GODUNOV TYPE SCHEMES

Godunov’s idea

§ Consider every 2 grid cells as a tiny shock tube

§ Solve the wave propagation at each interface (g Riemann problem)
t
tn
+1

t xi-1/2 xi+1/2 xi+3/2

n
xi-1 xi xi+1 x
u
uni+1
uni+2
uni
uni−1

x
xi-1/2 xi+1/2 xi+3/2
2 Godunov Type Schemes - 2nd order (MUSCL)
 Godunov’s method

§ Numerical flux in Godunov’s method:

gni+1/2 = f(uRP (0;ui ,ui+1))

uRP: solution of the Riemann problem at cell interface

§ Properties of Godunov’s method:

§ Exact conservation
§ Non-linear wave propagation
§ Adaptivity to the solution

§ Stability: CFL condition

3 Godunov Type Schemes - 2nd order (MUSCL)
 Approximate Godunov schemes

§ Godunov’s method is stable and conservative, but expensive

§ What is expensive? The non-linearity
ut + r · f (u) = 0
<latexit sha1_base64="cfiL/jJcv+fXdbPz0DTgJNTVWoE=">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</latexit>

<latexit sha1_base64="VLCymXpQhZl0LB/iqbGtLo7cVQ0=">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</latexit>

@f1 (u) @f2 (u)

ut + + =0
@x @y
<latexit sha1_base64="Lvq/t1gd9X6hpJipJedmwF3L3/A=">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</latexit>

@f1 (u) @u @f2 (u) @u

ut + + =0
@u @x @u @y
@u @u
<latexit sha1_base64="dJXXHOCvaidgKNP3dR50DNX3R8o=">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</latexit>

ut + A1 (u) + A2 (u) =0
@x @x

§ Idea: linearize g Roe Scheme

4 Godunov Type Schemes - 2nd order (MUSCL)
 Roe linearization

§ P. Roe (1981): Replace the exact solution of the original Riemann problem
by the exact solution of the linearized Riemann problem:

ì u l für x < 0
u t + A lr u x = 0, u (x ,0 ) = í
î u r für x > 0

§ The matrix Alr= Alr(ul,ur) is called Roe matrix, if

1. Alr(u,u) = A(u)
2. Alr can be diagonalized (i.e. Alr is hyperbolic)
3. Alr fulfills the mean value property: f (ur ) − f (ul ) = Alr ⋅ (ur − ul )

5 Godunov Type Schemes - 2nd order (MUSCL)

 Solving the linearized Riemann problem

§ The 4 constant states

x t x x
= a1 = a2 = a3
t t t
u1 u2
u1 = ul + g1r1
ul ur u2 = ul + g1r1 + g 2r2

1 1 3
§ The flux g Roe (u i , u i +1 ) = (f i +1 + f i ) - å a k g k rk
2 2 k =1
Central difference Correction according
to wave propagation
6 Godunov Type Schemes - 2nd order (MUSCL)
 Properties of the Roe scheme

§ A single shock or a single contact discontinuity is resolved exactly.

§ In rarefactions, it underestimates the wave speed of the rarefaction.

§ At strong rarefactions, the Roe scheme might violate the entropy condition.

"
$ a für a1,3 > δ
g Entropy fix: a 1,3 = # 1,3
$
% δ otherwise

7 Godunov Type Schemes - 2nd order (MUSCL)

 The Method of Harten, Lax and Van Leer

§ Simplest Godunov type scheme

x x
= al = ar
t t t

Dt
ulr

Dx Dx x
-
2 2

8 Godunov Type Schemes - 2nd order (MUSCL)

 The Method of Harten, Lax and Van Leer

§ Even further simplification: Only one state between left and right
( x
* ul for < al
* t
!x $ * x
w # ; ul ; ur & = ) ulr for al ≤ ≤ ar
"t % * t
* x
* ur for > ar
+ t

§ al, ar smallest and largest wave speed

§ ulr intermediate state

9 Godunov Type Schemes - 2nd order (MUSCL)

 Comparison of the 3 Riemann solvers

Godunov-scheme Roe-scheme HLL-scheme

Exact solution of the Exact solution of the A priori guess of
Riemann problem, lin. Riemann problem, highest/lowest
Fix point iteration Characteristics theorie wave speeds

t t t
Contact
discontinuity x = a2t
Rarefaction x = a1t
x = a 3t x = alt x = art
u1 u2 u1 ulr
u2

ul ur x ul ur x ul ur x

10 Godunov Type Schemes - 2nd order (MUSCL)

 Flux calculation – overview

Upwind differencing
§ Flux-difference splitting schemes: Godunov-type schemes
Godunov, Roe, HLL, Osher
§ Flux-vector splitting schemes
Steger-Warming, van Leer, AUSM, ...

