Linear Al Done Right Note

Solutions for exercises/Notes for Linear
Algebra Done Right by Sheldon Axler

Toan Quang Pham
mathtangents@gmail.com
Monday 10th September, 2018
Contents
1. Some note before reading the book 4
2. Terminology 4
3. Chapter 1 - Vector Spaces 4

3.1. Excercises 1.B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3.2. 1.C Subspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3.3. Exercises 1.C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4. Chapter 2 - Finite-dimensional vector spaces 9

4.1. 2.A Span and linear independence . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.1.1. Main theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.1.2. Important/Interesting results from Exercise 2.A . . . . . . . . . . . . . . . 11
4.2. Exercises 2.A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.3. 2.B Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.4. Addition, scalar multiplication of specific list of vectors . . . . . . . . . . . . . . . 15
4.5. Exercises 2B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.6. 2.C Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.7. Exercises 2C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5. Chapter 3: Linear Maps 21

5.1. 3.A The vector space of linear maps . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.2. Exercises 3A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.3. 3.B Null Spaces and Ranges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
5.4. Exercises 3B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.4.1. A way to construct (not) injective, surjective linear map . . . . . . . . . . 25
5.4.2. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1
Toan Quang Pham page 2
5.5. Exercises 3C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.6. 3D: Invertibility and Isomorphic Vector Spaces . . . . . . . . . . . . . . . . . . . 31
5.7. Exercises 3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.8. 3E: Products and Quotients of Vector Spaces . . . . . . . . . . . . . . . . . . . . 36
5.9. Exercises 3E . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.10. 3F: Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.11. Exercises 3F . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6. Chapter 4: Polynomials 48
6.1. Exercise 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
7. Chapter 5: Eigenvalues, Eigenvectors, and Invariant Subspaces 50

7.1. 5A: Invariant subspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
7.2. Exercises 5A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
7.3. 5B: Eigenvectors and Upper-Triangular Matrices . . . . . . . . . . . . . . . . . . 55
7.4. Exercises 5B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
7.5. 5C: Eigenspaces and Diagonal Matrices . . . . . . . . . . . . . . . . . . . . . . . . 58
7.6. Exercises 5C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
8. Chapter 6: Inner Product Spaces 64

8.1. 6A: Inner Products and Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
8.2. Exercises 6A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
8.3. 6B: Orthonormal Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
8.4. Exercises 6B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
8.5. 6C: Orthogonal Complements and Minimization Problems . . . . . . . . . . . . . 77
8.6. Exercises 6C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
9. Chapter 7: Operators on Inner Product Spaces 81

9.1. 7A: Self-adjoint and Normal Operators . . . . . . . . . . . . . . . . . . . . . . . . 81
9.2. Exercises 7A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
9.3. 7B: The Spectral Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
9.4. Exercises 7B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
9.5. 7C: Positive Operators and Isometries . . . . . . . . . . . . . . . . . . . . . . . . 88
9.6. Exercises 7C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
9.7. 7D: Polar Decomposition and Singular Value Decomposition . . . . . . . . . . . . 91
9.8. Exercises 7D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
10.Chapter 8: Operators on Complex Vector Spaces 97

10.1. 8A: Generalized Eigenvectors and NilpotentOperators . . . . . . . . . . . . . . . 97
10.2. Exercises 8A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
10.3. 8B: Decomposition of an Operator . . . . . . . . . . . . . . . . . . . . . . . . . . 100
10.4. Exercises 8B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
10.5. 8C: Characteristic and Minimal Polynomials . . . . . . . . . . . . . . . . . . . . . 105
10.6. Exercises 8C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
10.7. 8D: Jordan Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

10.8. Exercises 8D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
11.Chepter 9: Operators on Real Vector Spaces 113

11.1. 9A: Complexification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
11.2. Exercises 9A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
11.3. 9B: Operators on Real Inner Product Spaces . . . . . . . . . . . . . . . . . . . . . 116
11.4. Exercises 9B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
12.Chapter 10: Trace and Determinant 119

12.1. 10A: Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
12.2. Exercises 10A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
13.Summary 122
14.Interesting problems 123
15.New knowledge 123

1. Some note before reading the book

1. If the problem does not specify that vector space V is finite-dimensional then it means
V can be either finite-dimensional or infinite-dimensional, which is sometimes harder to
exer:3B:24
solve. For example, exercises 24 in 3B.
2. Carefully check whether the theorem/definition holds for real vector space or complex
vector space.
2. Terminology
1. There exists a basis of V with respect to which the matrix of T = the matrix of T with
respect to basis of V .
2. Eigenvectors v1 , . . . , vm corresponding to eigenvalues 1, . . . , m = Eigenvalues 1, . . . , m

and corresponding eigenvectors v1 , . . . , vm .
3. Chapter 1 - Vector Spaces

3.1. Excercises 1.B
1. ( v) = ( 1)( v) = ( 1)(( 1)v) = 1v = v.
2. If a = 0 then it’s obviously true. If a 6= 0 then v = 1

aa v = a1 (av) = a1 0 = 0.
3. Let 13 (w v) = x then 3x = w v so 3x + v = (m v) + v = m. We can see that x in

here is unique, otherwise if there exist such two x, x0 then x = 13 (w v) = x0 .
4. The empty set is not a vector space because it doesn’t contain any additive identity 0 (it
doesn’t have any vector).
5. Because the collection of objects satisfying the orgininal condition is all vectors in V , same
as collection of objects satisfying the new condition.
6. R [ {1} [ { 1} is a vector space over R.
3.2. 1.C Subspaces

Problem 3.2.1 (Example 1.35). (a) Obvious.
(b) Let S be set of all continuous real-valued functions on the interval [0, 1]. S has additive
identity (Example 1.24). If f, g 2 S then f + g, cf is continuous real-valued function on
the interval [0, 1] for any c 2 F.
(c) The idea is similar, if g, f are two differentiable real-valued functions then so are f + g, cf .
(d) Let g, f 2 S then g 0 (2) = f 0 (2) = b. According to Notion 1.23 then f + g 2 S so

b = (f + g)0 (2) but (f + g)(x) = f (x) + g(x) so (f + g)0 (x) = f 0 (x) + g 0 (x) implying
b = (f + g)0 (2) = f 0 (2) + g 0 (2) = 2b or b = 0.
(e) Let an , bn be to sequences of S then lim an = lim bn = 0 so lim (an +bn ) = lim can = 0
n!1 n!1 n!1 n!1
which follows an + bn , can 2 S.
Problem 3.2.2 (Page 19). What are subspaces of R2 ?
Answer. {(0, 0)} and R2 are obviously two of them. Are there any other subspaces? Let S
be a subspace of R2 and there is an element (x1 , y1 ) 2 S with (x1 , y1 ) 6= (0, 0). Note that
(x1 , y1 ) 2 S for each 2 F so this follows that all points in the line y = xy11 x belong to S. If
there doesn’t exist a point (x2 , y2 ) 2 S so that (x2 , y2 ) doesn’t lie in the line y = xy11 x then we
can state that S, set of points lie in y = xy11 x, is a subspace of R2 . Hence, since point (x1 , y1 ) is
an arbitrary point, we deduce that all lines in R2 through the origin are subspaces of R2 .
Now, what if there exists such (x2 , y2 ) 2 S then the line y = xy22 x also belongs to S. We will
state that if there exists two lines belong to S then S = R2 . Indeed, since R2 is the union of
all lines in R2 through the origin, it suffices to prove that all lines y = ax for some a belongs
to R, given that two lines y = xy11 x and y = xy22 x already belong to S. Let m = xy11 , n = xy22 .
We know that if (x1 , y1 ), (x2 , y2 ) 2 S then (x1 + x2 , y1 + y2 ) 2 S. We pick x1 , x2 so that
(m a)x1 = (a n)x2 then mx1 + nx2 = a(x1 + x2 ). We pick y1 = mx1 , y2 = nx2 then this
follows (y3 , x3 ) = (y1 + y2 , a(x1 + y2 )) 2 S which yields that line y = ax belongs to S. Since
this is true for arbitrary a so we deduce that S = R2 .
Thus, all subspaces of R2 are {(0, 0)}, R2 and all lines in R2 through the origin.
Problem 3.2.3 (Page 19). What are subspaces of R3 ?
Answer. Similar idea to previous problem. It’s obvious that {(0, 0, 0)}, R3 are two subspaces of
R3 . Let S be a different subspace.
1. If (x1 , y1 , z1 ) 2 S then (x1 , y1 , z1 ) 2 S implying line ` in R3 through the origin and

(x1 , y1 , z1 ) belongs to S.
2. If there exists (x2 , y2 , z2 ) 2 S and (x2 , y2 , z2 ) does not lie in the line ` on R3 then the
plane P through the origin, (x1 , y1 , z1 ) and (x2 , y2 , z2 ) belongs to S.
3. If there exists (x3 , y3 , z3 ) 2 S but not in the plane P then S = R3 .
Thus, {0}, R3 , all line in R3 through the origin and all planes in R3 through the origin are all
subspaces of R3 .
theo_7:1C:1.45 Theorem 3.2.4 (Sum of subspaces, page 20) Suppose U1 , . . . , Um are subspaces of V . Then
U1 + U2 + . . . + Um is the smallest subspace of V containing U1 , . . . , Um .
Proof. It’s not hard to prove that U1 + U2 + . . . + Um is subspace of V . Let U be a subspace

of V containing U1 , . . . , Um , we will prove that if v 2 U1 + . . . + Um then v 2 U . Indeed, since
v 2 U1 + . . . + Um then v = v1 + . . . + vmPfor vi 2 Ui for all 1  i  m. On the other hand,
vi 2 U since U contains U1 , . . . , Um so v = m i=1 vi 2 U . Thus, our claim is proven, which yields
that U contains U1 + . . . + Um .
Next, we will prove that U1 + . . . + Um contains U1 , . . . , Um . This is not hard since we know
that 0 2 Ui for all 1  i  m so for any v 2 Ui then
v = |0 + .{z . . + 0} 2 U1 + . . . + Um .
. . + 0} +v + 0| + .{z
i 1 m i
These two argument follows that U1 + . . . + Um is the smallest subspace of V containing

U1 , . . . , U m .
Theorem 3.2.5 (1.44, Condition of a direct sum, page 23)

Suppose U1 , U2 , . . . , Um are subspaces of V . Then U1 + . . . + Um is a direct sum if and only
if 0 can only be written as sum u1 + . . . + um where each ui = 0 in Ui .
Proof. If U1 + . . . + Um is a direct sum then obviously 0 is written uniquely as sum 0 + . . . + 0.

If 0 is written uniquely as 0 + 0 + . . . + 0. Assume the contrary that sum U = U1 + . . . + Um
is not a direct sum, which means there exists u 2 U so that u = v1 + . . . + vm = u1 + . . . + um
where ui , vi 2 Ui and there exists i so ui 6= vi . This follows
m
X m
X m
X
(vi ui ) = vi ui = u u = 0.
i=1 i=1 i=1
In other words, there is another way to write 0 as (v1 u1 ) + . . . + (vm um ) where v1 u1 2 Ui

and there exists j so vj uj 6= 0, a contradiction. Thus, U1 + . . . + Um is a direct sum.
Theorem 3.2.6 (1.45, Direct sum of two subspaces)

Suppose U, V are two subspaces of V . Then U +W is a direct sum if and only if U \W = {0}.
Proof. If U + W is a direct sum. If there exists v 2 U \ W and v 6= 0 then v = ( 1)v 2 U so

0 = v v, which contradicts to the unique representation of 0 as sum of vectors in U, W .
If U \ W = {0}. Assume the contrary that U + W is not a direct sum then 0 = u + v with
u 2 U, v 2 W, u, v 6= 0. This follows u = v 2 W or u 2 U \ W , a contradiction. We are
done.
3.3. Exercises 1.C

1. (a), (d) are subspaces of F3 . (b) is not subspace of F3 because it doesn’t contain 0. (c) is
not subspace of F3 because (0, 1, 1) + (1, 1, 0) = (1, 2, 1) 62 F3 .
2. Already did.
3. Let S be set of such functions. Let f, g 2 S then f + g is a differentiable real-valued

function and (f + g)0 ( 1) = f 0 ( 1) + g 0 ( 1) = 3f (2) + 3g(2) = 3(f + g)(2). Similarly to
f.
4. Similar to example 1.35.
5. Yes.
6. (a) Since a3 = b3 if and only if a = b so {(a, b, c) 2 R3 : a3 = b3 } = {(a, b, c) 2 R3 : a = b}.

It’s not hard to prove that⇣this is subspace
p
of R 3.
p ⌘ ⇣ p p ⌘
(b) No. We see that (1, 1), 1 + i, 1 2 3 + 1+2 3 i 2 S but 2 + i, 1 2 3 + 1+2 3 i 62 S.
7. Set {(a, b) : a, b 2 Z} is not subspace of R2 .
8. Set U = {(a, b) 2 R2 : a = b} [ {(a, b) 2 R2 : a = 2b} is not subspace of R2 because

(1, 1), (2, 1) 2 U but (3, 2) 62 U .
p p
9. We consider functionpf (x) so that f (x) = xp for x 2 [0, 2) and f (x) = f (x+ 2)., function
g(x) = x for x 2 [0, 3) and g(x) = g(x + 3) then f + g is not periodic. p
Indeed, assume that the function f (x) + g(x) is periodic
p with length Tp
. Note thatpif T / 2
is not an integer then f (0) < f (T ), similar to T / 3. Thus, since T / 2 and T / 3 can’t
be both integers so f (0) + g(0) < f (T ) + g(T ), a contradiction. Thus, f + g is non-periodic
function. This yields that set of periodic function from R to R is not a subspace of RR .
10. For any u, v 2 U1 \ U2 , then u + v 2 U1 , u + v 2 U2 so u + v 2 U1 \ U2 . Similar to

u 2 U1 \ U2 .
11. Similar to 10.
12. Assume the contrary, that means there exists two subspaces U1 , U2 of V so that U1 [ U2
is also a subspace and there exists u1 2 U1 , u2 2 U2 , u1 62 U2 , u2 62 U1 . We have u1 , u2 2
U1 [ U2 so u1 + u2 2 U1 [ U2 . WLOG, assume u1 + u2 2 U1 then (u1 + u2 ) u1 2 U1 or
u2 2 U1 , a contradiction. Thus, either U1 ✓ U2 or U2 ✓ U1 .
13. Let U1 , U2 , U3 be three subspaces of V so U1 [ U2 [ U3 is also a subspace. If there exists

two among these three subspaces, WLOG U1 ✓ U2 , then U1 [ U2 [ U3 = U2 [ U3 is a
subspace if and only if U3 ✓ U2 or U2 ✓ U3 , in other words either U2 contains U1 , U3 or
U3 contains U1 , U2 .
If there doesn’t exist any two among three subspaces U1 , U2 , U3 so that one contains the
other, then there exists a vector v 62 U2 , v 2 U1 [ U2 [ U3 which yields v 2 U1 [ U3 . WLOG,
v 2 U1 . Since U1 6✓ U2 , U2 6✓ U1 so there exists u 2 U2 , u 62 U1 . With similar argument
to 12, we find u + v 62 U2 , u + v 62 U1 so u + v 2 U3 . Similar u + 2v 2 U3 which we find
(u + 2v) (u + v) = v 2 U3 . Thus, we find that for any v 62 U2 then v 2 U1 \ U3 . This
follows U1 [ U2 = U1 \ U2 or U1 = U2 , a contradiction to the assumption.
14. Done
15. U +U is exactly U . From Theorem 1.39, U +U contains U . We also notice that U contains
U + U because every u 2 U + U then u 2 U .
16. Yes. As long as U, W are sets of vectors of V .
17. Yes. As long as U1 , U2 , U3 are sets of vectors of V .
18. Operation of the addition on the subspaces of V has additive identity, which is U = {0}.
Only {0} has additive inverse.
19. Let V = {(x, y, z) 2 R3 : x, y, z 2 R}, U1 = {(x, 0, 0) 2 R3 : x 2 R}, U2 {(0, x, 0) 2 R3 :

x 2 R} and W = {(x, y, 0) 2 R3 : x, y 2 R} then U1 + W = U2 + W = W but U1 6= U2 .
20. W = {(x, 2x, y, 2y) 2 F4 : x, y 2 F}. It is obvious that W + U is a direct sum. It

suffices to prove that F4 = W U or prove that for any (m, n, p, q) 2 F4 there exists
(x, 2x, y, 2y) 2 W, (x1 , x1 , y1 , y1 ) 2 U so (m, n, p, q) = (x, 2x, y, 2y) + (x1 , x1 , y1 , y1 ). We
can choose x = n m, x1 = 2m n, y = q p, y1 = 2p q.
21. W = {(x, y, z, z, x) 2 F5 : x, y, z 2 F}. Vector 0 can be represented uniquely as w1 + u1

where w1 2 W, u1 2 U and w1 = u1 = 0. Thus, W U is a direct sum. It suffices to
prove for each (m, n, p, q, r) 2 F5 there is w = (x, y, z, z, x) 2 W, u = (x1 , y1 , x1 + y1 , x1
y1 , 2x1 ) 2 U so (m, n, p, q, r) = w + u. We can choose x1 = r m, x = 2m r, z =
m r + 12 p + 12 q, y1 = 12 p 12 q, y = n + 12 q 12 p.
22. W1 = {(0, 0, x, 0, 0) 2 F5 : x 2 F}, W2 = {(0, 0, 0, x, 0) 2 F5 : x 2 F}, W3 = {(0, 0, 0, 0, x) 2

F5 : x 2 F}.
23. Let’s take Exercise 20 as counterexample, V = F4 , W = {(x, x, y, y) 2 F4 : x, y 2 F}, U1 =

{(x, 2x, y, 2y) 2 F4 : x, y 2 F} and U2 = {(x, 3x, y, 3y) 2 F4 : x, y 2 F}.
24. Since Ue \ Uo = {f (x) = 0} so Ue + Uo is a direct sum. Consider f (x) 2 RR then f (x) =

fe (x) + fo (x) where fe (x) = 12 [f (x) + f ( x)] 2 Ue and fo (x) = 12 [f (x) f ( x)] 2 Uo .
Thus, RR = Ue Uo .
4. Chapter 2 - Finite-dimensional vector spaces

4.1. 2.A Span and linear independence
4.1.1. Main theories
theo_8:2A:2.21 Theorem 4.1.1 (Linear Dependence Lemma) Suppose v1 , . . . , vm is linearly dependent list
in V . There exists j 2 {1, 2, . . . , m} such that the following holds:
(a) vj 2 span(v1 , . . . , vj 1 ).
(b) if the jth term is removed from v1 , . . . , vm , the span of the remaining list equals
span(v1 , . . . , vm ).
theo1 Theorem 4.1.2 (2.23) In a finite-dimensional vector space, the length of every linearly
independent list of vector is less than or equal to the length of every spanning list of
vectors.
This theorem is also called as Steinitz exchange lemma or Replacement’s theorem.
Proof. Let v1 , v2 , . . . , vm is linearly independent list and u1 , u2 , . . . , un is spanning list of vectors.

It suffices to prove for the case that u1 , u2 , . . . , un is also linearly independent, otherwise if
u1 , u2 , . . . , un is linearly dependent, according to Linear Dependence Lemma (2.21), we can take
out some vectors from the list without changing the span and bring back to linearly independent
list. Assume the contrary that n < m and from this assumption, we will prove that list v1 , . . . , vm
is linearly dependent.
Since u1 , u2 , . . . , un spans V and is also linearly independent so for each vi , there exists unique
ci1 , ci2 , . . . , cin 2 F so vi = ci1 u1 + ci2 u2 + . . . + cin un . Hence, for a1 , . . . , am 2 F, we have
m
X m
X m
X
a 1 v 1 + a 2 v 2 + . . . + a m v m = u1 ai ci1 + u2 ai ci2 + . . . + un ai cin .
i=1 i=1 i=1
We will show that there exists a1 , . . . , am not all 0 so a1 v1 + . . . + am vm = 0. Since u1 , . . . , un

is linearly independent so form the above, we follow
8
>
>a1 c11 + a2 c21 + . . . + am cm1 = 0,
>
>
<a c + a c + . . . + a c = 0,
2 12 2 22 m m2
a1 v1 + . . . + am vm = 0 () (1) eq1
>
> . . .
>
>
:a c + a c + . . . + a c = 0.
m 1n 2 2n m mn
This system of equation has m variable and n equation with n < m.eq1Here is an algorithm in
finding non-zero solution for a1 , . . . , am . Number the equations in (1) from top to bottom as
(11), (12), . . . , (1n).
1. For 1  i  m 1, get rid of am in (1i) by subtracting to (1(i + 1)) times some constant.
By doing this, we obtain a system of equations where the first n 1 equations only contain
m 1 variables, the last equation contains m variables.
2. For 1  i  m 2, get rid of am in (1i) by subtracting to (1(i + 1)) times some constant.
We obtain an equivalent system of equations where the first n 2 equations contain m 2
variables, equation (1(n 1)) contains m 1 variables and and equation (1n) contains m
variables.
3. Keep doing this until we get an system of equation so that the (1i) equation contains
m n + i variables.
4. Starting from equation (11) with m n+1 variables a1 , . . . , am n+1 , we can pick a1 , . . . , am+n 1
so that it satisfies equation (11) and a1 , . . . , am n+1 not all 0, which is possible since
m n + 1.
5. Come to (12) and pick am n+2 , to (13) and pick am n+3 until to (1n) and pick am .
Thus, there exists a1 , . . . am not all 0 so that a1 v1 + . . . + am vm = 0, a contradiction since

v1 , . . . , vm is linearly independent. Hence, we must have n m. We are done.
exer:2A:12
exer:2A:13
exer:2A:14
exer:2A:17
See Exercises 12, 13, 14, 17 for applications of this theorem.
theo:2 Theorem 4.1.3 (2.26 Finite dimensional subspaces) Every subspace of a finite-dimensional
vector space is finite-dimensional.
Proof. Let V = span(v1 , v2 , . . . , vm ) be a finite-dimensional vector space and U be its subspace.

Among vectors in the spanning list v1 , v2 , . . . , vm , let u1 , u2 , . . . un be vectors that are in subspace
U , k1 , k2 , . . . , km n be vectors that are not in U . If U = span(u1 , u2 , . . . , un ) then we are done.
If not, then there exists ` 2 U so ` 62 span(u1 , u2 , . . . , un ). Hence, since ` 2 span(v1 , . . . , vm ) so
n
X m
Xn
`= ui a i + k i bi
i=1 j=1
for ai , bi 2 F and there exists bi (1  i  m n) so that bi 6= 0. Since ui , ` 2 U so that

m
Xn
means h1 = ki bi 2 U and also in span(k1 , . . . , km n ) and h1 6= 0. We add h1 to the list
i=1
h1 , u1 , u2 , . . . , un in U . Similarly, we compare U and span(h1 , u1 , . . . , un ) and may add an-
other h2 2 span(k1 , k2 , . . . , km n ) onto the list. This process will keep going until we obtain
a spanning list of U . We will prove that after adding at most m n such hi , we can get
U = span(h1 , h2 , . . . , hj , u1 , . . . , un ) with j  m n.
Indeed, assume the contrary that after adding at least m n such hi 2 U so hi 6= 0, we

still can’t obtain a list of vectors that spans U . That means there exists ` 2 U and ` 2 /
span(h1 , . . . , hm n , u1 , . . . , un ). Similarly, we can construct another , hm+n 1 6= 0, hm n+1 2 U
and hm n+1 2 span(k1 , . . . , km n ) based on `. Let hi = ci1 k1 + ci2 k2 + . . . + ci,m n km n then
we have
a1 h1 + . . . + am n+1 hm n+1 = k1
8
>
> a1 c1,1 + a2 c2,1 + . . . + am n+1 cm n+1,1 = 1,
>
>
<a c + a c + . . . + a
1 1,2 2 2,2 m n+1 cm n+1,2 = 0,
()
>
> ...
>
>
:a c
1 1,m n + . . . + am n+1 cm n+1,m n = 0.
This is a system theo1

of m n equations with m n + 1 variables a1 , . . . , am n+1 2 F so similarly
to Theorem 4.1.2’s proof, there are infinite number of ai so a1 h1 + . . . + am n+1 hm n+1 = k1 .
Since hi 2 U so k1 2 U , a contradiction. Thus, there are at most m n such hi 2 U (1 
i  j) so that there does not exist ` 2 U but ` 2 / span(h1 , . . . , hj , u1 , . . . , un ). Note that
span(h1 , . . . , hj , u1 , . . . , un ) ✓ U so this can only mean U = span(h1 , . . . , hj , u1 , . . . , un ). Thus,
U is a finite-dimensional vector space.
Remark 4.1.4. Both my proofs for the two theorems use the result that a system of m equations
with n variables in R with m < n has infinite solution in R.
4.1.2. Important/Interesting results from Exercise 2.A
coroll:1 Corollary 4.1.5 (Example 2.20) If some vector in a list of vectors in V is a linear combi-
nation of the other vectors, then the list is linearly dependent.
P P
Proof.
Pm List v1 , . . . , vm has v1 = i ai vi then v1 i ai vi = 0, or there exists x1 , . . . , xm not all
0 so i=1 xi vi = 0. Hence, v1 , . . . , vm is linearly dependent.
This is another way to check if some list is linearly independent or not. See Exercises 10 (2A)
for application. The reverse is also true.
Corollary 4.1.6
exer:2A:11
coroll:2 (Exercise 11, 2A) Suppose v1 , . . . , vm is linearly independent in V and w 2 V then v1 , v2 , . . . , vm , w
is linearly independent if and only if w 62 span(v1 , v2 , . . . , vm ).
This is a corollary implying from Linear Dependence Lemma (2.21). With this, we can construct
aexer:2A:14
spanning list for any infinite-dimensional vector space. Also with this, we can prove Exercise
14, which proves existence of a sequence of vectors in infinite-dimensional vector space so that for
any positive integer m, the first m vectors
exer:2A:15
exer:2A:16
in the sequence is linearly independent. See Exercise
exer:2A:14
15, 16 for applications for Exercise 14.
4.2. Exercises 2.A

1. For any v 2 V then v = a1 v1 + a2 v2 + a3 v3 + a4 v4 for ai 2 F. Hence,
v = a1 (v1 v2 ) + (a1 + a2 )(v2 v3 ) + (a1 + a2 + a3 )(v3 v4 ) + (a1 + a2 + a3 + a4 )v4 ,
= b1 (v1 v2 ) + b2 (v2 v3 ) + b3 (v3 v 4 ) + b4 v 4 .
Thus, v 2 span(v1 v2 , v2 v3 , v3 v4 , v4 ). We follow V ✓ span(v1 v2 , v2 v3 , v3

v4 , v4 ). On the other hand, since v1 v2 , v2 v3 , v3 v4 , v4 so for any u 2 span(v1
v2 , v2 v3 , v3 v4 , v4 ) then u 2 V . We find span(v1 v2 , v2 v3 , v3 v4 , v4 ✓ V . Thus,
v1 v2 , v2 v3 , v3 v4 , v4 spans V .
2. Example 2.18: (a) True for v 6= 0. If v = 0 then for any a 2 F we always have av = 0, so
list of one vector 0 is not linearly independent.
(b) Consider u, v 2 V . We have au + bv = 0 () au = bv so in order to have only one
choice of (a, b) to be (0, 0) then u can’t be scalar multiple of v.
(c),(d) True.
3. It suffices to find t so there are 8 x, y, z 2 R other than x = y = z = 0 so x · (3, 1, 4) +

>
<3x + 2y + 5z = 0,
y · (2, 3, 5) + z · (5, 9, t) = 0 or x 3y + 9z = 0, Pick t = 2 then we find (x, y, z) =
>
:
4x + 5y + tz = 0.
( 3a, 2a, a) for any a 2 R. This follows (3, 1, 4), (2, 3, 5), (5, 9, t) is not linearly indepen-
dent in R3 .
4. 8
Similarly to 3, let say if x · (2, 3, 1) + y · (1, 1, 2) + z · (7, 3, c) = 0 for x, y, z 2 R then
>2x + y + 7z = 0,
<
3x y + 3z = 0, From two above equations, we find x = 2z, y = 3 and substitute
>
:
x + 2y + cz = 0
into the third equation to get (c 8)z = 0. This has at least two solutions of z if and only
if c = 8. Done.
5. (a) The only solution x, y 2 R so x(1 + i) + y(1 i) = 0 is x = y = 0. Hence list 1 + i, 1 i

is linearly independent in vertor space C over R.
(b) Beside x = y = 0 then x = i, y = 1 also satisfies x(1 + i) + y(1 i) = 0. Hence,
1 + i, 1 i is linearly dependent in vector space C over C.
exer:2A:6 6. Consider a, b, c, d 2 F so a(v1 v2 ) + b(v2 v3 ) + c(v3 v4 ) + dv4 = 0, which is equivalent

to av1 + (b a)v2 + (c b)v3 + (d c)v4 = 0. Since v1 , v2 , v3 , v4 is linearly dependent in
V so this follows a = b a = c b = d c = 0 or a = b = c = d = 0 and a, b, c, d are
uniquely determined. Hence, v1 v2 , v2 v3 , v3 v4 , v4 is linearly independent in V .
exer:2A:7 7. True. Let a1 (5v1 4v2 ) + a2 v2 + a3 v3 + . . . + am vm = 0 or 5a1 v1 + (a2 4a1 )v2 + a3 v3 +

. . . + am vm = 0, which implies 5a1 = a2 4a1 = a3 = . . . = am = 0 or a1 = a2 = . . . = am .
8. Consider a1 v1 + a2 v2 + . . . + am vm = 0, which implies a1 = a2 = . . . = am = 0

since a1 , a2 , . . . , am is linear independent. Thus, a1 = a2 = . . . = am so v1 , v2 , . . . , vm
is linearly independent.
9. Not true. Let v1 , v2 , . . . , vm be (1, 0, 0, . . . , 0), (0, 1, . . . , 0), . . . , (0, . . . , 0, 1), respectively
and let w3 , . . . , wm be (0, 0, 2, 0, . . . , 0), . . . , (0, . . . , 0, 2) and w1 = (0, 1, 0, . . . , 0) and
w2 = ( 1, 0, . . . , 0). Therefore, for i 3 then wi + vi = (0, . . . , 0, i, 0, . . . , 0) and w1 + v1 =

| {z }
i 1
(1, 1, 0, . . . , 0), w2 + v2 = ( 1, 1, 0, . . . , 0). It’s not hard to see (w1 + v1 ) + (w2 + v2 ) = 0
so w1 + v1 , w2 + v2 , . . . , wm + vm is linearly dependent.
10. If v1 + w, v2 + w, . . . , vm + w is linearly dependent then according to Linear Dependence

Lemma (2.21) there exists j 2 {1, 2, . . . , m} so that vj + m 2 span(v1 + m, . . . , vj 1 + m).
Hence, vj + m = a1 (v1 + m) + . . . + aj 1 (vj 1 + m) or
(a1 + . . . + aj 1 1)m = vj a1 v1 a2 v 2 ... aj 1 vj 1
If a1 + . . . + aj 1 1 = 0 then vj a1 v1 a2 v2 . . . aj 1 vj 1 = 0. If j = 1 then
that means vj = 0, so v1 , v2 , . . . , vm is linearly dependent, a contradiction. If j 2 then
vj = a1 v1 + . . . + aj 1 vj 1 which also follows that v1 , . . . , vm is linearly dependent, a
contradiction.
Thus, we must have a1 + . . . + aj 1 1 6= 0. Therefore, since RHS 2 span(v1 , . . . , vm ) so
m 2 span(v1 , . . . , vm ).
exer:2A:11 11. If v1 , . . . , vm , w is linearly independent then w 62 span(v1 , . . . , vm ) according to Corollary

coroll:1
4.1.5. The reverse, if w 62 span(v1 , . . . , vm ) then assume the contrary, if v1 , . . . , vm , w is
linearly dependent then it will lead to a contradiction according to Linear Dependence
Lemma (2.21). Done.
theo1
exer:2A:12 12. Because P4 (F) = span(1, x, x2 , x3 , x4 ) and from Theorem 4.1.2 (2.23) length of any linearly
independent list is less than length of a spanning list 1, x, x2 , x3 , x4 , which is 5. That
explains why we can’t have 6 polynomials that is linearly independent in P4 (F).
exer:2A:13 13. Because 1, x,theo1

x2 , x3 , x4 is a linearly independent list of length 5 in P4 (F) and according
to Theorem 4.1.2 (2.23) length of any spanning list is greater than or equal to length of
a linearly independent list, which is 5. Thus, we can’t have four polynomials that spans
P4 (F).
exer:2A:14 14. Prove that V is infinite-dimensional if and only if there is a sequence v1 , v2 , . . . of vectors
in V such that v1 , . . . , vm is linearly independent for every positive integer m.
Proof. If V is infinite-dimensional: Pick v1 2 V , pick vi so vi 2 V, vi 62 span(v1 , . . . , vi 1 ).

Such vi exists, otherwise any v 2 V must also in span(v1 , . . . , vi 1 ) so V ✓ span(v1 , . . . , vi 1 )
and we also have span(v1 , . . . , vi 1 ) ✓ V so a finite list of vectors span V , a contradic-
tion. Notecoroll:2
that since v1 , . . . , vi 1 is linearly independent and vi 62 span(v1 , . . . , vi 1 ) so by
Corollary 4.1.6 then v1 , . . . , vi is linearly independent. By continuing the process, we can
construct such sequence of vectors.
If there exists such sequence of vectors in V : Assume the contrary that V is finite-
dimensional then there exists finite spanning list in V with length `. However, from
existence of such sequence, we can find in V a linearly independent list with arbitrary
theo1
large length, which contradicts to Theorem 4.1.2 that length of any linearly independent
list less than or equal to length of spanning list in a finite-dimensional vector space. We
are done.
exer:2A:15 15. Withexer:2A:14

F1 , consider the sequence (1, 0, . . .), (0, 1, 0, . . .), (0, 0, 1, 0, . . .), . . . then from Exer-
cise 14, we follow F1 is infinite-dimensional.
exer:2A:16 16. Consider sequence of continuous real-valued functions defined on [0, 1]: 1, x, x2 , x3 , . . . and
we are done.
exer:2A:17 17. Nice problem. Since pj (2) = 0 for each j so that means x 2 divides each pj . Let qi = xpi 2
then q0 , q1 , . . . , qm are polynomials in Pm 1 (F). If q0 , q1 , . . . , qm is linear independent but
it has length m + 1 larger theo1
than length m of spanning list in Pm 1 (F), a contradiction
according to Theorem 4.1.2. P Thus, q0 , q1 , . . . , qm is linearP dependent. That Pmeans there
exists ai 2 F not all 0 so m a q
i=0 i i = 0. This follows (x 2) m
a q
i=0 i i = 0 or m
i=0 i i = 0.
a p
Hence, p0 , . . . , pm is linearly dependent.
4.3. 2.B Bases

theo_3:2B:2.31 Theorem 4.3.1 (2.31) Every spanning list in a vector space can be reduced to a basis of
the vector space.
theo_4:2B:2.32 Theorem 4.3.2 (2.32) Every finite-dimensional vector has a basis.
theo_5:2B:2.33 Theorem 4.3.3 (2.33) Every linearly independent list of vectors in a finite-dimensional
vector space can be extended to a basis of the vector space.
theo_6:2B:2.34 Theorem 4.3.4 (2.34) Suppose V is finite-dimensional and U is subspace of V . Then there
is a subspace W of V such that V = U W .
theo:2
Proof.theo_4:2B:2.32
According to Theorem 4.1.3 then U is also finite-dimensional. Hence, according to The-
orem 4.3.2 then U has a basis v1 , . . . , vm . Since v1 , . . . , vm is linearly independent so according
theo_5:2B:2.33
to Theorem 4.3.3, there exists vectors w1 , w2 , . . . , wn so v1 , . . . , vm , w1 , . . . , wn is a basis of V
Let W = span(w1 , . . . , wn ) then V = U W .
exer:2B:8 theo_6:2B:2.34
Exercise 8 chapter 2B gives another rephrasing of theorem 4.3.4, saying that if there exists sub-
spaces U, W of V so V = U W and each has basis {u1 , . . . , um } and {w1 , . . . , wn } respectively
then u1 , . . . , um , w1 , . . . , wn is a basis of V .
4.4. Addition, scalar multiplication of specific list of vectors

The book doesn’t seem to mention this. I just realised this while i was doing the exercises. It
makes the writing much easier if I write a generalisation in here so I can refer to it later in the
exercises. It basically says spanning list/basis/ linearly independent list has addition and scalar
multiplication properties.
prop:1 Proposition 4.4.1 Given (v1 , v2 , . . . , vm ) that is a basis/spanning list/linearly independent

list then (v1 , vP
2 , . . . , vi + w, vi+1 , . . . , vm ) is also a basis/spanning list/linearly independent
list with w = m j=1 bj vj with bi 6= 1 and bj 2 F for all 1  j  m..
Proof. It’s not hard to see that if (v1 , . . . , vm ) is a spanning list then so is (v1 , . . . , vi +w, vi+1 , . . . , vm ).
If (v1 , . . . , vm ) is linearly independent, we consider ai 2 F so a1 v1 + a2 v2 + . . . + ai (vi + w) +
. . . + am vm = 0 or
(a1 + ai b1 )v1 + (a2 + ai b2 )v2 + . . . + (ai + ai bi )vi + . . . + (am + ai bm )vm = 0.
This follows aj + ai bj = 0 for all 1  j  m. Consider ai + ai bi = 0 or ai (1 + bi ) = 0, since

bi 6= 1 so ai = 0. We also have aj + ai bj = 0 for all 1j  m so we find aj = 0 for all 1  j  m.
Hence, (v1 , v2 , . . . , vi + w, . . . , vm ) is linearly independent.
exer:2A:6
exer:2A:7
Using this proposition, we can solve exercises 6,exer:2B:3
7 exer:2B:4
in exer:2B:5
chapter 2A. We can construct a new basis
exer:2B:6
from known basis of a vector, such as exercises 3, 4, 5, 6 in chapter 2B.
4.5. Exercises 2B
1. If (v1 , . . . , vm ) is a basis in V then so is (v1 , . . . , vm 1 , vm + v1 ) according to proposition
prop:1
4.4.1. Hence in order for V to have only one basis, we must have vm = vm + v1 or v1 = 0.
Since vector 0 in the list so in order for this list to be linearly independent, the list must
contain 0 only. Thus, this follows {0} is the only vector space that has only one basis.
2. Not hard to do.
exer:2B:3 3. (a) U = {(x1 , x2 , x3 , x4 , x5 ) 2 R5 : x1 = 3x2 , x3 = 7x4 }. Basis of U is {(3, 1, 0, 0, 0), (0, 0, 7, 1, 0), (0, 0, 0,
(b) Add two more to basis of U which are (0, 1, 0, 0, 0) and (0, 0, 0, 1, 0) we will get a basis
of R5 .
theo_6:2B:2.34
(c) According to proof of theorem 4.3.4, pick W = span((0, 1, 0, 0, 0), (0, 0, 0, 1, 0)). then
R5 = U W .
exer:2B:4 4. (a) Basis of U = {(x1 , x2 , x3 , x4 , x5 ) 2 C5 : 6z1 = z2 , z3 + 2z4 + 3z5 = 0} can be

(1, 6, 0, 0, 0), (0, 0, 4, 1, 2), (0, 0, 7, 1, 3). We need to prove that this list is a basis,
i.e. there exists uniquely m, n 2 C so m(4, 1, 2) + n(7, 1, 3) = (z3 , z4 , z5 ) for
any z3 , z4 , z5 2 C so z3 + 2z4 3z5 = 0. Note that 4 + 2 · 1 + 3 · ( 2) = 0 and
7 + 2 · 1 + 3 · ( 3) = 0 so as long as we determine m, n for z3 , z4 then that m, n
also works for z5 . Hence, it deduces to m(4, 1) ( + n(7, 1) = (z3 , z4 ) or 4m + 7n = z3
4 Re(m) + 7 Re(n) = Re(z3 )
and m + n = z4 . We have system of equations and
Re(m) + Re(n) = Re(z4 )
(
4 Im(m) + 7 Im(n) = Im(z3 )
. Hence, m, n can be uniquely determined. Thus,
Im(m) + Im(n) = Im(z4 )
the list is a basis.
prop:1
(b) Proposition
0 4.4.1 can construct
1 a basis of C than contains three vectors in (a). Let
5
vi = @0, 0, . . . , 0, 1, 0, . . . , 0A then v1 , v2 , v3 , v4 , v5 is a basis of C5 then according

| {z }
i prop:1
1
proposition 4.4.1 so does {v1 + 6v2 , v2 , 4v3 + v4 2v5 , v4 , v5 } = {u1 , v2 , u3 , v4 , v5 },
which is
(1, 6, 0, 0, 0), (0, 1, 0, 0, 0), (0, 0, 4, 1, 2), (0, 0, 0, 1, 0), (0, 0, 0, 0, 1).
From here we find u1 , v2 , u3 , 74 u3 3

4 v4 + 12 v5 , v5 is also a basis, or
(1, 6, 0, 0, 0), (0, 1, 0, 0, 0), (0, 0, 4, 1, 2), (0, 0, 7, 1, 3), (0, 0, 0, 0, 1).
theo_6:2B:2.34
(c) According to proof of theorem 4.3.4, pick W = span((0, 1, 0, 0, 0), (0, 0, 0, 0, 1)) and
we are done.
prop:1
exer:2B:5 5. True. Note that 1, x, x2 , x3 is a basis of P3 (F) so according proposition 4.4.1 we find
1, x, x2 + x3 , x3 is also a basis of P3 (F) and one of the vectors has degree 2.
prop:1
exer:2B:6 6. True according to proposition 4.4.1.
7. It’s not true. A counterexample: Pick V = F4 and U = {(x1 , x2 , x3 , x4 ) 2 F4 : x2 = 2x1 }.

