A Comparison of Numerical Flux Formulas for the

Euler Equations
- Math 671 final assignment -
H. Nishikawa∗
December 1998

Abstract
In this report, eight basic numerical flux functions for 1D Euler equations are compared on Sod’s
shock tube problems. Some fundamental properties of the schemes will be discussed. Also a solution
obtained by a second-order method will be presented.

1 Introduction
In this study, we consider the 1D Euler equations for a perfect gas.
∂u ∂f
+ =0 (1)
∂t ∂x
where
T [ ]T p 1
u = [ρ, ρu, ρE] , f = ρu, ρu2+p, ρuH , ρE = ρH − p = + ρu2 , γ = 1.4. (2)
γ−1 2
We write a numerical scheme in the conservative form.
∆t ( )
Un+1
j = Unj − Fj+ 12 − Fj− 12 (3)
∆x
where Unj is a numerical approximation to the cell-average of the exact solution over the j th cell at t = tn ,
and Fj+ 12 is a numerical flux approximating the time-average of the exact flux over ∆t evaluated at xj+ 12 .
Various formulas for the numerical flux have been developed up to the present. We shall consider
eight of those in this report. Although these schemes are derived from different interpretations of the
numerical approximation; computing cell-averages(finite-volume method) or point-values(finite difference
method), it is not an important issue here since we consider only 1st/2nd order method.

2 Numerical Fluxes
In this section, we will give the formulas for the numerical flux, and describe some well-known properties
for each.

∗ Doctoral Candidate, Aerospace Engineering and Scientific Computing

1
CFD Notes by Hiroaki Nishikawa www.cfdnotes.com

2.1 Lax-Friedrichs scheme

Lax-Friedrichs’ scheme, written in the conservation form, is defined by the following flux function.
1[ ] ∆x [ n ]
Fj+ 12 = f (Unj ) + f (Unj+1 ) − Uj+1 − Unj . (4)
2 2∆t
It is well-known that the scheme is very dissipative. In the linear case, it can be shown that the
amplitude error gets larger as the CFL number goes down.

2.2 Richtmyer’s scheme

Richtmyer’s scheme is a two-step Lax-Wendroff type scheme defined by

n+ 1 1[ n ] ∆t [ ]
Uj+ 12 = Uj + Unj+1 − f (Unj+1 ) − f (Unj ) (5)
2 2 2∆x
∆t [ ]
n+ 12 n+ 12
Un+1
j = U n
j − f (U 1
j+ 2
) − f (U 1
j− 2
) . (6)
∆x
This is in the conservative form, and therefore we identify Richtmyer’s flux function as
n+ 1
Fj+ 12 = f (Uj+ 12 ) (7)
2

with (5).
The first step is the Lax-Friedrichs scheme, and the second is a leapfrog scheme. This scheme becomes
identical to the Lax-wendroff scheme in the linear case. The Lax-wendroff scheme is known as dispersive,
i.e. high frequency components lag behind. And due to the small dissipation they remain to be high
frequency. A linear analysis tells us that this behavior becomes more conspicuous as the CFL number
goes down.

2.3 MacCormack’s scheme

MacCormack’s scheme is also a two-step Lax-Wendroff type scheme defined by
∆t [ ]
U∗j = Unj − f (Unj+1 ) − f (Unj ) (8)
∆x
1[ n ] ∆t [ ]
Un+1
j = Uj + U∗j − f (U∗j ) − f (U∗j−1 ) . (9)
2 2∆x
MacCormack’s scheme is known as conservative. Indeed it can be written in the conservative form by
substituting (8) into (9) and defining the numerical flux as
1[ ]
Fj+ 21 = f (Unj+1 ) + f (U∗j ) (10)
2
with (8).
This also becomes identical to the Lax-wendroff scheme in the linear case. The first step is a first
order forward difference, and the second is a backward difference. Therefore, either step will be unstable
for one characteristic speed, but the overall scheme is stable and of second-order in the linear case. In
nonlinear cases however, it is well-known to suffer a nonlinear instability due to the lack of dissipation.
Another important property of the scheme is that it fails at the sonic expansion, as Roe’s method does,
due to the upwind nature.

⃝1998
c by Hiroaki Nishikawa 2
CFD Notes by Hiroaki Nishikawa www.cfdnotes.com

2.4 Steger-Warming scheme (the beam scheme)

We apply the upwind scheme for a positive and a negative part of the flux separately.

