MATHEMATICAL MODELING : ANALYSIS AND APPLICATIONS

Lecture 5 : Numerical Methods to Compute the Eigen Values

Dr. Ameeya Kumar Nayak

Department of Mathematics

1
Contents:
• Physical Significance of Eigen Values.

• Why Numerical Methods?

• Different Numerical Methods to Compute Eigen Values.

• Power Method.

• LR Method.

2
Physical Significance of Eigen Values:
• In the previous lectures, we have talked a lot on eigen values, characteristic
equations and stability.
• By the time, you might have realized the importance of eigen values in
finding the stability of solution and hence analyzing the behavior of the
same.
• Do these eigen values describe something? Or are they mere some
numbers?
• Eigen values (and eigen vectors) are very well related with the
transformation of a system.
• The need of transformation of system is to make it easier for study and
analysis. When you transform any matrix, you need two things:
1. By what amount are you going to transform your system?
2. In which direction are you transforming it?
3
Physical Significance of Eigen Values:
• The answer to your first question is given by the eigen values and to
second is given by the associated eigen vectors.

• Suppose you have a 2 x 2 matrix and it’s eigen values are λ1 ,λ2 and
corresponding eigen vectors are v1 ,v2 respectively.

• λ1 gives the amount by which one is transforming matrix in direction v1 .

Similar is the case with λ2 in direction of v2 .
If 0 < λ < 1, means system is contracted in the direction of v.
If λ > 1, means system is expended in the direction of v.
If -1 < λ < 0, means system is contracted in the opposite direction of v.
If λ < -1, means system is expended in the opposite direction of v.
If λ = 0, means system is not at all transformed in the direction of v.

4
Need of Numerical Methods:
• Now we have seen the physical and mathematical importance of eigen
values (and eigen vectors). We have also seen that how tedious job is to
calculate eigen values when matrix is of high dimension.

• Also, when degree of characteristic equation is large (more than or equal to

3), it’s not even easy to find eigen values.
Jury’s test only help us to know whether all the eigen values lie in a unit
disk or not. But to know the exact solution, we need eigen values.

• When finding the solution of a system or an equation becomes difficult

task with analytical methods, then we go with numerical methods.

• Numerical methods are the computer based approach which try to

converge near the exact solution as far as possible.

5
Numerical Methods to find Eigen Values:
• There are many numerical algorithms to compute the eigen values. Some
of the widely used algorithms are:

1. Power Method
2. LR Method
3. QR Method
4. Householder Method etc.

• Each algorithm has some limitations and some advantages.

• In this lecture we will study Power method (which is used to find

numerically largest eigen value) and LR Method (which can be used to
any arbitrary matrix to find all the eigen values).

6
Power Method:
• The power method is a computer oriented method that finds the numerically largest
(dominant) eigen value of a given matrix and the associated eigen vector.

• In order to find sub-dominant, say second largest eigen value, the size of matrix can
be reduced by one and the power method can be applied on newly formed matrix.

• Power method is an iterative method (in-direct method) so we need some initial

assumption to start the process.

• After each iteration, we will compare the solution obtained with previous iteration
solution for convergence. If after some iterations, difference between two successive
iteration solution are less than a certain quantity, then the solution is said to be
converging.

• This certain quantity relates the accuracy of method for a particular matrix.

7
Example on Power Method:
Question: Find the largest eigen value of
.
Solution: Assume the initial approximation to eigen vector is X(0) = [1 0 0]T .

8
Example on Power Method:
• By repeating the iteration process for some more times, we will get:

• Clearly, λ(5) = λ(6) so the largest value of

eigen value is 3.41 (up-to accuracy at
second decimal place).

• The corresponding eigen vector is X(6) =

[0.76 -1 0.65]T .

• To find the smallest eigen value, one

can use the power method on [A]-1 .

9
LR Method:
• The LR Method was suggested by a Swiss Mathematician H. Rutishauser
in 1958.

• This method can be used for any arbitrary matrix to compute all the eigen
values. This method generally takes two steps:
Ak = Lk Rk,
Ak+1 = Rk Lk. …. 5.1
where k = 1, 2, 3, … and A1 = A. Lk is an unit lower triangular matrix while
Rk is an upper triangular matrix.

• For large k, matrix Ak+1 generally approaches to an upper triangular

matrix. (Note: The determinant of an upper or a lower triangular matrix is
simply the product of its diagonal entries.) So the eigen values are given by
the diagonal elements.

10
Example on LR Method:
Question: Find the eigen values of matrix [A] given as:

Solution: A = A1 = L1R1 . Assume and .

From first iteration, one can find the values of α’s as:
α2 = 3, α3 = 2, α1 = 1/3 and α4 = 10/3.
Now A2 = R1L1 = . . Again find the L2, R2 by the same method

as followed for L2, R2. This process will be repeated until we get the same matrix [A] in
the successive iterations. After performing this iteration 10 times you will get,

You can observe that A9 and A10 are almost identical and diagonal matrices. Hence the
eigen values are 5 and 2. (Verify with exact eigen values and with Jury’s test!)

11
Summary:
• Physical significance of eigen values.

• Need of numerical methods.

• Numerical methods for computing eigen values.

• Power Method.

• LR Method.

 Now, we will present the over-all summary of this week lectures.

12
 Mathematical Models Deterministic Methods to Solve Linear Discrete Models:
1. Analytical solution
 Homogeneous difference equations.
Stochastic
 Non-homogeneous difference equations.
2. Graphical solution
Discrete Time System Continuous time system  Cob-web graphs (valid only for first order
(involve difference equations) (involve ODE’s, PDE’s) difference equations)

Equilibrium Point:
2. Linear Cell Division Model  occurs when xn+1 = 0.
Linear Models (1st order difference equation)  known as critical point.
3. Linear Prey-Predator Model
Non Age- (system of two 1st order
Structured difference equations) Stability Analysis:
Models 1. Depends on parameter (for single equation)
2. Depends on eigen values (for system of
Age- equations).
Structured 1. Fibonacci’s Rabbit Model 3. Jury’s condition. (determine whether all the
Models (2nd order difference equation) eigen values are with in unit disk or not).
4. Human Population Model 4. Numerical methods to compute eigen values.
Non-Linear (system of five 1st order difference  Power method
equations; Leslie matrix)  LR method
Models
13
14