Note that v1 = (0, 0, 0, 1), v2 = (0, 0, 1, 0) 2 U and v4 = (1, 0, 0, 0), v3 = (0, 1, 0, 0) 62 U
and (0, 0, 0, 1), (0, 0, 1, 0) is not a basis of U (it does not span U ).
exer:2B:8 8. Since V = U W so every v 2 V can be represented uniquely as v = ru + swPwith u 2

m
U, w 2 W . Since u1 , . . . , um is a basis of U so u can
Pnbe represented uniquely as i=1 ai ui .
Similarly, w can be represented uniquely as w = i=1 bi wi . Hence, v can be represented
uniquely as a linear combination of u1 , . . . , um , w1 , . . . , wn . Hence, u1 , . . . , um , w1 , . . . , wn
is a basis of V .
4.6. 2.C Dimension

theo_9:2C:2.35 Theorem 4.6.1 (2.35) Any two bases of a finite-dimensional vector space the a same
length.
theo1
Proof. This is true according to theorem 4.1.2. By considering a basis v1 , . . . , vm in finite-
dimensional vector space V then v1 , . . . , vm is a list length m which is both linearly independent
and spanning V . Other bases are also linearly independent lists so they must have length less
than length of any spanning list, which is m. They are also spanning lists so their lengh must
be greater than length of any independent list, which is m. Hence, other bases must also have
length m.
Definition 4.6.2. The dimension of a finite-dimensional vector space is the length of any basis
of the vector space. The dimension of V (if V is finite-dimensional) is denoted by dimV .
theo_10:2C:2.38 Theorem 4.6.3 (2.38) If V is finite-dimensional and U is a subspace of V then dim U 

dim V .
theo_5:2B:2.33
Proof. This is true according to theorem 4.3.3, since basis u1 , . . . , um of U is linearly independent
so there exists w1 , . . . , wn 2 V so u1 , . . . , um , w1 , . . . , wn is a basis of V . Therefore, dim U 
dim V .
theo_11:2C:2.39 Theorem 4.6.4 (2.39) Suppose V is finite-dimensional. Then every linearly independent
list of vectors in V with length dim V is a basis of V .
theo_5:2B:2.33
Proof. According to theorem 4.3.3, every linearly independent list can be extended to a basis.
Since the list has length dim V , that means no vector need to be added to the list to create a
basis, i.e. the list is a basis.
theo_12:2C:2.42 Theorem 4.6.5 (2.42) Suppose V is finite-dimensional. Then every spanning list of vectors
in V with length dim V is a basis of V .
theo_13:2C:2.43 Theorem 4.6.6 (Dimension of a sum, 2.43) If U1 and U2 are subspaces of a finite-dimensional
vector space, then
dim(U1 + U2 ) = dim U1 + dim U2 dim (U1 \ U2 ).
Proof. Note that U1 \ U2 is a subspacetheo_5:2B:2.33

of U1 , U2 . Hence, if we let v1 , . . . , vm be basis of vector
space U1 \U2 then according to theorem 4.3.3, there exists w1 , . . . , wn 2 U1 so v1 , . . . , vm , w1 , . . . , wn
is basis of U1 and there exists u1 , . . . , ul 2 U2 so v1 , . . . , vm , u1 , . . . , ul is basis of U2 . Thus,
this follows the right hand side of the identity will be m + n + l. Hence, it suffices to prove
dim(U1 + U2 ) = m + n + l. Indeed, we will prove that
v 1 , . . . , v m , u1 , . . . , u l , w 1 , . . . , w n (2) eq2:2C:2.43
is a basis of U1 + U2 . This list obviously spans U1 + U2 . Before, we go and prove the above
exer:2B:8
is is linearly independent, notice that form exercise 8 2B then U1 = (U \ U2 ) S where S =
span(w 1 , . . . , wn ) so this follows S \ (U1 \ U2 ) = {0} from theorem Direct sum of two subspaces
theo_7:1C:1.45
3.2.4. eq2:2C:2.43
Now, coming back to (2), assume the list is linearly depdendent,theo_8:2A:2.21 and note that v1 , . . . , vm , u1 , . . . , ul
is linearly independent so from Linear Dependence Lemma 4.1.1 we find that there exists
1  j  n so wi 2 span(v1 , . . . , vm , u1 , . . . , ul , w1 , . . . , wj 1 ), or there exists v 2 S, v 6= 0 so
v 2 U2 . Note that S ⇢ U1 so from this we find v 2 U1 \ U2 . Hence, v 2 (S \ (U1 \ U2 )) and
v 6= 0, a contradiction with above argument. Thus, the list is linearly independent. We obtain
eq2:2C:2.43
list (2) is a basis of U1 +U2 so dim (U1 +U2 ) = m+n+l = dim U1 +dim U1 dim (U1 +U2 ).
4.7. Exercises 2C
1. Let v1 , . . . , vm be basis of Utheo_11:2C:2.39
then this list is linearly independent which has length equal
to dim V so from theorem 4.6.4 the list is a basis of V , therefore, U = V .
theo_10:2C:2.38
2. We have dim R2 = 2 and from theorem 4.6.3 then if U is subspace of R2 then dim U  2.
If dim U = 0 then U = {0}, if dim U = 1 then U is U = span((x, y)) which is a line in
R2 through the origin and (x, y) 6= 0. If dim U = 2 then U = R2 .
3. Similarly, we find dim U  3 for U a subspace of R3 . If dim U = 0 then U = {0},

if dim U = 1 then U is set of points in line in R3 through the origin. If dim U = 2
then U = span((x1 , y1 , z1 ), (x2 , y1 , z2 )) so (x1 , y1 , z1 ) 6= k(x2 , y2 , x2 ) with kR, i.e. they are
not in the same line through the origin. This follows U is the plane through the origin,
(x1 , y1 , z1 ) and (x2 , y2 , z2 ).
4. a) With U = {p 2 P4 (F) : p(6) = 0} then x 6, (x 6)x, (x 6)x2 , (x 6)x3 is a basis

of subspace U of P4 (F). Indeed, it’s obvious that the list spans U . The list is also
linearly independent since it doesn’t exist a, b, c, d so a(x 6) + b(x 6)x + c(x
6)x2 + d(x 6)x3 = 0 for all x 2 R.
b) 1, x 6, (x 6)x, (x 6)x2 , (x 6)x3 is a basis of P4 (F) since this is spans P4 (F) and
it has length 5 = dim P4 (F).
theo_6:2B:2.34
c) According to theorem 4.3.4 then W = span (1).
5. a) If U = {p 2 P4 (F) : p00 (6) = 0} then 1, x, (x 6)3 , (x 6)4 is a basis of Utheo_10:2C:2.38

. This is
is obviously linearly independent. Since dim P4 (F) = 5 so from theorem 4.6.3 then
theo_11:2C:2.39
dim U  5 and if dim U = 5 then U = P4 (F) according to theorem 4.6.4, which is
theo_11:2C:2.39
a contradiction since x2 62 U, x2 2 P4 (F). Thus, dim U = 4 so again from 4.6.4 we
have 1, x, (x 6)3 , (x 6)4 is a basis of U .
b) Add x2 to 1, x, (x 6)3 , (x 6)4 we get a basis of P4 (F) because this list is linearly
independent and as length of 5.
theo_6:2B:2.34
c) According to theorem 4.3.4, we find W = span (x2 ).
exer:2C:6 6. a) With U = {p 2 P4 (F) : p(2) = p(5)} then 1, (x 2)(x 5), (x 2)(x 5)x, (x
2)(x 5)x2 is a basis of U . Indeed, this is linearly independent and similarly to above
argument, we find dim U  4 so we are done.
b) Add x to the list and we obtain a linearly independent list with length of 5 so it is a
basis of P4 (F).
theo_6:2B:2.34
c) According to theorem 4.3.4, we find W = span (x).
7. a) With U = {p 2 P4 (F) : p(2) = p(5) = p(6)} then 1, (x 2)(x 5)(x 6), (x
2)(x 5)(x 6)x is a basis of U . Indeed, this is is linearly independent and it is
exer:2C:6
a subspace of U2 = {p 2 P4 (F) : p(2) = p(5)}. From exercise 6 2C we
theo_11:2C:2.39
know
dim U2 = 4 so dim U  4. If dim U = 4 then from theorem 4.6.4 we have U = U2
but (x 2)(x 5) 2 U2 , (x 2)(x 5) 62 U , a contradiction. Thus, dim U  3 so
dim U = 3 which follows 1, (x 2)(x 5)(x 6), (x 2)(x 5)(x 6)x is a basis of
U.
b) Add x, x2 to the list and we get a linearly independent list with length of 5 so the
new list is a basis of P4 (F).
theo_6:2B:2.34
c) According to theorem 4.3.4 we find W = span (x, x2 ).
R1
8. a) With U = {p 2 P4 (F) : 1 p = 0} then the list 2x, 3x
2 1, 4x3 , 5x4 3x2 is a
basis of U . Indeed, this list is linearly independent and from dim U  4 because
we follow dim U = 4 which means the list is a basis of U
x2 62 U, x2 2 P4 (F), theo_11:2C:2.39
according to theorem 4.6.4.
An interesting remark is that if V = {p 2 P5 (F) : p( 1) = p(1), p(0) = 0} then for
a basis of V , a list of antiderivative of all polynomials in basis of V is a basis of U .
b) Add 1 to the list we get a basis of P4 (F).
c) Similarly, W = span (1).
9. If w 62 span (v1 , . . . , vm ) then w+v1 , . . . , w+vm is linearly independent in V so dim span(w+
+ vm ) = m. If w 2 span (v1 , . . . , vm ) then there w can be written uniquely
v1 , . . . , wP
m
as w = i=1 ai vi . WLOG, say the biggest i so ai 6= 0 is j (1  j  m). Consider
U = span(w + v1 , . . . , w + vj 1 , w + vj+1 , . . . , w + vm ). We see that the list w + v1 , . . . , w +
vj 1 , w + vj+1 , . . . , w + vm is linearly independent because otherwise
b1 (w + v1 ) + . . . + bj 1 (w + vj 1) + bj+1 (w + vj+1 ) + . . . + bm (w + vm ) = 0
is true for bi 2 F not all 0. However, this follows w = c1 v1 + . . . + cj 1 vj 1 + cj+1 vj +
. . . + cm vm , a contradiction since aj = 0 in this representation of w. Thus, w + v1 , . . . , w +
vj 1 , w + vj+1 , . . . , w + vm is a basis of U so dim U = m 1. On the other hand, U is a
subspace of span(w + v1 , . . . , w + vm ) so
dim span(w + v1 , . . . , w + vm ) dim U = m 1.
P
10. List p0 , p1 , . . . , pm is linearly independent. Indeed, consider P (x) = m i=1 ai pi for ai 2 F.
We have [x ]P = [x ]am pm so P (x) = 0 for all x 2 R iff [x ]P = 0 or [xm ]am pm = 0
m m m
which follows am = 0. This leads to [xm 1 ]P = [xm 1 ]am 1 pm 1 and we keep going
until ai = 0 for all 0  i  m. Thus, the list is linearly independent with length of
theo_11:2C:2.39
m + 1 = dim Pm (F) so from theorem 4.6.4 we follow p0 , . . . , pm is a basis of Pm (F).
theo_13:2C:2.43
11. From theorem 4.6.6 we follow dim (U \ W ) = 0, i.e. U \ W = {0}. This follows R8 =
U W.
12. We have theo_13:2C:2.43

dim U = dim W = 5 and if U \ W = {0} then dim (U \ W ) = 0. Hence, from
theorem 4.6.6 we find dim (U + W ) = 10. But U + W is a subspace of R9 which has
theo_10:2C:2.38
dimension of 9, a contradiction from theorem 4.6.3. Thus U \ W 6= {0}.
13. Since dim U = dim W = 4 and dim (U + W )  dim R6 = 6 so dim (U \ W ) 2. This

follows there exists two vectors u, w in U \ W so neither of these vector is a scalar multiple
of the other.
theo_13:2C:2.43
14. dim (U1 + . . . , +Um )  dim U1 + . . . + dim Um is true from theorem 4.6.6.
15. Since dim V = n so there exists a basis v1 , . . . , vn of V . Pick Ui = span(vi ) then of course
dimUi = 1 and V = U1 U2 . . . Un .
16. It’s not hard to prove U1 + . . . + Um is finite-dimensional and note that let vi,1 , vi,2 , . . . , vi,bi
be the basis of Ui then dim Ui = bi and
v1,1 , . . . , v1,b1 , . . . , vm,1 , . . . , vm,bm
spans U1 + . . . + Um and is linearly independent since U1 + . . . + Um is a direct sum. Thus,

dim U1 U2 . . . Um = b1 + . . . + bm = dim U1 + . . . + dim Um .
17. This is false, take three lines in R2 as U1 , U2 , U3 then LHS = 2 and RHS = 3.
5. Chapter 3: Linear Maps

5.1. 3.A The vector space of linear maps
Set of all linear maps from V to W is denoted as L(V, W ).
theo_14:3A:3.5 Theorem 5.1.1 (3.5 Linear maps and basis of domain) Suppose v1 , . . . , vn is a basis of V
and w1 , . . . , wn 2 W . There exists a unique linear map T : V ! W such that T vj = wj for
each j = 1, . . . , n.
prop:2 Proposition 5.1.2 (Linear maps take 0 to 0) Suppose T is a linear map from V to W .
Then T (0) = 0.
5.2. Exercises 3A
prop:2
1. We have T (0, 0, 0) = (0 + b, 0) but from proposition 5.1.2 then T (0, 0, 0) = (0, 0) so b = 0.
On the other hand, we have T (1, 1, 1) = (1, 6 + c) so T (2, 2, 2) = 2T (1, 1, 1) = (2, 12 + 2c)
but T (2, 2, 2) = (2, 12 + 8c). Thus, c = 0. If b = c = 0 then it’s obvious that T is linear.
2. Consider p = 1 2 P(R) then T (1) = 3 + b, 15

4 + c sin 1 so T (2) = 2T (1) = 6 + 2b, 2 + 2c sin 1 .
15
However, T (2) = 6 + 4b, 15

2 + c sin 2 . This follows b = 0, c = 0. The reverse is not hard.
0 1
3. Let T @0, . . . , 0, 1, 0, . . . , 0A = (A1,i , A2,i , . . . , Am,i ). Then, for any (x1 , . . . , xn ) 2 Fn we

| {z }
i 1
have
0 1
n
X
T (x1 , . . . , xn ) = xi · T @0, . . . , 0, 1, 0, . . . , 0A ,
| {z }
i=1 i 1
n
X
= xi · (A1,i , A2,i , . . . , Am,i ) ,
i=1
= (A1,1 x1 + . . . + A1,n xn , . . . , Am,1 x1 + . . . + Am,n xn ) .
4. Assume the contrary

P that v1 , . . . , vm is linearly dependent. Hence, there exists ai 2 R not
all 0 so that ni=1 ai vi = 0 in V . Hence,
n
! n
X X
T ai v i = ai T vi = 0.
i=1 i=1
This follows, T v1 , . . . , T vm is linearly dependent, a contradiction. Thus, v1 , . . . , vm is

linearly independent in V .
5. Yes, L(V, W ) is a vector space.

6. Prove 3.9, page 56: Associativity: (T1 T2 )T3 = T1 (T2 T3 ). Let T1 2 L(U, V ), T2 2 L(V, W ), T3 2
L(W, Z). Hence,
(T1 T2 )T3 (u) = (T1 T2 )(T3 (u)) = T1 (T2 (T3 (u))) = T1 (T2 T3 ).
7. Say we have for some v 2 V thentheo_14:3A:3.5

T v = w 2 V but since dim V = 1 so w = v.
Furthermore, according to theorem 5.1.1 there exists a unique linear map T 2 L(V, V ) so
T v = w which is T v = v.
p p
8. Let '(x, y) p
= 3 xy 2 then '(ax, ay) = a 3 xy 2 = a'(x, y) but '((x, y) + (z, t)) = '(x +
z, y + t) = 3 (x + z)(y + t)2 6= '(x, y) + '(z, t). Thus, ' is not linear map.
9. Let '(x + iy) = y + ix then '((x + iy) + (z + it)) = y + t + i(z + x) = '(x + iy) + '(z + it).
But '(i(x + iy)) = '(ix y) = iy + x 6= i'(x + iy) = iy x. Thus, ' is not linear.
10. Consider v1 2 U, v2 2 V \ U and v1 , v2 , Sv1 6= 0. Hence, T (v1 + v2 ) = 0 since v1 + v2 62 U
but T v1 + T v2 = Sv1 6= 0. Thus, T (v1 + v2 ) 6= T v1 + T v2 . Thus, T is not a linear map on
V.
11. Considertheo_5:2B:2.33
S 2 L(U, W ) and let v1 , . . . , vm be basis of U then Svi = wi . According to
theorem 4.3.3, there exists u1 , . . . , ul so v1 , . . . , vtheo_14:3A:3.5
m , u1 , . . . , ul is a basis of V . Pick arbitrary
l vectors `1 , . . . , `l in W . According to theorem 5.1.1, there exists a linear map T : V ! W
so T vi = wi = Svi , T ui = ì .
12. Since V is finite-dimensional so there exer:2A:14
exists a basis v1 , . . . , vm of V . Since W is infinite-
dimensional so according to Exercise 14 (2A), there exists sequence of vectors w1 , w2 , . . .
so that w1 , w2 , . . . , wk is linearly independent for any k 2 N. In order to show that
L(V, W ) is infinite-dimensional, we will prove the existence of seuquences T1 , T2 , . . . where
Ti 2 L(V, W ) and T1 , . . . , Tk is linearly independent for any k 2 N.
theo_14:3A:3.5
Indeed, according to theorem 5.1.1,Pthere exists Ti : V ! W so that Ti (vj ) = w(i 1)m+j
Pm
for all 1  j  m. Thus, for v = m ↵ v
j=1P j j then T i (v) = j=1 ↵j w(i 1)m+j . Thus, for
m
any k 2 N and for all v 2 V, v 6= 0, v = j=1 ↵j vj if
k
X k X
X m
i Ti (v) = 0v () i ↵j w(i 1)m+j = 0v.
i=1 i=1 j=1
Note that w1 , . . . , wkm is linearly independent so the above equation follows that i ↵j = 0
for any 1  i  k, 1  j  m. However, since v 6= 0 so there must exists an ↵l 6= 0 (1 
l  m). This deduces i = 0 for all 1  i  k, i.e. T1 , . . . , Tk is linearly independent. And
this is true for any k 2 N so L(V, W ) is infinite-dimensional.
P
13. Since v1 , . . . , vm is linear dependent so there exists vi = j6=iP ↵j vj with ↵j 2 F. Hence,
pick arbitrary wj (1  j  m, j 6= i) and then pick wi so wi 6= j6=i ↵j wj = T vi . Done.
14. Let v1 , . . . , vm (m 2) be basis of V then for i 3, let Svi = T vi = vi but Sv1 = T v2 =
v1 , Sv2 = T v1 = v2 . Hence, ST (v1 ) = Sv2 = v2 but T S(v1 ) = T ( v1 ) = v2 . Thus,
ST 6= T S.
5.3. 3.B Null Spaces and Ranges

Proposition 5.3.1
(3.14) Suppose T 2 L(V, W ) then null T is a subspace of V .
Proof. According to the definition of null space if v 2 null T then v 2 V . If v1 , v2 2 null T

then T v1 = T v2 = 0 so T (v1 + v2 ) = T v1 + T v2 = 0. Hence, v1 + v2 2 null T . We also have
T ( v1 ) = T v1 = 0 so v1 2 null T for any 2 F.
Proposition 5.3.2
(3.16) Let T 2 L(V, W ) then T is injective if and only if null T = {0}.
Proof. If T is injective. Assume the contrary that there exists v 6= 0, v 2 null T . then for any
u 6= v, u 2 V then T (u + v) = T u + T v = T u, a contradiction since T is injective.
Reversely, if null TT = {0}. For any u, v 2 V so T u = T v then T (u v) = 0. Thus, u = v.
This follows T is injective.
Proposition 5.3.3
(3.19) If T 2 L(V, W ) then range T is a subspace of W .
Proof. We have T (0) = 0 so 0 2 range T . For any v1 , v2 2 range T then there exists u1 , u2 2 V
so T u1 = v1 , T u2 = v2 so T (u1 + u2 ) = v1 + v2 . Hence, v1 + v2 2 range T . We also have
v1 = T u1 = T ( u1 ) so v1 2 range T for any 2 F. Thus, range T is a subspace of W .
Definition 5.3.4. A function T : V ! W is called surjective if its range equal W .
theo_15:3B:3.22 Theorem 5.3.5 (3.22, Fundamental Theorem of Linear Maps) Suppose V is finite-dimensional
and T 2 L(V, W ). Then range T is finite-dimensional and
dim V = dim null T + dim range T.

Pm
Proof. Let vP 1 , . . . , vm be the basis of V then for any v 2 V, v = i=1 ↵i vi then f (v) 2 range T
m
and f (v) = i=1 ↵i f (vi ). This follows range T = span (f (v1 ), . . . , f (vm )) or range T is finite-
dimensional. theo_5:2B:2.33
Let u1 , . . . , um k be basis of null T . Hence, acoording to theorem 4.3.3, linearly independent
list u1 , . . . , um k can be extented to a basis u1 , . . . , um k , w1 , . . . , wk of V . It suffices to prove
that dim range T = k.
Indeed, let W = span(w1 , . . . , wk ). Note that for any u, v 2 W, v 6= u then T v 6= T u, otherwise
T (u v) = 0 which means , u v 6= 0, u v 2 null T but u v 2 W , a contradiction since
u1 , . . . , um k , w1 , . . . , wk is linearly independent.
We prove that (T w1 , T w2 , . . . , T wk ) is a basis of range T . From the above argument, we find
T w1 , . . . , T wk is linearly independent. For any v 2 range T then there exists u 2 V so T u = v.
theo_6:2B:2.34
According to theorem 4.3.4, V = null T W so for any u 2 V there uniquelyP exists u1 2
null T, w 2 W so u = u1 + w. Hence, T u = T u1 + T w = T w or v = T u = T w = ki=1 ↵k T wk
for any v 2 range T . Thus, T w1 , . . . , T wk spans range T, which follows dim range T = k, as
desired.
theo_16:3B:3.23 Theorem 5.3.6 (3.23) Suppose V and W are finite-dimensional vector spaces such that
dim V > dim W . Then no linear map from V to W is injective.
Proof. Asusme the contrary, there is injective linear map T 2 L(V, W ). Hence, null T = {0}
theo_15:3B:3.22
or dim null T = 0. Thus, from theorem 5.3.5, dim V = dim range T  dim W since range T
is a subspace of W , a contradiction. Thus, there doesn’t exist injective linear map from V to
W.
theo_16:3B:3.24 Theorem 5.3.7 (3.24) Suppose V and W are finite-dimensional vector spaces such that
dim V < dim W . Then no linear map from V to W is surjective.
Proof. For any T 2 L(V, W ), we have dim range T = dim V dim null T < dim W , which
means range T 6= W , so T is not surjective.
Theorem 5.3.8
(3.26) A homogeneous system of linear equations with more variables than equations has
nonzero solutions.
Proof. Using the notation from the textbook. This means n > m, or dim Fn > dim Fm .
This follows no linear map from Fn to Fm is injective, i.e. dim nullT (x1 , . . . , xn ) > 0 for any
T 2 L(V, W ). This means the system of linear equations has nonzero solutions.
Theorem 5.3.9
(3.29) An inhomogeneous system of linear equations with more equations than variables
has no solution for some choice of constant terms.
Proof. We have dim Fn < dim Fm so no linear map from Fn to Fm is surjective. Hence,
for any T 2 L(V, W ) then range T 6= Fm . This follows there exists (c1 , . . . , cm ) 2 Fm but
(c1 , . . . , cm ) 62 range T . Thus, the system of linear equations with constant terms c1 , . . . , cm has
no solution.
5.4. Exercises 3B
5.4.1. A way to construct (not) injective, surjective linear map
theo_14:3A:3.5
That is based on theorem 5.1.1. T is injective if basis V is mapped one-to-one with basis of W
(which is also we need dim V  dim W for existence of injective linear map from V to W ). In
other words, if v1 , . . . , vm is basis of V then T 2 L(V, W ) is injective if and only if T v1 , . . . , T vm

are linearly independent in W . T is not injective otherwise.
Similarly, T is surjective if and only if
exer:3B:7
T v1 , . . . , T vm spans W .
exer:3B:8
exer:3B:9
exer:3B:10
Some examples are exercises 7, 8, 9,theo_14:3A:3.5
10.
Many exercises below use theorem 5.1.1!
5.4.2. Exercises
theo_15:3B:3.22
1. If T 2 L(V, W ) then from theorem 5.3.5, dim V = 5. We can let V = F5 and since
range T is a subspace of W so 2 = dim range T  dim W . Hence, we can let W = F2 .
Hence, we can write T (x, y, z, t, w) = (x + y, t + w). With this, null T = {(x, y, z, t, w) 2
F5 : x + y = t + w = 0}. Hence, basis of null T is (1, 1, 0, 0, 0), (0, 0, 1, 0, 0), (0, 0, 1, 1).
And range T = F2 .
2. Since range S ⇢ null T so (T S)(v) = T (Sv) = 0 because Sv 2 range S so Sv 2 null T .

Thus,
(ST )2 (v) = (ST )((ST )v) = ST (Su) = S(T Su) = S(0) = 0.
3. (a) If v1 , . . . , vm spans V then range T = V or T is surjective.

(b) If v1 , . . . , vm are linearly independent then T (z1 , . . . , zm ) = 0 if and only if (z1 , . . . , zm ) =
0, i.e. null T = {0} or T is injective.
4. Let T1 (x, y, z, t, w) = (x + y, 0, 0, t + z + w) then range T1 = {(x, 0, 0, y) 2 R4 } so

dim range T1 = 2 so dim null T1 = 3. Let T2 (x, y, z, t, w) = (x + y, 0, t, 0) then (T1 +
T2 )(x, y, z, t, w) = (2x + 2y, 0, t, t + z + w) so dim range (T1 + T2 ) = 3. This follows
dim null (T1 + T2 ) = 2 or T1 + T2 62 S = {T 2 L(R5 , R4 ) : dim null T > 2}. Hence, S is
not a subspace of L(R5 , R4 ).
theo_15:3B:3.22
5. We contruct such linear map based on proof of theorem 5.3.5. Note that basis of R4 is
(1, 0, 0, 0), (0, 1, 0, 0), (0, 0, 1, 0), (0, 0, 0, 1) so if we choose (1, 0, 0, 0), (0, 1, 0, 0) to be basis
of null T then we need T (0, 0, 1, 0), T (0, 0, 0, 1) to be basis of range T . Therefore, null T =
{(x, y, 0, 0) 2 R4 } = range T so we can set T (0, 0, 1, 0) = (1, 0, 0, 0), T (0, 0, 0, 1) =
(0, 1, 0, 0). Hence, T can be T (x, y, z, t) = (z, t, 0, 0).
6. Because that will mean dim null T = dim range T so 5 = dim R5 = 2 · dim null T , a
contradiction.
exer:3B:7 7. Let v1 , . . . , vm be basis of V and w1 , . . . , wn be basis of W with n m 2. Construct

T so T vi = wi for 1  i  m 1 but T vm = 2wm 1 . Next construct S so Svi = wi for
1  i  m 2 but Svm 1 = 2wm and Svm = wm . It’s obvious that T, S 2 R = {T 2
L(V, W ) : T is not injective}.
However, T + S is injective. Indeed, we have (T + S)(vi ) = wi for 1  i  m 2 and
(T + S)vm 1 = wm 1 + 2wm , (T + S)vm = 2wm 1 + wm . Note that w1 , . . . , wm 2 , wm 1 +
2wm , 2wm 1 + wm are linearly independent so dim null (T + S) = {0} or T + S is injective.
Thus, R is not subspace of L(V, W ).
exer:3B:7
exer:3B:8 8. Notation the same with exercise 7 but this time m n 2. Construct T so T vi = wi for
1  i  n 2, T vj = jwn 1 for n 1  j  m. Construct S so Svi = wi for 1  i  n 2,
T vj = (m + 1 j)wn for n 1  j  m. It’s obvious that T, S 2 R = {T 2 L(V, W ) :
T is not surjective}.
However, T + S is surjective since T v1 , T v2 , . . . , T vn is w1 , . . . , wn 2 , (n 1)wn 1 + (m
n + 2)wn , nwn 1 + (m n + 1)wn , . . . which spans W . Thus, S + T is surjective. Thus, R
is not subspace of L(V, W ).
exer:3B:9 9. T
Pm is injective so for any v 2 span (v1 , . . . , wm ) then T v = 0 if and only if v = 0, i.e.
i=1 ↵i T vi = 0 if and only if ↵i = 0 for all 1  i  m, i.e. T v1 , . . . , T vm is linearly
independent in W .
P Pm
exer:3B:10 10. For any u 2 range T there exists v 2 V or v = m i=1 ↵i vi so T v = u or u = i=1 ↵i T vi .
Thus, any u 2 range T can be expressed as linear combination of T v1 , . . . , T vm which
follows T v1 , . . . , T vm spans range T .
exer:3B:11 11. If we have S1 (S2 . . . Sn u) = S1 (S2 . . . Sn v) then we find S2 . . . Sn u = S2 . . . Sn v since S1 is

injective linear map. Similarly, we from this we deduce u = v. This follows S1 S2 . . . Sn is
injective.
exer:3B:12 12. Since null T is a subspace of finite-dimensional vector space V so according to theorem
theo_5:2B:2.33
4.3.3, a basis v1 , . . . , vm of null T can be extended to a basis of V , i.e. v1 , . . . vm , u1 , . . . , un .
Hence, let U = span (u1 , . . . , un ) then we have U \ null T = {0}. On the other hand, any
v 2 V there exists u 2 U, z 2 null T so v = u + z so T v = T (u + z) = T u. This follows
range T = {T v : v 2 V } = {T u : u 2 U }.
theo_15:3B:3.22
exer:3B:13 13. A basis for null T is (5, 1, 0, 0), (0, 0, 7, 1) so dim null T = 2. Combining with theorem 5.3.5
we find dim range T =theo_11:2C:2.39
2 = dim W and since range T is a subspace of W so range T = W
according to theorem 4.6.4. Thus, T is surjective.
theo_15:3B:3.22
exer:3B:14 14. Similarly, from theorem 5.3.5 to find dim range T = 5 = dim R5 so T is surjective.
exer:3B:15 15. Assume the contrary, there exists such linear map T then dim null T = 2. On the other
hand, dim range T  dim R2 = 2 so dim null T + dim range T  4, a contradiction to
theo_15:3B:3.22
theorem 5.3.5.
exer:2A:14
exer:3B:16 16. Assume the contrary that V is infinite-dimensional then according to exercise 14 there
exists a sequence v1 , v2 , . . . so v1 , . . . , vm is linearly independent for every positive integer
m. WLOG, say that v1 , . . . , vn is basis of null T .
Since range T of a linear map on V is finite-dimensional so there exists u1 , . . . , uk 2 V so
T u1 , . . . , T uk is a basis of range T . Note that there exists a M > n so all uj (1  j  k)
can be represented as linear combination of vi for i < M . Hence, since T vM 2 range T so
X̀ M
X1
T vM = ↵ i T ui = i T vi .
i=1 i=1
PM 1 PM 1 Pn
This follows vM i=1 i vi 2 null T or vM = i=1 i vi + j=1 j vj , a contradiction
since v1 , . . . , vM is linearly independent. Thus, V is finite-dimensional.
theo_16:3B:3.23
exer:3B:17 17. According to theorem 5.3.6, if there exists an injective linear map from V to W then
dim V  dim W . Conversely, if dim V  dim W , let v1 , . . . theo_14:3A:3.5
, vm be basis of V and
w1 , . . . , wn (n m) be basis of W then according to theorem 5.1.1, there exists a lin-
ear map T : V ! W such that T vj = wj for 1  j  m. With this, we will find that
null T = {0} so T is injective.
theo_16:3B:3.24
exer:3B:18 18. According to theorem 5.3.7, if there exists a surjective map from V onto W then dim V
dim W . Conversely, if dim V dim W , let v1theo_14:3A:3.5
, . . . , vm be basis of V and w1 , . . . , wn (n 
m) be basis of W then according to theorem 5.1.1, there exists a linear map T : V ! W
such that T vj = wj for 1  j  n. With this, we find range T = W so T is surjective.
theo_15:3B:3.22
exer:3B:19 19. If there exists such linear map T then according to theorem 5.3.5, we have dim null T =
dim U = dim V dim range T dim V dim W . Conversely, if dim U dim V dim W ,
let u1 , u2 , . . . , um , v1 , . . . , vn be basis of V where u1 , . .theo_14:3A:3.5
. , um be basis of U and let w1 , . . . , wk
be basis of W then k n. According to theorem 5.1.1, there exists a linear map T 2
L(V, W ) so T ui = 0, T vj = wj for 1  j  n, 1  i  m. With k n, we can show that
null T = U .
exer:3B:20 20. If there exists such linear map S then when T u = T v then ST u = ST v or u = v. Thus,
T is injective.theo_14:3A:3.5
Conversely, if T is injective, let T v1 , . . . , T vm be a basis of range T . Hence,
according to 5.1.1, there exists a linear map S 2 L(W, V ) so S(T vi ) = vi and S(wj ) = 0
for arbitrary uj 2 V where T v1 , . . . , T vm , w1 , . . . , wn is basis of W .
P Pm
For any v 2 V then T v = m i=1 ↵i T vi . Since T is injective so null T = 0 so v = i=1 ↵i vi .
Hence, !
Xm X m Xm
ST v = S ↵i T vi = ↵i ST vi = ↵i vi = v.
i=1 i=1 i=1
exer:3B:21 21. If there exists such linear map S then any w 2 W we have T (Sw) = w 2 W so w 2 range T .
Since range T is a subspace of W , this follows range T = W or T is surjective.
Conversely, if T is surjective then range T = W . There exists v , . . . , vm 2 V so T v1 , . . . , T vm
theo_14:3A:3.5 1
is basis of W . Hence, according to theorem 5.1.1, there exists a linear P map S 2 L(W, V )
so S(T vi ) = vi for all 1  i  m. This follows, for any w 2 W, w = m i=1 ↵i T vi then
m
! m
! m
X X X
T Sw = T ↵i S(T vi ) = T ↵i vi = ↵i T vi = w.
i=1 i=1 i=1
exer:3B:22 22. For any v 2 null ST then T v 2 null S. Hence, we define linear map T 0 2 L(null ST, null S)
so that for any v 2 null ST then T 0 v = T v. Note that since null ST is a subspace of theo_15:3B:3.22
V so
null T 0 is a subspace of null T so dim null T 0  dim null T . According to theorem 5.3.5,
we have
dim null ST = dim null T 0 + dim range T 0  dim null T + dim null S.
exer:3B:23 23. Note that range ST is a subspace of range S so dim range ST  dim range S. We can
write range ST = {S(T v) : v 2 U } = {Sv : v 2 range T }. Let u1 , . . . , um be basis of
range T then Su1 , . . . , Sum will span range ST so dim range ST  dim range T . Thus,
dim range ST  min{dim range S, dim range T }.
exer:3B:24 24. (note that V is not necessarily finite-dimensional so we can’t actually set a basis for V ) If
there exists such S, then for any v 2 null T1 then ST1 v = S(0) = 0 = T2 v or v 2 null T2 .
Thus, null T1 ⇢ null T2 .
Conversely, if null T1 ⇢ null T2 , let T1 v1 , . . . , T1 vm , w1 , . . . , wn be basis of W where T v1 , . . . , T vm
is basis of range T . Let S be linear map so S(T vi ) = T2 vi and Swj = 0 for 1  i  m, 1 
j  n.
We
Pm prove that T2 = PmST1 . Indeed, for any v 2 V , we have T v 2 range T so T1 v =
↵ T
i=1 i 1 iv so v i=1 ↵i vi 2 null T1 ⇢ null T2 . Hence,
m
! m m m
!
X X X X
T2 (v) = T2 ↵i vi = ↵ i T2 v i = ↵i ST1 vi = ST1 ↵i T1 vi = S(T1 v).
i=1 i=1 i=1 i=1
exer:3B:25 25. If there exists such S then for all v 2 V we have T1 v = T2 Sv 2 range T2 which follows
range T1 ⇢ range T2 . Conversely, for v1 , . . . , vm as basis of V , since range T1 ⇢ range T2
theo_14:3A:3.5
there exists ui so T1 vi = T2 ui for all 1  i  m. According to theorem 5.1.1, there exists
a linear map S 2 L(V, V ) so Svi = ui for all 1  i  m. This follows T2 Svi = T2 ui = T1 vi
for all 1  i  m, which leads to T2 Sv = T1 v for all v 2 V .
exer:3B:26 26. Let Dxi = pi for all i 1 then deg pi = i 1 so every

P polynomial canPbe represented
as linear combination of pi . For any p0 2 P(R), p0 = m
i=1 ↵ i p i then D( m ↵ xi ) = p0 .
i=1 i
Hence, p0 2 range D. This follows range D = P(R) or D is surjective.
exer:3B:26
exer:3B:27 27. Consider a linear map D 2 L(P(R), P(R)) so Dp = 5p00 +3p0 then according to exercise 26,
D is surjective so for any p 2 P(R) then there exists q 2 P(R) so Dq = p or 5q 00 + 3q 0 = p.
exer:3B:28 28. Let 'i (v) be scalar multiple of wi in representation P of T v. We prove Pthat 'i is linear.
Indeed, since T is linear so T (u + v) = T u + T v or m '
i=1 i (u + v)w i = m
i=1 i u + 'i v)wi .
('
Since w1 , . . . , wm is linearly independent so 'i (u + v) = 'i (u) + 'i (v). Similarly, 'i ( v) =
'i v for any v 2 V .
exer:3B:29 29. It’s obvious that null ' {au : a 2 F} is subset of V . On the other hand, for any v 2 V
then '(v) = a. We know '(u) =6 0 so x = v 'u a
u 2 V . Hence, we have
✓ ◆
a
a = '(v) = ' x + u = a + '(x).
'u
Thus, 'x = 0 or x 2 null '. Hence, any v 2 V can be represented uniquely as v = w + au

where w 2 null ' and a 2 F. This follows null ' {au : a 2 F} = V .
exer:3B:29
exer:3B:30 30. According to exercise 29, if there is u 2 V not in null '1 = null '2 then let c =
'1 (u)/'2 (u). For any v 2 V, v can be represented as v = z+au, (a 2 F) where z 2 null '1 .
Thus, '1 (v) = '1 (au) = a'1 (u) = c'2 (v).
exer:3B:31 31. Null space of T1 , T2 is {(x1 , x2 , x3 , 0, 0) : x1 , x2 , x3 2 R}. We have T1 (0, 0, 0, 1, 0) =

T2 (0, 0, 0, 0, 1) = (1, 0) and T1 (0, 0, 0, 0, 1) = T2 (0, 0, 0, 1, 0) = (0, 1).
5.5. Exercises 3C
exer:3C:1 1. Let v1 , . . . , vn be basis of V and w1 , . . . , wm be basis of W where w1 , . . . , wk (k =
dim range T  m) is basis of range T since range T is subspace of W . We know that
T v1 , . . . , T vn spans range T and for any 1  i  k then wi 2 range T , hence wi can be
written as linear combination of T vj (1  j  n). We have
n
X n
X m
X
wi = ↵j T vj = ↵j Ah,j wj .
j=1 j=1 h=1
In order for the above to hold, there must exist a 1  h  m so Ah,i 6= 0. This is true for
any 1  i  k. Thus, M(T ) has at least k = dim range T nonzero entries.
exer:3C:2 2. Let v1 , . . . , v4 be basis of P3 (R) and w1 , . . . , w3 be basis of P2 (R). From the matrix, since
Ai,j = 0 for i 6= j and Ai,i = 0 so we find Dvk = wk for 1  k  3 and Dv4 = 0. So we
can choose vi = xi for 1  i  3 and v4 = 0 then wi = ixi 1 for 1  i  3.
exer:3C:3 3. Let T v1 , . . . , T vm be basis of range T where vi 2 V then v1 , . . . , vm is linearly independent

in V , hence u1 , . . . , un can be added to the list to create a basis of V . This follows that
u1 , . . . , un is basis of null space of T . Thus, with v1 , . . . , vm , u1 , . . . , un as basis of V and
T v1 , . . . , T vm as basis of range T , we have Aj,j = 1 for 1  j  dim range T .
exer:3C:4 4. Since the first column of M(T ) (except for possibly A1,1 = 1) so we can let w1 = T v1 . It
doesn’t matter what should w2 , . . . , wn specifically be as long as w1 , . . . , wn is basis of W .
P
exer:3C:5 5. If there doesn’t exists v = w1 + ni=2 ↵i wi in range T then for any u 2 V , since T vi 2
range T so we must have A1,i = 0 for all 1  i  m, or entries of the first row of M(T )
are all 0.
P
If there does exist such w = w1 + ni=2 ↵i wi in range T then that means there exists v1 2 V
so T v1 = w. Extend v1 to a basis v1 , . . . , vm of V with T vi = ui for i 2. Note that
if v1 , . . . , vm is basis of of V then so does v1 , v2 v ,
2 1 3v v
3 1 , . . . , v m m v1 . Hence,
T (vi i v 1 ) = ui i w. By choosing the right i to remove w1 from ui , we find that
v 1 , v2 2 v 1 , v3 3 v1 , . . . , vm m v1 is the desired basis of V , which will gives A1,k = 0
for all 1  k  m.
exer:3C:6 6. If there exists such bases of V, W then T v1 = T v2 = . . . = T vm 6= 0. Since v1 , v2 , . . . , vm

is basis of V so v1 , v2 v1 , v3 v1 , . . . , vm v1 is also basis of V . Note that T (vi v1 ) = 0
for all 2 theo_15:3B:3.22
 i  m so that means dim null V = m 1 or dim range T = 1 according to
theorem 5.3.5.
Conversely, if dim range T = 1 then dim null V = m 1 so there exists basis v2 , . . . , vm

of null V . This list can be exented to basis v1 , . . . , vm of V . Hence, T v1 = T (v1 + v2 ) =
. . . = T (v1 + vm ) and v1 , v1 + v2 , . . . , v1 + vm is basis of V . From this, we can choose basis
of W to obtain matrix M(T ) of entries 1.
exer:3C:7 7. Let M(T ) = A, M(S) = C then T vP k = A1,k w1 + . . . + Am,k wm and Svk = C1,k w1 + . . . +
Cm,k wm . This follows (T + S)vk = m i=1 (Ai,k + Ci,k )wi . Thus, (i, k)th entry of M(T + S)
is Ai,k + Ci,k , which also equals to (i, k)th entry of M(T ) + M(S).
exer:3C:8 8. We have ( T )vk = (T vk ).
exer:3C:9 9. Done.
exer:3C:10 10. It suffices to prove (AC)j,k = (Aj,. C)1,k for any 1  k  p. Indeed,
n
X n
X
(Aj,. C)1,k = (Aj,. )1,i Ci,k = Aj,i Ci,k = (AC)j,k .
i=1 i=1
exer:3C:11 11. We have

(a1 C1,. + · · · + an Cn,. )1,k = (a1 C1,. )1,k + . . . + (an Cn,. )1,k ,
Xn
= ai Ci,k = (aC)i,k .
i=1
✓ ◆ ✓ ◆ ✓ ◆ ✓ ◆
1, 0 1, 1 1, 1 1, 0
exer:3C:12 12. Pick A = and C = then AC = but CA = .
0, 0 0, 0 0, 0 0, 0
exer:3C:13 13. Prove A(B + C) = AB + AC, say A is m-by-n matrix and B, C are n-by-p matrix. Hence,
n
X
(AB + AC)i,j = (AB)i,j + (AC)i,j = (Ai,h Bh,j + Ai,h Ch,j )
h=1
n
X
= Ai,h (B + C)h,j = (A(B + C))i,j .
h=1
exer:3C:14 14. A, B, C are m-by-n,n-by-p,p-by-q matrices then

p
X
((AB)C)i,j = (AB)i,h Ch,j ,
h=1
Xp n
X
= Ch,j Ai,l Bl,h ,
h=1 l=1
n
X p
X
= Ai,l Bl,h Ch,j ,
l=1 h=1
Xn
= Ai,l (BC)l,j , = (A(BC))i,j .
l=1
exer:3C:14
exer:3C:15 15. Just did in 14.
5.6. 3D: Invertibility and Isomorphic Vector Spaces

Theorem 5.6.1 (3.56)
A linear map is invertible or non-singular if and only if it is injective and surjective.
exer:3B:20
exer:3B:21
This is deduced from exercises 20 and 21 (3B).
theo:3.59:3D Theorem 5.6.2 (3.59) Two finite-dimensional vector space over F are isomorphic if and
only if they have same dimension.
theo:3.65:3D Theorem 5.6.3 (3.65) Suppose T 2 L(V, W ) and v inV . Suppose v1 , . . . , vn is basis of V
and w1 , . . . , wm is basis of W . Then
M(T v) = M(T )T (v).
theo:3.69:3D Theorem 5.6.4 (3.69) Suppose V is finite-dimensional and T 2 L(V ). Then the following
are equivalent:
(a) T is invertible.
(b) T is surjective.
(c) T is injective.
5.7. Exercises 3D
exer:3D:1 1. We have (ST )(T 1S 1) = I and (T 1 S 1 )(ST ) = I.
exer:3D:2 2. Let v1 , . . . , vn be basis of V . Consider noninvertible operators Ti 2 L(V ) so Ti (vj ) = 0

if j 6= i and Ti (vi ) = vi . Hence, linear map T1 + . . . + Tn is the identity map, which is
invertible. Thus, set of noninvertible operators on V is not a subspace of L(V ).
exer:3D:3 3. If there exists such invertible operator T then if Su = Sv then T u = T v which follows
u = v. Thus, S is injective.
If S is injective, since U is subspace of V , basis v1 , . . . , vm of U can be extended to basis
v1 , . . . , vn (m  n) of V . Since S is injective so Sv1 , . . . , Svm is linearly independent, so
this list can be extended to basis Sv1 , . . . , Svm , um+1 , . . . , un of V . From this, we let T be
a linear map so T vi = Svi for 1  i  m and T vi = ui for m < i  n. We find that T is
surjective so T is invertible and T u = Su for any u 2 U .
exer:3D:4 4. If there exists such invertible operator S then if u 2 null T1 then 0 = T1 u = ST2 u so
T2 u = 0. Similarly if u 2 null T2 then u 2 null T1 . Thus, null T1 = null T2 .
If null T1 = null T2 : Let T2 v1 , T2 v2 , . . . , T2 vm , w1 , . . . , wn be basis of W . Note that

