See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/1809170

Numerical Magnetohydrodynamics in Astrophysics: Algorithm and Tests for

One-Dimensional Flow

Article  in  The Astrophysical Journal · April 1994

DOI: 10.1086/175437 · Source: arXiv

CITATIONS READS
290 269

2 authors:

Dongsu Ryu Tom Jones

Ulsan National Institute of Science and Technology University of Minnesota Twin Cities
384 PUBLICATIONS   11,870 CITATIONS    252 PUBLICATIONS   9,569 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Shocked Radio Galaxies View project

All content following this page was uploaded by Dongsu Ryu on 24 January 2015.

The user has requested enhancement of the downloaded file.

 Numerical Magnetohydrodynamics in Astrophysics:
arXiv:astro-ph/9404074v1 29 Apr 1994

Algorithm and Tests for One-Dimensional Flow

Dongsu Ryu
Princeton University Observatory, Peyton Hall, Princeton, NJ 08544;
Department of Astronomy and Space Science, Chungnam National University,
Daejeon 305-764, (South) Korea
and
T. W. Jones
School of Physics and Astronomy, University of Minnesota, Minneapolis, MN 55455

February 1, 2008

Submitted to the Astrophysical Journal

 2
ABSTRACT

We describe a numerical code to solve the equations for ideal magnetohydrodynamics

(MHD). It is based on an explicit finite difference scheme on an Eulerian grid, called the
Total Variation Diminishing (TVD) scheme, which is a second-order-accurate extension
of the Roe-type upwind scheme. We also describe a nonlinear Riemann solver for ideal
MHD, which includes rarefactions as well as shocks and produces exact solutions for
two-dimensional magnetic field structures as well as for the three-dimensional ones. The
numerical code and the Riemann solver have been used to test each other.

Extensive tests encompassing all the possible ideal MHD structures with planar sym-
metries (i.e. one-dimensional flows) are presented. These include those for which the
field structure is two-dimensional (i.e., those flows often called “1 + 1/2 dimensional”)
as well as those for which the magnetic field plane rotates (i.e.,, those flows often called
“1 + 1/2 + 1/2 dimensional”). Results indicate that the code can resolve strong fast, slow,
and magnetosonic shocks within 2-4 cells while more cells are required if shocks become
weak. With proper stiffening, rotational discontinuities are resolved within 3-5 cells. Con-
tact discontinuities are also resolved within 3-5 cells with stiffening and 6-8 cells without
stiffening, while the stiffening on contact discontinuities in some cases generates numerical
oscillations. Tangential discontinuities spread over more than 10 cells.

Our tests confirm that slow compound structures with two-dimensional magnetic field
are composed of intermediate shocks (so called “2-4” intermediate shocks) followed by slow
rarefaction waves. Finally, tests demonstrate that in two-dimensional magnetohydrody-
namics fast compound structures, which are composed of intermediate shocks (so called
“1-3” intermediate shocks) preceded by fast rarefaction waves, are also possible.

subject headings: hydromagnetics - magnetohydrodynamics:MHD - methods:numerical -

shock waves
 3
1. INTRODUCTION

Many astronomical objects as diverse as planets, stars, and galaxies all possess mag-
netic fields which have important implications on their dynamics and evolution. However,
in many objects including the Earth the ohmic dissipation time tD = L/η 2 (η is the electric
resistivity) is smaller than their ages. So their magnetic fields should be continuously gen-
erated by some dynamo activity, otherwise the observed strength cannot be maintained.
The dynamo activity involves patterns of magnetohydrodynamic (MHD) flows (for exam-
ple, the interaction of differential rotation and convection as in some stars or accretion
disks) needed to produce spatially coherent magnetic fields of large scale. However, such
flows are usually highly nonlinear, so with several exceptions analytic approaches to un-
derstand dynamo theory and the origin of the magnetic field have failed (for review and
further references, see Parker 1979).

With the appearance of fast supercomputers it has become possible to study the MHD
flows numerically, but the development of numerical techniques to solve MHD equations has
been slower due to the intrinsic complexity of the MHD flows. For instance, until now most
numerical schemes to solve compressible MHD equations have been based on the methods
using artificial viscosity (e.g.DeVore 1991; Lind, Payne & Meier 1991; Stone et al. 1992;
Stone & Norman 1992) while schemes to solve compressible hydrodynamic equations have
been based on methods using more sophisticated linear or nonlinear Riemann solvers
(e.g. Roe 1981; Colella & Woodward 1984). It is rather recent that Brio & Wu (1988;
henceforth BW) and Zachary & Colella (1992) developed schemes for MHD equations
based on linear Riemann solvers and Dai & Woodward (1994; henceforth DW) one based
on a nonlinear Riemann solver. These new schemes have been proved to handle MHD
flows better, even though they are more expensive in CPU time and more difficult to code.

In this paper, we describe a one-dimensional code to solve numerically the ideal MHD
equations. It is based on an explicit second-order-accurate finite difference scheme on an
Eulerian grid, called the Total Variation Diminishing (TVD) scheme, which was originally
developed for numerical hydrodynamics by Harten (1983). It utilizes a Roe-type linear
Riemann solver described in BW. Here, we do not attempt to describe the theoretical
philosophy of the scheme which can be found in Harten (1983) and BW. Instead, we focus
on the description of the procedure necessary to write the code and the test results for its
performance.

We also describe a nonlinear Riemann solver for ideal magnetohydrodynamics (MHDs)

which has been developed to provide analytic solutions for shock tube tests of the code.
It has been built by following a procedure similar to that in DW, but using initial guesses
with the linear eigenvectors from BW and Zachary & Colella (1992).

The paper is organized in the following way. In §2 we describe the step by step
procedure of the code. In §3 we describe the nonlinear Riemann solver in details. We
intend to make the detailed descriptions of the code and the nonlinear Riemann solver so
others can recover them by following the descriptions. The tests using shock tubes are
presented in §4 to demonstrate the ability of the code to capture discontinuities as well
 4
as to follow smooth flows. The tests also serve to show the performance of the nonlinear
Riemann solver. Finally, discussion is followed in §5 including the possible existence of fast
compound structures as well as slow compound structures for two-dimensional magnetic
field structures.

2. TVD CODE FOR IDEAL MHDS

2.1. Ideal MHD Equations

The subject of MHDs describes the dynamics of electrically conducting fluids in the
presence of magnetic fields. The MHD equations represent coupling of the equations of fluid
dynamics with the Maxwell’s equations of electrodynamics. By neglecting the displacement
current, the separation between ions and electrons, and the effects of electrical resistivity,
viscosity, and thermal conduction, we get the following ideal MHD equations:
∂ρ
+ ∇ · (ρv) = 0, (2.1)
∂t
∂v 1 1
+ v · ∇v + ∇p − (∇ × B) × B = 0, (2.2)
∂t ρ ρ
∂p
+ v · ∇p + γp∇ · v = 0, (2.3)
∂t
∂B
− ∇ × (v × B) = 0, (2.4)
∂t
with an additional constraint ∇ · B = 0 for the absence of magnetic monopole (for details,
see Shu 1992). Here, we have chosen units so that factor of 4π does not appear in the
equations.

For plane-symmetric, or one-dimensional flows exhibiting variation along the x direc-

tion. The equations in (2.1) to (2.4) can be written in conservative form as

∂q ∂F
+ = 0, (2.5)
∂t ∂x

ρ
 
 ρvx 
 ρvy 
 

 ρvz 
 
q=  (2.6)
 By 
 
B 
z
E ,
 5

ρvx
 

 ρvx + p∗ − Bx
2 2 


 ρvx vy − Bx By 

F = ρvx vz − Bx Bz (2.7)
 


 By vx − Bx vy 

Bz vx − Bx vz
 
 
(E + p∗ )vx − Bx (Bx vx + By vy + Bz vz ) ,

where the total pressure and the total energy are given by
1 2
p∗ = p + Bx + By2 + Bz2 ,

(2.8)
2
1  p 1 2
E = ρ vx2 + vy2 + vz2 + Bx + By2 + Bz2 .
 
