A S URVIVAL G UIDE TO

L INEAR A LGEBRA
D ONE R IGHT
2015 S PRINGER E DITION
Sheldon Axler

PARTIAL SCRUTINY,
C OMMENTS , S UGGESTIONS AND E RRATA
José Renato Ramos Barbosa
2019

Departamento de Matemática
Universidade Federal do Paraná
Curitiba - Paraná - Brasil
[email protected]

1
================================================================================
================================================================================
1
================================================================================
================================================================================
EX. 20, p. 25
On the one hand, note that each vector u ∈ U can be written as u = xu1 + yu2 with u1 = (1, 1, 0, 0), u2 =
(0, 0, 1, 1) and x, y ∈ F. Then, since u1 and u2 are linearly independent (because none of them is a scalar
multiple of the other), {u1 , u2 } is a basis of U.1 On the other hand, suppose that w1 = (0, 1, 0, 0) and w2 =
(0, 0, 1, 0) spans W.2 So w1 and w2 are linearly independent since none of them is a scalar multiple of the other.
Now consider B = {u1 , u2 , w1 , w2 }. Therefore, on the one hand, the equality ( x, x, y, y) = (0, a, b, 0) implies
that x = y = a = b = 0 and thus U ∩ W = {(0, 0, 0, 0)}. On the other hand, by considering the matrix whose
columns are u1 , u2 , w1 and w2 written as column vectors, one has
   
1 0 0 0 1 0 0 0
 1 0 1 0   0 1 0 0 
 0 1 0 1  = − det  0 1 0 1 
det    

0 1 0 0 1 0 1 0
 
1 0 0 0
 0 1 0 0 
= det 
 1 0 1 0 


0 1 0 1
=1

and thus u1 , u2 , w1 and w2 are linearly independent, that is, B is a basis of U + W. Hence dim(U + W ) = 4 =
dim F4 .
================================================================================
EX. 24, p. 26
On the one hand, if f ∈ Ue ∩ Uo , then

f ( x ) = f (− x )
= − f (x)

for each x ∈ R and thus f = 0. On the other hand, if f ∈ RR is arbitrary, then

f ( x ) + f (− x ) f ( x ) − f (− x )
f (x) = +
2 2
for each x ∈ R with the first summand in Ue and the second summand in Uo .
================================================================================

1 See CHAPTER 2.
2 Idem.

2
================================================================================
================================================================================
2
================================================================================
================================================================================
EX. 1, p. 37
On the one hand,
span(v1 − v2 , v2 − v3 , v3 − v4 , v4 ) ⊂ span(v1 , v2 , v3 , v4 )
since

v1 − v2 = 1 · v1 − 1 · v2 + 0 · v3 + 0 · v4
v2 − v3 = 0 · v1 + 1 · v2 − 1 · v3 + 0 · v4
v3 − v4 = 0 · v1 + 0 · v2 + 1 · v3 − 1 · v4
v4 = 0 · v1 + 0 · v2 + 0 · v3 + 1 · v4 .

On the other hand,

span(v1 , v2 , v3 , v4 ) ⊂ span(v1 − v2 , v2 − v3 , v3 − v4 , v4 )
since

v1 = 1 · ( v1 − v2 ) + 1 · ( v2 − v3 ) + 1 · ( v3 − v4 ) + 1 · v4
v2 = 0 · ( v1 − v2 ) + 1 · ( v2 − v3 ) + 1 · ( v3 − v4 ) + 1 · v4
v3 = 0 · ( v1 − v2 ) + 0 · ( v2 − v3 ) + 1 · ( v3 − v4 ) + 1 · v4
v4 = 0 · ( v1 − v2 ) + 0 · ( v2 − v3 ) + 0 · ( v3 − v4 ) + 1 · v4 .

Therefore

span(v1 − v2 , v2 − v3 , v3 − v4 , v4 ) = span(v1 , v2 , v3 , v4 )
= V.

================================================================================
EX. 6, p. 37
Consider ai ∈ F, i = 1, 2, 3, 4, such that

a1 (v1 − v2 ) + a2 (v2 − v3 ) + a3 (v3 − v4 ) + a4 v4 = 0. (1)

So
a1 v1 + ( a2 − a1 ) v2 + ( a3 − a2 ) v3 + ( a4 − a3 ) v4 = 0.
Thus, due to the fact that v1 , v2 , v3 , v4 is linearly independent,

a1 = 0,
a2 − a1 = 0,
a3 − a2 = 0,
a4 − a3 = 0.

Then
ai = 0, i = 1, 2, 3, 4. (2)
Therefore, since (1) =⇒ (2), v1 − v2 , v2 − v3 , v3 − v4 , v4 is linearly independent.
================================================================================
EX. 7, p. 37
Consider ai ∈ F, i = 1, 2, 3, . . . , m, such that

a1 (5v1 − 4v2 ) + a2 v2 + a3 v3 + · · · + am vm = 0. (3)

So
5a1 v1 + ( a2 − 4a1 ) v2 + a3 v3 + · · · + am vm = 0.

3
Thus, due to the fact that v1 , v2 , . . . , vm is linearly independent,

5a1 = 0,
a2 − 4a1 = 0,
a3 = 0,
..
.
am = 0.

Then
ai = 0, i = 1, 2, 3, . . . , m. (4)
Therefore, since (3) =⇒ (4), 5v1 − 4v2 , v2 , v3 , . . . , vm is linearly independent.
================================================================================
EX. 6, p. 43
See EX. 1 and EX. 6, p. 37.
================================================================================
EX. 8, p. 43
First, let us show that the list of m + n vectors spans V. In order to do that, consider v ∈ V. Since V = U + W,
there exist u ∈ U and w ∈ W for which
v = u + w. (5)
On the other hand, the bases of U and W give us scalars α1 , . . . , αm , β 1 , . . . , β n in F such that

u = α1 u1 + · · · + α m u m and w = β 1 w1 + · · · + β n w n . (6)

Substituting (6) in (5), it folllows that v is a linear combination of the m + n vectors.

