PHYSICS 116A

Homework 8 Solutions
1. Boas, problem 3.9–4. Given the matrix,
 
0 2i −1
A = −i
 2 0 ,
3 0 0
find the transpose, the inverse, the complex conjugate and the transpose conjugate of A. Verify
that AA−1 = A−1 A = I, where I is the identity matrix.

We shall evaluate A−1 by employing Eq. (6.13) in Ch. 3 of Boas. First we compute the determinant
by expanding in cofactors about the third column:
0 2i −1
−i 2
det A ≡ −i 2 0 = (−1) = (−1)(−6) = 6 .
3 0
3 0 0
Next, we evaluate the matrix of cofactors and take the transpose:
 
2 0 2i −1 2i −1
−

 0 0 0 0 2 0  0 
  0 2
   
 −i 0 0 −1 0 −1 
adj A =  = 0 3

− 3 0 3 0
−
−i 0
i . (1)
   
 
  −6 6i −2
 −i 2 0 2i 0 2i 
−
3 0 3 0 −i 2
According to Eq. (6.13) in Ch. 3 of Boas the inverse of A is given by Eq. (1) divided by det(A), so
 1

0 0 3
 1 i

A−1 =  0 2 6 . (2)
−1 i − 13

It is straightforward to check by matrix multiplication that

  1
  1
   
0 2i −1 0 0 3 0 0 3 0 2i −1 1 0 0
  1 i
  1

i
   
 −i 2 0 6   −i 2 0 0 1 0
 0 6 =  0 = .
 
 2 2 
3 0 0 −1 i − 13 −1 i − 13 3 0 0 0 0 1

The transpose, complex conjugate and the transpose conjugate∗ can be written down by inspec-
tion:
     
0 −i 3 0 −2i −1 0 i 3
     
AT = 
 2i 2 0
, A∗ = 
 i 2 0
, A† = 
−2i 2 0
.
−1 0 0 3 0 0 −1 0 0
∗
The transpose conjugate is more often referred to as the hermitian conjugate or the adjoint.

1
 These have been obtained by noting that AT is obtained by A by interchanging the rows and
columns, A∗ is obtained from A by complex conjugating the matrix elements, and the definition
of the hermitian conjugate is A† = (A∗ )T .

Alternative method : One can also evaluate A−1 by employing Gauss Jordan elimination, which is
described in the class handout http://young.physics.ucsc.edu/116A/gauss jordan.pdf.
   
0 2i −1 1 0 0
A =  −i 2 0 , I = 0 1 0
3 0 0 0 0 1
First we interchange R1 ↔ R3 . Then we rescale the new row 1, R1 → 31 R1 to obtain
   1

1 0 0 0 0 3
   
 0 2i −1, 1 0 0 
   
−i 2 0 0 1 0
Next, we perform the elementary row operations, R3 → R3 + iR1 and R2 → − 12 iR2 to obtain:
   1

1 0 0 0 0 3
 1 
  
 0 1 − 1 i 0 0
 2 i ,  2 
1
0 2 0 0 1 3i
Next, we perform the elementary row operations, R3 → R3 − 2R1 to obtain:
   1

1 0 0 0 0 3
 1 
  
 0 1 − 1 i 0 0
 2 i ,  2 
1
0 0 −i i 1 3i

Finally, we perform the elementary row operations, R2 → R2 + 12 R3 , followed by R3 → iR3 to

obtain:    
1
1 0 0 0 0 3
   1

i
 0
 1 0
, A−1 = 
 0 2 6
0 0 1 −1 i − 13
The final step produces the inverse, which is indicated above. Note that we have reproduced
Eq. (2).
2. Boas, problem 3.9–5. Show that the product AAT is a symmetric matrix.

Using Eq. (9.10) on p. 139 of Boas, (AB)T = B TAT for any two matrices A and B. Hence,

(AAT )T = (AT )T AT = AAT , (3)

where we have used the fact† that (AT )T = A for any matrix A. Eq. (3) implies that AAT is a
symmetric matrix, since by definition a symmetric matrix is equal to its transpose [cf. the table
at the top of p. 138 of Boas].
†
The transpose of a matrix interchanges the rows and columns. Thus, if one performs the transpose operation twice,
the original matrix is recovered.

