You are on page 1of 42

ANALYSIS OF STRONG STABILITY PRESERVING RUNGE-KUTTA TIME DISCRETIZATION

Gabriel Leonardo

de Moraes

S. de B. Alves

Instituto Militar de Engenharia P os-Gradua c ao em Engenharia de Defesa Universidade Federal Fluminense Departamento de Engenharia Mec anica

VII Congresso Nacional de Engenharia Mec anica 2012, S ao Lu s, Maranh ao 31-03 de Agosto
Leonardo S. de B. Alves (UFF) Gabriel de Moraes (IME) CONEM 2012 1 / 42

Introduction

Motivation
Historically1,2 , the development of SSP methods was motivated in two ways:

Hyperbolic PDEs Nonlinear numerical stability

Sigal Gottlieb. 2005. On High Order Strong Stability Preserving Runge-Kutta and Multi Step Time Discretizations. J. Sci. Comput. 2 S. Gottlieb, C.W. Shu and E. Tadmor, Strong Stability Preserving High Order Time Discretization Methods SIAM review vol. 43 no. 1 (2001), pp. 89-112
Leonardo S. de B. Alves (UFF) Gabriel de Moraes (IME) CONEM 2012 2 / 42

Methodology

Forward Euler Method

From a method of lines approximation of hyperbolic conservation law : ut = f (u)x the rst-order forward Euler time discretization gives, un+1 = un + tF (un ) and linear stability dictates the larger allowable time step (CFL), t tF E . (3) (2) (1)

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

3 / 42

Methodology

SSP High-Order Methods

If is assumed that eq.2 is stable under a certain norm (or a convex functional), un+1 un (4)

then a SSP high order time discretization maintains this stability under a suitable restriction on the time step. t ctF E . (5)

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

4 / 42

Methodology

The Shu-Osher formulation


The Shu-Osher formulation rearranges each stage of the Runge-Kutta method into convex combinations of forward Euler steps ,
i 1 i 1

(i)

=
j =0

i,j u

(j )

+ti,j F (u

(j )

j =0

i,j u(j ) +t

i,j F (u(j ) ) i,j

(6) and, the solution obtained by the Runge-Kutta method satises the strong stability bound under the time step restriction: ij |ij |

t c(, )tF E

where

c(, ) = minij

(7)

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

5 / 42

Methodology

Total Variation Diminishing (TVD) property

The Total Variation (TV) of the one-dimension discrete solution does not increase in time, the so-called TVD property holds: T V (un+1 ) T V (un ), T V (un ) =
j n | un j +1 uj |

(8)

(9)

Thus, by the SSP approach is possible develop a TVD scheme for higher orders.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

6 / 42

Methodology

Explicit Scheme

Explicit SSPRK (2,2)

The optimal explicit two-stage second order (2,2) SSP Runge-Kutta method written in the Shu-Osher form: k (1) = un + tF (un ), 1 1 1 = un + k (1) + tF (k (1) ). 2 2 2

(10) (11)

un+1

for this method the SSP coecient is c = 1, t 1tF E (12)

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

7 / 42

Methodology

Implicit Scheme

Implicit SSPRK (1,2)

The second order implicit Runge-Kutta method based on midpoint rule is given, 1 k (1) = un + tF (k (1) ), 2 1 un+1 = k (1) + tF (k (1) ), 2 for this method the SSP coecient is c = 2. t 2tF E

(13) (14)

(15)

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

8 / 42

Methodology

Implicit Scheme

Spatial discretization

Conservative rst order upwind approximation, F (x) = 1 (f (ui ) f (ui1 )). x (16)

This spatial discretization is TVD for t x when coupled with forward Euler in time.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

9 / 42

Numerical Investigation

Numerical Investigation : Burgers Equation

Numerical simulations were made on the invicid Burgers equation: ut + f (u)x = 0, Considering, f (u) =
u2 2

(17)

, the initial data, u(x, 0) = 1 1 sin(x) 2 4 (18)

and periodic boundary conditions.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

10 / 42

Results

Time step barrier investigation

Explicit scheme - Solution


dt = 1.4 dx dt = 1.2 dx

0.7

0.6

u/U 0

0.5

0.4

0.3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x/L

Figure: Invicid Burgers solution 1.0s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

11 / 42

Results

Time step barrier investigation

Explicit scheme - Solution


dt = 1.4 dx dt = 1.2 dx

0.7

0.6

u/U 0

0.5

0.4

0.3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x/L

Figure: Invicid Burgers solution at t= 1.2s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

