Chapter 6: Inner Product Spaces

Linear Algebra Done Right, by Sheldon Axler

A: Inner Products and Norms

Problem 1


Show that the function that takes (x1 , x2 ), (y1 , y2 ) ∈ R2 ×R2 to |x1 y1 |+
|x2 y2 | is not an inner product on R2 .

Proof. Suppose it were. First notice

⟨(1, 1) + (−1, −1), (1, 1)⟩ = ⟨(0, 0), (1, 1)⟩

= |0 · 1| + |0 · 1|
= 0.

Next, since inner products are additive in the first slot, we also have

⟨(1, 1) + (−1, −1), (1, 1)⟩ = ⟨(1, 1), (1, 1)⟩ + ⟨(−1, −1), (1, 1)⟩
= |1 · 1| + |1 · 1| + |(−1) · 1| + |(−1) · 1|
= 4.

But this implies 0 = 4, a contradiction. Hence we must conclude that the

function does not in fact define an inner product.

Problem 3
Suppose F = R and V = ̸ {0}. Replace the positivity condition (which
states that ⟨v, v⟩ ≥ 0 for all v ∈ V ) in the definition of an inner product
(6.3) with the condition that ⟨v, v⟩ > 0 for some v ∈ V . Show that this
change in the definition does not change the set of functions from V × V
to R that are inner products on V .

Proof. Let V be a nontrivial vector space over R, let A denote the set of functions
V × V → R that are inner products on V in the standard definition, and let B
denote the set of functions V × V → R under the modified definition. We will
show A = B.
Suppose ⟨·, ·⟩1 ∈ A. Since V ̸= {0}, there exists v ∈ V −{0}. Then ⟨v, v⟩1 > 0,
and so ⟨·, ·⟩1 ∈ B. Thus A ⊆ B.
Conversely, suppose ⟨·, ·⟩2 ∈ B. Then there exists some v ′ ∈ V such that

1
⟨v ′ , v ′ ⟩2 > 0. Suppose by way of contradiction there exists u ∈ V is such that
⟨u, u⟩2 < 0. Define w = αu + (1 − α)v ′ for α ∈ R. It follows
⟨w, w⟩2 = ⟨αu + (1 − α)v ′ , αu + (1 − α)v ′ ⟩2
= ⟨αu, αu⟩2 + 2⟨αu, (1 − α)v ′ ⟩2 + ⟨(1 − α)v ′ , (1 − α)v ′ ⟩2
= α2 ⟨u, u⟩2 + 2α(1 − α)⟨u, v ′ ⟩2 + (1 − α)2 ⟨v ′ , v ′ ⟩2 .
Notice the final expression is a polynomial in the indeterminate α, call it p.
Since p(0) = ⟨v ′ , v ′ ⟩2 > 0 and p(1) = ⟨u, u⟩2 < 0, by Bolzano’s theorem there
exists α0 ∈ (0, 1) such that p(α0 ) = 0. That is, if w = α0 u + (1 − α0 )v ′ , then
⟨w, w⟩2 = 0. In particular, notice α0 = ̸ 0, for otherwise w = v ′ , a contradiction
′ ′
since ⟨v , v ⟩2 > 0. Now, since ⟨w, w⟩2 = 0 iff w = 0 (by the definiteness condition
of an inner product), it follows
α0 − 1
u= v.
α0
α0 −1
Letting t = α0 , we now have

⟨u, u⟩2 = ⟨tv ′ , tv ′ ⟩2

= t2 ⟨v ′ , v ′ ⟩2
> 0,
where the inequality follows since t ∈ (−1, 0) and ⟨v ′ , v ′ ⟩2 > 0. This contradicts
our assumption that ⟨u, u⟩2 < 0, and so we have ⟨·, ·⟩2 ∈ A. Therefore, B ⊆ A.
Since we’ve already shown A ⊆ B, this implies A = B, as desired.
Problem 5
√ T ∈ L(V ) is such that ∥T v∥ ≤ ∥v∥
Let V be finite-dimensional. Suppose
for every v ∈ V . Prove that T − 2I is invertible.

√
Proof. Let v ∈ null(T − ̸ 0.
2I), and suppose by way of contradiction that v =
Then
√ √
Tv − 2v = 0 =⇒ T v = 2v
√
=⇒ ∥ 2v∥ ≤ ∥v∥
√
=⇒ 2 · ∥v∥ ≤ ∥v∥
√
=⇒ 2 ≤ 1,
√ √
a contradiction. Hence v = 0 and null(T − 2I) = {0},√so that T − 2I is
injective. Since V is finite-dimensional, this implies T − 2I is invertible, as
desired.

Problem 7
Suppose u, v ∈ V . Prove that ∥au + bv∥ = ∥bu + av∥ for all a, b ∈ R if
and only if ∥u∥ = ∥v∥.

2
Proof. (⇒) Suppose ∥au + bv∥ = ∥bu + av∥ for all a, b ∈ R. Then this equation
holds when a = 1 and b = 0. But then we must have ∥u∥ = ∥v∥, as desired.
(⇐) Conversely, suppose ∥u∥ = ∥v∥. Let a, b ∈ R be arbitrary, and notice
∥au + bv∥ = ⟨au + bv, au + bv⟩
= ⟨au, au⟩ + ⟨au, bv⟩ + ⟨bv, au⟩ + ⟨bv, bv⟩
= a2 ∥u∥2 + ab ⟨u, v⟩ + ⟨v, u⟩ + b2 ∥v∥2 .


(1)
Also, we have
∥bu + av∥ = ⟨bu + av, bu + av⟩
= ⟨bu, bu⟩ + ⟨bu, av⟩ + ⟨av, bu⟩ + ⟨av, av⟩
= b2 ∥u∥2 + ab ⟨u, v⟩ + ⟨v, u⟩ + a2 ∥v∥2 .


(2)
Since ∥u∥ = ∥v∥, (1) equals (2), and hence ∥au + bv∥ = ∥bu + av∥. Since a, b
were arbitrary, the result follows.
Problem 9
Suppose u, v ∈ V and ∥u∥ ≤ 1 and ∥v∥ ≤ 1. Prove that
p p
1 − ∥u∥2 1 − ∥v∥2 ≤ 1 − |⟨u, v⟩|.

