Vous êtes sur la page 1sur 9

Generalized Eigenvectors and Jordan Form

We have seen that an n n matrix A is diagonalizable precisely when the dimensions of its
eigenspaces sum to n. So if A is not diagonalizable, there is at least one eigenvalue with a
geometric multiplicity (dimension of its eigenspace) which is strictly less than its algebraic
multiplicity. In this handout, we will discuss how one can make up for this deciency of
eigenvectors by nding what are called generalized eigenvectors, which can in turn be used
to nd the Jordan form of the matrix A.
First consider the following non-diagonalizable system.
Example 1. 3 The matrix
A =
_
3 1
0 3
_
has characteristic polynomial ( 3)
2
, so it has only one eigenvalue = 3, and the cor-
responding eigenspace is E
3
= span
__
1
0
__
. Since dim(E
3
) = 1 < 3, the matrix A is not
diagonalizable. Nevertheless, it is still possible to solve the system
dy
dt
= Ay
without much diculty. Writing out the two equations
dy
1
dt
= 3y
1
+ y
2
dy
2
dt
= 3y
2
we see that the second equation has general solution y
2
= c
1
e
3t
. Plugging this into the rst
equation gives
dy
1
dt
= 3y
1
+ c
1
e
3t
This is a rst order linear equation that can be solved by the method of integrating factors.
Its solution is y
1
= c
1
te
3t
+ c
2
e
3t
. Thus the general solution of the system is
y =
_
y
1
y
2
_
=
_
c
1
te
3t
+ c
2
e
3t
c
2
e
3t
_
= c
1
e
3t
_
t
_
1
0
_
+
_
0
1
__
+ c
2
e
3t
_
1
0
_
3
The matrix A in the previous example is said to be in Jordan form.
Denition 1. A 2 2 matrix of the form
J =
_
1
0
_
for some number is called a Jordan form matrix.
1
The following theorem generalizes the calculations of the previous example.
Theorem 1. The general solution of the system
dy
dt
= Jy, J =
_
1
0
_
is
y = c
1
e
t
_
t
_
1
0
_
+
_
0
1
__
+ c
2
e
t
_
1
0
_
In general, suppose A is a 22 matrix with a single repeated eigenvalue with dim(E

) =
1. Then, as we have seen before, making the change of variables y = Cx transforms the
system
dy
dt
= Ay
into the system
dx
dt
= C
1
ACx.
We now want to nd a matrix C such that the coecient matrix C
1
AC for the new system
is a Jordan form matrix J. That is, we want AC = CJ. Writing
C =
_
_
| |
v
1
v
2
| |
_
_
, J =
_
1
0
_
,
we have
AC =
_
_
| |
Av
1
Av
2
| |
_
_
, CJ =
_
_
| |
v
1
v
2
+v
1
| |
_
_
Therefore the columns of C must satisfy
Av
1
= v
1
Av
2
= v
2
+v
1
Thus the vector v
1
is an eigenvector with eigenvalue . Rewriting these equations
(A I)v
1
= 0
(A I)v
2
= v
1
it follows that (A I)
2
v
2
= (A I)v
1
= 0. Thus the vector v
2
must be in ker(A I)
2
.
2
Denition 2. A nonzero vector v which satises (AI)
p
v = 0 for some positive integer
p is called a generalized eigenvector of A with eigenvalue .
The vectors v
1
and v
2
form a generalized eigenvector chain, as the following diagram
illustrates:
v
2

