needs estimates of lower and upper bounds for the smallest and largest signal
velocities, aL and aR , respectively. Good estimates for aL and aR are essential to
guarantee robustness and, at the same time, the minimal amount of numerical
viscosity (the larger the difference ||aL | − |aR || is, the larger the viscosity of the
method; if the difference is too small, undesired numerical oscillations around
discontinuities appear). In the non-relativistic case, Einfeldt [37] proposed to
use the smallest and largest eigenvalues of Roe’s matrix (this corresponds to the
HLLE solver). Duncan & Hughes [33] have generalized the method to 2D SRHD.

3.3 Other HRSC schemes


Symmetric TVD schemes with nonlinear numerical dissipation. Sym-
metric TVD (sTVD) methods [32] are another subset of the HRSC methods.
As Godunov–type methods, they are written in conservation form, but they
are not based on solving Riemann problems. sTVD methods use standard finite
difference methods (e.g., Lax Wendroff scheme) and employ local conservative
dissipation terms in order to stabilize the algorithm across discontinuities. The
numerical dissipation term is local, free of problem dependent parameters and
does not require any characteristic information. Hence, they are simpler than
3D Relativistic Hydrodynamics 13

Godunov–type schemes. Extensions to 2D and 3D general relativistic magneto


hydrodynamics (GRMHD) can be found in, e.g., [74,73,116].

Relativistic beam scheme. In the beam scheme [136] and its relativistic
extension [167] the hydrodynamic equations are solved as the limit of the cor-
responding Botzmann equation. The velocity distribution functions are approx-
imated by several Dirac delta functions or discrete beams of particles in each
computational cell, which reproduce the appropriate moments of the distribu-
tion functions. This beam scheme, although being a particle method derived
from a microscopic kinetic description, has all the desirable properties of mod-
ern characteristic–based wave propagation methods based on a macroscopic con-
tinuum description. Yang et al. [167] show that the integration scheme for the
beams can be cast in the form of an upwind conservation scheme in terms of nu-
merical fluxes and build up high–order variants of the relativistic code in terms
of different TVD and ENO interpolations.

3.4 Other approaches.


Van Putten’s approach. Van Putten [159] solves the equations of (ideal)
special relativistic magneto–hydrodynamics (SRMHD) formulating Maxwell’s
equations as a hyperbolic system in divergence form. State vectors and fluxes
are decomposed into a spatially constant mean and a spatially dependent vari-
ational part. Then, the SRMHD equations become an evolution system for in-
tegrated (continuous) quantities for which standard integration methods can be
used. In order to update the state vectors of the original system, a numerical
differentiation of the integrated quantities is necessary. This process can lead
to oscillations in the case of strong shocks and a smoothing algorithm needs to
be supplied. The approach has been applied to SRMHD jets with values of the
Lorentz factor up to 4.25 [160,161].

Relativistic Smoothed Particle Hydrodynamics. Smoothed particle hy-


drodynamics (SPH; [89]) represents a fluid by a Monte Carlo sampling of its mass
elements. The motion and thermodynamics of these mass elements is governed
by the hydrodynamic equations. As the equations involve continuous proper-
ties of the fluid, it is necessary to estimate these quantities from the positions,
velocities and internal energies of the fluid mass elements, which are like par-
ticles moving with the flow. SPH treats the particle positions as a finite set of
interpolating points where the continuous fluid variables and their gradients are
estimated by an appropriately weighted average over neighboring particles. This
means that it is a free–Lagrange method, i.e., spatial gradients are evaluated
without the use of a computational grid. As a result, the method provides a
genuinely multidimensional (3D) description of the fluid with a lower computa-
tional cost (depending on the number of particles employed) than typical finite
volume, HRSC, multidimensional methods. SPH has been extended to SRHD by
Monaghan [109]. Other SRHD applications can be found in, e.g., [84,91,92,26].
14 M.A. Aloy and J.M. Martı́

Implementations of SPH in GRHD may be seen in [70,83,145]. The artificial vis-


cosity (necessary to handle discontinuities in the fluid) is the most critical issue
for relativistic SPH codes. The reason is that, unlike its Newtonian analogue, the
relativistic theory has not yet been developed to the degree required to predict –
by a relativistic kinetic theory – the form of the dissipation terms. To overcome
this problem, the most successful approach consists on taking the concepts from
Riemann solvers as a guide to improve the artificial viscosity required in SPH
(see [110,26]).

4 Computational issues and current 3D RHD codes


4.1 Computational issues
Due to the extreme computational resources demanded by typical 3D RHD
simulations – as compared to 1D or 2D RHD ones –, technical improvements
of the codes are required. Among the most important issues that should be
addressed we find the optimization of the memory and the algorithm itself and,
the design of suitable strategies for data analysis.

Memory handling optimization. Even small problems in 3D require huge


amounts of RAM memory. As an example, 15 variables (e.g., a set of five con-
served variables, five primitive variables and five fluxes) discretized in a volume
containing 2563 cells, in double precision (×8 bytes), would need ∼ 2 Gb of RAM.
Moreover, this number of 3D variables is a very moderate value that can easily
be multiplied by a factor of ∼ 4 if a GRHD problem with dynamical evolution
of the space–time is considered, or if several chemical species are present in the
fluid. Thereby, a reduction towards the minimal set of 3D variables is mandatory.
In addition, the algorithms should be designed to allow for an efficient memory
access (which is a computer dependent issue), the reason being that nowadays
the speed of the memory systems is about one order of magnitude smaller than
the actual CPU speeds. This last point is particularly critical when superscalar,
cache–oriented computers (the most commonly used systems in supercomputing,
at present) are used. Finally, high order methods are preferred in 3D, because
less grid points are needed to resolve fine structures in the fluid (as compared
with lower order methods).

Algorithm optimization. A typical 3D RHD code, using an ideal EoS, may


need about 10−4 s to update a numerical zone on a single processor. If one has
to update 2563 zones, the time per code iteration is roughly 30 minutes per
processor. This means that 1000 time steps may need several days to a few
weeks to be completed. This time estimate, can grow drastically if, e.g., realistic
microphysics and/or EoS are used. Hence, code efficiency becomes important
in order to reduce the total computational time. An obvious way of decreasing
the run time is to simplify analytically all the expressions, in particular those
that may lead to numerical cancellation errors [5]. However, the largest impact
3D Relativistic Hydrodynamics 15

on performance comes from computer architecture considerations, i.e., it is nec-


essary to write algorithms oriented to the type of computer that is going to
be used (cache based, vector based, etc.). Code parallelization and/or vector-
ization is unavoidable to work on current (both distributed or shared memory
based) supercomputers. Finally, given that the affordable resolution in fixed 3D
grids is still far from being appropriate to study many problems – even with the
largest supercomputers –, resolution enhancement algorithms are crucial (like,
e.g., adaptive mesh refinement [15], multiresolution [60]).