Proof. By the Cauchy-Schwarz Inequality, we have |⟨u, v⟩| ≤ ∥u∥∥v∥. Since

∥u∥ ≤ 1 and ∥v∥ ≤ 1, this implies
0 ≤ 1 − ∥u∥∥v∥ ≤ 1 − |⟨u, v⟩|,
and hence it’s enough to show
p p
1 − ∥u∥2 1 − ∥v∥2 ≤ 1 − ∥u∥∥v∥.
Squaring both sides, it suffices to prove
  
2
1 − ∥u∥2 1 − ∥v∥2 ≤ 1 − ∥u∥∥v∥ . (3)

Notice

2   
1 − ∥u∥∥v∥ − 1 − ∥u∥2 1 − ∥v∥2 = ∥u∥2 − 2∥u∥∥v∥ + ∥v∥2

2
= ∥u∥ − ∥v∥
≥ 0,
and hence inequality (3) holds, completing the proof.

Problem 11
Prove that  
1 1 1 1
16 ≤ (a + b + c + d) + + +
a b c d
for all positive numbers a, b, c, d.

3
Proof. Define
√ √ √ √  r r r r !
1 1 1 1
x= a, b, c, d and y = , , , .
a b c d
Then the Cauchy-Schwarz Inequality implies
r !2
1 √ 1 √
r r r
√ 1 √
 
1 1 1 1 1
(a + b + c + d) + + + ≥ a + b + c + d
a b c d a b c d
= (1 + 1 + 1 + 1)2
= 16,
as desired.

Problem 13
Suppose u, v are nonzero vectors in R2 . Prove that

⟨u, v⟩ = ∥u∥∥v∥ cos θ,

where θ is the angle between u and v (thinking of u and v as arrows with

initial point at the origin).

Proof. Let A denote the line segment from the origin to u, let B denote the line
segment from the origin to v, and let C denote the line segment from v to u.
Then A has length ∥u∥, B has length ∥v∥ and C has length ∥u − v∥. Letting θ
denote the angle between A and B, by the Law of Cosines we have
C 2 = A2 + B 2 − 2BC cos θ,
or equivalently
∥u − v∥2 = ∥u∥2 + ∥v∥2 − 2∥u∥∥v∥ cos θ.
It follows
2∥u∥∥v∥ cos θ = ∥u∥2 + ∥v∥2 − ∥u − v∥2
= ⟨u, u⟩ + ⟨v, v⟩ − ⟨u − v, u − v⟩


= ⟨u, u⟩ + ⟨v, v⟩ − ⟨u, u⟩ − 2⟨u, v⟩ + ⟨v, v⟩
= 2⟨u, v⟩.
Dividing both sides by 2 gives the desired result.
Problem 15
Prove that  2   
n n n 2
X X X bj 
 a j bj  ≤  jaj 2  
j=1 j=1 j=1
j

for all real numbers a1 , . . . , an and b1 , . . . , bn .

4
Proof. Let
 √ √  
1 1

u = a1 , 2a2 , . . . , nan and v = b1 , √ b2 , . . . , √ bn .
2 n
Pn
Since ⟨u, v⟩ = k=1 ak bk , the Cauchy-Schwarz Inequality yields
2
(a1 b1 + · · · + an bn ) ≤ ∥u∥2 ∥v∥2
!

2 2 2

2 b2 2 bn 2
= a1 + 2a2 + · · · + nan b1 + + ··· + ,
2 n

as desired.

Problem 17
Prove or disprove: there is an inner product on R2 such that the associated
norm is given by
∥(x, y)∥ = max{|x|, |y|}
for all (x, y) ∈ R2 .

Proof. Suppose such an inner product existed. Then by the Parallelogram

Equality, it follows
 
∥(1, 0) + (0, 1)∥2 + ∥(1, 0) − (0, 1)∥2 = 2 ∥(1, 0)∥2 + ∥(0, 1)∥2 .

After simplification, this implies 2 = 4, a contradiction. Hence no such inner

product exists.

Problem 19
Suppose V is a real inner product space. Prove that

∥u + v∥2 − ∥u − v∥2
⟨u, v⟩ =
4
for all u, v ∈ V .

Proof. Suppose V is a real inner product space and let u, v ∈ V . It follows

∥u + v∥2 − ∥u − v∥2 ∥u∥2 + 2⟨u, v⟩ + ∥v∥2 − ∥u∥2 − 2⟨u, v⟩ + ∥v∥2
=
4 4
4⟨u, v⟩
=
4
= ⟨u, v⟩,

as desired.

5
 Problem 20
Suppose V is a complex inner product space. Prove that

∥u + v∥2 − ∥u − v∥2 + ∥u + iv∥2 i − ∥u − iv∥2 i

⟨u, v⟩ =
4
for all u, v ∈ V .

Proof. Notice we have

∥u + v∥2 = ⟨u + v, u + v⟩
= ∥u∥2 + ⟨u, v⟩ + ⟨v, u⟩ + ∥v∥2

and

−∥u − v∥2 = −⟨u − v, u − v⟩

= −∥u∥2 + ⟨u, v⟩ + ⟨v, u⟩ − ∥v∥2 .

Also, we have

∥u + iv∥2 i = i ⟨u + iv, u + iv⟩

 
= i ∥u∥2 + ⟨u, iv⟩ + ⟨iv, u⟩ + ⟨iv, iv⟩
 
= i ∥u∥2 − i⟨u, v⟩ + i⟨v, u⟩ + ∥v∥2
= i∥u∥2 + ⟨u, v⟩ − ⟨v, u⟩ + i∥v∥2

and

−∥u − iv∥2 i = −i ⟨u − iv, u − iv⟩

 
= −i ∥u∥2 − ⟨u, iv⟩ − ⟨iv, u⟩ + ⟨iv, iv⟩
 
= −i ∥u∥2 + i⟨u, v⟩ − i⟨v, u⟩ + ∥v∥2
= −i∥u∥2 + ⟨u, v⟩ − ⟨v, u⟩ − i∥v∥2 .