12 / 42

Results

Time step barrier investigation

Explicit scheme - Solution


dt = 1.4 dx dt = 1.2 dx

0.7

0.6

u/U 0

0.5

0.4

0.3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x/L

Figure: Invicid Burgers solution at= 1.4s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

13 / 42

Results

Time step barrier investigation

Explicit scheme - Solution


dt = 1.4 dx dt = 1.2 dx

0.7

0.6

u/U 0

0.5

0.4

0.3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x/L

Figure: Invicid Burgers solution at= 1.6s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

14 / 42

Results

Time step barrier investigation

Explicit scheme - Solution


dt = 1.4 dx dt = 1.2 dx

0.754 0.752

0.7

0.75 0.748 0.746

0.6

0.744 0.41

0.42

0.43

0.44

u/U 0

0.5

0.4

0.3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x/L

Figure: Invicid Burgers solution at= 1.8s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

15 / 42

Results

Time step barrier investigation

Explicit scheme - Solution


dt = 1.4 dx dt = 1.2 dx

0.754
0.76

0.752
0.75

0.7

0.75
0.74

0.748
0.73

0.746

0.6

0.744 0.41

0.485

0.49

0.495

0.5

0.505

0.42

0.43

0.44

u/U 0

0.5

0.4

0.3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x/L

Figure: Invicid Burgers solution at= 2.0s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

16 / 42

Results

Time step barrier investigation

Explicit scheme - Solution


dt = 1.4 dx dt = 1.2 dx

0.754
0.76

0.752
0.75

0.7

0.75
0.74

0.748
0.73

0.746

0.6

0.744 0.41

0.485

0.49

0.495

0.5

0.505

0.42

0.43

0.44

u/U 0

0.5

0.4

0.3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x/L

Figure: Invicid Burgers solution at= 2.2s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

17 / 42

Results

Time step barrier investigation

Explicit scheme - Solution


dt = 1.4 dx dt = 1.2 dx

0.754
0.76

0.752
0.75

0.7

0.75
0.74

0.748
0.73

0.746

0.6

0.744 0.41

0.485

0.49

0.495

0.5

0.505

0.42

0.43

0.44

u/U 0

0.5

0.4

0.3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x/L

Figure: Invicid Burgers solution at= 2.4s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

18 / 42

Results

Time step barrier investigation

Explicit scheme - Solution


dt = 1.4 dx dt = 1.2 dx

0.754
0.76

0.752
0.75

0.7

0.75
0.74

0.748
0.73

0.746

0.6

0.744 0.41

0.485

0.49

0.495

0.5

0.505

0.42

0.43

0.44

u/U 0

0.5

0.4

0.3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x/L

Figure: Invicid Burgers solution at= 2.6s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

19 / 42

Results

Time step barrier investigation

Explicit scheme - Solution


dt = 1.4 dx dt = 1.2 dx

0.754
0.76

0.752
0.75

0.7

0.75
0.74

0.748
0.73

0.746

0.6

0.744 0.41

0.485

0.49

0.495

0.5

0.505

0.42

0.43

0.44

u/U 0

0.5

0.4

0.3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x/L

Figure: Invicid Burgers solution at= 2.8s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

20 / 42

Results

Time step barrier investigation

Explicit scheme - Solution


dt = 1.4 dx dt = 1.2 dx

0.754
0.76

0.752
0.75

0.7

0.75
0.74

0.748
0.73

0.746

0.6

0.744 0.41

0.485

0.49

0.495

0.5

0.505

0.42

0.43

0.44

u/U 0

0.5

0.4

0.3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x/L

Figure: Invicid Burgers solution at= 3.0s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

21 / 42

Results

Time step barrier investigation

Explicit scheme - dt = 1.0 dx

3 2.75 2.5

1 0.99 0.98 0.97 0.96

Average Order

2.25 2 1.75 1.5 1.25

0.95

TV
0.25 0.5 0.75 1

0.94 0.93 0.92 0.91 0.9 0.89 0.88

x/L

0.5

1.5

2.5

time(s)

(a) Temporal Order Preserved


Leonardo S. de B. Alves (UFF) Gabriel de Moraes (IME)

