On Flux Vector Splitting for the Euler Equations:

Some Recent Results *

Eleuterio Toro
Laboratory of Applied Mathematics
University of Trento, Italy
toro@ing.unitn.it
http://www.ing.unitn.it/toro
* E F Toro and M E Vazquez-Cendon. Flux Splitting Schemes for the Euler Equations. Computers and Fluids.
(Under review, April 2012).
1 / 46
Table of contents
1 The Euler Equations and Flux Splitting
2 The Liou-Steen Splitting
3 The Zha-Bilgen Splitting
4 A Flux-Splitting Framework
5 A Novel Splitting for the Euler Equations
6 Numerical Fluxes
7 Reinterpretation of Other Flux Splittings
8 Numerical Results for the Euler Equations
9 Other Potential Schemes for the Pressure System
10 Concluding Remarks
2 / 46
The Euler Equations and Flux Splitting
3 / 46
The Euler equations in one space dimension are

t
Q+
x
F(Q) = 0 , (1)
where Q is the vector of conserved variables and F(Q) is the ux vector,
both given as
Q =
_
_

u
E
_
_
, F(Q) =
_
_
u
u
2
+ p
u(E + p)
_
_
. (2)
Here is density, u is particle velocity, p is pressure and E is total energy
given as
E = (
1
2
u
2
+ e) . (3)
The specic internal energy e is, in general, a function of other variables
via an equation of state. For example, e may be taken to be a function of
density and pressure, namely
e = e(, p) . (4)
4 / 46
For ideal gases
e(, p) =
p
( 1)
, (5)
where 1 < < 3 is the ratio of specic heats. For air at moderate
pressures and temperatures one uses = 1.4.
For solving numerically equations of the type (1) we adopt a conservative
method of the form
Q
n+1
i
= Q
n
i

t
x
[F
i+
1
2
F
i
1
2
] , (6)
where F
i+
1
2
is the numerical ux. For background on the Euler equations
and conservative schemes of the form (6).
Flux vector splitting:
F(Q) = A(Q) +P(Q) , (7)
5 / 46
The Liou-Steen Splitting
6 / 46
Liou and Steen (1993) split the ux vector into and advection part A(Q)
and a pressure part P(Q) as follows
F(Q) = A(Q) +P(Q) ,
with
A(Q) =
_
_
u
u
2
uH
_
_
, P(Q) =
_
_
0
p
0
_
_
, (8)
where
H =
E + p

(9)
is the enthalpy.
From the numerical point of view the aim is to obtain a numerical ux for
(6) of the form
F
i+
1
2
= A
i+
1
2
+P
i+
1
2
(10)
by nding partial advection and pressure uxes A
i+
1
2
and P
i+
1
2
.
7 / 46
To begin with, Liou and Steen express A(Q) as
A(Q) = M
_
_
a
au
aH
_
_
, (11)
where M = u/a is the Mach number and a =
_
p/ is the speed of
sound.
The advection ux is then taken as
A
i+
1
2
= M
i+
1
2

A
i+
1
2
, (12)
with

A
i+
1
2
=
_

_
_
_
a
au
aH
_
_
n
i
if M
i+
1
2
0 ,
_
_
a
au
aH
_
_
n
i+1
if M
i+
1
2
< 0 .
(13)
8 / 46
The advection ux is upwinded according to advection speed implied in
the Mach number M
i+
1
2
, which is split as
M
i+
1
2
= M
+
i
+ M

i+1
, (14)
with
M

=
_

1
4
(M1)
2
if | M | 1 ,
1
2
(M | M |) if | M | > 1 .
(15)
The pressure vector P
i+
1
2
is constructed by splitting the pressure as
p
i+
1
2
= p
+
i
+ p

i+1
, (16)
with two choices for the negative and positive components as follows
p

