280 CHAPTER 5 Eigenvalues and Eigenvectors
Application to Dynamical Systems
Eigenvalues and eigenvectors hold the key to the discrete evolution of a dynamical
system, as mentioned in the chapter introduction.
:95
:03
EXAMPLE 5 Let A D . Analyze the long-term behavior of the dynam-
:05
:97
:6
ical system defined by xk C1 D Axk .k D 0; 1; 2; : : :/, with x0 D .
:4
SOLUTION The first step is to find the eigenvalues of A and a basis for each eigenspace.
The characteristic equation for A is
:95 :03
0 D det D .:95 /.:97 / .:03/.:05/
:05 :97
D 2 1:92 C :92
By the quadratic formula
p p
1:92 ˙ .1:92/2 4.:92/ 1:92 ˙ :0064
D D
2 2
1:92 ˙ :08
D D 1 or :92
2
It is readily checked that eigenvectors corresponding to D 1 and D :92 are multiples
of
3 1
v1 D and v2 D
5 1
respectively.
The next step is to write the given x0 in terms of v1 and v2 . This can be done because
fv1 ; v2 g is obviously a basis for R2 . (Why?) So there exist weights c1 and c2 such that
c
x0 D c1 v1 C c2 v2 D Πv1 v2 1 (3)
c2
In fact,
1
c1 1 3 :601
D Πv1 v2 x0 D
c2 5 :401
1 1 1 :60 :125
D D (4)
8 5 3 :40 :225
Because v1 and v2 in (3) are eigenvectors of A, with Av1 D v1 and Av2 D :92v2 , we
easily compute each xk :
x1 D Ax0 D c1 Av1 C c2 Av2 Using linearity of x 7! Ax
D c1 v1 C c2 .:92/v2 v1 and v2 are eigenvectors.
x2 D Ax1 D c1 Av1 C c2 .:92/Av2
D c1 v1 C c2 .:92/2 v2
and so on. In general,
xk D c1 v1 C c2 .:92/k v2 .k D 0; 1; 2; : : :/
Using c1 and c2 from (4),
3 1
xk D :125 C :225.:92/k .k D 0; 1; 2; : : :/ (5)
5 1
SECOND REVISED PAGES
5.2 The Characteristic Equation 281
This explicit formula for xk gives the solution ofthe difference
equation xk C1 D Axk .
:375
As k ! 1, .:92/k tends to zero and xk tends to D :125v1 .
:625
The calculations in Example 5 have an interesting application to a Markov chain
discussed in Section 4.9. Those who read that section may recognize that matrix A
in Example 5 above is the same as the migration matrix M in Section 4.9, x0 is the
initial population distribution between city and suburbs, and xk represents the population
distribution after k years.
Theorem 18 in Section 4.9 stated that for a matrix such as A, the sequence xk tends
to a steady-state vector. Now we know why the xk behave this way, at least for the
migration matrix. The steady-state vector is :125v1 , a multiple of the eigenvector v1 ,
and formula (5) for xk shows precisely why xk ! :125v1 .
NUMERICAL NOTES
1. Computer software such as Mathematica and Maple can use symbolic calcu-
lations to find the characteristic polynomial of a moderate-sized matrix. But
there is no formula or finite algorithm to solve the characteristic equation of a
general n n matrix for n 5.
2. The best numerical methods for finding eigenvalues avoid the characteristic
polynomial entirely. In fact, MATLAB finds the characteristic polynomial
of a matrix A by first computing the eigenvalues 1 ; : : : ; n of A and then
expanding the product . 1 /. 2 / . n /.
3. Several common algorithms for estimating the eigenvalues of a matrix A
are based on Theorem 4. The powerful QR algorithm is discussed in the
exercises. Another technique, called Jacobi’s method, works when A D AT
and computes a sequence of matrices of the form
A1 D A and Ak C1 D Pk 1 Ak Pk .k D 1; 2; : : :/
Each matrix in the sequence is similar to A and so has the same eigenvalues
as A. The nondiagonal entries of Ak C1 tend to zero as k increases, and the
diagonal entries tend to approach the eigenvalues of A.
4. Other methods of estimating eigenvalues are discussed in Section 5.8.
PRACTICE PROBLEM
1 4
Find the characteristic equation and eigenvalues of A D .
4 2
5.2 EXERCISES
Find the characteristic polynomial and the eigenvalues of the 2 1 3 4
5. 6.
matrices in Exercises 1–8. 1 4 4 8
2 7 5 3 5 3 7 2
1. 2. 7. 8.
7 2 3 5 4 4 2 3
Exercises 9–14 require techniques from Section 3.1. Find the
3 2 5 3 characteristic polynomial of each matrix, using either a cofactor
3. 4. expansion or the special formula for 3 3 determinants described
1 1 4 3
SECOND REVISED PAGES