(b) TVD Preserved


CONEM 2012 22 / 42

Results

Time step barrier investigation

Behavior Beyond time step Barrier - dt = 1.2 dx

1.5

1 0.99 0.98

1.25

0.97 0.96 0.95

Average Order

TV
0.25 0.5 0.75 1

0.94 0.93 0.92

0.75

0.91 0.9 0.89 0.88

x/L

0.5

1.5

2.5

time(s)

(c) Temporal Order Reduction


Leonardo S. de B. Alves (UFF) Gabriel de Moraes (IME)

(d) TVD Preserved


CONEM 2012 23 / 42

Results

Time step barrier investigation

Behavior Beyond time step Barrier - dt = 1.4 dx

-6 -7 -8 -9

13 12 11 10 9 8

Average Order

-10

TV
0.25 0.5 0.75 1 1.25 1.5 1.75 2

-11 -12 -13 -14 -15 -16 -17

7 6 5 4 3 2 1 0.5 1 1.5 2 2.5 3

x/L

time(s)

(e) Loss of Temporal Precision


Leonardo S. de B. Alves (UFF) Gabriel de Moraes (IME)

(f) TVD violated


CONEM 2012 24 / 42

Results

Time step barrier investigation

Implicit scheme - Solution


dt = 3 dx dt = 2 dx

0.7

0.6

u/U 0

0.5

0.4

0.3 0 0.25 0.5 0.75 1

x/L

Figure: Invicid Burgers solution at= 1.0s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

25 / 42

Results

Time step barrier investigation

Implicit scheme - Solution


dt = 3 dx dt = 2 dx

0.7

0.6

u/U 0

0.5

0.4

0.3 0 0.25 0.5 0.75 1

x/L

Figure: Invicid Burgers solution at= 1.2s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

26 / 42

Results

Time step barrier investigation

Implicit scheme - Solution


dt = 3 dx dt = 2 dx

0.7

0.6

u/U 0

0.5

0.4

0.3 0 0.25 0.5 0.75 1

x/L

Figure: Invicid Burgers solution at= 1.4s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

27 / 42

Results

Time step barrier investigation

Implicit scheme - Solution


dt = 3 dx dt = 2 dx

0.7

0.6

u/U 0

0.5

0.4

0.3 0 0.25 0.5 0.75 1

x/L

Figure: Invicid Burgers solution at= 1.6s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

28 / 42

Results

Time step barrier investigation

Implicit scheme - Solution


dt = 3 dx dt = 2 dx

0.7

0.6

u/U 0

0.5

0.4

0.3 0 0.25 0.5 0.75 1

x/L

Figure: Invicid Burgers solution at= 1.8s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

29 / 42

Results

Time step barrier investigation

Implicit scheme - Solution


dt = 3 dx dt = 2 dx

0.751 0.75

0.7

0.749 0.748 0.747 0.746

0.6

0.745

0.499

0.4995

0.5

u/U 0

0.5

0.4

0.3 0 0.25 0.5 0.75 1

x/L

Figure: Invicid Burgers solution at= 2.0s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

30 / 42

Results

Time step barrier investigation

Implicit scheme - Solution


dt = 3 dx dt = 2 dx
0.7485

0.751 0.74825 0.750.748


0.74775 0.749

0.7

0.748 0.747

0.7475

0.74725 0.747

0.746 0.74675

0.6

0.745

0.546 0.547 0.548 0.5490.550.551 0.552 0.499 0.4995 0.5

u/U 0

0.5

0.4

0.3 0 0.25 0.5 0.75 1

x/L

Figure: Invicid Burgers solution at= 2.2s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

31 / 42

Results

Time step barrier investigation

Implicit scheme - Solution


dt = 3 dx dt = 2 dx

0.751 0.75
0.742

0.7

0.749 0.748 0.747 0.746


0.741 0.598 0.599 0.6 0.7415

0.6

0.745

0.499

0.4995

0.5

u/U 0

0.5

0.4

0.3 0 0.25 0.5 0.75 1

x/L

Figure: Invicid Burgers solution at= 2.4s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

32 / 42

Results

Time step barrier investigation

Implicit scheme - Solution


dt = 3 dx dt = 2 dx

0.751 0.75

0.7

0.749 0.748 0.747 0.746

0.6

0.745

0.499

0.4995

0.5

u/U 0

0.5

0.4

0.3 0 0.25 0.5 0.75 1

x/L

Figure: Invicid Burgers solution at= 2.6s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