=
_
1
2
p(1M) if | M | 1 ,
1
2
p
(M|M|)
M
if | M | > 1 ,
(17)
and
p

=
_
1
4
p(M1)
2
(2 M) if | M | 1 ,
1
2
p
(M|M|)
M
if | M | > 1 .
(18)
9 / 46
The Zha-Bilgen Splitting
10 / 46
Zha and Bilgen (1993) split the ux vector into as in (7) with
A(Q) =
_
_
u
u
2
uE
_
_
, P(Q) =
_
_
0
p
pu
_
_
. (19)
Numerically, they propose uxes A
i+
1
2
and P
i+
1
2
as follows.
A
i+
1
2
= A
+
i
+A

i+1
, (20)
where
A

i
= min(0, u
n
i
)Q
n
i
, A
+
i
= max(0, u
n
i
)Q
n
i
. (21)
For the pressure ux vector Zha and Bilgen use the splitting
P
i+
1
2
= P
+
i
+P

i+1
. (22)
11 / 46
For the p component Zha and Bilgen adopt the Liou-Steen splitting (16),
(17), while for the pu component they propose
(pu)
i+
1
2
= (pu)
+
i
+ (pu)

i+1
, (23)
where
(pu)

i
= p
n
i
_

_
u
n
i
if M
n
i
1 ,
1
2
(u
n
i
a
n
i
) if 1 < M
n
i
< 1 ,
0 if M
n
i
1 ,
(24)
and
(pu)
+
i
= p
n
i
_

_
0 if M
n
i
1 ,
1
2
(u
n
i
+ a
n
i
) if 1 < M
n
i
< 1 ,
u
n
i
if M
n
i
1 .
(25)
Finally the Zha-Bilgen numerical ux is
F
i+
1
2
= A
+
i
+P
+
i
+A

i+1
+P

i+1
. (26)
12 / 46
A Flux-Splitting Framework
13 / 46
The framework
We propose to split system (1) via the ux splitting (7) into the two
systems

t
Q+
x
A(Q) = 0 ,

t
Q+
x
P(Q) = 0 ,
_
_
_
(27)
called respectively the advection system and the pressure system. The aim
is then to compute a numerical ux as
F
i+
1
2
= A
i+
1
2
+P
i+
1
2
, (28)
where A
i+
1
2
and P
i+
1
2
are obtained respectively from appropriate Cauchy
problems for the advection and pressure systems (27).
14 / 46
Consider the Cauchy problem for the linear advection equation

t
q(x, t) +
x
q(x, t) = 0 , < x < , t > 0 ,
q(x, 0) = h(x) ,
_
(29)
where is a constant. The exact solution of IVP (29) after a time t is
q(x, t) = h(x t) . (30)
We now decompose the characteristic speed as
= + (1 ) =
a
+
p
, 0 1 , (31)
with denitions

a
= ,
p
= (1 ) , (32)
so as to obtain two linear partial dierential equations, namely

t
q +
a

x
q = 0 ,
t
q +
p

x
q = 0 . (33)
15 / 46
Now consider rst the Cauchy problem for the advection equation

t
q +
a

x
q = 0 ,
q(x, 0) = h(x) ,
_
(34)
the solution of which after a time t
1
is
q(x, t
1
) = h(x
a
t
1
) . (35)
Consider the Cauchy problem

t
q +
p

x
q = 0 ,
q(x, 0) = h(x
a
t
1
) .
_
(36)
The exact solution of IVP (36), after a time t
2
, is
q(x, t
2
) = h(x
a
t
1