+ (2.9)
2 γ−1 2
With the state vector, q, and the flux function, F (q), the Jacobian matrix, A(q) = ∂F /∂q,
is formed. The above system of equations is called hyperbolic, since all the eigenvalues of
the Jacobian matrix are real and the corresponding set of the right eigenvectors is complete
(Jeffrey & Taniuti 1964). However, the eigenvalues may coincide in some limiting cases
(BW).

2.2. Eigenvalues and Eigenvectors for Plane-Symmetric MHD Equations

The first step to build a code based on the Harten’s TVD scheme (Harten 1983) for
the hyperbolic system of equations in (2.5) is to get the eigenvalues and the right and left
eigenvectors of the Jacobian A(q). The seven eigenvalues in nondecreasing order are

a1 = vx − cf , (2.10)

a2 = vx − ca , (2.11)
a3 = vx − cs , (2.12)
a4 = vx , (2.13)
a5 = vx + cs , (2.14)
a6 = vx + ca , (2.15)
a7 = vx + cf , (2.16)
where cf , ca , cs are the fast, Alfvén, and slow characteristic speeds. The above represent
the seven speeds with which information is propagated locally by three MHD wave families
and an entropy mode. The three characteristic speeds are expressed as
v
u B2
u
ca = t x (2.17)
ρ ,
 6
  v  1
!2  2
Bx2 + By2 + Bz2 Bx2 + By2 + Bz2 2
u
1 B x 

a2 +
 u
cf =  + t
a2 + − 4a 2 (2.18)
2 ρ ρ ρ 

 
,
  v  1
!2 2
Bx2 + By2 + Bz2 u Bx2 + By2 + Bz2 2
u
1 B

x
 
2
cs =  a + −
t 2
a + − 4a2

(2.19)
2 ρ ρ ρ 


,
where a is the sound speed given by
s
p
a= γ (2.20)
ρ.

The corresponding right eigenvectors are (see, e.g., Jeffrey & Taniuti 1964)
1
 

 vx ± cf 


 Bx By cf  
 vy ∓   

 ρ cf −ca 
2 2

 
 Bx Bz cf 
 vz ∓ 
 
 

 ρ cf −ca 
2 2

Rvx ±cf = 
By cf2

 (2.21)
  
 

 ρ c2f −c2a 

Bz cf
 2

 
   
 

 ρ cf −ca
2 2 

 vx +vy2 +vz2
 2 
±
2 + hf
,
0
 
 0 
∓B sign(B )
 
 z x 
±B sign(B )
 

 y x 

Rvx ±ca = 
 B
√z

(2.22)
ρ

 
B
 

√ y 

 − ρ 

   
∓ Bz vy − By vz sign(Bx )
,
1
 
 v x ± cs 
 vy ∓ Bx2By c2s 
 
 
 ρ(cs −ca ) 
 vz ∓ Bx2Bz c2s 
 
 
Rvx ±cs = 
 ρ (cs −ca ) 
(2.23)
By c2s

 
ρ (c2s −c2a )
 
 
B c 2
 
 z s 

 ρ(cs −ca )
2 2 

vx +vy +vz
 2 2 2
±

2 + h s ,
 7

1
 
 vx 
vy
 
 
vz
 
Rvx =  (2.24)
 
0

 
 2 02
 
 
vx +vy +vz2 
2 ,
where  
c2f Bx cf By vy + Bz vz γ−2 2
h± cf − a2 ,

f = ± cf vx ∓ + (2.25)
 
γ −1 ρ c2f − c2a γ−1
 
c2s Bx cs By vy + Bz vz γ−2 2
h± cs − a2 .

s = ± cs vx ∓   + (2.26)
γ−1 ρ c2s − c2a γ−1

Near the point where either Bx = 0 or By = Bz = 0, the above set of the right
eigenvectors is not well defined with columns becoming singular. By renormalizing the
eigenvectors, the singularities can be removed (BW). The renormalized eigenvectors are

αf
 
   

 αf vx ± cf 

 α vy ∓ αs βy ca sign(Bx ) 
 
 f 
 αf vz∓ αs βz ca sign(Bx ) 
 
 
Rvx ±cf =  αs βy cf  (2.27)
√
 

 ρ 

αs βz cf
 
 
√
ρ
 
 
vx +vy +vz
 2 2 2

±
 
αf 2 + g f ,

0
 
 0 
∓βz sign(Bx )
 
 
±βy sign(Bx )
 
 
 
Rvx ±ca = 
 βz
√

(2.28)
ρ

 
βy
 
 
− √
ρ
 
 
  
∓ βz vy − βy vz sign(Bx )
,
 8

αs
 
 α s x ± cs )
(v 
 αs vy ± α βy a sign(Bx ) 
 
 f 
 αs vz ± αf βz a sign(Bx ) 
 
 

Rvx ±cs =  αf βy a2 
(2.29)
− √ 

 cf ρ 

α β a
 2

f √z
 

 − c ρ 

 f 
vx +vy +vz
2 2 2
 
±
αs 2 + g s ,
1
 
 vx 
v
 
 y 
v
 
Rvx = 
 z 
(2.30)
0

 
 2 02
 
 
vx +vy +vz2 
2 ,
where
αf c2f γ−2  2
gf± αf cf − a2 ,
  
= ± αf cf vx ∓ αs ca sign(Bx ) βy vy + βz vz + (2.31)
γ−1 γ−1

αs c2s γ−2  2
gs± = αs cs − a2 .
  
± αs cs vx ± αf a sign(Bx ) βy vy + βz vz + (2.32)
γ−1 γ−1
Here α’s and β’s are given by
c2f − c2a
q

αf = q (2.33)
c2f − c2s
,
q
c2f − a2
αs = q (2.34)
c2f − c2s
,
By
βy = q (2.35)
By2 + Bz2
,
Bz
βz = q (2.36)
By + Bz2
2
.
At the points where By = Bz = 0, β’s are defined as the limiting values, i.e.,

1
βy = βz = √ if By = Bz = 0. (2.37)
2
 9

Similarly, at the point where By = Bz = 0 and Bx2 /ρ = a2 , α’s are defined as

Bx2
αf = αs = 1 if By = Bz = 0 and = a2 . (2.38)
ρ

Then the left eigenvectors, which are orthonormal to the right eigenvectors, Ll · Rm =
δlm , are  
(1) (2) (3) (4) (5) (6) (7)
Lvx ±cf = lvx ±cf , lvx ±cf , lvx ±cf , lvx ±cf , lvx ±cf , lvx ±cf , lvx ±cf (2.39)
,
1 αf 2 2 1 αf αs
 
(1) 
lvx ±cf = a v ∓ avx sign(Bx ) − cs βy vy + βz vz (2.40)
θ1 4 θ2 2 2 ,
(2) 1 αf 2 1 αf
lvx ±cf = − a vx ± a sign(Bx ), (2.41)
θ1 2 θ2 2
(3) 1 αf 2 1 αs
lvx ±cf = − a vy ∓ βy cs , (2.42)
θ1 2 θ2 2
(4) 1 αf 2 1 αs
lvx ±cf = − a vz ∓ βz cs , (2.43)
θ1 2 θ2 2
!
(5) 1 αs 2−γ 2 √
lvx ±cf = βy cf c2s + a ρ, (2.44)
θ1 2 γ−1
!
(6) 1 αs 2−γ 2 √
lvx ±cf = βz cf c2s + a ρ, (2.45)
θ1 2 γ−1
(7) 1 αf 2
lvx ±cf = a , (2.46)
θ1 2
 
(1) (2) (3) (4) (5) (6) (7)
Lvx ±ca = lvx ±ca , lvx ±ca , lvx ±ca , lvx ±ca , lvx ±ca , lvx ±ca , lvx ±ca (2.47)
,
(1) βz vy − βy vz
lvx ±ca = ± sign(Bx ), (2.48)
2
(2)
lvx ±ca = 0, (2.49)
(3) βz
lvx ±ca = ∓ sign(Bx ), (2.50)
2
(4) βy
lvx ±ca = ± sign(Bx ), (2.51)
2
√
(5) βz ρ
lvx ±ca = , (2.52)
2
√
(6) βy ρ
lvx ±ca = − , (2.53)
2
(7)
lvx ±ca = 0, (2.54)
 10
 