T1 v1 , . . . , T1 vm is linearly independent, otherwise there exists T1 v = 0 where v is linear
combination of v1 , . . . , vm . This means v 2 null T1 = null T2 so T2 v = 0, which leads to
T2 v1 , . . . , T2 vm being linearly dependent, a contradiction. Thus, T1 v1 , . . . , T1 vm is linearly
independent so it can be extended to basis T1 v1 , . . . , T1 vm , u1 , . . . un of W .
Let S 2 L(W ) be linear map so S(T2 vi ) = T1 vi for 1  i  m and Swj = uj for 1  j  n.
This follows S is surjective so S is invertible.
P Pm
Now we prove ST2 = T1 . Indeed, for any v 2 V then T2 v = m i=1 ↵i T2 vi so v i=1 ↵i vi 2
null T2 = null T1 so
m
! m m m
!
X X X X
T1 v = T1 ↵i vi = ↵ i T1 v i = ↵i ST2 vi = ST2 ↵i VI = S(T2 v).
i=1 i=1 i=1 i=1
exer:3D:5 5. If there exists such invertible operator S then for any u 2 range T1 , u = T1 v = T2 (Sv) so
u 2 range T2 . Similarly, if u 2 range T2 then u 2 range T1 . Thus, range T1 = range T2 .
Conversely, for v1 , . . . , vm as basis of V , since range T1theo_14:3A:3.5
= range T2 there exists ui so
T1 vi = T2 ui for all 1  i  m. According to theorem 5.1.1, there exists a linear map
S 2 L(V, V ) so Svi = ui for all 1  i  m. This follows T2 Svi = T2 ui = T1 vi for all
1  i  m, which leads to T2 Sv = T1 v for all v 2 V . On the other hand, since v1 , . . . , vm
is linearly independent so u1 , . . . , um is linearly independent so it is basis of V . Hence, S
is surjective so S is invertible.
exer:3D:6 6. If there exists such R, S. Let v1 , . . . , vm be basis of null T1 then ST2 Rvi = 0. Since S is
invertible so S is injective, which means T2 Rvi = 0. Thus, Rvi 2 null T2 and Rv1 , . . . , Rvm
is linearly independent so dim null T2 dim null T1 . Since R is invertible so R is surjective
so let Ru1 , . . . , Run be basis of null T2 . Hence S(T2 Rui ) = S(0) = 0 so T1 ui = 0 so
ui 2 null T1 . Since u1 , . . . , un is linearly independent so dim null T2  dim null T1 . Thus,
dim null T1 = dim null T2 .
Conversely, if dim null T1 = dim null T2 : Let v1 , . . . , vm be basis of V where vn+1 , . . . , vm (n 
m) is basis of null T1 . Let u1 , . . . , um be another basis of V where un+1 , . . . , um is basis
of null T2 . This folows that two lists T1 v1 , . . . , T1 , vn and T2 u1 , . . . , T2 un are linearly inde-
pendent so they can be extended to two bases of W , which are T1 v1 , . . . , T1 vn , w1 , . . . , wk
and T2 u1 , . . . , T2 un , z1 , . . . , zk .
theo_14:3A:3.5
According to theorem 5.1.1, there exists a linear map R 2 L(V ) so Rvi = ui for 1  i  m.
Since u1 , . . . , um is basis of V so R is surjective so R is invertible. There also exists linear
map S 2 L(W ) so S(T2 ui ) = T1 vi for 1  i  n and Szj = wj for 1  j  k. We can see
that S is also surjective so S is invertible.
P
Now we prove T1 = ST2 R. Indeed, for any v = m i=1 ↵i vi in V , we have
m
! n
!
X X
ST2 Rv = ST2 ↵i ui = ST2 ↵i ui (since T2 uj = 0 for m j > n),
i=1 i=1
n
! n
X X
=S ↵ i T2 ui = ↵ i T1 v i ,
i=1 i=1
m
X
= ↵i T1 vi = T1 v (since T1 vj = 0 for m j > n).
i=1
exer:3D:7 7. (a) For any T1 , T2 2 E then we have (T1 + T2 )(v) = T1 v + T2 v = 0 so T1 + T2 2 E and

( T1 )(v) = (T1 v) = 0 so T1 2 E. We also have 0 2 E since 0(v) = 0. Thus, E is
subspace of V .
(b) Let v, v1 , . . . , vn be basis of V and w1 , . . . , wm be basis of W . We prove that M with
respect to these bases is an isomorphism between two vector spaces E and D = {A 2
Fm,n : A.,1 = 0} (need to prove D is vector space first, which is not hard).
Indeed, M is linear. If M(T ) = 0 then T = P0. Thus, M is injective. For any A 2 D, let
T be the linear map from E to D so T vk = m i=1 Ai,k wi for 1  k  n and T v = 0. This
follows T 2 E. Thus, M is surjective, which follows M is invertible.
theo:3.59:3D
Hence, according to theorem 5.6.2, dim E = dim D. One can easily verify that dim D =
mn = (dim V 1)dim W so dim E = (dim V 1)dim W .
exer:3D:8 8. Since T is surjective so range T = W so W is finite-dimensional. Let v1 , . . . , vm be
basis of V where vn+1 , . . . , m (n  P m) is basis of null T . Let U = span Pm(v1 , . . . , vn ).
m
For any w 2 W , there exists v = ↵ v
i=1 i i so T v = w so u = v i=n+1 ↵i vi is
in U and T u = T v = w. Thus, T |U is surjective. If T u = T v for u, v 2 U then
u v 2 null T = span (vn+1 , . . . , vm ) so u v = 0 since v1 , . . . , vm is linearly independent.
Thus, T |U is injective. Thus, T |U is an isomorphism of U onto W .
exer:3D:9 9. If both S, T are invertible then T 1 S 1 is inverse of ST so ST is invertible. Conversely, if
ST is invertible then there is an inverse X of ST . For any v 2theo:3.69:3D
V , there exists T Xv 2 V so
S(T Xv) = v. Hence, S is surjective so according to theorem 5.6.4, S is invertible. On the
other hand, if T u = T v for u, v 2 V then XS(T u) = XS(T v) but u = XST u, v = XST v
theo:3.69:3D
so u = v. Thus, T is injective so according to theorem 5.6.4, T is invertible.
exer:3D:10 10. If ST = theo:3.69:3D
I then for any v 2 V , we have S(T v) = v so S is surjective so according to
theorem 5.6.4, S is invertible. Let S 1 be inverse of S then we have T = IT = (S 1 S)T =
S 1 (ST ) = S 1 . Thus, T is inverse of S so T S = I. Similarly, if T S = I then ST = I.
exer:3D:10
exer:3D:11 11. According to exercise 10 (3D) if S(T U ) = I then (T U )S = T (U S) = I then (U S)T = I
so T 1 = U S.
exer:3D:12 12. Define T, U, S 2 L(P(R)) so U v = vx3 for any v 2 P(R); T xi = xi 2 for i 2 and
i
T x = x, T 1 = 1; Sx = x i 1 for i 1 and S1 = 1. With this, we have ST U = I. However,
if T is not invertible since T is not injective as T (2x2 + x3 ) = T (2 + x3 ) = 2 + x.
theo:3.69:3D
exer:3D:13 13. Note that RST 2 L(V ) and RST is surjective exer:3D:9
so according to theorem 5.6.4, RST is
invertible operator. According to exercise 9 (3D), since RST is invertible so R and ST are
also invertible, which follows S, T, R are invertible. Thus, S is injective.
exer:3D:14 14. T is linear. We have T u = T v then W(u) = W(v) so u = v. Thus, T is injective. For any
A 2 Fn,1 there exists a v 2 V so T v = A. Thus, M is isomorphism of V onto Fn,1 .
exer:3D:15 15. Let A = M(T ) with respect to standard bases of both Fn,1 , Fm,1 . Then fortheo:3.65:3D
any v 2 Fn,1 ,
we have M(v) = v and T v = M(T v). Hence, according to theorem 5.6.3, we have
T v = M(T v) = M(T )M(v) = Av.
exer:3D:16 16. Let v1 , . . . , vn be basis of V . First, we prove that for any 1  i  n, there exists Ci 2 F
so T vi = Ci vi . Indeed, we can choose linear map S so Svi = vi P and Svj = 2vj for
1  j  n, j 6= i. This follows ST vi = T (Svi ) = T vi . Hence, if T vi = nk=1 ↵k vk then
n n
!
X X X
↵k vk = T vi = ST vi = S ↵k vk = ↵i vi + 2 ↵k vk .
k=1 k=1 k6=i
P
This follows k6=i ↵k vk = 0, which happens when ↵k = 0 for k 6= i. Thus, T vi = ↵i vi =
Ci vi .
Next, we prove that for any two 1  j, i  n so j 6= i then Ci = Cj = C. Indeed, pick an
invertible operator S so Svi = vj then T Svi = T vj = Cj vj but ST vi = S(Ci vi ) = Ci vj .
Hence, Ci vj = Cj vj so Ci = Cj = C. Thus, T vi = Cvi for all 1  i  n, or T is scalar
multiple of I.
exer:3D:17 17. (For this, it’s better to think linear maps as matrices) Let v1 , . . . , vn be basis of V . Denote
linear map Ti,j as T vj = vi , T vk = 0 for all k 6= j, 1  k  n (image this as n-by-n matrix
with 0 in all entries except for 1 in (i, j)-th entry).
P
If all T 2 E is 0 then E = {0}. If there exists T 6= 0, WLOG T v1 = ni=1 ↵i vi with ↵1 6= 0.
We find Ti,1 (T T1,1 ) 2 E. Note that Ti,1 T T1,1 = ↵1 Ti,1 so Ti,1 2 E for all 1  i  n. Since
Ti,1 2 E so Ti,j = Ti,1 T1,j 2 E for all 1  j  n. Thus, Ti,j 2 E for all 1  i, j  n. Since
T1,1 , . . . , Tn,n is linearly independent and dim V = n2 so E = V .
exer:3D:18 18. Define a map T from V onto L(F, V ) so for any v 2 V , T v = Rv 2 L(F, V ) so Rv 1 = v.
it’s not hard to verify that T is linear. If T u = T v then Rv = Ru so Rv (1) = Ru (1) so
u = v. Thus, T is injective. For any R 2 L(F, V ), then there is R(1) 2 V so V (R(1)) = R.
Thus, T is surjective. In conclusion, T is an isomorphism so V and L(F, V ) are isomorphic
vector spaces.
exer:3D:19 19. (a) For any q 2 P(R), which we can restrict to q 2 Pm (R). Define T 0 : Pm (R) ! Pm (R)
so T 0 p = T p for p 2 Pm (R). Since deg T p  deg p = m so T 0 is indeed an operator on
finite-dimensional Pm (R). Since T is injective so T 0 is also injective, which follows T 0 is
invertible. Hence, for q 2 Pm (R) there exists p 2 Pm (R) so T 0 p = T p = q. This follows T
is surjective.
(b) Assume the contrary, there is non-zero p so deg T p < deg p. Let T p = q. From
(a), there exists r 2 P(R) so deg r  deg q and T r = q. This follows T p = T r so
p = r. However deg p > deg T p = deg q deg r, a contradiction. Thus, we must have
deg T p = deg p for any non-zero p 2 P(R).
exer:3D:20 20. Denote n-by-n matrix A with Ai,j is the (i, j)-th entry. Define a linear map T 2 L(Fn,1 )
so T v = Av. Hence, (a) is equivalent to T being injective and (b) is equivalent to T being
theo:3.69:3D
surjective. Thus, (a) is equivalent to (b) according to theorem 5.6.4.
5.8. 3E: Products and Quotients of Vector Spaces

theo:3.85:3E Theorem 5.8.1 (3.85) Suppose U is a subspace of V and u, v 2 V . Then the following are
equivalent:
(a) v w 2 U.
(b) v + U = w + U .
(c) (v + U ) \ (w + U ) 6= 0.
theo:3.89:3E Theorem 5.8.2 (3.89) Suppose V is finite-dimensional and U is subspace of V . Then

dim V /U = dim V dim U .
theo:3.91:3E Theorem 5.8.3 (3.91) Suppose T 2 L(V, W ). Define T̃ : V /(null T ) ! W by T̃ (v +

null T ) = T v. Then:
(a) T̃ is a linear map from V /(null T ) to W .
(b) T̃ is injective,
(c) range T̃ = range T ,
(d) V /(null T ) is isomorphic to range T .
Remark 5.8.4. Be careful with vector space V /U , because there are more than one v 2 V to
give the same affine subset v + U of V . So if we define a map T from V /U to some W , be sure
that T makes senses, i.e. for v + U = u + U then T (v + U ) = T (u + U ).
5.9. Exercises 3E
exer:3E:1 1. If T is a linear map then for any two u, v 2 V we have (u, T u)+(v, T v) = (u+v, T (u+v)) 2
graph of T and (u, T u) = ( u, T u) 2 graph of T . We also have (0, T 0) 2 graph of T so
graph of T is a subspace of V ⇥ W .
Conversely, if graph of T is a subspace of V ⇥ W . We have (u, T u) + (v, T v) = (u + v, T u +
T v) 2 graph of T so T u + T v = T (u + v) (otherwise if T u + T v 6= T (u + v) then function
T maps u + v to two different results, a contradiction). Similarly, (u, T u) = ( u, T u) 2

range T so T u = T ( u). Thus, T is a linear map.
exer:3E:2 2. Let (v1,1 , v1,2 . . . , v1,m ), (v2,1 , . . . , v2,m ), . . . , (vk,1 , . . . , vk,m ) be basis of V1 ⇥ V2 ⇥ · · · ⇥ Vm .
Assume the contrary that V1 is infinite-dimensional. then there exists v 2 V1 so v 62
span (v1,1 , v2,1 , . . . , vk,1 ). This follows (v, u2 , . . . , um ) for any ui 2 Vi is not a linear com-
bination of (vi,1 , . . . , vi,m ), 1  i  k, a contradiction. Thus, Vi is finite-dimensional.
exer:3E:3 3. Let V = P(R), U1 = span (1, x, x2 ) and U2 = {p 2 P(R) : deg p 2}. We find U1 + U2 is
not a direct sum since 1 + x + 2x2 + x3 = (1 + x + x2 ) + (x2 + x3 ) = (1 + x) + (2x2 + x3 ).
Define a linear map T from U1 ⇥ U2 onto U1 + U2 so T (u1 , u2 ) = u1 + xu2 . We can
find that T is surjective. If T (u1 , u2 ) = T (v1 , v2 ) then u1 + xu2 = v1 + xv2 . Note that
deg u1 < deg xu2 for any u1 2 U1 , u2 2 U2 so this only happens when u1 = v1 , u2 = v2 .
Thus, T is injective which follows T is an isomorphism from U1 ⇥ U2 onto U1 + U2 .
exer:3E:4 4. Define a map S from L(V1 ⇥ · · · ⇥ Vm , W ) onto L(V1 , W ) ⇥ · · · ⇥ L(Vm , W ) so for any
T 2 L(V1 ⇥ · · · ⇥ Vm , W ), S(T ) = (R1 , . . . , Rm ) where Ri are map from Vi onto W so: for
any v = (0, . . . , 0, vi , 0, . . . , 0) 2 V1 ⇥ · · · ⇥ Vm then Ri vi = T v = w 2 W . It’s not hard to
check that Ri is a linear map.
Now we prove that S is a linear map. Say S(T1 ) = (R1 , . . . , Rm ) and S(T2 ) = (U1 , . . . , Um )
then S(T1 ) + S(T2 ) = (R1 + U1 , . . . , Rm + Um ). On the other hand, for any 1  i  m, we
have for any vi 2 Vi then (Ri + Ui )vi = Ri vi + Ui vi = T1 v + T2 v = (T1 + T2 )v where v =
(0, . . . , 0, vi , 0, . . . , 0). This follows S(T1 + T2 ) = (R1 + U1 , . . . , Rm + Um ) = S(T1 ) + S(T2 ).
Similarly, for any (R1 , . . . , Rm ) 2 L(V1 , W ) ⇥ · · · ⇥ L(Vm , W ) we can define respective
T 2 L(V1 ⇥ · · · ⇥ Vm , W ). Thus, S is surjective. If S(T1 ) = S(T2 ) then (R1 , . . . , Rm ) =
(U1 , . . . , Um ) so Ri = Ui or T1 vi = T2 vi where vi = (0, . . . , 0, vi , 0, . . . , 0) for all 1  i  m.
This leads to T1 = T2 . Thus, S is injective so S is an isomorphism from L(V1 ⇥· · ·⇥Vm , W )
onto L(V1 , W ) ⇥ · · · ⇥ L(Vm , W ) so these two vector spaces are isomorphic.
exer:3E:4
exer:3E:5 5. Similarly to previous exercise 4, define a linear map S from L(V, W1 ⇥ · · · ⇥ Wm ) onto
L(V, W1 ) ⇥ · · · ⇥ L(V, Wm ) so for T 2 L(V, W1 ⇥ · · · ⇥ Wm ), S(T ) = (R1 , . . . , Rm ) where
Ri 2 L(V, Wi ) so: if T v = (w1 , . . . , wm ) then Ri vi = wi .
exer:3E:6 6. Define a linear map T from V n onto L(Fn , V ) so T (v1 , . . . , vn ) = S 2 L(Fn , V ) so

S(0, . . . , 0, 1i , 0, . . . , 0) = vi . T is obviously surjective. If T (v1 , . . . , vn ) = T (u1 , . . . , un )
then Sv = Su so Sv (0, . . . , 0, 1i , 0, . . . , 0) = Su (0, . . . , 0, 1i , 0, . . . , 0) which means vi = ui .
This follows (v1 , . . . , vn ) = (u1 , . . . , un ). Thus, T is injective so two vector spaces V n and
L(Fn , V ) are isomorphic.
exer:3E:7 7. For any utheo:3.85:3E

2 U , we have v + u 2 v + U = x + W so v + u x 2 W which according to
theorem 5.8.1 follows v + u + W = x + W = v + U so u + W = U . Since u 2 U so W ✓ U .
Similarly, for any w 2 W then w + x 2 x + W = v + U so w + x v 2 U so w + x + U =
v + U = x + W so w + U = W . Thus, U ✓ W . In conclusion, we find U = W .
exer:3E:8 8. If A is an affine subset of V then A is r + U for some r 2 V and some subspace U of

V . For v, w 2 A then there exists u1 , u2 2 U so r + u1 = v, r + u2 = w. We find
v + (1 )w = r + u1 + (1 )u2 2 A for any 2 F.
Conversely, if v + (1 )w 2 A for all v, w 2 A and all 2 F: Pick arbitrary w 2 A, we
prove that U = {u w : u 2 A} is a subspace of V . Indeed, for any x1 , x2 2 U then x1 =
u1 w, x2 = u2 w for u1 , u2 2 A. Pick = 2 then 2u1 w 2 A, 2u2 w 2 A. Pick = 1/2
then 1/2(2u1 w) + 1/2(2u2 w) 2 A or u1 + u1 w 2 A. This follows u1 + u2 2w 2 u
or x1 + x2 2 U for any x1 , x2 2 U . On the other hand, (2u1 w) + (1 )w 2 A or
2 u1 2 A for any u1 2 A. Thus, U is a subspace of V . This follows A = w + U so A is an
affine subset of V .
exer:3E:8
exer:3E:9 9. If A1 \A2 is nonempty then for any x, y 2 A1 \A2 , according to exercise 8, x+(1 )y 2 A1
and x + (1 )y 2 A 2 so x + (1
exer:3E:8
)y 2 A1 \ A2 for any 2 F, x, y 2 A1 \ A2 . Thus,
according to exercise 8, we find A1 \ A2 is an affine subset of V .
exer:3E:9
exer:3E:10 10. This is deduced from exercise 9.
Pm Pm
exer:3E:11 11. (a) For any x, y 2
P A where x = i=1 i v i , y = i=1 i vi then for any ↵ 2 F, we have
(1 ↵)x + ↵y = m i=1 vi [ i (1 ↵) + ↵ i ]. Since
m
X m
X m
X
i (1 ↵) + ↵ i = (1 ↵) i +↵ i =1
i=1 i=1 i=1
exer:3E:8
so according to exercise 8, A is an affine subset of V .
(b) Let U be affine subset of V that contains v1 , . . . , vm . Hence, 1 + 2 v2 2 U for any
1 , 2 2 F; 1 + 2 = 1. This follows 3 v3 + (1 3 )( 1 v1 + 2 v2 ) 2 U for any 3 2 F.
Since 3 + (1 3 ) 1 + (1 3 ) 2 = 1 so we can say that 3 v3 + 2 v2 + 1 v1 2 U for any
1 , 2 , 3 2 F so 1 + 2 + 3 = 1. P Similarly, 4 v4 + (1 Pm 4 )( 1 v1 + 2 v2 + 3 v3 ) 2 U .
m
With this, we will eventually obtain i=1 i vi 2 U for i=1 i = 1 or U contains A.
P Pm
(c) We
Pm find v + 0 2 APso v = m i=1 ↵i vi with
P i=1 ↵i = 1. This follows for any u 2 U then
u = i=1 i vi with m i=1 i = 0 or u = m 1
i=1 i (vi vm ) since m = 1 ... m 1.
Thus, U is a subspace of span (v1 vm , v2 vm , . . . , vm 1 vm ) so dim U  m 1.
exer:3E:12 12. P
Let v1 + U, v2 + U, . . . , vm + U be basis of V /U then v1 , . . . , vm is linearly independent and
m
i=1 ↵i vi 62 U for any not-all-zero ↵i 2 F.
Pm
Define a linear
Pm map T from V onto U ⇥ (V /U ) so T v = (v
theo:3.85:3E Pi=1 ↵i vi , v + U ) where
m
v + U = ( i=1 ↵i vj ) + U . According to theorem 5.8.1 then v i=1 ↵i vi 2 U so T v 2
U ⇥ (V /U ).
P
For any (w, v + U ) 2PU ⇥ (V /U ) there exists ↵1 , . . . , ↵m 2 F so v + U = ( m i=1 ↵i vi ) + U .
By letting u = w + m ↵ v
i=1 i i 2 V then we have T u = (w, v + U ). Thus, T is surjective.
IfPT v1 = T v2 or (w1 , x1 + U ) = (w2 , x2 + U ) then w1 = w2 and x1 + U = x2 + U =
( m i=1 ↵i vi ) + U . Hence,
m
X m
X
v1 = w1 + ↵i vi = w 2 + ↵i vi = v2 .
i=1 i=1
Thus, T is injective. In conclusion, vector spaces V is isomorphic to U ⇥ (V /U ).

P
exer:3E:13 13. Since v1 + U, v2 + U, . . . , vP
m + U is basis of V /U P so if mi=1 ↵i (vi + U ) = 0 + U then ↵i = 0
for all 1  i  m, i.e. if m ↵ v
i=1 i i 6
= 0 then m
i=1 i i 62 U . We also find that v1 , . . . , vm
↵ v
is linearly independent. From all the above, we deduce v1 , . . . , vm , u1 , . . . , un is a linearly
theo:3.89:3E
dependent list. According to theorem 5.8.2 then m + n = dim V /U + dim U = dim V so
linearly independent list v1 , . . . , vm , u1 , . . . , un is basis of V .
exer:3E:14 14. (a) If (x1 , x2 , . . .) 2 U then there exists M1 2 N so for all i M1 then xi = 0. Sim-
ilarly, if (y1 , y2 , . . .) 2 U then there exists M2 . Thus, for (x1 + y1 , x2 + y2 , . . .), for all
i max{M1 , M2 } then xi + yi = 0, which means (x1 + y1 , x2 + y2 , . . .) 2 U . Similarly
(x1 , . . .) 2 U . And (0, 0, . . .) 2 U so U is subspace of F1 .
(b) Denote pi as i-th prime number and vi 2 F1 so for any k 2 N, k 1 then the kpi -th
coordinate of vi is 2, the rest of the coordinates of vi is 1. We will prove that prove that
v1 + U, . . . , vm + U is linearly independentexer:2A:14
for any m 1, which can follow that F1 /U is
infinite-dimensional according to exercise 14 (2A).
Pm Pm
Indeed, consider (
theo:3.85:3E i=1 ↵ i v i ) + U is equal to U when v = i=1 ↵i vi 2 U according to
theorem 5.8.1, which means there exists M 2 N so for all i M , i-th coordinate of v is
0. According to notation of vi , the x-th coordinate of v where
P
• x ⌘ 1 (mod p1 · · · pm ) is m i=1 ↵i .
Q P
• x ⌘ 0 (mod pj ), x ⌘ 1 (mod i6=j pi ) is 2↵j + i6=j ↵i .
Pm P
If we pick x M then from the above, we find that i=1 ↵i = 2↵j + i6=j ↵i = 0
for every 1  j  m. Hence, we find ↵i = 0 for all 1  i  m. In other words,
v1 + U, v2 + U, . . . , vm + U is linearly independent in F1 /U , as desired.
theo:3.91:3E
exer:3E:15 15. According to theorem 5.8.3, V /(null ') is isomorphic to range '. We find that range ' = F
since ' 6= 0 so for any c 2 F, there exists 2 F so '(1) = c. Hence, dim range ' =
dim F = 1, which follows dim V /(null ') = 1.
exer:3E:16 16. Since dim V /U = 1 so there exists u 2 V, u 62 U . Consider theo:3.85:3E

v 6= u, v 62 U, v 2 V then
v + U = u + U for some 2 F, hence according to theorem 5.8.1, v = u + x for some
x 2 U.
Define a linear map ' 2 L(V, F) so '(u) = 1, '(v) = 0 for all v 62 U , '(v) = '(u) if
v + U = u + U . Consider 'v = 0 for some v 2 V . If v 62 U then v = u + x for some
2 F, x 2 U so '(v) 6= 0, a contradiction. Thus, only v 2 U will give 'v = 0, which
means null ' = U .
exer:3E:17 17. Let v1 + U, v2 + U, . . . , vm +P

U be basis of U . For any v 2 V thentheo:3.85:3E
there exists
Pmuniquely
↵1 , . . . , ↵m 2 F so v + U = ( m ↵ v
i=1 i i ) + U . According to theorem 5.8.1, v I=1 ↵i vi =
u 2 U . Hence, every v 2 V can be represented uniquely as u+x where x 2 span (v1 , . . . , vm ).
Thus, V = U span (v1 , . . . , vm ).
exer:3E:18 18. If there exists such linear map S then 0 = T (0) = S(⇡0) = S(U ). For any u 2 U then
T (u) = S(⇡u) = S(U ) = 0 so u 2 null T . Thus, U ⇢ null T .
Conversely, if U ⇢ null T . Define S : V /U ! W so S(v + U ) = T (v). We show that the

definition of S makes sense, suppose u, v 2 V so u + U = v + U then according to theorem
theo:3.85:3E
5.8.1, v u 2 U so v u 2 null T so T v = T u.
exer:3E:19 19. Disjoint union is analogous to direct sum: If number of elements of A1 [ A2 [ . . . [ Am is
equal to sum of number of elements of Ai , then A1 [ . . . [ Am is a disjoint union.
exer:3E:20 20. (a) We have (S1 + S2 ) = (S1 + S2 ) ⇡ = S1 ⇡ + S2 ⇡ = (S1 ) + (S2 ). Similarly,
( S1 ) = ( S1 ) ⇡ = (S1 ). Thus, is a linear map.
(b) If (S1 ) = (S2 ) then S1 ⇡ = S2 ⇡ or S1 (v + U ) = S2 = (v + U ) or (S1 S2 )(v + U )
for any v + U 2 V /U . Hence, S1 = S2 so is injective.
exer:3E:18
(c) For any T 2 {T 2 L(V, W ) : T u = 0 for every u 2 U }, according to exercise 18, there
exists S 2 L(V /U, W ) so (S) = T = S ⇡. Thus, range = {T 2 L(V, W ) : T u =
0 for every u 2 U }.
5.10. 3F: Duality

theo:3.95:3F Theorem 5.10.1 (3.95) Suppose V is finite-dimensional. Then V 0 = L(V, F) is also finite-
dimensional and dim V 0 = dim V .
theo:3.105:3F Theorem 5.10.2 (3.105) Suppose U ⇢ V then U 0 is a subspace of V 0 .
theo:3.106:3F Theorem 5.10.3 (3.106) Suppose V is finite-dimensional and U is subspace of V . Then

dim U + dim U 0 = dim V .
theo:3.107:3F Theorem 5.10.4 (3.107) Suppose V and W are finite-dimensional and T 2 L(V, W ).
Then:
(a) null T 0 = (range T )0 (don’t need V, W being finite-dimensional).
(b) dim null T 0 = dim null T + dim W dim V .
theo:3.108,3.110:3F Theorem 5.10.5 (3.108,3.110) Suppose V, W are finite-dimensional and T 2 L(V, W ).

Then T is injective if and only if T 0 is surjective and T is surjective if and only if T 0
is injective.
theo:3.109:3F Theorem 5.10.6 (3.109) Suppose V and W are finite-dimensional and T 2 L(V, W ).
Then:
(a) dim range T 0 = dim range T ;
(b) range T 0 = null T .

theo:3.117:3F Theorem 5.10.7 (3.117) Suppose V and W are finite-dimensional and T 2 L(V, W ). Then
dim range T equals to column rank of M(T ).
5.11. Exercises 3F
exer:3F:1 1. Let ' 2 L(V, F). If ' is not a zero map then there exists v 2 V so 'v = c 6= 0. Hence, for
any 2 F then '( /c · v) = so ' is surjective.
exer:3F:2 2. Linear functionals on R[0,1] : '1 (f ) = f (0.5), '2 (f ) = f (0.1), '3 (f ) = f (0.2).
exer:3F:3 3. Pick ' 2 V 0 so '(v) = 1, value of '(u) for u 62 {av : a 2 F} can be picked arbitrarily.
exer:3F:4 4. Extend basis v1 , . . . , vm of U to basis v1 , . . . , vn (m < n) of V . Define ' from V onto F so

'(vi ) = 0 for 1  i  m, and '(vj ) = 1 for m < j  n. Hence, '(u) = 0 for every u 2 U .
exer:3E:4
exer:3F:5 5. It is a consequence of exercise 4 (3E) when W = F.
exer:3F:6 6. (a) If v1 , . . . , vm spans V : Consider ' 2 V 0 so (') = 0 then '(vi ) = 0 for all 1  i  m,
which leads to '(v) = 0 for all v 2 V since v1 , . . . , vm spans V . Hence, ' = 0, i.e. is
injective.
Conversely, if is injective. If v1 , . . . , vm does not span V then there exists vm+1 62
span (v1 , . . . , vm ), vm+1 2 V . Hence, there exists ' 2 V 0 so '(vm+1 ) = 1 and '(v) = 0 for
all v 2 V, v 62 {avm+1 : a 2 F}. We find (') = 0 but ' 6= 0, a contradiction to injectivity
of . Thus, v1 , . . . , vm spans V .
(b) If v1 , . . . , vm is linearly independent then for any (x1 , . . . , xm ) 2 Fm , we can construct
' 2 V 0 so '(vi ) = xi . Hence, is surjective.
Pm 1
If is surjective⇣ but v1 , . . . , vm is P
linearly dependent,
⌘ i.e. WLOG vm = i=1 ↵i vi .
m 1
Hence, (') = '(v1 ), . . . , '(vm 1 ), i=1 ↵i 'i (vi ) , which does not cover all Fm . Thus,
v1 , . . . , vm is linearly independent.
(xj )( j)(0)
exer:3F:7 7. Because 'j (xj ) = j! = 1 and 'j (xi ) = 0 for i 6= j.
exer:3F:8 8. (a) True.

p(j) (5)
(b) Dual basis of basis in (a) is '0 , '1 , . . . , 'm where 'j (p) = j! .
exer:3F:9 9. SinceP'1 , . . . , 'n is dual basis

Pnof V so for any
0 2 V 0 , there exists ↵1 , .P
. . , ↵n 2 F so
= i=1 ↵i 'i . Note that ( i=1 ↵i 'i ) (vi ) = ↵i so (vi ) = ↵i . Thus, = ni=1 (vi )'i .
n
exer:3F:10 10. We have (S + T )0 (') = ' (S + T ) = ' S + ' T = S 0 (') + T 0 (') = (S 0 + T 0 )('). Thus,
(S + T )0 = S 0 + T 0 . Similarly, ( T )0 = T 0 .
exer:3F:11 11. If there exists such (c1 , . . . , cm ) 2 Fm and (d1 , . . . , dn ) 2 Fn then A.,k = dk ·C where C is an
m-by-1 matrix so Ci = ci . Hence, rank of A is the dimension of span (A.,1 , A.,2 , . . . , A.,n ),
which is 1.
If rank of A is 1 then dimension of span (A.,1 , A.,2 , . . . , A.,n ) is 1, which means A.,k = dk · C
for some C 2 Fm,1 .
exer:3F:12 12. Let I be the identity map on V then the dual map I 0 (') = ' I = ' so I 0 is the identity
map on V 0 .
exer:3F:13 13. (a) (T 0 ('1 ))(x, y, z) = ('1 T )(x, y, z) = '1 (4x + 5y + 6z, 7x + 8y + 9z) = 4x + 5y + 6z
and (T 0 ('2 ))(x, y, z) = 7x + 8y + 9z.
exer:3F:9
(b) According to exercise 9 (3F) then since T 0 ('1 ) 2 (R3 )0 and 1, 2, 3 is dual basis of
(R3 )0 so
(T 0 ('1 )) = (T 0 ('1 ))(1, 0, 0) 1 + (T 0 ('1 ))(0, 1, 0) 2 + (T 0 ('1 ))(0, 0, 1) 3,

=4 1 +5 2 +6 3.
exer:3F:14 14. (a) We have (T 0 ('))(p) = (' T )(p) = '(T p) = '(x2 p(x)+p00 (x)) = (x2 p(x)+p00 (x))0 (4) =
(2xp(x) + x2 p0 (x) + p000 (x))(4) = 8p(4) + 16p0 (4) + p000 (4).
R1
(b) (T 0 ('))(x3 ) = '(T (x3 )) = '(x5 + 6x) = 0 (x5 + 6x)dx = 16 + 3.
exer:3F:15 15. If T = 0 then T 0 (') = ' T = 0 for any ' 2 W 0 . Hence, T 0 = 0.
Conversely,
Pm if T 0 = 0, let w1 , . . . , wm be basis of W . For any v, we can write T (v) =
i=1 ↵i wi . We can choose ' 2 W so '(wi ) = ↵i . Hence,
m
X m
X
0 = (T 0 ('))(v) = '(T v) = ↵i '(wi ) = ↵i2 .
i=1 i=1
Thus, ↵i = 0 for all 1  i  m. This follows T v = 0 for all v 2 V , i.e. T = 0.

exer:3F:16 16. Let S be the map that takes T 2 L(V, W ) to T 0 2 L(W 0 , V 0 ). S is a linear map since
S(T1 + T2 ) = (T1 + T2 )0 = (T1 )0 + (T2 )0 = S(T1 ) + S(T2 ) and S( T1 ) = ( T1 )0 = (T1 )0 =
S(T1 ).
exer:3F:15
We prove S is injective. Indeed, if S(T ) = T 0 = 0 then according to exercise 15 (3F) then
T = 0. Thus, S is injective.
We prove S is surjective. Let v1 , . . . , vn be basis of V and '1 , . . . , 'n be the dual basis of
V 0 , w1 , . . . , wm beP
basis of W and 1 , . . . , m be dual basis of W 0 . We want to determine
all Ai,j so T vk = m i=1 Ai,k wi . For every 1  i  m, 1  k  n, let Ai,k = (T ( i ))(vk ).
0
Now we will prove that such linear map T will satisfy T 0 ( ) = T for all ' 2 W 0 .
Indeed, we find that
m
!
X
j (T vk ) = j Ai,k wi ,
i=1
m
!
X
= j (T 0 ( i ))(vk )wi ,
i=1
= (T 0 ( j ))(vk ), (because j (wi ) = 0 for i 6= j)
This is true for all 1  k  n so for any v 2 V then j (T v) = (T 0 ( j ))(v) or T 0 ( j ) = j T .

This again is true for every 1  j  m so for any 2 W 0 then T 0 ( ) = T , which means
S(T ) = T 0 . Thus, S is surjective.
In conclusion, S is an isomorphism of L(V, W ) onto L(W 0 , V 0 ).
exer:3F:17 17. For any ' 2 U 0 then '(u) = 0 for any u 2 U , i.e. u 2 null ' for any u 2 U , i.e.
U ⇢ null '. Thus U 0 ⇢ {' 2 V 0 : U ⇢ null '}. Similarly, we can finally show that
U 0 = {' 2 V 0 : U ⇢ null '}.
exer:3F:18 18. U = {0} iff dim U = 0 iff dim U 0 = dim V dim U = dim V = dim V 0 iff U 0 = V 0 since
U 0 is subspace of V 0 .
exer:3F:19 19. U = V iff dim U = dim V iff dim U 0 = 0 iff U 0 = {0}.
exer:3F:20 20. If U ⇢ W then if ' 2 W 0 we will have '(w) = 0 for any w 2 W , which means '(u) = 0
for any u 2 U ⇢ W . Thus, ' 2 U 0 . Hence, W 0 ⇢ U 0 .
exer:3F:21 21. Let w1 , . . . , wn be basis of W . If U is not subspace of W then there exists v 2 V so v 62
W, v 2 U . Hence, basis w1 , . . . , wn of W can be extended to basis w1 , . . . , wn , v, v1 , . . . , vm
of V . Let ' 2 V 0 so '(wi ) = 0 and '(v) = 1. This follows ' 2 W 0 but ' 62 U 0 , a
contradiction since W 0 ⇢ U 0 .
exer:3F:20
exer:3F:22 22. Since U ⇢ U + W so according to exercise 20, (U + W )0 ⇢ U 0 . Similarly, (U + W )0 ⇢ W 0 .
Thus (U + W )0 ⇢ U 0 \ W 0 .
On the other hand, if ' 2 U 0 \ W 0 then '(u) = 0 for any u 2 U or u 2 W . Thus,
'(u + w) = 0 for any u 2 U, w 2 U so ' 2 (U + W )0 . We follow U 0 \ W 0 ⇢ (U + W )0 .
In conclusion, U 0 \ W 0 = (U + W )0 .
exer:3F:23 23. For any ' 2 U 0 , 2 W 0 then ' + 2 U 0 + W 0 and (' + )(u) = 0 for any u 2 U \ W .
Thus, U 0 + W 0 ⇢ (U \ W )0 .
Since U, W are subspaces of V so U \W is also subspace of V . Let v1 , . . . , vn be basis of U \
W , which can be extended to basis v1 , . . . , vn , u1 , . . . , um of U . Basis v1 , . . . , vn of U \W can
also be extended to basis v1 , . . . , vn , w1 , . . . , wk of W . Note that v1 , . . . , vn , u1 , . . . , um , w1 , . . . , wk
is linearly independent so this list can be extended further to basis v1 , . . . , vn , u1 , . . . , um , w1 , . . . , wk , z1 , . . . ,
of V . Hence, for any T 2 (U \ W )0 , then there exists ' 2 U 0 , 2 V 0 P so '(wi ) =
T wi , '(zi ) = 1 and (ui ) = T ui , '(zi ) = T zi 1. Hence, for any v 2 V, v = ni=1 ↵i vi +
Pm Pk Ph
i=1 i ui + i=1 i wi + i=1 ai zi then
m
X k
X h
X
(' + )(v) = i (ui ) + i '(wi ) + ai (' + )(zi ),
i=1 i=1 i=1
Xm k
X h
X
= i T ui + i T wi + ai T zi ,
i=1 i=1 i=1
= T v.
Thus, T = ' + . This follows (U \ W )0 ⇢ U 0 + W 0 . Thus, (U \ W )0 = U 0 + W 0 .
exer:3F:24 24. Done.
exer:3F:25 25. Let S = {v 2 V : '(v) = 0 for every ' 2 U 0 }. If v 2 U then obviously v 2 S. Thus,
U ⇢ S. If v 2 S but v 62 U , then there exists basis u1 , . . . , un , v, v1 , . . . , vm of V where
u1 , . . . , un is basis of U . Hence, there exists ' 2 U 0 so 'ui = 0 and 'v = 0. Thus, v 62 S,
a contradiction. We find that if v 2 S then v 2 U . In conclusion U = S = {v 2 V : '(v) =
0 for every ' 2 U 0 }.
exer:3F:26 26. Let S = {v 2 V : '(v) = 0 for every ' 2 }. If ' 2 then '(v) = 0 for every v 2 S so
' 2 S 0 . Thus, ⇢ S 0 .
Let v1 , . . . , vn be basis of S, which can be extended to basis v1 , . . . , vn , u1 , . . . , um of V .
Let ' , . . . , m be dual basis of V 0 . Hence, for any ' 2
P1 ,n. . . , 'n , 1P ⇢ V 0 , we have
m
' = i=1 ↵i 'i + j=1 j j . Note that 1 , . . . , m is basis of S so since vk 2 S, we have
0
j (vk ) = 0 for any 1  j  m. Thus,
n
X m
X
'(vk ) = ↵i 'i (vk ) + j j (vk ),
i=1 j=1
= ↵k .
On the other hand, since ⇢ S 0 and ' 2 , vk 2 S so '(vk ) = 0. Thus, ↵k = 0 for every
1  k  n. This follows 1 , . . . , m spans . Note that 1 , . . . , m is linearly independent.
Thus, 1 , . . . , m is basis of , which leads to = span ( 1 , . . . , m ) = S 0 .
theo:3.107:3F exer:3F:25
exer:3F:27 27. According to theorem 5.10.4 then (range T )0 = { ' : 2 R}. According to exercise 25
(3F) then
range T = {p 2 P(R) : ( ')(p) = 0 for any 2 R},

= {p 2 P(R) : p(8) = 0 for any 2 R},
= {p 2 P(R) : p(8) = 0}.
exer:3F:28 28. According to theorem 5.10.4 then (range T )0 = { ' : 2 F}. According to exercise 25
(3F) then
range T = {v 2 V : ( ')(v) = 0 for any 2 F},

= {v 2 V : '(v) = 0},
= null '.
exer:3F:29 29. According to theorem 5.10.6 then (null T )0 = { ' : 2 F}. According to exercise 25 (3F)
then
null T = {v 2 V : ( ')(v) = 0 for any 2 F},

= null '.
exer:3F:30 30. Let U = (null '1 ) \ (null '2 ) \ · · · \ (null 'm ) then U is a subspace of V . Hence, U 0 =
{' 2 V 0 : '(u) = 0 for any u 2 U }. Thus, we find that 'i 2 U 0 for every 1  i  m. Since
'1 , . . . , 'm is linearly independent so dim U 0 m. Thus, dim U = dim V dim U 0 
dim V m.
On the other hand, since V is finite-dimensional so let dim V = n m. If we let Ai be set
that contains a basis of null 'i . Then we will notice that dim U |A1 \ A2 \ · · · \ Am |.
It suffices to prove |A1 \ A2 \ · · · \ Am | n m.
Observe that since 'i in the linearly dependent list so range 'i = 1. Thus, |Ai | = null 'i =
dim V range 'i = n 1. We prove by induction on m that |A1 \ A2 \ · · · \ Am | n m.
Indeed, it’s true for m = 1. If it’s true for m 1, consider for m then we have:
|A1 \ · · · \ Am | = |A1 \ · · · \ Am 1| + |Am | |(A1 \ · · · \ Am 1) [ Am |,

n (m 1) + (n 1) n=n m.
Thus, we obtain dim U |A1 \ · · · \ Am | n m. Hence, we find dim U = dim V m.