2
3. Boas, problem 3.9–17.

(a) Show that if A and B are symmetric, then AB is not symmetric unless A and B commute.

If A and B are symmetric, then A = AT and B = B T . We now examine

(AB)T = B T AT = BA ,

after using the fact that A and B are symmetric matrices. We conclude that (AB)T = AB
if and only if AB = BA. That is, AB is not symmetric unless A and B commute.
(b) Show that a product of orthogonal matrices is orthogonal.

Consider orthogonal matrices Q1 and Q2 . By definition [cf. the table at the top of p. 138 of
Boas], we have Q−1 T −1 T
1 = Q1 and Q2 = Q2 . We now compute

T T T
(Q1 Q2 )−1 = Q−1 −1
2 Q1 = Q2 Q1 = (Q1 Q2 ) , (4)

after using the fact that Q1 and Q2 are orthogonal. In deriving Eq. (4), we have used the
following properties of the inverse and the transpose

(AB)−1 = B −1 A−1 , and (AB)T = B T AT ,

for any pair of matrices A and B. Thus, we have shown that

(Q1 Q2 )−1 = (Q1 Q2 )T , which implies that Q1 Q2 is orthogonal.
(c) Show that if A and B are Hermitian, then AB is not Hermitian unless A and B commute.

If A and B are Hermitian, then A = A† and B = B † . We now examine

(AB)† = B † A† = BA , (5)

after using the fact that A and B are Hermitian matrices. In deriving Eq. (5), we have used
the fact that:
(AB)† = ((AB)∗ )T = (A∗ B ∗ )T = (B ∗ )T (A∗ )T = B † A† . (6)

We conclude that (AB)† = AB if and only if AB = BA. That is, AB is not Hermitian un-
less A and B commute.
(d) Show that a product of unitary matrices is unitary.

Consider unitary matrices U1 and U2 . By definition [cf. the table at the top of p. 138 of
Boas], we have U1−1 = U1† and U2−1 = U2† . We now compute

(U1 U2 )−1 = U2−1 U1−1 = U2† U1† = (U1 U2 )† ,

after using the fact that U1 and U2 are unitary and employing the property of the Hermitian
conjugation given in Eq. (6). Thus, we have shown that (U1 U2 )−1 = (U1 U2 )† , which implies
that U1 U2 is orthogonal.

3
4. Boas, problem 3.10-5(a). Given two vectors,

~ = (3 + i , 1 , 2 − i , −5i , i + 1)
A and ~ = (2i , 4 − 3i , 1 + i , 3i , 1) ,
B

find the norms of A ~ and B

~ and the inner product of A
~ and B
~ , and note that the Schwarz
inequality is satisfied.

~ and B
Using eq. (1.07) on p. 146 of Boas, the norms of A ~ are given by:

√
~ = |3 + i|2 + |1|2 + |2 − i|2 + | − 5i|2 + |i + 1|2 1/2 = (10 + 1 + 5 + 25 + 2)1/2 = 43 ,
kAk

√
~ = |2i|2 + |4 − 3i|2 + |1 + i|2 + |3i|2 + 1 1/2 = (4 + 25 + 2 + 9 + 1)1/2 = 41 .
kBk

~ and B
Using eq. (10.6) on p. 146 of Boas, the inner product of A ~ is given by:

~·B
A ~ = (3 − i)(2i) + (1)(4 − 3i) + (2 + i)(1 + i) + (5i)(3i) + (1 − i)(1)
= (2 + 6i) + (4 − 3i) + (1 + 3i) − 15 + (1 − i)
= −7 + 5i . (7)

The Schwarz inequality [see eq. (10.9) on p. 146 of Boas] is:

~ · B|
|A ~ ≤ kAk
~ kBk
~ ,

where | · · · | indicates the magnitude of the corresponding complex number. Using Eq. (7), it
follows that √ √
~ · B|
|A ~ = | − 7 + 5i| = 49 + 25 = 74 .

Thus, the Schwarz inequality is satisfied for this example since

√ √ √ √
74 ≤ 43 41 = 1763 .

