On Flux Vector Splitting for the Eluer Equations

Eleuterio Toro
Laboratory of Applied Mathmatics
University of Trento, Italy

toro@ing.unitn.it
http://www.ing.unitn.it/toro
Table of Contents
1 The Euler equation and flux splitting
2 The Liou-Steffen 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 Application and Extension of the scheme
11 Concluding Remarks
Why another flux vector splitting scheme?
Intermediate characteristic fields have the answer.
Classical FVS schemes (Warming-Beam, van Leer, Zha-
Bilgen) are unsuitable . Excessive Numerical diffusion.
Liou-Steffen scheme (AUSM) is an exception but
We present a new framework for construct FVS schemes.
Clear Basis
The Zha-Bilgen splitting can be recovered using the new
frame
We present a new splitting (Toro-Vazquez) for the Euler
Equations

Test 6Stationary isolated contact. Exact (line) and numerical solutions
(symbols)
1/ 2
1/ 2
1/ 2
1
if 0

if 0
n
i
i
i
n
i
i
a
au M
aH
A
a
au M
aH

+
+
+
+

(
>

(

=

(
(
<
(
(

The advection flux is upwind according to advection speed implied in

the Mach number
+1/2
, which is split as


with



The pressure vector
+1/2
is constructed by splitting the pressure as
(16)
with two choices for the negative and positive components as follows

and (17)

1/ 2 1 i i i
M M M
+
+ +
= +
( )
( )
2 1
1 if 1
4
1
if 1
2
M M
M
M M M

>

1/ 2 1 i i i
p p p
+
+ +
= +
( )
( )
1
1 if 1
2
1
if 1
2
p M M
p
M M
p M
M

>

( ) ( )
( )
2 1
1 2 if 1
4
1
if 1
2
p M M M
p
M M
p M
M

>

The Zha-Bilgen Spliting

Zha and Bilgen (1993) split the flux vector into

with
=

, =
0

.
Numerically, they propose fluxes
+1/2
and
+1/2
as follows.


where


For he pressure flux vector Zha and Bilgen use the
( ) ( ) ( )
F Q A Q P Q = +
1/ 2 1 i i i
A A A
+
+ +
= +
( ) ( )
min 0, , max 0, .
n n n n
i i i i i i
A u Q A u Q
+
= =
1/ 2 1 i i i
P P P
+
+ +
= +
For the p component Zha and Bilgen adopt the Liou-Steffen Splitting
(16), (17), while for the pu compenent they propose


where


,


Finally the Zha-Bilgen numerical flux is
( ) ( ) ( )
1/ 2 1 i i i
pu pu pu
+
+ +
= +
( ) ( )
if 1
1
if -1< 1
2
0 if 1
n n
i i
n n n n
i i i i
i
n
i
u M
pu p u a M
M

= <

>

( ) ( )
0 if 1
1
if -1< 1
2
if 1
n
i
n n n n
i i i i
i
n n
i i
M
pu p u a M
u M
+
s

= + <

>

1/ 2 1 1 i i i i i
F A P A P
+ +
+ + +
= + + +
A Flux-Splitting Framework
The Framework
We propose to split system

= 0 via =
+ into two systems

= 0 ,

= 0 ,
, (27)

called respectively the advection system and the pressure system.
The aim is then to compute the numerical flux as

+1/2
=
+1/2
+
+1/2
,

where
+1/2
and
+1/2
are obtained respectively from
appropriate Cauchy problems for the advection and pressure
systems (27).
Consider the Cauchy problem for the linear advection equation

, +

, = 0, < < , > 0,

, =
(29)
where is a constant. The exact solution of IVP (29) after a time is
, = .
We now decompose the characteristic speed as
= + 1 =