exer:3F:30
exer:3F:31 31. According to exercise 30 (3F) then dim ((null '1 ) \ (null '2 ) \ · · · \ (null 'n )) = 0 and
dim ((null '2 ) \ · · · \ (null 'n )) = 1 so there exists v1 2 V so '1 (v1 ) = 1 but 'i (v1 ) = 0
for 2  i  n. Similarly, there exists v2 , v3 , . . . , vn .
P
Note that vk 62 span (v1 , . . . , vk 1 ) because otherwise 1 = 'k (vk ) = ki=11 ↵i 'i (vk ) = 0,
a contradiction. This follows v1 , . . . , vn is linearly independent so v1 , . . . , vn is basis of V
whose dual basis is '1 , . . . , 'n .
exer:3F:32 32. (a) T is surjective since range T = span (v1 , . . . , vn ) = V . Thus, operator T is invertible.
(b) Since T is invertible so T v1 , . . . , T vn is linearly independent, which follows M(T u1 ), . . . , M(T un )
is linearly independent, i.e. columns of M(T ) is linearly independent.
(c) Since dim span (M(T u1 ), . . . , M(T un )) = n = dim Fn,1 so column of M(T ) spans
Fn,1 .
(d) Let '1 , . . . , 'n be dual basis of u1 , . . . , un and 1 , . . . , n be dual basis of v1 , . . . , vn .
T 0 2 L(V 0 ) is a dual map of T with respect to these to dual bases. Since (M(T 0 ))t = M(T )
so i-th row of M(T ) is i-th column of M(T 0 ). Since T is invertible so T 0 is also invertible.
Hence, T 0 ('1 ), . . . , T 0 ('n ) is linearly independent. This follows M(T 0 '1 ), M(T 0 '2 ), . . . , M(T 0 'n )
is linearly independent or columns of M(T 0 ) are linearly independent or rows of M(T ) are
linearly idependent in Fn,1 .
(e) Row rank of M(T ) is n so rows of M(T ) spans Fn,1 .
exer:3F:33 33. Let S be a function that takes A 2 Fm,n to At 2 Fn,m then S(A + B) = (A + B)t =
At + B t = S(A) + S(B) and S( A) = ( A)t = At = S(A). Thus, S is a linear map
from Fm,n onto Fn,m .
S can be easily seen to be surjective, as for any A 2 Fn,m then there exists B = At 2 Fm,n
so S(B) = S(At ) = (At )t = A.
If S(A) = 0 then A = 0. Thus, S is injective so S is invertible.
exer:3F:34 34. (a) For any ' 2 V 0 then (⇤(v1 +v2 ))(') = '(v1 +v2 ) = '(v1 )+'(v2 ) = (⇤v1 )(')+(⇤v2 )(').
Thus, ⇤(v1 + v2 ) = ⇤v2 + ⇤v1 . Similarly, ⇤( v) = (⇤v). Thus, ⇤ is a linear map from V
to V 00 .
(b) For any v 2 V , we will prove that (⇤ T )(v) = (T 00 ⇤)(v) or ⇤(T v) = T 00 (⇤v). In
order to prove this, we show that for any ' 2 V 0 then (⇤(T v))(') = (T 00 (⇤v))('). Indeed,
we have (⇤(T v))(') = '(T v) and since T 00 = (T 0 )0 2 L(V 00 , V 00 ) so
(T 00 (⇤v))(') = ((⇤v) T 0 )(') = (⇤v)(T 0 (')) = (T 0 ('))(v) = (' T )(v) = '(T v).
Thus, (T 00 (⇤v))(') = (⇤(T v))(') for any ' 2 V 0 , which follows ⇤(T v) = T 00 (⇤v) for any
v 2 V , which leads to ⇤ T = T 00 ⇤.
(c) If ⇤v = 0 then (⇤v)(') = '(v) = 0 for any ' 2 V 0 , which leads to v = 0. Thus, ⇤ is
injective.
Let v1 , . . . , vn be basis of V and
Pn'1 , . . . , 'n be dual basis of V . For any S 2 V , we
0 00
show that ⇤v = S where v = i=1 S('i )vi . It suffices to show that. Indeed, we have
'i (v) = S('i ) for every 1  i  n. This follows '(v) = S(') for any ' 2 V 0 . Thus,
⇤v = S. Thus, ⇤ is surjective.
In conclusion, ⇤ is an isomorphism from V onto V 00 .
exer:3F:35 35. Denote 'i 2 (P(R))0 P so 'i (xi ) = 1 and 'i (xj ) = 0 for i 6= j. This follows for any
0
' 2 (P(R)) then ' = i 0 '(xi )'i .
P
Let ei 2 R1 where ei = (0, . . . , 0, 1, 0, . . .). Then any e 2 R1 we have e = i 0 ↵i ei .
| {z }
i 1
Define a linearPmap T from (P(R))0 ontoPF1 so T ('i ) = ei . T is surjective. If T (') = 0

for some ' = i 0 '(xi )'i then T (') = i 0 '(xi )ei = 0. Thus, '(xi ) = 0 for any i 0.
This follows ' = 0. Thus, T is injective. We find T is an ismorphism from (P(R))0 onto
F1 .
theo:3.107:3F
exer:3F:36 36. (a) Since range i = U so according to theorem 5.10.4 then null i0 = U 0 .
theo:3.109:3F
(b) Since null T = {0} so (null T )0 = U 0 . According to theorem 5.10.6 then range i0 =
(null T )0 = U 0 .
theo:3.91:3E
(c) Since null i0 = U 0 so according to theorem 5.8.3 then i˜0 is injective and range i˜0 =
range i0 = U 0 so i˜0 is surjective. Thus, i˜0 is an isomorphism from V 0 /U 0 = V 0 /(null i)0
onto U 0 .
theo:3.108,3.110:3F
exer:3F:37 37. (a) According to theorem 5.10.5, ⇡ 0 is injective iff ⇡ is surjective, which is true.
theo:3.109:3F
(b) According to theorem 5.10.6 then range ⇡ 0 = (null ⇡)0 = U 0 .
(c) From (a) and (b) then ⇡ 0 is an ismorphism from (V /U )0 onto U 0 .
6. Chapter 4: Polynomials
theo:4.17:4 Theorem 6.0.1 (4.17) Suppose p 2 P(R) is nonconstant polynomial. Then p has unique
factorization (except for the order of factors) of the form
2
p(x) = c(x 1 ) · · · (x m )(x + b1 x + c1 ) · · · (x2 + bM x + cM )
where c, 1, . . . , m , b1 , . . . , bM , c1 , . . . , cM 2 R with b2j < 4cj for each j.
6.1. Exercise 4
exer:4:1 1. Done.
exer:4:2 2. No since ( xm + x) + xm = x does not have degree of m.
exer:4:3 3. Yes.
Pm
exer:4:4 4. Pick p = (x 1)
a1 (x
2)
a2 · · · (x m)
am so ai 2 Z, ai 1 and i=1 ai = n.
exer:4:5 5. Let A be a (m + 1)-by-(m + 1) matrix so Ai,j = zij 1 . Define a linear map T 2 L(Rm+1,1 )
so T x = Ax.0 First,
1 we prove that T is invertible. Indeed, if there exists x 6= 0, x 2
a0
B a1 C
B C
Rm+1,1 , x = B . C so T x = Ax = 0 then the polynomial a0 + a1 x + a2 x2 + . . . + am xm
@ .. A
am
has m+1 distinct zeros z1 , . . . , zm , zm+1 , a contradiction. Thus, we must have x = 0 which
follows operator T is injective, which means T is invertible. Hence for any W 2 Rm+1,1 ,
Wi,1 = wi then there exists unique x 2 Rm+1,1 , x1,i = ↵i 1 so T x = Ax = W . Note
that Ai,. x = (Ax)i,1 = wi so polynomial p = ↵0 + ↵1 x + ↵2 x2 + . . . + ↵m xm satisfies the
condition. Since the x 2 Rm+1,1 is uniquely determined so p is uniquely determined.
exer:4:6 6. If p(x) has a zero then there exists q 2 P(C) so deg q = m 1 and p(x) = (x )q(x).
Hence, p0 (x) = q(x) + (x )q 0 (x).
If p(x), p0 (x) shares a same zero, say then p0 ( ) = 0 which follows q( ) = 0 so q(x) =
(x )r(x) so p(x) = (x )2 r(x) with r 2 P(C) and deg r = m 2. This follows p has
at most m 1 distinct zeros.
theo:4.17:4
exer:4:7 7. According to theorem 6.0.1 then p(x) has a unique factorization
2
p(x) = c(x 1 ) · · · (x m )(x + b1 x + c1 ) · · · (x2 + bM x + cM ).
With p(x) having no real zero then each p(x) has a unique factorization
p(x) = c(x2 + b1 x + c1 ) · · · (x2 + bM x + cM )
This follows degree of p is always even, a contradiction. Thus, polynomial with odd degree
must have a real zero.
exer:4:8 8. There exists q(x) 2 P(R) so p(x) = (x 3)q(x) + r for some r 2 R. We can see right
away that r = p(3). We prove that T p = q 2 P(R). Indeed, q 2 RR and for x 6= 3 then
q(x) = p(x) r
x 3 =
p(x) p(3)
x 3 . For x = 3 then observe that p0 (x) = q(x) + (x 3)q 0 (x) so
p (3) = q(3). Thus, essentially, T p = q is a polynomial in P(R).
0
We have (T (p + q))(3) = (p + q)0 (3) = p0 (3) + q 0 (3) = (T (p))(3) + (T (q))(3) and for x 6= 3
then (T (p + q))(x) = (p+q)(x)x (p+q)(3)
3 = p(x)x p(3)
3 + q(x)x q(3)
3 = (T (p))(x) + (T (q))(x).
Thus, T (p + q) = T (p) + T (q). Similarly, T ( p) = T (p). We find T is a linear map.
exer:4:9 9. If p 2 P(C) then p has unique factorization of the form p(z) = c(z 1 ) · · · (z m) where
c, 1 , . . . , m 2 C. Hence
q(z) = p(z)p(z),
m
Y m
Y
=c (z i )c (z i ),
i=1 i=1
Ym
⇥ ⇤
= cc (z i )(z i) ,
i=1
m
Y
2
= |c| z2 2(Re i )z + | i |2 .
i=1
Thus q(z) is a polynomial with real coefficients.

exer:4:5
exer:4:10 10. Let q 2 P(R) so q(xi ) = p(xi ) 2 R then according to exercise 5, q is uniquely determined
in P(R).
Let xq = M(q) with respect toexer:4:5 standard basis in P(C). Repeat the construction of
invertible operator T in exercise 5 (4) but this time T 2 L(Cm+1,1 ). Then T xp = T xq =
(p(x0 ), p(x1 ), . . . , p(xm ))t . Since T is injective so xp = xq , which follows p = q. Thus, p is
a polynomial with real coefficients.
exer:4:11 11. If deg q deg p then there exists r, s 2 P(F) so q = pr + s with deg s < deg p. We find
q + U = pr + s + U = s + U . And each such s 2 P(F), deg s < deg p can be represented as
linear combination of 1, x, . . . , xdeg(p) 1 . Thus, we follow that 1+U, x+U, . . . , xdeg(p) 1 +U
spans P(F)/U . And this list is linearly independent so this list is basis of P(F)/U . Hence,
dim P(F)/U = deg p.
7. Chapter 5: Eigenvalues, Eigenvectors, and Invariant Subspaces

7.1. 5A: Invariant subspaces
example:eigenvalue Example 7.1.1 Suppose T 2 L(F2 ) defined by T (w, z) = ( z, w) then T does not have
eigenvalue if F = R. However, T has eigenvalues ±i if F = C.
theo:5.6:5A Theorem 7.1.2 (5.6) Suppose V is finite-dimensional, T 2 L(V ) and 2 F. Then the
following are equivalent:
(a) is an eigenvalue of T .
(b) T I is not injective.
(c) T I is not surjective.
(d) T I is not invertible.

theo:3.69:3D
This is deduced from theorem 5.6.4.
theo:5.10:5A Theorem 7.1.3 (5.10) Let T 2 L(V ). Suppose 1 , . . . , m are distinct eigenvalues of T
and v1 , . . . , vm are corresponding eigenvectors. Then v1 , . . . , vm is linearly independent.
theo:5.13:5A Theorem 7.1.4 (5.13) Suppose V is finite-dimensional. Then each operator on V has at
most dim V distinct eigenvalues.
7.2. Exercises 5A
exer:5A:1 1. (a) For any u 2 U then T u = 0 2 U so U is invariant under T .
(b) For any u 2 U then T u 2 range T ⇢ U so U is invariant under T .
exer:5A:2 2. For every u 2 null S then ST u = T (Su) = T (0) = 0 so T u 2 null S. Thus, null S is
invariant under T .
exer:5A:3 3. For every u 2 range S then there exists v 2 V so Sv = u. Hence, T u = T (Sv) = S(T v) 2
range S so range S is invariant under T .
exer:5A:4 4. For any u1 2 U1 , u2 2 U2 , . . . , um 2 Um then T (u1 + . . . + um ) = T u1 + . . . + T um 2
U1 + . . . + Um . Thus, U1 + . . . + Um is invariant under T .
exer:5A:5 5. Let U1 , . . . , Um be invariant subspaces of V under T . Then for any v 2 S = U1 \ . . . \ Um
then T v 2 V so S is invariant under T .
exer:5A:6 6. If U 6= {0} then there exists basis u1 , . . . , um of U . If there exists v 2 V, v 62 U then we
can choose operator T 2 L(V ) so T u1 = v, which follows U is not invariant under T , a
contradiction. Thus, if v 2 V then v 2 U , i.e. U = V .
exer:5A:7 7. If T (x, y) = ( 3y, x) = (x, y) for (x, y) 6= (0, 0) then 3y = x, x = y which follows
3y = 2 y. If y = 0 then x = 0, a contradiction. Hence, y 6= 0 which means 3 = 2 .
Thus, T does not have any eigenvalue.
exer:5A:8 8. If T (w, z) = (z, w) = (w, z) for (w, z) 6= (0, 0) then z = w, w = z so z = 2 . Similarly,

since z 6= 0 so 2 = 1 which follows = ±1. If = 1 then eigenvector is (z, z), if = 1
then eigenvector is (z, z).
exer:5A:9 9. If T (z1 , z2 , z3 ) = (z1 , z2 , z3 ) = (2z1 , 0, 5z3 ) for (z1 , z2 , z3 ) 6= 0 then z1 = 2z2 , z2 =

0, z3 = 5z3 . If = 0 then z2 = z3 = 0 so eigenvector is (z, 0, 0). If ne0 then z1 = z2 = 0
and z3 = 5z3 so = 5. Respective eigenvector is (0, 0, z).
exer:5A:10 10. (a) If T (x1 , . . . , xn ) = (x1 , . . . , xn ) = (x1 , 2x2 , 3x3 , . . . nxn ) for (x1 , . . . , xn ) 6= (0, . . . , 0)
then ( i)xi = 0 for i = 1, . . . , n. This follows 2 {1, . . . , n}. If = i then eigenvector
is ↵ · ei where ei 2 Fn so ei ’s coordinates are 0 except the i-th coordinate is 1.
(b) Let U be an invariant subspace of V under T then for any u = (x1 , . . . , xn ) 2 U then
v = (x1 , 2x2 , . . . , nxn ) 2 U . With only these two vectors in U , we can choose certain
↵, 2 F to 1-st coordinate of w = ↵u + v is 0. We also have T w 2 U and from w, T w,
we can obtain a new vector in U so its 1st and 2nd coordinates are 0. Keep going like
that, we can eventually obtain that ei = (0, . . . , 0, 1, 0, . . . 0) 2 U if xi 6= 0. This follows
| {z }
i 1
subspace U is spanned by list of ei so xi 6= 0 for some (x1 , . . . , xn ) 2 U . Thus, any subset
of the standard basis of Fn is a spanning list of an invariant subspace of V .
exer:5A:11 11. If T p = p = p0 for p 6= 0. If = 0 then p0 = 0 so respective eigenvector p is a constant. If

6= 0 then since p 6= p0 for all p 2 P(R) so doesn’t exists eigenvector in this case. Thus,
the only answer is = 0 as eigenvalue and p = c 2 R as eigenvector.
exer:5A:12 12. If (T p)x = p(x) = xp0 (x) for some p(x) 6= 0. Let p = a0 +a1 x+. . . , a4 x4 then p0 (x) = a1 +
2a2 x+3a3 x2 +4a4 x3 . Thus, this follows ai = iai for i = 0, . . . , 4. Hence, 2 {0, 1, 2, 3, 4}
so that (a0 , . . . , a4 ) 6= (0, . . . , 0). If = i then p = axi is can eigenvector for any a 2 R.
theo:5.13:5A
exer:5A:13 13. Since V is finite-dimensional so according to theorem 7.1.4, there are at most dim V
different eigenvalues oftheo:5.6:5A
T , in other words, at most dim V ↵ 2 F so T ↵I is not invertible
according to theorem 7.1.2. Thus, we can clearly choose ↵ so |↵ 1
| < 1000 and T ↵I
is invertible.
exer:5A:14 14. If P (u + w) = (u + w) = u for u + w 6= 0, u 2 U, w 2 W then (1 )u = w. If = 6 0, 1

then w 2 U , a contradiction since V = U W . If = 0 then u = 0 so eigenvector is any
w 2 W . If = 1 then w = 0 so eigenvector is any u 2 W .
exer:5A:15 15. (a) If is an eigenvalue of T then there exists v 2 V, v 6= 0 so T v = v. Since S is

invertible so there exists u 2 V so Su = v which follows S 1 v = u. We have S 1 T Su =
S 1 T (Su) = S 1 T v = S 1 ( v) = u. Thus, is also eigenvalue of S 1 T S.
If is an eigenvalue of S 1 T S then there exists v 2 V, v 6= 0 so S 1 T Sv = v. or

S (T Sv) = S 1 (Sv). Since S 1 is invertible so T (Sv) = Sv. Thus, is also eigenvalue
1
of T .
(b) If v is eigenvalue of T then S 1v is eigenvector of S 1T S corresponding to .
Pn
exer:5A:16 16. Let v1 , . . . , vn be basis of V and for 1  k  n, let T vk = i=1 ↵i,k vPi so ↵i,j 2 R according
m
to problem’s condition. Pn If is eigenvalue of T with eigenvector v = i=1 i vi with i 2 C.
We show that v = i=1 i vi is the eigenvector corresponding to eigenvalue of T , i.e.
T v = · v.
P P P
Indeed, from T v = v, where T v = nk=1 k T vk = ni=1 k ni=1 ↵i,k vi , we follow for any
P P P
i = 1, . . . , n then nk=1 k ↵i,k = i so nk=1 k ↵i,k = i or nk=1 k ↵i,k = · i . Thus,
n n
! n
X X X
Tv = ↵
k i,k v i = ( · i vi ) = · v.
i=1 k=1 i=1
Thus, is also an eigenvalue of T .
exer:5A:17 17. T (x1 , x2 , x3 , x4 ) = (x2 , x3 , x4 , x1 ).
exer:5A:18 18. If T (z1 , z2 , . . .) = (0, z1 , . . .) = (z1 , z2 , . . .) for some (z1 , z2 , . . .) 6= (0, 0, . . .) then z1 = 0
and zi+1 = zi for all i 1. In both two cases = 0 and 6= 0, we all obtain zi = 0 for
all i 1, a contradiction. Thus, T does not have eigenvalue.
exer:5A:19 19. If T (x1 , . . . , xn ) = (x1 , . . . , xn ) = (x1 + . . . + xn , . . . , x1 + . . . + xn ) for some (x1 , . . . , xn ) 6=

(0, . . . , 0) then xi = x1 + . . . + xn for all 1  i  n. Adding all n equations we find
(x1 + . . . + xn ) = n(x1 + . . . + xn ). If 6= n then x1 + . . . + xn = 0, which follows
xi = 0 for all 1  i  n, a contradiction. Thus, = n, which leads to xi = xj for any
1  i < j  n. This follows respective eigenvector is ↵ · (1, 1, . . . , 1) for any ↵ 2 F.
exer:5A:20 20. If T (z1 , z2 , . . .) = (z1 , z2 , . . .) = (z2 , z3 , . . .) for some (z1 , z2 , . . .) 6= (0, 0, . . .) then zi =
zi+1 for all i 1. For any eigenvalue then corresponding eigenvector is ↵(1, , 2 , . . .)
for any ↵ 2 F.
exer:5A:21 21. If is eigenvalue of T then there exists v 2 V, v 6= 0 so T v = v. Hence, T 1 (v) =

T 1 1 T v = 1 v or 1 is eigenvalue of T 1 . Similarly if 1 is eigenvalue of T 1 then is
eigenvalue of T . We also find that T and T 1 have the same eigenvectors.
exer:5A:22 22. If w = v then 3 is eigenvalue of T . If w + v 6= 0 then T (v + w) = 3(v + w) so 3 is

eigenvalue of T .
exer:5A:23 23. If is eigenvalue of ST with ST v = v for v 2 V, v = 6 0. Then T S(T v) = T (ST v) =

T ( v) = T v. Since T v 6= 0 so is also eigenvalue of T S. Similarly, if is eigenvalue of
T S then is also eigenvalue of ST .
0 1
1
B .. C
exer:5A:24 24. (a) Let x = @ . A then T x = Ax = x so 1 is eigenvalue of T .
1
theo:3.117:3F
(b) Note that matrix A is precisely M(T ) and according to theorem 5.10.7, dim range (T
I) is equal to column rank of A I. Since sum of all entries in each column of A is 1 so
sum of the entries in each column of A I is 0. Let ei 2 Fn,1 be the i-th row of A I then
en = e1 + . . . + en 1 so row rank of A I is at most n 1, which means column rank of
A I is at most n 1 or dim range (T I)  n 1 so T I is not injective. According
theo:5.6:5A
to theorem 7.1.2, we follow 1 is an eigenvalue of T .
exer:5A:25 25. Let T u = 1 u, T v = 2 v and T (u + v) = 3 (u + v) then 1 u + 2 v = 3 (u + v) or

( 1 3 )u = ( 3 2 )v. If 1 = 3 then 2 = 3 = 1 , as desired. If 1 6= 3 then
3 6
= 2 which follows u = ↵v for ↵ 2 F, ↵ =
6 0. Hence, T u = ↵T v or 1 u = ↵ 2 v or
↵v( 1 2 ) = 0. Thus, 1 = 2 in all cases.
exer:5A:25
exer:5A:26 26. According to exercise 25 (5A) we can follows all nonzero vectors in V corresponds to same
eigenvalue , i.e. T v = v for all v 2 V . Thus, T = I.
P
exer:5A:27 27. Let v1 , . . . , vn be basis of V and T v1 = ni=1 ↵i vi . Consider n 1-dimensional invariant
subspace Ui = span(v1 , . . . , vi 1 , vi+1 , . . . , vn ) (i 6= 1) then since v1 2 Ui we follow T v1 2 Ui
which leads to ↵i = 0. This is true for all i 6= 1 so eventually we obtain T v1 = ↵1 vi .
Similarly, we can find T vk = ↵k vk for k = 1, . . . , n.
Now consider n 1-dimensional invariant subspace S1 = spanP(v1 v2 , v1 v3 , . . . , v1 vn )
n
then since v1 v2 2 S1 so ↵1 v1 ↵2 v2 = T (v1 v2 ) = i=2 i (v1 vi ). Note that
v1 , . . . , vn is linearly independent so we must have i = 0exer:5A:25
for all i 3. We obtain
↵1 v1 ↵2 v2 = 2 (v1 v2 ), which brings us back to exercise 25 (5A) and obtain ↵1 = ↵2 .
Similarly, we can obtain ↵i = ↵j = for all 1  i < j  n, which leads to T v = v for all
v 2 V . Thus, T = I.
exer:5A:27
exer:5A:28 28. Similarly to exercise 27 (5A), let v1 , . . . , vn (n 3) be basis of V then from v1 2
span (v1 , v2 ) and v1 2 span (v1 , v3 ), we deduce T v1 = ↵1 v1 . Similarly, T vi = ↵i vi for
all i = 1, . . . , n.
Consider v1 v2 in 2-dimensional invariant subspace span (v1 v2 , v1 v3 ) so ↵1 vi ↵2 v2 =
T (v1 v2 ) = 1 (v1 v2 ) + 2 (v1 v3 ). Since v1 , v2 , v3 is exer:5A:25
linearly independent so 2 = 0 or
↵1 v1 ↵2 v2 = 1 (v1 v2 ), which according to exercise 25 (5A) then ↵1 = ↵2 . Similarly,
we can show that ↵i = ↵j = ↵ for all 1  i < j  n. This follows T v = ↵v for every v 2 V
or T = ↵I.
exer:5A:29 29. Let 1 , 2 , . . . , m be all distinct eigenvalues of T and v1 , . . . , vm be corresponding eigen-

theo:5.10:5A
vectors of T then according to theorem 7.1.3, v1 , . . . , vm is linearly independent. If i 6= 0
for all i = 1, . . . , m then T vi = i vi 2 range T so vi 2 range T for all i = 1, . . . , m. This
follows m  dim range T = k. If there exists 1 = 0 then i 6= 0 for all i 2, which
leads to vi 2 range T for all i 2 and we obtain m 1  k in this case. In conclusion,
m  k + 1.
p
exer:5A:30 30. Let x1 , x2 , x3 be eigenvectors corresponding to 4, 5, 7 then x1 , x2 , x3 is linearly indepen-
dent so x1p , x2 , x3 is basis of R3 . Say x =p↵1 x1 + ↵2 x2 + ↵3 x3 then T x 9x = 13↵1 x1
4↵2 x2 + ( 7 9)↵3 x3 . We write ( 4,p5, 7) as linear combination of x1 , x2 , x3 then from p
that to find ↵i so T x 9x = ( 4, 5, 7). Thus, such x 2 R3 so T x 9x = ( 4, 5, 7)
exists.
exer:5A:31 31. If v1 , . . . , vm is linearly independent, since V is finite-dimensional, we can construct T 2

L(V ) so T vi = i vi for i = 1, . . . , m. Hence, T has eigenvectors v1 , . . . , vm corresponding
to 1 , . . . , m .
exer:5A:32 32. The hint says it all.
exer:5A:33 33. Because T (v + range T ) = T v + range T = 0 + range T .
exer:5A:34 34. If null T is injective but (null T ) \ (range T ) 6= {0}, i.e. there exists v 62 null T but
T v 2 null T , i.e. T v 2 (null T ) \ (range T ). Hence, (T /null T )(v + null T ) = null T .
Since v 62 null T so we find T /null T is not injective, a contradiction. Thus, we must have
(null T ) \ (range T ) = {0}.
Conversely, if (null T ) \ (range T ) = {0}. If T /(null T ) is not injective then there exists
v 62 null T so T v 2 null T . This follows T v 6= 0 and T v 2 (null T ) \ (range T ), a
contradiction. Thus, T /null T is injective.
exer:5A:35 35. If is eigenvalue of T /U then there exists v 2 V, v 62 U so (T /U )(v+U ) = T v+U = v+U .

This follows T v v 2 U.
Consider (T I)|U 2 L(U ), if (T I)|U is not injective then there exists u 2 U, u 6= 0
so (T I)u = 0 or T u = u so is eigenvalue of T . If (T
theo:3.69:3D
I)|U is injective then
(T I)|U is invertible according to theorem 5.6.4 (3D) as V is finite-dimensional. Hence
for T v v = u 2 U there exists w 2 U so (T I)w = u = T v v which follows
T (v w) = (v w). Since v w 6= 0 as v 2 / U so we also find as eigenvalue of T .
In conclusion, each eigenvalue of T /U is an eigenvalue of T .
exer:5A:36 36. Pick V = P(R) and 2 R, 6= 0. Let T 2 L(V ) so T (1) = 0, T (x) = x3 + x2 , T x2 =

x4 + x and T xk = x7+k for all k 3. Pick subspace of V as U = span (x3 , x4 , . . .).
Hence, with v = x + x2 we have T v v = x3 + x4 2 U so T v + U = v + U , i.e. is an
eigenvalue of T /U .
However, T v 6= v for all v 2 V . Indeed, if deg v 3 then deg T v 7 + deg v, which leads
to T v 6= v. If deg v  2, let v = a0 + a1 x + a2 x2 then T v = a1 (x3 + x2 ) + a2 (x4 + x)
and v = (a0 + a1 x + a2 x2 ). In this case, T v = v when a1 = a2 = 0 or T v = 0 = a0
and v = a0 6= 0, a contradiction.
Thus, in this example, is eigenvalue of T /U but not of T .
7.3. 5B: Eigenvectors and Upper-Triangular Matrices

theo:5.21:5B Theorem 7.3.1 (5.21) Every operator on a finite-dimensional, nonzero, complex vector
space has an eigenvalue.
theo:5.27:5B Theorem 7.3.2 (5.27) Suppose V is a finite-dimensional complex vector space and T 2
L(V ). Then T has an upper-triangular matrix with respect to some basis of V .
theo:5.30:5B Theorem 7.3.3 (5.30) Suppose T 2 L(V ) has an upper-triangular matrix with respect to
some basis of V . Then T is invertible if and only if all the entries on the diagonal of that
upper-triangular matrix are nonzero.
In general, T I is invertible iff all entries on the diagonal of that upper-triangular matrix

are different from .
theo:5.32:5B Theorem 7.3.4 (5.32) Suppose T 2 L(V ) has an upper-triangular matrix with respect to
some basis of V . Then the eigenvalues of T are precisely the entries on the diagonal of that
upper-triangular matrix.
Following theorems are not from the book. I found these from a blog post of tastymath75025
(AoPS).
theo:1:5B Theorem 7.3.5 Every operator T on a finite-dimensional real vector space V has invariant
subspace U of dimension 1 or 2.
theo:2:5B Theorem 7.3.6 Every operator T on a odd-dimensional real vector space has an eigenvalue.
From the above we follow

Proposition 7.3.7
If V is a real vector space with even dimension, then for any operator T on V , every
invariant subspace of V under T has even dimension.
7.4. Exercises 5B
exer:5B:1 1. (a) We have (I T )(I +T +T 2 +. . .+T n 1 ) = I T n = I and (I +T +. . .+T n 1 )(I T) = I
so I T is invertible and (I T ) 1 = I + T + T 2 + . . . + T n 1 .
(b) Looks like 1 xn = (1 x)(1 + x + . . . + xn 1 ).
exer:5B:2 2. If is eigenvalue of T then let v be the eigenvector corresponding to . We have ((x

2)(x 3)(x 4))(T ))v = 0 or ( 2)( 3)( 4) = 0. Thus, = 2 or = 3 or = 4.
exer:5B:3 3. We have T 2 I = (T + I)(T I) so for any v 2 V, v 6= 0 then (T + I)((T I)v) = 0. If
there exists v so (T I)v = w 6= 0 then (T + I)w = 0 which follows 1 is eigenvalue of T ,
a contradiction. Thus, (T I)v = 0 for all v 2 V , which follows T = I.
exer:5B:4 4. We have 0 = P 2 P = P (P I) so for every v 2 V then P (P I)v = 0 which follows

(P I)v 2 null P . Hence, any v 2 V can be written as v = P v (P I)v where
P v 2 range P, (P I)v 2 null P . We also note that (null P ) \ (range P ) = {0}, otherwise
there exists u 2
/ null P but P u 2 null P which leads to P 2 u = 0 6= P u, a contradiction.
Thus, V = null P range P .
exer:5B:5 5. Note that (ST S 1 )k = (ST S 1 )(ST S 1 ) · · · (ST S 1 ) = ST k S 1 so if p = a0 + a1 x +
a2 x2 + . . . + am xm then
1 1 1 m
p(ST S ) = a0 I + a1 ST S + . . . + am (ST S ) ,
1
= a0 I + a1 ST S + . . . + am ST m S m
,
m 1
= S(a0 + a1 T + . . . + am T )S ,
1
= Sp(T )S .
exer:5B:6 6. Since U is invariant under T so for any u 2 U then T n u 2 U for all i 0. This follows
p(T )u 2 U for every u 2 U , i.e. U is invariant under p(T ) for any p 2 P(F).
exer:5B:7 7. Since 9 is eigenvalue of T 2 so there exists v 2 V, v 6= 0 so ((z 2 9)(T ))v = (T 2 9I)v = 0
or (((z 3)(z + 3))(T ))v = (T 3I)(T + 3I)v = 0. If (T + 3I)v = 0 then 3 is eigenvalue
of T . If (T + 3I)v = u 6= 0 then (T 3I)u = 0 which follows 3 is eigenvalue of T .
exer:5B:8 8. Imagine T as rotation 45 clockwise in the plane R2 . This will give T 4 = I. Note that
T (x, y) = (x cos 45 y sin 45 , x sin 45 + y cos 45 ).
exer:5B:9 9. If is a zero of p then p = (z )q(z). Hence 0 = p(T )v = (T I)q(T )v. Since
deg q < deg p so q(T )v = u 6= 0 which follows (T I)u = 0 or is eigenvalue of T .
exer:5B:10 10. Let p = a0 + a1 z + . . . + am z m then from T v = v we have
p(T )v = (a0 + a1 T + . . . + am T m )v,
= a0 v + a1 T v + . . . + am T m v,
m
= a0 v + a1 v + . . . + am v,
= p( )v.
exer:5B:10
exer:5B:11 11. If ↵ = p( ) for some eigenvalue of T then from exercise 10 (5B) we have ↵ is eigvalue of
p(T ). Conversely, if ↵ is eigenvalue of p(T ) then there exists v 2 V, v 6= 0 so p(T )v = ↵v
or (p(T ) ↵I)v = 0. Since F = C so we can factorise p ↵ = c(x 1 ) · · · (x m ) which
follows 0 = (p(T ) ↵I)v = c(T 1 I) · · · (T m I)v. Hence, there exists 1  j  m so
j is eigenvalue of T . Hence p(T )v = p( j )v = ↵v so ↵ = p( j ). See MSE,
exer:5B:8 the prob-
exer:5B:12 12. Back to exercise 8 (5B), there exists T 2 L(R2 ) so T 4 = I, i.e. 1 is eigenvalue of lem is still
(x4 )(T ). However, 1 6= p( ) = 4 for any 2 R. true if T is
only trian-
exer:5B:13 13. If subspace U 6= {0} theo:5.21:5B
is invariant under T . U is finite-dimensional then T |U 2 L(U ) so gularizable
according to theorem 7.3.1, there exists eigenvalue of T |U , which follows is eigenvalue on any fi-
of T , a contradiction. Thus U is either {0} or infinite-dimensional. nite vector
space over
a field F
0 1
0 0 1
exer:5B:14 @
14. Pick T so M(T ) = 1 0 0A.
0 1 0
0 1
1 1 0
exer:5B:15 15. Pick T so M(T ) = @1 1 0A.
0 0 1
exer:5B:16 16. If dim V = n then consider linear map S : Pn (C) ! V sotheo_16:3B:3.23
Sp = (p(T ))v. Since
dim Pn (C) = n + 1 > n = dim V so according to theorem 5.3.6, S is not injective.
Hence, there exists p 2 Pn (C), p 6= 0 so Sp = (p(T ))v = 0, which leads to existence of
eigenvalue of T .
exer:5B:17 17. If dim V = n then consider linear map S : Pn2 (C) ! L(V ) sotheo_16:3B:3.23
Sp = p(T ). Since
dim Pn2 (C) = n2 + 1 > n2 = dim L(V ) so according to theorem 5.3.6, S is not injec-
tive, i.e. there exists p 2 Pn2 (C), p 6= 0 so Sp = p(T ) = 0. Pick v 6= 0, v 2 V then
p(T )v = 0 which follows existence of eigenvalue of T .
exer:5B:18 18. Since V is finite-dimensional complex vector space so there exists so T I is invertible or
is not invertible. If is not eigenvalue of T then T I is invertible so dim range (T I) =
dim v. If is eigenvalue of T then T I is not invertible so dim range (T I)  dim v 1.
Thus, f ( ) is not a continuous function.
exer:5B:19 19. If F = C then there exists eigenvalue of T with exer:5B:10

eigenvector v 6= 0. Hence, for any
p 2 P(C) then p(T )v = p( )v according to exercise 10 (5B). On the other hand, we can
choose S 2 L(V ) so Sv 62 span v as dim V > 1. Hence, S 62 {p(T ) : p 2 P(C)} so
{p(T ) : p 2 P(C)} =
6 L(V ).
If F = R, if there exists eigenvalue of T then it’s similar to previous case. If T has no
eigenvalues,
theo:1
i.e. T has no invariant subspace with dimension 1 then according to theorem
??, T has invariant subspace U with dimension 2. Hence, for any v 2 U then p(T )v 2 U
for any p 2 P(R). If dim V > 2 then we can choose S 2 L(V ) so Su 62 U for u 2 U , which
follows S 62 {p(T ) : p 2 P(R)}. Thus, {p(T ) : p 2 P(R)} =
6 L(V ).
theo:1:5B
If dim V = 2 then U = V . According to proof of theorem 7.3.5 and the fact that V has no
eigenvalue, we can choose v 2 V so V = U = span (v, T v) for some v 2 V, v 6= 0. We choose
S 2 L(V ) so Sv = v and S(T v) = v for , 2 R; , 6= 0. If S 2 {p(T ) : p 2 P(R)}
then S = a1 I + a2 T . Note that a2 6= 0, otherwise v = S(T v) = (a1 I)(T v) = a1 T v, a
contradiction since v, T v is linearly independent. Thus, from v = Sv = (a1 I + a2 T )(v)
we obtain T v = a2a1 v, a contradiction since V has no eigenvalue. Thus, S 62 {p(T ) : p 2
P(R)}.
Thus, in all cases, we have {p(T ) : p 2 P(F)} =
6 L(V ).
theo:5.27:5B
exer:5B:20 20. It is a consequence of theorem 7.3.2. Since V is finite-dimensional complex vector space so
there exists a basis v1 , . . . , vn of V so M(T ) with respect to this basis is upper-triangular
matrix. Hence, for any 1  i  n, i-dimensional subspace Ui = span (v1 , . . . , vi ) of V is
invariant under T .
7.5. 5C: Eigenspaces and Diagonal Matrices

theo:5.38:5C Theorem 7.5.1 (5.38) Suppose V is finite-dimensional and T 2 L(V ). Suppose also that
1 , . . . , m are distinct eigenvalues of T . Then E( 1 , T ) + . . . + E( m , T ) is a direct sum.
Furthermore,
dim E( 1 , T ) + . . . + dim E( m , T )  dim V.
theo:5.41:5C Theorem 7.5.2 (5.41) Suppose V is finite-dimensional and T 2 L(V ). Let 1, . . . , m

denote the distinct eigenvalues of T . Then the following are equivalent:
(a) T is diagonalizable;
(b) V has basis consisting of eigenvectors of T ;
(c) there exists 1-dimensional subspace U1 , . . . , Un of V , each invariant under T so V =

U1 · · · Un .
(d) V = E( 1, T ) ··· E( m , T ).
(e) dim V = dim E( 1, T ) + . . . + dim E( m , T ).
theo:5.44:5C Theorem 7.5.3 (5.44) If T 2 L(V ) has dim V distinct eigenvalues then T is diagonaliz-
able.
7.6. Exercises 5C
exer:5C:1 1. Since T is diagonalizable so V has basis v1 , . . . , vn consisting of eigenvectors of T . Let
T vi = i vi . List of vj so j = 0 is basis of null space of T . For all vj so j 6= 0 then we find
T vj = j vj spans range of T , hence vj is basis of range of T . Thus, V = null T range T .
exer:5C:2 2. The converse is not true. For example, if v1 , v2 , v3 is basis of V , let T 2 L(V ) so T v1 =
0, T v2 = v3 , T v3 = v2 then span (v1 ) is null space of T and span (v2 , v3 ) is range of T .
Hence, V = null T range T but T is not diagonalizable as 0 is the only eigenvalue of T
with E(0, T ) = span (v1 ).
exer:5C:3 3. If (a) is true then (b) is also true. Let u1 , . . . , un be basis of null T which can be extended
to basis u1 , . . . , un , v1 , . . . , vm of V .
If (b) is true then since vi 2 V = null T + range T so vi wi 2 range T for some
wi 2 null T = span (u1 , . . . , un ). Note that v1 w1 , . . . , vm wm is linearly dependent
as v1 , . . . , vm is linearly dependent and since dim range T = m so v1 w1 , .P . . , vm wm
n
is
Pm basis of range T . Hence if there is v 2 (null T ) \ (range T ) then v = i=1 ↵i ui =
j=1 j j (v w j ) which leads to j = 0 as v 1 , . . . , v ,
m 1u , . . . , u n is linearly dependent.
Thus, v = 0 or (null T ) \ (range T ) = {0}.
theo_7:1C:1.45
If (c) is true then from theorem 3.2.4 (1C), we find null T +range T is a direct sum. Hence,
if v1 , . . . , vm is basis of null T , u1 , . . . , un is basis of range T then v1 , . . . , vm , u1 , . . . , un
is linearly independent. On the other hand, since V is finite-dimensional so dim V =
dim range T + dim null T = m + n. Thus, v1 , . . . , vm , u1 , . . . , un is basis of V , or V =
null T range T .
exer:5C:4 4. Let V = P(R) and T 2 L(V ) so T (1) = 0, T x = 0, T xi = x2 for all i 2. Then null T =
span (1, x) and range T = span (x2 ) so null T \ range T = {0} but V 6= null T + range T .
exer:5C:5 5. If is not eigenvalue of T then T I is invertible, i.e. range (T I) = V for any

operator T 2 L(V ). Thus, the problem is equivalent to proving that T is diagonalizable
iff V = null (T I) range (T I) for all eigenvalue of T .
Claim 1. If T diagonalizable then V = null (T I) range (T I) for every eigenvalue
of T .
Proof 1.LIf T is diagonalizable. Let 1 , . . . , m be distinct eigenvalues of T . For 1 Pi  m,
let U = j6=i E( j , T ) and consider (T i I) restricted to U . For any u 2 U, u = j6=i uj
with uj 2 E( j , T ) then
X X X
(T i I)u = j uj i uj = ( j i )uj .
j6=i j6=i j6=i
Thus, from theo:5.6:5A

this, we follow (T i I)|U 2 L(U ) and (T i I)|U is injective, so according
7.1.2 (5A) then (T
to theorem exer:5C:3 i I)|U is invertible so range (T i I)|U = U . According
to exercise 3 (5C) then
m
M
V = E( i , T ) = null (T i I) U = null (T i I) range (T i I)|U .
i=1
exer:5C:3
This follows, V = null (T i I) + range (T i I) so again from exercise 3, V = null (T
i I) range (T i I). This is true for all 1  i  m.
Proof 2. Note that if T is diagonalizable
exer:5C:1
then T I is also diagonalizable for any 2 C.
Hence, according to exercise 1 (5C) then V = null (T I) range (T I) for all 2 C,
as desired.
Claim 2. If V = null (T I) range (T I) for all 2 C then T is diagonalizable.

Proof. Since V is complex vector space so there exists eigenvalue 1 of T . Let U1 =
range (T 1 I) be a subspace of V then dim U1 < dim V since dim null (T 1 I) 1.
If dim U1 1 then there exists exer:5C:3
eigenvalue 2 of T |U1 . Since V = null (T 1 I) range (T
1 I) so according to exercise 3 (5C) then null (T 1 I) \ range (T 1 I) = {0}. Let
v2 2 U1 is eigenvalue of T |U1 corresponding to 2 . If 2 = 1 then v2 2 null (T 1 I),
a contradiction to the fact that null (T 1 I) \ range (T 1 I) = {0}. Thus, 1 6= 2 .
This follows E( 2 , T ) = E( 2 , T |U1 ).
Let U2 = range (T 2 I)|U1 then U2 is subspace of U1 . From previous argument, we know

that null (T 2 I) \ range (T 2 I) = {0} so null (T 2 I)|U1 \ range exer:5C:3
(T 2 I)|U1 = {0}
so U1 = null (T 2 I)| U1 range (T 2 I)| U1 according to exercise 3 (5C). Combining
all of these, we obtain
V = null (T 1 I) range (T 1 I),

= E( 1, T ) U1 ,
= E( 1, T ) null (T 2 I)|U1 range (T 2 I)|U1 ,
= E( 1, T ) E( 2 , T |U1 ) U2 ,
= E( 1, T ) E( 2, T ) U2 .
We do the same thing for U2 . This keeps going as long as dim Ui 1 (since there always
theo:5.21:5B
exists eigenvalue of an operator
L on complex vector space according to theorem 7.3.1).theo:5.41:5C
At
the end, we will obtain V = m i=1 E( i , T ) for eigenvalues i of T . From theorem 7.5.2
(5C), this follows T is diagonalizable.
exer:5C:6 6. Since T has dim V distinct eigenvalues so T is diagonalizable, hence according to theorem
theo:5.41:5C
7.5.2 (5C), V has basis v1 , . . . , vn which are eigenvectors of T corresponding to eigenval-
ues 1 , . . . , n . Thus, v1 , . . . , vn are also eigenvectors of S analogously corresponding to
1 , . . . , n . Hence, ST vi = S( i vi ) = i i vi = T Svi for all 1  i  n. This follows
ST = T S as v1 , . . . , vn is basis of V .
exer:5C:7 7. T has diagonal matrix A with respect to basis v1 , . . . , vn of V . Let 1 , . . . , m be distinct

eigenvalues of T . Let Sk be number of vectors vi so vi 2 E( k , T ), i.e. number of
times k appears on the diagonal A. Note that v1 , . . . , vn is linearly independent so
Sk  dim E( k , T ). Hence, we have
m
X m
X
n= Sk  dim E( k, T ) = n.
k=1 k=1
The inequality holds when Sk = dim E( k , T ), i.e. k appears on the diagonal A precisely
dim E( k , T ) times.
theo:5.6:5A
exer:5C:8 8. If both T 2I and T 6I are not invertible, then according to theorem 7.1.2 (5A), 2 and
6 are eigevalues of T , i.e. dim E(2, T ) 1, dim E(6, T ) 1. However,
dim E(2, T ) + dim (6, T ) + dim T (8, T ) > 5 = dim F5 .

theo:5.38:5C
This contradicts to theorem 7.5.1 (5C). Thus, either T 2I or T 6I or both are invertible.
exer:5A:21
exer:5C:9 9. According to exercise 21 (5A), if is not eigenvalue of T then 1
is not eigenvalue of T 1.
Hence, E( , T ) = E( 1 , T 1 ) = {0}.
exer:5A:21
If is eigenvalue of T then according to exercise 21 (5A), if v 2 E( , T ) then v 2 E( 1 , T 1 ).
Similarly and we obtain E( , T ) = E( 1 , T 1 ).

theo:5.38:5C
exer:5C:10 10. We have null T = null (T 0I) = E(0, T ). Hence, according to theorem 7.5.1,
m
X
dim E( i , T ) + dim E(0, T )  dim V = dim range T + dim null T.
i=1
Pm
This follows i=1 dim E( i , T )  dim range T.
exer:5C:11 11. Done.
exer:5C:12 12. Let v1 , v2 , v3 be eigenvectors of R corresponding to 2, 6, 7 then v1 , v2 , v3 is basis of F3 .