Piecewise constant values

First order accurate, quite similar results

11 Godunov Type Schemes - 2nd order (MUSCL)

TIME STEP RESTRICTION
 Time step restrictions

§ Explicit time discretizations are numerically stable only under a condition. I.e. the max. time
step is restricted:

xi x
xi -1/ 2 xi +1/ 2

§ To calculate the flux over the boundary, we assume that the state on the cell boundary is
constant during the time step.
§ This is justified only as long as no waves of the neighbouring Riemann problem interact with
the ones of the local Riemann problem.
13 Godunov Type Schemes - 2nd order (MUSCL)
 Time step restriction

§ I.e. we search for the time, which the fastest wave needs from one cell
boundary to the next one:
t Δt i 1
=
< CFL condition
Δx i v1i + c i
v1i + c Δx i
⇒ Δt i <=
v1i + c i

xi x Δx i = x i+1/2 − x i−1/2
xi -1/ 2 xi +1/ 2

§ In practice, the CFL conditions (named after Courant, Friedrichs und Levy)
is multiplied by a security factor s to get from inequality to equation:
Δx i
Δt i = σ ⋅
v1i + c i
14 Godunov Type Schemes - 2nd order (MUSCL)
 Time step restriction

§ The states in the cells vary as well as the cell sizes in the computational domain.
Therefore, the time step has to be caculated per cell (local time step).
§ The entire simulation than has to use the smallest local time step:

$ '
Δx
Δt global = σ ⋅ min & max i )
&v )
% i + ci (

§ For stationary problems, it is possible to use the local time step in each cell (i.e.
allow for temporal inconsistency).
§ This is possible since for the fully converged solution, the change of the states is
zero.
§ For instationary problems, this is not possible.
15 Godunov Type Schemes - 2nd order (MUSCL)
 Time step calculation on arbitrary meshes

In multiple space dimensions, the time step restriction becomes more severe.
We have to use a characteristic lenth for each cell, i.e. the diameter of the
inner circle.

$ '
2 ⋅ r
Δt global = σ ⋅ min & max i )
&v + c )
% i i (

Be careful with strongly distorted cells (especially in Navier-Stokes).

The inner circle becomes very small, leading to tiny time steps.
16 Godunov Type Schemes - 2nd order (MUSCL)
2ND ORDER SCHEME: MUSCL

MONOTONIC UPWIND SCHEME

FOR CONSERVATION LAWS
 2nd order in space and time: MUSCL scheme – basic idea

§ Flux formulation: gi+1/2,j is an approximation to what flows in or out over

the boundary i+1/2,j of cell i,j during the entire time intervall.
§ Problem: we only know uij, i.e. what is inside cell i,j at time level tn, but not
its distribution inside the cell or how it varies during the time step.
§ Idea: assume a linear distribution inside the cell to determine the flux at
the boundary more accurately.

§ => explicit scheme 2nd order in space and time

18 Godunov Type Schemes - 2nd order (MUSCL)

 Upwind 2nd order

§ Explicit in time
§ Upwind in space

§ 2nd order in space

§ 2nd order in time

§ The MUSCL scheme

§ MUSCL: monotonic upwind scheme for conservation laws

19 Godunov Type Schemes - 2nd order (MUSCL)

 Finite volume discritzation – cell mean values

Real values at Real

cell boundaries distribution

integral
mean values

§ Problem: the constant representation (i.e. representation with polynomial of

order 0)is quite inexact. It leads to an overall scheme of 1st order

20 Godunov Type Schemes - 2nd order (MUSCL)

 Let’s try to use higher order polynomials

§ Let’s try to use higher order polynomials WITHIN a cell

§ Order of the polynomial is related to the order of the entire scheme.

§ Up to now:
Polynomial of degree 0 = constant values per cell, scheme of 1st order
§ Idea for next step:
Polynomial of degree 1 = linear distribution, scheme of 2nd order?

