Math 215 HW #11 Solutions

1. Problem 5.5.6. Find the lengths and the inner product of

   
2 − 4i 2 + 4i
~x = and ~y .
4i 4i

Answer: First,
 
2 − 4i
k~xk2 = ~xH ~x = [2 + 4i − 4i] = (4 + 16) + 16 = 36,
4i

so k~xk = 6. Likewise,
 
2 H 2 + 4i
k~y k = ~y ~y = [2 − 4i − 4i] = (4 + 16) + 16,
4i

so k~y k = 6.
Finally,
 
H 2 + 4i
h~x, ~y i = ~x ~y = [2 + 4i − 4i] = (2 + 4i)2 − (4i)2 = (4 − 16 + 16i) + 16 = 4 + 16i.
4i

2. Problem 5.5.16. Write one significant fact about the eigenvalues of each of the following:

(a) A real symmetric matrix.

Answer: As we saw in class, the eigenvalues of a real symmetric matrix are all real
numbers.
(b) A stable matrix: all solutions to du/dt = Au approach zero.
Answer: By the definition of stability, this means that the reals parts of the eigenvalues
of A are non-positive.
(c) An orthogonal matrix.
Answer: If A~x = λ~x, then

hA~x, A~xi = hλ~x, λ~xi = λ2 h~x, ~xi = λ2 k~xk2 .

On the other hand,

hA~x, A~xi = (A~x)T A~x = ~xT AT A~x = ~xT ~x = h~x, ~xi = k~xk2 .

Therefore,
λ2 k~xk2 − k~xk2 ,
meaning that λ2 = 1, so |λ| = 1.
(d) A Markov matrix.
Answer: We saw in class that λ1 = 1 is an eigenvalue of every Markov matrix, and that
all eigenvalues λi of a Markov matrix satisfy |λi | ≤ 1.

1
 (e) A defective matrix (nondiagonalizable).
Answer: If A is n × n and is not diagonalizable, then A must have fewer than n
eigenvalues (if A had n distinct eigenvalues and since eigenvectors corresponding to
different eigenvalues are linear independent, then A would have n linearly independent
eigenvectors, which would imply that A is diagonalizable).
(f) A singular matrix.
Answer: If A is singular, then A has a non-trivial nullspace, which means that 0 must
be an eigenvalue of A.

3. Problem 5.5.22. Every matrix Z can be split into a Hermitian and a skew-Hermitian part,
Z = A + K, just as a complex number z is split into a + ib. The real part of z is half of z + z,
and the “real part” (i.e. Hermitian part) of Z is half of Z + Z H . Find a similar formula for
the “imaginary part” (i.e. skew-Hermitian part) K, and split these matrices into A + K:
   
3 + 4i 4 + 2i i i
Z= and Z = .
0 5 −i i

Answer: Notice that

(Z + Z H )H = Z H + (Z H )H = Z H + Z,
so indeed 12 (Z + Z H ) is Hermitian. Likewise,

(Z − Z H )H = Z H − (Z H )H = Z H − Z = −(Z − Z H ),

is skew-Hermitian, so K = 12 (Z − Z H ) is the skew-Hermitian part of Z.

Hence, when  
3 + 4i 4 + 2i
Z= ,
0 5
we have
       
1 1 3 + 4i 4 + 2i 3 − 4i 0 1 6 4 + 2i 3 4 + 2i
A = (Z+Z H ) = + = =
2 2 0 5 4 − 2i 5 2 4 − 2i 10 4 − 2i 5
and
       
1 1 3 + 4i 4 + 2i 3 − 4i 0 1 8i 4 + 2i 4i 4 + 2i
K = (Z−Z H ) = − = = .
2 2 0 5 4 − 2i 5 2 −4 + 2i 0 −4 + 2i 0

On the other hand, when 

i i
Z=
−i i
we have
       
1 1 i i −i i 1 0 2i 0 i
A = (Z + Z H ) = + = =
2 2 −i i −i −i 2 −2i 0 −i 0
and        
1 1 i i −i i 1 2i 0 i 0
K = (Z − Z H ) = − = = .
2 2 −i i −i −i 2 0 2i 0 i