5. Boas, problem 3.11–9. Show that det(C −1 M C) = det M . What is the product of det(C −1 ) and
det C? Thus, show that the product of the eigenvalues of M is equal to det M .

Eq. (6.6) on p. 118 of Boas states that det(AB) = det A det B for any two matrices A and B. It
follows that
det(C −1 )det C = det(C −1 C) = det I = 1 ,
where I is the identity matrix. Hence,‡
1
det(C −1 ) = .
det C
Using Eq. (6.6) of Boas again along with the result just obtained,

det M det C
det(C −1 M C) = det(C −1 ) det M det C = = det M . (8)
det C
‡
By assumption, C −1 exists, in which case det C 6= 0 and so it is permissible to divide by det C.

4
 Finally, Boas asks you to show that the product of the eigenvalues of M is equal to det M . What
she is expecting you to do is to use eq. (11.11) on p. 150 of Boas, which states that
 
λ1 0 0 ··· 0
 0 λ2 0 ··· 0
 
C −1 M C = D =  0 0 λ3 · · · 0

, (9)
 .. .. .. . . ..
 . . . . .
0 0 0 ··· λn

where D is a diagonal matrix whose diagonal elements are the eigenvalues of M , denoted by
λ1 , λ2 , . . . , λn above. It follows that

det(C −1 M C) = det D = λ1 λ2 · · · λn . (10)

The results of Eqs. (8) and (10) then imply that

det M = λ1 λ2 · · · λn . (11)

Note that this proof is not completely general, since not all matrices M are diagonalizable and,
for these, the matrix of eigenvectors C does not have an inverse since two or more eigenvectors
are equal, with the result that det(C) = 0. However, one can show that Eq. (11) is also true for
non-diagonalizable matrices.

6. Boas, problem 3.11–10. Show that Tr(C −1 M C) = Tr M . Thus, show that the sum of the
eigenvalues of M is equal to Tr M .

Eq. (9.13) on p. 140 of Boas states that the trace of a product of matrices is not changed by
permuting them in cyclic order. In particular, Tr(ABC) = Tr(BCA) = Tr(CAB) for any three
matrices A, B and C. It follows that

Tr(C −1 M C) = Tr(M CC −1 ) = Tr(M I) = Tr M , (12)

where I is the identity matrix.

Finally, Boas asks you to show that the sum of the eigenvalues of M is equal to Tr M . What she
is expecting you to do is to use Eq. (9) to obtain

Tr(C −1 M C) = Tr D = λ1 + λ2 + · · · + λn , (13)

where D is a diagonal matrix whose diagonal elements are the eigenvalues of M , denoted by
λ1 , λ2 , . . . , λn above. The results of Eqs. (12) and (13) then imply that

Tr M = λ1 + λ2 + · · · + λn . (14)

As in the previous question, this proof is not general, since not all matrices M are diagonalizable.
However, one can show that Eq. (14) is also true for non-diagonalizable matrices.

7. Boas, problem 3.11–19. Find the eigenvalues and eigenvectors of

 
1 2 2
M= 2  3 0 .
2 0 3

5
To compute the eigenvalues, we evaluate the characteristic equation,

1−λ 2 2
det(M − λI) = 2 3−λ 0 = 0.
2 0 3−λ

Expanding in terms of the cofactors of the elements of the third row,

1−λ 2 2
2 3−λ 0 = −4(3 − λ) + (3 − λ) [(1 − λ)(3 − λ) − 4]
2 0 3−λ

= (3 − λ) [(1 − λ)(3 − λ) − 8]

= (3 − λ)(λ2 − 4λ − 5)

= −(λ − 3)(λ − 5)(λ + 1) .

Hence, the characteristic equation possesses three roots: λ = −1 , 3 , and 5.

To obtain the eigenvectors, we plug in the eigenvalues into the equation M~
v = λ~
v . First, we
consider the eigenvalue λ = 5.
    
1 2 2 x x
2 3 0  y = 5 y .
 
2 0 3 z z

Expanding out this matrix equation yields:

x + 2y + 2z = 5x ,

2x + 3y = 5y ,

2x + 3z = 5z .