(1) (2) (3) (4) (5) (6) (7)
Lvx ±cs = lvx ±cs , lvx ±cs , lvx ±cs , lvx ±cs , lvx ±cs , lvx ±cs , lvx ±cs (2.55)
,
(1) 1 αs 2 2 1 αs

αf 
lvx ±cs = c v ∓ ca vx sign(Bx) + c (βy vy + βz vz ) (2.56)
θ1 4 f θ2 2 2 f ,
(2) 1 αs 2 1 αs
lvx ±cs = − cf vx ± ca sign(Bx), (2.57)
θ1 2 θ2 2
(3) 1 αs 2 1 αf
lvx ±cs = − cf vy ± βy cf , (2.58)
θ1 2 θ2 2
(4) 1 αs 2 1 αf
lvx ±cs = − cf vz ± βz cf , (2.59)
θ1 2 θ2 2
1 αf
!
(5) 2−γ 2 √
lvx ±cs =− βy cf c2f + a ρ, (2.60)
θ1 2 γ−1
1 αf
!
(6) 2−γ 2 √
lvx ±cs =− βz cf c2f + a ρ, (2.61)
θ1 2 γ −1
(7) 1 αs 2
lvx ±cs = c , (2.62)
θ1 2 f
 
(1) (2) (3) (4) (5) (6) (7)
Lvx = lvx , lvx , lvx , lvx , lvx , lvx , lvx (2.63)
,
2 2 2 2
(1) 1 αf a + αs cf 2
lvx = 1 − v , (2.64)
θ1 2
(2) 1  2 2
αf a + α2s c2f vx ,

lvx = (2.65)
θ1
(3) 1  2 2
αf a + α2s c2f vy ,

lvx = (2.66)
θ1
(4) 1  2 2
αf a + α2s c2f vz ,

lvx = (2.67)
θ1
(5) 1 √
lvx = αf αs βy cf c2f − c2s

ρ, (2.68)
θ1
(6) 1 √
αf αs βz cf c2f − c2s

lvx = ρ, (2.69)
θ1
(7) 1  2 2
αf a + α2s c2f ,

lvx = − (2.70)
θ1
where ! !
2−γ 2 2−γ 2
θ1 = α2f a2 c2f + a + α2s c2f c2s + a (2.71)
γ−1 γ−1 ,

θ2 = α2f cf a sign(Bx ) + α2s cs ca sign(Bx ), (2.72)

v 2 = vx2 + vy2 + vz2 . (2.73)

 11
Note that, with the above normalization, some columns are not continuous. In order
to force them to be continuous, the following term
1, if By > 0 or By = 0 and Bz > 0

sign(BT ) = −1, if By < 0 or By = 0 and Bz < 0 , (2.74)

is multiplied Rvx ±cs and Lvx ±cs if a2 > c2a and to Rvx ±cf and Lvx ±cf if a2 < c2a .

It is interesting to see how the eigenvectors for the MHD equations reduce into those for
the hydrodynamic equations in the limit By → 0 and Bz → 0. If a2 > c2a , Rvx ±cf → RvHD x ±a
and Rvx → RvHD x
, that is, the characteristics associated with MHD fast waves become
those associated with hydrodynamic sound waves. The other two eigenvectors for the
three-dimensional hydrodynamic equations (Roe 1981) are obtained from the following
combinations of those for slow and Alfvén waves,
0
 
 0 
√  1 
 
2 sign(Bx ) Rvx +cs − Rvx −cs
 
 0 
 
− (Rvx +ca − Rvx −ca ) =   (2.75)
4 a  0 
 0 
 
vy ,
 
0
 0 
√ 
 0 

2 sign(Bx ) Rvx +cs − Rvx −cs
 
 
+ (Rvx +ca − Rvx −ca ) = 
 1 
 (2.76)
4 a  0 
 
 0 
vz .
Similarly, if a2 < c2a , Rvx ±cs → RvHD
x ±a
and Rvx → RvHD x
, that is, the characteristics
associated with MHD slow waves become those associated with hydrodynamic sound waves.
The other two eigenvectors for the three-dimensional hydrodynamic equations are obtained
from the following combinations of those for fast and Alfvén waves,
0
 
 0 
√  
 1 
 
2 sign(Bx )  Rvx +cf − Rvx −cf
 0 
 
− − (Rvx +ca − Rvx −ca ) =   (2.77)
4 ca  0 
 0 
 
vy ,
 
0
 0 
√   
 0 

2 sign(Bx )  Rvx +cf − Rvx −cf  
− + (Rvx +ca − Rvx −ca ) = 
 1 
 (2.78)
4 ca  0 
 
 0 
vz .
 12

In the cases with purely two-dimensional magnetic fields and motions (i.e., Bz = vz =
0), the eigenvectors associated with Alfvén waves becomes

0
 
 0 
0
 
 
±sign(B )
x 
 
Rvx ±ca = 
 (2.79)
0

 

 1
− √ρ


0 .

By combining them properly, they become (0, 0, 0, 1, 0, 0, 0)T and (0, 0, 0, 0, 0, 1, 0)T which
are trivial. This indicates that the flows with two-dimensional magnetic fields do not
produce structures associated with Alfvén modes like rotational discontinuities (see §4 and
5 for more discussions)

2.3. A Second-Order Explicit TVD Scheme

Here, we describe briefly the procedure to build the MHD-TVD code with the eigenval-
ues and eigenvectors in the previous subsection. The purpose of this section is to provide
a short but complete description of steps needed to build a code by the TVD scheme.
For the details, e.g., why and how each step works, the choices for the values of internal
parameters, etc., refer to the original work (Harten 1983).

The convention for indices used in this subsection is the following. The superscript n
represents the time step. The subscript i indicates the quantities defined in the cell centers
while i + 1/2 identifies those defined on the cell boundaries. The subscript k represents the
characteristic fields with the order that k = 1 is for the field associated with the eigenvalue
vx − cf , k = 2 for the field with vx − ca , k = 3 for the field with vx − cs , k = 4 for the field
with vx , k = 5 for the field with vx + cs , k = 6 for the field with vx + ca , and k = 7 for the
field with vx + cf .

In a code based on the TVD scheme, the physical quantities are defined in the cell
centers while the fluxes are computed on the cell boundaries. Implementation of Roe’s
linearization technique would result in a particular form of the averaged physical quantities
in the cell boundaries (Roe 1981). However, as pointed by BW, it is not possible to derive
the analytic form of the averaged quantities in MHDs for general cases with the adiabatic
index γ 6= 2. Instead, we should modify the Roe’s scheme by using the following simple
averaging scheme,
ρ + ρi+1
ρi+ 1 = i , (2.80)
2 2
vx,i + vx,i+1
vx,i+ 1 = , (2.81)
2 2
 13
vy,i + vy,i+1
vy,i+ 1 = , (2.82)
2 2
vz,i + vz,i+1
vz,i+ 1 = , (2.83)
2 2
By,i + By,i+1
By,i+ 1 = , (2.84)
2 2
Bz,i + Bz,i+1
Bz,i+ 1 = , (2.85)
2 2
p∗ + p∗i+1
p∗i+ 1 = i . (2.86)
2 2
Then, other quantities like momentum, gas pressure, total energy, etc are calculated by
combining the above quantities. Our tests for the cases with γ = 2 indicated that the
above simple averaging would do just as well when compared to the full implementation
of Roe’s linearization technique.