Data analysis. Applying the same data analysis than in 1D or 2D it is neither


appropriate nor feasible in 3D, the reason being the huge data sets to be stored
(an analyzed). A crude estimate of the typical storage requirements necessary to
save every variable at every time step leads to unpractical (if not at all affordable)
values of the order of terabytes of disc space. Albeit, it is convenient to design a
priori the analysis strategy, i.e., it is convenient to choose which variables have to
be saved. An example of this kind of procedure is the synthesis of emission maps
from simulations of relativistic jets including time–delay effects (see Sect. 5). In
such a case, the values of the state variables in many different time levels are
necessary in order to integrate the emission along rays parallel to the line of sight
(for a given observing angle). In practice the number of time levels required to
make the full integration is of the same order than the number of time steps of the
simulation, which means that terabytes of data should be saved. An alternative
is to choose a priori the viewing angles that are interesting and perform the
integration on run time – picking out the values from the appropriate zones at
each time level for every angle –.

4.2 Technical status of the existing 3D RHD codes

There are, according to our knowledge, four different 3D RHD codes, used in
astrophysical applications: Koide–Nishikawa’s code [74,76], GENESIS [3], Cactus
[44] and Shibata–Nakamura’s code [143]. In the following we will describe some
of the main features of each one.
The group of Koide and coworkers has developed a GRMHD code (for a
fixed background metric) based on a sTVD scheme (see Sect. 3.3). The code
has been applied to study the propagation of extragalactic jets through magne-
tized atmospheres [115] and also to simulate the early stages of jet formation by
magnetohydrodynamic mechanisms in background BH spacetimes [75]. A series
of tests of Koide et al.’s method involving mainly continuous solutions can be
found in ref. [76]. Koide et al.’s code has proven to be very stable (although very
difussive) when simulating mildly relativistic flows (maximum Lorentz factors
≈ 4) with discontinuities.
GENESIS [3] is a conservative 3D RHD code (used in astrophysical SRHD
and GRHD applications) based on HRSC techniques. It uses Marquina’s flux for-
mula to compute numerical fluxes and a third order PPM spatial interpolation.
The time advance is performed by means of a high–order Runge–Kutta method.
16 M.A. Aloy and J.M. Martı́

Relativistic Shock Tube Relativistic Shock Tube Relativistic Shock Tube


Marquina Solver [nx,ny,nz]=[128,128,128] t=0.4 MarquinaSolver
Marquina Solver[nx,
[nx, ny, nz]
nz]==[100,
[100,100,
100,100]
100],t=0.5
t=0.5
1.2

analytic solution
1.0 density/10
pressure/20
total velocity
0.8

0.6

0.4

0.2

0.0

−0.2
−0.5 −0.3 −0.1 0.1 0.3 0.5
main diagonal main
maindiagonal
diagonal

Fig. 3. Comparative performance of Cactus (left panel; figure from [44]) and GEN-
ESIS (right panel; figure from [3]) on a three–dimensional mildly relativistic shock
tube test. The panels show several primitive variables along the main diagonal of
the computational domain (solid line: analytic solution; symbols: numerical solu-
tion). The initial data are a constant left (L) and right (R) state characterized by
(ρL , pL , vL ) = (10, 13.3, 0) and (ρL , pL , vL ) = (1, 0.66 · 10−6 , 0). Numerical grid and
evolution time is similar in both cases.

GENESIS has been extensively tested in problems involving strong shocks even
in the ultrarelativistic regime, being able to handle Lorentz factors as large as
3 · 105 (in the one–dimensional wall reflection test). Simulations of extragalactic
jets (using SRHD, at both parsec [6] and kiloparsec [4] scales) and progenitors
of GRBs (in GRHD with a Schwarzschild background metric, [7]) have been
performed with GENESIS.
Cactus is a numerical tool developed within a collaboration of the Numerical
Relativity divisions at the National Center for Supercomputing Applications,
the Albert Einstein Institut and the Washington University [22]. Cactus is able
to solve the full set of Einstein field equations coupled to a perfect fluid source.
The hydrodynamic evolution employs HRSC methods and can be computed by
means of different Riemann solvers (Roe, Marquina) and flux–splitting schemes.
The metric evolution may be followed with several formalisms (ADM, hyperbolic
formulations, conformal–tracefree) combined with a number of integration meth-
ods (leapfrog, Crank–Nicholson, MacCormack along with Strang splitting) and
various gauges (algebraic, maximal slicing, etc.). The main goal of the code is
the simulation of astrophysical processes involving NSs and BHs. Font et al. [44]
have tested the capabilities of the hydro part computing shock tubes. In Fig. 3,
a comparison between Cactus and GENESIS is shown for a mildly relativistic
Riemann problem. Despite the slightly different grids (1003 for GENESIS, 1283
3D Relativistic Hydrodynamics 17

for Cactus) the results are still comparable, the reason being that both codes
use the same HRSC techniques. Font et al. [44] have also tested the GRHD cou-
pling in Cactus simulating Friedman–Robertson–Walker cosmologies with dust
and polytropic NSs (static and boosted). Alcubierre et al. [2] have performed a
number of experiments (Brill waves, single BH, static boson and neutron stars)
comparing different formulations of the Einstein equations. Recently, Landry &
Teukolsky [86] have performed simulations of coalescing binary NSs. The sta-
bility of the Einstein evolution is a main issue of this code, especially when the
ADM or hyperbolic formulations are considered.

Γ = 22360

Fig. 4. Comparative performance of Shibata–Nakamura’ code (left panel; figure from


[143]) and GENESIS (right panel; figure from [3]) on the shock reflection test. In both
cases the abscissas represent the distance to the reflection point. Left panel shows,
from top to bottom, the density, pressure and flow velocity distributions, for an inflow
velocity of 0.4c. In the right panel, the density distribution (in logarithmic scale) is
shown for a sample of inflow velocities –from 0.9c to 0.999999999c–.

Shibata–Nakamura’s code is based on the conformal–tracefree (CT) formu-


lation of the Einstein equations [141,140,142]. The hydrodynamic equations are
18 M.A. Aloy and J.M. Martı́

written (and solved) as a set of advection (i.e., non–conservative) equations to


which AV is added as in Wilson’s traditional approach (see § 3). The CT formu-
lation is employed for the Einstein equations which are evolved by means of a
leapfrog scheme. Several gauges are available (maximal slicing, minimum distor-
tion). Shibata & Nakamura [140] have used this code to study the gravitational
wave production of selected matter configurations. Rigidly rotating neutron star
stability and neutron star mergers in 3D were computed in [113,143]. For the first
time in 3D full GR, the dynamical collapse of a rotating neutron star has been
simulated with this scheme [144]. Although the code –and the CT formulation–
allows for the longest stable evolutions in dynamical space–times, the quality of
the results, as measured by the degree of fulfillment of the constraint Einstein
equations, is not as satisfactory as in the ADM or hyperbolic formulations. A
comparison between the quality of the results of Shibata–Nakamura’s code and
GENESIS (using HRSC methods) is shown in Fig. 4 for the shock reflection
test in a flat space–time. Shibata & Nakamura’ results (for an inflow velocity as
small as 0.4c) display the well known pathologies of traditional relativistic AV
methods in the treatment of shocks (diffusion, oscillations). HRSC techniques
are much less diffusive (even with Lorentz factors as large as 22360) and produce
more stable profiles in the postshock state.