Now, let us show that the list of m + n vectors is linearly independent. So, consider

α 1 u 1 + · · · α m u m + β 1 w1 + · · · + β n w n = 0 (7)

with, as before, scalars in F. Denote u and w as in (6). Then 0 = u + w and, due to the fact that the sum is
direct,3
u = w = 0. (8)
Therefore, substituting (8) in (6) and observing that the list of m vectors in U and the list of n vectors in W are
both linearly independent, all of the m + n scalars in (7) must be equal to zero.
================================================================================
EX. 16, p. 49
It is a direct consequence of the following generalization of EX. 8, p. 43:

If n o
B i = u i1 , . . . , u i n
i

is a basis of Ui , i = 1, . . . , m, then

B = ∪im=1 Bi

is a basis of U = ⊕im=1 Ui .

In fact, firstly, let u ∈ U. Then u = ∑im=1 ui with ui ∈ Ui , i = 1, . . . , m. So, since ui is a linear combination of
the elements of Bi , i = 1, . . . , m, B clearly spans U. Now, it remains to show that the elements of B are linearly
independent. Hence we first write

c11 u11 + · · · + c1n u1n + · · · + cm1 um1 + · · · + cmnm umnm = 0

1 1

with ci1 ui1 + · · · + cin uin ∈ Ui and ci j ∈ F, i = 1, . . . , m and j = 1, . . . , ni . Then ci1 ui1 + · · · + cin uin = 0,
i i i i
i = 1, . . . , m, by 1.44, p. 23. Therefore, since Bi is a basis for Ui , i = 1, . . . , m, ci j = 0 for every i and j.
================================================================================

3 See 1.44, p. 23.

4
================================================================================
================================================================================
3
================================================================================
================================================================================
EX. 3, p. 57
Let ek denote the k-th vector of the standard basis of Fn . Substitute

T (ek ) = ( A1,k , . . . , Am,k )

in !
n n
T ∑ xk ek = ∑ xk T ( ek ) .
k =1 k =1
================================================================================
EX. 4, p. 57
Consider a1 , . . . , am ∈ F such that
m
∑ a j v j = 0.
j =1

So
!
m m
∑ aj T ∑ aj vj


vj = T
j =1 j =1

= T (0)
=0
by the linearity of T and 3.11. Now use the hypothesis that Tv1 , . . . , Tvm is linearly independent.
================================================================================
EX. 6, p. 67
dim R5 = dim null T + dim range T by 3.22, p. 63, which implies that

5 = 2 (dim range T ) ,

contradicting the fact that dim range T is a non-negative integer.

================================================================================
EX. 9, p. 67
Consider a1 , . . . , an ∈ F such that
n
∑ a j T v j = 0.


j =1

Thus !
n
T ∑ aj vj = T (0)
j =1

by the linearity of T and 3.11, p. 57. Then, since T is injective,

n
∑ a j v j = 0.
j =1

Now use the hypothesis that v1 , . . . , vn is linearly independent.

================================================================================
EX. 10, p. 68
Let w ∈ range T. Thus there exist a1 , . . . , an ∈ F such that
!
n
w=T ∑ aj vj
j =1
n
∑ aj T


= vj
j =1

5
by the linearity of T.
================================================================================
EX. 14, p. 68
dim R8 = dim U + dim range T by 3.22, p. 63, which implies that
dim range T = 8 − 3
= 5.
Therefore range T = R5 , which implies that T is surjective.
================================================================================
Comment, p. 83, l. -1
We can use 3.59 to prove 3.61 provided that both Fm,n and L(V, W ) are finite-dimensional. The former satisfies
this condition by 3.40, p. 74. Concerning the latter, use 3.22, p. 63, with M−1 in place of T. Therefore
range M−1 = L(V, W )
is finite-dimensional.
================================================================================
Erratum, p. 85, ll. 1–9
M (vk ) should be M ( Tvk ), four times.
================================================================================
Comment, p. 97, 3.90
Note that T̃ ◦ π = T.
π
V V/(null T )
T̃
T
W
================================================================================
Comment, p. 98, 3.91, Proof, (b)
null T̃ = 0 is an abuse of notation. It means null T̃ = {null T }.
================================================================================
EX. 15, p. 100
The condition ϕ 6= 0 implies that range ϕ = F, which is a one-dimensional space. Now use 3.91(d), p. 98.
================================================================================
Erratum, p. 102, 3.98, Proof, l. -6
F should be F.
================================================================================
Comments, p. 112, 3.117, Proof
• 1st paragraph, 2nd sentence
Based on (the Proof of) 2.31, pp. 40–41, let
Tv1 , . . . , Tvr , r ≤ n,
be a basis of span( Tv1 , . . . , Tvn ). Thus
span( Tv1 , . . . , Tvr ) = span( Tv1 , . . . , Tvn )
and, if
w = a1 Tv1 + · · · + ar Tvr
with a1 , . . . , ar ∈ F, then, by 3.62, p. 84,


a1
M(w) =  ...  .
 

ar
Now, clearly,
span(M( Tv1 ) , . . . , M( Tvr )) = span(M( Tv1 ) , . . . , M( Tvn ))
and
span( Tv1 , . . . , Tvr ) 3 w 7→ M(w) ∈ Fr,1
is an isomorphism.

6
 • 2nd paragraph, 1st sentence
See EX. 10, p. 68.
================================================================================
EX. 32, p. 115

• (a) =⇒ (c) (and (b))

By 3.69,4 T is injective. So, by 3.16,5 null T = {0}. Then, by 3.22,6 dim range T = n. Thus, by 3.117 and
3.40,7 the column rank of M( T ) = dim Fn,1 . Therefore, by 3.115 and 3.64,8 M( Tu1 ) , . . . , M( Tun ) is a
basis of Fn,1 .
• (c) =⇒ (b)
Use 2.42.9
• (b) =⇒ (a)
By 3.115 and 3.117,10 dim range T = n. So, by 3.22 and 3.16,11 T is injective. Then, by 3.69,12 T is
invertible.

• (a) =⇒ (e) =⇒ (d) =⇒ (a)

It follows from:
* T is invertible ⇐⇒ T 0 is invertible;13
* M( T 0 ) = (M( T ))t by 3.114;14
* (a) =⇒ (c) =⇒ (b) =⇒ (a) with T 0 in the role of T and M( T 0 ) in the role of M( T ).