Fj+ 12 = f + (Unj ) + f − (Unj+1 ). (11)

In Steger-Warming splitting, we project the flux onto the space of eigenvectors using the homogeneity
property, and split the flux. This results
     
1 1 1
ρ  ρ (γ − 1)  ρ 
f ± (U) = λ(1)± u − c  + λ(2)± u  + λ(3)± u+c  (12)
2γ γ 2γ
H − uc 1 2
2 u H + uc
where
λ(k) ± |λ(k) |
λ(k)± = , k = 1, 2, 3. (13)
2
and λ(k) are defined by
λ(1) = u − c, λ(2) = u, λ(3) = u + c. (14)
The vectors are, as mentioned, associated eigenvectors, r(1) , r(2) , and r(3) respectively. This is however
not a unique splitting of the flux. It is a disadvantage of this splitting that it is not differentiable at a
sonic and stagnation points. Later Van Leer designed a splitting that is differentiable.

2.5 Van Leer’s scheme

Applying upwind differencing, we again write the numerical flux as

Fj+ 12 = f + (Unj ) + f − (Unj+1 ) (15)

where Van Leer’s flux vector splitting reads

   
( 1 ) ( 1 )
ρc  2c 1 + γ−1 M  ρc 2 
γ −1 + 2 M
− 2c γ−1
f + (U) = (M + 1)2   , f (U) = − (M − 1)   (16)
2 ( ) 2 ( )
γ 2
4 2 4 2
γ 2 −1 1 − 2 M
2c γ−1 2c γ−1
γ 2 −1 1 + 2 M

where M is the Mach number. This is differentiable at a sonic and a stagnation points.

2.6 Godunov’s scheme

Godunov’s scheme is the one that solves the Riemann problem at each interface exactly. The details of
the algorithm can be found in [1]. Note that this is a first-order method because of the assumption of
piecewise constant data.

2.7 Osher’s scheme

Osher’s scheme is an approximate Riemann solver. The derivation is so involed (the detailed derivation
is attached) that it will not be described here. It is written as
∫
1[ ] 1 j+1
Fj+ 21 = f (Unj ) + f (Unj+1 ) − |A(U)| dU (17)
2 2 j

∂f
where A is the Jacobian matrix ∂u . The integral is to be broken into three parts,
∫ j+1 ∫ Lm ∫ Rm ∫ j+1
|A(U)| dU = |A(U)| dU + |A(U)| dU + |A(U)| dU (18)
j j Lm Rm

⃝1998
c by Hiroaki Nishikawa 3
CFD Notes by Hiroaki Nishikawa www.cfdnotes.com

where the integrals are carried out along right eigenvectors, r(3) , r(2) , and r(1) respectively. After some
analysis, we arrive at the following practical(ready-to-use) formula.
1[ ] 1[ ]
Fj+ 12 = f (Unj ) + f (Unj+1 ) − F + F3-sonic + F1-sonic (19)
2 2
where
(3) (1)
F = sgn(λj ) f (Uj ) + sgn(λj+1 ) f (Uj+1 ) (20)
[ ]
(3) (2)
+ sgn(λj ) − sgn(λLm ) f (ULm ) (21)
[ ]
(2) (1)
+ sgn(λLm ) − sgn(λj+1 ) f (URm ) (22)

and
{
2sgn(λj ) [f (U∗ ) − f (ULm )] λj · λLm < 0
(3) (3) (3)
F3-sonic = (23)
0 otherwise
{
2sgn(λj+1 ) [f (URm ) − f (U∗ )] λj+1 · λRm < 0
(1) (1) (1)
F1-sonic = (24)
0 otherwise

The intermediate states, Lm and Rm, can be found easily. And U∗ is the state of the sonic point at
which λ(k) = 0, k = 1 or 3, which can also be determined easily.
Clearly, the approximate Riemann solution does not contain discontinuous transition(integrations
along r(k) , not the Hugoniot curve.). According to Van Leer’s analysis, the integration can be interpreted
as along an overturned shock in phase space whenever there exists a discontinuity in reality. The resulting
solution however will be sharp with at most two interior cells.