5 Applications

In this Section we will review briefly some of the most relevant 3D applications
of SRHD and SRMHD codes in astrophysical scenarios (i.e., relativistic jets) as
well as recent GRHD and GRMHD simulations in the context of progenitors of
GRBs and jet formation, respectively. Applications to dynamical space–times,
in which Einstein equations are coupled to the GRHD equations, are beyond the
scope of this article (for a review see, e.g., [45]).

5.1 Relativistic astrophysical jets

In the standard model [19,138] the elongated radio structures connected to the
center of AGNs in radio galaxies and radio–loud quasars are considered as con-
tinuous ejections of highly collimated, supersonic and very stable plasma. The
emission is produced by synchrotron and inverse Compton processes of electrons
accelerated up to ultrarelativistic energies in the vicinity of a central engine.
The asymmetries in the radio flux of the two oppositely directed jets of a source
and the superluminal motions observed in a few dozens of compact sources are
explained by assuming that both jets propagate with relativistic speeds along
directions at small angles to the line of sight. The relativistic Doppler beaming
of the emission in the direction of motion accounts for the observed emission
asymmetries whereas apparent superluminal speeds are explained by the com-
bination of a finite value of the speed of light and the relativistic motion of the
emitting source.
3D Relativistic Hydrodynamics 19

The formation, collimation and propagation of extragalactic jets involves


scales ranging from some microparsecs to hundreds of kiloparsecs. Thereby, the
study of the jet phenomenon is conveniently chopped into several pieces each
one covering a smaller range of length scales. The mechanism governing the jet
formation and collimation is still a challenge, mainly because the most detailed
high–frequency VLBI observations of nearby radio sources can resolve at most
the compact radio cores with a linear resolution of ≥ 0.1 pc [17] 1 , while the
Schwarzschild radius (Rs = GM/c2 ) of a 109 M⊙ galactic BH (the supposed
central engine in the most commonly accepted interpretation) is ∼ 10−5 pc.
Consistently, our theoretical view of jet formation is mainly constrained by the
fact that many jets are well collimated by the time they have propagated to a
distance ≤ 1pc from the nucleus (e.g., [69]). Several mechanisms of jet formation
have been proposed, all of which present some difficulties [13]. Additionally, there
is a wide variety in the observed properties of jets, so there may be a variety
of jet collimation mechanisms. The standard model assumes that jet formation
involves accretion onto a central compact object, such as a NS or a stellar BH
in the case of galactic microquasars (GRS 1915+105 [107] and GRO J1655-40
[156]), or a supermassive BH in AGNs. This accretion fuels bipolar outflows
which are further collimated and accelerated by MHD processes at subparsec
scales (see, e.g., [81]).
At parsec scales (from 0.1 pc to 100 pc) jets are observed using VLBI imaging.
VLBI radio maps display highly collimated jets whose morphology is character-
ized by a bright spot at the jet end and a series of components which separate
from the core, sometimes at superluminal speeds. Many parsec scale jets show
intraday variability of the radio flux, excess in brightness temperatures and one-
sidedness. The interpretation of these features is that the jet material moves at
small angles to the line of sight with bulk Lorentz factors Γ ≃ 15h−1 65 [49] (h65 is
Hubble’s constant in units of 65 km/s Mpc−1 ), or even larger (Γ ≃ 30 − 100) if
the intraday variability is intrinsic and a result of incoherent synchrotron radia-
tion [14]. The moving components are interpreted, within the shock–in–jet model
as traveling shock waves [95,54]. A worthy byproduct of this model is the expla-
nation of the complex multifrequency brightness and polarization variations in
blazars.
At kiloparsec scales, sources are shaped by the interaction with the external
medium, with a morphological dichotomy between FRI and FRII [42] sources
whose basis seems to be the source power. The morphology of FR I sources is
the result of a deceleration from relativistic to non–relativistic, transonic speeds
[16,85], occurring within the first kpc. In the case of the FRIIs, mildly relativistic
jet speeds (Lorentz factors between 2 and 3) well outside the galaxy have been
inferred from flux asymmetries between jets and counter–jets [21] and superlu-
minal motions at kpc scales (e.g., in 1928+738 [65] and 1055+201, 1830+285,
2209+080 [64]). The mechanism by which the relativistic flows inferred from
radio jets at pc scales remain collimated up to kpc scales is still unresolved.
1
The VLBI Space Observatory Program (VSOP), allows for a linear resolution of
∼ 0.1 − 1 pc even in distant sources like S5 0836+710 (z = 2.17; [88]).
20 M.A. Aloy and J.M. Martı́

Large scale jets. At far enough distances from the central object (pc and
kpc scales) the effects of gravity are negligible and we can assume (as a first
approximation to the problem) that magnetic fields are dynamically unimpor-
tant. In this approach, numerical simulations using a pure SRHD treatment have
been performed since the early 90s to study the morphology and dynamics of
relativistic jets.
The development of codes based on HRSC techniques has allowed 2D axisym-
metric time–dependent relativistic hydrodynamic simulations [33,38,99–101,80,134]
to be performed. These simulations led to the conclusion that both the internal
energy and the Lorentz factor of the beam enhance the stability of relativistic
jets compared to their classical counterparts, through the increase of the effective
inertial mass of the beam. Relativistic MHD simulations in 2D using pseudo–
spectral techniques [159,161] or sTVD methods [74,73] have been another step
forward in our understanding of relativistic astrophysical jets.
Only since 1998 the morphology and dynamics of relativistic jets is studied
with 3D SRHD [3,4] or SRMHD [115,116] simulations. Aloy et al. [3] did a 3D
simulation (in Cartesian coordinates) of an axisymmetric relativistic jet propa-
gating through an homogeneous atmosphere. The simulated jet is characterized
by a beam–to–external proper rest–mass density ratio η = 0.01, a beam Mach
number Mb = 6.0, and a beam flow speed vb = 0.99c. An ideal gas equation
of state with an adiabatic exponent γ = 5/3 is assumed to describe both the
jet matter and the ambient gas. The beam is in pressure equilibrium with the
ambient medium which fills a domain (X,Y,Z) with a size of 15Rb × 15Rb × 75Rb
(120 × 120 × 600 cells), where Rb is the beam radius. The jet is injected at z = 0
in the direction of the positive z-axis through a circular nozzle. Simulations were
typically performed with 16 R10000 processors (on a SGI-Origin 2000) and need
about ten thousand time iterations. The execution time was about 100 hours.
Genuine multidimensional effects were included by perturbing the axial injection
velocity.
Koide and collaborators [115,116] simulated the evolution of relativistic jets
through a magnetized uniform atmosphere during a very brief period of time. The
numerical setup [116] consisted on a cylindrical jet injected through a circular
nozzle into an oblique (45◦ with respect to the jet axis) magnetic atmosphere.
The computational domain was a cubic box of 20Rb ×20Rb ×20Rb (101×101×101
cells). The jet had the following parameters: vb = 0.98756c, ratio between the
beam and magnetic specific energy densities εb /εm = 1/3, Mb = 4.0, γ = 5/3
and considered two different strengths for the ambient magnetic field, weak and
strong. The simulation lasted for 35 CPU hours on a SGI–Power Challenge and
required about 1 Gb of RAM. It should be remarked that the coarse grid zoning
used in Nishikawa’s [116] simulations (5 cells/Rb ), prevented them from studying
genuine 3D effects in relativistic jets in detail.