The state vector q at the cell center is updated by calculating the modified flux f̄ at
the cell boundaries as follows:
∆tn
qin+1 = qin − (f̄ 1 − f̄i− 1 ), (2.87)
∆x i+ 2 2

1h 7
n ) − ∆x
F (qin ) + F (qi+1 n
i X
f̄i+ 1 = β 1R 1, (2.88)
2 2 2∆tn k=1 k,i+ 2 k,i+ 2

∆tn n
!
βk,i+ 1 = Qk a 1 + γk,i+ 1 αk,i+ 1 − (gk,i + gk,i+1), (2.89)
2 ∆x k,i+ 2 2 2

αk,i+ 1 = Lnk,i+ 1 · (qi+1

n − q n ),
i (2.90)
2 2

g −g
 k,i+1 k,i for αk,i+ 1 6= 0
αk,i+ 1

γk,i+ 1 = 2 , 2 (2.91)
2  0, for αk,i+ 1 = 0

2 ,
  
gk,i = sign(g̃k,i+ 1 ) max 0, min |g̃k,i+ 1 |, g̃k,i− 1 sign(g̃k,i+ 1 ) , (2.92)
2 2 2 2

∆tn n ∆tn n
" #
1 2
g̃k,i+ 1 = Qk ( ak,i+ 1 ) − ( a 1) αk,i+ 1 , (2.93)
2 2 ∆x 2 ∆x k,i+ 2 2

χ2 + ε,
(
Qk (χ) = 4ε for |χ| < 2ε (2.94)
|χ|, for |χ| ≥ 2ε ,

 0.1,
for k = 1 and 7


ε= 0.2, for k = 2 and 6 (2.95)
 0.1, for k = 3 and 5
0.0 for k = 4 .
 14
Here, the time step ∆tn is restricted by the usual Courant condition for the stability,
∆tn = Ccour ∆x/Max(|vx,i+
n
1| + c
n
f,i+ 1
) with Ccour < 1. Typically we use Ccour = 0.8.
2 2

The rotational discontinuities represented by the k = 2 and 6 fields are steepened by

replacing gk,i with gk,i + θk,i ḡk,i :

|αk,i+ 1 −αk,i− 1 |

2 2
for (|αk,i+ 1 | + |αk,i− 1 |) 6= 0


|αk,i+ 1 |+|αk,i− 1 |

θk,i = 2 2 ,
2 2 (2.96)

 0,

for (|αk,i+ 1 | + |αk,i− 1 |) = 0
2 2 ,
  
ḡk,i = sign(αk,i+ 1 ) max 0, min sign(αk,i+ 1 )σk,i− 1 αk,i− 1 , σk,i+ 1 |αk,i+ 1 | , (2.97)
2 2 2 2 2 2

 ∆tn
"  #
1 
σk,i+ 1 = 1− ak,i+ 1  (2.98)
 
2 2  ∆x 2
.
With the above steepening the rotational discontinuities are resolved within 3-5 cells,
otherwise within 6-8 cells. The contact discontinuities represented by the k = 4 field
could be also steepened by a similar scheme. However, our tests showed that the price for
steepening the contact discontinuities is additional numerical oscillations.

3. A NONLINEAR MHD RIEMANN SOLVER

As a means to test quantitatively the MHD-TVD code and to understand more fully
the properties of ideal MHD flows along one-dimension, we developed an accurate nonlinear
MHD Riemann solver. It is similar in many respects to that described in DW. The
most important improvement in our MHD Riemann solver is that it treats fast and slow
rarefactions properly, instead of approximating them as “rarefaction shocks” as did the
solver presented in DW. We present here only essential details of the “solver”, referring
readers to DW for other relevant background.

As with hydrodynamic Riemann solvers, the construction of an MHD Riemann solver is

based on the idea that two adjacent arbitrary states will evolve into a set of uniform states
separated by left and right facing shocks and rarefactions. For the MHD problems, there
are a total of eight states including the original pair. They are separated by six structures
representing left and right propagating shocks or rarefactions of the three wave families
and a structure representing the entropy mode (or the characteristic field associated with
the eigenvalues vx in the discussion of the previous section). The initial boundary moves
with the structure of the entropy mode which becomes a contact discontinuity or, in
degenerate cases, a tangential discontinuity (see §4 for more discussion). In fact the
eigenstates developed in §2 provide approximate solutions to this problem. The particular
difficulty in the MHD Riemann problem is that the equations are not strictly hyperbolic nor
strictly convex (see the discussion in BW). In practice this means that the wave speeds
of two families may sometimes coincide, and that compound wave structures involving
both shocks and rarefactions may sometimes develop. For the most part we can ignore
 15
this aspect, and consider those situations as special cases. Otherwise we can approach the
Riemann problem in pretty much the same manner as for the hydrodynamical problem,
except for the larger number of waves to consider.

The solution of the problem is obtained by finding the required set of fast and slow
shock jumps and rarefactions together with rotational discontinuities that self-consistently
lead to a proper contact discontinuity or tangential discontinuity at the “center” of the
structure. Our procedure consists of taking an initial guess for the resolved states of the
(six interior) zones, and then iterating towards the proper jump conditions of the contact
discontinuity or the tangential discontinuity. For this discussion, the left initial state is
identified with zone 1 and the right initial state is identified with zone 8. Fast structures
separate zones 1 and 2 as well as zones 7 and 8. Rotational discontinuities separate zones
2 and 3 as well as zones 6 and 7. Zones 3 and 4 and zones 5 and 6 are each separated
by slow structures. The contact discontinuity or the tangential discontinuity demarcates
zones 4 and 5.

Discontinuities are easier to handle than rarefactions, so we outline the methods to

compute discontinuities first. The problem is conceptually simpler in Lagrangian mass
coordinates, where the jump conditions across a discontinuity are (e.g., equation (2.7),
DW):
W [V ] = −[vx ], (3.1)
W [vx ] = [p∗ − Bx2 ], (3.2)
W [vy ] = −Bx [By ], (3.3)
W [vz ] = −Bx [Bz ], (3.4)
W [V By ] = −Bx [vy ], (3.5)
W [V Bz ] = −Bx [vz ], (3.6)
W [V E] = [vx p∗ ] − Bx [Bx vx + By vy + Bz vz ], (3.7)
where [Q] = Qd − Qu is the difference between the downstream and upstream values of
a quantity Q, W = −(ρvx )u = −(ρvx )d is the Lagrangian speed of the discontinuity in
the mass coordinates, and V = 1/ρ. The other quantities are as defined in §2. Thus,
given the upstream state and an estimate of the Lagrangian speed of the discontinuity, the
downstream state can be computed.

The fast shock speed Wf and the slow shock speed Ws in the Lagrangian mass
coordinates are given in DW as
1 1
 r 
2 =
Wf,s (Cs2 + Cf2 + S1 ) ± (Cs2 + Cf2 + S1 )2 − 4(1 + S0 )(Cs2 Cf2 − S2 ) , (3.8)
2 1 + S0
where Cf = ρcf and Cs = ρcs . The upper (lower) sign refers to fast (slow) waves. The
coefficients S0 , S1 , S2 can be written in terms of the jump in tangential magnetic field
across the shock, [B⊥], as
1 [B ]
S0 = − (γ − 1) ⊥ , (3.9)
2 B⊥
 16
( )
1 2 [B⊥ ] + 2C 2 − (γ − 4)C 2 − 2γC 2 [B⊥ ] ,
S1 = −(γ − 2)C⊥ o ⊥ a (3.10)
2 B⊥ B⊥
Ca2 ([B⊥ ])2 (γ + 2)C⊥ Ca2 [B⊥ ]
( )
1 2 2 [B⊥ ]
S2 = + √ + C⊥ Ca (γ + 1) + (γ + 1)Ca4 − 2Co2 Ca2 ,
2 V V B⊥
(3.11)
where Co = ρa is the Lagrangian sound speed, Ca = ρca is the Lagrangian Alfvén speed,
√ q
C⊥ = ρB⊥ , and B⊥ = By2 + Bz2 . All quantities except [B⊥ ] in equations (3.9) to
(3.11) are referred to the state upstream of the shock. The expressions used in equations
(3.9) to (3.11) are equivalent to those given by DW, but are somewhat simpler and in
our experience have provided more robust behavior in the Riemann solver, particularly in
switch-on or switch-off shocks. An expression analogous to equation (3.8) appropriate to
the special case Bx = 0 (magnetosonic shocks) is also given by DW. As is well known, fast
and slow shocks do not alter the plane of the magnetic field.