AI
v
1

AI
0
Therefore, to nd the columns of the matrix C that puts A in Jordan form, we must nd
a chain of generalized eigenvectors, as follows:
Find a nonzero vector v
2
in ker(A I)
2
that is not in ker(A I).
Set v
1
= (A I)v
2
.
Having found such vectors, we can then write down the solution of the system
dy
dt
= Ay.
From above, we know that the solution of the system
dx
dt
= Jx is
x = c
1
e
t
_
t
_
1
0
_
+
_
0
1
__
+ c
2
e
t
_
1
0
_
so multiplying this by C we get
y = Cx = c
1
e
t
(tv
1
+v
2
) + c
2
e
t
v
1
.
In summary, we have the following result.
Theorem 2. Let A be a 22 matrix and suppose v
2
v
1
0 is a chain of generalized
eigenvectors of A with eigenvalue . Then the general solution of the system
dy
dt
= Ay is
y(t) = c
1
e
t
(tv
1
+v
2
) + c
2
e
t
v
1
.
Example 2. 3 Let
A =
_
2 1
1 4
_
The characteristic polynomial of A is
2
6 +9 = ( 3)
2
, so = 3 is the only eigenvalue
of A. Next, we compute
A 3I =
_
1 1
1 1
_
, (A 3I)
2
=
_
0 0
0 0
_
Now we choose v
2
to be any vector in ker(A 3I)
2
that is not in ker(A 3I). One such
vector is v
2
=
_
0
1
_
. With this choice, we than have v
1
= (A 3I)v
2
=
_
1
1
_
. The solution of
the system
dy
dt
= Ay is therefore
y = c
1
e
3t
_
t
_
1
1
_
+
_
0
1
__
+ c
2
e
3t
_
1
1
_
.
3
3
For larger systems it turns out that any chain of generalized eigenvectors of the coecient
matrix can be used to write down a linearly independent set of solutions.
Theorem 3. Suppose {v
1
, v
2
, . . . , v
k
} is a chain of generalized eigenvectors of A with
eigenvalue . That is,
v
k

AI
v
k1

AI
v
k2

AI

AI
v
2

AI
v
1

AI
0.
Then
y
1
(t) = e
t
v
1
y
2
(t) = e
t
(v
2
+ tv
1
)
y
3
(t) = e
t
_
v
3
+ tv
2
+
t
2
2!
v
1
_
.
.
.
y
k
(t) = e
t
_
v
k
+ tv
k1
+
t
2
2!
v
k2
+ +
t
k2
(k 2)!
v
2
+
t
k1
(k 1)!
v
1
_
is a linearly independent set of solutions of
dy
dt
= Ay.
Proof. We show that y
k
is a solution. The proof that the others are solutions is similar.
Dierentiating gives
dy
k
dt
= e
t
_
(v
k
+v
k1
) + t(v
k1
+v
k2
) +
t
2
2!
(v
k2
+v
k3
) +
+
t
k2
(k 2)!
(v
2
+v
1
) +
t
k1
(k 1)!
(v
1
)
_
.
Multiplying by A gives
Ay
k
= e
t
_
(Av
k
) + t(Av
k1
) +
t
2
2!
(Av
k2
) + +
t
k2
(k 2)!
(Av
2
) +
t
k1
(k 1)!
(Av
1
)
_
.
The two expressions above are equal, since Av
1
= v
1
, Av
2
= v
2
+v
1
, and so on.
Next we show that the chain {v
1
, v
2
, . . . , v
k
} is necessarily linearly independent. Suppose
c
1
v
1
+ c
2
v
2
+ + c
k
v
k
= 0. (1)
The multiplying by (A I)
k1
sends v
1
through v
k
to zero, and v
k
to v
1
, so we are left
with c
k
v
1
= 0. Since v
1
is nonzero, this implies c
k
= 0. Similar reasoning shows that
the remaining coecients must also be zero. Thus the chain of generalized eigenvectors is
linearly independent. Now suppose
c
1
y
1
(t) + c
2
y
2
(t) + + c
n
y
n
(t) = 0
for all t. Then since y
j
(0) = v
j
, at t = 0 this equation is precisely equation (1) above, which
we just showed implies c
1
= c
2
= = c
k
= 0.
4
The next theorem guarantees that we can always nd enough solutions of this form to
generate a fundamental set of solutions.
Theorem 4. Let A be an nn matrix, and suppose is an eigenvalue of A with algebraic
multiplicity m. Then there is some integer p m such that dim(ker(A I)
p
) = m.
Example 3. 3 Let
A =
_
_
5 1 4
4 3 5
3 1 2
_
_
The characteristic polynomial of A is p
A
() = ( 2)
3
. Since
A 2I =
_
_
3 1 4
4 1 5
3 1 4
_
_