Parsec scale jets. The presence of emitting flows at almost light speed en-
hances the importance of relativistic effects in the apparency of jets. This fact
is stressed in the case of parsec scale jets, triggering the combination of syn-
3D Relativistic Hydrodynamics 21

chrotron emission models and hydrodynamic simulations to compare with ob-


servations. Theoretical models aimed at explaining the appearance of radio jets
[68,94,95,72] studied the spectrum of the synchrotron emission produced in dif-
ferent parts of the jet and the perturbations induced on the spectrum due to jet
inhomogeneities. The inhomogeneities were introduced to mimic the ejection of
components in real sources. The (frequency dependent) light curves of both total
and polarized flux of the ejected components are understood as enhanced emis-
sion behind relativistic shock waves propagating down the underlying steady jet
( shock–in–jet model).
The theoretical models gave rise to more detailed numerical models (see,
e.g., [67,96,53]) which, however, were restricted to a simplified (one-dimensional)
hydrodynamic evolution. In 1995, Gómez et al. [55] produced the first numerical
simulations of the pc scale synchrotron emission from 2D SRHD jets. Since then,
other works have followed the same approach [56,34,106,79].
Very recently, Aloy et al. [6] have computed, for the first time, the emis-
sion from 3D relativistic jets. As in the previous 2D simulations, the approach
followed by Aloy et al. rested on two main points. First the jet structure is
calculated using a relativistic time–dependent hydrodynamic code. Second, the
radio emission from the hydrodynamic jet models is computed by integrating
the transfer equations of synchrotron radiation, once the appropriate relations
between the computed hydrodynamic quantities and the emission/absorption
coefficients are established. The procedure accounts for the appropriate opacity
and relativistic effects, such as Doppler boosting and relativistic aberration. The
simulations were done with the code GENESIS [3] and the radio emission was
calculated with the same code as in Gómez et al. [55,56] (see also [58]). Aloy
et al. [6] discuss some observational consequences of the interaction between the
relativistic jet and the surrounding medium, leading to the development of a
shear layer. The presence of such a layer (with distinct kinematic properties and
magnetic field configuration) has been invoked ad hoc in the past by several
authors [78,85] to account for a number of observational characteristics in FR I
radio sources. However, its physical nature is still largely unknown. Remarkable
effects associated with this layer are, e.g., the presence of rails of low polariza-
tion intensity (see emission dips in Fig. 5–right) along the shear layer. Recently,
Swain, Bridle, & Baum [152] have found evidence of such shear layers in FR II
radio galaxies (3C353), and Attridge, Roberts, & Wardle [10] have inferred a
two-component structure in the parsec scale jet of the source 1055+018.

Jet formation Within the framework of General Relativity, GRMHD 2D ax-


isymmetric simulations of early phases of jet formation from magnetized accre-
tion disks around a non–rotating (Schwarzschild) BH [76] or rotating (Kerr) BH
[77] have been performed, assuming a fixed background metric. The magnetic
stress induced by the axial magnetic field anchored to the disk (in Keplerian
rotation) is responsible of the angular momentum loss of the disk material, trig-
gering the accretion process. In the non–rotating Schwarschild BH case [76] a jet
is formed with a maximum velocity of 0.93c (W = 2.7) if a hydrostatic corona is
22 M.A. Aloy and J.M. Martı́

Fig. 5. Figure from [6]. Logarithm of the integrated total ( left) and polarized ( right)
intensity across the jet for different viewing angles. Lines are plotted in intervals of 10◦
from an angle of 10◦ (top line in both plots), to 90◦ (showing a progressive decrease in
emission). Dashed lines (dot dashed) correspond to an observing angle of -100◦ (-140◦ ).
Units are normalized to the maximum in total intensity.

considered and with a maximum velocity of 0.4c if a free falling corona embed-
ding the black hole is assumed. A two layered jet structure (in agreement with
theoretical predictions – e.g., [18] –) is found. The inner part is pressure driven
and moves at relativistic speeds in the hydrostatic corona case. The outermost
part is magnetically driven and subrelativistic. This shell structure might be
the origin of the shear layer mentioned in the previous section. Nevertheless,
for a fast rotating BH [77], the maximum velocities obtained are 0.4c (counter–
rotating disk) and 0.3c (co–rotating disk), and again the two layered outflow
structure is formed (see Fig. 6). As the simulations had to be stopped due to nu-
merical problems quite early [77], the total evolution time was not large enough
to develop highly relativistic jets. Therefore, despite of the promising results of
Koide’s group, the mechanism of jet formation still remains an open and chal-
lenging question (see, e.g., [135]).

5.2 Gamma–ray bursts

GRBs are known observationally since over 30 years [71]. They consist of very
short, non–repeating events (except for a few soft gamma–ray repeaters), with
a typical duration between several milliseconds and several hundreds of seconds,
showing a large variability even at millisecond scale. They show a bimodal time–
distribution, the border between the two groups being at ∼ 2 s. The first group is
composed of short bursts centered around 0.1 s, while the second group consists
of long bursts (more numerous and softer than the first group) centered at
about 15 s. The time–structure is very different from burst to burst.
GRB spectra are non–thermal. The observed energy flux as a function of the
energy can be well described by one or a combination of several power laws.
The maximum of the energy distribution corresponds to an energy (the energy
3D Relativistic Hydrodynamics 23

Fig. 6. Figure from [76]. Initial and final time snapshots of the logarithm of the density
around a non rotating Schwarzschild BH. Top panels: free falling corona case. Bottom
panels: hydrostatic corona case. The solid lines are the magnetic field lines. The vector
plots show the flow velocity. On the initial state a uniform axial magnetic field is set
up. The disk (in white color) rotates around the BH with Keplerian velocity. In the
right panels the jet is formed almost along the magnetic field lines.

