Linear System Theory

Exercise Session 5
Jordan Form

Exercise 1. (Jordan Form)

Let:

1 0 0 0 0 0
0
0 0 0 1 1
1 1 1 1 1 1
A= 0

0 0 1 0 0
0 1 0 0 2 0
0 0 0 0 0 1

Find the Jordan form of A.

Solution 1. (Jordan Form)

We first compute:

det(I A) = ( 1)6

Then, the algebraic multiplicity is 1 = 6 for the only eigenvalue 1 = 1. The geometric
multiplicity can be compute by:

Rank (Null (A 1 I)) = i

In this case, we have:

0 0 0 0 0 0
0 1 0 0 1 1

1 1 0 1 1 1
A 1 I = A I =

0 0 0 0 0 0

0 1 0 0 1 0
0 0 0 0 0 0

Then, Rank (Null (A 1 I)) = 3 = d1 . Therefore, there are 3 linear independent

eigenvectors, corresponding to 1 = 1. Moreover, that means that there are 3 Jordan
chains for the eigenvalue 1 , or else, that there are 3 Jordan blocks of dimension at least
1.
Then, we compute:

1

0 0 0 0 0 0


0
0 0 0 0 1

2

0 0 0 0 0 1
Rank Null (A 1 I) = Rank Null
0
= 5 = d2 > d1

0 0 0 0 0

0 0 0 0 0 1

0 0 0 0 0 0
This tells us that there are d2 d1 = 5 3 = 2 Jordan blocks of dimension at least 2.
Finally

0 0 0 0 0 0
0 0 0 0 0 0

3

0 0 0 0 0 0

Rank Null (A 1 I) = Rank Null
0
= 6 = d3 > d2

0 0 0 0 0

0 0 0 0 0 0

0 0 0 0 0 0
This tells us that there is d3 d2 = 6 5 = 1 Jordan block of dimension at least 3.
The previous procedure helped us to find the lengths i,j of the Jordan chains. Indeed,
we have one Jordan chain of length 3, one of length 2, and one of length 1.
We then want to find the matrix P 1 s.t. A = P 1 JP .
We start by looking for the linear independent eigenvectors by solving:

0 0 0 0 0 0
0 1 0 0 1 1

1 1 0 1 1 1 1,1,1
(A 1 I) v = (A I) v 1,1,1 =
0
v =0
0 0 0 0 0

0 1 0 0 1 0
0 0 0 0 0 0
The system of the equations can be reduced to the following:

x2 x5 + x6 = 0

x1 x2 + x4 x5 + x6 = 0

x2 + x 5 = 0

Where xi are the components of the vector v 1,1,1 . One solution of such a system of
equations is:
1
0

0
v 1,1,1 =
1

0
0
We then check if there is any other generalized eigenvector v 1,1,2 s.t. (A 1 I) v 1,1,2 =
v 1,1,1 :

2

0 0 0 0 0 0 1
0 1 0 0 1 1 0

1 1 0 1 1 1 1,1,2 0
(A 1 I) v 1,1,2 = (A I) v 1,1,2 =

v
=
1 = v
1,1,1

0 0 0 0 0 0
0 1 0 0 1 0 0
0 0 0 0 0 0 0

Just by looking at the first row, we conclude that there exists no v 1,1,2 that satisfies
(A 1 I) v 1,1,2 = v 1,1,1 . Therefore, v 1,1,1 is actually the eigenvector that originates the
chain of length 1.
We can then move to a second chain, originates by another eigenvector, linearly indepen-
dent from v 1,1,1 .
Again, by solving the system:

x2 x5 + x6 = 0

x1 x2 + x4 x5 + x6 = 0

x2 + x5 = 0

We have that:

0
1

0
v 1,2,1 =
0


1
0
is another solution, and v 1,1,1 and v 1,2,1 are linear independent. We then check if there is
any other generalized eigenvector v 1,2,2 s.t. (A 1 I) v 1,2,2 = v 1,2,1 :

0 0 0 0 0 0 0
0 1 0 0 1 1 1

1 1 0 1 1 1 1,2,2 0
(A 1 I) v 1,2,2 = (A I) v 1,2,2 =

v
=
0 =v
1,2,1

0 0 0 0 0 0
0 1 0 0 1 0 1
0 0 0 0 0 0 0

We then notice that:

1
0

0
v 1,2,2 =
0


1
0

3
is a solution. We then check if there is any other generalized eigenvector v 1,2,3 s.t.
(A 1 I) v 1,2,3 = v 1,2,2 :

0 0 0 0 0 0 1
0 1 0 0 1 1 0

1 1 0 1 1 1 1,2,3 0
(A 1 I) v 1,2,3 = (A I) v 1,2,3 =

v
=
0 =v
1,2,2

0 0 0 0 0 0
0 1 0 0 1 0 1
0 0 0 0 0 0 0

Again, just by looking at the first row, we conclude that there exists no v 1,2,3 .
Finally, we consider the last linear independent eigenvector:

0
1

1,3,1
1
v =
0


1
0
Since the geometric multiplicity is 3, we know that this must be the last chain. Moreover,
we know it must have length 3, since the others have lengths 1 and 2. Indeed, we find:

0
1

1,3,2
0
v =
0

0
0

which solves (A 1 I) v 1,3,2 = v 1,3,1 , and:

1
0

0
v 1,3,3 =
0

0
1
which solves (A 1 I) v 1,3,3 = v 1,3,2 .
We finally write the Jordan canonical form of A. We first build P 1 , by putting the
eigenvector by columns, in sequential order:

1 0 1 0 0 1
0 1 0 1 1 0

 1,1,1 1,2,1 1,2,2 1,3,1 1,3,2 1,3,3  0 0 0 1 0 0
P 1

= v kv kv kv kv kv =
1 0

0 0 0 0

0 1 1 1 0 0
0 0 0 0 0 1

4
and then, computing the inverse:

0 0 0 1 0 0
1
0 1 1 1 1

1 0 0 1 0 1
P =
0
0 1 0 0 0

1 1 0 1 1 1
0 0 0 0 0 1

We finally, build J make in evidence the different Jordan blocks:

1 0 0 0 0 0
0 1 1 0 0 0

0 0 1 0 0 0
J = 0 0

0 1 1 0
0 0 0 0 1 1
0 0 0 0 0 1

noticing that the dimensions correspond to the lengths of the Jordan chains.
We can finally check that A = P 1 JP .

Remark 1 Note that, even before computing P 1 we could have determined:

1 0 0 0 0 0
0 1 1 0 0 0

0 0 1 0 0 0
J =
0

0 0 1 1 0

0 0 0 0 1 1
0 0 0 0 0 1

just by making use of the knowledge of the chain lengths. Indeed, we knew there is a chain
of length 1, one of length 2, and one of length 3.

5
General Algorithm (Jordan Form) Given a matrix A Rnn , a general algorithm to
find its Jordan decomposition is the following:

1. Find the characteristic polynomial:

PA (s) = ki=1 (s i )i

Where i are the distinct eigenvalues, and i is their algebraic multiplicity.

2. For each distinct i :

(a) compute the geometric multiplicity of i :

i = Rank(Null(A i I))

This gives the number of linear independent eigenvectors corresponding to i .

It holds: 1 i i . We call this i = di,1
(b) For r = 2, 3, . . . , l compute:

di,r = Rank(Null(A i I)r )

until di,l = i . Then, it holds:

0 < di,1 < ... < di,l = i

(c) The sequence di,1 , . . . , di,l determines how many Jordan blocks corresponding
to i we have, and their respective size. The way to interpret them is the
following:
i. di,1 = i tells us how many blocks of size at least 1 there are.
ii. di,2 di,1 tells us how many blocks of size at least 2 there are.
iii. di,3 di,2 tells us how many blocks of size at least 3 there are.
iv. ...
v. di,l di,l1 tells us how many blocks of size at least l there are.
(d) With this, you get to know the lengths of the chains i,j originated by the first
i linear independent eigenvector. Therefore, you can build Ji,j .

3. At this stage, you could already build the matrix J, by composing the different Ji,j .

Remark 2 Note that you can have different versions of J, depending on the order
you put the blocks Ji,j in the diagonal.

4. Then, again for each distinct i :

(a) Compute the i linear independent eigenvectors by solving:

(A i I) v i,j,1 = 0, j = 1, . . . , i

(b) For each j = 1, . . . , i :

6
 i. Compute the generalized eigenvectors by solving:
(A i I) v i,j,k = v i,j,k1 , k = 2, . . . , i,j
until it is solvable.
(c) Verify that the chain lengths are the same as the ones you expected from the
previous steps.
5. Build the P 1 matrix by placing the linear independent eigenvector and the gener-
alized eigenvectors in the following sequence:
Pi
n columns= ki=1 i = ki=1 ( j=1 i,j )
P P

z }| Pk
{
1 columns=Pj=1 1
1,j =algebraic multiplicity of 1 k columns==
j=1 k,,j =algebraic multiplicity of k

z }| { z }| {

1,1 columns 1,1 columns k,1 columns k,k columns
1

P = z
1,1,1 }| { z }| { z }| { z }| {
1,1,1,1 1,1 ,1 1,1 ,1,1 k,1,1 k,1,k,1 k,k ,1 1,k ,k,k
v k. . .kv k...kv k. . .kv k...kv k. . .kv k...kv k. . .kv
| {z } | {z } | {z } | {z }
1st chain ith chain 1st chain kth chain


| {z } | {z }
1st distinct eigenvalue1 kth distinct eigenvaluek

6. P 1 is invertible, therefore you can compute P .

7. Build the J matrix with the Ji,j blocks, accordingly to the sequence of eigenvectors
in P 1 :

J1,1 0 0 0 0 0
0 J1,2 0 0 0 0
...

0 0 0 0 0

0 0 0 J1,i 0 0

. .. .. .. ..
J =
.
. . . . .
0
0 0 0 0 Jk,1 0 0
0 0 0 0 0 0 Jk,2 0 0

. . .
.

0 0 0 0 0 0 0 . .
0 0 0 0 0 0 0 0 Jk,k

Where:

i 1 0

0
0 i 1 0 0

... ..
Ji,j = 0 0 1 .

. .
.. ..

i 1
0 0 0 i

and dim(Ji,j ) = i,j .

8. Finally, you can write A = P 1 JP