, 0 1, (31)
with definitions

= ,

= 1 , (32)
so as to obtain two linear partial differential equations, namely

= 0,

= 0.
Now consider the first Cauchy problem for the advection equation

= 0,
, 0 = ,
(34)
the solution of which after a time
1
is
,
1
=

1
.
Consider the Cauchy problem

= 0,
, 0 =

1
,
. (36)
The exact solution of IVP (36), after a time
2
, is
,
2
=

2
.
The combined solution of IVPs (34) and (36) for
1
=
2
= is
, =

= = , .
The above result can be started 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 different. For example, for > 0, by taking a very small
in (31) we would have

, 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 flux for IVP (29) by
computing numerical fluxes for IVPs (34) and (36). This would lead
to split flux 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.
A novel splitting for Euler equations
Here we propose a new splitting for the Euler equations by noting that the
flux may be decomposed thus
=

1
2
2
+ +
=

2
1/2
3
+
0

+
,
with the corresponding advection and pressure flux defined as
=

2
1/2
3
, =
0

+
, (40)
We note that the proposed advection flux A contains no pressure terms.
All pressure terms from the flux , including that of the total energy ,
are now included in the pressure . The advection flux may be
interpreted as representing advection of mass, momentum and kinetic energy.
For the ideal gas case (5) the pressure flux (40) becomes
=
0

/ 1
.
The advection system is

= 0,
where = , ,

and A as in (40) above. In quasi-linear form
the advection system becomes

= 0,
where M is the Jacobian matrix given as
=
0 0 0

2
2 0

3
3/2
2
0
.

It is easy to show that the eigenvalues of this matrix are

1
= 0,
2
=
3
= .
There are only two linearly independent right eigenvectors, namely

1
=
1
1
0
0
,
2
=
2
1

1/2
2
.
Thus the system is weakly hyperbolic, as there is no complete set of
linearly independent eigenvectors.
Regarding the nature of the characteristic fields, it is easy to show that
the
1
-filed is linearly degenerated and that the
2
-filed is genuinely
non-linear if
2
0 and 0; otherwise it is linearly degenerate.
The Pressure System
In terms of the conserved variables = , ,

the pressure
system is

= 0,
with as given in (40) above. In quasi-linear form the
pressure system becomes

= 0,
where is the Jacobian matrix given as
=
0 0 0
1/2 1
2
1 1

3
/

3/2
2

.
The eigenvalues of are always real and given as

1
=
1
2

1
2
,
2
= 0,
3
=
1
2
+
1
2
.(50)
where =
2
+4
2
,
2
= / .(51)
Here is the usual speed of sound for the full Euler equations. In
terms of physical variables the system reads

= 0,
where
=

, =
0 0 0
0 0 1/
0
.(53)
Note that since < =
2
+4
2
, the system is always subsonic,
that is

1
=
1
2

1
2
< 0 <
2
= 0 <
3
=
1
2
+
1
2
.
The right eigenvectors of in (53) corresponding to the
eigenvalues (50) are

1
=
0
2

,
2
=
1
0
0
,
3
=
0
2
+
.
Numerical Fluxes
In order to compute advection and pressure fluxes
+1/2
and

+1/2
we consider the Riemann problem for each subsystem. We
start with the pressure system.
To compute the flux for the pressure system we consider the
Riemann problem in terms of physical variables




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

<
`

=


>

)
(56)
with


where

and

are computed from (51).

Fig. 1. Structure of the solution of the Riemann problem for the pressure system
( )
( )
R R L L
* R L
R L R L
R L L R R L
* R L
R L R L
2
,
1
.
2
C u C u
u p p
C C C C
C p C p C C
p u u
C C C C

=



`

= +

)
( ) ( )
L L R R
; ,
L L R R
C u A C u A = = +
(57)
The Advection System
Recall that in our splitting (40) the advection operator may be
written thus
=

2
1/2
3
= , =

1/2
2
,
Advection of (mass, momentum and kinetic energy) with speed
u. Here we propose two algorithms.
Algorithm 1 (TV scheme)

