Notes for Chapter 6: The Eigenvalue Problem and Linear Systems

5.1 Eigenvalues and Eigenvectors

Given an n × n matrix A a number λ is said to be an eigenvalue of A if there exists
a nonzero vector V satisfying the linear system AV = λV . In case the pair (λ, V )
exist then we the vector V is called a eigenvector corresponding to the eigenvalue λ.

The linear system AV = λV can be written as the homogeneous linear system (λIn − A)V = 0.
We have learned several characterizations that allow us to conclude when a homogeneous system of
equations will have a nonzero solution, e.g. this will be the case if the coefficient matrix is singular.
The characterization we will use here is that the system will have a nonzero solution if and only if
the determinant of the coefficient matrix is zero.

Given an n × n matrix A a number λ is an eigenvalue if and only if it is a root of the

polynomial equation det(λIn − A) = 0.

Notice that det(λIn − A) = 0 is a polynomial, called the Characteristic Polynomial, of degree n

whose leading term is λn :

det(λIn − A) = λn + an−1 λn−1 + · · · + a1 λ + a0 = 0.

The fundamental theorem of algebra (see any college algebra book) tells us that every polynomial
(with real coefficients) of degree n has exactly n roots in the complex number system. Now the
main complication that arises is that some of the roots may be repeated and this can lead to serious
technical difficulties when it comes to finding the associated eigenvectors.

We will not be able to fully address this issue which is usually handled completely in a graduate
version of this class.

We will completely analyze the special case of 2 × 2 matrices and you will be asked to find eigen-
values and eigenvectors for a few higher order cases.
 
a11 a12
Let us consider the special case of a 2 × 2 matrix A = . We define the trace of A by
a21 a22
tr(A) = a11 + a22 and the determinant by det(A) = a11 a22 − a12 a21 , then a simple calculation shows
that the general characteristic equation for a 2 × 2 matrix A is

p(λ) = |λI − A| = λ2 − tr(A)λ + det(A) = 0.

We know that for a quadratic equation there are three cases depending on the discriminant of the
quadratic:

1
Set ∆ = det(A), τ = trace(A) then the Discriminant is

D = τ 2 − 4∆.

There are three cases depending on the cases D > 0, D = 0 and D < 0.

1. D > 0 implies there are two real distinct eigenvalues λ2 < λ1 with associated eigenvectors V1
and V2 .

2. D = 0 implies there is real distinct eigenvalue (a double root) λ0 and there are two possibilities:

(a) There may be two linearly independent eigenvectors V1 and V2 .

(b) There may only be one linearly independent eigenvector V0 . This is the most complicated
case. In this case we will also find a so-called generalized eigenvector requiring us to find
a vector P which involves solving (A − λ0 )P = V0 .

3. D < 0 implies there are complex eigenvalues λ0 = α ± iβ. In this case we solve (A − λ0 )V = 0
where V will contain complex numbers, i.e., V = W + iZ where W and Z are the real and
imaginary parts of V .

Here are some examples:

 
−1 2 (λ + 1) −2
1. A = |λI − A| = = λ2 − 7λ + 6 = (λ − 1)(λ − 6) = 0 ⇒ λ1 = 1
−7 8 7 (λ − 8)
and λ2 = 6.

(a) We seek V1 6= 0 satisfying (λ1 I − A) V1 = 0 this gives

    
2 −2 a 0
=
7 −7 b 0
 
1
We see immediately that we can take a = 1 and b = 1 so V1 = .
1
(b) We seek V2 6= 0 satisfying (λ2 I − A) V2 = 0 this gives
    
7 −2 a 0
=
7 −2 b 0
 
2
We see immediately that we can take a = 2 and b = 7 so V1 = .
7
 
−8 −1 (λ + 8) 1
2. A = |λI − A| = = λ2 + 8λ + 16 = (λ + 4)2 = 0 ⇒ λ1 = λ2 = −4.
16 0 −16 λ
We seek V1 6= 0 satisfying (λ1 I − A) V1 = 0 this gives
    
4 1 a 0
=
−16 −4 b 0
 
1
We see immediately that we can take a = 1 and b = −4 so V1 = .
−4
but notice there is no second linearly independent vector V2 .