Rewrite the above equations as a set of homogeneous equations,

− 4x + 2y + 2z = 0 ,

2x − 2y = 0 ,

2x − 2z = 0 .

The solution can be determined by inspection,

x = y = z.

Normalizing the eigenvector yields (up to an overall sign):

 
1
1
~
v (λ=5) = √ 1 ,
3 1

6
which satisfies M~
v = λ~
v with eigenvalue λ = 5.

Next, we consider the eigenvalue λ = 3.

    
1 2 2 x x
2 3 0  y  = 3  y  .
2 0 3 z z
Expanding out this matrix equation yields:
x + 2y + 2z = 3x ,

2x + 3y = 3y ,

2x + 3z = 3z .
Rewrite the above equations as a set of homogeneous equations,
− 2x + 2y + 2z = 0 ,

2x = 0 ,

2x = 0 .
The solution can be determined by inspection,
x = 0, and y = −z .
Normalizing the eigenvector yields (up to an overall sign) :
 
0
1  
v (λ=3) = √
~ 1 ,
2 −1

which satisfies M~
v = λ~
v with eigenvalue λ = 3.

Finally, we consider the eigenvalue λ = −1.

    
1 2 2 x x
2 3 0  y  = −  y  .
2 0 3 z z

Expanding out this matrix equation yields:

x + 2y + 2z = 5x ,

2x + 3y = 5y ,

2x + 3z = 5z .
Rewrite the above equations as a set of homogeneous equations,
2x + 2y + 2z = 0 ,

2x + 4y = 0 ,

2x + 4z = 0 .

7
 After subtracting the last two equations, the solution can be determined by inspection,

y=z and x = −2y .

Normalizing the eigenvector yields (up to an overall sign):

 
2
1  
~
v (λ=−1) =√ −1 ,
6 −1

which satisfies M~
v = λ~
v with eigenvalue λ = −1.
Note that the three eigenvectors form an orthonormal set. This is the expected behavior for a
real symmetric (or complex hermitian) matrix.

8. Boas, problem 3.11–32. The matrix,

 
6 −2
M= ,
−2 3

describes a deformation of the (x, y) plane. Find the eigenvalues and eigenvectors of the trans-
formation, the matrix C that diagonalizes M and specifies the rotation to new axes (x′ , y ′ ) along
the eigenvectors, and the matrix D that gives the deformation relative to the new axes. Describe
the deformation relative to the new axes.

First, we compute the eigenvalues by solving the characteristic equation,

6−λ −2
= (6 − λ)(3 − λ) − 4 = λ2 − 9λ + 14 = (λ − 7)(λ − 2) = 0 .
−2 3−λ

Hence, the two eigenvalues are λ = 7 and λ = 2. Next, we work out the eigenvectors. Since M is
a real symmetric matrix, we know that the eigenvectors will be orthogonal. By normalizing them
to unity, the eigenvectors will then be orthonormal. First, we examine:
    
6 −2 x x
=2 ,
−2 3 y y

which is equivalent to

6x − 2y = 2x ,
−2x + 3y = 2y .

These equations yield one independent relation, y = 2x. Hence, the normalized eigenvector is
   
x 1 1
=√ .
y λ=2 5 2

The second normalized eigenvector is orthogonal to the first one, and thus is given by
   
x 1 2
=√ .
y λ=7 5 −1

8
 One can check this by verifying that,
    
6 −2 2 2
=7 .
−2 3 −1 −1

The columns of the diagonalizing matrix C are given by the two eigenvectors. Thus,
 
1 1 2
C=√ .
5 2 −1

We expect that C −1 M C = D is a diagonal matrix. Let’s check this. First, we note that the
columns of C are orthonormal. This implies that C is an orthogonal matrix so that C −1 = C T .
In particular, for C given above we have C T = C, in which case C −1 = C. Hence,
   
−1 1 1 2 6 −2 1 2
C MC =
5 2 −1 −2 3 2 −1

    
1 1 2 2 14 2 0
= = .
5 2 −1 4 −7 0 7

Note that the order of the eigenvectors appearing as columns in C determines the order of the
eigenvalues appearing along the diagonal of
 
2 0
D= .
0 7

The deformation described by D is a stretching of the vectors by a factor of 2 along the x′ axis
and by a factor of 7 along the y ′ axis [cf. eq. (11.19) on p. 152 of Boas].