================================================================================

4 See p. 87.
5 See p. 61.
6 See p. 63.
7 See p. 112 and p. 74.
8 See p. 111 and p. 85.
9 See p. 46.
10 See pp. 111–112.
11 See p. 63 and p. 61.
12 See p. 87.
13 See 3.108 and 3.110, pp. 107–108.
14 See p. 110.

7
================================================================================
================================================================================
4
================================================================================
================================================================================
Erratum, p. 123, last paragraph, 1st sentence
“..., uses analysis its proof.” should be “..., uses analysis in its proof.”.
================================================================================
Erratum, p. 125, 4.14, Proof, 3rd paragraph, 3rd sentence
“... So we need only show that ...” should be “... So we need only to show that ...”.
================================================================================
EX. 2 and EX. 3, p. 129
The answer to both questions is “no”!15
================================================================================

15 Consider, for example, the sum of p( x ) = x2 + x and q( x ) = − x2 + 1.

8
================================================================================
================================================================================
5
================================================================================
================================================================================
Erratum, p. 134, 5.6
F should be F.
================================================================================
EX. 2, p. 138
v ∈ null S =⇒ S(v) = 0
=⇒ T (S(v)) = T (0)
=⇒ TS(v) = 0
=⇒ ST (v) = 0
=⇒ S( T (v)) = 0
=⇒ T (v) ∈ null S.
================================================================================
EX. 3, p. 139
v ∈ range S =⇒ ∃w ∈ V such that v = S(w)
=⇒ T (v) = T (S(w))
=⇒ T (v) = TS(w)
=⇒ T (v) = ST (w)
=⇒ T (v) = S( T (w))
=⇒ T (v) = S(u) with u = T (w)
=⇒ ∃u ∈ V such that T (v) = S(u)
=⇒ T (v) ∈ range S.
================================================================================
EX. 4, p. 139
m m
v∈ ∑ Ui =⇒ v = ∑ ui with ui ∈ Ui , i = 1, . . . , m
i =1 i =1
!
m
=⇒ T (v) = T ∑ ui
i =1
m
=⇒ T (v) = ∑ T ( ui ) with T (ui ) ∈ Ui , i = 1, . . . , m, since each Ui is invariant under T
i =1
m
=⇒ T (v) ∈ ∑ Ui .
i =1
================================================================================
EX. 33, p. 142
First, since range T is invariant under T,16 we can consider T/(range T ).17 So, for v ∈ V,
( T/(range T ))(v + range T ) = T (v) + range T
= range T
since T (v) ∈ range T.
================================================================================
Comment, p. 145, 5.21, Proof, penultimate sentence
If the m operators are injective, the image of a nonzero vector under T − λ j I is obtained for each index j.
Therefore 

Πnj=1 T − λ j I v 6= 0.
16 See 5.3(d), p. 132.
17 See 5.14, p. 137.

9
================================================================================
Comment, p. 149, Proof 1, l. 7
To be more specific, ‘... (see 3.69) ...’ should be ‘... (see 5.6) ...’.18
================================================================================
Comment, p. 150, Proof 2, sentence that begins with ‘Unraveling’
Consider a2 , . . . , a j ∈ F, j = 2, . . . , n, such that


Tv j + U = ( T/U ) v j + U


= a2 ( v2 + U ) + · · · + a j v j + U



= (( a2 v2 )+ U ) + · · · + a j v j + U


= a2 v2 + · · · + a j v j + U,

where the first equality comes from 5.14, p. 137, and the last two equalities come from 3.86, p. 96.
Thus, for each j ∈ {2, . . . , n}, since
Tv j − a2 v2 + · · · + a j v j ∈ U,19

there exists a1 ∈ F such that

Tv j = a1 v1 + a2 v2 + · · · + a j v j .
================================================================================
Comment - Erratum, p. 151

• 5th paragraph (“To prove the other direction, ...”)

Tv1 = 0 =⇒ null T 6= {0}

=⇒ T is not injective (by 3.16, p. 61)
=⇒ T is not invertible (by 3.69, p. 87),

which contradicts the assumption that T is invertible.

• 6th paragraph: (“Let 1 < j ≤ n, ...”)

“... T restricted to dim span v1 , . . . , v j ...” should be “... T restricted to span v1 , . . . , v j ...”.

================================================================================
Comment, p. 152, 5.32, Proof, last sentence
Consider 5.6, p. 134.
================================================================================
EX. 3, p. 153
Suppose that T − I 6= 0. So there is a nonzero vector v ∈ V such that

w := ( T − I )(v)
6= 0.

Then, by 5.20, p. 144,

( T + I )(w) = ( T + I )( T − I )(v)
= ( T 2 − I )(v)
=0

since T 2 = I. Therefore Tw = −w, which implies that −1 is an eigenvalue of T!

================================================================================
EX. 7, p. 153
By 5.20, p. 144,

T 2 − 9I = ( T + 3I )( T − 3I )
= ( T − 3I )( T + 3I ).
18 See p. 134.
19 See 3.85, p. 95.

10
Therefore, by 5.6, p. 134,

9 is an eigenvalue of T 2 ⇐⇒ T 2 − 9I is not injective

⇐⇒ ( T − 3I ) or ( T + 3I ) is not injective
⇐⇒ 3 or − 3 is an eigenvalue of T.

================================================================================
EX. 14, p. 154
Define T ∈ L F2 by T ( x, y) = (y, x ). Obviously, T 2 = I and the matrix of T with respect to the standard basis

is  
0 1
M( T ) = .
1 0
================================================================================
Comment, p. 155, sentence immediately preceding 5.37
See 5.6, p. 134, and 3.16, p. 61.
================================================================================
Erratum, p. 156, 5.38, Proof, 1st
sentence
‘E(λ, T )’ should be ‘E λ j , T ’.
================================================================================
EX. 6, p. 160
Consider 2.39, p. 45, and m = dim V in 5.10, p. 136. Now let λi0 be an eigenvalue of S with Svi = λi0 vi ,
i = 1, . . . , dim V. Therefore, for each basis vector vi ,

STvi = S( Tvi )
= S ( λi vi )
= λi Svi
= λi λi0 vi
= λi0 λi vi
= λi0 Tvi
= T λi0 vi

= T (Svi )
= TSvi .