2
  
−1 2 (λ + 1) −2
3. A = |λI − A| = = λ2 + 9 = 0 ⇒ λ1 = 3i and λ2 = −3i.
−5 1 5 (λ − 1)

(a) We seek V1 6= 0 satisfying (λ1 I − A) V1 = 0 this gives

    
(3i + 1) −2 a 0
=
5 (3i − 1) b 0
 
2
We see immediately that we can take a = 2 and b = (3i + 1) so V1 = .
(3i + 1)
(b) We seek V2 6= 0 satisfying (λ2 I − A) V2 = 0. Notice that λ2 = λ1 and A = A. From
this we can write
AV1 = λ1 V1 ⇒ AV1 = λ2 V1 .
 
2
We see immediately conclude that V2 = V1 = .
(−3i + 1)
(c) We can write V1 in terms of its real and imaginary parts as
     
2 2 0
V1 = = +i = W + i Z.
(3i + 1) 1 3
 
5 −1 0 (λ − 5) 1 0
4. A = 0 −5 9
  |λI − A| = 0 (λ + 5) 0 = λ(λ − 4)(λ + 4) = 0.
5 −1 0 −5 1 λ
We can easily see that λ1 = −4, λ2 = 0 and λ3 = 4.

Remark 1. When you are doing eigenvalue and eigenvector problems it is always a good idea
to keep a few things in mind. For example, if you have a 3×3 matrix A then the characteristic
polynomial will always be a cubic polynomial. That means there will be exactly 3 eigenvalues
but only, at most, 3 eigenvectors. On the other hand, distinct eigenvalues always have an
eigenvector so if the characteristic polynomial has 3 real distinct or 1 real root and a pair of
complex eigenvectors then (since these must all be distinct, i.e. one is real and the other two
are complex with one having + and the other − the imaginary part) there must be exactly 3
eigenvectors.
This give some useful information. When you write out (λI − A)v = 0 , you know there
is exactly one nonzero solution (up to constant multiples). In other words the null space of
(λI − A) is one dimensional. That means that there will be one free parameter. In particular
when you are working with the augmented matrix it will always happen that in the reduced
row echelon form the last row will become all zeros - It has to.

This can usually be used to your advantage to simplify the process of finding the eigenvector.
Usually you can just pick two of the equations and treat one variable as a free variable and
solve a 2 × 2 system for the other variables in terms of this free variable.

3
(a) We seek V1 6= 0 satisfying (λ1 I − A) V1 = 0 this gives
    
−9 1 0 a 0
0 1 −9   b = 0
 
−5 1 −4 c 0
 
−9 1 0 0
 0 5 −9 0 
−5 1 −4 0
Now we know that the Null space of (λ1 I − A) is one dimensional so the last row of
the RREF will be all zeros so let us just take the first two equations and take the terms
involving c to the right hand side.

−9a + b = 0
b = 9c
 
1
So c is arbitrary and b = 9c and a = c. Thus we have V1 = 9 c.
1
(b) We seek V2 6= 0 satisfying (λ2 I − A) V2 = 0 this gives
    
−5 1 0 a 0
0 5 −9   b = 0
 
−5 1 0 c 0
 
−5 1 0 0
 0 5 −9 0 
−5 1 0 0
Once again we know that the Null space of (λ2 I − A) is one dimensional so the last row
of the RREF will be all zeros so let us just take the first two equations and take the
terms involving c to the right hand side.

−5a + b = 0
5b = 9c
So c is arbitrary and b = 9/5c and a − 1/5b = 0. So a = 9/25c. So, for example, if we
don;t like fractiosn
  we could take we take c = 25 which gives b = 45 and a = 9. Thus we
9
have V2 = 45.

25
(c) We seek V3 6= 0 satisfying (λ3 I − A) V3 = 0 this gives
    
−1 1 0 a 0
0 9 −9  b  = 0
−5 1 4 c 0

4
  
−1 1 0 0
 0 9 −9 0 
−5 1 4 0
Once again we know that the Null space of (λ2 I − A) is one dimensional so the last row
of the RREF will be all zeros so let us just take the first two equations and take the
terms involving c to the right hand side.

a−b =0
9b = 9c
 
1
So c is arbitrary and b = c and a = b. So Thus we have V3 = 1.

1

5.2 Constant Coefficient 2D First Order Systems

In this section we consider constant coefficient two dimensional first order systems of ordinary
differential equations.     
d x a b x
= (1)
dt y c d y
If we introduce matrices we can write the system in a simple form
   
x a b dX
X= , A= , ⇒ = AX. (2)
y c d dt

Look for a simple solution X = eλt v with v 6= 0 then we have

dX
λX = = AX, λv = Av
dt
which implies
(λI − A)v = 0. (3)

Notice that this is just an eigenvalue problem as discussed in the previous section. Recalling the
earlier discussion we have three cases depending on whether the discriminant D > 0, D = 0, D < 0:
Here ∆ = det(A), τ = trace(A) and the Discriminant is

D = τ 2 − 4∆.