2
4. Problem 5.5.28. If A~z = ~0, then AH A~z = ~0. If AH A~z = ~0, multiply by ~zH to prove that
A~z = ~0. The nullspaces of A and AH are . AH A is an invertible Hermitian matrix
when the nullspace of A contains only ~z = .
Answer: Suppose A A~z = ~0. Then, multiplying both sides by ~zH yields
H

0 = ~zH AH A~z = (A~z)H (A~z) = hA~z, A~zi = kA~zk2 ,

meaning that A~z = ~0.

Therefore, we see that if A~z = ~0, then AH A~z = ~0 and if AH A~z = ~0, then A~z = ~0, so the
nullspaces of A and AH are equal. AH A is an invertible matrix only if its nullspace is {~0}, so
we see that AH A is an invertible matrix when the nullspace of A contains only ~z = ~0.
5. Problem 5.5.48. Prove that the inverse of a Hermitian matrix is again a Hermitian matrix.

Proof. If A is Hermitian, then

A = U ΛU H ,
where U is unitary and Λ is a real diagonal matrix. Therefore,

A−1 = (U ΛU H )−1 = (U H )−1 Λ−1 U −1 = U Λ−1 U H

since U −1 = U H . Note that Λ−1 is just the diagonal matrix with entries 1/λi (where the λi
are the entries in Λ). Hence,

(A−1 )H = (U Λ−1 U H )H = U (Λ−1 )H U H = U Λ−1 U H = A−1

since Λ−1 is a real matrix, so we see that A−1 is Hermitian.

6. Problem 5.6.8. What matrix M changes the basis V ~1 = (1, 1), V~2 = (1, 4) to the basis
~P ~ ~2 as combinations
v1 = (2, 5), ~v2 = (1, 4)? The columns of M come from expressing V1 and V
mij ~vi of the ~v ’s.
Answer: Since      
~ 1 2 1
V1 = = − = ~v1 − ~v2
1 5 4
and  
~2 = 1 = ~v2 ,
V
4
we see that 

1 0
M= .
−1 1

7. Problem 5.6.12. The identity transformation takes every vector to itself: T ~x = ~x. Find the
corresponding matrix, if the first basis is ~v1 = (1, 2), ~v2 = (3, 4) and the second basis is
w
~ 1 = (1, 0), w
~ 2 = (0, 1). (It is not the identity matrix!)
Answer: Despite the slightly confusing way this question is worded, it is just asking for the
matrix M which converts the ~v basis into the w
~ basis. Clearly,
     
1 1 0
~v1 = = +2 =w ~ 1 + 2w
~2
2 0 1

3
 and      
3 1 0
~v2 = =3 +4 = 3w
~ 1 + 4w
~ 2,
4 0 1
so the desired matrix is  
1 3
M= .
2 4

8. Problem 5.6.38. These Jordan matrices have eigenvalues 0, 0, 0, 0. They have two eigenvectors
(find them). But the block sizes don’t match and J is not similar to K.
   
0 1 0 0 0 1 0 0
0 0 0 0
 and K = 0 0 1 0 .
 
J =0 0 0 1 0 0 0 0
0 0 0 0 0 0 0 0

For any matrix M , compare JM and M K. If they are equal, show that M is not invertible.
Then M −1 JM = K is impossible.
Answer: First, we find the eigenvectors of J and K. Since all eigenvalues of both are 0,
we’re just looking for vectors in the nullspace of J and K. First, for J, we note that J is
already in reduced echelon form and that J~v = ~0 implies that ~v is a linear combination of
   

 1 0 
   
  , 0 .
0

 0 1
 
0 0
 

Hence, these are the eigenvectors of J.

Likewise, K is already in reduced echelon form and K~v = ~0 implies that ~v is a linear combi-
nation of    

 1 0 
   
0
  , 0 .

 0 0
 
0 1
 

Hence, these are the eigenvectors of K.

