Solutions to Assignment 9

Math 217, Fall 2002

5.1.24 Construct an example of a 2 × 2 matrix with only one distinct eigenvalue.

 
a b
We know that if A then the eigenvalues of A are the roots of the char-
c d
acteristic equation (a − λ)(d − λ) − bc = 0. This equation will have a repeated
a 0
root if a = d and bc = 0. So for example, A will work. To give a specific
  c a
1 0
instance of this, A = has only one distinct eigenvalue, λ = 1.
1 1

5.1.26 Show that if A2 is the zero matrix, then the only eigenvalue of A is 0.

Suppose that λ is an eigenvalue of A. Then there is an eigenvector v such that

Av = λv. Multiplying both sides by A, we get that 0 = A2 v = Aλv = λAv =
λ(λv) = λ2 v. Thus λ2 v = 0. Recall that v is non-zero by the definition of an
eigenvector. We conclude that λ2 = 0, and hence that λ = 0.

5.1.32 Let A be the matrix of the linear transformation T : R3 → R3 that rotates points
about some line through the origin. Without writing A, find an eigenvalue of
A and describe the eigenspace.

Suppose we use L to indicate the line about which T rotates the points of R3 .
Then if v ∈ R3 lies on L, we have that T (v) = Av = v (rotating points around
a line does not move points on the line). Thus we can see that λ = 1 is an
eigenvalue of A. The corresponding eigenspace is either just L if T does not
rotate by 360z degrees for z ∈ N, and all of R3 otherwise (in the first case,
points end up in a different place then they begin unless they lie on L; in the
second, rotation by 360z just takes every point back to its starting position so
that every vector is an eigenvector for λ = 1).

1
  
0 3 1
5.2.10 Find the characteristic equation of 3 0 2.
1 2 0

 
−λ 3 1
This amounts to finding det  3 −λ 2 . I will do this by cofactor ex-
1 2 −λ     
−λ 2 3 2 3 −λ
pansion. The answer is −λ det − 3 det + det =
2 −λ 1 −λ 1 2
−λ(λ2 − 4) − 3(−3λ − 2) + (6 + λ) = −λ3 + 14λ + 12.

 
5 −2 6 −1
0 3 h 0 
5.2.18 Find h in the matrix A = 
0 0 5 4  such that the eigenspace for λ = 5


0 0 0 1
is two-dimensional.

The eigenspace corresponding to λ = 5 is the Null space of the matrix A − 5I.

We can calculate the dimension of Nul(A − 5I) row reducing
 
0 −2 6 −1
0 −2 h 0 
A − 5I = 0 0 0 4  .


0 0 0 −4

Going as far as we can without knowing h, we get

 
0 1 −h/2 0
0 0 6 − h 0
B= 0 0
.
0 1
0 0 0 0

Now rank(A) + dim(Nul(A)) = rank(B) + dim(Nul(B)) = 4, so unless h = 6 we

see that B has 3 pivot columns, and then dim(Nul(B)) = dim(Nul(A−5I)) < 2.
If h = 6, then rank(B) = 2 so that dim(Nul(B)) = 2. We conclude that h must
equal 6 in order for the eigenspace corresponding to λ = 5 to be two-dimensional.

         
.5 .2 .3 .3 1 −1 1
5.2.27 Let A = .3 .8 .3, v1 = .6, v2 = −3, v3 =  0 , and w = 1.
.2 0 .4 .1 2 1 1

(a) Show that v1 , v2 , and v3 are eigenvectors of A.

2
      
.3 .5 −.2
To do this we calculate Av1 = .6, Av2 = −1.5, and Av3 =  0 .
 .1    1    .2
.3 .3 .5 1
It is not difficult to see that  6  = .6, −1.5 = (1/2) −3, and
    1 .1 1 2
−.2 −1
 0  = (1/5)  0 . Thus v1 , v2 , and v3 are eigenvectors of A.
.2 1
(b) Let x0 be an vector in R3 with nonnegative entries whose sum is 1. Ex-
plain why there are constants c1 , c2 , c3 such that x0 = c1 v1 + c2 v2 + c3 v3 .
Compute wT x0 and conclude that c1 = 1.

We saw in the part (a) that A has 3 distinct eigenvalues (1, 1/2, and
1/5). Thus we know that the three eigenvectors corresponding to these
eigenvalues are linearly independent (theorem 2, pg 307). Because the
dimension of R3 is 3, we know by the basis theorem that v1 , v2 , and v3
give a basis for R3 (in particular, they must span R3 ). Thus there must
exist constants c1 , c2 , c3 as required.
 
a1
If we write x0 = a2 , then wT x0 = a1 + a2 + a3 = 1. Now multiplying
a3
both sides of the equation x0 = c1 v1 + c2 v2 + c3 v3 by wT , we obtain
1 = wT x0 = c1 wT v1 + c2 wT v2 + c3 wT v3 = c1 (1) + c2 (0) + c3 (0). From this
we see that c1 = 1.
(c) For k = 1, 2, . . . , define xk = Ak x0 with x0 as in part (b). Show that
xk → v1 as k increases.

