today’s class boundary value problems eigenvalue problems numerical methods lecture 18 prof. jinbo...

35
Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Upload: jeffry-mclaughlin

Post on 26-Dec-2015

220 views

Category:

Documents


4 download

TRANSCRIPT

Page 1: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Today’s class

• Boundary Value Problems• Eigenvalue Problems

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

1

Page 2: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Boundary Value Problems

• Auxiliary conditions• n-th order equation requires n conditions• Initial value conditions

• All n conditions are specified at the same value of the independent variable

• Boundary value conditions• The n conditions are specified at different values of

the independent variable

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

2

Page 3: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Initial value conditions

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

3

Page 4: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Boundary value conditions

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

4

Page 5: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Types of Boundary Conditions• Simple B.C.: The value of the unknown function

is given at the boundary.

• Neumann B.C.: The derivative of the unknown function is given at the boundary.

• Mixing B.C.: The combination of the unknown function’s value and derivative is given at the boundary.

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

5

Page 6: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Boundary Value Problems

• Shooting Method• Convert the boundary value problem into an

equivalent initial value problem by choosing values for all dependent variables at one boundary

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

6

Page 7: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Shooting Method Example

• Convert to first-order

• Guess a value for z(0)• Use RK method or any other method to

solve the ODE at y=10

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

7

Page 8: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Shooting Method Example• Using RK method with an initial guess of

z(0)=10 we get y(10)=168.3797• The specified condition was y(10)=200, so try

again• Using RK method with an initial guess of

z(0)=20 we get y(10)=285.8980• With two guesses and a linear ODE, we can

interpolate to find the correct value

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

8

Page 9: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Shooting Method Example

z(0)=10 z(0)=20

z(0)=12.69

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

9

Page 10: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Shooting Method

• If the ODE is non-linear, then you have to recast the problem as solving a series of root problems

• Think of the guessed values as a vector V

• That vector V will generate a y(yL) vector after solving the ODE

• The y(yL) vector should equal the boundary value conditions at yL

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

10

Page 11: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Shooting Method

• Since y is a function of V, you can create a new discrepancy function F that is dependent on V

• The problem then becomes to zero the discrepancy

• We don’t know F so we cannot use normal root-solving methods

• But we can approach it by iteratively looking at the finite difference derivatives

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

11

Page 12: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Finite-Difference Method (FDM)

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

12

Page 13: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Finite-Difference Method (FDM)

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

13

Page 14: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Finite-Difference Method (FDM)

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

14

Page 15: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Matrix Form of Equation System

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

15

Page 16: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

FDM Example

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

16

Page 17: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Eigenvalue Problems

• Eigenvalue problems are a special class of boundary-value problem

• Common in engineering systems with oscillating behavior• Springs• Vibrations

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

17

Page 18: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Physics of Eigenvalues & Eigenvectors

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

18

Page 19: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Physics of Eigenvalues

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

19

Page 20: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Physics of Eigenvalues

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

20

Page 21: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Steps of Solving Eigen-value Problems

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

21

Page 22: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Eigenvalue Example

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

22

Page 23: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Eigenvalue Example

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

23

Page 24: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Boundary value eigenvalue problems• Polynomial Method

• Power Method

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

24

Page 25: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Power Method

• The Power Method is an iterative procedure for determining the dominant (largest) eigenvalue of the matrix A and the corresponding eigenvector.

• Example

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

25

Page 26: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Example of Power Method (Cont.)

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

26

Page 27: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Example of Power Method (Cont.)

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

27

Page 28: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Example of Power Method (Cont.)

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

28

Page 29: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Example of Power Method (Cont.)

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

29

Page 30: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Inverse Power Method

• The Power Method can be modified to provide the smallest (lowest) (magnitude) eigenvalue and its eigenvector. The modified method is known as the inverse power method

• The smallest eigenvalue λ in the matrix A corresponds to 1/ λ the largest eigenvalue in the inverse matrix A-1.

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

30

Page 31: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

General I-P Method• The combination of Power Method and I-P

method can be use to determine all the eigenvalues and their related eigenvectors.

• Deflation Hotelling’s Method for symmetric matrices • First, normalize the largest eigenvector found

by the sum of the squares of the elements in the eigenvector

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

31

Page 32: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

General I-P Method

• Then create new A2 matrix

• The new A2 matrix has the same eigenvalues as A1 except that the largest eigenvalue has been replaced with 0

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

32

Page 33: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Hotelling’s Method

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

33

Page 34: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Hotelling’s Method

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

34

Page 35: Today’s class Boundary Value Problems Eigenvalue Problems Numerical Methods Lecture 18 Prof. Jinbo Bi CSE, UConn 1

Next class

• Curve Fitting

Numerical MethodsLecture 18

Prof. Jinbo BiCSE, UConn

35