================================================================================
EX. 8, p. 160
Suppose T − 2I and T − 6I are not invertible. Thus 2 and 6 are eigenvalues of T by 5.6, p. 134. So, since
dim E(8, T ) = 4, dim F5 = 5 and

dim E(2, T ) + dim E(6, T ) + dim E(8, T ) ≤ dim F5

by 5.38, p. 156, it follows that

E(2, T ) = {0} or E(6, T ) = {0} ,
which is a contradiction.20
================================================================================
EX. 9, p. 161
0 is in both eigenspaces and

Tv = λv ⇐⇒ T −1( Tv) = T −1(λv)

⇐⇒ v = λT −1 v
⇐⇒ T −1 v = λ−1 v.

================================================================================

20 See the sentence immediately preceding 5.37, p. 155.

11
================================================================================
================================================================================
6
================================================================================
================================================================================
EX. 13 and EX. 14, p. 176
Let V be an arbitrary inner product space. Consider u, v ∈ V are nonzero. Thus, by the Cauchy-Schwarz
Inequality,21
−||u|| ||v|| ≤ hu, vi ≤ ||u|| ||v||.

Then, since ||u|| and ||v|| are positive,

hu, vi
−1 ≤ ≤ 1.
||u|| ||v||

Now let θ = θ (u, v) be the unique number in [0, π ] (measured in radians) such that

hu, vi
cos θ = .
||u|| ||v||

1
hu,vi
||u|| ||v||

0 π
θ (u, v)

−1

hu,vi
Therefore hu, vi = ||u|| ||v|| cos θ and θ = arccos ||u|| ||v|| .
================================================================================
EX. 16, p. 176
Use the Parallelogram Equality.22
================================================================================
EX. 17, p. 177
If x < 0 and y < 0, then ||( x, y)|| < 0. However, by 6.8 and 6.3,23 a norm assigns to any v ∈ V a non-negative
real number ||v||.
================================================================================
EX. 30, p. 179
See 3.69, p. 87.
================================================================================
EX. 2, p. 189
Consider
m
||v||2 = ∑ |hv, e j i|2
j =1

and
m
u= ∑ hv, e j ie j .
j =1

21 Cf. 6.15, p. 172.

22 Cf. p. 174.
23 See pp. 168 and 166.

12
Thus
m
hu, vi = h ∑ hv, e j ie j , vi
j =1
m
= ∑ hv, e j ihe j , vi
j =1
m
= ∑ hv, e j ihv, e j i
j =1
m
= ∑ |hv, e j i|2
j =1

= ||v||2
= hv, vi.

So hu − v, vi = 0. Then, by 6.13,24
||u||2 = ||u − v||2 + ||v||2 .
Therefore, since ||u||2 = ||v||2 by 6.25,25
||u − v||2 = 0,
that is u = v, which implies that
v ∈ span(e1 , . . . , em ).
For the converse, since e1 , . . . , em is orthonormal, it is also linearly independent in U = span(e1 , . . . , em ). Then
e1 , . . . , em is an orthonormal basis of U. Therefore, if v ∈ U,
m
||v||2 = ∑ |hv, e j i|2
j =1

by 6.30, p. 182.
================================================================================
Comments, pp. 193–194
Concerning the example (where U is a line or a plane in R3 ), since U is not necessarily a subspace of V by 6.45,
it is not necessary for U to contain the zero vector. Similarly, that is why

U ∩ U ⊥ ⊃ {0}

does not necessarily hold for 6.46(d). By the way, the proof of 6.46(d) for U ∩ U ⊥ = ∅ is trivial.
================================================================================
EX. 1, p. 201
On the one hand, from {v1 , . . . , vm } ⊂ span(v1 , . . . , vm ), it follows that

(span(v1 , . . . , vr ))⊥ ⊂ {v1 , . . . , vm }⊥

by 6.46(e), p. 193. On the other hand, consider v ∈ {v1 , . . . , vm }⊥ and let u be a linear combination of v1 , . . . , vm ,
that is, consider c1 , . . . , cm ∈ F such that u = c1 v1 + . . . + cm vm . So

hv, ui = c1 hv, v1 i + · · · + cm hv, vm i

= c1 · 0 + · · · + c m · 0
= 0.

Then v ∈ (span(v1 , . . . , vm ))⊥ . Therefore

{v1 , . . . , vm }⊥ ⊂ (span(v1 , . . . , vm ))⊥ .

================================================================================
24 See p. 170.
25 See p. 180.

13
EX. 4, p. 201
For v1 = (1, 2, 3, −4) and v2 = (−5, 4, 3, 2), consider

u1 = v1 and u2 = v2 − Pspan(v1 ) (v2 )

as in example 6.54, p. 196. So  

u1 u2
,
||u1 || ||u2 ||
is an orthonormal basis of U. On the other hand, since

U ⊥ = { v1 , v2 } ⊥

by EX. 1, p. 201, and dim U ⊥ = 2 by 6.50, p. 195, if w1 , w2 is linearly independent with

hwi , v j i = 0, i, j ∈ {1, 2} ,

then {w1 , w2 } is a basis of U ⊥ . So let w = ( a, b, c, d) be a vector in R4 for which

hw, v j i = 0, j = 1, 2.

Then, since 
a + 2b + 3c − 4d = 0
−5a + 4b + 3c + 2d = 0
and 
a = 15c − 14d
b = − 9c + 9d
are equivalent systems,

w = (15c − 14d, −9c + 9d, c, d)

= c(15, −9, 1, 0) + d(−14, 9, 0, 1)

for all real numbers c and d. Therefore, if

w1 = (15, −9, 1, 0) and w2 = (−14, 9, 0, 1) − Pspan(w1 ) (−14, 9, 0, 1),

 
w1 w2
,
||w1 || ||w2 ||
is an orthonormal basis of U ⊥ .
================================================================================

14
================================================================================
================================================================================
7
================================================================================
================================================================================
Comment, p. 204
Consider
h−,wi
W 3 x 7−→ h x, wi.
So h−, wi ∈ L(W, F). Then
ϕ = h−, wi ◦ T ∈ L(V, F). (9)
On the other hand, by 6.42,26 let u be the vector in V such that

ϕ = h−, ui. (10)

Therefore, denoting u by T ∗ w, T ∗ is well-defined and, via (9) and (10),

h Tv, wi = hv, T ∗ wi.