p
t
2
) . (37)
The combined solution of IVPs (34) and (36) for t
1
= t
2
= t is
q(x, t) = h(x (
a
+
p
)t) = h(x t) = q(x, t) . (38)
16 / 46
The above result can be stated as the following proposition.
Proposition 3.1. The exact solution of the initial value problem (29) can
be obtained by solving in sequence the initial-value problems (34) and (36).
Remark 3.1. We note that in the wave decomposition (31),(32) of the
model problem (29) one can accept the characteristic speeds to be
arbitrarily dierent. For example, for > 0, by taking a very small in
(31) we would have
a
<<
p
, situation that resembles the slow advection
waves and the fast pressure waves.
Remark 3.2. From the numerical point of view, Proposition 3.1 suggests
a way to compute a numerical ux for IVP (29) by computing numerical
uxes for IVPs (34) and (36). This would lead to split ux vector splitting
methods and could potentially be of use to deal with systems in which
there is large disparity in the magnitude of the wave speeds present. This
line of enquiry has not been pursued, to my knowledge.
17 / 46
Remark 3.3. For the full non-linear problem, such as the Euler equations,
the proposed framework has two components: (a) the particular way the
full system is split into two subsystems, called here the advection and
pressure systems, and (b) the numerical treatment of each subsystem to
produce corresponding advection and pressure numerical uxes to make up
the numerical ux for the full system.
18 / 46
A Novel Splitting for the Euler Equations
19 / 46
Here we propose a new splitting for the Euler equations by noting that the
ux may be decomposed thus
F(Q) =
_
_
u
u
2
+ p
u(
1
2
u
2
+ e + p)
_
_
=
_
_
u
u
2
1
2
u
3
_
_
+
_
_
0
p
u(e + p)
_
_
, (39)
with the corresponding advection and pressure uxes dened as
A(Q) = u
_
_

u
1
2
u
2
_
_
, P(Q) =
_
_
0
p
u(e + p)
_
_
. (40)
We note that the proposed advection ux A(Q) contains no pressure
terms. All pressure terms from the ux F(Q), including that of the total
energy E, are now included in the pressure ux P(Q). The advection ux
may be interpreted as representing advection of mass, momentum and
kinetic energy. For the ideal gas case (5) the pressure ux (40) becomes
P(Q) =
_
_
0
p

1
pu
_
_
. (41)
20 / 46
The advection system is

t
Q+
x
A(Q) = 0 , (42)
where Q = [, u, E]
T
and A(Q) as in (40) above. In quasi-linear form
the advection system becomes

t
Q+M(Q)
x
Q = 0 , (43)
where M(Q) is the Jacobian matrix given as
M(Q) =
_

_
0 0 0
u
2
2u 0
u
3 3
2
u
2
0
_

_
. (44)
21 / 46
It is easy to show that the eigenvalues of this matrix are

1
= 0 ,
2
=
3
= u . (45)
There are only two linearly independent right eigenvectors, namely
R
1
=
1
_
_
1
0
0
_
_
, R
2
=
2
_
_
1
u
1
2
u
2
_
_
. (46)
Thus the system is weakly hyperbolic, as there is no complete set of
linearly independent eigenvectors.
Regarding the nature of the characteristic elds, it is easy to show that the

1
-eld is linearly degenerated and that the
2
-eld is genuinely non-linear
if
2
= 0 and u = 0; otherwise it is linearly degenerate.
22 / 46
The pressure system
In terms of the conserved variables Q = [, u, E]
T
the pressure system is

t
Q+
x
P(Q) = 0 , (47)
with P(Q) as given in (40) above. In quasi-linear form the advection
system becomes

t
Q+N(Q)
x
Q = 0 , (48)
where N(Q) is the Jacobian matrix given as
N(Q) =
_

_
0 0 0
1
2
( 1)u
2
( 1)u 1
u
3
uE/ E/
3
2
u
2
u
_

_
. (49)
23 / 46
The eigenvalues of N(Q) are always real and given as

1
=
1
2
u
1
2
A ,
2
= 0 ,
3
=
1
2
u +
1
2
A , (50)
where
A =
_
u
2
+ 4a
2
, a
2
=
p