Where is the intercell advection velocity taken as from
solution (57) of the Riemann problem (56)
( )
*
1/ 2
,
i
A Q u K
+
=
*
1/ 2 *
1/ 2 1/ 2
*
1 1/ 2
if 0 ,
if 0 .
n
i i
i i
n
i i
K u
A u
K u
+
+ +
+ +
>

=

s

*
1/ 2 i
u
+
*
1/ 2 i
u
+
(62)
(61)
(63)
Algorithm 2 (TV-AWS scheme). Here propose a weighted splitting
scheme, which is a simple modification of the scheme proposed by
Zha and Bilgen (1963) for their advection system. The modified
scheme is given as follows

with

Here


with a small positive quantity, , for example. The function
allows a smooth transition from upwinding fully to the left and fully to
the right, in the vicinity of .
The resulting scheme from Algorithm 2, called the TV-AWS scheme,
is effectively a weighted averaged scheme and the Zha-Bilgen scheme
is recovered from it by simply setting the weight to be
1/ 2 1 i i i
A A A
+
+ +
= +
( ) ( )
1/ 2 1 , 1/ 2 1 .
n n n n
i i i i i i
A w u K A w u K
+
= = +
( )
( )
2
,
n
n
i
i
n
i
u
u
u
e e
c
= =
+
=0.1 c c ( )
n
i
u e
0
n
i
u =
( )
.
n
i
sign u e =
(64)
(65)
(66)
Summary of the present scheme
In order to compute a numerical flux
+1/2
for the conservative
scheme (6) we proceed as follows:
Present flux. Evaluate the intercell pressure and velocity
from the solution of the Riemann problem given in (57) to
compute the present flux as in (60).
Advection. We have proposed two options. From algorithm 1
(TV scheme) we evaluate the advection flux as in (63).
Algorithm 2 (TV-AWS) is described in equations (64) to (66).
Intercell flux. Compute the intercell flux as in (28), namely

*
1/ 2 i
p
+
1/ 2 i
P
+
1/ 2 1/ 2 1/ 2 i i i
F A P
+ + +
= +
*
1/ 2 i
u
+
1/ 2 i
A
+
1/ 2 i
F
+
(67)
Reinterpretation of other flux splitting
The Liou-Steffen scheme
The Liou-Steffen splitting (1993) may be interpreted in our
framework defining the advection system as

= 0,
with
=

2
+
.
and the pressure system as

= 0,
with
=
0

0
.
(68)
(71)
(69)
(70)
In terms of primitive variables = , ,

the pressure system
can be shown to hyperbolic with eigenvalues

1
=
2
= 0,
3
= 1 . (72)
and three linearly independent eigenvectors

1
=
1
1
0
0
,
2
=
2
0
1
0
,
3
=
3
0
1
1
. (73)
Here
1
,
2
and
3
are scaling factors. Simple calculations show
that the characteristic fields associated with
1
and
2
are linearly
degenerate and the characteristic field associated with
3
is
genuinely non-linear.
Unfortunately we have not been able to find a straightforward
pressure numerical flux by solving the Riemann problem for this
unusual hyperbolic system . Thus the re-interpretation of the
Liou-Steffen splitting in our framework has not been productive.
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

= 0,
with
=

.
and the pressure system as

= 0,
with
=
0

.
(75)
(74)
In terms of primitive variables = , ,

the pressure system
can be shown to hyperbolic with eigenvalues

1
= ,
2
= 0,
3
= , (76)
with
C = 1 / , (77)
and three linearly independent right eigenvectors

1
=
1
0
1

,
2
=
2
1
0
0
,
3
=
3
0
1

. (78)
Here
1
,
2
and
3
are scaling factors.
The Riemann problem for the Zha-Bilgen pressure system in terms
of primitive variables is





The structure of the solution is analogous to that shown in Fig. 1,
( )
( )
L
R 1
0,
if 0 ,
, 0
if 0 .
t x
n
i
n
i
V Z V V
V V x
V x
V V x
+
c + c =