peak), which is characteristic of each GRB and usually is about several hundreds
of keV. The observed fluence on earth is 10−5 − 10−7 erg/cm2 .
For years it was unclear whether GRBs take place at local or cosmological
distances (see e.g., [105]). However, a galactic origin can ve excluded, because
the BATSE catalog shows an isotropic distribution of GRBs over the sky [104].
BeppoSAX spacecraft [29] has provided accurate coordinates (∼ arc minutes) of
the fading X–ray counterparts of GRBs, which has allowed for subsequent ground
based observations of faint GRB afterglows at optical and radio wavelengths.
The recent redshift determinations, obtained from the optical spectra, prove
24 M.A. Aloy and J.M. Martı́

the cosmological origin at least of the majority of GRBs (see [48] for more
information). Observed redshift values are in the range 0.7 ≤ z ≤ 3.4 implying
emitted gamma-ray energies of 2 × 1051 ≤ E ≤ 2.3 × 1054 erg for an isotropically
radiating source. The cosmological origin of the GRBs is consistent with the
distribution of bursts in the log N − log P plane.
Nonetheless, the accuracy of the positioning is neither sufficient to determine
the host galaxies of GRBs nor their progenitors. This picture was challenged by
the detection of the Type Ib/c supernova SN 1998bw [46,47] within the error box
of GRB 980425 [149,126] whose explosion time and location is consistent with
that of the GRB. This suggests a relationship between GRBs and SNe Ib/c,
i.e., core collapse supernovae of massive stellar progenitors which have lost their
hydrogen and helium envelopes [47,66]. However, the observation of a second fad-
ing X–ray source within the error box of GRB 980425 (different from SN 1998bw)
still causes some doubts on the GRB–supernova connection, although the prob-
ability of chance coincidence of GRB 980425 and SN 1998bw is almost negligible
[126].
Another clue on the nature of the progenitors of GRBs comes from the du-
ration of the shorter bursts and the temporal substructure of the longer bursts
(∼ 1 msec). If the time variation is intrinsic, the length scales involved in the
production of a GRB are of about 1 light–millisecond, which in turn points to-
wards compact objects, like NSs or BHs. Furthermore, the non recurrence of the
events points towards cataclysmic astrophysical events.
The compact nature of GRB sources, the observed flux and the cosmological
distance taken together imply a large photon density and, therefore, a large op-
tical depth for pair production. This is, however, inconsistent with the optically
thin source indicated by the non–thermal gamma–ray spectrum, which extends
well beyond the pair production threshold at 0.5 MeV. This problem ( compact-
ness problem) can be resolved by assuming an ultra–relativistic expansion of the
emitting region. The bulk Lorentz factor required are Γ > 100 (see, e.g., [128])
Various catastrophic collapse events have been proposed in order to ex-
plain the energies released in a GRB. Among those proposals we find neutron–
star/neutron–star mergers [120,52], neutron–star/black–hole mergers [108], col-
lapsars [166,90] and hypernovae [121]. These models rely on the existence of
a stellar mass BH hole which accretes several solar masses of matter from
a disk (formed during a merger or by a non–spherical collapse) at a rate of
∼ 1 M⊙ s−1 [131]. A fraction of the gravitational binding energy released by
accretion is converted into neutrino and anti–neutrino pairs, which in turn an-
nihilate into electron–positron pairs. This creates a pair fireball, which will also
include baryons present in the environment surrounding the black hole. If the
baryon load (the ratio of the fireball mass to its energy) of the fireball is small
enough, the baryons are accelerated together with the e+ e− pairs to ultra–
relativistic speeds with Lorentz factors > 102 [23,129]. The existence of such
relativistic flows is supported by radio observations of GRB 980425 [82]. The
rapid temporal decay of several GRB afterglows is inconsistent with spherical
(isotropic) blast wave models propagating through the interstellar medium, and
3D Relativistic Hydrodynamics 25

instead is more consistent with the evolution of a relativistic jet after it slows
down and spreads laterally [137]. Finally, the bulk kinetic energy of the fireball is
thought to be converted into gamma–rays via cyclotron radiation and/or inverse
Compton processes (see, e.g., [105,128]).
One–dimensional numerical simulations of spherically symmetric relativis-
tic fireballs have been performed by several authors to model GRB sources
(e.g., [129,123,124,59]). Multi–dimensional modeling of ultra–relativistic jets in
the context of GRBs has for the first time been attempted by Aloy et al. [7]. Us-
ing a collapsar progenitor model (from [90]) they have simulated the propagation
of an axisymmetric jet through the mantle and envelope of a collapsing massive
star using a version of GENESIS [3] that includes a background Schwarzschild
metric. The jet forms as a consequence of an assumed energy deposition rate
of 1050 − 1051 erg/sec within a 30◦ cone around the rotation axis. When the
jet reaches the surface of the stellar progenitor, the maximum Lorentz factor
attained by the flow is about 20. The latter fact implies that Newtonian sim-
ulations of this phenomenon [90] are clearly inadequate. The simulations also
try to address the ulterior acceleration of the fireball when the jet propagates
through an atmosphere of declining density. At the end of the simulations (when
the jet has gone over ∼ 1011 cm) the maximum Lorentz factor is about 50 (for an
energy deposition rate of 1051 erg/sec). The baryonic contamination is very het-
erogenous having an average value of 1. However, there are regions (coincident
with the parts of the flow having the largest Lorentz factor) where this value is
as small as 10−5 , which is in agreement with the theoretical expectations (see
above). Although the final Lorentz factor is small compared with the predictions
of the standard model [23,129], the distance up to which the jet propagation has
been tracked (8 × 1010 cm) is much smaller than the one assumed to produce
efficient internal shocks (1012 −1014 cm) and the fireball to become optically thin
(∼ 1013 cm, [128]). Therefore, there is still room for further acceleration of the
jet until it becomes transparent. Finally, there are recent claims pointing to the
possibility of GRB generation without extremely high Lorentz factors and with
much smaller masses than in the standard model [153].

6 Summary
Hydrodynamic relativistic processes are on the basis of a variety of challenging
astrophysical phenomena. On the other hand, relativistic astrophysics has ben-
efited of the recent development of accurate numerical techniques which have
allowed, for the first time, the simulation of ultrarelativistic multidimensional
flows. Two main applications are currently addressed. One is in the field of
relativistic jets, where very important advances have been made in problems
like the jet formation mechanisms or the nature of superluminal sources. The
other main application is in the field of GRBs. In order to stress the importance
of relativistic hydrodynamic simulations in these fields, let us remind that the
generation of both relativistic jets and GRBs is hidden to present observations
making numerical simulations the only means to confront the theoretical models.
26 M.A. Aloy and J.M. Martı́

