On Flux Vector Splitting for the Euler Equation
Transcript of On Flux Vector Splitting for the Euler Equation
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
1/58
On Flux Vector Splitting for the Eluer Equations
Eleuterio ToroLaboratory of Applied Mathmatics
University of Trento, Italy
http://www.ing.unitn.it/toro
mailto:[email protected]:[email protected] -
8/14/2019 On Flux Vector Splitting for the Euler Equation
2/58
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 scheme11 Concluding Remarks
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
3/58
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
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
4/58
Test 6Stationary isolated contact. Exact (line) and numerical solutions
(symbols)
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
5/58
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
6/58
1/2
1/2
1/2
1
if 0
if 0
n
i
i
i n
i
i
a
au M
aHA
a
au M
aH
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
7/58
The advection flux is upwind according to advection speed implied in
the Mach number +/, which is split as
with
The pressure vector +/is constructed by splitting the pressure as(16)
with two choices for the negative and positive components as follows
and (17)
1/2 1i i iM M M
211 if 1
4
1 if 1
2
M M
M
M M M
1/2 1i i ip p p
11 if 1
2
1 if 1
2
p M M
pM M
p MM
211 2 if 1
4
1 if 1
2
p M M M
pM M
p MM
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
8/58
The Zha-Bilgen Spliting
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
9/58
Zha and Bilgen (1993) split the flux vector into
with
= , = 0 .
Numerically, they propose fluxes
+/
and
+/as follows.
where
For he pressure flux vector Zha and Bilgen use the
F Q A Q P Q
1/2 1i i iA A A
min 0, , max 0, .n n n ni i i i i iA u Q A u Q
1/2 1i i iP P P
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
10/58
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 1i i ipu pu pu
if 1
1
if -1< 12
0 if 1
n n
i i
n n n n
i i i ii
n
i
u M
pu p u a M
M
0 if 1
1
if -1< 12
if 1
n
i
n n n n
i i i ii
n n
i i
M
pu p u a M
u M
1/2 1 1i i i i iF A P A P
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
11/58
A Flux-Splitting Framework
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
12/58
The Framework
We propose to split system
= 0via
= 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+/= +/ +/,where +/and +/are obtained respectively fromappropriate Cauchy problems for the advection and pressure
systems (27).
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
13/58
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.
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
14/58
Now consider the first Cauchy problem for the advection equation
= 0, ,0 = , (34)the solution of which after a time is , = .Consider the Cauchy problem
= 0, , 0 = ,. (36)The exact solution of IVP (36), after a time , is
, = .
The combined solution of IVPs (34) and (36) for = = is , = = = , .
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
15/58
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 slowadvection 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 leadto 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.
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
16/58
A novel splitting for Euler equations
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
17/58
Here we propose a new splitting for the Euler equations by noting that the
flux may be decomposed thus
=
=
1/2 0 ,with the corresponding advection and pressure flux defined as
=
1/2 , = 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
.
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
18/58
The advection system is = 0,where = , , and A as in (40) above. In quasi-linear formthe advection system becomes
= 0,where M is the Jacobian matrix given as = 0 0 0 2 0
3/2 0.
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
19/58
It is easy to show that the eigenvalues of this matrix are= 0, = = .There are only two linearly independent right eigenvectors, namely
= 100 , =
11/2
.
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 -filed is linearly degenerated and that the -filed is genuinelynon-linear if 0and 0; otherwise it is linearly degenerate.
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
20/58
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 01/2 1 1 1 / 3/2
.
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
21/58
The eigenvalues of are always real and given as= , = 0, = .(50)where = 4, = /.(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 00 0 1 / 0 .(53)
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
22/58
Note that since < = 4, the system is always subsonic,that is= < 0 < = 0 < = .
The right eigenvectors of
in (53) corresponding to the
eigenvalues (50) are
= 02 , =
100 , = 02 .
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
23/58
Numerical Fluxes
In order to compute advection and pressure fluxes
+/and
+/we consider the Riemann problem for each subsystem. Westart 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. Thewave 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 xV x
V V x
(56)
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
24/58
with
whereandare 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 uu p pC C C C
C p C p C C p u u
C C C C
L L R R ; ,L L R RC u A C u A
(57)
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
25/58
The Advection System
Recall that in our splitting (40) the advection operator may be
written thus
=
1/2 = , =
1/2 ,
Advection of (mass, momentum and kinetic energy) with speedu. Here we propose two algorithms.Algorithm 1 (TV scheme)
Where is the intercell advection velocity taken as fromsolution (57) of the Riemann problem (56)
* 1/2 ,iA 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 uA u
K u
*
1/2iu *
1/2iu
(62)
(61)
(63)
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
26/58
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 1i i iA A A
1/ 2 1 , 1/ 2 1 .n n n ni i i i i iA w u K A w u K
2,
nn ii
n
i
uu
u
=0.1 niu
0niu
.nisign u
(64)
(65)
(66)
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
27/58
Summary of the present scheme
In order to compute a numerical flux
+/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
*
1/2ip
1/2iP
1/2 1/2 1/2i i iF A P
*
1/2iu
1/2iA
1/2iF
(67)
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
28/58
Reinterpretation of other flux splitting
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
29/58
The Liou-Steffen scheme
The Liou-Steffen splitting (1993) may be interpreted in our
framework defining the advection system as = 0,with
=
.and the pressure system as = 0,with
= 00 .