. (51)
Here a is the usual speed of sound for the full Euler equations.
In terms of physical variables the system reads

t
V+B(V)
x
V = 0 , (52)
where
V =
_
_

u
p
_
_
, B =
_
_
0 0 0
0 0 1/
0 p u
_
_
. (53)
24 / 46
Note that, since u < A =

u
2
+ 4a
2
, the system is always subsonic, that
is

1
=
1
2
u
1
2
A < 0 <
3
=
1
2
u +
1
2
A . (54)
The right eigenvectors of matrix B in (53) corresponding to the
eigenvalues (50) are
R
1
=
_
_
0
2
(u A)
_
_
, R
2
=
_
_
1
0
0
_
_
, R
3
=
_
_
0
2
(u + A)
_
_
. (55)
25 / 46
Numerical Fluxes
In order to compute advection and pressure uxes A
i+
1
2
and P
i+
1
2
we
consider the Riemann problem for each subsystem. We start with the
pressure system.
To compute the ux for the pressure system we consider the Riemann
problem in terms of physical variables

t
V+B(V)
x
V = 0 ,
V(x, 0) =
_
V
L
V
n
i
if x < 0 ,
V
R
V
n
i+1
if x > 0 .
_

_
(56)
The solution of this problem has structure as shown in Fig. 1. The wave
pattern is always subsonic, with a stationary contact discontinuity and two
non-linear waves to the left and right of the contact wave.
26 / 46
t
x
p

(
2
=
1
2
(u + A))
(
1
=
1
2
(u A))
x = 0
Fig. 1. Structure of the solution of the Riemann problem for the pressure
system.
u

=
C
R
u
R
C
L
u
L
C
R
C
L

2
C
R
C
L
(p
R
p
L
) ,
p

=
C
R
p
L
C
L
p
R
C
R
C
L
+
1
2
C
R
C
L
C
R
C
L
(u
R
u
L
) ,
_

_
(57)
with
C
L
=
L
(u
L
A
L
) ; C
R
=
R
(u
R
+ A
R
) , (58)
where A
L
and A
R
are computed from (51).
27 / 46
One could improve upon the linear approximation by applying the exact
generalised Riemann invariants throughout. The result is the 2 2
non-linear system for p

and u

_
u
2

+ 4
p

L
= u
L

_
u
2
L
+ 4
p
L

L
,
u

+
_
u
2

+ 4
p

R
= u
L
+
_
u
2
R
+ 4
p
R

R
.
_

_
(59)
Finally, the numerical ux for the pressure system is given as
P
i+
1
2
= p

_
_
0
1

1
u

_
_
. (60)
Remark: no visible dierence between exact and approximate
solutions.
28 / 46
The advection system
Recall that in our splitting (40) the advection operator may be written thus
A(Q) =
_
_
u
u
2
1
2
u
3
_
_
= uK(Q) , K(Q) =
_
_

u
1
2
u
2
_
_
, (61)
Advection of K (mass, momentum and kinetic energy) with speed u. Here
we propose two algorithms.
Algorith 1 (TV scheme).
A(Q) = u

i+
1
2
K , (62)
where u

i+
1
2
is the intercell advection velocity taken as u

i+
1
2
from solution
(57) of the Riemann problem (56)
A
i+
1
2
= u

i+
1
2
_

_
K
n
i
if u

i+
1
2
> 0 ,
K
n
i+1
if u

i+
1
2
0 .
(63)
29 / 46
Algorith 2 (TV-AWS scheme). Here propose a weighted splitting scheme,
which is a simple modication of the scheme proposed by Zha and Bilgen
[?] for their advection system. The modied scheme is given as follows
A
i+
1
2
= A
+
i
+A

i+1
, (64)
with
A

i
=
1
2
(1 )u
n
i
K
n
i
, A
+
i
=
1
2
(1 + )u
n
i
K
n
i
. (65)
Here
= (u
n
i
) =
u
n
i
_
+ (u
n
i
)
2
, (66)
with a small positive quantity, = 0.1, for example. The function (u
n
i
)
allows a smooth transition from upwinding fully to the left and fully to the
right, in the vicinity of u
n
i
= 0.
The resulting scheme from Algorithm 2, called the TV-AWS scheme, is
eectively a weighted averaged scheme and the Zha-Bilgen scheme is
recovered from it by simply setting the weight to be = sign(u
n
i
).
30 / 46
Summary of the present scheme
In order to compute a numerical ux F
i+
1
2
for the conservative scheme (6)
we proceed as follows:
Pressure ux. Evaluate the intercell pressure p