9. Boas, problem 3.11–42. Verify that the matrix,

 
3 1−i
H= ,
1+i 2

is Hermitian. Find its eigenvalues and eigenvectors, write a unitary matrix U that diagonalizes H
by a similarity transformation, and show that U −1 HU is the diagonal matrix of eigenvalues.

A Hermitian matrix satisfies H † = H, where H † ≡ H ⋆ T is the complex-conjugate transpose of H.

Since (1 − i)⋆ = 1 + i, it follows that H is Hermitian. We compute the eigenvalues by solving the
characteristic equation,

3−λ 1−i
= (3 − λ)(2 − λ) − 2 = λ2 − 5λ + 4 = (λ − 4)(λ − 1) = 0 ,
1+i 2−λ

which yields two roots: λ = 4 and λ = 1. Next, we work out the eigenvectors. Since M is an
Hermitian matrix, we know that the eigenvectors are orthogonal.§ After normalizing them to
§ P ⋆
For two complex vectors v and w, the inner product is defined by hv , wi = i vi wi , where vi and wi are the
components of the vectors v and w. Note the appearance of the complex conjugate, vi⋆ , in the expression for the inner
product. Two complex vectors are then orthogonal if hv , wi = 0. See p. 146 of Boas for further details.

9
 unity, the two eigenvectors are orthonormal.¶ First, we examine:
    
3 1−i x x
=4 ,
1+i 2 y y

which yields
−x + (1 − i)y = 0 , (1 + i)x − 2y = 0 .
These two equations are not independent, since if you multiply the first equation by −1 − i, you
obtain the second equation. It follows that x = (1 − i)y. Normalizing the eigenvector to unity
yields:
   
x 1 1−i
=√ .
y λ=4 3 1

The second normalized eigenvector is orthogonal to the first one (keeping in mind the footnotes
at the bottom of this page), and thus is given by
   
x 1 −1
=√ .
y λ=1 3 1+i

One can check this by verifying that,

    
3 1−i −1 −1
= .
1+i 2 1+i 1+i

The columns of the unitary diagonalizing matrix are given by the orthonormal eigenvectors. Hence,
 
1 1−i −1
U=√ .
3 1 1+i

Finally, we check that U −1 HU is diagonal. Since U is unitary, U −1 = U † . Hence

   
−1 1 1+i 1 3 1−i 1−i −1
U HU =
3 −1 1 − i 1+i 2 1 1+i

    
1 1+i 1 4(1 − i) −1 4 0
= = .
3 −1 1 − i 4 1+i 0 1

As expected, the diagonal elements are the eigenvalues of H.

10. Boas, problem 3.11–58. Consider the matrix,

 
5 −2
M= . (15)
−2 2

Evaluate f (M ) where f (M ) is a series comprised of powers of M . In particular, compute M 4 ,

M 10 and eM .
¶
vi⋆ vi
P

The length of a complex vector v is given by kvk = i
1/2 . A complex vector v normalized to unity satisfies
kvk = 1.

10
 In Eq. (11.10) on p. 150 of Boas, the diagonalization of M is performed. The result is:
   
−1 1 1 −2 1 0
C MC = D , where C=√ and D= .
5 2 1 0 6

Then for any function consisting of sums of powers of M , we have

f (M ) = f (CDC −1 ) = Cf (D)C −1 = C diag (f (λ1 ) , f (λ2 ) , . . . , f (λn )) C −1 , (16)

where the λi are the eigenvalues of M , and diag is a diagonal matrix, whose diagonal entries are
given as the arguments. The second step above follows from the observation that

(CDC −1 )n = CDC
|
−1
CDC −1 CDC
{z
−1
· · · CDC −1} = CDn C −1 ,
n terms

after noting that C −1 C = I. Applying Eq. (16) to the matrix given in Eq. (15), we have
 
f (1) 0
f (M ) = C C −1
0 f (6)