Similarly, let w1 , w2 , w3 be eigenvectors of T corresponding to 2, 6, 7 then w1 , w2 , w3 is
basis of V . Let S 2 L(F3 ) so Svi = wi then S is injective so S is invertible. We have
S 1 T Svi = S 1 T wi = S 1 (↵i wi ) = ↵i vi = Rvi . Thus, S 1 T S = R.
exer:5C:13 13. Let T, R 2 L(F4 ) so T v1 = Rv1 = 2v1 , T v2 = Rv2 = 6v2 , T v3 = Rv3 = 7v3 and T v4 =
v1 + 7v4 , Rv4 = v2 + 6v4 where v1 , . . . , v4 is basis of F4 . It’s not hard to check that T, R
only have 2, 6, 7 as eigenvalues. We show that for any if there is S 2 L(F4 ) so SR = T S
then S is not invertible, which can follow that there doesn’t exists invertible operator S
so R = S 1 T S.
P
Indeed, if Sv1 = 4i=1 ↵i vi then
4
!
X
T Sv1 = T ↵i vi = 2↵1 v1 + 6↵2 v2 + 7↵3 v3 + ↵4 (v1 + 7v4 ).
i=1
And
4
X
SRv1 = 2Sv1 = 2 ↵i vi .
i=1
With this, we follow ↵4 = 0, which leads to ↵2 = ↵3 = 0. Thus, Sv1 = ↵1 v1 . Similarly,

Sv2 = ↵2 v2 (note that we can’t imply Sv3 = ↵3 v3 , because of the way we choose T v4 =
v1 + 7v4 ).
P
Now, consider T Sv4 and SRv4 . If Sv4 = 4i=1 i vi then
4
!
X
T Sv4 = T i vi = 2 1 v1 + 6 2 v2 + 7 3 v3 + 4 (v1 + 7v4 ).
i=1
And
4
X
SRv4 = S(6v4 + v2 ) = ↵2 v2 + 6 i vi .
i=1
If T Sv4 = SRv4 then we must have 4 = 3 = 0. This follows Sv4 = 1 v1 + 2 v2 , i.e.

Sv4 2 span (Sv1 , Sv2 ) since Sv1 = ↵1 v1 , Sv2 = ↵2 v2 . Hence, S is not invertible.
In conclusion, with such T, R then there does not exists invertible operator S so R =
S 1 T S.
exer:5C:14 14. If T does not have a diagonal matrix for any basis of C3 then that means dim E(6, T ) =
dim E(7, T ) = 1 and T only has eigenvalues 6, 7. This makes the construction for T much
easier: For arbitrary basis of C3 then choose T 2 L(C3 ) so T v1 = 6v1 , T v2 = 7v2 , T v3 =
v1 + v2 + 6v3 .
Indeed, we show that T has only two eigenvalues 6 and 7. If there exists v = a1 v1 + a2 v2 +
a3 v3 6= 0 so
T v = (a1 v1 + a2 v2 + a3 v3 ) = 6a1 v1 + 7a2 v2 + a3 (v1 + v2 + 6v3 )
Then a3 = 6a3 , a1 = 6a1 + a3 , a2 = 7a2 + a3 . Hence, if 6= 6, 7 then a3 = a2 = a1 = 0,

a contradiction.
Next, we show that dim E(6, T ) = dim E(7, T ) = 1. Indeed, if there is v = a1 v1 + a2 v2 +
a3 v3 so T v = 6v = 6a1 v1 + 7a2 v2 + a3 (v1 + v2 + 6v3 ). This follows, a3 = 0, a2 = 0. Thus,
span (v1 ) = E(6, T ). Similarly, span (v2 ) = E(7, T ).
theo:5.41:5C
These two claims combining with theorem 7.5.2 (5C) deduce that T is not diagonalizable.
exer:5C:14
exer:5C:15 15. As claimed in exercise 14 (5C), since T is not diagonalizable, T has only 6, 7 as eigenvalues.
theo:5.6:5A
Hence, as 8 is not eigenvalue of T , T 8I is invertible according p to theorem 7.1.2 (5A),
which means there exists (x, y, z) 2 C3 so (T 8I)(x, y, z) = (17, 5, 2⇡).
exer:5C:16 16. (a) Induction on n, if T n 1 (0, 1) = (Fn 1 , Fn ) then
T n (0, 1) = T (Fn 1 , Fn ) = (Fn , Fn 1 + Fn ) = (Fn , Fn+1 ).
(b) If there is (x, y) 6= (0, 0) so T (x, y) = (x, y) = (y, x + y) then

p
x = y, y = x + y.
1± 5
This follows y( 2 1) = 0. Hence, eigenvalues of T is = 2 .
⇣ p ⌘ ⇣ p ⌘ ⇣ p ⌘ ⇣ p ⌘
(c) We have 1, 1 2 5 2 E 1 2 5 , T and 1, 1+2 5 2 E 1+2 5 , T .
⇣⇣ p ⌘ ⇣ p ⌘⌘
(d) We have (0, 1) = p15 1, 1+2 5 1, 1 2 5 . Hence,
p ! p !!
n 1 1+ 5
n n 1 5
T (0, 1) = p T 1, T 1, ,
5 2 2
" p !n p ! p !n p !#
1 1+ 5 1+ 5 1 5 1 5
=p 1, 1,
5 2 2 2 2
h⇣ p ⌘n ⇣ p ⌘n i
1+ 5 1 5
This follows Fn = p1
5 2 2 .
⇣ p ⌘n ⇣p ⌘n p
1 5 5 1 5
(e) It is equivalent to prove that p1
5 2 < 12 , which is true since 2 <1< 2 .
8. Chapter 6: Inner Product Spaces

8.1. 6A: Inner Products and Norms
hu, vi hu, vi
theo:6.14:6A Theorem 8.1.1 (6.14) Suppose u, v 2 V , with v 6= 0. Set c = and w = u v
kvk2 kvk2
then hw, vi = 0 and u = cv + w.
theo:6.15:6A Theorem 8.1.2 (6.15, Cauchy-Schwarz Inequality) Suppose u, v 2 V . Then | hu, vi | 

kuk kvk. The inqualitiy holds when one of u, v is a scalar multiple of the other.
8.2. Exercises 6A
exer:6A:1 1. Definiteness is not true, i.e. |1 · 0| + |1 · 0| = 0 but (1, 0) 6= (0, 0).
exer:6A:2 2. For (x1 , y1 , z1 ) 2 R3 so x1 y1 < 0 then h(x1 , y1 , z1 ), (x1 , y1 , z1 )i = 2x1 y1 < 0, a contradic-
tion to condition of positivity in definition of inner product.
exer:6A:3 3. (INCORRECT - NEED TO CHECK) According to definition of inner product then

hv, vi > 0 for v 6= 0, hence any inner product on V satisfying the old definition will
also satisfies the new definition.
Conversely, consider an inner product on V that satisfies the new definition, there exists
v 2 V, v 6= 0 so hv, vi > 0. Note that with new definition, we can’t define norm
p of vector,
however, we can avoid this notation by only woking on hu, ui rather than hu, ui. With
this in mind, we can check that the orthogonal decomposition strill works in the new
definition, but only for v, as shown:
hu, vi hu, vi
For any u 2 V then with c = and w = u v then hw, vi = 0 and u = cv + w.
hv, vi hv, vi
Hence, with this, we have hu, ui = hcv + w, cv + wi = c2 hv, vi + hw, wi
exer:6A:4 4. (a) hu + v, u vi = kuk2 hu, vi + hv, ui kvk2 . Since we’re talking about real inner
product, so hu, vi = hv, ui, done.
(b) This is deduced from (a).
(c) A rhombus made by two vectors u, v 2 R2 with same norm has two diagonals u v, u+v.
From (b), we find u + v is orthogonal to u v, which follows two diagonals of the rhombus
are perpendicular to each other.
p ⌦p p ↵
exer:6A:5 5. If there exists v 2 V, v 6= 0 so T v = 2v then kT vk2 = 2v, 2v = 2 kvk2 > kvk2 ,
p p
a contradiction. Thus there does not existspv 2 V, v 6= 0 so T v = 2v, theo:5.6:5A
i.e. 2 is
not eigenvalue of T , which follows that T 2I is invertible from theorem 7.1.2 under
assumption that V is finite-dimensional.
exer:6A:6 6. If hu, vi = 0 then according to Pythagorean’s theorem, ku + avk2 = |a|2 kvk2 +kuk2 kuk2
for all a 2 F.
If ku + avk2 kuk2 for all a 2 F then it is equivalent to 0  a2 kvk2 + 2aRe hv, ui. If
hv, ui 6= 0 then we can choose a so 0 > a2 kvk2 + 2aRe hv, ui, a contradiction. Thus,
hv, ui = 0.
exer:6A:7 7. Since a, b 2 R so kau + bvk2 = |a|2 kuk2 + 2ab hu, vi + |b|2 kvk2 , similarly, kbu + avk2 =
|b|2 kuk2 + 2ab hu, vi + |a|2 kvk2 . Hence, 2 2 2
⌘ = kbu + avk iff |a| kuk + |b| kvk =
⇣ kau + bvk
2 2 2
|b|2 kuk2 + |a|2 kvk2 iff |a|2 |b|2 kuk2 kvk2 = 0 for all a, b 2 R iff kvk = kuk.
theo:6.15:6A
exer:6A:8 8. We have | hu, vi | = 1 = kuk kvk so according to theorem 8.1.2, u = av for a 2 F. From
here we find a = 1 or u = v.
theo:6.15:6A
exer:6A:9 9. Let kuk = a, kvk = b then we have according to theorem 8.1.2 (6A), we have | hu, vi |  ab
so
(1 | hu, vi |)2 (1 ab)2 (1 a2 )(1 b2 ).
theo:6.14:6A h(1, 2), (1, 3)i 7 7

exer:6A:10 10. Using theorem 8.1.1 (6A), we have with c = 2 = then u = (1, 3) and
k(1, 3)k 10 10
✓ ◆
7 3 1
v = (1, 2) (1, 3) = , .
10 10 10
exer:6A:11 11. For x = theo:6.15:6A

(a1/2 , b1/2 , c1/2 , d1/2 ), y = (a 1/2 , b 1/2 , c 1/2 , d 1/2 ) 2 R4 then according to
theorem 8.1.2 (6A) we have
✓ ◆
2 2 2 1 1 1 1
16 = |hx, yi|  kxk kyk = (a + b + c + d) + + + .
a b c d
theo:6.15:6A
exer:6A:12 12. Apply theorem 8.1.2 to (x1 , . . . , xn ), (1, . . . , 1) 2 Rn with respect to Euclidean inner prod-
uct.
exer:6A:13 13. Draw triangle formed by u, v and u v. Let ✓ be angle between u and v then applying
law of cosines, we have
ku vk2 = kuk2 + kvk2 2 cos ✓ kuk kvk .
On the other hand, for u, v 2 R2 then
ku vk2 = kuk2 + kvk2 2 hu, vi .
Thus, hu, vi = kuk kvk cos ✓.

hx, yi
exer:6A:14 14. Since 1  cos ✓  1 so we must have 1   1, which is true according to
theo:6.15:6A kxk kyk
theorem 8.1.2.
exer:6A:15 15. Consider x = (a1 , 21/2 a2 , . . . , n1/2 an ), y = (b1 , 2 1/2 b2 , . . . , n 1/2 bn ) 2 R2 then according
theo:6.15:6A
to theorem 8.1.2 with respect to Euclidean inner product, we have the desired inequality.
p
exer:6A:16 16. 2kvk2 = ku + vk2 + ku vk2 2kuk2 = 34 so kvk = 17.
exer:6A:17 17. We have k(x, y)k = max{|x|, |y|} = 1

2 (|x + y| + |y x|) so k(x, y)k2 = 12 (x2 +y 2 +|y 2 x2 |).
If such inner product on R2 exists, then for any u, v 2 R2 then ku + vk2 + ku vk2 =
2(kuk2 + kvk2 ). Let u = (x, y), v = (a, b) then
ku+vk2 +ku vk2 = x2 +y 2 +a2 +b2 +1/2 (x + a)2 (y + b)2 +1/2 (x a)2 + (y b)2 ,
and
2(kuk2 + kvk2 ) = x2 + y 2 + a2 + b2 + |x2 y 2 | + |a2 b2 |.
Then we must have
(x + a)2 (y + b)2 + (x a)2 + (y b)2 = 2|x2 y 2 | + 2|a2 b2 |.
However, if x = 6, y = 5, a = 3, b = 2 then LHS = 92 32 + 72 32 = 112 but

RHS = 2(62 52 + 32 22 ) = 32, a contradiction. Thus, such inner product on R2 does
not exist.
exer:6A:19
exer:6A:18 18. Exercise 19 (next exercise) makes it easier, in which we have
1
h(1, 1), (1, 1)i = k(2, 0)k2 k(0, 2k2 = 0.
4
So (1, 1) is othorgonal to (1, 1), hence according to Pythagorean’s theorem then
k(1, 1) + (1, 1)k2 = k(2, 0)k2 = k(1, 1)k2 + k(1, 1)k2 .
We have LHS = 22 = 4 and RHS = 2k(1, 1)k2 = 2 · 22/p . This leads to p = 2.

exer:6A:19
If p = 2 then according to exercise 19, it suffices to show
1⇥ ⇤
h(x, y), (a, b)i = (x + a)2 + (y + b)2 (x a)2 (y b)2 = xa + yb.
4
is an inner product on R2 . This is obviously true as this inner product is Euclidean inner
product.
exer:6A:19 19. We have
ku + vk2 ku vk2 = hu + v, u + vi hu v, u vi ,
= kuk + hu, vi + hv, ui + kvk2
2
kuk2 hu, vi hv, ui + kvk2 ,
= 2 hu, vi + 2 hv, ui .
In real inner product, we find hu, vi = hv, ui, which will obtain the desired identity.
exer:6A:19
exer:6A:20 20. From exercise 19, we know that ku + vk2 ku vk2 = 2 hu, vi + 2 hv, ui. Next, we have
ku + ivk2 ku ivk2 = hu + iv, u + ivi2 hu iv, u ivi ,
2 2
= kuk + hu, ivi + hiv, ui + kivk
kuk2 + kivk2 hu, ivi hiv, ui ,
=2 [hu, ivi + hiv, ui] ,
=2i [hv, ui hu, vi] .
Thus,
1
hu, vi = ku + vk2 ku vk2 + ku + ivk2 i ku ivk2 i .
4
exer:6A:21 21. We will only prove for the case F = R, the other case F = C is similar (which proof
for this case can be seen here) . Define a function h , i : U ⇥ U ! R so hu, vi =
1
4 ku + vk
2 ku vk2 . Then it’s obvious that kuk = hu, ui1/2 . It suffices to show that
h; , i is an inner product.
Indeed, the positivity and definiteness conditions satisfies since hu, ui = kuk2 is 0 iff
kuk = 0 iff u = 0. Now we check the condition of additivity in first slot, we have
1
hu, vi + hw, vi = ku + vk2 + kw + vk2 ku vk2 kw vk2 .
4
Since the norm satisfies the parallelogram equality so
1 2
ku + vk2 + kw + vk2 = kw + u + 2vk2 + ku wk ,
2
1⇥ ⇤
= ku wk2 + 2 kw + u + vk2 + kvk2 kw + uk2 ,
2
1 1
= kw + u + vk2 + ku wk2 + kvk2 kw + uk2
2 2
Similarly, we obtain
1 1
ku vk2 + kw vk2 = kw + u vk2 + kvk2 + ku wk2 kw + uk2 .
2 2
Thus,
1
hu, vi + hw, vi = kw + u + vk2 kw + u vk2 = hw + u, vi .
4
Next we check the condition of homogeneity in first slot. From the condition of additivity of
first slot, we follow that hnu, vi = n hu, vi for any n 2 Z. For = a/b 2 Q, a, b 2 Z, b 6= 0
then for any u, v 2 U , we have u/b 2 U so b hu/b, vi = hu, vi. Hence,
a Da E
hu, vi = hu, vi = u, v = h u, vi .
b b
So far, homogeneity in first slot is true on Q. Since kuk + kvk ku + vk so (kuk + kvk)2
ku + vk2 so kukkvk hu, vi. From kuk + kvk ku + vk we also follow that the norm is
continuous.
Next, consider 2 C and consider sequence (rn )n 1 ! where ri 2 Q then we have

hrn u, vi h u, vi = h(rn )u, vi  k(rn )ukkvk = |rn |kukkvk.
Since limn!1 |rn |kukkvk = 0 so limn!1 hrn u, vi = h u, vi. Since rn 2 Q so limn!1 hrn u, vi =
hu, vi limn!1 rn = hu, vi. Thus, h u, vi = hu, vi for all 2 R.
Thus, the function h, i is indeed an inner product.
exer:6A:22 22. We need to show that for any a1 , . . . , an 2 R then (a1 + . . . + an )2  n atheo:6.15:6A
2 + . . . + a2 .
1 n
Consider x = (a1 , . . . , an ), y = (1, 1, . . . , 1) 2 Rn then according to theorem 8.1.2, we have
q
|a1 + . . . + an | = | hx, yi |  kxkkyk = n(a21 + . . . + a2n ).
exer:6A:23 23. We have h(v1 , . . . , vm ), (v1 , . . . , vm )i = hv1 , v1 i + . . . + hvm , um , i equals 0 iff hvi , vi i = 0 iff

vi = 0 for all 1  i  m. One can easily check other conditions in the definition of inner
space.
exer:6A:24 24. We have hu, ui1 = 0 iff hSu, Sui = 0 iff Su = 0 iff u = 0 since S is injective. We also have
hu + w, vi1 = hS(u + w), Svi = hSu, Svi + hSw, Svi = hu, vi1 + hw, vi1 .
Similarly, h u, vi1 = hu, vi1 and hu, vi1 = hSu, Svi = hSv, Sui = hv, ui1 .
exer:6A:25 25. Since S is not injective so there exists u 2 V, u 6= 0 so Sv = 0. Hence, hv, vi1 = 0 but
v 6= 0, which fails the definiteness condition of inner product.
exer:6A:26 26. (a) This is talking about Euclidean inner product space. So
hf (t), g(t)i0 = h(f1 (t), . . . , fn (t)) , (g1 (t), . . . , gn (t))i0 ,
= [f1 (t)g1 (t) + . . . + fn (t)gn (t)]0 ,
Xn
⇥ 0 ⇤
= fi (t)gi (t) + fi (t)gi0 (t) ,
i=1
n
X n
X
= fi0 (t)gi (t) + fi (t)gi0 (t),
i=1 i=1
⌦ ↵ ⌦ ↵
= f (t), g(t) + f (t), g 0 (t) .
0
(b) We have hf (t), f (t)i = kf (t)k2 = c2 so hf (t), f (t)i0 = 0. On the other hand, from (a)
then hf (t), f (t)i0 = 2 hf 0 (t), f (t)i so done.
(c) (DON’T KNOW WITH CURRENT KNOWLEDGE)
exer:6A:27 27. Since kw uk + kw vk2 = 1
2 k2w u vk2 + ku vk2 so we are done.
exer:6A:28 28. If there is two points u, v 2 C, u 6= v closest to w 2 C then kw uk = kw vk  kw rk for
2
all r 2 C. Since 12 (u+v 2 C so we have kw uk2  w 12 (u + v) = kw uk2 14 ku vk2 ,
which follows ku vk2 = 0 or u = v, a contradiction. Thus, there is at most one point in
C that is closest to w.
exer:6A:29 29. (a) d(u, v) 0 and d(u, v) = 0 iff ku vk = 0 iff u = v; d(u, v) = d(v, u); d(u, v)+d(v, w)exer:6A:21
=
ku vk + kv wk ku wk = d(u, w) according to definition of norm on exercise 21.
Thus d is metric on V .
(b) (I couldn’t solveexer:6A:21
this so I did some research and found the proof, note that P norm is
defined in exercise 21 P(6A) ). Let v1 , . . . , vn be basis of V . For all x 2 V, x = ni=1 ↵i vi ,
define a norm kxk1 = ni=1 |↵i | (can verify that it is indeed P a norm). Consider a Cauchy
sequence {xk }k 1 in the metric space (V, d) with xk = ni=1 ↵i,k vi , i.e. for any " > 0 there
exists N > 0 so for all m, n > N we have
n
X
d(xm , xn ) = kxm xn k = (↵i,m ↵i,n )vi < ".
i=1
Since any two norms on some finite-dimensional vector space V over F are equivalent, there
exists pairs of real numbers 0 < C1  C2 so for all x 2 V then C1 kxk1  kxk  C2 kxk1 .
Hence, for any j = 1, . . . , n, we have
k
X n
X
(↵i,m ↵i,n )vi C1 |↵i,m ↵i,n | C1 |↵j,m ↵j,n |.
i=1 i=1
This follows that for any " > 0, there exists N > 0 so for all m, n > N then |↵j,m ↵j,n | < ".
This follows that {↵j,i }i 1 is a Cauchy sequence in F. Since F is complete (see page 2 in
here P
for proof of case F = R) so for any j = 1, . . . , n, {↵j,i }i 1 converges to j . Consider
a = ni=1 i vj . We show that Cauchy sequence {xk }k 1 converges, i.e. for any " > 0,
there exists N > 0 so for all n > N then d(xn , a) = kxn ak < ". This is true because
n
X n
X n
X
kxn ak = (↵i,n i )vi  C2 |↵i,n i | < C2 "i .
i=1 i=1 i=1
(c) It suffices to prove that any Cauchy sequence {xk }k 1 in U converges to x in U . This
is true if we choose u1 , . . . , un as basis of U and prove similarly to (b), which will gives
u 2 span (u1 , . . . , un ) = U .
exer:6A:30 30. (Followed from the hint in the book) Let p = q + 1 kxk2 r for some polynomial r on Rn .
It suffices to choose r so p is a harmonic polynomial on Rn , i.e. 1 kxk2 r = ( q).
Let V be set of polynomial r on Rn so deg r P  deg q. Hence V is a finite-dimensional
n
vector space with basis xm1
1 m1
x 2 · · · x mn where
n i  deg q. Define an operator T on
i=1 mP
n
V so for r 2 V then T r P = 1 kxk r = 2 1 i=1 xi r. Consider polynomial r
2
n
so T r = 0, i.e. h = 1 i=1 xi r is harmonic polynomial. Since harmonic polynomial
2
h(x) = 0 for every x 2 Rn so kxk = 1, we follow h(x) = 0. Thus, T is injective, which

leads to T being surjective. Hence, for polynomial ( q), there exists polynomial r so
Tr = 1 kxk2 r = ( q). By letting p = q + 1 kxk2 r, we find p = 0, i.e. p
is a harmonic polynomial on Rn .
exer:6A:27
exer:6A:31 31. Use exercise 27 (6A), with a = w u, b = w v, c = u v and d = w 1
2 (u + v).
8.3. 6B: Orthonormal Bases

theo:6.30:6B Theorem 8.3.1 (6.30) Suppose e1 , . . . , en is an orthonormal basis of V and v 2 V . Then
v = hv, e1 i e1 + · · · + hv, e2 i and kvk2 = | hv, e1 i |2 + · · · + | hv, en i |2 .
theo:6.31:6B Theorem 8.3.2 (6.31, Gram-Schmidt Procedure) Suppose v1 , . . . , vm is a linearly inde-

pendent list of vectors in V . Let e1 = v1 /kv1 k. For j = 2, . . . , m, define ej inductively
by
vj hvj , e1 i e1 · · · hvj , ej 1 i ej 1
ej = .
kvj hvj , e1 i e1 · · · hvj , ej 1 i ej 1 k
Then span (v1 , . . . , vj ) = span (e1 , . . . , ej ) for every 1  j  m.
Gram-Schmidt Procedure helps us to find an orthonormal basis of a vector space. From this,
we follow:
theo:6.34:6B Theorem 8.3.3 (6.34) Every finite-dimensional inner product space has an orthonormal
basis.
theo:6.37:6B Theorem 8.3.4 (6.37) Suppose T 2 L(V ). If T has an upper-triangular matrix with
respect to some basis of V , then T has an upper-triangular matrix with respect to some
orthonormal basis of V .
theo:6.42:6B Theorem 8.3.5 (6.42, Riesz Representation Theorem) Suppose V is finite-dimensional and
' is a linear functional on V . Then there is a unique vector u 2 V such that 'v = hv, ui.
In particular, u = '(e1 )e1 + · · · + '(en )en for ANY orthonormal basis e1 , . . . , en of V .
8.4. Exercises 6B
exer:6B:1 1. p
(a) We have h(cos ✓, sin ✓), ( sin ✓, cos ✓)i = 0 and k(cos ✓, sin ✓)k = k( sin ✓, cos ✓)k =
(cos ✓)2 + (sin ✓)2 = 1. Similarly with the other orthonormal basis.
(b) If (x1 , y1 ), (x2 , y2 ) is an orthonormal basis of R2 then x1 x2 + y1 y2 = 0 and x21 +
y12 = x22 + y22 = 1. Since |x1 |  1 so there exists ✓ 2 R so x1 = cos ✓, which follows
y12 = (sin ✓)2 . WLOG, say y1 = sin ✓ and x1 6= 0 then x2 /y2 = y1 /x1 = tan ✓. Hence,
x22 + y22 = y 2 (tan ✓)2 + 1 = 1 so y22 = (cos ✓)2 . WLOG y2 = cos ✓ then x2 = sin ✓.
Thus, (x1 , y1 ), (x2 , y2 ) is (cos ✓, sin ✓), ( sin ✓, cos ✓).
theo:6.30:6B
exer:6B:2 2. If v 2 span (e1 , . . . , en ) then it’s true according to theorem 8.3.1.
Conversely, if kvk2 = | hv, e1 i |2 + · · · + | hv, en i |2 . Let u = hv, e1 ie1 + · · · + hv, en ien then
we have
hu, vi = hv, ui = | hv, e1 i |2 + · · · + | hv, en i |2 = kvk2 .
Hence, hv, ui = kvk2 . On the other hand, we also have kuk2 = kvk2 . This follows
ku vk2 = kuk2 + kvk2 hu, vi hv, ui = 0, hence v = u 2 span (e1 , . . . , en ).
theo:6.37:6B
exer:6B:3 3. According to the proof of theorem 8.3.4 (6B), it suffices to apply the Gram-Schmidt Pro-
theo:6.31:6B
cedure 8.3.2 to (1, 0, 0), (1, 1, 1), (1, 1, 2). We have e1 = (1, 0, 0) and
h(1, 0, 0), (1, 1, 1)i (1, 0, 0) = (0, 1, 1)

(1, 1, 1)
p ⇣ q q ⌘
with k(0, 1, 1)k = 2 so e2 = 0, 12 , 12 . We have
r r !
3 1 1
e2 h(1, 1, 2), e1 i e1 h(1, 1, 2), e2 i e2 = (1, 1, 2) (1, 0, 0) p 0, , ,
2 2 2
= (0, 1, 2) (0, 3/2, 3/2) = (0, 1/2, 1/2).
q ⇣ ⌘
Since k(0, 1/2, 1/2)k = 1
2 so e3 = 0, p1 , p1
2 2
.
exer:6B:4 4. For postive integers i, j so i > j, we have

Z ⇡ Z
cos ix cos jx 2 ⇡
dx = [cos(i j)x cos(i + j)x] dx,
⇡ ⇡ ⇡ ⇡

2 sin(i j)x sin(i + j)x ⇡
= = 0.
⇡ i j i+j ⇡
Z ⇡ Z
sin ix sin jx 2 ⇡
dx = [cos(i j)x cos(i + j)x] dx,
⇡ ⇡ ⇡ ⇡

2 sin(i j)x sin(i + j)x ⇡
= = 0.
⇡ i j i+j ⇡
Z ⇡ Z
sin ix cos jx 2 ⇡
= [sin(i + j)x + sin(i j)x] dx,
⇡ ⇡ ⇡ ⇡

2 cos(i + j)x cos(i j)x ⇡
= + = 0.
⇡ i j j i ⇡
Z 
⇡
cos ix 1 sin ix ⇡
p dx = p = 0.
⇡ 2⇡ 2⇡ i ⇡
Z ⇡ 
sin ix 1 cos ix ⇡
p dx = p = 0.
⇡ 2⇡ 2⇡ i ⇡
Z ⇡ Z ⇡
sin ix cos ix 1
dx = sin(2ix) dx = 0.
⇡ ⇡ 2⇡ ⇡
Similarly, we can find that
1 cos ix sin ix
p = p = p = 1.
2⇡ ⇡ ⇡
R1
exer:6B:5 5. We have k1k2 = 0 1 dx = 1 so e1 = 1. We have
Z 1
1
x hx, e1 i e1 = x x dx = x .
0 2
And
2 Z 1✓ ◆2 Z 1✓ ◆
1 1 2 1 1 1 1 1
x = x dx = x x+ dx = + =
2 0 2 0 4 3 2 4 12
p
so e2 = (2x 1) 3. We have
Z 1 Z 1
2
⌦ 2
↵ ⌦ 2
↵ 2 2
x x , e1 e 1 x , e2 e 2 = x x dx 3(2x 1) x2 (2x 1) dx,
0 0
1 1
= x2 (2x 1),
3 2
1
= x2 x .
6
Since
2 Z 1✓ ◆2
2 1 2 1
x x = x x dx,
6 0 6
Z 1✓ ◆
4 2 1 1
= x 2x + x2 + x +
3
dx,
0 3 3 36
7
= .
60
p
2 105
Thus, e3 = 7 x2 x 1
6 .
exer:6B:6 6. Since differentiation operator on P2 (R) has an upper-triangular
theo:6.37:6B
matrix with respect exer:6B:5
to
the basis 1, x, x2 so according to proof of theorem 8.3.4 (6B), the answer is in exercise 5.
exer:6B:7 7. Define a linear functional T on P2 (R) so T p = p 12 . Consider the inner product hp, qi =
R1 theo:6.42:6B
0 p(x)q(x) dx then according to theorem 8.3.5 (6B), if T p = hp, qi then q(x) = e1 (1/2)e1 +
e3 (1/2)e3 where e1 , e2 , e3 is orthonormal basis of P2 (R), which can be found
e2 (1/2)e2 +exer:6B:5
in exercise 5 (6B).
exer:6B:8 8. Similarly, we find
✓Z 1 ◆ ✓Z 1 ◆
q(x) = (cos ⇡x)dx + (2x 1) cos ⇡x dx 3(2x 1),
0 0
✓Z 1 ✓ ◆ ◆ ✓ ◆
1 60 1
+ x2 x cos ⇡x dx x2 x .
0 6 7 6
exer:6B:9 9. If the Gram-Schmidt Procedure is applied to a list of vectors v1 , . . . , vm that is not

linearly independent. WLOG say that v1 , . . . , wm 1 is linearly independent and vm 2
span (v1 , . . . , vm 1 ) then we can’t define em because kvm hvm , e1 i e1 · · · hvm , em 1 i em 1 k =
0.
exer:6B:10 10. Since span (v1 ) = span (e1 ) so v1 = hv1 , e1 i e1 or e1 = hv1v,e1 1 i . Observe that ke1 k = 1
so that means | hv1 , e1 i | = kv1 k, which leads to hv1 , e1 i = ±kv1 k. Thus, there are two
choices for e1 . Similarly, from vj = hvj , e1 i e1 + · · · + hvj , ej i ej we find hvj , ej i = ±kvj
hvj , e1 i e1 · · · hvj , ej 1 i ej 1 k so there are two ways to choose ej . In total, there are 2m
orthonormal lists e1 , . . . , em of vectors in V .
exer:6B:11 11. Choose a u 2 V, u 6= 0, let c = kuk21 /kuk22 . For any v 2 V, v 6=theo:6.14:6A

0 can be decomposed into
v = ku + w with hu, wi1 = 0 = hu, w, i2 according to theorem 8.1.1 (6A). Hence,
hv, ui1 = kkuk21 = ckkuk22 = c hv, ui2 .
On the other hand, we can also write u = kv + w with hv, wi1 = 0 = hv, w, i2 . We find
hv, ui1 = kkvk21 and hv, ui2 = kkvk22 . Combining with the above, we find kvk21 = ckvk22 for
any v 2 V .
Therefore, for any v, w 2 V , if one of them is 0 then hv, wi1 = c hv, wi2 = 0. If v, w 6= 0
then v = kw + u with hw, ui1 = hw, ui2 = 0. Hence,
hv, wi1 = kkwk21 = ckkwk22 = hv, wi2 .
exer:6B:12 12. Since V is finite-dimensional, there exists an othornormal basis e1 , . . . , en of V with respect
to inner product h·, ·i2 . We apply Gram-Schmidt Procedure to e1 , . . . , en with respect to
inner product h·, ·i1 and obtain another orthonormal basis u1 , . . . , un of V . With this, we
theo:6.30:6B
have span (e1 , . . . , ej )P
= span (u1 , . . . , uj ) for all 1  j  n. According to theorem 8.3.1
(6B) we have kvk22 = ni=1 |hv, ei i|2 and
n
X
kvk21 = |hv, uk i|2 ,
k=1
n
* k
+ 2
X X
= v, huk , ei i ei ,
k=1 i=1
n X
k 2
X
= huk , ei i hv, ei i ,
k=1 i=1
n k
!2
X X
 huk , ei i · |hv, ei i| ,
k=1 i=1
Xn X
= Ak |hv, ek i|2 + Bi,j |hv, ei i| · |hv, ej i| .
k=1 1i<jn
In above, Ak , Bi,j are constants obtained from |hui , ej i| so they will be the same regardless
of which
Pn v 2 V is Pchosen. Note that 2 |hv, ei i hv, ej i|  hv, ei i2 + hv, ej i2 so if we choose
C = k=1 Ak + 1i<jn Bi,j we will obtain Ckvk22 kvk21 for all v 2 V .
theo:6.31:6B
exer:6B:13 13. Apply Gram-Schmidt Procedure 8.3.2 (6B) to linearly independent list of vectors v1 , . . . , vm ,
we find an orthonormal list e1 , . . . , em so span (e1 , . . . , ej ) = span (v1 , . . . , vj ) for all
theo:6.30:6B
1  j  m. Hence, according to theorem 8.3.1 (6B), say w 2 span (v1 , . . . , vm ) then
we have w = hw, e1 i e1 + · · · + hw, em i em and vj = hvj , e1 i e1 + · · · + hvj , ej i ej so
j
X
hw, vj i = hw, ei i hvj , ei i = Cj + hw, ej i hvj , ej i.
i=1
It suffices to choose hw, ei i so hw, vj i > 0 for all 1  j  m.

For any 1  j  m, we have hvj , ej i = 6 0, otherwise vj 2 span (e1 , . . . , ej 1 ) = span (v1 , . . . , vj 1 ),
a contradiction since v1 , . . . , vm is linearly independent. Hence, we can actually choose
hw, ei i inductively on i so hw, vi i > 0. Thus, there exists w so hw, vi i > 0 for all 1  i  m.
exer:6B:14 14. Since kvi ei k2 < 1

n so we follow kvi k2 + n 1
n < 2| hvi , ei i | for all 1  i  n.
Assume the contrary P that v1 , . . . , vn is linearly dependent, i.e. there exsits
P not-all-zero
a1 , . . . , an 2 F so ni=1 ai vi = 0. From this, for all 1  j  n, we obtain ni=1 ai hvi , ej i =
h0, ej i = 0. Hence, if we consider the linear map T : Fn,1 ! Fn,1 so T x = M(T )x where
M(T )i,j = hvj , ei i for all 1  i, j  n. We know that with x = (a1 , . . . , an )T , x 6= 0 then
T x = 0 so T is not injective, hence T is not invertible so dim range T  n 1. Note that
theo:3.117:3F
according to theorem 5.10.7 (3F), dim range T equals column rank of M(T ) equals row
rank of M(T ). Hence, n rows Pn of M(T ) is linearly dependent in F , i.e.
1,n
P there exists not-
all-zero b1 , . . . , bn 2 F so i=1 bi (hv1 , ei i , · · · , hvn , ei i) = 0. Hence, ni=1 bi hvj , ei i = 0
P
for all 1  j  n so |bj hvj , ej i| = k6=j bk hvj , ek i for all 1  j  n. On the other hand,
we have
2 0 12
X X
bk hvj , ek i  @ |bk | · |hvj , ek i|A ,
k6=j k6=j
X X
 |bk |2 | hvj , ek i |2 ,
k6=j k6=j
X
2
= kvj k | hvj , ej i |2 |bk |2 ,
k6=j
 X
1
< (1 | hvj , ej i |)2 |bk |2 .
n
k6=j
P
Note that for all 1  j  n, k6=j kbk k 6= 0, otherwise we will have bk = 0 for all
2
k 6= j, which leads to bj hvj , ej i = 0 or hvj , ej i = 0 since bj 6= 0, a contradiction since

0 < kvi k2 + nn 1 < 2| hvi , ei i |. From this and the above inequality, we obtain
|bj |2 1/n (1 | hvj , ej i |)2

P 2
<
k6=j |bk | | hvj , ej i |2
This leads to
|b |2 1/n (1 | hvj , ej i |)2
Pn j < .
k=1 |bk |
2
| hvj , ej i |2 + 1/n (1 | hvj , ej i |)2
Next, we will show that
1/n (1 | hvj , ej i |)2 1

2  .
| hvj , ej i |2 + 1/n (1 | hvj , ej i |) n
Indeed, the above is equivalent to (n| hvj , ej i | n + 1)2 0, which is true. Therefore, for
Pn|bj |2
all 1  j  n then 2 < n,
1
which will leads to a contradiction if we sums up all the
k=1 |bk |
inequalities. Thus, v1 , . . . , vn is linearly independent so v1 , . . . , vn is basis of V .
exer:6B:15 15. We prove for the general CR ([a, b]) and '(f ) = f (c) for some c 2 [a, b]. Assume the
contrary, there exists such function g then pick f (x) = (x c)2 g(x), we have f (c) = 0
Rb Rb
and f (x) 2 CR ([a, b]). Hence, a f (x)g(x)dx = a [(x c)g(x)]2 dx = f (c) = 0 so this
follows (x c)g(x) = 0 or g(x) = 0 for x 2 [a, b] since g(x) is continuous on [a, b]. Hence,
hf, gi = 0 for all f 2 CR ([a, b]), which is a contradiction if we choose f so f (c) 6= 0. Thus,
there doesn’t exist such function g.
exer:6B:16 16. We induct on dim V . For dim V = 1 then there exists v 2 V so kvk = 1 and T v = v
with | | < 1. Hence, since 0 < | | < 1 so limm!1 | |m = 0 so for any ✏ > 0, there exists
M so for all m > M then kT m vk = k m vk = | |m < ✏.
If the problem is true for dim V = n 1. For dim V = n, since F = C,theo:5.27:5B
T has an upper-
triangular matrix with respect to some basis of V according to theorem 7.3.2 (5B), which
follows T has an upper-triangular matrix with respect to some orthonormal basis e1 , . . . , en
theo:6.37:6B
of V according to theorem 8.3.4 (6B). Hence, T is invariant under U = span (e1 , . . . , en 1 )
so according to inductive hypothesis, for 0 < ✏ < 1, there exists m 2 Z, m 1 so
kT m vk  ✏kvk for all v 2 U .
We can easily show that T m en = u + ↵n,n m where u 2 U . Hence, we can prove inductively
P im T (k 1 i)m u + ↵m(k 1) T m e . Since
that for any positive integer k then T mk en = ki=02 ↵n,n n,n n
max{✏, |↵n,n |} < 1 so limx!1 x max{✏, |↵n,n |}x = 0. Hence, there exists X so for all
x > X then x max{✏, |↵n,n |}x < kuk"
for sufficiently small " > 0. If we choose large enough
k, then for any 0  i  m 2, we have
im (k 1 i)m
k↵n,n T uk = |↵n,n |im T (k 1 i)m
u ,
 |↵n,n |im ✏ T (k 2 i)m

u ,
 |↵n,n |im ✏k 1 i kuk,
"
 .
i(m 1) + k 1
We can simultaneously choose large enough k so |↵n,n |m(k 1) kT m e

nk  which will lead
to
k 2
X
mk im (k 1 i)m m(k 1)
kT en k  ↵n,n T u + ↵n,n T m en ,
i=0
(k 1)"
 + .
i(m 1) + k 1
With this we follows that P

for any " > 0, there exists positive integer k so kT mk en k  ".
Hence, for any v 2 V, v = ni=1 i ei = w + i ei then
kT mk vk  kT mk wk + | n| · kT mk en k,
 ✏mk kwk + | n |",
p
= ✏mk | 1 |2 + · · · + | n 1 |2 + | n |",
p
 | 1 |2 + · · · + | n |2 = kvk.
We can choose to be really small with proper choice of ✏ and k.
exer:6B:17 17. (a) We show that for u, v 2 V then u+ v = (u + v). Indeed, we have for any w 2 V
then
( u + v)(w) = ( u)(w) + ( v)(w) = hw, ui + hw, vi = hw, u + vi = ( (u + v))(w).
Since F = R so we also have
( u)(w) = hw, ui = hw, ui = ( u)(w).
Thus, is a linear map from V to V 0 .

(b) If F = C, for v 2 V, V 6= 0, pick = 1 i then ( v)(v) = hv, vi = (1 + i)kvk2 6=
(1 i)kvk2 = ( v)(v). This is not a linear map.
(c) If u = 0 for some u 2 V then ( u)(u) = kuk2 = 0 which leads to u = 0. Thus, is
injective.
theo:3.95:3F
(d) Since V is finite-dimensional so according to theorem 5.10.1 (3F) then dim V = dim V 0
so dim V 0 = dim V = dim null + dim range = dim range so is surjective. Thus,
is invertible so is an isomorphism from V onto V 0 .
theo:6.42:6B
How is (d) an alternative proof of the Riesz Representation Theorem (8.3.5 (6B)) when
F = R?: Note since is surjective so for any linear functional ' in V 0 , there exists v 2 V
so v = ', which means '(u) = ( v)(u) = hu, vi for any u 2 V .
8.5. 6C: Orthogonal Complements and Minimization Problems
theo:6.47:6C Theorem 8.5.1 (6.47) Suppose U is finite-dimensional subspace of V . Then V = U U ?.

theo:6.51:6C Theorem 8.5.2 (6.51) Suppose U is finite-dimensional subspace of V . Then U = (U ? )? .
def:6.53:6C Definition 8.5.3 (orthogonal projection). Suppose U is Ra finite-dimensional subspace of V .

The orthogonal projection of V onto U is the operator PU L(V ) defined as follows: For v 2 V ,
write v = u + w where u 2 U, w 2 U ? . Then PU v = u.
theo:6.56:6C Theorem 8.5.4 (6.56) Suppose U is a finite-dimensional subspace of V , v 2 V and u 2 U .