i+
1
2
and velocity u

i+
1
2
from the solution of the Riemann problem given in (57) to compute
the pressure ux P
i+
1
2
as in (60).
Advection ux. We have proposed two options. From algorithm 1
(TV scheme) we evaluate the advection ux A
i+
1
2
as in (63).
Algorith 2 (TV-AWS) is described in equations (64) to (66).
Intercell ux. Compute the intercell ux F
i+
1
2
as in (28), namely
F
i+
1
2
= A
i+
1
2
+P
i+
1
2
. (67)
31 / 46
Reinterpretation of Other Flux Splittings
32 / 46
The Liou-Steen scheme
The Liou-Steen splitting (1993) may be interpreted in our framework
dening the advection system as

t
Q+
x
A(Q) = 0 , (68)
with
A(Q) =
_
_
u
u
2
u(E + p)
_
_
(69)
and the pressure system as

t
Q+
x
P(Q) = 0 , (70)
with
P(Q) =
_
_
0
p
0
_
_
. (71)
33 / 46
In terms of primitive variables V = [, u, p]
T
the pressure system can be
shown to hyperbolic with eigenvalues

1
=
2
= 0 ,
3
= ( 1)u (72)
and three linearly independent eigenvectors
R
1
=
1
_
_
1
0
0
_
_
, R
2
=
2
_
_
0
1
0
_
_
, R
3
=
3
_
_
0
1
( 1)u
_
_
. (73)
Here
1
,
2
and
3
are scaling factors. Simple calculations show that the
characteristic elds associated with
1
and
2
are linearly degenerate and
the characteristic eld associated with
3
is genuinely non-linear.
Unfortunately we have not been able to nd a straightforward pressure
numerical ux by solving the Riemann problem for this unusual hyperbolic
system. Thus the re-interpretation of the Liou-Steen splitting in our
framework has not been productive.
34 / 46
The Zha-Bilgen splitting
The Zha-Bilgen splitting (1993) assumes a very natural splitting that may
be interpreted in our framework as follows. The advection system is

t
Q+
x
A(Q) = 0 , A(Q) =
_
_
u
u
2
uE
_
_
(74)
and the pressure system is

t
Q+
x
P(Q) = 0 , P(Q) =
_
_
0
p
pu
_
_
. (75)
35 / 46
In terms of primitive variables V = [, u, p]
T
the pressure system can be
shown to be hyperbolic with real eigenvalues

1
= C ,
2
= 0 ,
3
= C , (76)
with
C =
_
( 1)p/ , (77)
and three linearly independent right eigenvectors
R
1
=
1
_
_
0
1
C
_
_
, R
2
=
2
_
_
1
0
0
_
_
, R
3
=
3
_
_
0
1
C
_
_
. (78)
Here
1
,
2
and
3
are scaling factors.
36 / 46
The Riemann problem for the Zha-Bilgen pressure system in terms of
primitive variables is

t
V+Z(V)
x
V = 0
V(x, 0) =
_
_
_
V
L
V
n
i
if x < 0 ,
V
R
V
n
i+1
if x > 0 .
_

_
(79)
The structure of the solution is analogous to that shown in Fig. 1,
u

=

L
C
L
u
L
+
R
C
R
u
R

L
C
L
+
R
C
R

(p
R
p
L
)

L
C
L
+
R
C
R
,
p

=

R
C
R
p
L
+
L
C
L
p
R

L
C
L
+
R
C
R

L
C
L

R
C
R

L
C
L
+
R
C
R
(u
R
u
L
) ,
_

_
(80)
with
C
L
=
_
( 1)p
L
/
L
, C
R
=
_
( 1)p
R
/
R
. (81)
37 / 46
Contact Discontinuity
Proposition 4.1. The Zha-Bilgen splitting along with the Zha-Bilgen
numerical scheme cannot sustain isolated stationary contact discontinuities
for the Euler equations.
Proof. Dene the problem for a stationary, isolated contact discontinuity
for the ideal gas Euler equations with initial condition
u(x, 0) = 0 , p(x, 0) = p : constant , x such that x
L
< x < x
R
,
(x, 0) =
_
_
_