      
1 1 −2 f (1) 0 1 2 1 1 −2 f (1) 2f (1)
= =
5 2 1 0 f (6) −2 1 5 2 1 −2f (6) f (6)

 
1 f (1) + 4f (6) 2f (1) − 2f (6)
= . (17)
5 2f (1) − 2f (6) 4f (1) + f (6)

We apply Eq. (17) to three cases:

   
1 4 14 + 4·64 2·14 − 2·64 1037 −518
M = = ,
5 2·14 − 2·64 4·14 + 64 −518 260

   
10 1 110 + 4·610 2·110 − 2·610 48, 372, 941 −24, 186, 470
M = = ,
5 2·110 − 2·610 4·110 + 610 −24, 186, 470 12, 093, 236

   
M 1 e1 + 4e6 2e1 − 2e6 e 1 + 4e5 2(1 − e5 )
e = = .
5 2e1 − 2e6 4e1 + e6 5 2(1 − e5 ) 4 + e5

11. Boas, problem 3.11–60. The Cayley-Hamilton theorem states that a matrix satisfies its own
characteristic equation. Verify this theorem for the matrix
 
5 −2
M= .
−2 2

The characteristic equation of M is given in eq. (11.4) of Boas,

5−λ −2
= (5 − λ)(2 − λ) − 4 = λ2 − 7λ + 6 = (λ − 6)(λ − 1) = 0 .
−2 2−λ

11
Thus, according to the Cayley-Hamilton theorem,

M 2 − 7M + 6I = 0 ,

where I is the 2 × 2 identity matrix and 0 is the 2 × 2 zero matrix. To verify this, we evaluate:
     
2 29 −14 5 −2 1 0
M − 7M + 6I = −7 +6
−14 8 −2 2 0 1

   
29 − 35 + 6 −14 + 14 0 0
= = .
−14 + 14 8 − 14 + 6 0 0

Boas suggests an alternate way of verifying the Cayley-Hamilton theorem. First, Boas diagonalizes
M [cf. eq. (11.10) on p. 150]. The result is:
   
1 1 −2 1 0
C −1 M C = D , where C=√ and D= .
5 2 1 0 6

Then, using M = CDC −1 , it follows that

M 2 − 7M + 6I = C(D2 − 7D + 6I)C −1 , (18)

Noting that the eigenvalues λ = 6 and λ = 1 satisfy the characteristic equation (which is how the
eigenvalues are determined in the first place), it immediately follows that

D2 − 7D + 6I = 0 . (19)

You can also verify this explicitly:

 2   
2 1 − 7·1 + 6 0 0 0
D − 7D + 6I = = .
0 62 − 7·6 + 6 0 0

Consequently, Eqs. (18) and (19) yield:

M 2 − 7M + 6I = 0 ,

as required by the Cayley-Hamilton theorem.

Note: Using the technique above suggested by Boas, it is straightforward to prove the Cayley-
Hamilton theorem for any diagonalizable matrix M .
If f (λ) is the characteristic polynomial, then it follows that f (λ) = 0 for any of the eigenvalues
of M , since this is how the eigenvalues are determined. Consider the diagonal matrix D with the
eigenvalues on the diagonal. Then, noting that Dn has values λni on the i-th diagonal element
and zero on all off-diagonal elements, we have
 
f (λ1 ) 0 0 ··· 0

 0 f (λ2 ) 0 ··· 0 
f (D) = 
 0 0 f (λ3 ) · · · 0  = 0. (20)
 .. .. .. . . ..
 . . . . . 
0 0 0 ··· f (λn )

12
Next one writes M = CDC −1 and notes that M k = (CDC −1 )k = (CDC −1 ) (CDC −1 ) · · · (CDC −1 ) =
CDk C −1 , since the internal factors of C −1 C cancel. Hence, because f is a polynomial, we have

f (M ) = f (CDC −1 ) = Cf (D)C −1 = 0

since f (D) = 0 as shown above. However, this proof is not applicable to a matrix that is not
diagonalizable. Nonetheless, one can show that the theorem is valid even for a non-diagonalizable
matrix.

13