Thus it follows

∥u + v∥2 − ∥u − v∥2 + ∥u + iv∥2 i − ∥u − iv∥2 i = 4⟨u, v⟩.

Dividing both sides by 4 yields the desired result.

Problem 23
Suppose V1 , . . . , Vm are inner product spaces. Show that the equation

⟨(u1 , . . . , um ), (v1 , . . . , vm )⟩ = ⟨u1 , v1 ⟩ + · · · + ⟨um , vm ⟩

defines an inner product on V1 × · · · × Vm .

6
Proof. We prove that this definition satisfies each property of an inner product
in turn.
Positivity: Let (v1 , . . . , vm ) ∈ V1 × . . . Vm . Since ⟨vk , vk ⟩ is an inner product
on Vk for k = 1, . . . , m, we have ⟨vk , vk ⟩ ≥ 0. Thus
⟨(v1 , . . . , vm ), (v1 , . . . , vm )⟩ = ⟨v1 , v1 ⟩ + · · · + ⟨vm , vm ⟩ ≥ 0.
Definiteness: First suppose ⟨(v1 , . . . , vm ), (v1 , . . . , vm )⟩ = 0 for (v1 , . . . , vm ) ∈
V1 × · · · × Vm . Then
⟨v1 , v1 ⟩ + · · · + ⟨vm , vm ⟩ = 0.
By positivity of each inner product on Vk (for k = 1, . . . , m), we must have
⟨vk , vk ⟩ ≥ 0. Thus the equation above holds only if ⟨vk , vk ⟩ = 0 for each k,
which is true iff vk = 0 (by definiteness of the inner product on Vk ). Hence
(v1 , . . . , vm ) = (0, . . . , 0). Conversely, suppose (v1 , . . . , vm ) = (0, . . . , 0). Then
⟨(v1 , . . . , vm ), (v1 , . . . , vm )⟩ = ⟨v1 , v1 ⟩ + · · · + ⟨vm , vm ⟩
= ⟨0, 0⟩ + · · · + ⟨0, 0⟩
= 0 + ··· + 0
= 0,
where the third equality follows from definiteness of the inner product on each
Vk , respectively.
Additivity in first slot: Let
(u1 , . . . , um ), (v1 , . . . , vm ), (w1 , . . . , wm ) ∈ V1 × · · · × Vm .
It follows
⟨(u1 , . . . , um )+(v1 , . . . , vm )), (w1 , . . . , wm )⟩
= ⟨(u1 + v1 , . . . , um + vm ), (w1 , . . . , wm )⟩
= ⟨u1 + v1 , w1 ⟩ + · · · + ⟨um + vm , wm ⟩
= ⟨u1 , w1 ⟩ + ⟨v1 , w1 ⟩ + · · · + ⟨um , wm ⟩ + ⟨vm , wm ⟩
= ⟨(u1 , . . . , um ), (w1 , . . . , wm )⟩ + ⟨(v1 , . . . , vm ), (w1 , . . . , wm )⟩,
where the third equality follows from additivity in the first slot of each inner
product on Vk , respectively.
Homogeneity in the first slot: Let λ ∈ F and
(u1 , . . . , um ), (v1 , . . . , vm ) ∈ V1 × · · · × Vm .
It follows
⟨λ(u1 , . . . , um ), (v1 , . . . , vm )⟩ = ⟨(λu1 , . . . , λum ), (v1 , . . . , vm )⟩
= ⟨λu1 , v1 ⟩ + · · · + ⟨λum , vm ⟩
= λ⟨u1 , v1 ⟩ + · · · + λ⟨um , vm ⟩
= λ(⟨u1 , v1 ⟩ + · · · + ⟨um , vm ⟩)
= λ⟨(u1 , . . . , um ), (v1 , . . . , vm )⟩,

7
where the third equality follows from homogeneity in the first slot of each inner
product on Vk , respectively.
Conjugate symmetry: Again let

(u1 , . . . , um ), (v1 , . . . , vm ) ∈ V1 × · · · × Vm .

It follows

⟨(u1 , . . . , um ), (v1 , . . . , vm )⟩ = ⟨u1 , v1 ⟩ + · · · + ⟨um , vm ⟩

= ⟨v1 , u1 ⟩ + · · · + ⟨vm , um ⟩
= ⟨u1 , v1 ⟩ + · · · + ⟨um , vm ⟩
= ⟨(v1 , . . . , vm ), (u1 , . . . , um )⟩,

where the second equality follows from conjugate symmetry of each inner product
on Vk , respectively.

Problem 24
Suppose S ∈ L(V ) is an injective operator on V . Define ⟨·, ·⟩1 by

⟨u, v⟩1 = ⟨Su, Sv⟩

for u, v ∈ V . Show that ⟨·, ·⟩1 is an inner product on V .

Proof. We prove that this definition satisfies each property of an inner product
in turn.
Positivity: Let v ∈ V . Then ⟨v, v⟩1 = ⟨Sv, Sv⟩ ≥ 0.
Definiteness: Suppose ⟨v, v⟩ = 0 for some v ∈ V . This is true iff ⟨Sv, Sv⟩ = 0
(by definition) which is true iff Sv = 0 (by definiteness of ⟨·, ·⟩), which is true iff
v = 0 (since S is injective).
Additivity in first slot: Let u, v, w ∈ V . Then

⟨u + v, w⟩1 = ⟨S(u + v), Sw⟩

= ⟨Su + Sv, Sw⟩
= ⟨Su, Sw⟩ + ⟨Sv, Sw⟩
= ⟨u, w⟩1 + ⟨v, w⟩1 .

Homogeneity in first slot: Let λ ∈ F and u, v ∈ V . Then

⟨λu, v⟩1 = ⟨S(λu), Sv⟩

= ⟨λSu, Sv⟩
= λ⟨Su, Sv⟩
= λ⟨u, v⟩1 .