rref
_
_
1 0 1
0 1 1
0 0 0
_
_
,
the eigenspace E
2
= ker(A 2I) has dimension 1. Next, since
(A 2I)
2
=
_
_
1 0 1
1 0 1
1 0 1
_
_

rref
_
_
1 0 1
0 0 0
0 0 0
_
_
,
the space ker(A 2I)
2
has dimension 2. Finally, since
(A 2I)
3
=
_
_
0 0 0
0 0 0
0 0 0
_
_
,
the space ker(A 2I)
3
has dimension 3, which matches the algebraic multiplicity of the
eigenvalue = 2.
We can now form a chain of 3 generalized eigenvectors by choosing a vector v
3
in ker(A
2I)
3
and dening v
2
= (A 2I)v
3
and v
1
= (A 2I)v
2
= (A 2I)
2
v
3
. To ensure that v
2
and v
1
are both non-zero, we need v
3
to not be in ker(A 2I)
2
(which in turn implies that
v
3
is not in ker(A 2I)). Since ker(A 2I)
3
= R
3
, we can choose v
3
to be any vector not
in ker(A 2I)
2
. Since the rst column of (A 2I)
2
is non-zero, we may choose
v
3
=
_
_
1
0
0
_
_
= v
2
= (A 2I)v
3
=
_
_
3
4
3
_
_
= v
1
= (A 2I)v
2
=
_
_
1
1
1
_
_
.
5
Then
y
1
(t) = e
2t
_
_
1
1
1
_
_
y
2
(t) = e
2t
_
_
t
_
_
1
1
1
_
_
+
_
_
3
4
3
_
_
_
_
y
3
(t) = e
2t
_
_
t
2
2!
_
_
1
1
1
_
_
+ t
_
_
3
4
3
_
_
+
_
_
1
0
0
_
_
_
_
is a fundamental set of solutions of
dy
dt
= Ay. 3
There may in general be more than one chain of generalized eigenvectors corresponding
to a given eigenvalue. Since the last vector in each chain is an eigenvector, the number of
chains corresponding to an eigenvalue is equal to the dimension of the eigenspace E

.
Example 4. 3 Let
A =
_
_
4 1 1
2 1 2
1 1 4
_
_
.
The characteristic polynomial of A is p
A
() = ( 3)
3
. Since
A 3I =
_
_
1 1 1
2 2 2
1 1 1
_
_

rref
_
_
1 1 1
0 0 0
0 0 0
_
_
,
the eigenspace E
3
= ker(A 3I) has dimension 2, so there will be two chains. Next, since
(A 3I)
2
=
_
_
0 0 0
0 0 0
0 0 0
_
_
,
the space ker(A 3I)
2
has dimension 3, which matches the algebraic multiplicity of = 3.
Thus one of the chains will have length 2, so the other must have length 1.
We now form a chain of 2 generalized eigenvectors by choosing v
2
in ker(A 3I)
2
such
that v
2
is not in ker(A 3I). Since every vector is in ker(A 3I)
2
and the rst column of
A 3I is non-zero,we may again choose
v
2
=
_
_
1
0
0
_
_
= v
1
= (A 3I)v
2
=
_
_
1
2
1
_
_
To form a basis for R
3
, we now need one additional chain of one generalized eigenvector.
This vector must be an eigenvector independent from v
1
. Since
E
3
= span
_
_
_
_
1
1
0
_
_
,
_
_
1
0
1
_
_
_
_
,
6
and neither of these spanning vectors is itself a scalar multiple of v
1
, we may choose either
one of them. So let
w
1
=
_
_
1
1
0
_
_
Then we have the two chains
v
2
v
1
0
w
1
0.
The rst chain generates the two solutions
y
1
(t) = e
3t
_
_
1
2
1
_
_
y
2
(t) = e
3t
_
_
t
_
_
1
2
1
_
_
+
_
_
1
0
0
_
_
_
_
and the second chain generates a third solution
y
3
(t) = e
3t
_
_
1
1
0
_
_
and together y
1
, y
2
and y
3
form a fundamental set of solutions of
dy
dt
= Ay. 3
Example 5. 3 Let
A =
_

