You are on page 1of 4

42 MATRIX AN AL YSI S

Example 1.16

Solution

For the symmetric matrix A considered in Example 1.9 write down the corresponding
T
orthogonal modal matrix K and show that K AM = , where is the spectral matrix.
From Example 1.9 the eigenvalues are 1 = 6, 2 = 3 and 3 = 1, with corresponding
normalized eigenvectors
1 = [1

2 = [0

0]T/5,

1]T,

3 = [2

0]T/5

The corresponding modal matrix is


 1--5-

2 1--5-

K = 2 1--5-

 1--5-

and, by matrix multiplication,


6
T
K AK = 0
0

1.6.2

0
3
0

0
0 =
1

The Jordan canonical form


If an n n matrix A does not possess a full set of linearly independent eigenvectors
then it cannot be reduced to diagonal form using the similarity transformation M 1AM.
In such a case, however, it is possible to reduce A to a Jordan canonical form, making
use of generalized eigenvectors.
As indicated in (1.11), if a matrix A has an eigenvalue i of algebraic multiplicity
mi and geometric multiplicity qi, with 1  qi  mi, then there are qi linearly independent
eigenvectors corresponding to i. Consequently, we need to generate mi qi generalized
eigenvectors in order to produce a full set. To obtain these, we rst obtain the qi linearly
independent eigenvectors by solving
(A iI )ei = 0
Then for each of these vectors we try to construct a generalized eigenvector e*
i such
that
(A iI )e*
i = ei
If the resulting vector e*i is linearly independent of all the eigenvectors (and generalized
eigenvectors) already found then it is a valid additional generalized eigenvector. If
further generalized eigenvectors corresponding to i are needed, we then repeat the
process using
(A iI )e**
= e*
i
i
and so on until sufcient vectors are found.

1.6 REDUCTION TO CANONICAL FORM

Example 1.17
Solution

43

Obtain a generalized eigenvector corresponding to the eigenvalue = 2 of Example 1.7.


For
A=

1
0
1

2
2
2

2
1
2

we found in Example 1.7 that corresponding to the eigenvalue i = 2 there was only
one linearly independent eigenvector
e1 = [2

1 0]T

and we need to nd a generalized eigenvector to produce a full set. To obtain the generalized eigenvector e*1 , we solve
(A 2I )e*1 = e1
that is, we solve
1

e*
11

= 1
e*
12

e*
13

2
0

At once, we have e*
13 = 1 and e*
11 = 2e*
12, and so
e*
1 = [2

1]T

Thus, by including generalized eigenvectors, we have a full set of eigenvectors for the
matrix A given by
e1 = [2

1 0]T,

e2 = [2

1]T,

e3 = [1

1]T

If we include the generalized eigenvectors, it is always possible to obtain for


an n n matrix A a modal matrix M with n linearly independent columns e1, e2,
. . . , en. Corresponding to (1.18), we have
AM = MJ
where J is called the Jordan form of A. Premultiplying by M 1 then gives
M 1AM = J

(1.22)

The process of reducing A to J is known as the reduction of A to its Jordan normal, or


canonical, form.
If A has p distinct eigenvalues then the matrix J is of the block-diagonal form
J = [J1

J2

...

Jp]

where each submatrix Ji (i = 1, 2, . . . , p) is associated with the corresponding eigenvalue i . The submatrix Ji will have i as its leading diagonal elements, with zeros
elsewhere except on the diagonal above the leading diagonal. On this diagonal the
entries will have the value 1 or 0, depending on the number of generalized eigenvectors

44 MATRIX AN AL YSI S
used and how they were generated. To illustrate this, suppose that A is a 7 7 matrix
with eigenvalues 1 = 1, 2 = 2 (occurring twice), 3 = 3 (occurring four times), and
suppose that the number of linearly independent eigenvectors generated in each case is

1 = 1, 1 eigenvector
2 = 2, 1 eigenvector
3 = 3, 2 eigenvectors
with one further generalized eigenvector having been determined for 2 = 2 and two
more for 3 = 3.
Corresponding to 1 = 1, the Jordan block J1 will be just [1], while that corresponding
to 2 = 2 will be
J2 =

2
0

1
2

Corresponding to 3 = 3, the Jordan block J3 can take one of the two forms

J 3,1 = 0
0

or

J 3,2 =

depending on how the generalized eigenvectors are generated. Corresponding to 3 = 3,


we had two linearly independent eigenvectors e3,1 and e3, 2. If both generalized eigenvectors are generated from one of these vectors then J3 will take the form J3,1, whereas
if one generalized eigenvector has been generated from each eigenvector then J3 will
take the form J3,2.
Example 1.18

Solution

Obtain the Jordan canonical form of the matrix A of Example 1.17, and show that
M 1AM = J where M is a modal matrix that includes generalized eigenvectors.
For
A =

1
0
1

2
2
2

2
1
2

from Example 1.17 we know that the eigenvalues of A are 1 = 2 (twice) and 3 = 1.
The eigenvector corresponding to 3 = 1 has been determined as e3 = [1 1 1]T in
Example 1.7 and corresponding to 1 = 2 we found one linearly independent eigenvector e1 = [2 1 0]T and a generalized eigenvector e*1 = [2 1 1]T. Thus the modal
matrix including this generalized eigenvector is
2
M = 1
0

2
1
1

1
1
1

1.6 REDUCTION TO CANONICAL FORM

45

and the corresponding Jordan canonical form is

J =

To check this result, we compute M 1 as

2
= 1
1

3
2
2

1
1
0

and, forming M 1AM, we obtain J as expected.

In MATLAB the command J=jordan(A) computes the Jordan form of A; including


the case when J is diagonal and all the eigenvectors of A are linearly independent.
The command
[M,J]=jordan(A)

also computes the similarity transformation or modal matrix M that may include
generalized eigenvectors.
Numerical calculation of the Jordan form is very sensitive to round-off errors, etc.
This makes it very difcult to compute the Jordan form reliably and almost any
change in A causes it to be diagonal.
For the matrix A in Example 1.18 the sequence of commands
A=[1 2 2; 0 2 1; -1 2 2];
[M,J]=jordan(A)

returns
-1 -2
2
M=-1 -1
1
1
0 -1
1
J= 0
0

0
2
0

0
1
2

which is equally acceptable to the solution given in Example 1.18. (This can be
checked by evaluating M 1AM.)
Using the Symbolic Math Toolbox in MATLAB the sequence of commands
A=[1 2 2; 0 2 1; -1 2 2];
AS=sym A
[M,J]=jordan(AS)

returns the same output as above. In practice, this sequence of commands is only
really effective when the elements of the matrix A are integers or ratios of small
integers.

You might also like