2.8 Roe’s scheme

Roe’s scheme is an approximate Riemann solver based on a linearization. The scheme reads

1[ ] 1 ∑ (k) 3
Fj+ 12 = f (Unj ) + f (Unj+1 ) − |λ̂ |∆V̂ (k) r̂(k) (25)
2 2
k=1

where ∆V (k) is the discretized kth component of the differential of the characteristic vector,
[ ]
∆p − ρ̂ĉ∆u ∆p − ĉ2 ∆ρ ∆p + ρ̂ĉ∆u
∆V̂ (k) = , − , (26)
2ĉ2 ĉ2 2ĉ2

where ∆p = pR − pL = pj+1 − pj , and similarly for others. And r̂(k) are right eigenvectors
[ ]T [ ]T [ ]T
û2
r̂ (1)
= 1, û − ĉ, Ĥ − ûĉ , r̂ = 1, û,
(2)
, r̂(3) = 1, û + ĉ, Ĥ + ûĉ . (27)
2
with the eigenvalues
λ̂(1) = û − ĉ, λ̂(2) = û, λ̂(3) = û + ĉ. (28)
All the quantities with a hat are evaluated by the Roe-averages,
( )
√ uL + R uR HL + R HR û2
ρ̂ = ρL ρR = R ρL , û = , Ĥ = , ĉ2 = (γ − 1) Ĥ − . (29)
1+R 1+R 2
In Roe’s method, all the waves are replaced by linear waves, and therefore no rarefaction waves(nonlinear
waves) are contained in the scheme. This results in the failure of detecting the sonic point, and it needs
to be modified in such a case.

⃝1998
c by Hiroaki Nishikawa 4
CFD Notes by Hiroaki Nishikawa www.cfdnotes.com

3 Results and Discussions

We consider the following shock tube problem(no sonic rarefaction).
{
ρL = 1.0, pL = 1.0, uL = 0.0 x≤0
(30)
ρR = 0.125, pR = 0.1, uR = 0.0 x > 0

We define the CFL number by

max cells (|uj | + cj )∆t
CFL = . (31)
∆x
We will use a uniform grid with 80 cells, and use the CFL number as a parameter (0 <CFL≤ 1). Therefore
∆t will be computed, from the above relation, at each time step to keep the CFL number as specified.
And we will stop the calculation when time exceeds 1.7. As a boundary condition, we place two ghost
cells at each end and store the same solutions as the adjacent cells’. We computed the solutions for two
CFL numbers, CFL=0.95 and 0.1, to see how the properties of the schemes in the linear case are carried
over to the nonlinear case. Finally the solid lines in all the figures that follow indicate the exact solutions.

Lax-Friedrichs (Figures 1 and 2)

We see clearly the effect of its large dissipation. And as we expected from the linear analysis, it
gets worse as we reduce the CFL number. Also the staircase-like solution, which is particular to the
Lax-Friedrichs scheme, is observed.

Two-step LW-type schemes(Richtmyer, MacCormack) (Figures 3-6)

In both cases, the solution is oscillatory due to the dispersive property of second-order schemes
although the solutions are more accurate than those obtained by first-order methods in the smooth
regions. As we expected, the oscillations grow (in fact remain due to the lack of dissipation) as the
CFL number goes down. In particular, MacCormack’s scheme breaks down for CFL < 0.6518. This is
often called nonlinear instability since the scheme is theoretically stable for CFL ≤ 1 in the linear case.
On the other hand, Richtmyer does not break down due to the large dissipation introduced in its first
step(Lax-Friedrichs), but still exhibits a highly oscillatory behavior. It is nevertheless well-known that
the oscillations can be surpressed by introducing some artificial viscosity at the cost of the introduction
of a problem-dependent tuning parameter[1].

Upwind schemes(Steger-Warming, Van Leer, Godunov, Osher, Roe) (Figures 7-16)

In the linear case, we know the upwind scheme works quite well, i.e. less dissipative than Lax-
Friedrichs’ scheme, and low dispersion error. It is clearly seen from the results that these properties
remain in the nonlinear case.

From the above observations, we see that the properties of the schemes well-known in the linear case
are indeed carried over to the nonlinear case. We then could conclude that we obtain the best results by
the upwind schemes. Next we compare the upwind schemes.

