You are on page 1of 28

ENO and WENO schemes.

Further topics
and time Integration

Tefa Kaisara

CASA Seminar

29 November, 2006
Short review Further topics Time Integration

Outline

1 Short review
ENO/WENO

2 Further topics
Subcell resolution
Other building blocks

3 Time Integration
TVD Runge-Kutta methods
TVD multistep methods
Lax-Wendroff procedure
Short review Further topics Time Integration

Outline

1 Short review
ENO/WENO

2 Further topics
Subcell resolution
Other building blocks

3 Time Integration
TVD Runge-Kutta methods
TVD multistep methods
Lax-Wendroff procedure
Short review Further topics Time Integration

ENO/WENO

Solving hyperbolic conservation laws,consider;


Class of piecewise smooth functions,v (x)
Fixed stencil,S(i) = {Ii−r , ..., Ii+s }, approximation may lead to
oscillations near the discontinuity
Adaptive stencil; left shift r
Short review Further topics Time Integration

ENO/WENO

ENO Procedure
Compute the divided differences, using the cell averages.
For cell Ii start with the two-point stencil S̃(i) = {xi− 1 , xi+ 1 }.
2 2

Add one of the two neighbouring points,using Newton divided


difference measure.
Repeat previous step until a k-th order stencil is obtained.
Use reconstruction method to compute the approximation to v (x)

In the stencil choosing process, k candidate stencils are


considered covering 2k − 1 cells, but only one stencil is used to
reconstruct.
Short review Further topics Time Integration

ENO/WENO

To improve the accuracy:

Assume a uniform grid, i.e. ∆xi = ∆x, i = 1, . . . , N. Consider the k


candidate stencils

Sr (i) = {xi−r , . . . , xi−r +k −1 } r = 0, . . . , k − 1.

The reconstruction produces k different approximations to the value


v (xi+ 1 ) :
2

(r )
X
k−1
vi+ 1 = crj v̄i−r +j r = 0, . . . , k − 1 (1.1)
2
j=0

(r )
I WENO takes a convex combination of the vi+ 1 ’s.
2
Short review Further topics Time Integration

Outline

1 Short review
ENO/WENO

2 Further topics
Subcell resolution
Other building blocks

3 Time Integration
TVD Runge-Kutta methods
TVD multistep methods
Lax-Wendroff procedure
Short review Further topics Time Integration

Subcell resolution

Instead of using the reconstruction polynomial Pi (x) in the


shocked cell Ii , find the location of discontinuity inside Ii , say xs
Use neighbouring reconstructions Pi−1 (x) and Pi+1 (x)

Extend Pi−1 (x) and Pi+1 (x) into the cell Ii


Require that the cell average v̄i is preserved:
Z xs Z xi + 12
Pi−1 (x)dx + Pi+1 (x)dx = ∆xi v̄i
xi − 12 xs
Short review Further topics Time Integration

Other building blocks

Examples
rational functions
trigonometric polynomials
exponential functions
radial functions

idea
Find suitable smooth indicators similar to Newton divided difference
for the polynomial case.
Short review Further topics Time Integration

Outline

1 Short review
ENO/WENO

2 Further topics
Subcell resolution
Other building blocks

3 Time Integration
TVD Runge-Kutta methods
TVD multistep methods
Lax-Wendroff procedure
Short review Further topics Time Integration

TVD Runge-Kutta methods

Consider,
system of ODEs
ut = L(u)
resulting from spatial approximation to PDE;

ut = −f (u)x

Construction
Assume: First order Euler forward time stepping

u n+1 = u n + ∆tL(u n )

is stable in a certain norm

||u n+1 || ≤ ||u n ||

under a suitable restriction on ∆t : ∆t ≤ ∆t1


I For higher order in time Runge-Kutta methods,
Short review Further topics Time Integration

TVD Runge-Kutta methods

General Runge-Kutta method

X
i−1  
u (i)
= αik u k + ∆tβik L(u (k ) ) , i = 1, . . . , m (3.2)
k=0
n
u (0) = u ,u(m)
= u n+1

For αik ≥ 0, βik ≥ 0 ⇒ convex combination of Euler forward operators


αik
∆t replaced by ∆t
βik

Consistency;
X
i−1
αik = 1.
k =0
Short review Further topics Time Integration

TVD Runge-Kutta methods

Lemma
The Runge-Kutta method (3.2) is TVD under the CFL coefficient (3.1)
αik
c = min
i,k βik

provided; αik ≥ 0 and βik ≥ 0.


Short review Further topics Time Integration

TVD Runge-Kutta methods

Optimal schemes
Second order TVD Runge-Kutta

u (1) = u n + ∆tL(u n )
1 n 1 (1) 1
u n+1 = u + u + ∆tL(u (1) )
2 2 2
with CFL c = 1.
Third order TVD R-K

u (1) = u n + ∆tL(u n )
3 n 1 (1) 1
u (2) = u + u + ∆tL(u (1) )
4 4 4
1 n 2 (2) 2
u n+1 = u + u + ∆tL(u (2) )
3 3 3
Fourth order TVD Runge-Kutta method does not exist with for
positive αik and βik
Short review Further topics Time Integration