================================================================================
Comment, p. 213, 7.21
Since T − λI is normal, another Proof comes from the small box in the bottom left corner of p. 212. In fact, note
that
null ( T − λI ) = null T ∗ − λI .

================================================================================
EX. 2, p. 214
Firstly, consider the following result

If T is invertible, then T ∗ is invertible

and  ∗
( T ∗ ) −1 = T −1 .

In fact, by 7.6, items (d) and (e), p. 206, if

TT −1 = I
= T −1 T,

then
 ∗
T −1 T∗ = I
 ∗
= T ∗ T −1 .

Therefore, by 5.6, p. 134, and by 7.6, items (a), (b) and (c), p. 206,

λ is an eigenvalue of T ⇐⇒ T − λI is not invertible

⇐⇒ ( T − λI )∗ = T ∗ − λI is not invertible
⇐⇒ λ is an eigenvalue of T ∗ .

================================================================================
EX. 4, p. 214
It is a direct consequence of 7.7, p. 207.
================================================================================
Comments, p. 220, Proof
26 See p. 188.

15
 • m+M ≥ 1
In fact, since V 6= {0},

n = dim V
≥ 1.

Therefore a0 + a1 x + · · · + an x n is a nonconstant polynomial.

• m>0
In fact, if there is no ( T − λi I ) factor, then M > 0 and the invertible operator
M  
c ∏ T2 + bj T + c j I
j =1

is the zero vector at v 6= 0, which is a violation of 3.69, p. 87.

================================================================================
Comment, p. 222, Proof, (c) =⇒ (a)
Suppose dim V = n and let B be the orthonormal basis mentioned in the 1st paragraph. Then, since M( T, B) ∈
Rn,n is diagonal,
M( T ∗ , B) = M( T, B)
by 7.10, p. 208. Furthermore, from the fact that

M(−, B) : L(V ) −→ Rn,n

S 7−→ M(S, B)

is an isomorphism,27 it follows that T ∗ = T.

================================================================================
EX. 6, p. 223
Let T be a normal operator on a complex inner product space and, firstly, suppose that

T ∗ = T.

Furthermore, consider that λ is an eigenvalue of T. Then, by 7.21, p. 213, λ is an eigenvalue of T and

E(λ, T ) = E λ, T ∗




= E λ, T .

So, by the Complex Spectral Theorem, p. 218, T is diagonalizable. Then, by 5.41(d), p. 157, λ = λ, that is,

λ ∈ R.

T HE PREVIOUS REASONING USES THE NORMALITY

HYPOTHESIS . A MORE STRAIGHTFORWARD PROOF
FOLLOWS FROM 7.13, P. 210.

Now, for the converse, suppose that

λ is an eigenvalue of T =⇒ λ ∈ R.

Hence, by the Complex Spectral Theorem, there exists an orthonormal basis B such that

D = M( T, B)

is diagonal (with its (real) eigenvalues lying on the main diagonal). Then, by 7.10, p. 208,
t
M( T ∗ , B) = D
= D,
27 Cf. p. 83.

16
which implies that
T ∗ = T.
================================================================================
EX. 13, p. 224
Consider the following modifications in the Proof of page 222:

• In the 1st paragraph, right before the last sentence, insert the sentence: Then T is normal. ;28

• In the 2nd paragraph, replace the word real with the word complex ;

• In the 3rd paragraph, replace:

– the word self-adjoint with the word normal , twice;

– the result 7.27 with the result 5.21 ;

– the result 7.28(c) with the result 9.30(d) .
================================================================================
Comment, p. 226, Proof, (b) =⇒ (c)
Let B = {e1 , . . . , en }. Then, since M( R, B) ∈ Rn,n is diagonal,29

M( R∗ , B) = M( R, B)
by 7.10, p. 208. Furthermore, from the fact that

M(−, B) : L(V ) −→ Fn,n

S 7−→ M(S, B)

is an isomorphism,30 it follows that R∗ = R, that is, R is self-adjoint.

Now, the positivity of R follows from the fact that
p
h Rei , e j i = λi hei , e j i
is nonnegative for each indices i, j.
================================================================================
EX. 4, p. 231
Firstly, T ∗ T and TT ∗ are self-adjoint operators since, by 7.6, (c) and (e).31

( T ∗ T )∗ = T ∗ ( T ∗ )∗
= T∗ T
and

( TT ∗ )∗ = ( T ∗ )∗ T ∗
= TT ∗ .
Now, for v ∈ V and w ∈ W, the positivity follows from

h( T ∗ T ) v, vi = h T ∗ ( Tv), vi
= h Tv, Tvi
≥0
and

h( TT ∗ ) w, wi = h T ( T ∗ w) , wi
= h T ∗ w, T ∗ wi
≥ 0.
28 Every self-adjoint operator is normal!
29 For
p
each j ∈ {1, . . . , n}, the diagonal entry λ j is the square root of a nonnegative number. Now see the last paragraph of p. 225.
30 Cf. p. 83.
31 See p. 206.

17
================================================================================
EX. 10, p. 232
By 7.42, p. 229, (a) ⇐⇒ (g). Then one can replace S by S∗ in (a), (b), (c) and (d) of 7.42.
================================================================================
Comment, p. 233
Concerning the sentence preceding 7.45, see EX. 4, p. 231.
================================================================================
Comment, p. 236, 7.50, Solution
• 1st sentence

( z 1 , z 2 , z 3 , z 4 ) , T ∗ ( w1 , w2 , w3 , w4 ) = T ( z 1 , z 2 , z 3 , z 4 ) , ( w1 , w2 , w3 , w4 )
= (0, 3z1 , 2z2 , −3z4 ) , (w1 , w2 , w3 , w4 )
= 3z1 w2 + 2z2 w3 − 3z4 w4
= (z1 , z2 , z3 , z4 ) , (3w2 , 2w3 , 0, −3w4 )i

w

T ∗ (w1 , w2 , w3 , w4 ) = (3w2 , 2w3 , 0, −3w4 )
w


T ∗ T (z1 , z2 , z3 , z4 ) = T ∗ (0, 3z1 , 2z2 , −3z4 )

= (9z1 , 4z2 , 0, 9z4 ) .