Steger-Warming, Van Leer, Godunov, Osher, Roe (Figures 7, 9, 11, 13, and 15)
As can be seen in figure 7 and 9, Van Leer’s scheme improves the solution at the contact discontinuity
and at both ends of the expansion fan, compared with those by Steger-Warming’s scheme in which the
solution is strongly smeared out at these regions. For Godunov-type methods, Godunov, Osher, and
Roe, the solutions are almost indistinguishable. This leads us to conclude that Roe’s method be the
practical choice since it requires the least computational work among the three. Now, we compare Van
Leer and Roe, Figures 7 and 15. There is no significant differences except for a minor difference in the
expansion fan. It is hard to tell at this point which is superior in a general sense. Finally note that the
contact discontinuity is smeared significantly in all the schemes, which can be seen clearly in the entropy
plot. On the other hand, the shock has been resolved quite sharply by most of the schemes. This is

⃝1998
c by Hiroaki Nishikawa 5
CFD Notes by Hiroaki Nishikawa www.cfdnotes.com

a typical result for a nonlinear discontinuity because of its self-sharpening nature (combination of the
characteristics merging into the shock and the upwind method) while there is no such mechanism in the
linearly degenerate field (combination of the characteristics running parallel to the discontinuity and the
upwind method, i.e. upwind in the same direction across the discontinuity).
In the sense of extending the first-order method to the second-order, it would be reasonable to choose
Roe’s method because of its linearization property. A second-order method developed for a linear system
can be applied directly to the Euler equations linearized by the method of Roe.

MUSCL scheme with Roe-linearization (Figures 17)

There are various ways to extend the first- order method to the second-order accuracy. In this study,
the method described in [2] was used. We first construct a set of piecewise linear data within each cell
with the minmod limiter, then solve the linearized characteristic equations (linear advection) exactly, and
finally average the solutions. All these steps being put together, the flux can be written, on a uniform
grid,
1 ∑ (p) ( )
3
(p)
Fj+ 12 = Fupwind + λ̂ sgn(ν̂ (p) ) − ν̂ (p) σip r̂(p) (32)
2 p=1

where Fupwind is the upwind flux given by (25), ν̂ (p) = λ̂(p) dt/dx, and

(p) (p) (p)

σj = minmod(dV̂j+ 1 , dV̂j− 1 ) (33)
2 2

{
j if λ̂(p) > 0
ip = (34)
j+1 if λ̂(p) < 0
Note that all the quantities with a hat are evaluated by the Roe-averages between jth and (j + 1)th cells
(p) (p)
except for dV̂j− 1 and dV̂j+ 3 which must be evaluated by the Roe-averages between their corresponding
2 2
two neighboring cells.
As shown in figures 17 and 18, a noticeable improvement can be seen especially at the contact
discontinuity. The oscillations typical to second-order schemes have been successfully removed by the
minmod limiter.

Sonic rarefaction (Figures 19-24)

Finally, we compare some of the schemes for the case of sonic rarefaction. The initial values are taken
from [3], {
ρL = 3.857, pL = 10.333, uL = 0.92 x ≤ 0
(35)
ρR = 1.0, pR = 1.0, uR = 3.55 x>0
First of all, MacCormack’s and Roe’s schemes keep the nonphysical discontinuity as expected. Also Van
Leer’s scheme breaks down for any value of the CFL number. As shown in Figure 19., Lax-Friedrichs
works, but not quite well. It is too dissipative and also creates a large peak of entropy behind the contact
discontinuity. Richtmyer(Figure 20) gives a good solution except the oscillation in the entropy behind the
contact discontinuity. On the other hand, Steger-Warming scheme produces completely wrong solutions
as can be seen in Figure 21. As in the previous test problem, Godunov and Osher give almost identical
solutions. And they have a glitch at the soonic point as usually observed, and also create an entropy
peak behind the contact discontinuity although small compared with others. Finally, Roe’s method was
applied with an entropy fix, modifying the wave speed by a smooth quadratic function near the sonic
point as described in [1]. It is clearly seen in Figure 24 that the solution is much smoother at the sonic
point than other schemes. However, it creates a quite large entropy peak behind the contact discontinuity.

⃝1998
c by Hiroaki Nishikawa 6
CFD Notes by Hiroaki Nishikawa www.cfdnotes.com

4 Concluding Remarks
The eight basic schemes for the Euler equations were compared for two shock tube problems. From this
study, the following conclusions may be drawn. First, the upwind schemes would be the best choice
for the Euler equations. Second, the flux-vector splitting methods are not as robust as Godunov-type
methods as shown in the case of the sonic rarefaction. But Godunov-type methods can produce a spurious
solution in some cases.