L
if x < x
0
,

R
if x > x
0
,
_

_
(82)
with x
L
< x
0
< x
R
. Assume the discretisation of [x
L
, x
R
] such that the
contact discontinuity is between cells i and i + 1. Application of the
Zha-Bilgen scheme to any cell away from cells i and i + 1 leaves the ow
undisturbed.
38 / 46
However, application of the scheme to cell i for one time step gives
E
n+1
i
=
p
1
+
i
p ,
i
=
1
2
t
x
_
p(
1

R
) . (83)
Application of the scheme to cell i + 1 gives an analogous expression but
with
i+1
=
i
. In order to preserve the contact discontinuity unaltered
one requires
i
= 0, which is not satised by the Zha-Bilgen scheme, as
seen in (83).
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0.7
0.9
1.1
1.3
1.5
ZB-orig
Exact
Fig. 2. Test 6: Stationary isolated contact. Exact (line) and numerical
solution (symbols) using the Zha-Bilgen original scheme (ZB-orig).
39 / 46
Proposition 4.2. The Toro-Vazquez splitting (TV) along with their
numerical method can recognise exactly isolated stationary contact
discontinuities for the Euler equations.
Proof. Dene the problem for a stationary, isolated contact discontinuity
for the ideal gas Euler equations with initial condition as in (82). Assume
the discretization of the domain [x
L
, x
R
] is such that the cell just to the
left of the discontinuity is i and that immediately to the right of the
discontinuity is i + 1. Application of the TV scheme to any cell away from
cells i and i + 1 leaves the ow indisturbed. Let us now apply the scheme
to cell i for one time step. First we need the solution (57) of the Riemann
problem with initial data (82). Clearly u

= u
i+
1
2
= 0 and p

= p
i+
1
2
= p.
Then it is easy to verify that the state Q
n+1
i
= Q
n
i
and thus the isolated
stationary contact is preserved exactly. Application of the scheme to cell
i + 1 gives an analogous result and the proposition is thus proved.
40 / 46
Proposition 4.3. The Zha-Bilgen splitting along with the Godunov-type
numerical method of section 4.2 can recognise exactly isolated stationary
contact discontinuities for the Euler equations.
Proof. The proof is straightforward and is thus omitted.
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0.9
1
1.1
1.2
1.3
1.4
1.5
TV
Exact
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0.7
0.9
1.1
1.3
1.5
ZB-God
Exact
Fig. 3. Test 6: Stationary isolated contact. Exact (line) and numerical
solutions (symbols).
41 / 46
Numerical Results for the Euler Equations
42 / 46
Two classes of test problems
Test
L
u
L
p
L

R
u
R
p
R
1 1.0 0.75 1.0 0.125 0.0 0.1
2 1.0 -2.0 0.4 1.0 2.0 0.4
3 1.0 0.0 1000.0 1.0 0.0 0.01
4 5.99924 19.5975 460.894 5.99242 -6.19633 46.0950
5 1.0 -19.59745 1000.0 1.0 -19.59745 0.01
6 1.4 0.0 1.0 1.0 0.0 1.0
Test problems with exact solution (Toro, 2009)
Test of Woodward and Colella (1984). Reference solution: WAF.
0 x 0.1 0.1 < x 0.9 0.9 < x 1.0

L
= 1.0
M
= 1.0
R
= 1.0
u
L
= 0.0 u
M
= 0.0 u
R
= 0.0
p
R
= 1000.0 p
M
= 0.01 p
R
= 100.0
_