In the present review we have rewritten the equations governing the dynam-
ics of relativistic perfect fluids paying special attention to their conservative and
hyperbolic character (two properties which are extensively exploited by mod-
ern numerical techniques). We have also reviewed the evolution of numerical
relativistic hydrodynamics, that started as a branch of relativistic astrophysics
more than thirty years ago with the pioneering simulations of May and White
of stellar core collapse, and which has culminated in the last decade with the
introduction of high–resolution shock–capturing (HRSC) methods. The basics
of HRSC methods have been summarized, too.
Most of the problems that may be treated by means of numerical simula-
tions require a multidimensional modeling of the flows. Fully three–dimensional
simulations in relativistic hydrodynamics are particularly challenging because of
their intrinsic, technical difficulties. These difficulties have been addressed only
by a few scientific groups, so far. We have listed those codes that have been used
in 3D special or general RHD simulations. Two of them (Cactus and Shibata–
Nakamura codes) have implemented a full consistent evolution of the metric of
the space–time coupled to the GRHD equations. This fact has allowed them to
study, for the first time, problems as complex as the coalescence of simplified
models of NSs. The other two codes (GENESIS and Koide codes) allow for a
static background metric of the space–time and even for the inclusion of the
magnetic field in the equations (Koide’s code).
The last Section has been devoted to outline the main results in the simula-
tion of relativistic extragalactic jets and GRBs.
Acknowledgements
This work has been supported in part by the Spanish DGES (grant PB 97-
1432). M.A. Aloy thanks to MPA for financial support under its guest program
and to the Spanish MEC for the a grant (EX 00 22566499). The authors thank
J.M. Ibáñez and E. Müller for the critical reading of the manuscript.

References
1. C.E. Akujor: Astron. Astrophys. 259, L61 (1992)
2. M. Alcubierre, B. Bruegmann, et al. :Phis. Review D 62, 044034 (2000)
3. M.A. Aloy, J.Ma . Ibáñez, J.Ma . Martı́, E. Müller: Astrophys. J. Suppl. 122, 151
(1999)
4. M.A. Aloy, J.Ma . Ibáñez, J.Ma . Martı́, J.L. Gómez, E. Müller: Astrophys. J. Lett.
523, L125 (1999)
5. M.A. Aloy, J. Pons, J.Ma . Ibáñez, Comput. Phys. Commun. 120, 115 (1999)
6. M.A. Aloy, J.Ma . Ibáñez, J.Ma . Martı́, J.L. Gómez, E. Müller: Astrophys. J. Lett.
528, L85 (2000)
7. M.A. Aloy, E. Müller, J.Ma . Ibáñez, J.Ma . Martı́, A. MacFadyen: Astrophys. J.
Lett. 531, L119 (2000)
8. A.M. Anile. Relativistic Fluids and Magnetofluids, (Cambridge University Press,
Cambridge 1989)
9. R. Arnowitt, S. Deser, C.W. Misner. Gravitation: An Introduction to Current Re-
search, ed., Witten, L., (John Wiley, New York, 1962) p. 227
10. J.M. Attridge, D.H. Roberts, J.F.C. Wardle: Astrophys. J. Lett. 518, L87 (1999)
3D Relativistic Hydrodynamics 27

11. D.S. Balsara: J. Comp. Phys. 114, 284 (1994)


12. F. Banyuls, J.A. Font, J.Ma . Ibáñez, J.Ma . Martı́, J.A. Miralles: Astrophys. J.
476, 221 (1997)
13. M.C. Begelman, R.D. Blandford, M.J. Rees: Rev. Mod. Phys., 56, 255 (1984)
14. M.C. Begelman, M.J. Rees, M. Sikora: Astrophys. J. Lett. 429, L57 (1994)
15. M.J. Berger, P. Colella: J. Comp. Phys. 82, 64 (1989)
16. G.V. Bicknell. In Energy Transport in Radio Galaxies and Quasars, eds. P.E.
Hardee, A.H. Bridle, J.A. Zensus. (ASP Conference Series, 1996) 100, p. 253
17. R.D. Blandford. In Active Galactic Nuclei, eds. T.K.-L. Courvoisier and M. Mayor
(Springer, Berlin, 1990), p. 161
18. R.D. Blandford, D.G. Payne: Mon. Not. Roy. Astronom. Soc. 199, 883 (1982)
19. R.D. Blandford, M.J. Rees: Mon. Not. Roy. Astronom. Soc. 169, 395 (1974)
20. J.P. Boris, D.L. Book: J. Comp. Phys. 11, 38 (1973)
21. A.H. Bridle, D.H. Hough, et al. : Astronom. J. 108, 766 (1994)
22. More information can be found under the URL: http://www.cactuscode.org.
23. G. Cavallo, M.J. Rees: Mon. Not. Roy. Astronom. Soc. 183, 359 (1978)
24. J. Centrella, J.R. Wilson: Astrophys. J. 54, 229 (1984)
25. A.J. Chorin: J. Comp. Phys. 22, 517 (1976)
26. E. Chow, J.J. Monaghan: J. Comp. Phys. 134, 296 (1997)
27. P. Colella: SIAM J. Sci. Comput. 3, 76 (1982)
28. P. Colella, P.R. Woodward: J. Comp. Phys. 54, 174 (1984)
29. E. Costa, et al. : Nature 387, 783 (1997)
30. R. Courant, K.O. Friedrichs. Supersonic Flows and Shock Waves, (Springer, Berlin
1976)
31. W. Dai, P.R. Woodward: SIAM J. Sci. Comput. 18, 982 (1997)
32. S.F. Davis: ICASE Report No. 84–20 (1984)
33. G.C. Duncan, P.A. Hughes: Astrophys. J. Lett. 436, L119 (1994)
34. G.C. Duncan, P.A. Hughes, J. Opperman. In Energy Transport in Radio Galaxies
and Quasars, eds. P.E. Hardee, A.H. Bridle and J.A. Zensus, ASP Conference Series
(1996), 100, p. 143-148
35. R. Donat, J.M. Font, J.Ma . Ibáñez, A. Marquina: J. Comp. Phys. 146, 58 (1998)
36. R. Donat, A. Marquina: J. Comp. Phys. 125, 42 (1996)
37. B. Einfeldt: SIAM J. Numer. Anal. 25, 294 (1988)
38. F. Eulderink, G. Mellema: Astron. Astrophys. 284, 654 (1994)
39. F. Eulderink, G. Mellema: Astron. Astrophys. Suppl. 110, 587 (1995)
40. C.R. Evans. In Dynamical Space–Times and Numerical Relativity, ed. J. Centrella,
(Cambridge University Press, Cambridge, 1986) p. 3
41. S.A.E.G. Falle, S.S. Komissarov: Mon. Not. Roy. Astronom. Soc. 278, 586 (1996)
42. B.L. Faranoff, J.M. Riley: Mon. Not. Roy. Astronom. Soc. 167, 31 (1974)
43. J.A. Font, J.Ma . Ibáñez, J.Ma . Martı́, A. Marquina: Astron. Astrophys. 282, 304
(1994)
44. J.A. Font, M. Miller, W.-M. Suen, M. Tobias: Phys. Rev. D 61, 044011 (1999)
45. J.A. Font, Living Reviews in Relativity (2000), under the URL:
http://www.livingreviews.org/Articles/Volume3/2000-2font
46. T.J. Galama, P.M. Vreeswijk, E. Pian, F. Frontera, V. Doublier, J.-F. Gonzalez:
IAU Circ., 6985 (1998)
47. T.J. Galama, P.M. Vreeswijk, et al. : Nature 395, 670 (1998)
48. T.J. Gallama, (2000), this volume.
49. G. Ghisellini, P. Padovani, A. Celotti, L. Maraschi: Astrophys. J. 407, 65 (1993)
50. J. Glimm: Comm. Pure Appl. Math. 18, 697 (1965)
28 M.A. Aloy and J.M. Martı́