8
Conjugate symmetry Let u, v ∈ V . Then

⟨u, v⟩1 = ⟨Su, Sv⟩

= ⟨Sv, Su⟩
= ⟨v, u⟩1 .

Problem 25
Suppose S ∈ L(V ) is not injective. Define ⟨·, ·⟩1 as in the exercise above.
Explain why ⟨·, ·⟩1 is not an inner product on V .

Proof. If S is not injective, then ⟨·, ·⟩1 fails the definiteness requirement in the
definition of an inner product. In particular, there exists v ̸= 0 such that Sv = 0.
Hence ⟨v, v⟩1 = ⟨Sv, Sv⟩ = 0 for a nonzero v.

Problem 27
Suppose u, v, w ∈ V . Prove that
2 2 2 2
1 ∥w − u∥ +∥w − v∥ ∥u − v∥
w − (u + v) = − .
2 2 4

Proof. We have
2     2
1 w−u w−v
w − (u + v) = +
2 2 2
2 2     2
w−u w−v w−u w−v
=2 +2 − −
2 2 2 2
2 2 2
∥w − u∥ +∥w − v∥ −u + v
= −
2 2
2 2 2
∥w − u∥ +∥w − v∥ ∥u − v∥
= − ,
2 4
where the second equality follows by the Parallelogram Equality.
The next problem requires some extra work to prove. We first include a
definition and prove a theorem.
Definition. Suppose ∥·∥1 and ∥·∥2 are norms on vector space V . We say ∥·∥1
and ∥·∥2 are equivalent if there exist 0 < C1 ≤ C2 such that

C1 ∥v∥1 ≤ ∥v∥2 ≤ C2 ∥v∥1

for all v ∈ V .

9
Theorem. Any two norms on a finite-dimensional vector space are equivalent.
Proof. Let V be finite-dimensional with basis e1 , . . . , en . It suffices to prove that
every norm on V is equivalent to the ℓ1 -style norm ∥·∥1 defined by

∥v∥1 = |α1 | + · · · +|αn |

for all v = α1 e1 + · · · + αn en ∈ V .
Let ∥·∥ be a norm on V . We wish to show C1 ∥v∥1 ≤ ∥v∥ ≤ C2 ∥v∥1 for all
v ∈ V and some choice of C1 , C2 . Since this is trivially true for v = 0, we need
only consider v ̸= 0, in which case we have

C1 ≤ ∥u∥ ≤ C2 , (*)

where u = v/∥v∥1 . Thus it suffices to consider only vectors v ∈ V such that

∥v∥1 = 1.
We will now show that ∥·∥ is continuous under ∥·∥1 and apply the Extreme
Value Theorem to deduce the desired result. So let ϵ > 0 and define M =
max{∥e1 ∥ , . . . ,∥en ∥} and
ϵ
δ= .
M
It follows that if u, v ∈ V are such that ∥u − v∥1 < δ, then

∥u∥ −∥v∥ ≤ ∥u − v∥
≤ M ∥u − v∥1
≤ Mδ
= ϵ,

and ∥·∥ is indeed continuous under the topology induced by ∥·∥1 . Let S = {u ∈
V | ∥u∥1 = 1} (the unit sphere with respect to ∥·∥1 ). Since S is compact and ∥·∥
is continuous on it, by the Extreme Value Theorem we may define

C1 = min∥u∥ and C2 = max∥u∥ .

u∈S u∈S

But now C1 and C2 satisfy (*), completing the proof.

Problem 29
For u, v ∈ V , define d(u, v) = ∥u − v∥.
(a) Show that d is a metric on V .

(b) Show that if V is finite-dimensional, then d is a complete metric on

V (meaning that every Cauchy sequence converges).
(c) Show that every finite-dimensional subspace of V is a closed subset
of V (with respect to the metric d).

10
Proof. (a) We show that d satisfies each property of the definition of a metric
in turn.
Identity of indiscernibles: Let u, v ∈ V . It follows
p
d(u, v) = 0 ⇐⇒ ⟨u − v, u − v⟩ = 0
⇐⇒ ⟨u − v, u − v, =⟩0
⇐⇒ u − v = 0
⇐⇒ u = v.

Symmetry: Let u, v ∈ V . We have

d(u, v) = ∥u − v∥
= (−1)(u − v)
= ∥v − u∥
= d(v, u).

Triangle inequality: Let u, v, w ∈ V . Notice

d(u, v) + d(v, w) = ∥u − v∥ +∥v − w∥

≤ (u − v) + (v − w)
= ∥u, w∥
= d(u, w).

(b) Suppose V is a p-dimensional vector space with basis e1 , . . . , ep . Assume

{vk }∞
k=1 is Cauchy. Then for ϵ > 0, there exists N ∈ Z
+
such that
∥vm − vn ∥ < ϵ whenever m, n > N . Given any vi in our Cauchy sequence,
we adopt the notation that αi,1 , . . . , αi,p ∈ F are always defined such that

vi = αi,1 e1 + · · · + αi,p ep .

By our previous theorem, ∥·∥ is equivalent to ∥·∥1 (where ∥·∥1 is defined in

that theorem’s proof). Thus there exists some c > 0 such that, whenever
m, n > N , we have

c∥vm − vn ∥1 ≤ ∥vm − vn ∥ < ϵ,

and hence  
Xp
c αm,i − αn,i  < ϵ.
i=1

This implies that {αk,i }∞

k=1 is Cauchy in R for each i = 1, . . . , p. Since R
is complete, these sequences converge. So let αi = limk→∞ αk,i for each i,
and define v = α1 e1 + · · · + αp ep . It follows

vj − v = (αj,1 − β1 )e1 + · · · + (αj,p − βp )ep

≤ αj,1 − α1 ∥e1 ∥ + · · · + αj,p − αp ep .

11
 Since αj,i → αi for i = 1, . . . , p, the RHS can be made arbitrarily small by
choosing sufficiently large M ∈ Z+ and considering j > M . Thus {vk }∞ k=1
converges to v, and V is indeed complete with respect to ∥·∥.
(c) Suppose U is a finite-dimensional subspace of V , and suppose {uk }∞
k=1 ⊆ U
is Cauchy. By (b), limk→∞ uk ∈ U , hence U contains all its limit points.
Thus U is closed.