1. D > 0 implies there are two real distinct eigenvalues λ2 < λ1 with associated eigenvectors v1
and v2 . The general solution of (1) is

X(t) = a1 eλ1 t V1 + a2 eλ2 t V2 . (4)

2. D = 0 implies there is real distinct eigenvalue (a double root) λ0 and there are two possibilities:

5
 (a) There may be two linearly independent eigenvectors V1 and V2 . If so the general solution
of (1) is
X(t) = a1 eλ0 t V1 + a2 eλ0 t V2 . (5)

(b) There may only be one linearly independent eigenvector V0 . This is the most complicated
case. In this case the general solution of (1) is then

X(t) = a1 eλ0 t V0 + a2 eλ0 t (tV0 + P ) (6)

where w is any solution of (A − λ0 )P = V0 .

3. D < 0 implies there are complex eigenvalues λ0 = α ± iβ. In this case we solve (A − λ0 )v = 0
where v will contain complex numbers, i.e., v = W + iZ where W and Z are the real and
imaginary parts of v. Then the general solution can be written as

X(t) = a1 eαt cos(βt)W − sin(βt)Z + a2 eαt cos(βt)Z + sin(βt)W .

   
(7)

6
 Constant Coefficient First Order Systems: the General Case
The case of a general n × n matrix is somewhat more complicated. We will not attempt a complete
discussion. Consider
   
x1 a11 · · · a1n
dX  ..   .. ..  .
= AX where X =  .  , A =  . .  (8)
dt
xn an1 · · · ann
Once again seeking simple solutions X = eλt V with V 6= 0 leads to

(λI − A)V = 0

The homogeneous system has a nonzero solution if and only if

p(λ) = det(λI − A) = λn + an−1 λn−1 + · · · a1 λ + a0 = 0.

Since we assume that n > 2 there are many more possibilities than in the case n = 2. We only
consider a few cases.
1. Distinct Real Eigenvalues If A has n distinct real eigenvalues λ1 , λ2 , · · · , λn with associated
eigenvectors V1 , V2 , · · · , Vn then the general solution is given by

X(t) = c1 eλ1 t V1 + c2 eλ2 t V2 + · · · + cn eλn t Vn .

2. A Repeated Eigenvalue Suppose that one of the eigenvalues, say λj is an mth order root
of p(λ) then there are many possibilities.
(a) If there are m linearly independent eigenvectors Vj1 , Vj2 , · · · , Vjm for λj then the general
solution contains an expression of the form

Xj (t) = cj1 eλj t Vj1 + cj2 eλj t Vj2 + · · · + cjm eλj t Vjm .

(b) If for example, if λj has multiplicity two and there is only one eigenvector Vj then
Xj = a1 eλj t Vj + a2 eλj t (tVj + P ) where (A − λj )P = Vj . But if λj has multiplicity three
and there is only one eigenvector Vj then
t2
 
λj t λj t λj t
X j = a1 e Vj + a2 e (tVj + P ) + a3 e Vj + tP + Q
2
where (A − λj )P = Vj and (A − λj )Q = P .
3. A Pair of Complex Eigenvalues Suppose λj = α + iβ is an eigenvalue of multiplicity one
(note that also α−iβ is an eigenvalue if A has real coefficients) with eigenvector Vj = W +iZ.
The we write

Xj1 (t) = [W cos(βt) − Z sin(βt)] eαt , Xj1 (t) = [Z cos(βt) + W sin(βt)] eαt

where W = Re (Vj ) and Z = Im (Vj ). Then the general solution contains terms

Xj (t) = cj1 Xj1 (t) + cj2 Xj2 (t).

Notice this is not the most general case since λj could have multiplicity m > 1.