19
MATH 175: NUMERICAL ANALYSIS II Lecturer: Jomar Fajardo Rabajante 2 nd Sem AY 2012-2013 IMSP, UPLB

MATH 175: NUMERICAL ANALYSIS II

  • Upload
    sitara

  • View
    43

  • Download
    0

Embed Size (px)

DESCRIPTION

MATH 175: NUMERICAL ANALYSIS II. Lecturer: Jomar Fajardo Rabajante 2 nd Sem AY 2012-2013 IMSP, UPLB. STIFF ODEs. - PowerPoint PPT Presentation

Citation preview

Page 1: MATH 175: NUMERICAL ANALYSIS II

MATH 175: NUMERICAL ANALYSIS II

Lecturer: Jomar Fajardo Rabajante2nd Sem AY 2012-2013

IMSP, UPLB

Page 2: MATH 175: NUMERICAL ANALYSIS II

STIFF ODEs

Remember that in our discussion from Variable Step-Size Methods, I told you that if h* becomes very small (say 10^-10) then halt the process. This is an indicator that your IVP might be stiff!

There is no precise definition of stiffness but one instance that stiffness may occur is when solution to your IVP changes values drastically (slopes are too steep).

Page 3: MATH 175: NUMERICAL ANALYSIS II
Page 4: MATH 175: NUMERICAL ANALYSIS II

STIFF ODEsThere are methods that for some h>stab, the method

is unstable (where stab is the value of h at the boundary between stability and instability).

Example: Given the following system of ODE, we need to choose h<0.2 in Euler’s method, so that we can get stable solutions. (Note: h<0.2 only gives us stability and not yet our desired accuracy.)

1)0( '1)0( 10'

yyyxyxx

Page 5: MATH 175: NUMERICAL ANALYSIS II

STIFF ODEs

But sometimes we do not want our step size to be so small and sacrifice our computing time just to attain stability (especially when our time period considered is to wide).

We want method that has larger stability region (graphed in a complex plane).

Page 6: MATH 175: NUMERICAL ANALYSIS II

Stability Region for Euler Method

Page 7: MATH 175: NUMERICAL ANALYSIS II

Stability Region for RK 3

Page 8: MATH 175: NUMERICAL ANALYSIS II

Stability Region for RK 4

Page 9: MATH 175: NUMERICAL ANALYSIS II

Stability Region for Implicit Trapezoidal Method

Page 10: MATH 175: NUMERICAL ANALYSIS II

Stability Region for Backward Euler

The shaded area is the stability region.

Page 11: MATH 175: NUMERICAL ANALYSIS II

STIFF ODEsNote: Stiffness is a property of the ODE not of the

numerical methods we have discussed (remember that all our discussed methods are consistent and convergent as h0). It is just for stiff ODEs, we cannot endure having very small h because of computational concerns.

However, note that for Implicit Trapezoidal and Backward Euler’s methods, the left-hand side of their complex planes are shaded. These mean that they are A-stable methods, i.e. for any value of h the approximates will converge to the long-term behavior of the solution.

Page 12: MATH 175: NUMERICAL ANALYSIS II

Euler’s method requires tiny step

Note: this is not the solution to our previous example.

Page 13: MATH 175: NUMERICAL ANALYSIS II

Euler’s method with too large a step

Note: this is not the solution to our previous example.

Page 14: MATH 175: NUMERICAL ANALYSIS II

Large steps OK with Backward Euler’s

Note: this is not the solution to our previous example.

Page 15: MATH 175: NUMERICAL ANALYSIS II

Very large steps OK, too for Backward Euler’s

Note: this is not the solution to our previous example.

Page 16: MATH 175: NUMERICAL ANALYSIS II

Numerical Methods (Implicit) for Stiff ODEs

One way to solve stiff ODEs is to bring information from the right side of the interval [ti , ti+h] instead of relying solely on the left side.• We have already seen the Implicit Trapezoidal method.

Now, we will see Backward Euler’s Method (Implicit Euler’s Method).

• Implicit R-K methods are known to be A-stable. • Berkeley Madonna uses Rosenbrock Method.

Note: Implicit methods can be made adaptive.

Page 17: MATH 175: NUMERICAL ANALYSIS II

BACKWARD EULER’S METHOD

),( 111 iiii wthfww

Example 1: Given the following IVP:

5.0)0()1(10'

yyy

10110

)1(105.0

1

11

0

hhww

whwww

ii

iii

Page 18: MATH 175: NUMERICAL ANALYSIS II

BACKWARD EULER’S METHOD

Notice that we need to do some root-finding to get the approximates.

Example 2:

Then solve for wi+1 using Chapter 1, use wi as initial guess.

098

98'

13

12

11

32

iiiii wwwwhw

yyyy

Page 19: MATH 175: NUMERICAL ANALYSIS II

BDFActually, Backward Euler’s Method is a 1st

order Backward Differentiation Formula (BDF). The 2nd order BDF is

),(32

31

34

2212 iiiii yxhfwww