Problem 31
Use inner products to prove Apollonius’s Identity: In a triangle with
sides of length a, b, and c, let d be the length of the line segment from
the midpoint of the side of length c to the opposite vertex. Then
1 2
a2 + b2 = c + 2d2 .
2

Proof. Consider a triangle formed by vectors v, w ∈ R2 and the origin such that
∥w∥ = a, ∥v∥ = c, and ∥w − v∥ = b. The identity follows by applying Problem 27
with u = 0.

B: Orthonormal Bases
Problem 1
(a) Suppose θ ∈ R. Show that (cos θ, sin θ), (− sin θ, cos θ) and
(cos θ, sin θ), (sin θ, − cos θ) are orthonormal bases of R2 .
(b) Show that each orthonormal basis of R2 is of the form given by one
of the two possibilities of part (a).

Proof. (a) Notice

⟨(cos θ, sin θ), (− sin θ, cos θ)⟩ = − sin θ cos θ + sin θ cos θ = 0

and

⟨(cos θ, sin θ), (sin θ, − cos θ)⟩ = sin θ cos θ − sin θ cos θ = 0,

hence both lists are orthonormal. Clearly the three distinct vectors
contained in the two lists all have norm 1 (following from the identity
cos2 θ + sin2 θ = 1). Since both lists have length 2, by Theorem 6.28 both
lists are orthonormal bases.
(b) Suppose e1 , e2 is an orthonormal basis of R2 . Since ∥e1 ∥ = ∥e2 ∥ = 1, there
exist θ, φ ∈ [0, 2π) such that

e1 = (cos θ, sin θ) and e2 = (cos φ, sin φ).

12
 Next, since ⟨e1 , e2 ⟩ = 0, we have
cos θ cos φ + sin θ sin φ = 0.
1
Since cos θ cos φ = 2 (cos(θ + φ) + cos(θ − φ)) and sin θ sin φ = cos(θ − φ) −
cos(θ + φ), the above implies
cos(θ − φ) = 0
and thus φ = θ + 3π 2 − nπ, for n ∈ Z. Since θ, φ ∈ [0, 2π), this implies
φ = θ ± π2 . If φ = θ + π2 , then
   !
π π
e2 = cos θ + , sin θ +
2 2
= (− sin θ, cos θ),
and if φ = θ − π2 , then
   !
π π
e2 = cos θ − , sin θ −
2 2
= (sin θ, − cos θ).
Thus all orthonormal bases of R2 have one of the two forms from (a).

Problem 3
Suppose T ∈ L(R3 ) has an upper-triangular matrix with respect to
the basis (1, 0, 0), (1, 1, 1), (1, 1, 2). Find an orthonormal basis of R3
(use the usual inner product on R3 ) with respect to which T has an
upper-triangular matrix.

Proof. Let v1 = (1, 0, 0), v2 = (1, 1, 1), and v3 = (1, 1, 2). By the proof of 6.37, T
has an upper-triangular matrix with respect to the the basis e1 , e2 , e3 generated
by applying the Gram-Schmidt Procedure to v1 , v2 , v3 . Since ∥v1 ∥ = 1, e1 = v1 .
Next, we have
v2 − ⟨v2 , e1 ⟩e1
e2 =
v2 − ⟨v2 , e1 ⟩e1
(1, 1, 1) − ⟨(1, 1, 1), (1, 0, 0)⟩(1, 0, 0)
=
(1, 1, 1) − ⟨(1, 1, 1), (1, 0, 0)⟩(1, 0, 0)
(1, 1, 1) − (1, 0, 0)
=
(1, 1, 1) − (1, 0, 0)
(0, 1, 1)
=
(0, 1, 1)
 
1 1
= 0, √ , √
2 2

13
and
v3 − ⟨v3 , e1 ⟩e1 − ⟨v3 , e2 ⟩e2
e3 =
v3 − ⟨v3 , e1 ⟩e1 − ⟨v3 , e2 ⟩e2
    
(1, 1, 2) − ⟨(1, 1, 2), (1, 0, 0)⟩(1, 0, 0) − (1, 1, 2), 0, √1 , √1 0, √12 , √12
2 2
=     
(1, 1, 2) − ⟨(1, 1, 2), (1, 0, 0)⟩(1, 0, 0) − (1, 1, 2), 0, √12 , √12 0, √12 , √12
 
(1, 1, 2) − (1, 0, 0) − √3 0, √12 , √12
2
=  
(1, 1, 2) − (1, 0, 0) − √3 0, √12 , √12
2

0, − 12 , 21


=
0, − 12 , 12

√ √ !
2 2
= 0, − , ,
2 2

and we’re done.

Problem 4
Suppose n is a positive integer. Prove that
1 cos x cos 2x cos nx sin x sin 2x sin nx
√ , √ , √ ,..., √ , √ , √ ,..., √
2π π π π π π π

is an orthonormal list of vectors in C[−π, π], the vector space of continuous

real-valued functions on [−π, π] with inner product
Z π
⟨f, g⟩ = f (x)g(x)dx.
−π

Proof. First we show that all vectors in the list have norm 1. Notice
sZ
π
1 1
√ = dx
2π −π 2π
s Z π
1
= dx
2π −π
= 1.