Now, suppose  
m11 m12 m13 m14
m21 m22 m23 m24 
M =
m31

m32 m33 m34 
m41 m42 m43 m44
such that JM = M K. Then
    
0 1 0 0 m11 m12 m13 m14 m21 m22 m23 m24
0 0 0 0 m21 m22 m23 m24 
= 0 0 0 0 
 
JM = 
0 0

0 1 m31 m32 m33 m34  m41 m42 m43 m44 
0 0 0 0 m41 m42 m43 m44 0 0 0 0

4
 and     
m11 m12 m13 m14 0 1 0 0 0 m11 m12 0
m21 m22 m23 m24  0 0 1 0 0 m21 m22 0
MK = 
m31
 = .
m32 m33 m34  0 0 0 0 0 m31 m32 0
m41 m42 m43 m44 0 0 0 0 0 m41 m42 0

Therefore JM = M K means that

   
m21 m22 m23 m24 0 m11 m12 0
 0 0 0 0  0
  m21 m22 0
m41 m42 m43 m44  = 0
 
m31 m32 0
0 0 0 0 0 m41 m42 0

and so we have that

m21 = m24 = m22 = m41 = m44 = m42 0.
Plugging these back into M , we see that
 
m11 m12 m13 m14
 0 0 m23 0 
M = m31 m32
.
m33 m34 
0 0 m43 0

Clearly, the second and fourth rows are multiples of each other, so M cannot possibly have
rank 4. However, M not having rank 4 means that M cannot be invertible. Therefore,
M −1 JM = K is impossible, so it cannot be the case that J and K are similar.

9. Problem 5.6.40. Which pairs are similar? Choose a, b, c, d to prove that the other pairs aren’t:
       
a b b a c d d c
.
c d d c a b b a

Answer: The second and third are clearly similar, since

 −1  
0 1 0 1
=
1 0 1 0

and         
0 1 b a 0 1 0 1 a b c d
= = .
1 0 d c 1 0 1 0 c d a b
Likewise, the first and fourth are similar, since
        
0 1 d c 0 1 0 1 c d a b
= = .
1 0 b a 1 0 1 0 a b c d

There are no other similarities, as we can see by choosing

a = 1, b = c = d = 0.

5
 Then the matrices are, in order
       
1 0 0 1 0 0 0 0
.
0 0 0 0 1 0 0 1
Each of these is already a diagonal matrix, and clearly the first and fourth have 1 as an
eigenvalue, whereas the second and third have only 0 as an eigenvalue. Since similar matrices
have the same eigenvalues, we see that neither the first nor the fourth can be similar to either
the second or the third.
10. (Bonus Problem) Problem 5.6.14. Show R x that every number is an eigenvalue for T f (x) =
df /dx, but the transformation T f (x) = 0 f (t)dt has no eigenvalues (here −∞ < x < ∞).

Proof. For the first T , note that, if f (x) = eax for any real number a, then
df
T f (x) = = aeax = af (x).
dx
Hence, any real number a is an eigenvalue of T .
Turning to the second T , suppose we had that T f (x) = af (x) for some number a and some
function f . Then, by the definition of T ,
Z x
f (t)dt = af (x).
0
Now, use the fundamental theorem of calculus to differentiate both sides:
f (x) = af 0 (x).
Solving for f , we see that
f 0 (x)dx
Z Z
1
= dx,
f (x) a
so
x
ln |f (x)| = + C.
a
Therefore, exponentiating both sides,
|f (x)| = ex/a+C = eC ex/a .
We can get rid of the absolute value signs by substituting A for eC (allowing A to possibly
be negative):
f (x) = Aex/a .
Therefore, we know that
Z x Z x ix
T f (x) = f (t)dt = Aet/a dt = aAet/a = aAex/a − aA = a(Aex/a − A) = a(f (x) − A).
0 0 0

On the other hand, our initial assumption was that T f (x) = af (x), so it must be the case
that
af (x) = a(f (x) − A) = af (x) − aA.
Hence, either a = 0 or A = 0. However, either implies that f (x) = 0, so T has no eigenvalues.