21 Godunov Type Schemes - 2nd order (MUSCL)

 Piecewise linear reconstruction

Monotonic Upwind Scheme for Conservation Laws

xi-1 xi xi+1 xi+2 x

Instead of assuming u=const in [xi-1/2, xi+1/2], we now assume that u is linearly

distributed. I.e. we determine a straight line, evaluate it at the boundaries and
determine the fluxes with that.
22 Godunov Type Schemes - 2nd order (MUSCL)
 1st step: get the slope of the line

§ To determine the straight line, we need a point and a slope.

§ The point is xij with the value uij.
§ Slope: 2 possibilities, left sided or right sided difference:
ui, j −ui−1, j ui+1, j −ui, j
Lij = , Rij =
Δx Δx
§ We have to chose one of them or a linear combination
§ This is done by the€so-called limiter
§ For 2nd order, it has to fulfill a condition (TVD property)

23 Godunov Type Schemes - 2nd order (MUSCL)

 Limiter: TVD property

§ TVD property (Total Variation Diminishing)

n
∑u i+1 − uni ≤ ∑ ui+1
0
− ui0
all i all i

§ Sufficient condition
€ (A.Harten)
% Δx si Δx si (
0≤& , )≤ 2
u
' i − ui−1 ui+1 − ui*

€
24 Godunov Type Schemes - 2nd order (MUSCL)
 Limiters

§ Minmod limiter (
si = minmod Li, j , Ri, j )
#a for a < b , ab > 0
%%
€ ( )
minmod a , b = $b for a ≥ b , ab > 0
%
%&0 otherwise

§ Sweby‘s limiter: weighted minmod

€
( ) { ( )
sk a,b = sign(a) max minmod a , kb , minmod ka , b ( )}
with 1 ≤ k ≤ 2

€
25 Godunov Type Schemes - 2nd order (MUSCL) €
 Reconstruction in space and time

§ At time tn, extrapolate from middle of the cell (i,j) to boundary of the cell
(i+1/2,j)
Δx n
uni± = uni ± si slope sin
2
§ To get the 2nd order also in time, the basic idea is the same: extrapolate from
time level tn to time level tn+1:
n+ 1 Δt
ui±
2
= uni± + u
2 t

§ But: we can not compute a slope in time, because we only have 1 time level.
Therefore, we replace €the time derivatives by spatial derivatives
Δt Δt
t n → t n+1/2: un+1/2
i±
= uni± −
2 ⋅ Δx
( ( ) ( ))
f uni+ − f uni− = uni± −
2 ⋅ Δx
(
ai±1/2 uni+ − uni− )
26 Godunov Type Schemes - 2nd order (MUSCL)
 How to get the flux with that

u (in ++11)/ 2- , j ui+1,j

u in++,1j/ 2
uij

xi-1 xi xi+1 xi+2 x

§ Now we have to change the way we think from cell-oriented to boundary

oriented: the flux at boundary i+1/2,j is now

gn+1/2 (n+1/2
i+1/2 = g ui+ , un+1/2
(i+1) − )
27 Godunov Type Schemes - 2nd order (MUSCL)

€
 Upwind scheme with MUSCL

gi+ 12, j = gn+1/2 (

i+1/2, j = g ui+, j
n+1/2
, un+1/2
(i+1) −, j )
#un+1/2 if ai+ 12, j > 0
%% i+, j
= ai+ 12, j $ 21 (un+1/2
i+, j + un+1/2
(i+1)−, j ) if ai+ 12, j = 0
% n+1/2
%&u(i+1)−, j if ai+ 12, j < 0

hi, j+ 12 analogously

28
€
Godunov Type Schemes - 2nd order (MUSCL)
 MUSCL procedure (summary)

Δx n
§ Boundary values at tn uni± = uni ± si Slope sin
2

§ tn g tn+1/2 uni±+1/2 = uni± -

Δt
2 × Dx
( ( ) ( ))
f uni+ - f uni-

§ Final step: uni+1 = uni - (

Δt n+1/2 n+1/2
gi+1/2 - gi-1/2 ) with gni++1/2
1/2
(
= g uni++1/2 , un(i++1/2
1)- )
Δx

29 Godunov Type Schemes - 2nd order (MUSCL)

 Summary – General FV scheme

Determine the time step

Set boundary conditions

Reconstruction

Set boundary conditions

Flux calculation

Update the state at next time level

30 Godunov Type Schemes - 2nd order (MUSCL)