Rotational discontinuities, which are not compressive, can be handled by setting the
jump [vx ] = 0 in equations (3.1) and (3.2). That leads necessarily to Wa = ±Ca . The
jump conditions required at the contact discontinuity can be found by setting W = 0 in
equations (3.1) to (3.7). With these results, it is clear that if the jumps of [B⊥ ] across
shocks and the rotations of [B⊥ ] across rotational discontinuities are known then it should
be possible to exactly determine the structure of any ideal MHD Riemann problem that
involves only discontinuous interfaces.

DW included rarefactions in their MHD Riemann solver by assuming they could also
be treated as discontinuities (i.e., as “rarefaction shocks”). So long as the rarefactions
are weak this is reasonably accurate, but not exact. In fact, it is also straightforward
to include fast and slow rarefactions exactly, just as for hydrodynamics. By conserving
all Riemann invariants through the rarefactions except that associated with the particular
wave involved, one can derive a simple set of differential equations to be integrated through
the rarefactions. The transitions computed in this way then replace the jumps given in
equations (3.1) to (3.7).

The relations appropriate to right facing (upper sign) and left facing (lower sign) fast
rarefactions are (e.g., Jeffrey 1966)
2 2 2
γ + 1√ C⊥ Cs2 γ + 1 √ Cs (Cf − Ca ) γ + 1 ρp′
Co′ =− ρ = ρ = , (3.12)
2 Co (Cs2 − Ca2 ) 2 Ca2 C⊥ Co 2 Co

1 C⊥ Ca2 Cf2 − Ca2 2 Co Ca2 Co′

u′x = ∓√ = ± √ = ± , (3.13)
ρ Cf (Cs2 − Ca2 ) ρC⊥ Cf γ + 1 Cs2 Cf ρ
u′y u′ 1 Ca
= z = ∓√ . (3.14)
cos ψ sin ψ ρ Cf
 17
For right and left facing slow rarefactions one finds

γ + 1√ C⊥ Cf2 2 2
γ + 1 √ Cf (Cs − Ca )
2
γ + 1 ρp′
Co′ =− ρ = ρ = , (3.15)
2 Co (Cf2 − Ca2 ) 2 Ca2 C⊥ Co 2 Co

1 C⊥ Ca2 Cs2 − Ca2 2 Co Ca2 Co′

u′x = ∓√ = ± √ =± , (3.16)
ρ Cs (Cf2 − Ca2 ) ρC⊥ Cs γ + 1 Cf2 Cs ρ

u′y u′ 1 Ca
= z = ∓√ . (3.17)
cos ψ sin ψ ρ Cs
In these expressions, tan ψ = Bz /By , and primes represent derivatives with respect to B⊥ .
It is also useful to have the relationship

(Cf2 − Ca2 )(Cs2 − Ca2 ) = −Ca2 C⊥

2
. (3.18)

Our procedure for obtaining an accurate solution to a Riemann problem is similar to

that employed by DW. We utilize the following steps:

1) As an initial guess, we found it convenient and mostly reliable to use the eigenvectors
defined in §2. In particular if the state vector, q, defined in equation (2.6) is represented
in each of the intermediate regions by q(i) (i = 2, 3, ...7) then an estimate of q(i) is simply
7
X
q(i) = q(8) − α k Rk , (3.19)
k=i+1

where αk is defined as in equation (2.90), and q(8) is the initial right state. Occasionally
the states found by equation (3.19) can have unphysical properties (i.e., jumps in B⊥ that
violate equations (3.1) to (3.7)), so some simple physical constraints need to be applied.

2) The quantities B⊥ (2), B⊥ (4), B⊥ (7) and ψ(3) from that solution are applied to deter-
mine jumps across each of the six waves (e.g. [B⊥ ]1→2 ), subject to the constraints that
B⊥ (5) = B⊥ (4) and ψ(4 − 6) = ψ(3). This is accomplished by starting with the two fast
waves, followed by the two rotations of B⊥ and finally the two slow waves.

3) For the solution to be considered exact, we require that the resulting jump conditions
at the contact discontinuity precisely satisfy those expected from equations (3.1) to (3.7).
In particular, we test if all the jumps [vx ], [vy ], [vz ] and [p∗ ] = 0.

4) If the contact discontinuity jump conditions are not satisfied, we vary B⊥ (2), B⊥ (4),
B⊥ (7) and ψ(3) using a Newton-Raphson scheme based on a numerical approximation
to the associated Jacobian matrix for an improved guess in the quantities B⊥ (2), B⊥ (4),
B⊥ (7) and ψ(3).
 18
5) The procedure, beginning with step 2), is then repeated until convergence is obtained.
The accuracy of the initial guess is the single most important aspect controlling the num-
ber of iterations required. Once a reasonably approximate solution is found, usually only
a couple of iterations lead to very good convergence.

When Bx = 0 the same scheme is applied, but in somewhat simplified form, since there
are no slow wave features or rotational discontinuities and the constraint ψ(4) = ψ(5) is
removed.

With these procedures, we are able in most cases to obtain solutions such that the
contact discontinuity jump conditions are very well satisfied, even to near the limits of
machine accuracy. In practice, we establish a convergence criterion of 10−6 relative to the
larger of a zone averaged vx or fast wave speed for velocities or total pressure for [p∗ ]. Only
for Riemann problems involving switch-off or switch-on waves, we are significantly limited,
since an exact switch-on or switch-off feature requires either the upstream or downstream
flow speed to exactly equal the Alfvén speed. In those cases, even very small errors in
[B⊥ ] lead to significant errors in W , and so the remaining downstream state variables. We
found that we could obtain precise solutions only if we permitted the smaller B⊥ next to
such a wave to have a value ∼ 10−3 of the larger B⊥ . These waves are for almost any
practical purpose indistinguishable from exact switch-waves, however.

We will discuss tests of the Riemann solver along with the tests of the MHD TVD
code in the next section, since we test them against each other in many cases. Suffice
it to say here that, to start with, we examined our Riemann solver against all those
solutions presented in DW and against a number of other examples kindly given to us by
Dr. Dai which were also generated with the Riemann solver described in DW. In the cases
involving only discontinuities, our solutions agree exactly with the DW results. When
there are rarefactions, we find some differences. But those are attributable to the fact
that we treated rarefactions exactly, whereas they did not. Mostly those differences are
relatively small, indicating that treating rarefactions as shocks would produce a reasonable
result in an approximate MHD Riemann solver.

4. NUMERICAL TESTS

To test the code described in §2 as well as the Riemann solver described in §3, we chose
MHD shock tube problems including those considered in BW and DW. In all the tests,
we set the adiabatic index γ = 5/3 and used a one-dimensional box with x = [0, 1]. The
numerical calculations were done with a Courant constant 0.8 and without the stiffening
of the contact discontinuity. The results of the numerical calculations with the code are
plotted as dots and the analytic solutions of the Riemann solver are plotted as lines. The
plotted quantities are density, gas pressure, total (thermal, kinetic, and magnetic) energy,
x-velocity (parallel to the direction of structure propagation), y-velocity, z-velocity, y-
magnetic field, z-magnetic field, and the orientation angle of tangential magnetic field
(ψ = tan−1 (Bz /By )) in the plane perpendicular to the propagation vector. Numerical
 19
values of the analytic solutions in the regions between the structures (e.g. between the left
moving fast shock and the left moving rotational discontinuity, etc) are listed in the tables
with the same labels as the figures. To simplify the discussion of this section we refer to
solutions as two-dimensional when the magnetic field remains in one plane through the
entire structure, or three-dimensional when the field cannot be so described.