14
And for k ∈ Z+ , we have
s Z π
cos(kx) 1
√ = cos(kx)2 dx
π π −π
s  π
1 sin(2kx) x
= +
π 4k 2 −π
v " #
u
u1 π  π
= t − −
π 2 2

= 1,

and
s Z π
sin(kx) 1
√ = sin(kx)2 dx
π π −π
s
 π
1 x cos(2kx)
= −
π 2 4k −π
v " #
u  
u1 π π
=t − −
π 2 2

= 1,

so indeed all vectors have norm 1. Now we show them to be pairwise orthogonal.
Suppose j, k ∈ Z are such that j ̸= k. It follows from basic calculus

1 π
  Z
sin(jx) sin(kx)
√ , √ = sin(jx) sin(kx)dx
π π π −π
 π
1 k sin(jx) cos(kx) + j cos(jx) sin(kx)
=
π j 2 − k2 −π
= 0,

  Z π
sin(jx) cos(kx) 1
√ , √ = sin(jx) cos(kx)dx
π π π −π
 π
1 k sin(jx) sin(kx) + j cos(jx) cos(kx)
=−
π j 2 − k2 −π
"   #
1 j cos(jπ) cos(kπ) j cos(−jπ) cos(−kπ)
=− −
π j 2 − k2 j 2 − k2
= 0,

15
   Z π
cos(jx) cos(kx) 1
√ , √ = cos(jx) cos(kx)dx
π π π −π
 π
1 j sin(jx) cos(kx) − k cos(jx) sin(kx)
=
π j 2 − k2 −π
= 0,

  Z π
sin(jx) cos(jx) 1
√ , √ = sin(jx) cos(jx)dx
π π π −π
" #π
cos2 (jx)
= −
2j
−π
= 0,

  Z π
1 cos(jx) 1
√ , √ =√ cos(jx)dx
2π π 2π −π
 π
1 sin(jx)
= √
2π j −π
= 0,

and
  Z π
1 sin(jx) 1
√ , √ =√ sin(jx)dx
2π π 2π −π
 π
1 cos(jx)
=√ −
2π j −π
 
1 cos(jπ) − cos(−jπ)
=√ −
2π j
= 0.

Thus the list is indeed an orthonormal list in C[−π, π].

Problem 5
On P2 (R), consider the inner product given by
Z 1
⟨p, q⟩ = p(x)q(x) dx.
0

Apply the Gram-Schmidt Procedure to the basis 1, x, x2 to produce an

orthonormal basis of P2 (R).

16
Proof. First notice ∥1∥ = 1, hence e1 = 1. Next notice

v2 − ⟨v1 , e1 ⟩e1 = x − ⟨x, 1⟩

Z 1
=x− x dx
0
1
=x−
2
and
s 
1 1 1
x− = x − ,x −
2 2 2
s
Z 1  
1 1
= x− x− dx
0 2 2
s
Z 1 
2
1
= x −x+ dx
0 4
r
1 1 1
= − +
3 2 4
1
= √ ,
2 3
and therefore we have
√
 
1
e2 = 2 3 x − .
2
To compute e3 , first notice
"  #
1 √ Z 1
Z 
2 2 2 1
v3 − ⟨v3 , e1 ⟩e1 − ⟨v3 , e2 ⟩e2 = x − x dx − 2 3 x x− dx e2
0 0 2
 ! " #
√ Z 1 2 √

2 1  3 x 1
=x − − 2 3 x − dx 2 3 x −
3 0 2 2
  
1 1 1 1
= x2 − − 12 − x−
3 4 6 2
 
2 1 1
=x − − x−
3 2
1
= x2 − x +
6

17
and
s 
1 1 2 1
x2 − x + = x2
− x + ,x − x +
6 6 6
s
Z 1  
1 1
= x2 − x + x2 − x + dx
0 6 6
s
Z 1 
4 3
4 2 x 1
= x − 2x + x − + dx
0 3 3 36
r
1 1 4 1 1
= − + − +
5 2 9 6 36
1
=√
180
1
= √ .
6 5
Thus
√
 
1
e3 = 6 5 x2 − x + ,
6
and we’re done.

Problem 7
Find a polynomial q ∈ P2 (R) such that
  Z 1
1
p = p(x)q(x) dx
2 0

for every p ∈ P2 (R).

R1
Proof. Consider the inner product ⟨p, q⟩ = 0 p(x)q(x) dx on P2 (R). Define
φ ∈ L(P2 (R)) by φ(p) = p 21 and let e1 , e2 , e3 be the orthonormal basis found

in Problem 5. By the Riesz Representation Theorem, there exists q ∈ P2 (R)

such that φ(p) = ⟨p, q⟩ for all p ∈ P2 (R). That is, such that
  Z 1
1
p = p(x)q(x) dx.
2 0

Equation 6.43 in the proof of the Riesz Representation Theorem fashions a way

18
to find q. In particular, we have

q(x) = φ(e1 ) e1 + φ(e2 ) e2 + φ(e3 ) e3

√ √
   
1 1 1 1 1
= e1 + 2 3 − e2 + 6 5 − + e3
2 2 4 2 6
" #
√ √
 
−1 1
=1+6 5 6 5 x2 − x +
12 6
3
= −15(x2 − x) − ,
2
as desired.

Problem 9
What happens if the Gram-Schmidt Procedure is applied to a list of
vectors that is not linearly independent?

Proof. Suppose v1 , . . . , vm are linearly dependent. Let j be the smallest integer

in {1, . . . , m} such that vj ∈ span(v1 , . . . , vj−1 ). Then v1 , . . . , vj−1 are linearly
independent. The first j − 1 steps of the Gram-Schmidt Procedure will produce
an orthonormal list e1 , . . . , ej−1 . At step j, however, notice

vj − ⟨vj , e1 ⟩e1 − · · · − ⟨vj , ej−1 ⟩ej−1 = vj − vj = 0,

and we are left trying to assign ej to 00 , which is undefined. Thus the procedure
cannot be applied to a linearly dependent list.

Problem 11
Suppose ⟨·, ·⟩1 and ⟨·, ·⟩2 are inner products on V such that ⟨v, w⟩1 = 0
if and only if ⟨v, w⟩2 = 0. Prove that there is a positive number c such
that ⟨v, w⟩1 = c⟨v, w⟩2 for every v, w ∈ V .

