1 Jordan blocks and Jordan form

A Jordan Block of size m and value λ is a matrix Jm (λ) having the value
λ repeated along the main diagonal, ones along the superdiagonal and zeros
everywhere else. For example:
 1 
−2 1 0 0
 0 −1 1
µ ¶
4 1 1 2 0 
J2 (4) = and J4 (− ) =  1
.
0 4 2  0 0 −2 1 
0 0 0 − 12
A Jordan Form is a block diagonal matrix consisting of several Jordan
blocks. For example, we can form a Jordan Form from two copies of J2 (4)
and one copy of J4 (− 12 ) as follows.
4 1 0 0 0 0 0 0
 
0 4 0 0 0 0 0 0 
 
  0 0 4 1 0 0 0 0 
J2 (4) 0 0 
0 0 0 4 0

0 0 0 
J= 0 J2 (4) 0 =
 .
0 0 0 0 − 12 1
 
1 0 0 
0 0 J4 (− 2 ) 
0 0 0 0 0 − 1 1

 2 0 

0 0 0 0 0 1
0 − 1 
2
0 0 0 0 0 0 0 − 12
Notice that the eigenvalues and eigenvectors of a matrix in Jordan Form
can be read off without your having to do any work.
1. The eigenvalues are along the main diagonal (this is true of any upper-
triangular matrix).
2. Eigenvectors can be found on the columns at the beginning of each
block.
For example, in the above form J, we have the eigenvalues λ = 1 with
multiplicity 4 and λ = 21 with multiplicity 4. Furthermore, there are two
eigenvectors associated with λ = 1, namely
1 0
   
0 0
   
0 1
   
0 0
v1 = 
0 , v2 = 0
  
   
0 0
   
0 0
0 0

1
and one eigenvector associated with the eigenvalue λ = − 12 , namely

0
 
0
 
0
 
0
v3 = 
0 .

 
0
 
0
1

Note that any two basis vectors for span(v1 , v2 ) can be chosen as the eigen-
vectors for λ = 1 and any non-zero scalar multiple of v3 can be chosen as
the eigenvector for λ = − 21 .

2 Any matrix can be put in Jordan Form

Theorem: Let A ∈ Rn×n and suppose that A has k ≤ n linearly in-
dependent eigenvectors v1 , ..., vk associated with (not necessarily distinct)
eigenvalues λ1 , ..., λk . Then A is similar to a Jordan form with blocks

Jm1 (λ1 ), ..., Jmk (λk )

where m1 + ..., mk = n.
Note that there may be several choices that need to be evaluated for
what the mi s are. These are discussed below. It will be useful to define the
following notions.

1. The algebraic multiplicity am(λ) of an eigenvalue λ is the number of

times it shows up in the characteristic polynomial |λI − A|.

2. The geometric multiplicity gm(λ) of an eigenvalue λ is the dimension

of the eigenspace associated with λ.

2.1 The geometric multiplicity equals algebraic multiplicity

In this case, there are as many blocks as eigenvectors for λ, and each has size
1. For example, take the identity matrix I ∈ n × n. There is one eigenvalue
λ = 1 and it has n eigenvectors (the standard basis e1 , .., en will do). So

2
am(λ) = gm(λ) = n and I is similar to (and equal to) the Jordan form
 
J1 (1) 0 ... 0
 0 J1 (1) ... 0 
J = . . .. 
 
 .. .. . 
0 0 ... J1 (1)

2.2 The geomestric multiplicity equals 1

In this case, there is one block for the eigenvalue and its size is mj = am(λj )
– that is, the block is the size of the algebraic multiplicity. For example, say
 
1 1 2 1
0 1 2 0
A= 0 0 1 1 .


0 0 0 1

This system has a signle eigenvalue λ = 1 with algebraic multiplicity 4

(which is clear since A is upper diagonal). Solving the equation

Av = λv

gives  
1
0
v= 
0
0
as an eigenvector and geometric multiplicity of λ = 1 is therefore 1. Thus,
A is similar to the matrix
 
1 1 0 0
0 1 1 0
J =0 0 1 1 .


0 0 0 1

To find the transformation M so that AM = M J we have to solve for M ,

which we do as follows. First, write out the columns of M in the similarity
statement.
 
1 1 0 0
0 1 1 0
A(m2 m2 m3 m4 ) = (m2 m2 m3 m4 )  0 0 1 1 .


0 0 0 1

3
Then note that

Am1 = m1
Am2 = m1 + m2
Am3 = m2 + m3
Am4 = m3 + m4 .

Clearly, the vector m1 is an eigenvector for λ = 1. The rest of the columns

are called generalized eigenvectors. To find them, we put m1 = v in the
second equation to get  
0
1
m2 = 
0


0
for example (any nonzero multiple of this will v work, although it will change
what follows). Then we put m2 in the third equation and solve for m3 to
get  
0
−1
m3 =   1 .

2
0
And finally, putting m3 into the last equation gives
 
0
0
m4 =  0 .

1
2

The resulting transformation from A to J is then

 
1 0 0 0
0 1 −1 21 
M = .
0 0 12 − 21 
1
0 0 0 2

2.3 The geometric multiplcity not 1 but is less than the al-
gebraic multiplicity
In this case, there is one Jordan block for each dimension of the eigenspace
(as usual). However, the sizes of those blocks are not determined by the

4
theorem above. For example, it may be that gm(λ) = 2 < 4 = am(λ).
Then there are either two blocks of size 2, or there are one two blocks of size
one and three.
Here is a more detailed example. Suppose that
 
1 1 1 −2
0 1 0 −1
A= 0 0 1 1  .


0 0 0 1

Once again, you can see by inspection that there is one eigenvalue λ = 1
with algebraic multiplicity 4. Finding the eigenvectors by solving Av = v
for v gives a two dimensional eigenspace spanned by the vectors
   
1 0
0 −1
v1 = 0 and v2 =  1 
  

0 0

for example (once again, any basis will do).

Now we have a choice to make. Either there are two blocks of size two,
or one block of size one and one block of size three. Suppose it is the latter,
then we have
 
1 0 0 0
0 1 1 0
A(m2 m2 m3 m4 ) = (m2 m2 m3 m4 )  0 0 1 1 .


0 0 0 1

and

Am1 = m1
Am2 = m2
Am3 = m2 + m3
Am4 = m3 + m4 .

The first two equations here are just eigenvalue equations, but we don’t know
which eigenvalue is which. Once again, we have choices. Either m2 = v1 or
m2 = v2 . Let’s try both:
1. Possibility 1: m2 = v1 . In this case, we get

Am3 = v1 + m3

5
 and
Am4 = m3 + m4 .
You can check (do this) that a solution to the first equation is m3 =
(0 1 0 0)T but that putting this solution into the second equation
yields no solution!

2. Possibility 2: m2 = v2 . In this case you can also check (do this) that
there is no solution for m3 and m4 .

Therefore, our initial guess that there was one block of size 1 and one block
of size three is wrong.
Now: You check that you can solve
 
1 1 0 0
0 1 0 0
A(m2 m2 m3 m4 ) = (m2 m2 m3 m4 )  0
.
0 1 1
0 0 0 1

There are more sophisticated wasy to find the Jordan form of a matrix
than just guessing. Try looking some up on the web. Unfortunately, they
take a bit more machinery than we have time to cover in this class, so for
now we will just use this method.