51. S.K. Godunov: Mat. Sb., 47, 271 (1959)


52. J. Goodman: Astrophys. J. Lett. 308, L47 (1986)
53. J.L. Gómez, A. Alberdi, J.M. Marcaide: Astron. Astrophys. 274, 55 (1993)
54. J.L. Gómez, A. Alberdi, J.M. Marcaide: Astron. Astrophys. 284, 51 (1994)
55. J.L. Gómez, J.Ma . Martı́, A.P. Marscher, J.Ma . Ibáñez, J.M. Marcaide: Astrophys.
J. Lett. 449, L19 (1995)
56. J.L. Gómez, J.Ma . Martı́, A.P. Marscher, J.Ma . Ibáñez, A. Alberdi: Astrophys. J.
Lett. 482, L33 (1997)
57. J.L. Gómez, A.P. Marscher, A. Alberdi, D.C. Gabuzda: Astrophys. J. 519, 642
(1999)
58. J.L. Gómez, (2000), this volume.
59. J. Granot, M. Miller, T. Piran, W.-M. Suen. In Gamma–Ray Bursts. Eds. R.M.
Kippen, R.S. Mallozzi and G.J. Fishman, (AIP conference proceedings 2000), 526,
540.
60. A. Harten: Comp. Pure Appl. Math. 48, 12 (1995)
61. A. Harten: SIAM J. Numer. Anal. 21, 1 (1984)
62. A. Harten, B. Engquist, S. Osher, S. Chakravarthy: J. Comp. Phys. 71, 231 (1987)
63. J.F. Hawley, L.L. Smarr, J.R. Wilson: Astrophys. J. Suppl. 55, 211 (1984)
64. J.R.A. Hooimeyer, P.D. Barthel, R.T. Schilizzi, G.K. Miley: Astron. Astrophys.
261, 25 (1992)
65. C.A. Hummel, et al. : Astron. Astrophys. 257, 489 (1992)
66. T.J. Iwamoto, P.A. Mazzali, et al. : Nature 395, 672 (1998)
67. T.W. Jones. In Supermassive black holes, (Cambridge University Press, Cambridge
1988), p. 59
68. T.W. Jones, S.L. O’Dell: Astrophys. J. 214, 522 (1977)
69. W. Junor, J.A. Bireta: Astronom. J. 109, 500 (1995)
70. A. Kheyfets, W.A. Miller, W.H. Zurek: Phys. Rev. D 41, 451 (1990)
71. R.W. Klebesadal, I.B. Strong, R.A. Olsen: Astrophys. J. Lett. 182, L85 (1973)
72. A. Königl: Astrophys. J. 243, 700 (1981)
73. S. Koide: Astrophys. J. 487, 66 (1997)
74. S. Koide, K.-I. Nishikawa, R.L. Muttel: Astrophys. J. Lett. 463, L71 (1996)
75. S. Koide, K. Shibata, T. Kudoh: Astrophys. J. Lett. 495, L63 (1998)
76. S. Koide, K. Shibata, T. Kudoh: Astrophys. J. 522, 727 (1999)
77. S. Koide, D.L. Meier, K. Shibata, T. Kudoh: Astrophys. J. 536, 668 (2000)
78. S.S. Komissarov: Sov. Astron. Lett. 16(4), 284 (1990)
79. S.S. Komissarov, S.A.E.G. Falle: Mon. Not. Roy. Astronom. Soc. 288, 833 (1997)
80. S.S. Komissarov, S.A.E.G. Falle: Mon. Not. Roy. Astronom. Soc. 297, 1087 (1998)
81. A.K. Kembhavi, J.V. Narlikar. Quasars and Active Galactic Nuclei. An introduc-
tion (Cambridge University Press, Cambridge 1999)
82. S.R. Kulkarni, D.A. Frail, et al. : Nature 395, 663 (1998)
83. P. Laguna, W.A. Miller, W.H. Zurek: Astrophys. J. 404, 678 (1993)
84. N.K. Lahy, N.K. A Particle Method for Relativistic Fluid Dynamics, M.Sc. Thesis,
Monash University (1989)
85. R.A. Laing. In Radio Galaxies and Quasars, eds. P.E. Hardee, A.H. Bridle, J.A.
Zensus. ASP Conference Series, Vol. 100 (1996) p. 241
86. W. Landry, S.A. Teukolsky: submitted (1999). See also: gr-qc/9912004
87. R.J. LeVeque. Numerical Methods for Conservation Laws, 2nd. edn. (Birkhäuser,
Basel 1992)
88. A.P. Lovanov, T.P. Krichbaum, et al. : Astron. Astrophys. 340, L60 (1998)
89. L.B. Lucy: Astronom. J. 82, 1013 (1977)
3D Relativistic Hydrodynamics 29