The first set of tests (also found in DW) has been done with two-dimensional field and
velocity structure in the x − y plane but without change in the direction of the tangential
magnetic field (By in these tests). Fig. 1a shows the solution of √ the MHD shock tube test
with the left state (ρ, √ vx , vy , vz , By , Bz , E) = √
(1, 10, 0, 0, 5/ 4π, 0, 20) and the right
state (1, −10, 0, 0, 5/ 4π, 0, 1) with Bx = 5/ 4π at time t = 0.08. The plot shows a
pair of fast shocks, a left facing slow rarefaction, a right facing slow shock and a contact
discontinuity. Fig.1 b shows the solution of the √ MHD shock tube test with the left state
(ρ,√vx , vy , vz , By , Bz , E) = √ (1, 0, 0, 0, 5/ 4π, 0, 1) and the right state (0.1, 0, 0, 0,
2/ 4π, 0, 10) with Bx = 3/ 4π at time t = 0.03. The plot shows one fast shock and one
fast rarefaction, one slow shock and one slow rarefaction, and a contact discontinuity. As
expected from the two-dimensional nature of the field and velocity structure, there is no
rotational discontinuity in either of these flows. In the numerical calculations, fast shocks
that are strong with a large parallel velocity jump, [vx ], are resolved within 2-4 cells, while
slow shocks that are weak with a small jump require more cells to be resolved. Contact
discontinuity spread typically spread over 6-8 or so cells.

The second set of tests (also from DW) involves three-dimensional field and velocity
structure where the magnetic field plane rotates. The solution of the MHD √ shock √
tube
test with the left state (ρ, vx , vy , vz , B , B
√y z √, E) = (1.08, 1.2, 0.01, 0.5,
√ 3.6/ 4π, 2/ 4π,
0.95) and the right state (1, 0, 0, 0, 4/ 4π, 2/ 4π, 1) with Bx = 2/ 4π at time t = 0.2 is
plotted in Fig. 2a. Fast shocks, rotational discontinuities, and slow shocks propagate from
each side of the contact discontinuity. The solution √ of the MHD shock tube test with the
left state
√ (ρ, vx , vy , vz , By , Bz , E)
√ = (1, 0, 0, 0, 6/ 4π, 0, 1) and the right state (0.1, 0,
2, 1, 1/ 4π, 0, 10) with Bx = 3/ 4π at time t = 0.035 is plotted in Fig. 2b. A fast shock,
a rotational discontinuity, and a slow shock propagate from the left side of the contact
discontinuity, while a fast rarefaction, a rotational discontinuity, and a slow rarefaction
propagate to the right. The rotation across the initial discontinuity of the magnetic field
generates two rotational discontinuities. As in the previous case, strong fast shocks are
resolved within 2-4 cells, while weak slow shocks take more cells. With proper stiffening,
the rotational discontinuities spread over only 3-5 cells while the contact discontinuity
spreads over more cells.

In the third set, handling of magnetosonic structures with vanishing tangential flow
velocity and parallel magnetic field is tested. A test from DW for magnetosonic √ √shocks is
set up with the left state (ρ, vx , vy , √
vz , By , B
√ z , E) = (0.1, 50, 0, 0, −1/ 4π, −2/ 4π, 0.4)
and the right state (0.1, 0, 0, 0, 1/ 4π, 2/ 4π, 0.2) and with Bx = 0. The solution has
a pair of magnetosonic shocks propagating from a tangential discontinuity and is plotted
at time t = 0.01 in Fig. 3a. The test for magnetosonic rarefactions are set up with the left
state (ρ, vx , vy , vz , By , Bz , E) = (1, −1, 0, 0, 1, 0, 1) and the right state (1, 1, 0, 0, 1, 0,
1) and with Bx = 0. The solution has only two identical magnetosonic rarefactions and is
 20
plotted at time t = 0.1 in Fig. 3b. Magnetosonic shocks are fast shocks with zero parallel
field (Bx = 0) and they are resolved within 2-4 cells like fast shocks. The tangential
discontinuity, which is a degenerate combination of one contact discontinuity, two slow
structures, and two rotational discontinuities, spreads typically more than 10 cells. Even
with the stiffening described in §2, we could not prevent the spreading of the tangential
discontinuity in numerical calculations. But the Riemann solutions show that even slight
differences from an exact tangential magnetic field, Bx = 0, lead to the formation of the
more complex and expanding set of features already mentioned.

In the fourth set, we test how the MHD-TVD code and the Riemann solver deal with a
special category of fast and slow structures; namely, the so-called “switch-on” and “switch-
off” structures. The Tangential magnetic field turns on in the region behind switch-on fast
shocks and switch-on slow rarefactions. On the other hand, it turns off in the region behind
switch-off slow shocks and switch-off fast rarefactions. The test involving a switch-on fast
shock has been set up with the left state (ρ, vx , vy , vz , By , Bz , E) = (1, 0, 0, 0, 1, 0,
1) and the right state (0.2, 0, 0, 0, 0, 0, 0.1) and with Bx = 1. The switch-on fast shock
propagates to the right. Other structures formed in this test include a fast rarefaction, a
slow rarefaction, a contact discontinuity, and a slow shock. The test results are plotted
at time t = 0.15 in Fig. 4a, showing that the code and the Riemann solver handle the
switch-on fast shock without any trouble. The test involving a switch-off fast rarefaction
has been set up with the left state (ρ, vx , vy , vz , By , Bz , E) = (0.4, −0.66991, 0.98263,
0, 0.0025293, 0, 0.52467) and the right state (1, 0, 0, 0, 1, 0, 1) and with Bx = 1.3. It has
been designed to generate only a right-moving, switch-off fast rarefaction with a contact
discontinuity with an accuracy (the ratio of residual strength to pre-rarefaction strength of
tangential magnetic field) better than 0.3% (see Table 4b). The test results are plotted at
time t = 0.15 in Fig. 4b. The numerical calculation shows small, yet noticeable, signatures
of the right-moving slow structure and the left-moving hydrodynamic structure. The test
involving a switch-off slow shock has been set up with the left state (ρ, vx , vy , vz , By ,
Bz , E) = (0.65, 0.667, −0.257, 0, 0.55, 0, 0.5) and the right state (1, 0.4, −0.94, 0, 0,
0, 0.75) and with Bx = 0.75. The solution of this test has, from left to right, a fast
shock, a switch-off slow shock, a contact discontinuity, and a hydrodynamics shocks. In
the region to the right of the switch-off slow shock hydrodynamic structures form with
vanishing tangential magnetic field. The test results are plotted at time time t = 0.15
in Fig. 4c, showing good agreement between the numerical calculation and the analytic
solution. Our Riemann solver turns off the tangential magnetic field up to an accuracy
of about 0.04% as indicated in Table 4c. The test involving a switch-on slow rarefaction
has been set up with the left state (ρ, vx , vy , vz , By , Bz , E) = (1, 0, 0, 0, 0, 0, 1) and
the right state (0.3, 0, 0, 1, 1, 0, 0.2) and with Bx = 0.7. The structures formed in this
test include a hydrodynamic rarefaction, a switch-on slow shock, a contact discontinuity,
a slow shock, a rotational discontinuity, and a fast rarefaction. With the hydrodynamic
nature in the region between the two left-moving rarefactions, no rotational discontinuity
is produced on the left side of the contact discontinuity. Hence this test has a solution with
only one rotational discontinuity, while two discontinuities are expected in most tests with
three-dimensional field structure (e.g. the tests in Fig. 2). The test results are plotted at
time t = 0.16 in Fig. 4d.
 21
The final set of tests involves compound structures, whose existence was first discussed
in BW. The initial setups includes two-dimensional field and velocity structure in the
x − y plane and the sign of the tangential magnetic field (By ) changes across the initial
discontinuity. The test involving a slow compound structure has been set up with the left
state (ρ, vx , vy , vz , By , Bz , E) = (1, 0, 0, 0, 1, 0, 1) and the right state (0.125, 0, 0, 0,
−1, 0, 0.1) and with Bx = 0.75. It is the same test as in BW, except we set the adiabatic
index, γ = 5/3, instead of γ = 2. The results are plotted at time t = 0.1 in Fig. 5a. Here
lines are the solution of the Riemann solver which has a rotational discontinuity and a slow
shock instead of a slow compound structure, and numerical values of the solution are listed
in Table 5a. Clearly the numerical calculations and the analytic solution produce different
results around the compound structure, even though the agreement in other structures:
two fast rarefactions, a slow shock, and a contact discontinuity, is acceptable for most
purposes. Similarly, we set up a test involving a fast compound with the left state (ρ,
vx , vy , vz , By , Bz , E) = (1, 0, 0, 0, 1, 0, 1) and the right state (0.4, 0, 0, 0, −1, 0, 0.4)
and with Bx = 1.3. The results are plotted at time t = 0.16 in Fig. 5b. Again lines are
the solution of the Riemann solver with a rotational discontinuity and a fast rarefaction
instead of a fast compound structure, and numerical values of the solution are listed in
Table 5b. Further discussion on compound structures will follow in the next section.