References
[1] Hirsch, C., Numerical Computation of Internal and External Flows, Volume 2, John Wiley & Sons,
1990.
[2] LeVeque, R. J., Numerical Methods for Conservation Laws, Birkhäuser, 1992.

[3] Huynh, H. T., Accurate Upwind Methods for the Euler Equations, SIAM J. Num. Anal., Vol. 32,
No. 5, pp. 1565-1619, October 1995.

⃝1998
c by Hiroaki Nishikawa 7
 CFD Notes by Hiroaki Nishikawa www.cfdnotes.com

Density Velocity Density Velocity

1 1 1 1
Lax−Friedrichs Lax−Friedrichs
CFL=0.95 CFL=0.1
0.8 30 time Steps 0.8 0.8 221 time Steps 0.8

Velocity u

Velocity u
Density ρ

Density ρ
0.6 0.6 0.6 0.6

0.4 0.4 0.4 0.4

0.2 0.2 0.2 0.2

0 0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Pressure Entroby/R Pressure Entroby/R

gas gas
1 2 1 2

0.8 0.8
1.5 1.5
Entropy/Rgas

Entropy/Rgas
Pressure P

Pressure P
0.6 0.6
1 1
0.4 0.4

0.5 0.5
0.2 0.2

0 0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Figure 1: The solutions by Lax-Friedrichs’ Figure 2: The solutions by Lax-Friedrichs’

scheme with CFL= 0.95, t = 1.74833. scheme with CFL= 0.1, t = 1.70464.

Density Velocity Density Velocity

1 1.4 1 1.4
Richtmyer Richtmyer
CFL=0.95 1.2 CFL=0.1 1.2
0.8 32 time Steps 0.8 333 time Steps
1 1
Velocity u

Velocity u
Density ρ

Density ρ

0.6 0.8 0.6 0.8

0.4 0.6 0.4 0.6

0.4 0.4
0.2 0.2
0.2 0.2

0 0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Pressure Entroby/Rgas Pressure Entroby/Rgas

1 2 1 2

0.8 1.5 0.8 1.5

Entropy/Rgas

Entropy/Rgas
Pressure P

Pressure P

0.6 1 0.6 1

0.4 0.5 0.4 0.5

0.2 0 0.2 0

0 −0.5 0 −0.5
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Figure 3: The solutions by Richtmyer’s Figure 4: The solutions by Richtmyer’s

scheme with CFL= 0.95, t = 1.70154. scheme with CFL= 0.1, t = 1.70036.

⃝1998
c by Hiroaki Nishikawa 8
 CFD Notes by Hiroaki Nishikawa www.cfdnotes.com

Density Velocity Density Velocity

1 1.5 1 2
MacCormack MacCormack
CFL=0.95 CFL=0.6518
0.8 33 time Steps 0.8 54 time Steps
1.5
1

Velocity u

Velocity u
Density ρ

Density ρ
0.6 0.6
1
0.4 0.4
0.5
0.5
0.2 0.2

0 0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Pressure Entroby/R Pressure Entroby/R

gas gas
1 2 1 2

0.8 1.5 0.8 1.5

1
Entropy/Rgas

Entropy/Rgas
Pressure P

Pressure P
0.6 1 0.6
0.5
0.4 0.5 0.4
0
0.2 0 0.2 −0.5

0 −0.5 0 −1
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Figure 5: The solutions by MacCormack’s Figure 6: The solutions by MacCormack’s

scheme with CFL= 0.95, t = 1.74408. scheme with CFL= 0.6518, t = 1.73141.

Density Velocity Density Velocity

1 1 1 1
Steger−Warming Steger−Warming
CFL=0.95 CFL=0.1
0.8 31 time Steps 0.8 0.8 284 time Steps 0.8
Velocity u

Velocity u
Density ρ

Density ρ

0.6 0.6 0.6 0.6

0.4 0.4 0.4 0.4

0.2 0.2 0.2 0.2

0 0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Pressure Entropy/Rgas Pressure Entropy/Rgas

1 2 1 2

0.8 1.5 0.8 1.5

Entropy/Rgas

Entropy/Rgas
Pressure P

Pressure P

0.6 1 0.6 1

0.4 0.5 0.4 0.5

0.2 0 0.2 0

0 −0.5 0 −0.5
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Figure 7: The solutions by Steger-Warming Figure 8: The solutions by Steger-Warming

scheme with CFL= 0.95, t = 1.73133. scheme with CFL= 0.1, t = 1.70308.