In part (a) we learned that Av2 = (1/2)v2 , so

Ak v2 = Ak−1 Av2 = Ak−1 (1/2)v2 = (1/2)Ak−1 v2 = (1/2)Ak−2 Av2 =

(1/2)2 Ak−2 v2 = · · · = (1/2)k v2 .

We can similarly show that Ak v3 = (1/5)k v3 . From part (b) we know
that x = v1 + c2 v2 + c3 v3 . Thus xk = Ak x0 = Ak (v1 + c2 v2 + c3 v3 ) =
Ak v1 + c2 Ak v2 + c3 Ak v3 = v1 + c2 (1/2)k v2 + c3 (1/5)k v3 . For k → ∞, this
limits to v1 .

 
−7 −16 4
5.3.18 Diagonalize A =  6 13 −2 if possible. One of the eigenvalues is λ = 5,
12  16  1
−2
and one eigenvector is 1 .

2

3
We need to calculate the eigenvalues, and then demonstrate that there are 3
linearly independent eigenvectors.
First we form the characteristic equation:
 
−7 − λ −16 4
det  6 13 − λ −2 
12 16 1−λ

= (−7 − λ) det A1,1 − (−16) det A1,2 + 4 det A1,3

     
13 − λ −2 6 −2 6 13 − λ
= (−7−λ) det +16 det +4 det
16 1−λ 12 1 − λ 12 16
= (−7 − λ)((13 − λ)(1 − λ) + 32) + 16(6 − 6λ + 24) + 4(96 − 12(13 − λ))
= −λ3 + 7λ2 + 5λ − 75.

We know that λ = 5 is a root of this equation, so we can write −λ3 + 7λ2 + 5λ −

75 = (λ − 5)(−λ2 + 2λ + 15) (you can either do this by inspection, or use long
division of polynomials). We can factor this further as (λ−5)(−λ+5)(λ+3)) =
−(λ − 5)2 (λ + 3). Thus A has two distinct eigenvalues.
So now we look for the eigenvectors of A. We do this by row reducing [(A−λI) 0]
for each of the eigenvalues of A.
For λ = 5,
   
−7 − λ −16 4 0 −12 −16 4 0
 6 13 − λ −2 0 =  6 8 −2 0 ,
12 16 1−λ 0 12 16 −4 0
 
1 4/3 1/3 0
which has row reduced echelon form 0 0 0 0. From this matrix we
0 0 0 0
can read the eigenspace corresponding to the eigenvalue λ = 5. It is the set of
all vectors of the form
 4  −4   −1 
− 3 x2 − 13 x3

3 3
= x2  = x2  1  + x3  0  .
x3 0 1
 
4
We choose two linearly independent eigenvectors from this set, v1 = 3 and
  0
−1
v2 =  0  (I just cleared denominators).
3
For λ = −3,
   
−7 − λ −16 4 0 −4 −16 4 0
 6 13 − λ −2 0 =  6 16 −2 0 ,
12 16 1−λ 0 12 16 4 0

4
  
1 0 1 0
which has row reduced echelon form 0 1 −1/2 0. From this matrix we
0 0 0 0
can read the eigenspace corresponding to the eigenvalue λ = −3. It is the set
of all vectors of the form    
−x3 −1
 1 x3  = x3  1  .
2 2
x3 1
 
−2
We choose a representative vector from this set, v3 = 1 .

2
   
4 −1 −2 1 0 0
Now row reducing the matrix 3 0 1 . The result is 0 1 0, so by
0 3 2 0 0 1
the invertible matrix theorem the vi are linearly independent.
   
4 −1 −2 5 0 0
So let P = [v1 v2 v3 ] = 3 0 1  and D = 0 5 0 . By the diagonal-
0 3 2 0 0 −3
−1
ization theorem, A = P DP , so we have diagonalized A.

5.3.28 Show that if A has n linearly independent eigenvectors, then so does AT .

A matrix A has n linearly independent eigenvectors if and only if A is diag-

onalizable, so it is enough to show that AT is diagonalizable. We are given
that A is diagonalizable, so there is a diagonal matrix D and an invertible
matrix P such that A = P DP −1 . Taking the transpose of both side yields
AT = (P DP −1 )T = (P −1 )T DT P T = (P T )−1 DP T . For the last equality I used
that fact that (BC)T = C T B T , and (B −1 )T = (B T )−1 (see theorems 3, pg 115
and 6, pg 121). I also used that the transpose of a diagonal matrix D is simply
D. So let Q = (P T )−1 . Then our equation says that AT = QDQ−1 , and hence
AT is diagonalizable.

5.3.28 Construct a nondiagonal 2 × 2 matrix that is diagonalizable but not invertible.

 
a b
Let A = . Then A will fail to be invertible if ad − bc = 0. So con-
c d  
2 6
sider the matrix A = . Clearly A is not invertible. Note also that
  1 3
2−λ 6
det = λ2 − 5λ = λ(λ − 5). Thus A has two distinct eigenvectors
1 3−λ
and hence by theorem 2 (pg 307), two linearly independent eigenvectors. This
implies by the Diagonalization theorem that A is diagonalizable.