In the above tests with the MHD shock tube problems, the general and detailed
agreement between the numerical calculations and the analytic solutions are satisfactory,
and mostly excellent, except in the tests involving compound structures. Especially, the
positions of shocks, rarefactions, and discontinuities agree very well, indicating the both
generate the same solutions. Numerical results show that our code resolves strong shocks
(fast, slow, or magnetosonic) typically within 2-4 cells, while more cells are required if
shocks are weak with small jump in the parallel flow velocity. Rotational discontinuities
are resolved within 3-5 cells with proper stiffening, and contact discontinuities are resolved
within 6-8 cells without stiffening. While the stiffening of contact discontinuities make
them look sharper in some cases, it generates spurious numerical oscillations in many cases.
Tangential discontinuities spread over more than 10 cells. Further work could be done to
develop proper stiffening schemes for contact discontinuities and tangential discontinuities
to improve the ability of the MHD-TVD code to handle these discontinuities.

5. DISCUSSION

One test problem that has attracted considerable attention was one originally presented
by BW involving a slow compound structure, which is related to structures known as
“intermediate” shocks (e.g., Wu 1988; Kennel et al. 1990). This structure reverses the
direction of B⊥ and leads to a flow that passes from super-Alfvénic to sub-Alfvénic.
Numerical MHD codes, which are always dissipative, seem generally to find the BW
compound structure out of that shock-tube initial condition. For the compound wave
it does not appear necessary that the magnetic field anywhere exist outside the upstream
or downstream field plane. Thus, it would appear that this structure might be represented
as a “1 + 1/2 dimensional flow problem”. However, ideal MHD Riemann solvers that
allow rotational discontinuities will generally find a solution to that shock-tube problem
 22
that interprets the compound structure as a rotational discontinuity followed by a slow
shock, making it a “1 + 1/2 + 1/2 dimensional” flow.

It is not our intent to add to the controversy surrounding the reality of intermediate
shocks. However, in order to test the performance as well as understand the properties of
our code, we have done a high resolution calculation for the test of the slow compound
structure in Fig. 5a with 8192 cells. The magnified region around the slow compound
structure has been plotted in Fig. 6a. From the plot, the numerical values of several
flow quantities on the both sides of the “shock” read: ρ = 0.647 (density), pg = 0.484
(gas pressure), u = 0.963 (flow velocity in the shock frame), By = 0.536 (tangential
magnetic field), cf = 1.423 (fast speed), ca = 0.932 (Alfvén speed), cs = 0.732 (slow
speed), a = 1.117 (sound speed) in the preshock region, ρ = 0.827, pg = 0.758, u = 0.751,
By = −0.0738, cf = 1.241, ca = 0.825, cs = 0.822, a = 1.236 in the postshock region,
and shock velocity in lab frame is 0.284. In the preshock region the flow velocity is
ca < u < cf (sub-fast but super-Alfvénic) with Alfvénic Mach number Ma = 1.03, while
in the postshock region u < cs (sub-slow) with Ma = 0.910. Hence, the shock in the
numerical calculation with our code should be considered as a “2-4” intermediate shock
and the slow compound structure as a intermediate shock followed by a slow rarefaction.

We have done also a high resolution calculation for the test of the fast compound
structure in Fig. 5b with 8192 cells and the magnified region around the fast compound
structure has been plotted in Fig. 6b. The numerical values in the plot of several flow
quantities on the both sides of the “shock” read: ρ = 0.6820, pg = 0.5290, u = 1.617,
By = 0.0555, cf = 1.577, ca = 1.574, cs = 1.135, a = 1.137 in the preshock region,
ρ = 0.7361, pg = 0.6031, u = 1.498, By = −0.3442, cf = 1.622, ca = 1.515, cs = 1.092,
a = 1.169 in the postshock region, and shock velocity in lab frame is 0.935. In the preshock
region the flow velocity is u > cf (super-fast) with Ma = 1.027, while in the postshock
region cs < u < ca (sub-Alfvénic but super-slow) with Ma = 0.989. Hence, the shock in
the numerical calculation with our code should be considered as a “1-3” intermediate shock
and the fast compound structure as a intermediate shock preceded by a fast rarefaction.