⃝1998
c by Hiroaki Nishikawa 9
 CFD Notes by Hiroaki Nishikawa www.cfdnotes.com

Density Velocity Density Velocity

1 1 1 1
Van Leer Van Leer
CFL=0.95 CFL=0.1
0.8 31 time Steps 0.8 0.8 282 time Steps 0.8

Velocity u

Velocity u
Density ρ

Density ρ
0.6 0.6 0.6 0.6

0.4 0.4 0.4 0.4

0.2 0.2 0.2 0.2

0 0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Pressure Entroby/R Pressure Entroby/R

gas gas
1 2 1 2

0.8 0.8
1.5 1.5
Entropy/Rgas

Entropy/Rgas
Pressure P

Pressure P
0.6 0.6
1 1
0.4 0.4

0.5 0.5
0.2 0.2

0 0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Figure 9: The solutions by Van Leer’s scheme Figure 10: The solutions by Van Leer’s scheme
with CFL= 0.95, t = 1.74534. with CFL= 0.1, t = 1.70217.

Density Velocity Density Velocity

1 1 1 1
Godunov Godunov
CFL=0.95 CFL=0.1
0.8 31 time Steps 0.8 0.8 283 time Steps 0.8
Velocity u

Velocity u
Density ρ

Density ρ

0.6 0.6 0.6 0.6

0.4 0.4 0.4 0.4

0.2 0.2 0.2 0.2

0 0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Pressure Entroby/Rgas Pressure Entroby/Rgas

1 2 1 2

0.8 0.8
1.5 1.5
Entropy/Rgas

Entropy/Rgas
Pressure P

Pressure P

0.6 0.6
1 1
0.4 0.4

0.5 0.5
0.2 0.2

0 0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Figure 11: The solutions by Godunov’s Figure 12: The solutions by Godunov’s
scheme with CFL= 0.95, t = 1.73846. scheme with CFL= 0.1, t = 1.70409.

⃝1998
c by Hiroaki Nishikawa 10
 CFD Notes by Hiroaki Nishikawa www.cfdnotes.com

Density Velocity Density Velocity

1 1 1 1
Osher Osher
CFL=0.95 CFL=0.1
0.8 31 time Steps 0.8 0.8 283 time Steps 0.8

Velocity u
Density ρ
0.6 0.6

Velocity u
Density ρ

0.6 0.6

0.4 0.4 0.4 0.4

0.2 0.2 0.2 0.2

0 0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Entroby/R Pressure Entroby/R

Pressure gas gas
1 2 1 2

0.8 0.8
1.5 1.5

Entropy/Rgas
Entropy/Rgas

Pressure P
Pressure P

0.6 0.6
1 1
0.4 0.4

0.5 0.5
0.2 0.2

0 0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Figure 13: The solutions by Osher’s scheme Figure 14: The solutions by Osher’s scheme
with CFL= 0.95 at t = 1.73402. with CFL= 0.1, t = 1.70201.

Density Velocity Density Velocity

1 1 1 1
Roe Roe
CFL=0.95 CFL=0.1
0.8 31 time Steps 0.8 0.8 283 time Steps 0.8
Velocity u

Velocity u
Density ρ

Density ρ

0.6 0.6 0.6 0.6

0.4 0.4 0.4 0.4

0.2 0.2 0.2 0.2

0 0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Pressure Entroby/Rgas Pressure Entroby/Rgas

1 2 1 2

0.8 0.8
1.5 1.5
Entropy/Rgas

Entropy/Rgas
Pressure P

Pressure P

0.6 0.6
1 1
0.4 0.4

0.5 0.5
0.2 0.2

0 0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Figure 15: The solutions Roe’s scheme with Figure 16: The solutions by Roe’s scheme with
CFL= 0.95, t = 1.73448. CFL= 0.1, t = 1.70268.