33 / 42

Results

Time step barrier investigation

Implicit scheme - Solution


dt = 3 dx dt = 2 dx

0.751 0.75

0.7

0.749 0.748 0.747 0.746

0.6

0.745

0.499

0.4995

0.5

u/U 0

0.5

0.4

0.3 0 0.25 0.5 0.75 1

x/L

Figure: Invicid Burgers solution at= 2.8s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

34 / 42

Results

Time step barrier investigation

Implicit scheme - Solution


dt = 3 dx dt = 2 dx

0.751 0.75

0.7

0.749 0.748 0.747 0.746

0.6

0.745

0.499

0.4995

0.5

u/U 0

0.5

0.4

0.3 0 0.25 0.5 0.75 1

x/L

Figure: Invicid Burgers solution at= 3.0s.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

35 / 42

Results

Time step barrier investigation

Implicit scheme - dt = 2.0 dx

3 2.75 2.5

1.03 1.02 1.01 1 0.99 dt = 2 dx

Average Order

2.25 2 1.75 1.5 1.25

TV
0 0.25 0.5 0.75 1

0.98 0.97 0.96 0.95 0.94 0.93 0.92 1 2 3

x/L

Time (s)

(a) Temporal Order Preserved


Leonardo S. de B. Alves (UFF) Gabriel de Moraes (IME)

(b) TVD Preserved


CONEM 2012 36 / 42

Results

Time step barrier investigation

Behavior Beyond time step Barrier dt = 3.0 dx

3 2.75 2.5

1.03 1.02 1.01 1 0.99 dt = 3 dx

Average Order

2.25 2 1.75 1.5 1.25

TV
0 0.5 1 1.5 2

0.98 0.97 0.96 0.95 0.94 0.93 0.92 1 2 3

x/L

Time (s)

(c) Temporal Order Preserved


Leonardo S. de B. Alves (UFF) Gabriel de Moraes (IME)

(d) TVD violated


CONEM 2012 37 / 42

Explicit VS. Implicit

Explicit or Implicit, the question remains

For classical stability properties (such as linear stability or B-stability), implicit methods exist that are stable under arbitrarily large timesteps.... ...The question is, then, whether the allowable step-size can be large enough to oset the extra computational eort required in the implicit solution of the resulting system at each iteration.

Reference: Sigal Gottlieb, David I. Ketcheson, and Chi-Wang Shu. 2009. High Order Strong Stability Preserving Time Discretizations. J. Sci. Comput. 38, 3 (March 2009), 251-289. DOI=10.1007/s10915-008-9239-z http://dx.doi.org/10.1007/s10915-008-9239-z

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

38 / 42

Explicit VS. Implicit

Explicit VS. Implicit

100 10 10
-1

100 Explicit Implicit 10-1 10


-2

Explicit Implicit

-2

10-3

10-3

L2 error

10 10 10 10 10

-5

L2 error
-2 -1 0 1 2 3 4

10

-4

10

-4

10-5 10-6 10 10 10
-7

-6

-7

-8

-8

-9

-9

10

10

10

10

10

10

10

10

-2

10

-1

10

10

10

10

10

CPU Time (s)

CPU Time (s)

(e) t = 0.5s
Leonardo S. de B. Alves (UFF) Gabriel de Moraes (IME)

(f) t = 1.0s
CONEM 2012 39 / 42

Explicit VS. Implicit

Explicit VS. Implicit

100
x

100
x x x x x x x x x x

10 10

-1

Explicit Implicit

10-1 10
-2

Explicit Implicit

-2

10-3

10-3

L2 error

10 10 10 10 10

-5

L2 error
x
3 4

10

-4

10

-4

10-5 10-6 10 10 10
-7

-6

-7

-8

-8

-9

-9

10

-2

10

-1

10

10

10

10

10

10

-2

10

-1

10

10

10

10

10

CPU Time (s)

CPU Time (s)

(g) t = 2.0s
Leonardo S. de B. Alves (UFF) Gabriel de Moraes (IME)

(h) t = 3.0s
CONEM 2012 40 / 42

Conclusions and future work

Conclusions

All theoretical SSP barriers were numerically veried ; Comparisons are going to be extended to new schemes (higher-order in time and space) and test cases (incompressible and compressible ows) Implicit-Explicit (IMEX) schemes must be included in the analysis as well.

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

41 / 42

Thank you

Leonardo S. de B. Alves (UFF)

Gabriel de Moraes (IME)

CONEM 2012

42 / 42

You might also like