<
`

=


>

)
(79)
with


( )
L L L R R R R L
*
L L R R L L R R
R R L L L R L L R R
* R L
L L R R L L R R
,
.
C u C u p p
u
C C C C
C p C p C C
p u u
C C C C




+

=

+ +

`
+

=

+ +
)
( ) ( )
L L L R R R
1 / ; 1 / . C p C p = =
(80)
(81)
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. Define the problem for a stationary, isolated contact
discontinuity for the ideal gas Euler equations with initial
condition




with

<
0
<

. Assume the discretization of

such that
the contact discontinuity is between cells and +1. Application
of the Zha-Bilgen scheme to any cell away from cells and +1
leaves the flow undisturbed.
(82)
( ) ( )
( )
L R
L 0
R 0
, 0 0, , 0 : constant, such that ,
if ,
, 0
if ,
u x p x p x x x x
x x
x
x x

= = < <

<
`
=


>

)
However application of the scheme to cell for one time step
gives



Application of the scheme to cell +1 gives an analogous
expression but with
+1
=

. In order to preserve the contact

discontinuity unaltered one requires

= 0, which is not satisfied

by the Zha-Bilgen scheme, as seen in (83).
1
R R
1 1 1
, .
1 2
n
i i i
p t
E p p
x
o o

+
| |
A
= + =
|
|
A
\ .
(83)
Fig. 2. Test 6: Stationary isolated contact. Exact (line) and numerical
solution (symbols) using the Zha-Bilgen original scheme (ZB-orig)
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. Define the problem for a stationary, isolated contact
discontinuity for the ideal gas Euler equations with initial
condition as (82). Assume the discretization of the domain

such that the cell just to the left of the discontinuity is and that
immediately to the right of the discontinuity is +1. Application
of the TV scheme to any cell away from cells and +1 leaves
the flow undisturbed. Let us now apply the scheme to cell for
one time step. First we need the solution (57) of the Riemann
problem with initial data (82). Clearly

=
+1/2
= 0 and

=
+1/2
= . Then it is easy to verify that the state
and thus the isolated stationary contact is preserved exactly.
Application of the scheme to cell +1 gives analogous result and
the proposition is thus proved.
1 n n
i i
Q Q
+
=
Proposition 4.3. The Zha-Bilgen splitting along with the
Godunov-type numerical method of section 4.2 can recognize
exactly isolated stationary contact discontinuities for the Euler
equations.
Proof. The proof is straightforward and is thus omitted.
Fig. 3. Test 6: Stationary isolated contact. Exact (line) and numerical
solution (symbols) .
Numerical Results for the Euler Equations
Two classes of test problems
Test of Woodward and Colella (1984). Reference solution:WAF.
(84)
Test

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
L M R
L M R
L M R
0 0.1 0.1 0.9 0.9 1.0
1.0 1.0 1.0
0.0 0.0 0.0
1000.0 0.01 100.0
x x x
u u u
p p p

s s < s < s

= = =

`
= = =

= = =
)
Test 1 (sonic flow). Exact (line) and numerical solution (symbols) using
two numerical schemes (TV and TV-AWS) for the flux splitting of this
paper.
Test 1 (sonic flow). Exact (line) and numerical solution (symbols) using
two numerical schemes: Liou-Steffen (LS) and Zha-Bilgen (ZB-orig).
Test 2 (low density). Exact (line) and numerical solution (symbols)
using two numerical schemes (TV and TV-AWS) for the flux splitting of
this paper.
Test 3 (very strong shock). Exact (line) and numerical solution (symbols)
using two numerical schemes (TV and TV-AWS) for the flux splitting of
this paper.
Test 4 (collision of two strong shocks). Exact (line) and numerical
solution (symbols) using two numerical schemes (TV and TV-AWS) for
the flux splitting of this paper.
Test 5 (non-isolated stationary contact discontinuity). Exact (line) and
numerical solution (symbols) using two numerical schemes (TV and
TV-AWS) for the flux splitting of this paper.
Test 6 (isolated stationary contact discontinuity). Exact (line) and
numerical solution (symbols) using two numerical schemes: the Zha-
Bilgen original scheme (ZB-orig) and Zha-Bilgen splitting with present
Godunov-type numerical approach (ZB-God).
Test 7 (Woodward and Colellar 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).
Other Potential Schemes for
the Pressure System
Lax-Friedrichs
FORCE
Rusanov
HLL
Godunov Centered
Applications and Extension of the Scheme
General Equation of State
For the Euler equations for general equation of state
= , (85)
the flux may be decomposed as
=