• 2nd sentence √
On the one hand, T ∗ T denotes the unique √ positive square root of the (self-adjoint) positive operator
T ∗ T.32 On the other hand, the operator T ∗ T presented in the solution is (self-adjoint)33 positive since
√
T ∗ T (z1 , z2 , z3 , z4 ) , (z1 , z2 , z3 , z4 ) = (3z1 , 2z2 , 0, 3z4 ) , (z1 , z2 , z3 , z4 )
= 3z1 z1 + 2z2 z2 + 3z4 z4
= 3 | z1 |2 + 2 | z2 |2 + 3 | z4 |2
≥ 0.

Furthermore
√ √  √ √ 
T ∗ T T ∗ T ( z1 , z2 , z3 , z4 ) = T ∗ T T ∗ T ( z1 , z2 , z3 , z4 )
√
= T ∗ T (3z1 , 2z2 , 0, 3z4 )
= (3 (3z1 ) , 2 (2z2 ) , 0, 3 (3z4 ))
= T ∗ T ( z1 , z2 , z3 , z4 ) .

================================================================================
Comment, p. 238, 7.52, Proof
If B = {e1 , . . . , en }, then
 √  2
∗
M T T, B = M( T ∗ T, B)
√ 2
since T∗ T = T ∗ T. Therefore, for each index j,
√
T ∗ Te j = α j e j =⇒ α2j = λ j
q
=⇒ α j = λ j .
32 See p. 233. √ 
33 If B is the standard basis, M T ∗ T, B is diagonal with real entries!

18
================================================================================
EX. 1, p. 238
In the exercise,
Tv = hv, ui x. (11)
So, on the one hand,

hv, T ∗ wi = h Tv, wi
= hhv, ui x, wi
= hv, uih x, wi
= hv, hw, x iui

w
w


T ∗ v = hv, x iu (12)
w
w


( T ∗ T ) v = T ∗ ( Tv) (13)
= T ∗ (hv, ui x ) (14)
= hv, ui T ∗ x (15)
= hv, uih x, x iu (16)
= || x ||2 hv, uiu. (17)
√
On the other hand, T ∗ T denotes
√ the unique positive square root of the (self-adjoint) positive operator T ∗ T.34
Furthermore, the operator T T presented in the exercise is self-adjoint,35 positive (since
∗

√ D || x || E
T ∗ Tv, vi = hv, uiu, v
||u||
|| x ||
= hv, uihu, vi
||u||
|| x ||
= |hv, ui|2
||u||
≥0

for each v ∈ V) and a root of T ∗ T (since, for each v ∈ V,

√ 2 √ √ 
T∗ T v= T∗ T T ∗ Tv
√ || x ||
 
= T∗ T hv, uiu
||u||
|| x || √
= hv, ui T ∗ Tu
||u||
|| x || || x ||
= hv, ui hu, uiu
||u|| ||u||
|| x ||2
= hv, ui||u||2 u
||u||2
= (T∗ T ) v

by (17)).
================================================================================
34 See p. 233.
35 Use (11) and (12)

19
EX. 3, p. 239
By 7.45, p. 233, there exists an isometry Σ ∈ L(V ) such that
q
T ∗ = Σ ( T ∗ )∗ T ∗ (18)
√
= Σ TT ∗ , (19)

where (19) holds by 7.6(c).36 So

T = ( T ∗ )∗
 √ ∗
= Σ TT ∗
√ ∗
= TT ∗ Σ∗
√
= TT ∗ Σ∗ ,

where the first equality holds by 7.6(c), the second equality holds by (19) (as stated above), the third equality
holds by 7.6(e),37 and the last equality holds because each positive operator is self-adjoint by definition. Now
use S to denote Σ∗ .
================================================================================
EX. 17, p. 240
(a)

hv, T ∗ wi = h Tv, wi
D n E
= ∑ s j hv, e j i f j , w
j =1
n
= ∑ s j hv, e j ih f j , wi
j =1
n
= ∑ s j hw, f j ihv, e j i
j =1
D n E
= v, ∑ s j hw, f j ie j
j =1
w
w

n
T∗ w = ∑ s j hw, f j ie j
j =1
√
since, for each j ∈ {1, . . . , n}, s j is an eigenvalue of T ∗ T, which is self-adjoint.38
(b)

( T ∗ T ) v = T ∗ ( Tv)
!
n
=T ∗
∑ si hv, ei i fi
i =1
n
= ∑ si hv, ei iT ∗ fi
i =1
n
= ∑ si hv, ei isi ei
i =1
n
= ∑ s2i hv, ei iei .
i =1
36 See p. 206.
37 See p. 206.
38 Cf. 7.13, p. 210.

20
(c) Consider B = {e1 , . . . , en }. Then

s21
 
zeros
M( T ∗ T, B) = 
 .. 
. 
zeros s2n

by (b). Now use 7.52, p. 238.

(d) Firstly, note that, for each j ∈ {1, . . . , n}, s j 6= 0 if T −1 exists. In fact, by 3.43, p. 75, if B = {e1 , . . . , en } and
B 0 = { f 1 , . . . , f n }, then
   
M T −1 , B 0 , B M T, B , B 0 = M T −1 T, B

= M ( I, B)

does not hold if some diagonal entry of the matrix M( T, B , B 0 ) shown on page 237, last line, is zero.
Now, consider the operator ‘T −1 ’ presented in the exercise. Then
 
T −1 T v = T −1 ( Tv)
!
n
= T −1 ∑ s j hv, e j i f j
j =1
n
∑ s j hv, e j iT −1


= fj
j =1
n h f j , f j ie j
= ∑ s j hv, e j i sj
j =1
n
= ∑ hv, e j ie j
j =1

=v

and
   
TT −1 v = T T −1 v
!
n
hv, f j ie j
=T ∑
j =1
sj


n h v, f i T e
j j
=∑
j =1
sj
n
hv, f j is j f j
= ∑ sj
j =1
n
= ∑ hv, f j i f j
j =1

= v.

Therefore T −1 T = I and TT −1 = I.