Then kv PU vk  kv uk. The inequality holds iff u = PU v.
theo:6.55:6C Theorem 8.5.5 (Properties of orthogonal projection PU ) Suppose U is a finite-dimensional

subspace of V and v 2 V . Then
1. PU 2 L(V ).
2. PU u = u for every u 2 U and PU w = 0 for every w 2 U ? .
3. range PU = U and null PU = U ? .
4. v PU v 2 U ? .
5. PU2 = PU .
6. kPU vk  kvk.
7. For every orthonormal basis e1 , . . . , em of U then PU v = hv, e1 i e1 + · · · + hv, em i em .
8.6. Exercises 6C
exer:6C:1 1. If v 2 {v1 , . . . , vm }? iff hv, vi i = 0 for all 1  i  m iff hv, wi = 0 for all w 2
span (v1 , . . . , vm ) iff v 2 span (v1 , . . . , vm )? . Thus, {v1 , . . . , vm }? = span (v1 , . . . , vm )? .
exer:6C:2 2. If U = V then it’s obviously U ? = {0}. Conversely, if U ? = {0}. Let e1 , . . . , em be
orthonormaltheo:6.31:6B
basis of U , if U 6= V then there exists v 62 U, v 2 U . Apply Gram-Schmidt
Procedure 8.3.2 (6B) to linear independent list e1 , . . . , em , v to get a orthonormal list
e1 , . . . , em , em+1 which leads to em+1 2 U ? , a contradiction. Thus, U = V .
exer:6C:3 3. According to Gram-Schmidt Procedure then span (u1 , . . . , um ) = span (e1 , . . . , em ) = U
so e1 , . . . , em is orthonormal basis of U . Since each fi is orthogonal to each ej so fi 2 U ?
for each 1  i  n. Hence, f1 , . . . , fn is an orthonormal list in U ? . On the other hand,
dim U ? = dim V dim U = n so f1 , . . . , fn is orthonormal basis of U ? .
exer:6C:3
exer:6C:4 4. According to exercise 3 (6C), apply Gram-Schmidt Procedure to (1, 2, 3, 4), ( 5, 4, 3, 2),
(1, 0, 0, 0), (0, 1, 0, 0).
exer:6C:5 5. Since V is finite-dimensional so U, U T is finite-dimensional.
theo:6.51:6C
Hence, since U = (U ? )?
theo:6.47:6C
according to theorem 8.5.2 (6C), so from theorem 8.5.1 (6C), we have V = U U ? =
(U ? )? U ? , i.e. we can represent v uniquely as v = u + w where u 2 U = (U ? )? , w 2 U ?

hence PU ? v = w, PU v = u so (PU ? + PU ) v = u + w = Iv. Thus, PU ? v = I PU .
exer:6C:6 6. For any w 2 W then PU PW w = PU w = 0 so w 2 null PU = U ? , or hw, ui = 0 for all
u 2 U . Hence, hu, wi = 0 for all u 2 U , all w 2 W .
exer:6C:7 7. Let U = range P then null P ⇢ U ? since every vector in null P is orthogonal to every
exer:5B:4
vector in range P . On the other hand, since P 2 = P so accoording to exercise 4 (5B), we
find V = range P null P . Hence, each v 2 V can be written uniquely as v = P v + w
where P v 2 U, w 2 null P ⇢ U ? . Thus, since null P ⇢ U ? so PU v = PU (P v + w) = P v.
exer:6C:8 8. Let U = range P . We will show that every vectorexer:5B:4 in null P is orthogonal to every vector
in range P . Indeed, from P = P and exercise 4 (5B), we find V = range P null P .
2
Let e1 , . . . , em be orthonormal basis of range P and f1 , . . . , fn be orthonormal basis of

null P then e1 , . . . , em , f1 , . . . , fn is basis of V . It suffices to prove that hei , fj i = 0 for all
1  i  m, 1  j  n.
From P 2 = P we follow v P v 2 null P for all v 2 V , i.e. each v can be represented
uniquely
P as v = P Pvn+ w where w 2 nullPPm. Hence, for any ↵12, . . .P
, ↵m , 1 , . . . , n 2 F so
v= m ↵ e
i=1 i i + f
j=1 j j then P v = ↵ e
i=1 i i . Thus, kP vk = m
i=1 |↵i | and
2
n
X m X
X n
kvk2 = kP vk2 + | i |2 + 2Re ↵i j hei , fj i .
j=1 i=1 j=1
If there exists at least one hei , fj i 6= 0 then choose ↵x = y = 0 for all x 6= i, y 6= j and
↵i , j so ↵i j = Khei , fj i for some real number K < 0 so 2K| hei , fj i |2 + | j |2 < 0. Hence,
kvk2 kP vk2 = | j |2 + 2K| hei , fj i |2 < 0.
This contradicts to assumption that kvk kP vk. Thus, we must have hei , fj i = 0 for
every 1  i  m, 1  j  n. Thus, every vector in null P is orthogonal to every vector
in range P = U , or null P ⇢ U ? . Combining with v = P v + w for any v 2 V , we find
PU v = PU (P v + w) = P v for all v 2 V .
theo:6.47:6C
exer:6C:9 9. Since U is finite-dimensional subspace of V so according to theorem 8.5.1 (6C) we have
V = U U ? , or each v can be written uniquely as v = u + w for u 2 U, w 2 U ? .
If U is invariant under T then we have T u 2 U so for any v 2 V, v = u + w then PU TU v =
PU T u = T u = T PU v which leads to PU T PU = T PU . Conversely, if PU T PU u = T PU u or
PU T u = T u then that means T u 2 V for all u 2 U . Thus, U is invariant under T .
theo:6.47:6C
exer:6C:10 10. Since V is finite-dimensional so according to theorem 8.5.1 (6C) we have V = U U ? , i.e.
each v can be represented uniquely as v = u + w with u 2 U, w 2 U ? .
If U and U ? are both invariant under T then for any v 2 V, T v = T u + T w with T u 2
U, T w 2 U ? . Hence, PU T v = PU (T u + T w) = T u = T PU v so PU T = T PU . Conversely,
if PU T = T PU then for any u 2 U we have PU T u = T PU u = T u so T u 2 U . This
follows U is invariant under T . For any w 2 U ? we have PU T w = T PU w = T (0) = 0 so
T w 2 null PU = U ? . Thus, U ? is invariant under T .
theo:6.56:6C
exer:6C:11 11. Such u is exactly PU (1, 2, 3, 4) according to theorem 8.5.4 (6C). We do this step by step:
• Apply Gram-Schmidt Procedure to (1, 1, 0, 0), (1, 1, 1, 2) and obtain orthonormal basis
of U which is e1 , e2 .
• Calculate h(1, 2, 3, 4), e1 i , h(1, 2, 3, 4), e2 i (note that since V = R4 so we’re talking
about Euclidean inner product).
• PU (1, 2, 3, 4) = h(1, 2, 3, 4), e1 i e1 + h(1, 2, 3, 4), e2 i e2 is the answer.
exer:6C:12 12. Define subspace U of P3 (R) as U = {p(x) 2 P3 (R) deg p  3, p0 (0) = 0, p(0) = 0}. This
follows U = {ax3 + bx2 : a, b 2 R} = span (x3 , x2 ). Hence, similarly to previous exercise,
we find orthonormal basis of U from basis x3 , x2 and inner product on P3 (R):
Z ⇡
hp, qi = p(x)q(x)dx.
⇡
From that we find PU (3x + 2) = h3x + 2, e1 i e1 + h3x + 2, e2 i e2 .
exer:6C:13 13. It is exercise 6.58 in the book.
exer:6C:14 14. (a) If U ? 6= {0}, then we show that for any g 2 U ? , g 6= 0, we have xg(x) 2 U ? . Indeed,
for any f 2 U then f (x)x 2 U so
Z 1
hf (x), xg(x)i = [xf (x)] g(x)dx = hxf (x), g(x)i = 0.
1
Hence, xg(x) 2 U ? . On the other hand, (xg(x))(0) = 0 so xg(x) 2 U . Thus, we must

have xg(x) = 0 or g(x) = 0, a contradiction. Thus, U ? = {0}. Another proof for (a) can
be seen here.
theo:6.47:6C
theo:6.51:6C
(b) Since U ? = {0} so theorems 8.5.1, 8.5.2 (6C) do not hold in this case where CR ([ 1, 1])
is infinite-dimensional.
9. Chapter 7: Operators on Inner Product Spaces

9.1. 7A: Self-adjoint and Normal Operators
V, W are assumed to be finite-dimensional inner product spaces over F.
def:7.2:7A Definition 9.1.1 (adjoint). Suppose T 2 L(V, W ). The adjoint of T is the function T ⇤ : W !
V such that hT v, wi = hv, T ⇤ wi .
def:7.11:7A Definition 9.1.2 (self-adjoint operator). An operator T 2 L(V ) is called self-adjoint or Her-
mitian if T = T ⇤ .
theo:7.7:7A Theorem 9.1.3 (7.7) Suppose T 2 L(V, W ) then:
(a) null T ⇤ = (range T )? .
(b) range T ⇤ = (null T )? .
(c) null T = (range T ⇤ )? .
(d) range T = (null T ⇤ )? .
theo:7.7:7A
From theorem 9.1.3, we find that if T is self-adjoint, then that means null Tdef:6.53:6C
= (range T )? . If
adding the condition T = T , we obtain that T is an orthogonal projection 8.5.3.
2
theo:7.10:7A Theorem 9.1.4 (7.10) Let T 2 L(V, W ). Suppose e1 , . . . , en is an orthonormal basis of V

and f1 , . . . , fm is an orthonormal basis of W . Then M(T ⇤ , (f1 , . . . , fm ), (e1 , . . . , en )) is the
conjugate transpose of M(T, (e1 , . . . , en ), (f1 , . . . , fm )).
theo:7.10:7A
From theorem 9.1.4 (7A), we find that for self-adjoint operator A, i.e. A = A⇤ then matrix
of A with respect to some orthonormal basis is equal to its conjugate transpose. This is called
Hermitian matrix. If we’re talking about real inner product space then matrix of self-adjoint
operator A (wrt orthonormal basis) is equal to its own transpose. This kind of matrix is called
real symmetric matrix.
theo:7.13:7A Theorem 9.1.5 (7.13) Every eigenvalue of a self-adjoint operator is real.
theo:7.21:7A Theorem 9.1.6 Suppose T 2 L(V ) is normal and v 2 V is eigenvector of T with eigenvalue
. Then v is also eigenvector of T ⇤ with eigenvalue .
theo:7.22:7A Theorem 9.1.7 (7.22) Suppose T 2 L(V ) is normal. Then eigenvectors of T corresponding
to distinct eigenvalues are orthogonal.
How to find the adjoint T ⇤ of T 2 L(V, W )?: There are two ways:
exer:7D:6 exer:7A:1
1. Directly: Find T ⇤ from hT v, wi = hv, T ⇤ wi. See exercise 6 (7D) or exercise 1 (7A) as
examples.
2. Indirectly: Find orthonormal bases of V and W . Write M(T ) with respect two these two
theo:7.10:7A
bases then M(T ⇤ ) is the conjugate transpose of M(T ) according to theorem 9.1.4 (7A).
9.2. Exercises 7A
exer:7A:1 1. For any (a1 , . . . , an ) 2 Fn then
h(a1 , . . . , an ), T ⇤ (z1 , . . . , zn )i = hT (a1 , . . . , an ), (z1 , . . . , zn )i ,

= h(0, a1 , . . . , an 1 ), (z1 , . . . , zn )i ,
n
X1
= ai zi+1 ,
i=1
= h(a1 , . . . , an ), (z2 , . . . , zn , 0)i .
Thus, T ⇤ (z1 , . . . , zn ) = (z2 , . . . , zn , 0).

theo:7.7:7A
exer:7A:2 2. According to theorem 9.1.3 (7A), we have range (T ⇤ I) = (null (T I))? so that
means
dim null (T ⇤ I) = dim V dim range (T ⇤ I),

= dim V (dim null (T I))? ,
= dim null (T I).
Thus, is eigenvalue of T iff dim null (T I) 1 iff dim null (T ⇤ I) 1 iff is

eigenvalue of T ⇤ .
exer:7A:3 3. U is invariant under T iff T u 2 U for every u 2 U iff for every u 2 U , hu, T ⇤ vi = hT u, vi = 0

for all v 2 U ? iff T ⇤ v 2 U ? for every v 2 U ? iff U ? is invariant under T ⇤ .
theo:7.7:7A
exer:7A:4 4. T is injective iff null T = {0} iff (range T ⇤ )theo:6.51:6C
? = {0} (according to theorem 9.1.3) iff
range T = {0} = V (according to theorem 8.5.2 6C) iff T ⇤ is surjective. Similarly, T is

⇤ ?
surjective iff T ⇤ is injective.
exer:7A:5 5. We have dim range T = dim V

theo:7.7:7A
dim null T = dim (null T )? but according to theo-
rem 9.1.3 then dim (null T )? = dim range T ⇤ so dim range T = dim range T ⇤ . Since
dim range T = dim (null T ⇤ )? = dim W dim null T ⇤ and dim range T ⇤ = dim (null T )? =
dim V dim null T so we obtain the desired equality.
exer:7A:6 6. (a) Write out orthonormal basis of P2 (R) using Gram-Schmidt Procedure to 1, x, x2 then
write out M(T ) with respect to this orthonormal basis.
(b) This is not a contradiction because the matrix of T is created from nonorthonormal
basis of P2 (R).
exer:7A:7 7. Since S, T are self-adjoint so for any v, w 2 V then hST v, wi = hT v, Swi = hv, T Swi.
Thus, ST is self-adjoint iff for any v, w 2 V then hST v, wi = hv, ST wi iff hv, T Swi =
hv, ST wi iff ST = T S.
exer:7A:8 8. We have 0 is a self-adjoint operator on V . For any self-adjoint operators S, T on V then

S +T = S ⇤ +T ⇤ = (S +T )⇤ so S +T is self-adjoint. For any 2 R then T = T ⇤ = ( T )⇤
so T is self-adjoint. Thus, the set of self-adjoint operators on V is a subspace of L(V ).
exer:7A:9 9. For any self-adjoint operator T 6= 0 on V , then with =1 i, we have T = (1 i)T =

6 (1 + i)T ⇤ = ( T )⇤ .
(1 i)T ⇤ =
exer:7A:10 10. We show that the set of normal operators

✓ on◆V is not closed under addition. It suffices to
a b
prove this when dim V = 2. Note that is the matrix of a normal operator iff b2 = c2
c d
✓ ◆
1 2
and ac + bd = ab + cd. Hence, with two normal operators S, T so M(S) = and
2 3
✓ ◆ ✓ ◆
2 3 3 1
M(T ) = then M(S + T ) = so S + T is not a normal operator. Thus,
3 2 5 5
for dim V = 2, set of normal
✓ operators
◆ on V is not a subspace of L(V ). For dim V > 2, we
M(S)
can choose M(X) = (i.e. the 2-by-2 matrix at top left of M(X) is exactly
0
✓ ◆
M(T )
M(S) while the remaining entries of M(X) are 0) and M(Y ) = .
0
theo:7.7:7A
exer:7A:11 11. If P is self-adjoint then P = P ⇤ so according to theorem 9.1.3 (7A) then null P = null P ⇤ =
(range T )? , i.e. every vector in null P is orthogonal to every vector in range P , which
exer:6C:7
brings us back to exercise 7 (6C), which proves the existence of U .
Conversely, if such subspace U exists then since P = PU so range P = range PU = U . This
follows null P = null PU = U ? . Thus, range P = (null P )? = range P ⇤ and similarly,
null P = null P ⇤ . On the other hand, since P = P 2 so P ⇤ = (P 2 )⇤ = (P ⇤ )2 so according
exer:5B:4
to exercise 4 (5B) we find every v can be represented uniquely as v = P ⇤ v + w where
w 2 null P ⇤ . We also have v = P v + w since P = P 2 and null P ⇤ = null P and range P =
range P ⇤ . Thus, P v = P ⇤ v for every v 2 V so P = P ⇤ or P is self-adjoint.
exer:7A:12 12. Let v1 , v2 be eigenvectors of T corresponding to eigenvalues 3, 4. theo:7.22:7A

Then for e1 = v1 /kv1 k, e2 =
v2 /kv2 k then T e1 = 4e1 and T e2 = 4e2 . According to theorem 9.1.7 (7A) then he1 , e2 i = 0
so hT e1 , T e2 i = 0 so according to Pythagorean’s theorem we find 52 = 32 + 42 = kT e1 k2 +
kT e2 k2 = kT (e1p+ e2 )k2 so kT (e1 + e2 )k = 5 and note that ke1 + e2 k2 = ke1 k2 + ke2 k2 = 2
so ke1 + e2 k = 2.
0 1
2 3 0 0
B3 2 0 0C
exer:7A:13 13. M(T ) = B @0 0 0 0A.
C
0 0 0 0
theo:7.22:7A
exer:7A:14 14. Since T is normal so according to theorem 9.1.7 (7A) then hT v, T wi = hv, wi = 0 so
22 (32 + 42 ) = kT vk2 + kT wk2 = kT (v + w)k2 so kT (v + w)k = 10.
exer:7A:15 15. First, we will find adjoint of T . For any w 2 V then
hv, T ⇤ wi = hT v, wi = hv, ui hx, wi = hv, hw, xi ui .
Thus, T ⇤ w = hw, xi u.
(a) With F = R, T is self-adjoint then T v = T ⇤ v for every v 2 V so hv, ui x = hv, xi u so
u, x is linearly dependent.
If u, x is linearly dependent with u = x then for any v 2 V we have hv, ui x = hv, xi u/ =
hv, xi u so T = T ⇤ or self-adjoint.
(b) We have T T ⇤ v = hv, xi T u = hv, xi kuk2 x and T ⇤ T v = hv, ui T ⇤ x = hv, ui kxk2 u. If T
is normal then T T ⇤ = T ⇤ T then u, x is linearly dependent.
If u = x then for every v 2 V , we have
⌦ ↵
hv, xi kuk2 x = v, 1 u | |2 kxk2 x,
⌦ ↵
= v, /| |2 u | |2 kxk2 x,
= /| |2 hv, ui | |2 kxk2 x,
= hv, ui kxk2 u.
This follows T T ⇤ v = T ⇤ T v for every v 2 V so T is normal.
exer:7A:16 16. First, we will show that null T = null T ⇤ . Indeed, we have v 2 null T iff 0 = kT vk2 =
kT ⇤ vk2 iff T ⇤ v = 0 or v 2 null T ⇤ . Thus, theo:7.7:7A
null T = null T ⇤ so (null T )? = (null T ⇤ )? so
range T = range T according to theorem 9.1.3 (7A).
⇤
exer:7A:17 17. We show that for any positive integer k then null T k = null T k+1 . Indeed, it’s obvious
that if v 2 null T k then v 2 null T k+1 .exer:7A:16
If v 2 null T k+1 then T k v 2 null T and since
null T = null T according to exercise 16 (7A) so T k v 2 null T ⇤ . On the other hand,
⇤
we also have T k v 2 range T and null T ⇤ = (range T )? so we must have kT k vk2 = 0 so

T k v = 0, i.e. v 2 null T k . Thus, we obtain null T k = null T k+1 for every positive integer
k.
Note that if T is normal then T k is also normal for every positive integer k. Hence,
null T k = null (T k )⇤ = (range T k )? and similarly, null T k+1 = (range T k+1 )? . Thus,
range T k = range T k+1 for every positive integer k 1.
exer:7A:18 18. Consider 2-dimensional

✓ vector space
◆ V over C so with✓ respect to ◆ orthonormal basis e1 , e2
1 1 i 1 1+i
of V then M(T ) = and M(T ⇤ ) = . Then we have kT e1 k =
1+i 1 1 i 1
p p
12 + |1 + i|2 = 12 + |1 i|2 = kT ⇤ e1 k and similarly kT e2 k = kT ⇤ e2 k. However, T is
not normal.
exer:7A:19 19. Since T (1, 1, 1) = 2(1, 1, 1) so (1, 1, 1) is eigenvector of T corresponding to eigenvalue 2.

If (z1 , z2 , z3 ) 2 null T then (ztheo:7.22:7A
1 , z2 , z3 ) is eigenvector of T corresponding to eigenvalue 0.
Hence, according to theorem 9.1.7 (7A) then h(z1 , z2 , z3 ), (1, 1, 1)i = z1 + z2 + z3 = 0.
exer:7A:20 20. Note that both V T ⇤ and T 0 w are linear map from W to V . Hence, it suffices to
0
prove that for every w 2 W then ( V T )(w) = (T

⇤ 0
W )(w) or V (T w) = T ( W w).
⇤ 0
Since there two are linear functionals on V so it again suffices to prove that for every
v 2 V then ( V (T ⇤ w))v = (T 0 ( W w))v. Indeed, according to definition of V , we
have ( V (T ⇤ w))v = hv, T ⇤ wi. On the other hand, (T 0 ( W w))v = (( W w) T )v =
( W w)(T v) = hT v, wi. Since hv, T ⇤ wi = hT v, wi so we obtain the desired equality.
exer:6B:4
exer:7A:21 21. From exercise 4 (6B), we find the list
1 cos x cos nx sin x sin nx

p , p ,..., p , p ,··· , p
2⇡ ⇡ ⇡ ⇡ ⇡
is the orthonormal basis of V .

(a) Note that we are talking about vector space over R so M(D⇤ ) with respect to the
mentioned basis is eventually transpose of M(D). Hence, to show D⇤ = D, it suffices to
prove that M(D)j,i + M(D)i,j = 0. Note that the entry in row i, column j of M(D) is
hDej , ei i where ek is the k-th vector in the orthonormal basis mentioned above. It sufficesexer:6B:4
to show hDej , ei i + hDei , ej i = 0 for every 1  i, j  2n + 1. From integrals in exercise 4
(6B), we know that if |j i| = 6 n then hDei , ej i = 0. Hence, we only need to consider when
p p
|i j| = n, i.e. if ei = cos(mx)/ ⇡ and ej = sin(mx)/ ⇡ for some 1  m  n. We have
hDei , ej i = m kej k2 = m and hDej , ei i = mkei k2 = m so hDei , ej i + hDej , ei i = 0.
Thus, we obtain D⇤ = D as desired. Note that M(D) 6= 0 so D⇤ 6= D so D is not
self-adjoint but D is normal.
(b) It suffices to prove T = T ⇤ , i.e. hT ej , ei i = hT ei , ej i for every 1  i, j  2n + 1. From
exer:6B:4
integrals in exercise 4 (6B), when i 6= j then hT ej , ei i = hT ei , ej i = 0. Thus, we obtain
T = T ⇤ or T is self-adjoint.
9.3. 7B: The Spectral Theorem

theo:7.24:7B Theorem 9.3.1 (7.24, Complex Spectral Theorem) Suppose F = C and T 2 L(V ). Then
the following are equivalent :
(a) T is normal.
(b) V has an orthonormal basis consisting of eigenvectors of T .
(c) T has a diagonal matrix with respect to some orthonormal basis of V .

theo:7.29:7B Theorem 9.3.2 (7.29, Real Spectral Theorem) Suppose F = R and T 2 L(V ). Then the
following are equivalent:
(a) T is self-adjoint.
(b) V has an orthonormal basis consisting of eigenvectors of T .
(c) T has a diagonal matrix with respect to some orthonormal basis of V .
How to find orthonormal basis consisting of eigenvectors of a normal/self-adjoint operator T :

• Solve T v = v to find all eigenvalues.
• Find a basis of E( , T ) for each eigenvalue of T .

theo:6.31:6B
• Using Gram-Schidmt Procedure 8.3.2 (6B), find orthonormal basis of E( , T ) for each
eigenvalue of T .
• Combines all the bases of E( , T ), given that T is self-adjoint/normal, we will obtain a

orthonormal
exer:7B:4
exer:7B:5
basis consisting of eigenvectors of T . This is true according to two exercises
4, 5 (7B).
9.4. Exercises 7B
exer:7B:1 1. Pick T 2 L(R3 ) so T (1, 1, 0) = 2(1, 1, 0), T (0, 1, 0) = 3(0, 1, 0) and T (0, 0, 1) = 4(0, 0, 1)
then with respect to standard basis
0 (which is1 also orthonormal 0 basis with1respect to usual
2 0 0 2 1 0
inner product) we have M(T ) = @ 1 3 0 so M(T ) = 0 3 0A. Thus T is not
A ⇤ @
0 0 4 0 0 4
self-adjoint.
exer:7B:2 2. Since T is self-adjoint so according to Real/Complex Spectral Theorem, T has an diagonal

matrix so V = E(2, T ) E(3, T ). So every v can be represented as u + w where u 2
E(2, T ), w 2 E(3, T ). Hence (T 2 5T + 6I)v = (T 3I)(T 2I)(u + w) = (T 3I)[(T
2I)u+(T 2I)w] = (T 3I)(T 2I)w = (T 2I)[(T 3I)w] = 0. Thus, T 2 5T +6I = 0.
exer:7B:3 3. Pick T 2 L(C)3 ) so T (1, 0, 0) = 3(1, 0, 0), T (0, 1, 0) = 2(0, 1, 0) and T (0, 0, 1) = 2(1, 1, 1).
Then (T 2 5T + 6I)(0, 0, 1) = ( 2, 0, 0) so T 2 5T + 6I 6= 0. 2 and 3 are the only
eigenvalues of T .
theo:7.22:7A
exer:7B:4 4. If T is normal then according to theorem 9.1.7 (7A), every pairs of eigenvectors correspond-
ing to distinct eigenvalues of T are orthogonal. According
theo:7.24:7B
to Complex Spectral Theorem
theo:5.41:5C
9.3.1 then T has diagonal matrix so from theorem 7.5.2 (5C), V = E( 1 , T ) · · · E( m , T ).
theo:6.34:6B
Conversely, according to theorem 8.3.3 (6B), each finite-dimensional vector space E( i , T )
has an orthonormal basis. Combining all orthonormal bases of vector spaces E( i , T ) for
1  i  m we obtain a new orthonormal basis (since every vector in E( i , T ) is orthogonal
to every vector in E( j , T ) for j 6= i). This orthonormal basis has length dim E( 1 , T ) +
. . . + dim E( m , T ), which equals to dim V because V = E( 1 , T ) · · · E( m , T ). Thus,

we obtain an orthonormal basis of V consisting of eigenvectors of T so T is normal.
exer:7B:4
exer:7B:5 5. Completely the same as previous exercise 4, except this time we use Real Spectral Theorem
instead of Complex Spectral Theorem and note that any self-adjoint operator is normal.
theo:7.13:7A
exer:7B:6 6. One direction is already proven by theorem 9.1.5 (7A). Thus, if everytheo:7.24:7B
eigenvalue of a
normal operator T is real then according to Complex Spectral Theorem 9.3.1 (7B), T has
theo:7.10:7A
diagonal matrix M(T ) consisting of all real numbers. With this and theorem 9.1.4 (7A),
we find M(T ) = M(T ⇤ ) so T is self-adjoint.
exer:7A:16
exer:7A:17
exer:7B:7 7. Since T is normal so according to exercise 16, 17 (7A) then = range T ⇤ = range
theo:7.7:7A
T =
theo:6.47:6C
range T which follows V = null T
7 7 range T from theorem 9.1.3 (7A) and 8.5.1 (6C).
7
Thus, each v 2 V can be written as v = T 7 u + w where w 2 null T 7 . Thus, T 2 v =

T 2 (T 7 u + w) = T 9 u = T 8 u = T (T 7 u + w) = T v. We conclude T 2 = T . If v 6= 0
is an eigenvector of T corresponding to eigenvalue 2 C then T v = 2 v = v = T v
so ( 1)v = 0 so exer:7B:6
2 {0, 1}. Thus, all eigenvalues of normal operator T are real so
according to exercise 6 (7B), T is self-adjoint.
exer:7B:8 8. UNSOLVEd
exer:7B:9 9. Every normal operator T has a diagonal matrix according to Complex Spectral Theorem
theo:7.24:7B
9.3.1
p (7B). Pick operator S 2 L(V ) so S has diagonal matrix M(S) and M(S)i,i =
M(T )i,i for every 1  i  n (this is possible because we are considering complex vector
space). Hence, M(S 2 ) = (M(S))2 = M(T ) so S 2 = T .
theo:1:5B
exer:7B:10 10. Theorem 7.3.5 (look at the proof) makes the construction easier. It suffices to choose
an operator T so that T has no eigenvalue. Pick V = R2 and T (x, y) = ( y, x) then if
T (x, y) = (x, y) we find y = x, x = y so y( 2 + 1) = 0, which happens when y = 0
so x = 0. Thus, T has no eigenvalues. Hence, since for any v 2 R2 , v 6= 0 then T 2 v, T v, v
is linearly dependent so there exists real numbers b, c so b2 < 4c and T 2 v + bT v + cv = 0.
The condition b2 < 4c guarantees that T has no eigenvalue, otherwise we can factorise
T 2 v + bT v + cv and then obtain eigenvalue for T .
exer:7B:11 11. If T is self-adjoint then T is also normal so according to Real/Complex Spectral Theorem, T
has a diagonal matrix M(T ) with respect to some p orthonormal basis of V . Pick S 2 L(V )
so S also has a diagonal matrix so M(S)i,i = M(T )i,i for every i. We can conclude
3
S 3 = T from here.
exer:7B:12 12. Since T is self-adjoint so according to Complex/Real Spetral Theorem, V has an orthonor-
mal basis
P consisting of eigenvectors of T , i.e. each v 2 V, kvk = 1 can be represented as
vP= m i=1 ai ui where ui 2 E( i , T ) and hui , uj i = 0 for 1  i < j  m and ai 2 F so
m
i=1 |ai | = 1. Hence,
2
m 2 m
X X
2
kT v vk = ( i )ai ui = |( i )ai |2 .
i=1 i=1
P
If | i | ✏ for all 1  i  m then kT v vk2 ✏2 m i=1 |ai | = ✏ so kT v
2 2 vk ✏, a
contradiction. Thus, there must exists i so | i | < ✏.
exer:7A:3
exer:7B:13 13. According to exercise 3 (7A), if U is invariant under T then U ? is invariant under T ⇤ .
Hence, if T is normal then T ⇤ |U ? 2 L(U ? ) is normal.
We prove Complex Spectral Theorem by induction on dim V . In a complex inner product
space V , there always exists an eigenvector u 6= 0 corresponding to eigenvalue of T . Let
U = span (u) then since dim U ? < dim V , there exists an orthonormal basis e1 , . . . , en exer:7A:2
1
of U ? consisting of eigenvectors of T ⇤ |U ? . Since T is normal so according to exercise 2
(7A), e1 , . . . , en 1 is also orthonormal basis consisting of eigenvectors of T . Note that
eigenvalues of T of these eigenvectors must be different from , otherwise u 2 U ? which
means u = 0, a contradiction. Thus, V has orthonormal basis u, e1 , . . . , en 1 consisting of
eigenvectors of T .
exer:7B:14 14. One direction is true according to Real Spectral Theorem. If U has basis u1 , . . . , un
consisting of eigenvectors of T , consider inner product on U so hui , uj i = 0 for all 1 
i < j  n then this inner product makes T into a self-adjoint operator according to Real
Spectral Theorem.
exer:7B:15 15. It suffices to find a 2 R so

0 10 1 0 10 1
1 1 0 1 0 1 1 0 1 1 1 0
@0 1 1A @1 1 0A = @1 1 0A @0 1 1A .
1 0 a 0 1 a 0 1 a 1 0 a
9.5. 7C: Positive Operators and Isometries

Definition 9.5.1. An operator T 2 L(V ) is called positive if T is self-adjoint and hT v, vi 0
for all v 2 V .
Another terminology for the above definition is positive semidefinite. A positive semidefi-
nite operator that satisfies hT v, vi > 0 for all non-zero v is called positive definite.
theo:7.35:7C Theorem 9.5.2 (7.35, Positive Operators) Let T 2 L(V ). Then the following are equiva-
lent:
(a) T is positive;
(b) T is self-adjoint and all the eigenvalues of T are nonnegative;
(c) T has positive square root;
(d) T has self-adjoint square root;
(e) There exists an operator R 2 L(V ) such that T = R⇤ R.

theo:7.42:7C Theorem 9.5.3 (7.42, Isometries) Suppose S 2 L(V ). Then the following are equivalent:
(a) S is an isometry;
(b) hSu, Svi = hu, vi for all u, v 2 V ;
(c) Se1 , . . . , Sen is orthonormal for every orthonormal list of vectors e1 , . . . , en in V ;
(d) there exists an orthonormal basis e1 , . . . , en of V such that Se1 , . . . , Sen is orthonor-
mal;
(e) S ⇤ S = I;
(f) SS ⇤ = I;
(g) S ⇤ is an isometry;
(h) S is invertible and S 1 = S⇤.
theo:7.43:7C Theorem 9.5.4 (7.43) Suppose V is a complex inner product space and S 2 L(V ). Then
the following are equivalent:
(a) S is isometry.
(b) There is an orthonormal basis of V consisting of eigenvectors of S whose corresponding

eigenvalues all have absolute value 1.
9.6. Exercises 7C
exer:7C:1 1. Let e1 , e2 be orthonormal basis of V . Let T e1 = xe1 + ye2 and T e2 = ye1 + ze2 with
x, y, z 2 R so 2y + x + z < 0, x, z 0. Hence, T is self-adjoint and hT e1 , e1 i = x
0, hT e2 , e2 i = z 0. On the other hand,
hT (e1 + e2 ), e1 + e2 i = h(x + y)e1 + (y + z)e2 , e1 + e2 i = x + 2y + z < 0.
Thus, T is not positive.

exer:7C:2 2. We have
hT (v w), v wi = hw v, v wi = kv wk2 0.
Thus, we must have kv wk2 = 0 or v = w.
exer:7C:3 3. We have for every u 2 U then hT |U u, ui = hT u, ui 0 so T |U is positive operator on U .
exer:7C:4 4. For every v 2 V , we have hT ⇤ T v, vi = hT v, T vi 0 and (T ⇤ T )⇤ = (T ⇤ T so T ⇤ T is positive
operator on V . Similarly, T T ⇤ is positive operator on V .
exer:7C:5 5. If T, S is positive operators on V then for any v 2 V we have h(T + S)v, vi = hT v, vi +
hSv, vi 0 and (T + S)⇤ = T ⇤ + S ⇤ = T + S so T + S is positive operator on V .
theo:7.35:7C
exer:7C:6 6. If T is positive operator on V then according to theorem 9.5.2 (7C), T has self-adjoint
square root R 2 L(V ). Hence, we have Rk = (R⇤ )k = (Rk )⇤ so Rk is self-adjoint for every
theo:7.35:7C
positive integer k. Hence, T k = (R2 )k = (Rk )2 so again from theorem 9.5.2 (7C), T k is a
positive operator on V .
exer:7C:7 7. If hT v, vi > 0 for every v 2 V, v 6= 0 then if T is invertible (otherwise there exists v 2

V, v 6= 0 so T v = 0 which leads to hT v, vi = 0, a contradiction).
theo:7.35:7C
Conversely, if T is invertible. Since T is positive so according to theorem 9.5.2 (7C), there
exists operator R 2 L(V ) so R⇤ R = T . Hence, hT v, vi = kRvk2 . Since T is invertible so
Rv 6= 0 for all v 2 V , otherwise T v = R⇤ Rv = 0 a contradiction. Thus, kRvk2 > 0 or
hT v, vi > 0 for every v 2 V, v 6= 0.
exer:7C:8 8. It suffices to check the definiteness condition of inner product. We have hv, viT = 0 iff
exer:7C:7
hT v, vi = 0. Applying previous exercise 7 (7C), the condition hT v, vi = 0 for only v = 0
is equivalent to T being invertible.
exer:7C:9 9. Let S be a self-adjoint square root of identity operator I then according to Real/Complex
theo:7.24:7B
Spectral Theorem 9.3.1 (7B), V has an orthonormal basis e1 , e2 consisting of eigenvectors
of S corresponding to eigenvalues 1 , 2 . Thus, we have S 2 e1 = 21 e1 = Ie1 = e1 so 21 = 1.
Similarly, 22 = 1. Hence, identity operator on F2 has finitely many self-adjoint square
roots.
theo:7.42:7C
exer:7C:10 10. If (a) holds, i.e. S is isometry then from theorem 9.5.3 (7C), S ⇤ is also isometry which
leads to (b). If any of (b),(c) or (d) holds, we can follow that S ⇤ is isometry which can
theo:7.42:7C
leads back to other conditions according to theorem 9.5.3 (7C).
exer:5C:12
exer:7C:11 11. Similar to exercise 12 (5C). Since T1 is normal so according to Complex Spectral Theorem
theo:7.24:7B
9.3.1 (7B), there exists an orthonormal basis v1 , v2 , v3 consisting of eigenvectors 2, 5, 7 of T1 .
Similarly, there exists an orthonormal basis w1 , w2 , w3 consisting of eigenvectors 2, 5, 7 of
T2 . Let Sexer:5C:15
2 L(F3 ) so Sv1 = w1 , Sv2 = w2 and Sw3 = w3 then S is invertible. Similarly to
exercise 15 (5C), we can show that T1 = S 1 T S. Now it suffices to show that S is isometry.
P P3 2 P
Indeed, for any v 2 V, v = 3i=1 ai ei then kSvk2 = i=1 ai wi = 3i=1 |ai |2 = kvk2 .
theo:7.42:7C
Thus, S is isometry so according to theorem 9.5.3 (7C), S 1 = S ⇤ so T1 = S ⇤ T2 S.
exer:5C:13
exer:7C:12 12. It is essentially exercise 13 (5C).
exer:7C:13 13. It is false. Pick S 2 L(V ) for Sej = ej for all j 6= 2 and Se2 = x1 e1 +x2 e2 where x1 , x2 2 R
so |x1 |2 + |x2 |2 = 1 to guarantee kSe2 k = 1. Hence kS(e1 + e2 )k2 = k(x1 + 1)e1 + x2 e1 k2 =
|x1 + 1|2 + |x2 |2 = 2 + 2x1 6= 2 = ke1 + e2 k2 . Thus, S is not an isometry.
exer:7C:14 14. We’ve already shown that T is self-adjoint so T is also self-adjoint. We have
h( T ) sin ix, sin ixi = hsin ix, sin ixi > 0, h( T ) cos ix, cos ixi = hcos ix, cos ixi > 0.
exer:7A:21
And h( T )ei , ej i = 0 for i 6= j (see exercise 21 (7A) for more details). With this, we can
conclude that h( T )v, vi > 0 for all v 2 V , i.e. T is positive.
9.7. 7D: Polar Decomposition and Singular Value Decomposition

theo:7.45:7D Theorem 9.7.1 (7.45, Polar Decomposition)
p Suppose T 2 L(V ). Then there exists an
isometry S 2 L(V ) such that T = S T T .
⇤
theo:7.51:7D Theorem 9.7.2 (7.51 Singular Value Decomposition) Suppose T 2 L(V ) has singular val-
ues s1 , . . . , sn . Then there exist orthonormal bases e1 , . . . , en and f1 , . . . , fn of V such
that
T v = s1 hv, e1 i f1 + . . . + sn hv, en i fn
for every v 2 V .
theo:7.52:7D Theorem 9.7.3 (7.52) Suppose T 2 L(V ). Then the singular values of T are the nonnega-
tive square roots of the eigenvalues of T ⇤ T , with each eigenvalue repeated dim E( , T ⇤ T )
times.
How to find positive operator R if R2 is known:
• Since R2 is also positive so first is to find orthonormal basis e1 , . . . , en of V consisting of

eigenvectors of R2 (see 7B).
p
• If R2 ei = i ei then Rei = i ei . Hence,
n
! n
X X p
Rv = R hv, ei i ei = hv, ei i i ei .
i=1 i=1
How to find singular values of an operator T :
• Find T ⇤ (see 7A) then find T ⇤ T then find all eigenvalues of T ⇤ T . Hence, singular values
theo:7.52:7D
of T are nonnegative square roots of the eigenvalues of T ⇤ T according to theorem 9.7.3
(7D).
• If T is self-adjoint: The singular values of T equals the absolute values of the eigenvalues
exer:7D:10
of T according to exercise 10 (7D).
9.8. Exercises 7D
exer:7A:15
exer:7D:1 1. According to solution in exercise 15 (7A) then T ⇤ w = hw, xi u so T ⇤ T v = hv, ui T ⇤ x =
p kxk
hv, ui hx, xi u. One can verify that T ⇤ T v = kuk hv, ui u is indeed true.
exer:7D:2 2. p
Let T (a, b) = (5b, 0) then T ⇤ (x, y) = (0, 5x). Hence, T ⇤ T (x, y) = (0, 25y). Hence,
T ⇤ T (x, y) = (0, 5y) which has two singular values 0, 5 while 0 is the only eigenvalue
of T .
theo:7.45:7D
exer:7D:3 3. Apply Polar Decomposition 9.7.1 (7D) to T ⇤ we find there ⇣p exists
⌘⇤ an isometry S 2 L(V )
p p p
so T ⇤ = S (T ⇤ )⇤ T ⇤ = S T T ⇤ . Hence, T = (T ⇤ )⇤ = T T ⇤ S ⇤ = T T ⇤ S ⇤ because
p
T T ⇤ is positive. Note that S is an isometry so S ⇤ is also an isometry.
p
exer:7D:4 4. Since s is singular
p value of T so there exists v 2 V, kvk = 1 so T ⇤ T v = sv. Hence, we
have kT vk = k T T vk = |s|kvk = s.
⇤
exer:7D:5 5. We have
h(a, b), T ⇤ (x, y)i = hT (a, b), (x, y)i = h( 4b, a), (x, y)i = 4bx + ay = h(a, b), (y, 4x)i .
Hence, T ⇤ (x, y) = (y, 4x). Hence, T ⇤ T (x, y) = T ⇤ ( 4y, x) = (x, 16y). Thus, singular
values of T are 1, 4.
exer:7D:6 6. We have for all a2 , a1 , a0 2 R then
⌦ ↵ ⌦ ↵
a2 x2 + a1 x + a0 , T ⇤ (b2 x2 + b1 x + b0 ) = T (a2 x2 + a1 x + a0 ), b2 x2 + b1 x + b0 ,
⌦ ↵
= 2a2 x + a1 , b2 x2 + b1 x + b0 ,
Z 1
= (2a2 x + a1 )(b2 x2 + b1 x + b0 ) dx,
1
2
= (2a2 b1 + a1 b2 ) + 2a1 b0 ,
3 ✓ ◆
4 2
= a 2 · b1 + a 1 b2 + 2b0 .
3 3
On the other hand, we have
⌦ ↵ 2 2
a1 x2 + a1 x + a0 , c2 x2 + c1 x + c0 = 2a0 c0 + (a1 c1 + a2 c0 + a0 c2 ) + a2 c2 ,
✓ 3 ◆ 5
2 2 2
= a2 c2 + c0 + a1 · c1 + 2a0 c0 .
5 3 3
Thus, we must have c0 = 0, 43 b1 = 25 c2 + 23 c0 and 23 b1 + 2b0 = 23 c1 . This follows c0 = 0, c2 =
10
3 b1 and c1 = b1 + 3b0 . We obtain T (b2 x + b1 x + b0 ) = 3 b1 x + (b1 + 3b0 )x. Thus,
⇤ 2 10 2
T ⇤ T (a2 x2 + a1 x + a0 ) = T ⇤ (2a2 x + a1 ) = (2a2 + 3a1 )x.