Acknowledgments

The work by DR was supported in part by David and Lucille Packard Foundation Fel-
lowship through Jeremy Goodman at Princeton University and in part by Non-Directed
Research Fund of Korea Research Foundation 1993 at Chungnam National University.
Work by TWJ was supported in part by NASA through grant NAGW-2548, the NSF
through grant AST-9100486, by a travel grant from the University of Minnesota Inter-
national Programs Office and by the University of Minnesota Supercomputer Institute
(UMSI). The UMSI also generously contributed support for travel by DR.
 23
REFERENCES
Brio, M., & Wu, C. C. 1988, J. Comp. Phys., 75, 500.
Colella, P., & Woodward, P. R. 1984, J. Comp. Phys., 54, 174.
Dai W., & Woodward P. R. 1994, J. of Comput. Phys., in press.
DeVore, C. R. 1991, J. of Comput. Phys., 92, 142.
Harten, A. 1983, J. Comp. Phys., 49, 357.
Jeffrey, A. 1966, Magnetohydrodynamics (London: Oliver and Boyd).
Jeffrey A., & Taniuti, T. 1964, Nonlinear Waves Propagation (New York: Academic Press).
Kennel, C. F., Blandford, R. D. & Wu, C. C. 1990, Phys. Fluids B, 2, 253.
Lind, K. R., Payne, D. G. & Meier, D. L. 1991, preprint.
Parker, E. N. 1979, Cosmical Magnetic Fields: Their Origin and Their Activity (Oxford:
Oxford University Press).
Roe, P. L. 1981, J. Comp. Phys., 43, 357.
Shu, F. H. 1992, The Physics of Astrophysics, Volume II: Gad Dynamics (Mill Valley:
University Science Books).
Stone, J. M., Hawley, J. F., Evans, C. R. & Norman, M. L. 1992, ApJ, 388, 415.
Stone, J. M., & Norman, M. L. 1992, ApJS, 80, 791.
Wu, C. C. 1988, J. Geophys. Research, 93, 987.
Zachary A. L., & Colella P. 1992, J. of Comput. Phys., 99, 341.
 24
FIGURE CAPTIONS
Fig. 1a.— Solution of the√MHD shock tube test with the left state (ρ, vx ,√vy , vz , By , Bz ,
E) = (1,√10, 0, 0, 5/ 4π, 0, 20) and the right state (1, −10, 0, 0, 5/ 4π, 0, 1) with
Bx = 5/ 4π and γ = 5/3 at time t = 0.08 (test in DW table 7). Dots are the result
of a numerical calculation with the MHD-TVD code described in §2 using 512 cells
and a Courant constant of 0.8. Lines are the result with the nonlinear Riemann solver
described in §3. Plots show from left to right (1) fast shock, (2) slow rarefaction, (3)
contact discontinuity, (4) slow shock, and (5) fast shock.
Fig. 1b.— Solution of the MHD √ shock tube test with the left state (ρ, vx , √vy , vz , By ,
Bz , E) = (1, √0, 0, 0, 5/ 4π, 0, 1) and the right state (0.1, 0, 0, 0, 2/ 4π, 0, 10)
with Bx = 3/ 4π and γ = 5/3 at time t = 0.03 (test in DW table 3a). Dots are the
result of a numerical calculation with the MHD-TVD code described in §2 using 512
cells and a Courant constant of 0.8. Lines are the result with the nonlinear Riemann
solver described in §3. Plots show from left to right (1) fast shock, (2) slow shock, (3)
contact discontinuity, (4) slow rarefaction, and (5) fast rarefaction.
Fig. 2a.— Solution of the MHD shock √ tube test√with the left state (ρ, vx , vy , vz , By ,
Bz , E) √ 1.2, 0.01, 0.5, 3.6/ √4π, 2/ 4π, 0.95) and the right state (1, 0, 0,
√ = (1.08,
0, 4/ 4π, 2/ 4π, 1) with Bx = 2/ 4π and γ = 5/3 at time t = 0.2 (test in DW
table 1a). Dots are the result of a numerical calculation with the MHD-TVD code
described in §2 using 512 cells and a Courant constant of 0.8. Lines are the result
with the nonlinear Riemann solver described in §3. Plots show from left to right (1)
fast shock, (2) rotational discontinuity, (3) slow shock, (4) contact discontinuity, (5)
slow shock, (6) rotational discontinuity, and (7) fast shock.
Fig. 2b.— Solution of the MHD √ shock tube test with the left state (ρ, vx , √vy , vz , By ,
Bz , E) = (1, 0,√ 0, 0, 6/ 4π, 0, 1) and the right state (0.1, 0, 2, 1, 1/ 4π, 0, 10)
with Bx = 3/ 4π and γ = 5/3 at time t = 0.035 (test in DW table 5a). Dots
are the result of a numerical calculation with the MHD-TVD code described in §2
using 512 cells and a Courant constant of 0.8. Lines are the result with the nonlinear
Riemann solver described in §3. Plots show from left to right (1) fast shock, (2)
rotational discontinuity, (3) slow shock, (4) contact discontinuity, (5) slow rarefaction,
(6) rotational discontinuity, and (7) fast rarefaction.
Fig. 3a.— Solution of the MHD √ shock √ tube test with the left state (ρ, vx , vy , vz , By√
, Bz ,
E)√= (0.1, 50, 0, 0, −1/ 4π, −2/ 4π, 0.4) and the right state (0.1, 0, 0, 0, 1/ 4π,
2/ 4π, 0.2) with Bx = 0 and γ = 5/3 at time t = 0.01 (test in DW table 2a). Dots
are the result of a numerical calculation with the MHD-TVD code described in §2
using 512 cells and a Courant constant of 0.8. Lines are the result with the nonlinear
Riemann solver described in §3. Plots show from left to right (1) magnetosonic shock,
(2) tangential discontinuity, and (3) magnetosonic shock.
Fig. 3b.— Solution of the MHD shock tube test with the left state (ρ, vx , vy , vz , By ,
Bz , E) = (1, −1, 0, 0, 1, 0, 1) and the right state (1, 1, 0, 0, 1, 0, 1) with Bx = 0
and γ = 5/3 at time t = 0.1. Dots are the result of a numerical calculation with the
MHD-TVD code described in §2 using 512 cells and a Courant constant of 0.8. Lines
are the result with the nonlinear Riemann solver described in §3. Plots show from left
to right (1) magnetosonic rarefaction, and (2) magnetosonic rarefaction.
 25
Fig. 4a.— Solution of the MHD shock tube test with the left state (ρ, vx , vy , vz , By , Bz ,
E) = (1, 0, 0, 0, 1, 0, 1) and the right state (0.2, 0, 0, 0, 0, 0, 0.1) with Bx = 1 and
γ = 5/3 at time t = 0.15. Dots are the result of a numerical calculation with the
MHD-TVD code described in §2 using 512 cells and a Courant constant of 0.8. Lines
are the result with the nonlinear Riemann solver described in §3. Plots show from left
to right (1) fast rarefaction, (2) slow rarefaction, (3) contact discontinuity, (4) slow
shock, and (5) switch-on fast shock.
Fig. 4b.— Solution of the MHD shock tube test with the left state (ρ, vx , vy , vz , By , Bz ,
E) = (0.4, −0.66991, 0.98263, 0, 0.0025293, 0, 0.52467) and the right state (1, 0, 0, 0,
1, 0, 1) with Bx = 1.3 and γ = 5/3 at time t = 0.15. Dots are the result of a numerical
calculation with the MHD-TVD code described in §2 using 512 cells and a Courant
constant of 0.8. Lines are the result with the nonlinear Riemann solver described
in §3. Plots show from left to right (1) contact discontinuity, and (2) switch-off fast
rarefaction.
Fig. 4c.— Solution of the MHD shock tube test with the left state (ρ, vx , vy , vz , By , Bz ,
E) = (0.65, 0.667, −0.257, 0, 0.55, 0, 0.5) and the right state (1, 0.4, −0.94, 0, 0, 0,
0.75) with Bx = 0.75 and γ = 5/3 at time t = 0.15. Dots are the result of a numerical
calculation with the MHD-TVD code described in §2 using 512 cells and a Courant
constant of 0.8. Lines are the result with the nonlinear Riemann solver described in
§3. Plots show from left to right (1) fast shock, (2) switch-off slow shock, (3) contact
discontinuity, and (4) hydrodynamic shock.
Fig. 4d.— Solution of the MHD shock tube test with the left state (ρ, vx , vy , vz , By , Bz ,
E) = (1, 0, 0, 0, 0, 0, 1) and the right state (0.3, 0, 0, 1, 1, 0, 0.2) with Bx = 0.7
and γ = 5/3 at time t = 0.16. Dots are the result of a numerical calculation with the
MHD-TVD code described in §2 using 512 cells and a Courant constant of 0.8. Lines
are the result with the nonlinear Riemann solver described in §3. Plots show from
left to right (1) hydrodynamic rarefaction, (2) switch-on slow rarefaction, (3) contact
discontinuity, (4) slow shock, (5) rotational discontinuity, and (6) fast rarefaction.
Fig. 5a.— Solution of the MHD shock tube test with the left state (ρ, vx , vy , vz , By , Bz , E)
= (1, 0, 0, 0, 1, 0, 1) and the right state (0.125, 0, 0, 0, −1, 0, 0.1) with Bx = 0.75 and
γ = 5/3 at time t = 0.1 (test in BW). Dots are the result of a numerical calculation
with the MHD-TVD code described in §2 using 512 cells and a Courant constant of
0.8. Lines are the result with the nonlinear Riemann solver described in §3. Plots show
from left to right (1) fast rarefaction, (2) slow compound, (3) contact discontinuity,
(4) slow shock, and (5) fast rarefaction.
Fig. 5b.— Solution of the MHD shock tube test with the left state (ρ, vx , vy , vz , By , Bz ,
E) = (1, 0, 0, 0, 1, 0, 1) and the right state (0.4, 0, 0, 0, −1, 0, 0.4) with Bx = 1.3
and γ = 5/3 at time t = 0.16. Dots are the result of a numerical calculation with the
MHD-TVD code described in §2 using 512 cells and a Courant constant of 0.8. Lines
are the result with the nonlinear Riemann solver described in §3. Plots show from left
to right (1) fast compound, (2) slow shock, (3) contact discontinuity, (4) slow shock,
and (5) fast rarefaction.
Fig. 6a.— Same slow compound structure in the MHD shock tube test as that in Fig. 5a.
The calculation has been done with the MHD-TVD code described in §2 using 8192
cells and only the region around the slow compound structure has been plotted. Plots
 26
show that the slow compound structure from the numerical calculation is composed
of a “2-4” intermediate shock followed by a slow rarefaction.
Fig. 6b.— Same fast compound structure in the MHD shock tube test as that in Fig. 5b.
The calculation has been done with the MHD-TVD code described in §2 using 8192
cells and only the region around the fast compound structure has been plotted. Plots
show that the fast compound structure from the numerical calculation is composed of
a “1-3” intermediate shock preceded by a fast rarefaction.

View publication stats