_
(84)
43 / 46
Frame 002 23 Jul 2011
Distance
D
e
n
s
i
t
t
y
0 0.25 0.5 0.75 1
0.1
0.4
0.7
1
TV
Exact
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0
0.2
0.4
0.6
0.8
1
TV-AWS
Exact
Test 1 (sonic ow). Exact (line) and numerical solutions (symbols) using
two numerical schemes (TV and TV-AWS) for the ux splitting of this
paper.
43 / 46
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0.1
0.4
0.7
1
LS
Exact
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0.1
0.4
0.7
1
ZB-orig
Exact
Test 1 (sonic ow). Exact (line) and numerical solutions (symbols) using
two numerical schemes: Liou-Steen (LS) and Zha-Bilgen (ZB-orig).
43 / 46
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0
0.2
0.4
0.6
0.8
1
TV
Exact
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0
0.2
0.4
0.6
0.8
1
Exact
TV-AWS
Test 2 (low density). Exact (line) and numerical solutions (symbols) using
two numerical schemes (TV and TV-AWS) for the ux splitting of this
paper.
43 / 46
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0
1
2
3
4
5
6
TV
Exact
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0
1
2
3
4
5
6
TV-AWS
Exact
Test 3 (very strong shock). Exact (line) and numerical solutions (symbols)
using two numerical schemes (TV and TV-AWS) for the ux splitting of
this paper.
43 / 46
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
5
10
15
20
25
30
35
TV
Exact
Distance
V
2
0 0.25 0.5 0.75 1
5
10
15
20
25
30
35
TV-AWS
Exact
Test 4 (collision of two strong shocks). Exact (line) and numerical
solutions (symbols) using two numerical schemes (TV and TV-AWS) for
the ux splitting of this paper.
43 / 46
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0
1
2
3
4
5
6
7
TV
Exact
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0
1
2
3
4
5
6
7
TV-AWS
Exact
Test 5 (non-isolated stationary contact discontinuity). Exact (line) and
numerical solutions (symbols) using two numerical schemes (TV and
TV-AWS) for the ux splitting of this paper.
43 / 46
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0.9
1
1.1
1.2
1.3
1.4
1.5
TV
Exact
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0.9
1
1.1
1.2
1.3
1.4
1.5
TV-AWS
Exact
Test 6 (Isolated stationary contact discontinuity). Exact (line) and
numerical solutions (symbols) using two numerical schemes (TV and
TV-AWS) for the ux splitting of this paper.
43 / 46
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0.7
0.9
1.1
1.3
1.5
ZB-orig
Exact
Distance
D
e
n
s
i
t
y
0 0.25 0.5 0.75 1
0.7
0.9
1.1
1.3
1.5
ZB-God
Exact
Test 6 (Isolated stationary contact discontinuity). Exact (line) and
numerical solutions (symbols) using two numerical schemes: the
Zha-Bilgen original scheme (ZB-orig) and the Zha-Bilgen splitting with
present Godunov-type numerical approach (ZB-God).
43 / 46
Distance
D
e
n
s
i
t
y
0.5 0.6 0.7 0.8 0.9
0
1
2
3
4
5
6
7
WAF
God+Ex. RS
TV
Distance
D
e
n
s
i
t
y
0.5 0.6 0.7 0.8 0.9
0
1
2
3
4
5
6
7
WAF
God+Ex. RS
TV-AWS
Test 7 (Woodward and Colella blast wave problem). Reference solutions
(WAF and Godunovs method with exact Riemann solver) and numerical
solutions from two numerical schemes of this paper: TV (top) and
TV-AWS (bottom).
44 / 46
Other Potential Schemes for the Pressure System
44 / 46
Lax-Friedrichs
FORCE
Rusanov
HLL
Godunov Centred
45 / 46
Concluding Remarks
New ux splitting
New way of dealing with pressure term
Scheme captures contact discontinuity as well as AUSM
Our scheme is more robust and more accurate than AUSM
Other schemes for pressure system under study
Best combination: explicit for advection implicit for pressure
46 / 46