Proof. Let v, w ∈ V be arbitrary. By hypothesis, if v and w are orthogonal

relative to one of the inner products, they’re orthogonal relative to the other.
Hence any choice of c ∈ R would satisfy ⟨v, w⟩1 = c⟨v, w⟩2 . So suppose v and w
are not orthogonal relative to either inner product. Then both v and w must
be nonzero (by Theorem 6.7, parts b and c, respectively). Thus ⟨v, v⟩1 , ⟨w, w⟩1 ,

19
⟨v, v⟩2 , and ⟨w, w⟩2 are all nonzero as well. It now follows
⟨v, w⟩1
0 = ⟨v, w⟩1 − ⟨v, v⟩1
⟨v, v⟩1
*   +
⟨v, w⟩1
= ⟨v, w⟩1 − v, v
⟨v, v⟩1
1
*   +
⟨v, w⟩1
= v, w − v
⟨v, v⟩1
1
*   +
⟨v, w⟩1
= v, w − v
⟨v, v⟩1
2
*   +
⟨v, w⟩1
= ⟨v, w⟩2 − v, v
⟨v, v⟩1
2
⟨v, w⟩1
= ⟨v, w⟩2 − ⟨v, v⟩2
⟨v, v⟩1
⟨v, v⟩2
= ⟨v, w⟩2 − ⟨v, w⟩1 ,
⟨v, v⟩1
where the fifth equality follows by our hypothesis. Thus
2
∥v∥1
⟨v, w⟩1 = 2 ⟨v, w⟩2 . (4)
∥v∥2
By a similar computation, notice
⟨v, w⟩1
0 = ⟨v, w⟩1 − ⟨w, w⟩1
⟨w, w⟩1
 
⟨v, w⟩1
= ⟨v, w⟩1 − w, w
⟨w, w⟩1 1
 
⟨v, w⟩1
= v− w, w
⟨w, w⟩1
 1
⟨v, w⟩1
= v− w, w
⟨w, w⟩1 2
 
⟨v, w⟩1
= ⟨v, w⟩2 − w, w
⟨w, w⟩2 2
⟨v, w⟩1
= ⟨v, w⟩2 − ⟨w, w⟩2
⟨w, w⟩2
⟨w, w⟩2
= ⟨v, w⟩2 − ⟨v, w⟩1 ,
⟨w, w⟩1
and thus
2
∥w∥1
⟨v, w⟩1 = 2 ⟨v, w⟩2 (5)
∥w∥2

20
as well. By combining Equations (4) and (5), we conclude

⟨v, v⟩1 ⟨w, w⟩1

= .
⟨v, v⟩2 ⟨w, w⟩2
2 2
Since v and w were arbitrary nonzero vectors in V , choosing c = ∥u∥1 /∥u∥2
for any u ̸= 0 guarantees ⟨v, w⟩1 = c⟨v, w⟩2 for every v, w ∈ V , as was to be
shown.

Problem 13
Suppose v1 , . . . , vm is a linearly independent list in V . Show that there
exists w ∈ V such that ⟨w, vj ⟩ > 0 for all j ∈ {1, . . . , m}.

Proof. Let W = span(v1 , . . . , vm ). Given v ∈ W , let a1 , . . . , am ∈ F be such

that v = a1 v1 + · · · + am vm . Define φ ∈ L(W ) by

φ(v) = a1 + · · · + am .

By the Riesz Representation Theorem, there exists w ∈ W such that φ(v) =

⟨v, w⟩ for all v ∈ W . But then φ(vj ) = 1 for j ∈ {1, . . . , m}, and indeed such a
w ∈ V exists.

Problem 15
Suppose CR ([−1, 1]) is the vector space of continuous real-valued functions
on the interval [−1, 1] with inner product given by
Z 1
⟨f, g⟩ = f (x)g(x)dx
−1

for f, g ∈ CR ([−1, 1]). Let φ be the linear functional on CR ([−1, 1])

defined by φ(f ) = f (0). Show that there does not exist g ∈ CR ([−1, 1])
such that
φ(f ) = ⟨f, g⟩
for every f ∈ CR ([−1, 1]).

Proof. Suppose not. Then there exists g ∈ CR ([−1, 1]) such that

φ(f ) = ⟨f, g⟩

for every f ∈ CR ([−1, 1]). Choose f (x) = x2 g(x). Then f (0) = 0, and hence
Z 1 Z 1
f (x)g(x)dx = [xg(x)]2 dx = 0.
−1 −1

Now, let h(x) = xg(x). Since h is continuous on [−1, 1], there exists an interval
[a, b] ⊆ [−1, 1] such that h(x) ̸= 0 for all x ∈ [a, b]. By the Extreme Value

21
Theorem, h(x)2 has a minimum at some m ∈ [a, b]. Thus h(m)2 > 0, and we
now conclude
Z 1 Z b
0= h(x)2 dx = h(x)2 dx ≥ (b − a)h(m)2 > 0,
−1 a

which is absurd. Thus it must be that no such g exists.

Problem 17
For u ∈ V , let Φu denote the linear functional on V defined by

(Φu )(v) = ⟨v, u⟩

for v ∈ V .

(a) Show that if F = R, then Φ is a linear map from V to V ′ .

(b) Show that if F = C and V ̸= {0}, then Φ is not a linear map.
(c) Show that Φ is injective.
(d) Suppose F = R and V is finite-dimensional. Use parts (a) and (c)
and a dimension-counting argument (but without using 6.42) to
show that Φ is an isomorphism from V to V ′ .

Proof. (a) Suppose F = R. Let u, u′ ∈ V and α ∈ R. Then, for all v ∈ V , we

have
Φu+u′ (v) = ⟨v, u + u′ ⟩ = ⟨v, u⟩ + ⟨v, u′ ⟩ = Φu (v) + Φu′ (v)
and
Φαu (v) = ⟨v, αu⟩ = α⟨v, u⟩ = α⟨v, u⟩ = αΦu (v).
Thus Φ is indeed a linear map.
(b) Suppose F = C and V ̸= {0}. Let u ∈ V . Then, given v ∈ V , we have
Φiu (v) = ⟨v, iu⟩ = i⟨v, u⟩,
whereas
iΦu (v) = i⟨v, u⟩.
Thus Φiu ̸= iΦu , and indeed Φ is not a linear map, since is is not homoge-
neous.
(c) Suppose u, u′ ∈ V are such that Φu = Φu′ . Then, for all v ∈ V , we have
Φu (v) = Φu′ (v)
=⇒ ⟨v, u⟩ = ⟨v, u′ ⟩
=⇒ ⟨v, u⟩ − ⟨v, u′ ⟩ = 0
=⇒ ⟨v, u − u′ ⟩ = 0.