Thus, T ⇤ T has eigenvalues 3pwhich repeated 1 time and 0 which repeated 2 times. Thus,
theo:7.52:7D
singular values of T are 0, 0, 3 according to theorem 9.7.3 (7D).
exer:7D:7 7. We have
h(z1 , z2 , z3 ), T ⇤ (x1 , x2 , x3 )i = hT (z1 , z2 , z3 ), (x1 , x2 , x3 )i ,
= h(z3 , 2z1 , 3z2 ), (x1 , x2 , x3 )i ,
= h(z1 , z2 , z3 ), (2x2 , 3x3 , x1 )i .
Thus, Tp⇤ (x1 , x2 , x3 ) = (2x2 , 3x3 , x1 ). Hence, T ⇤ T (z1 , z2 , z3 ) = T ⇤ (z3 , 2z1 , 3z2 ) = (4z1 , 9z2 , z3 ).
Thus, T ⇤ T (z1 , z2 , z3 ) = (2z1 , 3z2 , z3 ). Let S 2 L(V ) so S(2z1 , 3z2 , z3 ) = (z3 , 2z1 , 3z2 )
then S is an isometry.
p
exer:7D:8 8. Since T = SR so T ⇤ = (SR)⇤ = R⇤ S ⇤ = RS ⇤ so T ⇤ T = RS ⇤ SR = R2 so R = T ⇤ T .
p p p
exer:7D:9 9. Since kT p vk = k T T vk so T is invertible iff T T is invertible
⇤ ⇤ iff null T ⇤ T = {0} iff
range T ⇤ T = V . Thus, it suffices p to show that range T ⇤ T = V iff there exists a unique
isometry S 2 L(V ) so T = S T T . ⇤
p
Indeed, if range T ⇤ T = p V then obviously isometry operator S is uniquely determined
since Spmust satisfy T =pS T ⇤ T . Conversely, if there exists p a unique isometry S 2 L(V ) so
T = S T ⇤ T but range T ⇤ T 6= V ,theo:7.45:7D
which means range T ⇤ T )? 6= {0}. Hence, by looking
back proof of Polar Decomposition 9.7.1 (7D) p from the book, for m ? 1, let e1 , . . . , em and
f1 , . . . , fm be orthonormal bases of (range T ⇤ T )? and (range p T ) , respectively. Hence,
we have at least two ways to choose linear map S2 : (range T ⇤ T )? ! (range T )? , one is
S2 (a1 e1 + . . . am em ) = a1 f1 + . . . + am fm or S2 (a1 e1 + . . . , am fm ) p
= a1 f1 . . . am fm .
This follows there are at least p two possible isometries S so T = S T ⇤ T , a contradiction.
Thus, we must have range T ⇤ T = V or T is invertible.
theo:7.52:7D
exer:7D:10 10. According to theorem 9.7.3 (7D), the singular values of T are the nonnegative square roots
of the eigenvalues of T ⇤ T . Since T is self-adjoint so T ⇤ T = T 2 so eigenvalues of T are the
nonnegative square roots of the eigenvalues of T 2 . Thus, the singular values of T equal
the absolute values of the eigenvalues of T , repeated appropriately.
exer:7D:11 11. It suffices to show that T ⇤ T and T T ⇤ have the same eigenvalues values that are repeated
under same number of times. Since T ⇤ T is self-adjoint so according to Spectral Theorem,
there exists an orthonormal basis e1 , . . . , en consisting of eigenvectors of T ⇤ T . WLOG,
among these vectors, say f1 , . . . , fm are eigenvectors of T ⇤ T corresponding to eigenvalue
. Note that m = dim E( , T ⇤ T ).
One the other hand, T T ⇤ (T fi ) = T (T ⇤ T fi ) = T fi so T f1 , . . . , T fm are eigenvectors
of T T ⇤ corresponding to eigenvalues . For any 1  i < j  m then hT fi , T fj i =
hfi , T ⇤ T fj i = hfi , fj i = 0 so T f1 , . . . , T fm is also an orthonormal list which means
T f1 , . . . , T fm is linearly independent. This follows that eigenvalue under operator T T ⇤
is repeated at least dim E( , T ⇤ T ) times, i.e. dim E( , T T ⇤ ) dim E( , T ⇤ T ). Similarly
and we obtain dim E( , T T ⇤ ) = dim E( , T ⇤ T ) and can conclude that T T ⇤ and T ⇤ T have
same eigenvalues that are repeated same number of times. Thus, T and T ⇤ have same
singular values.
theo:7.52:7D
exer:7D:12 12. The statement is false. According to theorem 9.7.3 (7D), the singular values of T 2 are
the nonnegative square roots of the eigenvalues of (T 2 )⇤ T 2 = (T ⇤ )2 T 2 and the singular
values of T are the nonnegative square roots of the eigenvalues of T ⇤ T . Hence, the singular
values of T 2 equal the squares of the singular values of T when the eigenvalues of (T ⇤ )2 T 2
equals the squares of the eigenvalues of T ⇤ T . Note that (T ⇤ )2 T 2 and T ⇤ T are self-adjoint
operators so we must have (T ⇤ T )2 = (T ⇤ )2 T 2 , which is incorrect when T is not normal.
Choose a nonnormal operator T in 2-dimensional vector space V as a counterexample.
p p
exer:7D:13 13. Since for all v 2 V then
p kT vk = k T T vk so T is invertible iff T T is invertible iff 0 is
⇤ ⇤
not an eigenvalue of T ⇤ T iff 0 is not s singular value of T .

p p p
exer:7D:14 14. Again, from kTp vk = k T ⇤ T vk so dim range T = dim range
p p T ⇤ T and dim range T ⇤T =
n dim E(0, T T ) from the fact that T T is positive sop T T is diagonalizable. Note
⇤ ⇤ ⇤
that according to the definition of singular values, dim E(0, pT ⇤ T ) is number of repetition
of singular value 0 of T . Thus, dim range T = n dim E(0, T ⇤ T ) equals the number of
nonzero singular values of T .
theo:7.51:7D
exer:7D:15 15. If S has singular values s1 , . . . , sn then according to theorem 9.7.2 (7D), there exist
orthonormal bases e1 , . . . , en and f1 , . . . , fn of V such that Sv = s1 hv,
P e1 i f 1 + . . . +
sn hv, en i fn for every v 2 V . This follows for every v 2 V , kSvk2 = ni=1 |si hv, ei i |2 .
Hence, if all singular values of S equal 1 then kSvk = kvk so S is an isometry. Conversely,
if S is an isometry then kSei k = si = 1 for all 1  i  n so all singular values of S equal
1.
exer:7D:16 16. Note that with T1 exer:7C:11
= S1 T2 S2 then T1⇤ T = S2⇤ (T2⇤ T2 )S2 . The idea to solve this problem is
similar to exercise 11 (7C).
If T1 , T2 have same singular values then two positive operators T1⇤ T1 and T2⇤ T2 have same
eigenvalues 1 , . . . , n . According to Spectral Theorem, let e1 , . . . , en be orthonormal
basis consisting of eigenvectors of T1⇤ T1 and f1 , . . . , fn be orthonormal basis consisting of
eigenvectors of T2⇤ T2 . Let S2 2 L(V ) so S2 ei = fi for all 1  i  n then similarly to
exer:7C:11
exercise 11 (7C), we can show that S2 is an isometry and that T1⇤ T1 = S2⇤ (T2⇤ T2 )S2 . This
leads to kT1 vk = kT2 S2 vk for every v 2 theo:7.45:7D
V . With this, we can construct isometry S1
similarly to proof of Polar Decomposition 9.7.1 (7D) from the book so that S1 T2 S2 = T1 .
Conversely, if there exists isometries S1 , S2 so S1 T2 S2 = T1 then T1⇤ T1 = S2⇤ T2⇤ T2 S2 . If
f1 , f2 , . . . , fn is basis of V and are also eigenvectors of T2⇤ T2 corresponding to eigenvalues
1 , . . . , n . Since S2 is invertible, for each fi , there exists ei 2 V so S2 ei = fi . This follows
e1 , . . . , en is basis of V . Hence,
T1⇤ T1 ei = S2⇤ T2⇤ T2 S2 ei = S2⇤ T2⇤ T2 fi = S2⇤ fi = i ei .
This follows T1⇤ T1 and T2⇤ T2 have same eigenvalues so T1 and T2 have same singular values.
exer:7D:17 17. (a) We have
hT v, ui = hs1 hv, e1 i f1 + . . . + sn hv, en i fn , hu, f1 i f1 + . . . + hu, fn i fn i ,

= s1 hv, e1 i hu, f1 i + . . . + sn hv, en i hu, fn i ,
= hv, s1 hu, f1 i e1 + . . . + sn hu, fn i en i .
This follows T ⇤ u = s1 hu, f1 i e1 + . . . + sn hu, fn i en .

(b) We have
T ⇤ T v = T ⇤ (s1 hv, e1 i f1 + . . . + sn hv, en i fn ) ,

Xn
= si hv, ei i T ⇤ fi ,
i=1
= s21 hv, e1 i e1 + . . . + s2n hv, en i en .
p
(c) We have T ⇤ T ei = si ei so
p p
T ⇤ T v = T ⇤ T (hv, e1 i e1 + . . . + hv, en i en ) ,
= s1 hv, e1 i e1 + . . . + sn hv, en i en .
ei
(d) We have T ei = si fi so T 1f
i = si . Hence,
1
T v = T 1 (hv, f1 i f1 + . . . + hv, fn i fn ) ,
Xn
= hv, fi i T 1 fi ,
i=1
hv, f1 i e1 hv, fn i en
= + ... + .
s1 sn
P
exer:7D:18 18. (a) If s1  s2P · · ·  sn are all singular values of T then we have kT vk2 = ni=1 s2i | hv, ei i |2
and kvk2 = ni=1 | hv, ei i |2 so it can be easily seen that s1 kvk  kT vk  sn kvk.
(b) Applying (a) we obtain s1  | |  sn .
exer:7D:19 19. It suffices to show that for every " > 0 there exists > 0 so for all ku vk < then exer:7D:18
kT (u v)k < ". Indeed, if s the largest singular value of T then from previous exercise 18
(7D), we find kT (u v)k  sku vk. Hence, if we choose = "/s, we are done. Thus, T
is uniformly continuous with respect to the metric d on V .
exer:7D:18
exer:7D:20 20. According to exercise 18 (7D), for any v 2 V , we have k(S + T )vk  kSvk + kT vk 
exer:7D:4
(t + s)kvk. On the other hand, from exercise 4 (7D), there exists v 2 V, kvk = 1 so
k(S + T )vk = r so this follows r  t + s.
10. Chapter 8: Operators on Complex Vector Spaces

In this chapter, V denotes a finite-dimensional nonzero vector space over F.
10.1. 8A: Generalized Eigenvectors and NilpotentOperators

theo:8.2:8A Theorem 10.1.1 (8.2) Suppose T 2 L(V ). Then
{0} = null T 0 ⇢ null T 1 ⇢ · · · ⇢ null T k ⇢ null T k+1 ⇢ · · ·
theo:8.3:8A Theorem 10.1.2 (8.3) Suppose T 2 L(V ). Suppose m is a nonnegative integer such that
null T m = null T m+1 . Then
null T m = null T m+1 = null T m+2 = null T m+3 = . . .
theo:8.4:8A Theorem 10.1.3 (8.4) Suppose T 2 L(V ). Let n = dim V . Then null T k = null T k+1 for
all positive integer k n.
theo:8.5:8A Theorem 10.1.4 (8.5) Suppose T 2 L(V ). Let n = dim V . Then V = null T n range T n .
theo:8.13:8A Theorem 10.1.5 (8.13) Let T 2 L(V ). Suppose 1 , . . . , m are distinct eigenvalues of
T and v1 , . . . , vm are corresponding generalized eigenvectors. Then v1 , . . . , vm is linearly
independent.
theo:8.18:8A Theorem 10.1.6 (8.18) Suppose N 2 L(V ) is nilpotent. Then N dim V = 0.
theo:8.19:8A Theorem 10.1.7 (8.19) Suppose N is a nilpotent operator on V . Then there exists a basis
of V with respect to which the matrix of N is an upper-triangular matrix with only 0’s in
the diagonal.
How two determine all the generalized eigenspaces corresponding to distinct eigenvalues of
operator T 2 L(V ):
1. Find all eigenvalues of T .
2. For each eigenvalue of T , find null space of (T I)dim V , which is equal to generalized
eigenspace G( , T ) of T corresponding to .
10.2. Exercises 8A
exer:8A:1 1. 0 is the only eigenvalue of T . We have T 2 (w, z) = T (z, 0) = 0 for all (w, z) 2 C2 so
G(0, T ) = null T 2 = C2 .
exer:8A:2 2. If T (w, z) = (w, z) = ( z, w) then w = z, z = w. This follows 2 + 1 = 0 so
eigenvalues of T are = ±i. We have
(T + iI)2 (w, z) = (T + iI)( z + iw, w + iz),
=( w iz, z + iw) + i( z + iw, w + iz),
= ( 2w 2iz, 2iw 2z).
Thus, (T + iI)2 (w, z) = 0 when w + iz = 0. Thus, G( i, T ) = span ( i, 1). We have
(T iI)2 (w, z) = (T iI)( z iw, w iz),
= (iz w, z iw) i( z iw, w iz),
= (2iz 2w, 2z 2iw).
Thus, (T iI)2 (w, z) = 0 when iz w = 0 so G(i, T ) = span (i, 1).
exer:8A:3 3. Let dim V = n. If v 2 G( , T ) then (T I)n v = 0. Hence, ( T ) n (T I)n v = 0 so
( 1 I T ) v = 0 so v 2 G(
1 n 1 , T ). Similarly, if v 2 G(
1 1 , T ) then v 2 G( , T ).
1
Thus, G( , T ) = G( , T ).
1 1
theo:8.13:8A
exer:8A:4 4. The proof for this is completely similar to proof of theorem 10.1.5 (8A). Indeed, assume the
contrary there is v 2 G(↵, T ) \ G( , T ), v 6= 0 then let k be the largest nonnegative integer
so (T ↵I)k v 6= 0. Hence, with w = (T ↵I)k v we have (T ↵I)w = (T ↵I)k+1 v = 0
so T w = ↵w. This follows (T I)w = (↵ )w for any 2 F. Hence, we obtain,
(T I)n w = (↵ )n w. On the other hand, since v 2 G( , T ) so (T I)n w =
(T ↵I)k (T I)n v = 0. This follows (↵ )n w = 0 which leads to w = 0, a contradiction.
Thus, G(↵, T ) \ G( , T ) = {0}.
exer:8A:5 5. Let a0 , . . . , am 1 2 F so a0 v + a1 T v + . . . + am 1 T m 1 v = 0. Apply the the operator
T m 1 to both side of this we obtain a0 T m 1 v = 0 which leads to a0 = 0. Similarly,
we apply T m 2 to obtain a1 = 0, ..., T to obtain am 2 = 0. Thus, ai = 0 for all i so
v, T v, . . . , T m 1 v is linearly independent.
exer:8A:6 6. Since T (z1 , z2 , z3 ) = (z2 , z3 , 0) so T 2 theo:8.18:8A
= 0. If T has square root when S 4 = T 2 = 0 so S
is nilpotent so according to theorem 10.1.6 (8A) then S 2 = 0 or T = 0, a contradiction.
Thus, T has no square root.
theo:8.18:8A
exer:8A:7 7. If N is nilpotent then from theorem 10.1.6 (8A), N dim V = 0 or G(0, N ) = null N dim V = V
so there exists a basis ofexer:8A:4
V consisting of generalized eigenvectors corresponding to 0.
Combining with exercise 4 (8A), we find 0 is the only eigenvalue of N .
exer:8A:8 8. It is false. When V = C2 , consider to nilpotent operators T, S 2 L(C2 ) so T (x, y) = (0, x)
and S(x, y) = (y, 0). Then (T + S)2 (x, y) = (T + S)(y, x) = (x, y). Thus, T + S is not a
nilpotent operator. Thus, set of nilpotent operators on C2 is not a subspace of L(C2 ).
exer:8A:9 9. Since ST is nilpotent so ST dim V = 0 which leads to T S dim V +1 = T (ST )dim V S = 0 so

T S is nilpotent.
exer:8A:10 10. It suffices to show that null T n 1 \ range T n 1 = {0}. Assume the contrary, if there
exists v 2 null T n 1 \ range T n 1 , v 6= 0 then T n 1 v = 0 and there exists utheo:8.4:8A
2 V, u 6= 0
so T n 1 u = v. This follows T 2n 2 u = 0 so T n u = 0 according exer:8A:5
to theorem 10.1.3 (8A).
Note that v = T n 1 u 6= 0 and T n u = 0 so according to exercise 5 (8A), u, T u, . . . , T n 1 u
is linearly independent so it is basis of V . On the other hand, not that T i u 2 null T n
so that means T n = 0, a contradiction since T is not nilputent. Thus, we must have
null T n 1 \ range T n 1 = {0}, which leads to V = null T n 1 range T n 1 .
exer:8A:11 11. It is false. The idea is to pick T so T 2 doesn’t have enough eigenvectors to form a basis
of V . Let T 2 L(C2 ) so T (x, y) = (x, x + y) then T 2 (x, y) = T (x, x + y) = (x, 2x + y).
Hence, if (x, y) = (x, 2x + y) then ( 1)x = 0 and ( 1)y = 2x. With this, we obtain
1 is the only eigenvalue of T and that E(1, T ) = span {(0, 1)}. This follows T 2 is not
2 2
diagonalizable.
exer:8A:12 12. Let v1 , . . . , vn be such basis of V . We prove inductively on i  n that there exists positive
integer ki so T ki vi = 0. Indeed, since T v1 = 0 so it’s true for i = 1. If it’s true for
all i  m < n. Consider vm+1 then according to the assumption, we have T vm+1 2
span (v1 , . . . , vm ) so with k = max{k1 , . . . , km } we obtain T k+1 vm+1 = 0. Thus, this
follows N is nilpotent.
exer:8A:13 13. Since N is nilpotent so there exists a basis of V with respect to which the matrix of theo:6.37:6B
N is
an upper triangular matrix with only 0’s in the diagonal. Therefore, from theorem 8.3.4
(look at the proof of it, because only apply the theorem itself doesn’t help), there exists
an orthonormal basis of V with respect to which the matrix of N is an upper triangular
matrix with only 0’s in the diagonal. With this and the from the proof of Complex Spectral
theo:7.24:7B
Theorem 9.3.1 given that N is normal, we can deduce that matrix of N is the 0 matrix,
in other words N = 0.
exer:8A:13
exer:8A:14 14. (copy from part of exercise 13 (8A)) Since N is nilpotent, there exists a basis of V with
respect to which the matrix of N theo:6.37:6B
is an upper triangular matrix with only 0’s in the
diagonal. Therefore, from theorem 8.3.4 (look at the proof of it, because only apply the
theorem itself doesn’t help), there exists an orthonormal basis of V with respect to which
the matrix of N is an upper triangular matrix with only 0’s in the diagonal.
exer:8A:15 15. Since null N dim V 1 6= null N dim V exer:8A:5

so there exists v 2 V so N dim V v = 0 but N dim V 1 v 6=
0. Therefore, according to exercise 5 (8A), v, T v, . . . , T dim V 1 v is linearly independent so
it is basis of V . Note that T i v 2 null T dim V so that means V ⇢ null T dim V so T dim V = 0
so T is nilpotent.
theo:8.3:8A
Since null N dim V 1 6= null N dim V so according to theorem 10.1.2 (8A), we obtain null N j 6=
theo:8.2:8A
null N j+1 for all 0  j  dim V 1. Hence, from theorem 10.1.1 (8A), we obtain
dim null N j < null N j+1 for every 0  j  dim V 1, which leads to dim null N i = i for
every 1  i  dim V .
exer:8A:16 16. For any nonnegative integer k, if v 2 range T k+1 then v 2 range T k so range T k+1 ⇢
range T k .
exer:8A:16
exer:8A:17 17. From exercise 16 (8A), we already know range T m+k+1 ⇢ range T m+k . Now, if v 2
range T m+k then v = T m+k u. Since T m u 2 range T m = range T m+1 so there exists w 2 V
so T m u = T m+1 w so v = T m+k u = T m+k+1 w so v 2 range T m+k+1 . Thus, we obtain
range T m+k ⇢ range T m+k+1 . With this, we conclude range T m+k = range T m+k+1 for
every nonnegative integer k, in other words range T k = range T m for all k > m.
exer:8A:17
exer:8A:18 18. According to exercise 17 (8A), it suffices
exer:8A:16
to prove range T n = range T n+1 . Assume it is
exer:8A:17
not true, then from exercises 16 and 17 (8A), we find
V = range T 0 ) range T 1 ) · · · ) range T n+1 .
This follows dim range T n+1 < 0, a contradiction. Thus, we must have range T n =
range T n+1 .
exer:8A:19 19. null T m = null T m+1 iff dim null T m = dim null T m+1 iff dim range T m = dim range T m+1
iff range T m = range T m+1 .
exer:8A:19 exer:8A:15
exer:8A:20 20. From exercise 19 (8A), range T 4 6= range T 5 follows null T 4 6= null T 5 so from exercise 15
(8A), T is nilpotent.
exer:8A:21 21. W 2 Cn so T (z1 , . . . , zn ) = (z2 , . . . , zn , 0) then null T i = {(z1 , . . . , zi , 0, . . . , 0) : zi 2 C}.
10.3. 8B: Decomposition of an Operator

theo:8.21:8B Theorem 10.3.1 (8.21, Description of operators on complex vector spaces) Suppose V is
complex vector space and T 2 L(V ). Let 1 , . . . , m be the distinct eigenvalues of T .
Then:
(a) V = G( 1, T ) ··· G( m , T );
(b) each G( i , T ) is invariant under T ;
(c) each (T i I)G( i ,T )

is nilpotent.
exer:5C:5
The idea for the proof of (a) is similar to of exercise 5 (5C).
theo:8.5:8A
Proof of (a). We induct on dim V . From theorem 10.1.4 (8A) then V = null (T i I)
n
range (T i I) for all 1  i  m. Let U = range (T

n
i I) then dim U < dim V .
n
First, we show that for all j 6= i then G( j , T ) = G( j , T |U ). It’s obvious that G( j , T |U ) ⇢

G( j , T ). If v 2 G( j , T ) then (T j I) v = 0. With V = null (T
n
i I)
n range (T
i I) , we
n
can write v = u+(T i I) w where u 2 G( i , T ) so this follows (T

n n
j I) u = 0 (T
n
j I) (T
I) n w 2 U so (T I) n u 2 U . This follows there exists s 2 V so (T I) n u = (T n
i j j i I) s.
Note that u 2 G( i , T ) so that meanstheo:8.4:8A
(T 2n
i I) s = (T
n
j I) (T i I) u = 0 which leads to
n
i I) s = 0 according to theorem 10.1.3 (8A) or (T j I) u = 0 or u 2 G( j , T ). Therefore,

(T n n
exer:8A:4
u 2 G( i , T ) \ G( j , T ) so from exercise 4 (8A) we find u = 0. Thus, v = (T i I) w 2 U . We
n
conclude that any generalized eigenvector v of T corresponding to j 6= i is in U , which means

v 2 G( j , T |U ). With this, we deduce, G( j , T ) ⇢ G( j , T |U ) so G( j , T ) = G( j , T |U ) for all
j 6= i.
Second, note that j (j 6= i) are the only eigenvalues of operator T |U over U . Hence, since
dim U < dim V so from inductive hypothesis, we obtain
M M
n
range (T i I) = U = G( j , T |U ) = G( j , T ).
j6=i j6=i
Lm
Thus, V = G( i , T ) range (T i I)
n = j=1 G( j , T ).
theo:8.26:8B Theorem 10.3.2 (8.26) Suppose V is a complex vector space and T 2 L(V ). Then the
sum of multiplicities of all eigenvalues of T equals dim V .
theo:8.29:8B Theorem 10.3.3 (8.29) Suppose V is a complex vector space and T 2 L(V ). Let 1 , . . . , m
be the distinct eigenvalues of T , with multiplicities d1 , . . . , dm . Then there is a basis of V
with respect to which T has a block diagonal matrix of the form
0 1
A1 0
B .. C
@ . A,
0 Am
where each Aj is a dj -by-dj upper triangular matrix of the form

0 1
j ?
B .. C
Aj = @ . A.
0 j
theo:8.31:8B Theorem 10.3.4 (8.31) Suppose N 2 L(V ) is nipotent. Then I + N has kth root.
theo:8.33:8B Theorem 10.3.5 (8.33) Suppose V is a complex vector space and T 2 L(V ) is invertible.
Then T has a kth root.
10.4. Exercises 8B
exer:8B:1 1. Since 0 is the only
theo:5.27:5B
eigenvalue of N and that V is a complex vector space, from theorem
theo:5.32:5B
7.3.2 (5B) and 7.3.4 (5B), we follow there exists a basis of V with respect to which the
matrix of N is an upper-triangular matrix with only 0’s on the diagonal. From exercise
exer:8A:12
12 (8A), we follow N is nilputent.
exer:8B:2 2. The idea is choose operator T so T has 0 as the only eigenvalue and that there doesn’t
exists any basis of V with respect to which T has upper-triangular matrix. Hence:
• First pick null T = span (v1 ).
• Choose U = range T so T |U 2 L(U ) does not have any eigenvalue. Let dim U = 2
then say T v2 = v3 , T v3 = v2 , we can find that T |U does not have any eigenvalue.
• Confirm that T is not nilpotent. Now we have V = span (v1 , v2 , v3 ) so it suffices to
show T 3 6= 0. Indeed, we have
T 3 (av1 + bv2 + cv3 ) = T 2 (bv3 cv2 ) = T ( bv2 cv3 ) = bv3 + cv2 .
exer:8B:3 3. If is eigenvalue of T and v 2 E( , T ) then since S is invertible, there exists u 2 V so

Su = v. This follows S 1 T Su = S 1 T v = S 1 v = u so is also eigenvalue of T .
Similarly, if is eigenvalue of S 1 T S and v 2 E( , S 1 T S) then T Sv = Sv so is also
eigenvalue of T . Thus, T and S 1 T S have same eigenvalues.
Now it suffices to show that exer:5B:5
dim G( , T ) = dim G( , S 1 T S) for every eigenvalue of
. Note that from exercise 5 (5B) we have (S 1 T S I)n = S 1 (T I)n S. Let
v1 , . . . , vm be basis of G( , S T S) then (S T S
1 1 n
I) vi = S (T1 I)n Svi = 0
so (T I) Svi = 0. This follows Svi 2 G( , T ) and also observe that Sv1 , . . . , Svm
n
is linearly independent so that means dim G( , S 1 T S)  dim G( , T ). Similarly, if

Sv1 , . . . , Svm is basis of dim G( , T ) then v1 , . . . , vm 2 G( , S 1 T S) and v1 , . . . , vm is
linearly independeny. With this, we obtain dim G( , T )  dim G( , S 1 T S). Hence, we
obtain dim G( , T ) = dim G( , S 1 T S).
theo:8.3:8A
exer:8B:4 4. If null T n 2 6= null T n 1 then from theorem 10.1.2 (8A), we find null T i 6= null T i+1 for
all 0  i  n 2. This follows dim G(0, T ) n 1, i.e. the multiplicity of 0 of T is at
theo:8.26:8B
least n 1. Hence, from theorem 10.3.2 (8B), T has at most 2 eigenvalues.
exer:8B:5 5. If every generalized eigenvector of T is an eigenvector of T then obviously V is has basis

consisting of eigenvectors of T .PFor the other direction,
Pmif V has basis consisting of eigen-
m
vectors of T then that means i=1 dim E( i , T ) = i=1 dim G( i , T ) and the equality
holds when E( i , T ) = G( i , T ) for every 1  i  m, i.e. every generalized eigenvector of
T is an eigenvector of T .
theo:8.31:8B
exer:8B:6 6. We have N 5 = 0 so N is nilpotent. According to proof of theorem 10.3.4 (8B), let
R = I + a1 N + a2 N 2 + a3 N 3 + a4 N 4 be the square root of N + I then
R2 = I + 2a1 N + (a21 + 2a2 )N 2 + (2a3 + 2a1 a2 )N 3 + (2a4 + 2a3 a1 + a22 )N 4 .
We find a1 = 1/2, a2 = 1/8, a3 = 1/16, a4 = 5/128.
exer:8B:7 7. First is show thattheo:8.33:8B

I + N where N is nilpotent has a cube root then follow similarly to
proof of theorem 10.3.5 (8B).
exer:8B:8 8. If 3, 8 are the only eigenvalues of T then null T n

exer:8A:19
2 = null T 0 = {0} so according to exercise
19 (8A), range T n 2 = range T 0 = V .
exer:8B:4
If T has at least 3 eigenvalues then from exercise 4 (8B), we obtain null T n 2 = null T n 1 =
exer:8A:19
null T n so from exercise 19 (8A) again, we have range T n 2 = range T n 1 = range T n .
theo:8.5:8A
Thus, from theorem 10.1.4 (8A), we find V = null T n range T n = null T n 2 range T n 2 .
exer:8B:9 9. Not hard.

theo:8.29:8B
exer:8B:10 10. Since T is an operator on a complex vector space, from theorem 10.3.3 (8B), there exists
a basis of V with respect to which T has a block diagonal matrix.
Let Di be a diagonalizable dj -by-dj matrix with its diagonal same as matrix Aj . Then let
D 2 L(V ) so matrix of D is 0 1
D1 0
B .. C
@ . A.
0 Dm
Let Ni = A Di = i Ii where Ii is di -by-di identity matrix. Let N 2 L(V ) so matrix of
N is 0 1 0 1
N1 0 1 I1 0
B .. C B .. C
@ . A=@ . A
0 Nm 0 m Im
exer:8B:9
From here, we obtain N is nilpotent, T = N + D and from exercise 9 (8B), we have matrix
of N D is
0 10 1 0 1
1 I1 0 D1 0 1 D1 0
B .. CB .. C B .. C
@ . A@ . A=@ . A,
0 m Im 0 Dm 0 m Dm
which is also equal to matrix of DN . Thus, N D = DN .
exer:8B:11 11. First we show that number of times that 0 appears on the diagonal of the matrix of T is
at most dim null T n . For a basis of V with respect to which T has an upper-triangular
matrix, let va1 , . . . , vam be subset of that basis so the i-th 0 on the diagonal (from top left
to bottom right) is also in ai -th column of matrix of T .
We prove inductively on i that there exist ui 2 span (vP 1 , . . . , vai ) so T u1 = 0 and T ui 2
a1 1
span (u1 , . . . , ui 1 ) for i 2. For i = 1, let u1 = i=1 ↵i vi + va1 . Since T va1 2
span (v1 , . . . , va1 1 ) so we can write
aX
! a 1
1 1 X
1 aX
1 1
T (u1 ) = T ↵i vi + v
i i = (↵i T vi + i vi ) .
i=1 i=1 i=1
I suffices to choose ↵i so T (u1 ) = 0. Indeed, observe that the first a1 1 numbers

P on the
diagonal (from top left) are nonzero, i.e. for every 1  i  a1 1 then T vi = ij=1 cj,i vj
with cj,i 6= 0 for all 1  j  i. Hence,

0 1
aX
1 1 aX
1 1
T (u1 ) = @ ↵j ci,j + i A vi .
i=1 j=i
Therefore, since ci,j 6= 0 for all 1  i  a1 1, 1  j P  i, we can choose ↵a1 1 so

↵a1 1 ca1 1,a1 1 + a1 1 = 0, then we can choose ↵a1 2 so aj=a 1 1
1 2
↵j ca1 2,j + ai 2 = 0.
Continuing this, we guarantee to find u1 so T u1 = 0.
P i 1
If the statement is true for all numbers less than i, consider ui = vai + aj=1 ↵j vj . We
show that we can find ↵j so T ui 2 span (u1 , . . . , ui 1 ). Indeed, we have
0 1 0 1
ai 1 aX
i 1
X
T ui = T @ ↵j vj A + T @ ↵ j v j A + T v ai .
j=1 j=ai 1 +1
Observe that on the diagonal, there is no 0 between ai 1 -th number and ai -th number, so
with similar approach to case i = 1, we can choose ↵ai 1 +1 , . . . , ↵ai 1 so scalar multipli-
cation of vj (ai 1 + 1  j  ai 1) in representation of T ui is 0. With this and note that
ui 1 2 span (v1 , . . . , vai 1 ), we obtain:
0 1
ai 1 ai 1
X X
T ui = T @ A
↵j vj + i vi ,
j=1 i=1
0 1
ai ai
1 1
X1 X
=T@ ↵j vj A + ki vi + i 1 ui 1 ,
j=1 i=1
0 1 0 1
ai ai ai
1 1
X2 X1 X
=T@ ↵j vj A + T @ ↵j vj A + ki vi + i 1 ui 1 .
j=1 j=ai 2 +1 i=1
where ki 2 F. Similarly, we can choose ↵ai 2 +1 , . . . , ↵ai 1 so scalar multiplications of

vj (ai 2 + 1  j  ai 1 1 in T ui are 0. With this, we obtain:
0 1
ai 2 ↵i 2 1
X X
T ui = T @ A
↵j vj + ` i v i + i 2 ui 2 + i 1 ui 1 .
j=1 i=1
By continuing this method, we can deduce T ui 2 span (u1 , . . . , ui 1 ).
From the above, since T u1 = 0 and T ui 2 span (u1 , . . . , ui 1 ), we obtain u1 , . . . , um 2

null T n . On the other hand, we also find that u1 , . . . , um is linearly independent so that
means m  dim nullT n .
Back to our original question, for i as an eigenvalue of T , M(T ) is an upper-triangular
matrix with respect to some basis of V then M(T i I) is also an upper-triangular
matrix. Furthermore, number of times i appears on the diagonal of M(T ) (denote this
as Si ) equals to number of times 0 appears on the diagonal of M(T i I), which is at

most dim null (T i I) times according P
n to what we’ve
P proven above. We write Si 
dim null (T i I) n . On the other hand, S
i i = i dim null (T i I) = n so that
n
means Si = dim G( i , T ) for all i. We’re done.
10.5. 8C: Characteristic and Minimal Polynomials

Definition 10.5.1 (8.34, characteristic polynomial). Suppose V is a complex vector space and
T 2 L(V ). Let 1 , . . . , m denote the distinct eigenvalues of T , with multiplicities d1 , . . . , dm .
The polynomial (z 1)d1 · · · (z m)
dm is called the characteristic polynomial of T .
37:8C:cayley_hamilton Theorem 10.5.2 (Cayley-Hamilton Theorem) Suppose V is a complex vector space and
T 2 L(V ). Let q denote the characteristic polynomial of T . Then q(T ) = 0.
Definition 10.5.3 (8.43, minimal polynomial). Suppose T 2 L(V ). Then the minimal popy-
nomial of T is the unique monic polynomial p of smallest degree such that p(T ) = 0.
theo:8.46:8C Theorem 10.5.4 (8.46) Suppose T 2 L(V ) and q 2 P(F). Then q(T ) = 0 if and only if q
is a polynomial multiple of the minimal polynomial of T .
theo:8.46:8C theo:8.37:8C:cayley_hamilton
With F = C, theorem 10.5.4 and Cayley-Hamilton Theorem 10.5.2 follows that the charac-
teristict polynomial of T is a polynomial multiple of the minimal polynomial of T .
theo:8.49:8C Theorem 10.5.5 (8.49) Let T 2 L(V ). Then the zeros of the minimal polynomial of T
are precisely the eigenvalues of T .
Not from the book:
theo:3:8C Theorem 10.5.6 Let p(z) = (z d1

1 ) · · · (z m)
dm be the minimal polynomial of T
then G( i , T ) = null (T i I) for all 1  i  m. In particular, di is the smallest number

di
so null (T i I) = null (T
d i
i ) , or we can say that (z
n
i ) is the minimal polynomial
di
of T |G( i ,T ) .
exer:8C:12
The idea for the proof is similar to proof of exercise 12 (8C).
How to find characteristic/minimal polynomial of an operator.
Good sum-
mary ques-
Question 10.5.7. What can you tell about an operator from looking at its minimal/characteristic tion
polynomial?
10.6. Exercises 8C
exer:8C:1 1. If 3, 5, 8 are the only eigenvalues of T then each of them has multiplicity of at most 2. This
follows characteristic polynomial of T is (z 3)d1 (z 5)d2 (z 8)d3 with di  2. Hence,
(T 3I)2 (T 5I)2 (T 8I)2 = 0.
exer:8C:2 2. If 5, 6 are the only eigenvalues of T then each of them has multiplicity of at most n 1.
This follows (T 5I)n 1 (T 6I)n 1 = 0.
exer:8C:3 3. Let T (1, 0, 0, 0) = 7(1, 0, 0, 0), T (0, 1, 0, 0) = 7(0, 1, 0, 0) and T (0, 0, 1, 0) = 8(0, 0, 1, 0),
T (0, 0, 0, 1) = 8(0, 0, 0, 1).
exer:8C:4 4. If the minimal polynomial is (z 1)(z 5)2 , this could mean that G(5, T ) = null (T 5I)2 .
Choose T (a, b, c, d) = (a, 5b, 5c, c + 5d) then T 2 (a, b, c, d) = (a, 25b, 25c, 10c + 25d). Hence,
we find G(1, T ) = span ((1, 0, 0, 0)) and G(5, T ) = span ((0, 1, 0, 0), (0, 0, 1, 0), (0, 0, 0, 1)),
in particular (T 5I)(0, 0, 1, 0) 6= 0 but (T 5I)(0, 0, 1, 0) = 0; (T 5I)(0, 1, 0, 0) =
(T 5I)(0, 0, 0, 1) = 0 so indeed G(5, T ) = null (T 5I)2 . The characteristic polynomial
equals (z 1)(z 5)3 .
Note that (z 1)(z 5) is not minimal polynomial because (T I)(T 5I)2 (0, 0, 1, 0) 6= 0
and (T I)(T 5I)2 = 0 because G(5, T ) = null (T 5I)2 . Hence (z 1)(z 5)2 is the
minimal polynomial.
exer:8C:5 5. This means G(1, T ) = null (T I)2 6= null (T I). Hence, choose T (a, b, c, d) = (0, 3b, c, c+
d).
exer:8C:6 6. This means G(1, T ) = null (T I). Choose T (a, b, c, d) = (0, 3b, c, d).
exer:8C:7 7. From P 2 = P we follow 0, 1 are the only eigenvalues of P . Since P 2 = P so dim G(0, T ) =
dim null P so the characteristic polynomial of P is z dim null P (z 1)dim V dim null P . From
exer:5B:4
P 2 = P we also find that V = range P null P according to exercise 4 (5B). Hence
dim null P + dim P = dim V so the characteristic polynomial of P is z dim null P (z
1)dim range P .
exer:8C:8 8. T is invertible iff 0 is not an eigenvalue of T iff 0 iso not a zero of the minimal polynomial
of T iff the constant term in the minimal polynomial of T is nonzero.
exer:8C:9 9. If p(z) = 4 + 5z 6z 2 7z 3 + 2z 4 + z 5 is minimal polynomial of T then q(z) = 14 z 5 p(z 1 )

the is minimal polynomial of T 1 . Indeed, becausetheo:8.46:8C
p(T )v = 0 for all v 2 V so q(T 1 )v =
T p(T )v = 0 for all v 2 V . Hence, from theorem 10.5.4 (8C), q(z) is polynomial multiple
5
of the minimal polynomial of T 1 .