_
5 1 3 2
0 5 0 3
0 0 5 1
0 0 0 5
_

_
The characteristic polynomial of A is ( 5)
4
. Since
A 5I =
_

_
0 1 3 2
0 0 0 3
0 0 0 1
0 0 0 0
_

_

rref
_

_
0 1 3 0
0 0 0 1
0 0 0 0
0 0 0 0
_

_
,
the eigenspace E
5
= ker(A 5I) has dimension 2, and there are two chains. Since
(A 5I)
2
=
_

_
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
_

_
the dimension of the space ker(A 5I)
2
is 4, matching the algebraic multiplicity of = 5.
7
To form a chain of length 2, we rst choose a vector v
2
in ker(A 5I)
2
which is not in
ker(A 5I). Let
v
2
=
_

_
0
1
0
0
_

_
= v
1
= (A 5I)v
2
=
_

_
1
0
0
0
_

_
.
We now need one more chain of length 2. To nd a second chain, we need to choose another
vector w
2
in ker(A5I)
2
which is not in ker(A5I), in such a way that w
2
is independent
of v
2
and w
1
= (A 5I)w
2
is independent of v
1
. One such choice is
w
2
=
_

_
0
0
0
1
_

_
= w
1
= (A 5I)w
2
=
_

_
2
3
1
0
_

_
.
Thus
y
1
(t) = e
5t
_

_
1
0
0
0
_

_
y
2
(t) = e
5t
_
_
_
_
t
_

_
1
0
0
0
_

_
+
_

_
0
1
0
0
_

_
_
_
_
_
y
3
(t) = e
5t
_

_
2
3
1
0
_

_
y
4
(t) = e
5t
_
_
_
_
t
_

_
2
3
1
0
_

_
+
_

_
0
0
0
1
_

_
_
_
_
_
is a fundamental set of solutions of
dy
dt
= Ay. 3
Jordan Form for n n Matrices
Denition 3. A matrix of the form
_

_
1
1
.
.
.
.
.
.
.
.
.
1

_
consisting of in each entry along the main diagonal, 1 in each entry directly above the
main diagonal, and 0 elsewhere, is called an elementary Jordan block. A matrix of the
form
_

_
J
1
J
2
.
.
.
J
k
_

_
8
where each J
i
is an elementary Jordan block (with possibly dierent values of ) is called a
Jordan form matrix.
Example 6. 3 The following are examples of Jordan form matrices:
_
_
2 1 0
0 2 0
0 0 3
_
_
,
_
_
2 1 0
0 2 1
0 0 2
_
_
,
_
_
3 1 0
0 3 0
0 0 3
_
_
,
_
_
2 0 0
0 1 0
0 0 3
_
_
3
Theorem 5. For any nn matrix A, there is a matrix C and a Jordan form matrix J such
that C
1
AC = J.
The Jordan matrix J is determined by the number and length of the generalized eigen-
vector chains for A.
Example 7. 3 Let A be the matrix in Example 3. Since
Av
1
= 2v
1
Av
2
= v
1
+ 2v
2
Av
2
= v
2
+ 2v
3
,
letting
C =
_
_
| | |
v
1
v
2
v
3
| | |
_
_
implies
AC =
_
_
| | |
2v
1
v
1
+ 2v
2
v
2
+ 2v
3
| | |
_
_
=
_
_
| | |
v
1
v
2
v
3
| | |
_
_
_
_
2 1 0
0 2 1
0 0 2
_
_
= CJ,
so C
1
AC = J consists of a single elementary Jordan block. 3
Example 8. 3 Let A be the matrix in Example 4. Since
Av
1
= 3v
1
Av
2
= v
1
+ 3v
2
Aw
1
= 3w
1
,
letting
C =
_
_
| | |
v
1
v
2
w
1
| | |
_
_
implies
AC =
_
_
| | |
3v
1
v
1
+ 3v
2
w
1
| | |
_
_
=
_
_
| | |
v
1
v
2
w
1
| | |
_
_
_
_
3 1 0
0 3 0
0 0 3
_
_
= CJ,
so C
1
AC = J consists of 2 elementary Jordan blocks, one 2 2 and one 1 1. 3
9