(68)
(71)
(69)
(70)
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
30/58
In terms of primitive variables = , , the pressure systemcan be shown to hyperbolic with eigenvalues
= = 0, = 1 . (72)and three linearly independent eigenvectors=
100
, = 010
, = 01
1 .(73)
Here ,and are scaling factors. Simple calculations showthat the characteristic fields associated with and are linearlydegenerate and the characteristic field associated with isgenuinely non-linear.
Unfortunately we have not been able to find a straightforwardpressure 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.
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
31/58
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)
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
32/58
In terms of primitive variables = , , the pressure systemcan be shown to hyperbolic with eigenvalues= , = 0, = , (76)with
C = 1 / , (77)and three linearly independent right eigenvectors=
01
, = 100
, = 01
.(78)
Here ,and are scaling factors.
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
33/58
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 xV x
V V x
(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 pu
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)
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
34/58
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 < < . Assume the discretization of , such thatthe contact discontinuity is between cells and 1. Applicationof the Zha-Bilgen scheme to any cell away from cells and 1leaves 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 xx
x x
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
35/58
However application of the scheme to cell for one time stepgives
Application of the scheme to cell 1gives an analogousexpression but with + = . In order to preserve the contactdiscontinuity unaltered one requires = 0, which is not satisfiedby the Zha-Bilgen scheme, as seen in (83).
1
R R
1 1 1 , .
1 2
n
i i i
p tE p p
x
(83)
Fig. 2. Test 6: Stationary isolated contact. Exact (line) and numerical
solution (symbols) using the Zha-Bilgen original scheme (ZB-orig)
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
36/58
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 contactdiscontinuity 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 thatimmediately to the right of the discontinuity is 1. Applicationof the TV scheme to any cell away from cells and 1leavesthe flow undisturbed. Let us now apply the scheme to cell forone time step. First we need the solution (57) of the Riemann
problem with initial data (82). Clearly
= +/= 0and
= +/= . Then it is easy to verify that the stateand thus the isolated stationary contact is preserved exactly.
Application of the scheme to cell 1 gives analogous result andthe proposition is thus proved.
1n n
i iQ Q
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
37/58
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) .
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
38/58
Numerical Results for the Euler Equations
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
39/58
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 0.016 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
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
40/58
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.
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
41/58
Test 1 (sonic flow). Exact (line) and numerical solution (symbols) using
two numerical schemes: Liou-Steffen (LS) and Zha-Bilgen (ZB-orig).
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
42/58
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.
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
43/58
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.
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
44/58
Test 4 (collision of two strong shocks). Exact (line) and numericalsolution (symbols) using two numerical schemes (TV and TV-AWS) for
the flux splitting of this paper.
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
45/58
Test 5 (non-isolated stationary contact discontinuity). Exact (line) andnumerical solution (symbols) using two numerical schemes (TV and
TV-AWS) for the flux splitting of this paper.
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
46/58
Test 6 (isolated stationary contact discontinuity). Exact (line) andnumerical 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).
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
47/58
Test 7 (Woodward and Colellar blast wave problem). Referencesolutions (WAF and Godunovs method with exact Riemann solver) and
numerical solutions from two numerical schemes of this paper: TV (top)
and TV-AWS (bottom).
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
48/58
Other Potential Schemes forthe Pressure System
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
49/58
Lax-Friedrichs
FORCE
Rusanov
HLL
Godunov Centered
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
50/58
Applications and Extension of the Scheme
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
51/58
General Equation of State
For the Euler equations for general equation of state
= , (85)the flux may be decomposed as
=
=
0
,(86)
with the corresponding advection and pressure flux
=
, =
0
. (87)
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
52/58
The advection system is = 0,(88)This must be solved to :
1) Find advection speed to be used in the advection numerical
oprerator+/2) Find the pressure operator +/
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
53/58
2
1 1
,
... ...
... ...
ll
m m
u
u puuvv
uww
E u E pQ F Qq uq
q uq
q uq
(89)
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
54/58
2
1
0
0
0
1/ 2,
0
... ...
0
... ...
0
l
m
u p
v
w
u e puA Q P Q
q
q
q
(90)
With the corresponding advection and pressure flux defined as
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
55/58
1 1 11 1
2
1 1 1 11 1 1 1
1 1 1 11 1 1 1 2
2 2 2
2 2 22 2
212 2 2 2 2 2 2
1 22 2 22 2 2 2
0 0
0
0
0 0
00
0
0
t x x
u
u pu p
u E pE p u
u
u
pu u p
p uE u E p
(91)
: density, : velocity, : pressure, : total energy,
: volume fraction, with +=1 (93): specific energy, given by an EOS, e.g.= , . (94)
21/ 2k k k k E u e (92)
u
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
56/58
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
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
(95)
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
57/58
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.
-
8/14/2019 On Flux Vector Splitting for the Euler Equation
58/58
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 forpressure terms