TVD Runge-Kutta methods

Consider
αik ≥ 0 and βik negative
introduce an adjoint operator L̃
it approximates the same spatial derivatives as L
TVD for first order Euler, backward in time:

u n+1 = u n − ∆t L̃u n
Short review Further topics Time Integration

TVD Runge-Kutta methods

Lemma
The Runge-Kutta method (3.2) is TVD under the CFL coefficient (3.1)
αik
c = min
i,k |βik |

provided; αik ≥ 0 and L replaced by L̃


Short review Further topics Time Integration

TVD Runge-Kutta methods

Fourth order TVD R-K

1
u (1) = un + ∆tL(u n )
2

649 (0) 10890423 951 (1) 5000


u (2) = u − ∆t L̃(u n ) + u + ∆tLu (1)
1600 25193600 1600 7873

53989 n 102261 4806213 (1) 5121


u (3) = u − ∆t L̃(u n ) + u − ∆t L̃(u (1) )
2500000 5000000 20000000 20000
23619 (2) 7873
+ u + ∆tL(u (2) )
32000 10000

1 n 1 6127 (1) 1 7873 (2)


u n+1 = u + ∆tL(u n ) + u + ∆tL(u (1) ) + u
5 10 30000 6 30000
1 1
+ u (3) + ∆tL(u (3) )
3 6
Short review Further topics Time Integration

TVD Runge-Kutta methods

Example
Burgers’ equation  
1 2
ut + u =0
2 x

with Riemann initial data



1, x ≤0
u(x, 0) =
−0.5, x > 0

Spatial discretization : 2nd order ENO Scheme


Monotone flux h: Godunov flux
Time discretization
TVD R-K method :Optimal 2nd order
Non-TVD R-K

u (1) = u n − 20∆tL(u n )
41 (1) 1
u n+1 = un + u − ∆tL(u (1) )
40 40
Short review Further topics Time Integration

TVD Runge-Kutta methods

conclusion
It is much safer to use TVD Runge-Kutta Method for solving
hyperbolic problem.
Short review Further topics Time Integration

TVD multistep methods

Construction
Similar to Runge-kutta Methods

General From
X
m−1
u n+1 = (αk u n−k + ∆tβk L(u n−k )) (3.3)
k =0

For αk ≥ 0, βk ≥ 0 ⇒ convex combination of Euler Forward operators


βk
∆t replaced by ∆t
αk
Consistency;
X
m
αk = 1.
k =0
Short review Further topics Time Integration

TVD multistep methods

Lemma
The multi-step method (3.3) is TVD under the CFL coefficient (3.1)
αk
c = min
k βk
provided; αk ≥ 0 and βk ≥ 0
Short review Further topics Time Integration

TVD multistep methods

Examples
(m=2) three step , second order scheme, TVD with c=0.5

3 n 3 1
u n+1 =u + ∆tL(u n ) + u n−2 (3.4)
4 2 4
(m=4) five step,third order scheme, TVD with c=0.5

25 n 25 7 n−4 5
u n+1 = u + ∆tL(u n ) + u + ∆tL(u n−4 )
32 16 32 16

There are no multi-step schemes of order four or higher satisfying


positive αk and βk
Short review Further topics Time Integration

TVD multistep methods

Lemma
The multi-step method (3.3) is TVD under the CFL coefficient (3.1)
αk
c = min
k |βk |
provided; αk ≥ 0 and , L is replaced by L̃ for βk negative.

one residue evaluation is needed per time step


storage requirement is much bigger that Runge-Kutta methods
Short review Further topics Time Integration

Lax-Wendroff procedure

Idea
Taylor series expansion in time:

∆t 2
u(x, t + ∆t) = u(x, t) + ut (x, t)∆t + utt (x, t) + ... (3.5)
2

Illustration
ut = −f (u)x

second order Taylor series expansion


Replace time derivatives by the spatial derivatives
Short review Further topics Time Integration

Lax-Wendroff procedure

ut (x, t) = −f (u(x, t))x = −f 0 (u(x, t))ux (x, t)


utt (x, t) = 2f 0 (u(x, t))f 00 (u(x, t))(ux (x, t))2 + (f 0 (u(x, t)))2 uxx (x, t)
(3.6)

Substituting (3.6) into (3.5)


Discretize the spatial derivatives of u(x, t) by ENO/WENO
Integrate the pde
ut (x, t) + fx (u(x, t)) = 0
over the region [xi− 1 , xi+ 1 ] × [t n , t n+1 ] to obtain
2 2

Z t n+1 Z t n+1 !
1
ūin+1 = ūin − f (u(xi+ 1 , t)) dt − f (u(xi− 1 , t)) dt
∆xi tn
2
tn
2

(3.7)
Discretize time using suitable Gaussian quadrature
Z n+1 X
1 t
f (u(xi+ 1 , t)) dt ≈ ωα f (u(xi+ 1 , t n + βα ∆t))
∆t t n 2
α
2
Short review Further topics Time Integration

Lax-Wendroff procedure

replacing f (u(xi+ 1 , t n + βα ∆t)) by a monotone flux


2

±
f (u(xi+ 1 , t n + βα ∆t)) ≈ h(u(xi+ 1,t
n
+ βα ∆t)
2 2

use the Lax-Wendroff procedure to convert


± n
u(xi+ 1,t + βα ∆t) (3.8)
2

± n n
to u(xi+ 1 , t ) and its spacial derivatives at t .
2
Short review Further topics Time Integration

Lax-Wendroff procedure

limitations
Algebra is very complicated for multi dimensional systems
difficult to prove stability properties for higher order methods
Short review Further topics Time Integration