⃝1998
c by Hiroaki Nishikawa 11
 CFD Notes by Hiroaki Nishikawa www.cfdnotes.com

Density Velocity Density Velocity

1 1 1 1
2nd−order−MUSCL 2nd−order−MUSCL
CFL=0.95 CFL=0.1
0.8 31 time Steps 0.8 0.8 290 time Steps 0.8

Velocity u

Velocity u
Density ρ

Density ρ
0.6 0.6 0.6 0.6

0.4 0.4 0.4 0.4

0.2 0.2 0.2 0.2

0 0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Pressure Entropy/R Pressure Entropy/R

gas gas
1 2 1 2

0.8 0.8
1.5 1.5
Entropy/Rgas

Entropy/Rgas
Pressure P

Pressure P
0.6 0.6
1 1
0.4 0.4

0.5 0.5
0.2 0.2

0 0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5
x x x x

Figure 17: Second-order solution by MUSCL Figure 18: Second-order solution by MUSCL
with CFL= 0.95, t = 1.72672. with CFL= 0.1, t = 1.70463.

Density Velocity
4 4 Density Velocity
4 4
3.5 3.5
3.5 3.5
3 3
3 3
Velocity u
Density ρ

2.5 2.5
Velocity u
Density ρ

2.5 2.5
2 2
2 2
1.5 1.5
1.5 1.5
1 1
1 1
0.5 0.5
−5 0 5 −5 0 5 0.5 0.5
x x −5 0 5 −5 0 5
x x
Pressure Entropy/Rgas
Pressure Entropy/Rgas
12 2
12 2
10
1.5 10
8 1.5
Entropy/Rgas
Pressure P

8
Entropy/Rgas
Pressure P

6 1
6 1
4
0.5 4
2 0.5
2
0 0
−5 0 5 −5 0 5 0 0
x x −5 0 5 −5 0 5
x x

Figure 19: Lax-Friedrichs for the sonic rar-

Figure 20: Richtmyer for the sonic rarefaction
efaction with CFL= 0.95, 29 time steps, t =
with CFL= 0.95, 30 time steps, t = 0.711278.
0.711396.

⃝1998
c by Hiroaki Nishikawa 12
 CFD Notes by Hiroaki Nishikawa www.cfdnotes.com

Density Velocity
4 4 Density Velocity
4 4
3.5 3.5
3.5 3.5
3 3
3 3

Velocity u
Density ρ

2.5 2.5

Velocity u
Density ρ
2.5 2.5
2 2
2 2
1.5 1.5
1.5 1.5
1 1
1 1
0.5 0.5
−5 0 5 −5 0 5 0.5 0.5
x x −5 0 5 −5 0 5
x x
Pressure Entropy/R
gas
Pressure Entropy/R
12 1.5 gas
12 1.4
10
10 1.2
1
8 1
Entropy/Rgas
Pressure P

Entropy/Rgas
Pressure P
6 0.5 0.8
6
0.6
4
0 4
0.4
2
2 0.2
0 −0.5
−5 0 5 −5 0 5 0 0
x x −5 0 5 −5 0 5
x x

Figure 21: Steger-Warming for the sonic

Figure 22: Godunov for the sonic rarefaction
rarefaction CFL= 0.3, 77 time steps, t =
with CFL= 0.95, 29 time steps, t = 0.70471.
0.705027.

Density Velocity
Density Velocity 4 4
4 4
3.5 3.5
3.5 3.5
3 3
3 3

Velocity u
Density ρ

2.5 2.5
Velocity u
Density ρ

2.5 2.5
2 2
2 2
1.5 1.5
1.5 1.5
1 1
1 1
0.5 0.5
0.5 0.5 −5 0 5 −5 0 5
−5 0 5 −5 0 5 x x
x x
Pressure Entropy/Rgas
Pressure Entropy/Rgas
12 1.5
12 1.4
10
10 1.2
8 1
Entropy/Rgas

1
Pressure P

8
Entropy/Rgas
Pressure P

0.8 6
6
0.6 4 0.5
4
0.4
2
2 0.2
0 0
0 0 −5 0 5 −5 0 5
−5 0 5 −5 0 5 x x
x x

Figure 24: Roe’s scheme with an entropy fix

Figure 23: Osher for the sonic rarefaction with
for the sonic rarefaction CFL= 0.95, 30 time
CFL= 0.95, 29 time steps, t = 0.70471.
steps, t = 0.719955.

⃝1998
c by Hiroaki Nishikawa 13