21
================================================================================
================================================================================
8
================================================================================
================================================================================
Comment, p. 247, Proof, 2nd paragraph
By 5.20, p. 144, concerning the operator

( T − λ1 I ) k ( T − λ2 I ) n · · · ( T − λ m I ) n ,
the factors ( T − λ1 I )k , ( T − λ2 I )n , etc, commute with each other.
================================================================================
Comment, p. 248, Proof
There is some positive integer p such that N p = 0. Therefore
 
null N dim V = null ( N p )
=V
by 8.2 and 8.4, pp. 242–3.
================================================================================
EX. 4, p. 250
Firstly, note that
E(α, T ) ∩ G ( β, T ) = {0} . (20)
In fact, consider v ∈ E(α, T ) ∩ G ( β, T ). So

0 = ( T − βI )dim V v
= ( T − βI )dim V −1 ( T − βI )v
= (α − β)( T − βI )dim V −1 v.
Repeated use of these steps eventually give

0 = (α − β)dim V v,
and since α 6= β, we must have v = 0. Then (20) holds.
Now consider
G (α, T ) ∩ G ( β, T ) = W.
By 8.21, (b), T |W ∈ L(W ). As such, if W 6= {0}, provided that V is a complex vector space, T |W has at least
one eigenvalue with an associated eigenvector v ∈ W, which contradicts the fact that G ( β, T ) contains all the
eigenvectors of T corresponding to β but no other eigenvectors (by (20)). Therefore W = {0}.
================================================================================
EX. 5, p. 250
Firstly, note that:
• v 6= 0 since
T m−1 v 6= 0; (21)
• T m v = 0 implies that
T m+ j v = 0, j = 0, 1, 2, . . . . (22)
Now consider
m −1
∑ a j +1 T j v = 0 (23)
j =0

with a j+1 ∈ F, j = 0, 1, . . . , m − 1. Then

!
m −1 m −1
T m −1
∑ j
a j +1 T v = T m−1 0 =⇒ ∑ a j +1 T m −1+ j v = 0
j =0 j =0

=⇒ a1 T m−1 v = 0
=⇒ a1 = 0

22
by (22) and (21). Then (23) becomes
m −1
∑ a j+1 T j v = 0.
j =1

Therefore
!
m −1 m −1
T m −2
∑ j
a j +1 T v = T m−2 0 =⇒ ∑ a j +1 T m −2+ j v = 0
j =1 j =1

=⇒ a2 T m−1 v = 0
=⇒ a2 = 0

by (22) and (21). Continuing with the same reasoning, we also have

a3 = · · · = am = 0.

================================================================================
EX. 7, p. 250
Consider λ ∈ F , 0 6= v ∈ V and N p = 0 for a positive integer p. Therefore

Nv = λv =⇒ N p v = λ p v
=⇒ λ p v = 0
=⇒ λ = 0.

================================================================================
Comment, p. 252, Proof
By 5.20, p. 144,
p ( T ) T = T p ( T ).
================================================================================
Comment, p. 253, Proof,
4th sentence
Firstly, note that G λ j , T is invariant under T − λ j I by (b). Therefore




T − λ j I |G(λ ,T ) ∈ L G λ j , T .
j

Now, if v ∈ G λ j , T , that is, if

dim V
T − λj I v = 0,
then

dim V
dim V
T − λ j I |G(λ ,T ) v = T − λj I v
j

= 0,


which implies that T − λ j I |G(λ ,T ) is nilpotent.
j
================================================================================
EX. 1, p. 267
By 8.36, the characteristic polynomial of T belongs to
n o
(t − 3)2 (t − 5)(t − 8), (t − 3)(t − 5)2 (t − 8), (t − 3)(t − 5)(t − 8)2 .

Now use 8.37 and the fact that, by commutativity,

h i
( T − 3I )2 ( T − 5I )2 ( T − 8I )2 = ( T − 5I )( T − 8I ) ( T − 3I )2 ( T − 5I )( T − 8I )
h i
= ( T − 3I )( T − 8I ) ( T − 3I )( T − 5I )2 ( T − 8I )
h i
= ( T − 3I )( T − 5I ) ( T − 3I )( T − 5I )( T − 8I )2 .

================================================================================

23
EX. 2, p. 267
By 8.36, the characteristic polynomial of T belongs to
n o
(t − 5)n−i (t − 6)i : i = 1, . . . , n − 1 .

Now use 8.37 and the fact that, by commutativity,

h i
( T − 5I )n−1 ( T − 6I )n−1 = ( T − 5I )i−1 ( T − 6I )n−(i+1) ( T − 5I )n−i ( T − 6I )i , i = 1, . . . , n − 1.

================================================================================

24
================================================================================
================================================================================
9
================================================================================
================================================================================
Erratum, p. 277, 9.4, Proof, l. -4
‘Imn ’ should be ‘Im λn ’.
================================================================================
Comment, p. 279, 9.10, Proof, 2nd sentence
Consider p(t) = ∑m j
j=0 a j t with a j ∈ R, j = 0, 1, . . . , m, and u, v ∈ V. Therefore


p( T ) C
(u + iv) = p( T )u + ip( T )v
m m
= ∑ aj T j u + i ∑ aj T j v
j =0 j =0
m  
= ∑ aj T j u + iT j v
j =0
m
= ∑ a j (TC ) j (u + iv)
j =0

= p( TC ) (u + iv).

================================================================================
Erratum, p. 284, 9.23, Proof, last sentence
‘8.36(a)’ should be ‘8.36(b)’.
================================================================================
EX. 8, p. 285
On the one hand, 5 and 7 are eigenvalues of TC by 9.11. On the other hand, the sum of the multiplicities of all
the eigenvalues of TC equals 3 by 8.26. So there are only two possibilities:
1. 5 and 7 are the eigenvalues of TC , one of multiplicity 1 and one of multiplicity 2;
2. 5, 7 and another scalar λ are the eigenvalues of TC , each with multiplicity 1.
By 9.16, λ ∈ R.
================================================================================
EX. 11, p. 286
Consider q(t) = t2 + bt + c and let p denote the minimal polynomial of T. So, by 8.46, either q = p or there is
s ∈ P (R) such that q = ps.39 Therefore, by 8.49,

T has an eigenvalue ⇐⇒ p has a zero

⇐⇒ b2 − 4c ≥ 0.