If the minimal polynomial of T 1 has degree less than degree of q(z) then with similar
construction as above, we can show that there exists a minimal polynomial of T whose
degree is less than degree of q(z), which is 5, a contradiction. Thus, we obtain q(z) is the
minimal polynomial of T 1 .
exer:8C:10 10. Since T is invertible so the constant term in the characteristc polynomial of T is nonzero,
i.e. p(0) 6= 0. Let 1 , . . . , m be eigenvalues of T with corresponding multiplicities
exer:8A:3
d1 , . . . , dm . Hence, p(z) = (z 1 ) d1 · · · (z
m ) dm . From exercise 3 (8A), we have
G( i , T ) = G( 1i , T 1) so characteristic polynomial of T 1 is
1 d1 1 dm
q(z) = (z 1 ) · · · (z m ) ,
= 1
d1
···
m
dm
· z dim V · ( 1 z 1 d1
) ···( m z 1 dm
) ,
( 1)dim V
= · z dim V ( 1 z 1 d1
) ···( m z 1 dm
) ,
p(0)
✓ ◆
1 dim V 1
= z p .
p(0) z
exer:8C:11 11. Since T is invertible so the minimal polynomial of T p(z) = a0 + a1 z + . . . + am z m has

exer:8C:8
a0 6= 0 according to exercise 8 (8C). Therefore T 1 p(T ) = 0 or a0 T 1 = a1 I a2 T
. . . am T m 1 .
exer:8B:5
exer:8C:12 12. According to exercise 5 (8B), V has basis consisting of eigenvectors of T if and only if
every generalized eigenvector of T is an eigenvector of T .
Let p(z) = (z d1
1) · · · Q
(z m)
dm be the minimal polynomial of T . Since p(T )v = 0 for
all v 2 V so that means j6=i (T j I) v 2 null (T

dj
i I) for all v 2 V .
n
Next, we show that p(z) has no repeated zero iff all generalized eigenvector of T is an
eigenvalue of T . Indeed, if p(z) has a repeated zero i , i.e. Q di 2 thendjthere exists v 2 V
so v 62 null (T i I) but v 2 null (T i I) di , otherwise
j6=i (T j I) v 2 null (T i I)
Q
for all v 2 V which leads to q(T ) = (T i I) j6=i (T j I)
d j = 0, where q(z) is
a polynomial with degree less than p(z), a contradiction. Conversely, if there exists a
generalized eigenvector v of T corresponding to eigenvalues i but v is not an eigenvector
of T corresponding to i , i.e. (T i I)v 6= 0 but v 2 G( i , T ). Note that for any 1  j  m
then T j I is invariant underexer:8A:4
G( i , T ). Combining with Q the fact that G(↵, T )\G( , T ) =
{0} for ↵ 6= from exercise 4 (8A), we follow that j6=i (T dj
j I) (T i I)v 6= 0 and
Q
i I)v 2 G( i , T ). This follows p(z) has a repeated zero i .
d
j6=i (T j I) (T
j
From these two claims, we are done.

theo:7.24:7B
exer:8C:13 13. If F = C then according Complex Spectral Theorem 9.3.1 (7B), T has a diagonal matrix
with respect to some orthonormal basis of V which means V has a basis consisting of
eigenvectors of T . Hence, according to previous exercise, the minimal polynomial of T has
no repeated zero.
theo:7.43:7C
exer:8C:14 14. Since S is an isometry so according to theorem 9.5.4 (7C), there exists a basis of V
consisting of eigenvalues of S whose corresponding eigenvectors all have absolute value
1. On the other hand, the constant term in the characteristic polynomial of S is p(0) =
( 1)n d11 · · · dmm so |p(0)| = 1.
exer:8C:15 15. (a) We know that there exists a monic polynomial p(z) (such as minimal polynomial of
T ) so p(T )v = 0. Hence, it suffices the uniqueness of such polynomial of smallest degree.
Assume the contrary, there exists two polynomial p, q with same and smallest degree so
p(T )v = q(T )v = 0. We follow (p q)(T )v = 0 but p q 6= 0 and p q has degree less

than of p, q, which is a contradiction. Thus, uniqueness is proven.
(b) Let q be the minimal polynomial of T . We follow hat deg p  deg q. Hence, there exists
polynomials r, s so deg s < deg p and q = pr + s. Since p(T )v = q(T )v = 0 so s(T )v = 0
but deg s < deg p so s = 0. This follows p divides the minimal polynomial of T .
exer:8C:16 16. Let p(z) = a0 + a1 z + . . . + am 1 z m 1 + am be the minimal polynomial of T and q(z) =
a0 + a1 z + . . . + am 1 z m 1 + am . Then we find q(T ⇤ ) = (p(T ))⇤ . Note that since p(T )v = 0
for all v 2 V so 0 = hp(T )q(T ⇤ )v, vi = theo:8.46:8C
hq(T ⇤ )v, (p(T ))⇤ vi = kq(T ⇤ )vk2 for all v 2 V so
q(T )v = 0. Therefore, from theorem 10.5.4 (8C), q is the polynomial multiple of the
⇤
minimal polynomial h of T ⇤ .
Now it suffices to show that deg q = deg h. Assume the contrary deg h < deg q then with
similar approach, we can construct a polynomial g from h so g(T ) = 0 and deg g = deg h <
deg q = deg p, a contradiction. Thus, we must have deg g = deg h so q(z) is the minimal
polynomial of T ⇤ .
exer:8C:17 17. Let q(z) = (z k1
1 ) · · · (z m)
km be the minimal polynomial of T with k + . . . + k = n
1 m
and p(z) theo:8.46:8C
= (z d
1 ) · · · (z
1
m)
dm be the characteristic polynomial of T . According to
Pm
theorem
Pm 10.5.4 (8C), we follow d i k i for all 1  i  m. On the other hand, i=1 ki =
i=1 di = n so that means di = ki for all 1  i  m, i.e. the characteristic polynomial of
T equals the minimal polynomial of T .
exer:8C:18 18. Call such operator T and let e1 , . . . , en be the standard basis of Cn with ei has 1 in
the i-th coordinate and 0’s in the remaining coordinates. We can easily find that for
1  i  n 1, 1  k  n then T k ei = ei+k if i + k  n.
We have T en = (a0 , a1 , . . . , an 1) so
T 2 en = T (a0 , . . . , an 2 , 0) an 1 T en = (0, a0 , . . . , an 2) an 1 T en .
Hence,
T 3 en = T (0, a0 , . . . , an 3 , 0) an 2 T en an 1T
2
en ,
2
= (0, 0, a0 , . . . , an 3) an 2 T en an 1T en .
Continuing doing this, we will obtain
T i en = (0, 0, . . . , 0, a0 , . . . , an i ) an i+1 T en ... an 1T
i 1
en .
For i = n then T n en = a0 en a1 T en ... an 1T
n 1e
n or (T n + an 1T
n 1 + ... +
a1 T + a0 I)en = p(T )en = 0.
For 1  i < n, notice that T k ei = ei+k for 0  k  n i. Hence, we have
p(T )ei = (T n + an 1T
n 1
+ . . . + an i+1 T n i+1 )ei + an i en + . . . + a1 ei+1 + a0 ei ,
= (T i + an 1 T i 1 + . . . + an i+1 T )(T n i ei ) + (0, . . . , 0, a0 , . . . , an i 1 ),
= (T i + an 1 T i 1 + . . . + an i+1 T )en + (0, . . . , 0, a0 , . . . , an i ),
= 0.
Thus, p(T )v = 0 for all v 2 V . Since deg p = n so that means p(z) is both the minimal
polynomial and the characteristic polynomial of T .
exer:8B:11
exer:8C:19 19. According to exercise 11 (8B), the number of times that appears on the diagonal of the
matrix of T equals the multiplicity of as an eigenvalue of T. Combining with the defi-
nition of characteristic polynomial for complex vector space, we obtain the characteristic
polynomial of T is (z 1 ) · · · (z n ).
theo:5.27:5B
exer:8C:20 20. Since we are talking about complex vector space, according to theorem 7.3.2 (5B), T |Vj
has an upper-triangular matrix M(T |Vj ) with respect to some basis of Vj .
Therefore, if we combines all the bases of Vj we obtain a basis of V since V = V1 · · · Vm
which will give an block diagonal matrix for T as follow:
0 1
M(T |V1 )
B .. C
M(T ) = @ . A.
0 M(T |Vm )
Notice that M(T ) is an upper-triangular matrix since M(T |Vj ) are upper-triangular matri-
exer:8C:19
ces. Thus, by applying exercise 19 (8C) to M(T ), M(T |Vj ) we can conclude that p1 · · · pm
is the characteristic polynomial of T .
10.7. 8D: Jordan Form

theo:8.55:8D Theorem 10.7.1 (8.55, basis correspoding to a nilpotent operator) Suppose N 2 L(V ) is
nilpotent. There exists vectors v1 , . . . , vn 2 V and nonnegative m1 , . . . , mn such that
(a) N m1 v1 , . . . , N v1 , v1 , . . . , N mn vn , . . . , N vn , vn is a basis of V .
(b) N m1 +1 v1 = . . . = N mn +1 vn = 0.
exer:8D:6
Furthermore, according to exercise 6 (8D), n = dim null N and N m1 v1 , . . . , N mn vn is basis of
null N .
theo:8.60:8D Theorem 10.7.2 (8.60, Jordan Form) Suppose V is a complex vector space. If T 2 L(V ),
then there is a basis of V that is a Jordan basis for T .
In particular, there is a Jordan basis of V for T so

0 1
M(T |G( 1 ,T ) ) 0
B . .. C
M(T ) = @ A.
0 M(T |G( m ,T )
)
In here, M(T |G( i ,T )
) is Jordan matrix of T |G( on G( i , T ), i.e.
i ,T )
0 1
A1 0
B .. C
M(T |G( i ,T )
)=@ . A,
0 Ap
where Aj is an upper-triangular matrix of the form:

0 1
i 1 0
B .. .. C
B
B . . C
C.
B .. C
@ . 1A
0 i
theo:8.55:8D
Indeed, since (T i I)|G( i ,T ) is nilpotent so from theorem 10.7.1, there exists such basis to
form Jordan matrix for T |G( i ,T ) .
Q
theo:4:8D Theorem 10.7.3 The minimal polynomial of T is m i=1 (z i)
mi +1 where m is the length
i
of the longest consecutive string of 1’s that appears on the line directly above the diagonal
in the Jordan matrix of M(T |G( i ,T ) ).
theo:3:8C
Proof. From theorem 10.5.6 (8C), minimal polynomial of T equals product of minimal polyno-
mials of T |G( i ,T )exer:8D:3
for all eigenvalues i of T .
From exercise 3 (8D), minimal polynomial of the nilpotent operator (T i I)|G( i ,T ) is z
mi +1
where mi is the length of the longest consecutive string of 1’s that appears on the line directly
above the diagonal in the Jordan matrix of M(T |G( i ,T ) ). We follow (T i I)
mi +1 = 0 but
mi 6= 0, which leads to (z mi +1 as the minimal polynomial of T |
(T i i) i) G( i ,T ) .
Qm
Thus, the minimal polynomial of T is i=1 (z i)
m i +1 where mi is the length of the longest
consecutive string of 1’s that appears on the line directly above the diagonal in the Jordan
matrix of M(T |G( i ,T ) ).
example_jordan_form Example 10.7.4 we can find the minimal polynomial of T from its Jordan matrix as shown:
N 2 v1 N v1 v1 N v2 v2 N v3 v3
N 2v 0 1 0 0 0 0 01
1 1
N v1 B 0 1 1 0 0 0 0C
B C
v1 B 0 0 1 0 0 0 0C
M(T ) = B C
N v2 B 0 0 0 1 1 0 0C
B C
v2 B 0 0 0 0 1 0 0C
@ A
N v3 0 0 0 0 0 2 1
v3 0 0 0 0 0 0 2
So N 2 v1 , N v1 , v1 , N v2 , v2 2 G( 1 , T ) and N v3 , v3 2 G( 2 , T ). In the top left submatrix of

M(T ) that only has 1 on the diagonal, the longest length of a consecutive strings of 1’s is 2,
theo:4:8D
which follows (z 1 ) 3 is the minimal polynomial of T |
G( 1 ,T ) according to theorem 10.7.3.
Similarly, (z 2 ) is the minimal polynomial of T |G( 2 ,T ) . Thus, the minimal polynomial
2
of T is (z 3
1 ) (z 2) .
2
10.8. Exercises 8D
exer:8D:1 1. Since N is nilpotent so N 4 = 0 and since N 3 6= 0, we find z 4 is both the minimal polynomial
and the characteristic polynomial of T .
exer:8D:2 2. In example 8.54, with the basis N 2 v1 , N v1 , v1 , N v2 , v2 , v3 given that N 3 v1 = N 2 v2 =

N v3 = 0, we follows that N 3 = 0 and N 2 6= 0. Hence, N 3 is the minimal polynomial of T
and N 6 is the characteristic polynomial of T .
theo:8.55:8D theo:8.60:8D
exer:8D:3 3. We know so far according to theorem 10.7.1 (8D) and 10.7.2 (8D), Jordan basis of V is a
combination of linear independent lists of the form N m v, N m 1 v, . . . , N v, v. Each of this
list will produce a consecutive string of 1’s with length m that appears on the line directly
above the diagonal in the Jordan matrix of T . Therefore, the longest consecutive string of
1’s corresponds to the list N m v, N m 1 v, . . . , N v, v with maximal m. We follow N m+1 = 0
and N m 6= 0 (because N m v is in the Jordan basis of V ). Therefore, z m+1 is the minimal
polynomial of T . Note that m here is the length of the longest consecutive string of 1’s.
exer:8D:4 4. If the Jordan basis v1 , . . . , vn is reversed then the columns of N j vi are reversed, i.e. if it
is (a1 , a2 , . . . , an )T then its reverse is (an , . . . , a1 )T . All of the columns are then arranged
in reversed order. Thus, with this, we conclude that if we rotate 180 the Jordan matrix
of T , we will obtain a matrix with respect to the reversed order of Jordan basis of V .
exer:8D:5 5. In the matrix of T 2 with respect to Jordan basis of V for T , all the eigenvalues on the
diagonal are squared and all the 1’s that is in the same column with changes to 2 . All
the 0’s in row of N k vi and column N k 2 vi changes to 1. Below is an example:
N 2 v1 N v1 v1 N v2 v2 N v3 v3
N 2 v1 0 21 2 1 1 0 0 0 0 1
N v1 B 0 2 2 1 0 0 0 0 C
1
B 2 C
v1 B 0 0 1 0 0 0 0 C
M(T ) = N v2 B
2
B 0 0 0 2
2 2 2 0 0 C
C
B 2 C
v2 B 0 0 0 0 2 0 0 C
@ 2 A
N v3 0 0 0 0 0 3 2 3
v3 0 0 0 0 0 0 2
3
exer:8D:6 6. Since we already know that N m1 v1 , . . . , N mn vn is linearly independent list in null N , it

suffices to show that this list spans null N . Indeed, consider P
theo:8.55:8D
v 2 null N then from the
basis given in theorem 10.7.1 (8D), we can represent v as v = 0imj ,1jn ↵i,j N i vj so
P
0 = N v = 1imj ↵i,j N i vj . We follows ↵i,j = 0 for all 1  j  n, 1  i  mj , which
P
means v = nj=1 ↵mj ,j N mj vj . Thus, N m1 v1 , . . . , N mn vn is basis of null N .
theo:4:8D
exer:8D:7 7. From theorem 10.7.3 (8D), draw out the Jordan matrix for T that makes p the minimal
polynomial and q the characteristic polynomial. After that, we can choose the standard
basis e1 , . . . , en as the Jordan basis for this matrix. This will define an operator T that
satisfies the condition.
exer:8D:8 8. If there does not exist a direct sumtheo:8.21:8Bdecomposition of V into two proper subspaces in-
variant under T then from theorem 10.3.1 (8B), T must have only one eigenvalue and
V = G( , T ) = null (T I)dim V . We follow (T I) is a nilpotent operator on V . If
theo:8.55:8D
dim null (T I) 2 then according to theorem 10.7.1 (8D), for n = dim null (T I) 2,
V has a Jordan basis (T I)m1 v1 , . . . , v1 , . . . , (T I)mn vn , . . . , vn for T . Let U =
span ((T I)m1 v1 , . . . , v1 ) and W = span ((T I)m2 v2 , . . . , v2 , . . . , (T I)mn vn , . . . , vn )
then V = U W . If u 2 U then T u = (T I)u + u 2 U so U is invariant under T . Simi-
larly, W is invariant under T . This contradicts to the original theo:8.55:8D
assumption. Hence, we must
have dim null (T I) = 1, which again from theorem 10.7.1 (8D), (T I)dim V 1 v, . . . , v
is a basis of V . This follows the smallest d so (T I) = 0 is d = dim V . We find
d
(z )dim V is the minimal polynomial of T .
If (z )dim V is the minimal polynomial of T then there exists v so (T I)n 1 v, . . . , v is

a basis of V with n = dim V . Assume the contrary that V has a direct sum P decomposition
into two proper subspaces U, W invariant under T . If u 2 U , write u = ni=01 ai (T I)i v
where ai 2 C for 0  i  n 1. Let ai be the first nonzero number in a0 , . . . , an 1 . Since
U is invariant under T so U is also invariant under T I, which means (T I)n i 1 u =
ai (T I) n 1 v 2 U . Therefore, (T I) n 1 v 2 U . We can apply similar method to W
can conclude that (T I) n 1 v 2 W . Hence, U \ W 6= {0}, so we can’t have V = U W .
Thus, there does not exist a direct sum decomposition of V into two proper subspaces
invariant under T .
11. Chepter 9: Operators on Real Vector Spaces

11.1. 9A: Complexification
Definition 11.1.1 (Complexification VC of V ). The complexification VC of V is V ⇥ V , where
each element is of the form (u, v) or u + iv for u, v 2 V . We define addition on VC as (u1 + iv1 ) +
(u2 + iv2 ) = (u1 + v1 ) + i(u2 + v2 ) for ui , vi 2 V and scalar multiplication as (a + ib)(u + iv) =
(au bv) + i(by + av) for a, b 2 R and u, v 2 V .
Construction of VC from V can be thought as generalizing the construction of Cn from

Rn .
Theorem 11.1.2
Complexification of V is a complex vector space. Any basis of V is a basis of VC . Even
better, any subspace of VC has a basis containing vectors in V .
Definition 11.1.3 (Complexification of operator T ). For real vector space V , complexification

TC of any T 2 L(V ) is the operator TC 2 L(VC ) such that TC (u + iv) = T u + iT v for any
u, v 2 V .
lexification_property Theorem 11.1.4 An operator T on a real vector space V has minimal/characteristic poly-
nomial same as its complexification TC . This means that:
1. Minimal/characteristic polynomials of TC have real coefficients.
2. Real zeros of the characteristic polynomial are eigenvalues of T .
3. Multiplicity of 2 C as eigenvalue of TC equals the multiplicity of as eigenvalue of

TC . In other words dim G( , TC ) = dim G( , TC ).
4. If dim V is odd then T has at least one eigenvalue.
11.2. Exercises 9A
exer:9A:1 1. Done.
exer:9A:2 2. Show TC 2 L(VC ). For u1 + iv1 , u2 + iv2 2 VC then
TC (u1 + iv1 + u2 + iv2 ) = T (u1 + u2 ) + iT (v1 + v2 ) = TC (u1 + iv1 ) + TC (u2 + iv2 ).
Similarly, TC ( (u + iv)) = TC (u + iv).
exer:9A:3 3. Not hard.
exer:9A:4 4. Not hard.
exer:9A:5 5. (S + T )C (u + iv) = (S + T )u + i(S + T )v = (SC + TC )(u + iv).
exer:9A:6 6. Prove injectivity and surjectivity.
exer:9A:7 7. Since (NC )n (u + iv) = N n u + iN n v so N is nilpotent iff NC is nilpotent.
exer:9A:8 8. This follows 5, 7 are eigenvalues of TC and since characteristic polynomial of TC of degree
3 so TC has no nonreal eigenvalues.
exer:9A:9 9. Since T is an operator on odd-dimensional real vector space so T has a real eigenvalue.
On the other hand, since T 2 + T + I is nilpotent so minimal polynomial p(T ) of T divides
(T 2 + T + I)7 . Since minimal polynomial p(T ) of T has real coefficients so p(T ) = (T 2 +
T + I)k for some k < 7. However, this implies p(T ) has no real roots, which means T has
theo:8.49:8C
no real eigenvalue according to 10.5.5, a contradiction to previous claim. Thus, there does
not exists T 2 L(R7 ) so T 2 + T + I is a nilpotent.
exer:9A:9
exer:9A:10 10. The difference between this exercise and the previous exercise 9 is that minimal polynomial
of T need not to have real coefficients. Since T 2 +T +I is nilpotent, we aim to construct T so
⇣ p ⌘k ⇣ p ⌘k
1+i 3 1 i 3
its minimal polynomial is of the form p(T ) = (T 2 +T +I)k = T 2 I T 2 I
where k < 7/2 because minimal polynomial must have degree at most dim C7 = 7. Say
example_jordan_form 1+ip3 p
1 i 3
k = 3. From example 10.7.4, let + = 2 and = 2 , we can construct a
Jordan matrix as follow:
0 1
+ 1
B + 1 C
B C
B + C
B C
M(T ) = B
B +
C
C
B 1 C
B C
@ 1A
WLOG, say that (0, . . . , 0, 1, 0, . . . , 0) is Jordan basis of T then from the above matrix we
can construct T .
exer:9A:11 11. The minimal polynomial p of T must divide x2 + bx + c and since eigenvalue of T are
roots of p so T has real eigenvalue if is root of x2 + bx + c, i.e. b2 4c.
exer:9A:11
exer:9A:12 12. Similarly to exercise 11, minimal polynomial p of T must divide (z 2 + bz + c)dim V , and
since (z 2 + bz + c)dim V has no real root, p has no real root, i.e. T has no eigenvalues.
exer:9A:13 13. Since W = null (T 2 + bT + cI)jexer:9A:12

is a vector space so T 2 + bT + cI is nilpotent on W , which
according to previous exercise 12 implies that T has no eigenvalues. This follows dim W
is even.
exer:9A:14 14. T 2 +T +I is nilpotent implies TC has exactly two eigenvalues and with same multiplicity
complexification_property
according to theorem 11.1.4. Therefore, the characteristic polynomial of TC is p(z) =
(z 2 + z + 1)k . Since dim V = 8 so deg p = 8 or k = 4. Thus, the characteristic polynomial
of T is (z 2 + z + 1)4 so (T 2 + T + I)4 = 0.
exer:9A:15 15. Let U be an invariant subspace ofcomplexification_property

V . Since T has no eigenvalue so T |U 2 L(U ) has no
eigenvalue. According to theorem 11.1.4, U has even dimension.
exer:9A:12
exer:9A:16 16. If T 2 = I then T has no eigenvalue according to exercise 12, which implies V has even
complexification_property
dimension according to theorem 11.1.4.
example:eigenvalue
Conversely, if V has even dimension. Inspired from example 7.1.1, for any basis a1 , b1 , a2 , b2 , . . . , an , bn
of V , let T 2 L(V ) such that T (ai ) = bi , T (bi ) = ai . With this, we can see that T 2 +I = 0.
exer:9A:17 17. (a) We already know that V is a real vector space. For any a, b, c, d 2 R and any v, u 2 V
then (a + ib)v = av + bT v 2 V and
(a + bi)[(c + di)v] = (a + bi)(cv + dT v),

= (acv + bcT v) + (adT v + bdT 2 v),
= (ac bd)v + (bc + ad)T v,
= (ac bd + i(ad + bc))v,
= [(a + bi)(c + di)]v.
We also have (a + bi)(u + v) = a(u + v) + bT (u + v) = (au + bT u) + (av + bT v) =

(a + bi)u + (a + bi)v and similarly (a + bi + c + di)u = (a + bi)u + (c + di)u. Thus, V is a
complex vector space.
(b) Let dim V = 2n. Pick ai 2 V inductively such that ai+1 62 span (a1 , T a1 , a2 , T a1 , . . . , ai , T ai ).
We will first show that a1 , T a1 , . . . , an , T an is linearly independent.
Assume the contrary that the list is linearly dependent, P then there exists i such that
T ai+1 2 span (a1 , T a1 , . . . , ai , T ai , ai+1 ). Let T ai+1 = ij=1 (↵i ai + i T ai ) + cai+1 where
↵i , i 2 R. Since T 2 = I so
i
X
ai+1 = T 2 ai+1 = (↵i T ai i ai ) + cT ai+1 .
j=1
Substituting T ai+1 into the above to obtain ai+1 2 span (a1 , T a1 , . . . , ai , T ai ), a contra-
diction to our condition. Thus, a1 , T a2 , . . . , an , T an must be linearly independent and
therefore it is a basis of real vector space V .
We prove that with such definition of complex scalar multiplication then a1 , . . . , an is
the basis of V as complex vector space. Since aj 2 V so iaj = T aj 2 V . Therefore,
a1 , T a1 , . . . , an , T an 2PV which implies that
P a1 , . . . , an spans V . The list is also linear
independent because ni=1 (↵i + i i )ai = ni=1 ↵i ai + i T ai .
Thus, the dimension of V as a complex vector space is half the dimension of V as real
vector space.
exer:9A:18 18. (b) =) (a): If there exists a basis of V wrt to which T has an upper-triangular matrix,
then it is also a basis of theo:5.32:5B
VC with respect to which TC has an upper-triangular matrix.
Therefore, from theorem 7.3.4 (5B), we find TC has real eigenvalues.
(a) =) (c): If all eigenvalues of TC are real. Consider G( , TC ) where 2 R is an
eigenvalue of TC . Consider basis x1 + iy1 , . . . , xn + iyn of G( , TC ) where xj , yj 2 V . We
have TC (xj + iyj ) = T xj + iT yj = (xj + iyj ) which implies T xj = xj , T yj = yj for all
1  j  n. On the other hand, note that xj + iyj 62 U = span (x1 + iy1 , . . . , xj 1 + iyj 1 )
so at least one of xj , yj must not be in U . From this, we can inductively choose either xi
or yi to create a basis of G( , TC ) that are all in V .
theo:8.21:8B Lm
From theorem 10.3.1 (8B), since VC = i=1 G( i , TC ), we can construct a basis of VC
consisting generalized eigenvectors that are in V . This follows V has a basis consisting of
generalized eigenvectors of T .
L (a): If there is a basis of V consisting of generalized eigenvectors of T then

(c) =)
V = m i=1 G( i , T ) where i 2 R are eigenvalues of T .
If v1 , . . . , vn is basis of G( , T ) then since 2 R, we have (TC I)k (u + iv) = 0
when L u, v 2 G( , T ). Therefore, v1 , . . . , vn is a basis of G( , TC ). Hence, we obtain
VC = m i=1 G( i , TC ). This follows that TC does not have any nonreal eigenvalue.
(a) =) (c) =) (b): If all eigenvalues of TC are real andL then there exists a basis of V
consisting of generalized eigenvectors of T . Hence, V = m
i=1 G( i , T ) where i 2 R are
theo:8.60:8D
eigenvalues of T . With this and from the proof of theorem 10.7.2 (8D), we find that there
is a Jordan basis for T of V . This proves (b).
exer:9A:19 19. Bring back to working with complex vector space. Since null T n 2 6= null T n 1 so
exer:8B:4
n 2 n 1
null TC 6= null TC . Hence, from exercise 4, we find TC has at most 2 eigenvalues,
one of which must be 0. This follows the other must be real eigenvalue. Therefore, T has
at most 2 eigenvalues and TC has no nonreal eigenvalues.
11.3. 9B: Operators on Real Inner Product Spaces

theo:9.34:9B Theorem 11.3.1 (Characterization of normal vectors when F = R) Suppose V is a real
inner product space and T 2 L(V ). Then the following are equivalent:
1. T is normal.
2. There is an orthonormal basis of V with respect to which T has a block✓diagonal ◆

a b
matrix such that each block is a 1-by-1 matrix or a 2-by-2 matrix of the form
b a
with b > 0.
By looking at the characteristic polynomial of normal operator T , one can count number of
2-by-2 block diagonal theo:9.34:9B
matrices. theo:7.10:7A
Using this
theo:7.29:7B
theorem 11.3.1 and theorem 9.1.4, one can easily prove the Real Spectral Theorem
9.3.2.
theo:9.36:9B Theorem 11.3.2 (Description of isometries when F = R) Suppose V real inner product
space and S 2 L(V ). Then the following are equivalent:
1. S is an isometry.
2. There is an orthonormal basis of V with respect to which S has a block diagonal

matrix such that each block on✓the diagonal is◆a 1-by-1 matrix containing 1 or 1 or
cos ✓ sin ✓
is a 2-by-2 matrix of the form with ✓ 2 (0, 2⇡).
sin ✓ cos ✓
exer:9B:3
We can also define an inner product for complexification VC as indicated in exercises 3, i.e.
hu + iv, x + iyi = hu, xi + hv, yi + (hv, xi hu, yi) i.
11.4. Exercises 9B
theo:9.36:9B
exer:9B:1 1. From theorem 11.3.2, S has block diagonal 3-by-3 matrix so at least one block on the
diagonal is a 1-by-1 matrix. If it’s the first block then pick x = (1, 0, 0), if it’s the last
block then x = (0, 0, 1).
theo:9.36:9B
exer:9B:2 2. Obviously true from theorem 11.3.2.
exer:9B:3 3. Jump back to the definition of inner product.

Check positivity and definiteness: hu + iv, u + ivi = hu, ui + hv, vi 0 and is 0 when
u = v = 0.
Check additivity in first slot and homogeneity in first slot.
Check conjugate symmetry:
hu + iv, x + iyi = hu, xi + hv, yi (hv, xi hu, yi) i,

= hx, ui + hy, vi + (hy, ui hx, vi) i,
= hx + iy, u + ivi .
exer:9B:4 4. It suffices to show hTC (u + iv), x + iyi = hu + iv, TC (x + iy)i.

exer:9B:4
exer:9B:5 5. If T is self-adjoint then TC is self-adjointtheo:7.13:7A
according to exercise 4, which meanstheo:7.24:7B
all eigenval-
ues of TC are real according to theorem 9.1.5. By Complex Spectral Theorem Lm 9.3.1, VC has
orthonormal basis consisting of eigenvectors of TC . This implies VC = i=1 E( i , TC ).
In each subspace E( i , TCtheo:6.31:6B
), there exists a basis containing vectors in V . By applying
Gram-Schmidt procedure 8.3.2 to this basis, we obtain an orthonormal basis of E( , TC )
containing vectors in V . Combining all these vectors and the fact that eigenvectors cor-
theo:7.22:7A
responding to distinct eigenvalues of normal operator are orthorgonal 9.1.7, we obtain an
orthonormal basis for V containing all eigenvectors of T .
✓ ◆
1 2
exer:9B:6 6. Look at the proof and write out the matrix so B is nonzero matrix. Say M(T ) =
0 1
wrt to standard basis where T 2 L(R ). Then U = span (1, 0) is invariant under T but
2
not U ? = span (0, 1).
exer:9B:7 7. Let such basis of V be v1 , . . . , vn and show that T vi = (T1 · · · Tm )vi .

exer:7A:21
exer:9B:8 8. According to exercise 21 (7A), the list
1 cos x cos nx sin x sin nx

p , p ,..., p , p ,··· , p
2⇡ ⇡ ⇡ ⇡ ⇡
is the orthonormal basis of V , and the matrix of D with⇣ this respect ⌘ to this basis will give
theo:9.34:9B
the desired form in 11.3.1. In particular, Ui = span cos ix
p , p
⇡
sin ix
⇡
is invariant under T
✓ ◆
a b
where M(T |Ui ) = .
b a
12. Chapter 10: Trace and Determinant

12.1. 10A: Trace
prop:10.4:10A Proposition 12.1.1 (Matrix of product of linear maps) Suppose u1 , . . . , un and v1 , . . . , vn
and w1 , . . . , wn are bases of V . Suppose S, T 2 L(V ). Then
M(ST, (u1 , . . . , un ), (w1 , . . . , wn )) =M(S, (v1 , . . . , vn ), (w1 , . . . , wn ))

M(T, (u1 , . . . , un ), (v1 , . . . , vn )).
coroll:10.7:10A Corollary 12.1.2 (Change of basis) Suppose T 2 L(V ). Let u1 , . . . , un and v1 , . . . , vn be

bases of V .
Let A = M(I, (u1 , . . . , un ), (v1 , . . . , vn )) then
1
M(T, (u1 , . . . , un )) = A M(T, (v1 , . . . , vn ))A.
theo:10.16:10A Theorem 12.1.3 (Trace of an operator equals to trace of its matrix) For T 2 L(V ) then
trace T = trace M(T ) = (trace M(TC )). In other words, for any basis of T , the sum of
diagonal entries of M(T ) is constant and is equal to sum of eigenvalues of T (of TC if V is
real vector space), with each eigenvalue repeated according to its multiplicity.
12.2. Exercises 10A

exer:10A:1 1. If A = M(T, v1 , . . . , vn )) is invertible there exists inverse B of A. Weprop:10.4:10A
define an operator
S 2 L(V ) such that M(S, (v1 , . . . , vn )) = B. By using proposition 12.1.1, we can show
that B is inverse of T .
prop:10.4:10A
If T is invertible, then there exists inverse T 1 of T . Hence, using proposition 12.1.1 to ma-
trices M(T T 1 , (v1 , . . . , vn )) and M(T 1 T, (v1 , . . . , vn )) we obtain that M(T, (v1 , . . . , vn ))
is invertible.
exer:10A:2 2. Bring back to woring with operators. Define S, T 2 L(V ) so M(S, (v1 , . . . , vn )) = A, M(T, (v1 , . . . , vn )) =
exer:3D:10
B then ST = I as M(ST ) = I. Applying exercise 10, we implies T S = I so BA = I.
P
exer:10A:3 3. Consider an arbitrary basis v1 , . . . , vn of V . Let T v1 = ni=1 ↵i vi . Since M(T, (v1 , . . . , vn )) =
M(T, (2v1 , . . . , vn )) so ↵i = 2↵i for all i 6= 1 which implies ↵i = 0 for all i 6= 1. Hence,
T v1 = ↵1 v1 . Similarly, we obtain T vj = ↵j vj for all j.
On the other hand, since M(T, (v1 , v2 , . . . , vn )) = M(T, (v2 , v1 , . . . , vn )) so ↵1 = ↵2 . As a
result, we obtain T = ↵I.
exer:10A:4 4. Since T vk = uk so M(T, (v1 , . . . , vn ), (u1 , . . . , un )) = I. Therefore, by applying proposition

prop:10.4:10A
12.1.1, we obtain
M(T, (v1 , . . . , vn )) = M(I, (u1 , . . . , un ), (v1 , . . . , vn )).

exer:10A:5 5. Bring back to working with operator. Let T 2 L(V ) be an operatortheo:8.29:8B so M(T, (vtheo:8.60:8D
1 , . . . , vn )) =
B. Since V is complex vector space so according to theorem 10.3.3 (or 10.7.2), there
exists basis u1 , . . . , un so M(T, (u1 , . . . , un )) is an upper-triangular matrix. Now applying
coroll:10.7:10A
corollary 12.1.2 and we are done.
✓ ◆
0 1
exer:10A:6 6. Let T 2 L(R ) so M(T ) =
2 . Then trace (T 2 ) = 2 < 0.
1 0
P
exer:10A:7 7. trace (T 2 ) = m i=1 i
2 0.
exer:10A:8 8. If w = 0 then trace T = 0. If w 6= 0 then choose basis of V contatining w, with this, we

obtain trace T = hw, vi.
exer:8C:7
exer:10A:9 9. If P 2 = P then according to exercise 7 (8C), the characteristic polynomial of P (or PC ) is
z m (z 1)n where n = dim range P = dim range PC . This follows trace T = dim range P
according to the definition of trace of an operator.
theo:7.10:7A
exer:10A:10 10. For orthonormal basis v1 , . . . , vn of V , then according to theorem 9.1.4 (7A), M(T, (v1 , . . . , vn ))
is conjugate transpose of M(T ⇤ , (v1 , . . . , vn )). Therefore, trace T ⇤ = trace T according to
theo:10.16:10A
theorem 12.1.3 (10A)
theo:7.35:7C
exer:10A:11 11. From theorem 9.5.2, if T is positive then all eigenvalues of T are nonnegative. As trace T
is sum of all eigenvalues and trace T = 0, we obtain all eigenvalues are 0. Thus, T = 0.
exer:10A:12 12. We have trace (P Q) = trace M(P Q)C =theo:8.29:8B

trace (M(PC )M(QC )). With this, we can
choose basis for PC and QC as in theorem 10.3.3 (8B) so M(PC ) and M(QC ) are upper-
triangular matrices. Since P, Q are orthogonaltheo:6.55:6C
projections then so are PC , QC . Hence,
PC2 = P and Q2 = Q according to theorem 8.5.5 (6C). This implies that eigenvalues of
C C C
PC and QC are only 0 and 1. This means trace M(PC )M(QC ) 0, as desired.
exer:10A:13 13. 51 40 + 1 = 12 and 12 24 ( 48) = 36 so the third eigenvalue if 36.
exer:10A:14 14. We have trace (cT ) = trace M(cT ) = trace cM(T ) = ctrace T .
exer:10A:15 15. We have
trace (ST ) = trace M(ST ),

= trace (M(S)M(T )),
= trace (M(T )M(S)),
= trace M(T S),
= trace T S.
exer:10A:16 16. Let ST = T = S = I with dim V = 2 then trace S = trace T = trace ST = 2 and
therefore, trace ST 6= trace Strace T .
exer:10A:17 17. Work with M(T ) and M(S). Choose Si,j 2 L(V ) so M(Si,j )i,j = 1 while the rest entries
are 0. Hence, this implies M(T )j,i = 0. Thus, this concludes T = 0.
theo:7.10:7A
exer:10A:18 18. For orthonormal basis e1 , . . . , em of V then from theorem 9.1.4 (7A), M(T, (e1 , . . . , em ))
is the conjugate transpose of M(T ⇤ , (e1 , . . . , em )). This means, the i-th row of M(T ⇤ )
is the conjugate of the i-th column of M(T ). Hence, kT ei k2 equals the inner product of
these
Pm two vectors, which is the (i, i-th entry of M(T ⇤ T, (e1 , . . . , em )). Thus, trace (T ⇤ T ) =
i=1 kT ei k . This implies that then sum is independent from the choice of orthonormal
2
basis.
exer:10A:19 19. From previous exercise, we have hT, T i 0 and is 0 when when trace T T ⇤ = 0, which
follows kT ei k = 0 for orthonormal basis e1 , . . . , em of V . This implies T ⇤ = 0 so T = 0.
⇤
Thus, positivity and definiteness are proven.

Addivitity in first slot exer:3C:13
is true because of the distributive property for matrix addition and
matrix multiplication 13 (3C).
exer:10A:14
Homogeneity in first slot is true using exercise 14 (10A).
Conjugate
exer:10A:10
Symmetry is true because ST ⇤ is adjoint of T S ⇤ , which according to exercise
10 (10A) then
hS, T i = trace (ST ⇤ ) = trace (T S ⇤ ) = hT, Si.
exer:10A:18
exer:10A:20 20. Let e1 , . . . , en be orthonormal basis of V which gives the desired matrix. From exercise 18
(10A), we find
Xn Xn X n
⇤ 2
trace (T T ) = kT ei k = |Aj,k |2 .
i=1 k=1 j=1
Pn
It suffices to show i=1 | i |2  trace (T ⇤ T ).theo:8.29:8B
We go and pick different basis of V . Pick
orthonormal basis e1 , . . . , en as in theorem 10.3.3 (8B) so M(T ) is an upper triangular
matrix where all the eigenvalues are on its diagonal. Hence, kT ei k2 | i |2 for all 1  i  n.
exer:10A:18
Combining with exercise 18 (10A), we obtain the desired inequality.
exer:10A:18
exer:10A:21 21. Applying exercise 18 (10A), for any orthonormal basis e1 , . . . , en of V so e1 = v, we have
n
X n
X
⇤ ⇤
trace (T T ) = 2
kT ei k  kT ei k2 = trace (T ⇤ T ).
i=1 i=1
exer:10A:15
On the other hand, according to exercise 15 (10A), we find trace (T T ⇤ ) = trace (T ⇤ T ).
Thus, the equality holds when kT ⇤ ei k = kT ei k so kT ⇤ vk = kT vk. Since this is true for
any v 2 V , we obtain that T is normal
13. Summary
Example 13.0.1 (Inner product space of continuous real-valued functions)

Let C[ ⇡, ⇡] be the vector space of continuous real-valued functions on [ ⇡, ⇡] with inner
product Z ⇡
hf, gi = f (x)g(x)dx.
⇡
For some positive integer integer n then
exer:6B:4
1. (Exercise 4 (6B)) p12⇡ , cos
p x , . . . , cos
⇡
p nx , sin
⇡
p x , sin
⇡
p 2x , . . . , sin
⇡
pnx is an orthonormal list of
⇡
vectors in C[ ⇡, ⇡].
2. Let V = span (1, cos x, cos 2x, . . . , cos nx, sin x, sin 2x, . . . , sin nx) be a subspace of
C[ ⇡, ⇡].
a) Define D 2 L(V ) by Df = exer:7A:21
f 0 . Then D⇤ = D which implies D is normal
but not self-adjoint (exercise 21 (7A)). From previous remark, we can also find
an orthonormal basis of V such
theo:9.34:9B
that the matrix of D has the form described in
exer:9B:8
theorem 11.3.1 (see exercise 8 (9B)).
b) Define T 2 L(V ) by T f = f 00 . Then T is self-adjoint and
exer:7C:14
T is positive operator
(exercise 14 (7C)).
Example 13.0.2 (Projection operator)

A linear operator P on vector space V is a projection on V if P 2 = P . That is, it leaves its
image unchanged.
exer:5B:4
1. V = null P range P (exercise 4(5B)).
2. Characteristic polynomial of P is z m (z 1)n where m = dim null P and n = dim range P

exer:8C:7
(exercise 7 (8C)).
exer:10A:9
3. trace P = dim range P (exercise 9 (10A)).
def:6.53:6C exer:7A:11
4. P is an orthogonal projection 8.5.3 iff it is self-adjoint (exercise 11 (7A)).
All norms are equivalent?? https://math.stackexchange.com/questions/112985/every-linear-mapping-

833995#833995
All we just learned is about normed vector space, i.e. vector space over the real or complex
numbers, on which a norm is defined. See wiki for more. exer:6A:29
Two norms are equivalent on finite dimensional space?? See exercise 29 (6A).
Every inner product space is a normed space, but not every normed space is an inner product
space. See here.
The motivation for the equivalence of norms on V come from the fact that we want the two
norms to define the same open subsets of V . See here for more. See here for proofs about norms
being equivalent in finite dimensional normed space.
14. Interesting problems

1. How many there are k-dimensional subspaces in n-dimensional vector space V over Fp for
p is a prime. (See here)
Proof. So there are (pn 1)(pn p) · · · (pn pk 1 ) linearly independent lists of length k in
V and since each k-dimensional subspace of V contributes (pk 1)(pk p) · · · (pk pk 1 )
n 1)(pn p)···(pn pk 1 )
such lists, so there are (p
(pk 1)(pk p)···(pk pk 1 )
k-dimensional subspaces of V .
2. (AoPS) Let f 2 Q[X] be a polynomial of degree n > 0.Let p1 , p2 , .., pn+1 be distinct
prime
Pn+1 numbers.Show that there exists a non-zero polynomial g 2 Q[X] such that f g =
c X pi with c 2 Q.
i=1 i i
Proof. Let Pn (Q) be the Q-vector space of polynomials with degree at most n. For each i,
let X pi = f qi +ri where ri is the remainder of X pi modulo f . Since deg f = n so ri 2 Pn (Q)
for all i. On the other hand, since p1 , . . . , pn+1 are primes, this follows r1 , . . . , rn+1 are
linearly independent in Pn (Q), which means P they form a basis of Pn (Q). Therefore, as
f 2 Pn (Q), there exists ci 2 Q such that f = n+1 i=1 ci ri .
Therefore, we have
n+1 n+1 n+1
!
X X X
pi
ci X = ci (f qi + ri ) = f c i qi + 1 = f g.
i=1 i=1 i=1
Since pn+1 > n so deg qn+1 1 so deg g 1.
3. (MSE) Show that any n ⇥ n marix A can be written as sum of two non-singular matrices.
Proof. There are two ways presented in the link.
15. New knowledge

1. Rational canonical form, AoPS
2. Difference between R and C? algebraic closed field?

Linear Al Done Right Note

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Linear Al Done Right Note

Uploaded by

Copyright:

Available Formats

Solutions for exercises/Notes for Linear

Algebra Done Right by Sheldon Axler

Monday 10th September, 2018

3. Chapter 1 - Vector Spaces 4

4. Chapter 2 - Finite-dimensional vector spaces 9

5. Chapter 3: Linear Maps 21

7. Chapter 5: Eigenvalues, Eigenvectors, and Invariant Subspaces 50

8. Chapter 6: Inner Product Spaces 64

9. Chapter 7: Operators on Inner Product Spaces 81

10.Chapter 8: Operators on Complex Vector Spaces 97

10.7. 8D: Jordan Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

11.Chepter 9: Operators on Real Vector Spaces 113

12.Chapter 10: Trace and Determinant 119

14.Interesting problems 123

15.New knowledge 123

1. Some note before reading the book

2. Eigenvectors v1 , . . . , vm corresponding to eigenvalues 1, . . . , m = Eigenvalues 1, . . . , m

3. Chapter 1 - Vector Spaces

2. If a = 0 then it’s obviously true. If a 6= 0 then v = 1

3. Let 13 (w v) = x then 3x = w v so 3x + v = (m v) + v = m. We can see that x in

6. R [ {1} [ { 1} is a vector space over R.

3.2. 1.C Subspaces

(d) Let g, f 2 S then g 0 (2) = f 0 (2) = b. According to Notion 1.23 then f + g 2 S so

Problem 3.2.2 (Page 19). What are subspaces of R2 ?

Problem 3.2.3 (Page 19). What are subspaces of R3 ?

1. If (x1 , y1 , z1 ) 2 S then (x1 , y1 , z1 ) 2 S implying line ` in R3 through the origin and

3. If there exists (x3 , y3 , z3 ) 2 S but not in the plane P then S = R3 .

Proof. It’s not hard to prove that U1 + U2 + . . . + Um is subspace of V . Let U be a subspace

These two argument follows that U1 + . . . + Um is the smallest subspace of V containing

Theorem 3.2.5 (1.44, Condition of a direct sum, page 23)

Proof. If U1 + . . . + Um is a direct sum then obviously 0 is written uniquely as sum 0 + . . . + 0.

In other words, there is another way to write 0 as (v1 u1 ) + . . . + (vm um ) where v1 u1 2 Ui

Theorem 3.2.6 (1.45, Direct sum of two subspaces)

Proof. If U + W is a direct sum. If there exists v 2 U \ W and v 6= 0 then v = ( 1)v 2 U so

3.3. Exercises 1.C

3. Let S be set of such functions. Let f, g 2 S then f + g is a diﬀerentiable real-valued

4. Similar to example 1.35.

6. (a) Since a3 = b3 if and only if a = b so {(a, b, c) 2 R3 : a3 = b3 } = {(a, b, c) 2 R3 : a = b}.

7. Set {(a, b) : a, b 2 Z} is not subspace of R2 .

8. Set U = {(a, b) 2 R2 : a = b} [ {(a, b) 2 R2 : a = 2b} is not subspace of R2 because

10. For any u, v 2 U1 \ U2 , then u + v 2 U1 , u + v 2 U2 so u + v 2 U1 \ U2 . Similar to

11. Similar to 10.

13. Let U1 , U2 , U3 be three subspaces of V so U1 [ U2 [ U3 is also a subspace. If there exists

16. Yes. As long as U, W are sets of vectors of V .

17. Yes. As long as U1 , U2 , U3 are sets of vectors of V .

19. Let V = {(x, y, z) 2 R3 : x, y, z 2 R}, U1 = {(x, 0, 0) 2 R3 : x 2 R}, U2 {(0, x, 0) 2 R3 :

20. W = {(x, 2x, y, 2y) 2 F4 : x, y 2 F}. It is obvious that W + U is a direct sum. It

21. W = {(x, y, z, z, x) 2 F5 : x, y, z 2 F}. Vector 0 can be represented uniquely as w1 + u1

22. W1 = {(0, 0, x, 0, 0) 2 F5 : x 2 F}, W2 = {(0, 0, 0, x, 0) 2 F5 : x 2 F}, W3 = {(0, 0, 0, 0, x) 2

23. Let’s take Exercise 20 as counterexample, V = F4 , W = {(x, x, y, y) 2 F4 : x, y 2 F}, U1 =

24. Since Ue \ Uo = {f (x) = 0} so Ue + Uo is a direct sum. Consider f (x) 2 RR then f (x) =

4. Chapter 2 - Finite-dimensional vector spaces

This theorem is also called as Steinitz exchange lemma or Replacement’s theorem.

Proof. Let v1 , v2 , . . . , vm is linearly independent list and u1 , u2 , . . . , un is spanning list of vectors.

We will show that there exists a1 , . . . , am not all 0 so a1 v1 + . . . + am vm = 0. Since u1 , . . . , un

Thus, there exists a1 , . . . am not all 0 so that a1 v1 + . . . + am vm = 0, a contradiction since

Proof. Let V = span(v1 , v2 , . . . , vm ) be a finite-dimensional vector space and U be its subspace.

for ai , bi 2 F and there exists bi (1  i  m n) so that bi 6= 0. Since ui , ` 2 U so that

Indeed, assume the contrary that after adding at least m n such hi 2 U so hi 6= 0, we

This is a system theo1

4.1.2. Important/Interesting results from Exercise 2.A

4.2. Exercises 2.A

Thus, v 2 span(v1 v2 , v2 v3 , v3 v4 , v4 ). We follow V ✓ span(v1 v2 , v2 v3 , v3