2
+

1
2
2
+ +
=

2
1
2

3
+
0

+
,(86)
with the corresponding advection and pressure flux
=

2
1
2

3
, =
0

+
. (87)
The advection system is

= 0,(88)
This must be solved to :
1) Find advection speed to be used in the advection numerical
oprerator
+1/2

2) Find the pressure operator
+1/2


( )
( )
2
1
1
,
...
...
...
...
l
l
m
m
u
u p u
uv v
uw w
E u E p
Q F Q
q
uq
q
uq
q
uq

(
(
(
(
+
(
(
(
(
(
(
(
(
(
(
+
(
= = (
(
(
(
(
(
(
(
(
(
(
(
(
(
(


(89)
( ) ( )
( )
2
1
0
0
0
1/ 2
,
0
...
...
0
... ...
0
l
m
u p
v
w
u e p u
A Q P Q
q
q
q

( (
( (
( (
( (
( (
( (
( (
+
( ( = =
( (
( (
( (
( (
( (
( (
( (

(90)
With the corresponding advection and pressure flux defined as
( )
( )
( )
( )
1 1 1
1 1
2
1 1 1 1
1 1 1 1
1 1 1 1
1 1 1 1 2
2 2 2
2 2 2 2 2
2
1 2 2 2
2 2 2 2
1 2 2 2 2
2 2 2 2
0 0
0
0
0 0
0 0
0
0
t x x
u
u p
u p
u E p
E p u
u
u
p u
u p
p u E
u E p
o
o
o
o
o
o
o o
o o
o
o
o
o
(
( ( (
(
( ( +
(
( (
(
( + (
(
( (
(
c + c + c =
( (
(
( (
(
( (
(
( ( +
(
( (
(

+

(
(
(
(
(
(
(
(
(
(91)

: density,

: velocity,

: pressure,

: total energy,

: volume fraction, with

1
+
2
=1 (93)

: specific energy, given by an EOS, e.g.

. (94)
( )
2
1/ 2
k k k k
E u e = +
(92)
( )
( )
( )
( )
1 1 1
1 1 1 1
2
1 1 1 1
2 2 2
2 2 2 2
2
2 2 2 2
1
2
0
1
2
u
u u
u u
A
u
u u
u u
o
o
o
o
o
o
(
(
(
(
| |
(
|
(
\ .
(
=
(
(
(
(
(
| |
(
|
(
\ .

( ) ( ) ( )
( )
( ) ( ) ( )
1 1
1 1 1 1 1 1 1 1 1 2 1
2 2
2 1 2 2 2
2 2 2 2 2 2 1 2 2 2 2
0
0
x
x x x
x
x x
x x x
p
u e u p p u u
P u
p p p
u e u p p u p
o
o o o
o o
o o
o o o
(
(
c
(
(
c + c + c
(
= c (
(
(
(
c + c
(
c + c + c (

(95)
Explicit/Implicit Version
Could separate slow advetion from fast pressure waves
Treat slow advection explicitly
Treat fast pressure waves implicitly
Could capture interfaces with explicit sheme at CFL unity.
Potential improvements.
New flus 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
Potential applications:
general equation of state
multicomponent flow
two-phase flow
Extension of the scheme: explicit for advection and implicit for
pressure terms