================================================================================
Comment, p. 288, penultimate paragraph before 
 
a −b
M( T, (e1 , e2 )) =
b a

Te1 = ae1 + be2

= ae1 + (−b)(−e2 ),
Te2 = −be1 + ae2
= (−1) (be1 + a (−e2 ))

⇓
39 In this case, p and s are monic polynomials of degree one.

25
 
a b
M( T, (e1 , −e2 )) = .
−b a
================================================================================
EX. 4, p. 294
For u, v, x, y ∈ V,

TC (u + iv), x + iy = h Tu + iTv, x + iyi

= h Tu, x i + h Tv, yi + (h Tv, x i − h Tu, yi)i
= hu, Tx i + hv, Tyi + (hv, Tx i − hu, Tyi)i
= hu + iv, Tx + iTyi
= u + iv, TC ( x + iy)i.

================================================================================

26
================================================================================
================================================================================
10
================================================================================
================================================================================
Comment, 10.10, p. 299
Because the use of Determinants to obtain eigenvalues is established in the next section, 10.B, in order to verify
here that λ ∈ {1, 2 ± 3i } is an eigenvalue of T, we may determine a basis (vλ ) of null( T − λI ) and verify that
 
3−λ −1 −2
 3 2 − λ −3  vλ = λvλ .
1 2 −λ
For example, consider λ = 1. So the reduced row echelon form of
 
2 −1 −2
 3 1 −3 
1 2 −1

is  
1 0 −1
 0 1 0 
0 0 0
and  
1
vλ =  0  .
1
================================================================================
EX. 4, p. 304



M T, (v1 , . . . , vn ) = M IT, (v1 , . . . , vn ) , (v1 , . . . , vn )



= M I, (u1 , . . . , un ) , (v1 , . . . , vn ) M T, (v1 , . . . , vn ) , (u1 , . . . , un ) (by 10.4)


= M I, (u1 , . . . , un ) , (v1 , . . . , vn ) I.
================================================================================
EX. 13, p. 305
By 10.9, 10.13 and 10.16, we may consider the eigenvalue λ of T such that

−48 + 24 + λ = 51 + (−40) + 1.
================================================================================
EXERCISES 14–5, p. 305

trace(cT ) = trace M(cT )

= trace(cM( T ))
= c trace M( T )
= c trace T,
where the first and last equalities come from 10.16, the second equality comes from 3.38, and the third equality
is trivial.

trace(ST ) = trace M(ST )

= trace(M(S)M( T ))
= trace(M( T )M(S))
= trace M( TS)
= trace( TS),
where the first and last equalities come from 10.16, the second and penultimate equalities come from 3.43, and
the third equality comes from 10.14.

27
================================================================================
Comment, 10.26, Solution, 3rd paragraph, 1st sentence, p. 310
Here P( j) represents the claim
T n v j = a1 · · · an v j for each j.

So P(1) is valid since T n−1 v1 = a1 · · · an−1 vn implies that

T n v1 = a1 · · · an−1 Tvn
= a 1 · · · a n −1 a n v 1 .

Now suppose P( j − 1) holds for a fixed j ∈ {2, . . . , n}. Therefore

Tv j−1 = a j−1 v j =⇒ T n v j−1 = a j−1 T n−1 v j

=⇒ a1 · · · an v j−1 = a j−1 T n−1 v j
=⇒ a1 · · · a j−2 a j · · · an v j−1 = T n−1 v j
=⇒ a1 · · · an v j = T n v j
=⇒ P( j) holds.

================================================================================
Erratum, 10.54, Proof, 1st sentence, p. 326
‘V’ should be ‘Rn ’.
================================================================================
EX. 1, p. 330
T is assumed to have no real eigenvalues. So TC has no real eigenvalues by 9.11. However, by 5.21 and 9.16,
TC has a pair of complex conjugate eigenvalues. Therefore, by 10.20, det T = det TC is a product of positive
real numbers, with each number expressed in the form λλ where λ is a complex eigenvalue of TC .
================================================================================
EX. 8, p. 331
Suppose firstly that λ1 , . . . , λn are the eigenvalues of T (or of TC if V is a real vector space) with each eigenvalue
repeated according to its multiplicity. So

det T = λ1 · · · λn
= λ1 · · · λ n
= det T ∗

by definition and by exercise 2 in SECTION 7.A.40 Therefore

|det T |2 = det T det T

= det T det T ∗
= det ( TT ∗ )
√ √ 
= det TT ∗ TT ∗
√  √ 
= det TT ∗ det TT ∗
 √ 2
= det TT ∗

where the first equality holds because |z|2 = zz for each complex number z, the third and fifth equalities hold
by 10.44, p. 320, and the fourth equality holds by the remarks made before 10.46, p. 322.
================================================================================
EX. 9, p. 331
Firstly, in 10.56, p. 327, replace T with Ti ∈ L(Rn ), i = 1, 2. Now, by the Triangle Inequality,41

|| T1 y − T2 y|| ≤ ||σ( x + y) − σ( x ) − T2 y|| + || T1 y − σ( x + y) + σ( x )||

40 See pages 307 and 214.
41 6.18, p. 173.

28
for x ∈ Ω and y ∈ Rn such that x + y ∈ Ω. So

||( T1 − T2 )(y)||
lim = 0.
y →0 ||y||

Then, in particular, for each nonzero vector y and all sufficiently small nonzero real number t,

||( T1 − T2 )(ty)||
0 = lim
t →0 ||ty||
|t| ||( T1 − T2 )(y)||
  
= lim
t →0 | t | ||y||
||( T1 − T2 )(y)||
= .
||y||

Hence ( T1 − T2 )(y) = 0 for each y 6= 0. Therefore, along with the fact that every linear transformation maps
the zero vector to the zero vector, T1 − T2 = 0.
================================================================================
EX. 10, p. 331
By 10.56, p. 327, T ∈ L(Rn ) is differentiable at x ∈ Rn if

|| T ( x + y) − Tx − Ly||
lim =0 (24)
y →0 ||y||

for an operator L ∈ L(Rn ). Furthermore, by EX. 9, p. 331, this L := T 0 ( x ) is unique. So, since

x, y ∈ Rn =⇒ || T ( x + y) − Tx − Ty|| = 0,

(24) holds and T 0 ( x ) = T for each x ∈ Rn .

================================================================================

29