90. A. MacFadyen, S.E. Woosley: Astrophys. J. 524, 262 (1999)


91. P.J. Mann: Comput. Phys. Commun. 67, 245 (1991)
92. P.J. Mann: Comput. Phys. Commun. 107, 188 (1993)
93. A. Marquina: SIAM J. Sci. Comput. 15, 892 (1994)
94. A.P. Marscher: Astrophys. J. 235, 386 (1980)
95. A.P. Marscher, W.K. Gear: Astrophys. J. 298, 114 (1985)
96. A.P. Marscher, W.K. Gear, J.P. Travis. In Variability of Blazars, ed. by E. Vataoja
and M. Valtonen (Cambridge University Press, Cambridge 1992)
97. J.Ma . Martı́, J.Ma . Ibáñez, J.A. Miralles: Phis. Review D 43, 3794 (1991)
98. J.Ma . Martı́, E. Müller: J. Fluid Mech. 258, 317 (1994)
99. J.Ma . Martı́, E. Müller, J.Ma . Ibáñez: Astron. Astrophys. 281, L9 (1994)
100. Martı́, J.Ma , E. Müller, J.A. Font, J.Ma . Ibáñez: Astrophys. J. 448, L105 (1995)
101. Martı́, J.Ma , Müller, E., J.A. Font, J.Ma . Ibáñez, A. Marquina: Astrophys. J.
479, 151 (1997)
102. J.Ma , Martı́, E. Müller: Living reviews in Relativity (1999), under the URL:
http://www.livingreviews.org/Articles/Volume2/1999-3marti
103. M.M. May, R.H. White: Methods Comput. Phys. 7, 219 (1967)
104. C.A. Meegan, G.J. Fishman, et al. : Nature 355, 143 (1992)
105. P. Mèszáros. In Proc. of the 17th Texas Symp. on Relativistic Astrophysics and
Cosmology, eds. H. Böhringer, G.E. Morfill and J.E. Trümper, (N. Y. Acad. Sci.,
1995) p. 440
106. A.J. Mioduszewski, P.A. Hughes, G.C. Duncan: Astrophys. J. 476, 649 (1997)
107. I.F. Mirabel, L.F. Rodriguez: Nature 371, 46 (1994)
108. R. Mochkovitch, M. Hernanz, J. Isern, X. Martin: Nature 361, 236 (1993)
109. J.J. Monaghan: Comput. Phys. Rep. 3, 71 (1985)
110. J.J. Monaghan: J. Comp. Phys. 136, 298 (1997)
111. T. Nakamura: Prog. Theor. Phys. 65, 1876 (1981)
112. T. Nakamura, K. Oohara, Y. Kojima: Prog. Theor. Phys. Suppl. 90, 76 (1987)
113. T. Nakamura, K. Oohara. In Numerical Astrophysics. Ed. by S.M. Miyama, K.
Tomisaka and T. Hanawa (Kluwer Academic, Astrophysics and space science library
1999) 240, p.247
114. T. Nakamura, K. Maeda, S. Miyama, M. Sasaki: Prog. Theor. Phys. 63, 1229
(1980)
115. K.-I. Nishikawa, S. Koide, et al. : Astrophys. J. 483, L45 (1997)
116. K.-I. Nishikawa, S. Koide, et al. : Astrophys. J. 498, 166 (1998)
117. M.L. Norman, L. Smarr, K.-H.A. Winkler, M.S. Smith: Astron. Astrophys. 113,
285 (1982)
118. M.L. Norman, K.-H.A. Winkler. In: Astrophysical Radiation Hydrodynamics, ed.
by M.L. Norman, K.-H.A. Winkler (Reidel, Dordrecht, 1986) p. 227
119. S. Osher, S. Chakravarthy: SIAM J. Numer. Anal. 21, 995 (1984)
120. B. Pacyński: Astrophys. J. 308, L43 (1986)
121. B. Pacyński: Astrophys. J. 494, L45 (1998)
122. A. Panaitescu, L. Wen, P. Laguna, P. Mészáros: Astrophys. J. 482, 942 (1999)
123. A. Panaitescu, P. Mészáros: Astrophys. J. 492, 683 (1998)
124. A. Panaitescu, P. Mészáros: Astrophys. J. 526, 707 (1999)
125. L.I. Petrich, S.L. Shapiro, R.F. Stark, S.A. Teukolsky: Astrophys. J. 336, 313
(1989)
126. E. Pian, L. Amati, et al. : Astrophys. J. 538, 638 (2000)
127. T. Piran: J. Comp. Phys. 35, 254 (1980)
128. T. Piran: Physics Reports 314, 575 (1999)
30 M.A. Aloy and J.M. Martı́

129. T. Piran, A. Shemi, R. Narayan: Mon. Not. Roy. Astronom. Soc. 263, 861 (1993)
130. J.A. Pons, J.Ma Martı́, E. Müller: J. Comp. Phys. 422, 125 (2000)
131. R. Popham, S.E. Woosley, C. Fryer: Astrophys. J. 518, 356 (1999)
132. P.L. Roe: J. Comp. Phys. 43, 357 (1981)
133. R.D. Richtmyer, K.W. Morton: Difference Methods for Initial–value Problems,
(Wiley–Interscience, New York 1967)
134. Rosen, A., Hughes, et al. : Astrophys. J. 516, 729 (1999)
135. C. Sauty (2000), this volume.
136. R.H. Sanders, K.H. Prendergast: Astrophys. J. 188, 489 (1974)
137. R. Sari, T. Piran, J.P. Halpern: Astrophys. J. Lett. 519, L17 (1999)
138. P.A.G. Scheuer: Mon. Not. Roy. Astronom. Soc. 166, 513 (1974)
139. V. Schneider, U. Katscher, et al. : J. Comp. Phys. 105, 92 (1993)
140. M. Shibata, T. Nakamura: Phis. Review D 52, 5428 (1995)
141. M. Shibata: Prog. Theor. Phys. 101, 251 (1999)
142. M. Shibata: Prog. Theor. Phys. 101, 1199 (1999)
143. M. Shibata: Phis. Review D 60, 104052 (1999)
144. M. Shibata, T.W. Baumgarte, S.L. Shapiro: Phis. Review D 61, 044012 (2000)
145. S. Siegler, H. Riffert: Astrophys. J. Suppl. 531, 1053 (2000)
146. C.W. Shu: Math. Comp. 49, 105 (1987)
147. C.W. Shu, S.J. Osher: J. Comp. Phys. 83, 32 (1989)
148. L. Smarr, J.R. Wilson, R.T. Barton, R.L. Bowers: Astrophys. J. 246, 515 (1981)
149. P. Soffitta, M. Feroci, et al. : IAU Circ. 6884 (1998)
150. J. Sollfrank, P. Huovinen, M. Kataja, P.V. Ruuskanen, M. Prakash, R. Venu-
gopalan: Phys. Rev. C 55, 392 (1997)
151. R.F. Stark, T. Piran: Comput. Phys. Rep. 5, 221 (1987)
152. M.R. Swain, A.H. Bridle, S.A. Baum: Astrophys. J. 507, L29 (1998)
153. J.C. Tan, C.D. Matzner, C.F. McKee: Astrophys. J. , submitted, (2000)
154. M. Taub: Phys. Rev. 74, 328 (1948)
155. K.W. Thomson: J. Fluid Mech. 171, 365 (1986)
156. S.J. Tingay, D.L. Jauncey, R.A. Preston, J.E. Reynolds, D.L. Meier et al. : Nature
374, 141 (1995)
157. E. Toro: Riemann solvers and numerical methods for fluid dynamics: a practical
introduction, 1st edn. (Springer, Berlin 1997)
158. B. van Leer: J. Comp. Phys. 32, 101 (1979)
159. M.H.P.M. van Putten: J. Comp. Phys. 105, 339 (1993a)
160. M.H.P.M. van Putten: Astrophys. J. Lett. 408, L21 (1993b)
161. M.H.P.M. van Putten: Astrophys. J. Lett. 467, L57 (1996)
162. J. von Neumann, R.D. Richtmyer: J. Appl. Phys. 21, 232 (1950)
163. L. Wen, A. Panaitescu, P. Laguna: Astrophys. J. 486, 919 (1997)
164. J.R. Wilson: Astrophys. J. 173, 431 (1972)
165. J.R. Wilson. In: Sources of Gravitational Radiation, ed. by L.L. Smarr, (Cam-
bridge University Press, Cambridge 1979) p. 423
166. S.E. Woosley: Astrophys. J. 405, 273 (1993)
167. Yang, J.Y., Chen, M.H., Tsai, I-N. and J.W. Chang: J. Comp. Phys. 136, 19
(1997)

You might also like