22
 In particular, choosing v = u − u′ , the above implies ⟨u − u′ , u − u′ ⟩ = 0,
which is true iff u − u′ = 0. Thus we conclude u = u′ , so that Φ is indeed
injective.
(d) Suppose F = R and dim V = n. Notice that since Φ : V ,→ V ′ , we have

dim V = dim null Φ + dim range Φ = dim range Φ.

Thus Φ is surjective as well, and we have V ∼

= V ′ , as was to be shown.

C: Orthogonal Complements and Minimization

Problems
Problem 1
Suppose v1 , . . . , vm ∈ V . Prove that

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

Proof. Suppose v ∈ {v1 , . . . , vm }⊥ . Then ⟨v, vk ⟩ = 0 for k = 1, . . . , m. Let

u ∈ span(v1 , . . . , vm ) be arbitrary. We want to show ⟨v, u⟩ = 0, since this implies
v ∈ (span(v1 , . . . , vm ))⊥ and hence {v1 , . . . , vm }⊥ ⊆ (span(v1 , . . . , vm ))⊥ . To
see this, notice

⟨v, u⟩ = ⟨v, α1 v1 + · · · + αm vm ⟩
= α1 ⟨v, v1 ⟩ + · · · + αm ⟨v, vm ⟩
= 0,

as desired. Next suppose v ′ ∈ (span(v1 , . . . , vm ))⊥ . Since v1 , . . . , vm are all

clearly elements of span(v1 , . . . , vm ), clearly v ′ ∈ {v1 , . . . , vm }⊥ , and thus
(span(v1 , . . . , vm ))⊥ ⊆ {v1 , . . . , vm }⊥ . Therefore we conclude {v1 , . . . , vm }⊥ =
(span(v1 , . . . , vm ))⊥ .

Problem 3
Suppose U is a subspace of V with basis u1 , . . . , um and

u1 , . . . , um , w1 , . . . , wn

is a basis of V . Prove that if the Gram-Schmidt Procedure is applied

to the basis of V above, producing a list e1 , . . . , em , f1 , . . . , fn , then
e1 , . . . , em is an orthonormal basis of U and f1 , . . . , fn is an orthonormal
basis of U ⊥ .

Proof. By 6.31, span(u1 , . . . , um ) = span(e1 , . . . , em ). Since e1 , . . . , em is an

orthonormal list by construction (and linearly independent by 6.26), e1 , . . . , em
is indeed an orthonormal basis of U . Next, since each of fi is orthogonal to

23
each ej , so too is each fi orthogonal to any element of U . Thus fk ∈ U ⊥ for
k = 1, . . . , n. Now, since dim U ⊥ = dim V − dim U = n by 6.50, we conclude
f1 , . . . , fn is an orthonormal list of length dim U ⊥ and hence an orthonormal
basis of U ⊥ .

Problem 5
Suppose V is finite-dimensional and U is a subspace of V . Show that
PU ⊥ = I − PU , where I is the identity operator on V .

Proof. For v ∈ V , write v = u + w, where u ∈ U and w ∈ U ⊥ . It follows

PU ⊥ (v) = w
= (u + w) − u
= Iv − PU v,
and therefore PU ⊥ = I − PU , as desired.
Problem 7
Suppose V is finite-dimensional and P ∈ L(V ) is such that P 2 = P and
every vector in null P is orthogonal to every vector in range P . Prove
that there exists a subspace U of V such that P = PU .

Proof. By Problem 4 of Chapter 5B, we know V = null P ⊕ range P . Let v ∈ V .

Then there exist u ∈ null P and w ∈ range P such that v = u + w and hence
P v = P (u + w)
= Pu + Pw
= P w.
Let U = range P and notice that null P ⊆ null PU = U ⊥ by 6.55e. Then
P v = P w = PU (v), and so U is the desired subpace.

Problem 9
Suppose T ∈ L(V ) and U is a finite-dimensional subspace of V . Prove
that U is invariant under T if and only if PU T PU = T PU .

Proof. (⇐) Suppose PU T PU = T PU and let u ∈ U . It follows

T Pu (u) = PU T PU (v)
and thus
T u = PU T u.
Since range PU = U by 6.55d, this implies T u ∈ U . Thus U is indeed invariant
under T .
(⇒) Now suppose U is invariant under T and let v ∈ V . Since PU (v) ∈ U , it
follows that T PU (v) ∈ U . And thus PU T PU (v) = T PU (v), as desired.

24
 Problem 11
In R4 , let


U = span (1, 1, 0, 0), (1, 1, 1, 2) .

Find u ∈ U such that u − (1, 2, 3, 4) is as small as possible.

Proof. We first apply the Gram-Schmidt Procedure to v1 = (1, 1, 0, 0) and

v2 = (1, 1, 1, 2). This yields
v1
e1 =
∥v1 ∥
(1, 1, 0, 0)
=
(1, 1, 0, 0)
 
1 1
= √ , √ , 0, 0
2 2
and
v2 − ⟨v2 , e1 ⟩e1
e2 =
v2 − ⟨v2 , e1 ⟩e1
    
1 √1 √1 , √1 , 0, 0
(1, 1, 1, 2) − (1, 1, 1, 2), √
2
, 2
, 0, 0 2 2
=     
(1, 1, 1, 2) − (1, 1, 1, 2), √12 , √12 , 0, 0 √1 , √1 , 0, 0
2 2
 
(1, 1, 1, 2) − √2 √1 , √1 , 0, 0
2 2 2
=  
(1, 1, 1, 2) − √2 √1 , √1 , 0, 0
2 2 2

(0, 0, 1, 2)
=
(0, 0, 1, 2)
 
1 2
= 0, 0, √ , √ .
5 5

Now, with our orthonormal basis e1 , e2 , it follows by 6.55(i) and 6.56 that

25
u − (1, 2, 3, 4) is minimized by the vector

u = PU (1, 2, 3, 4)
= ⟨(1, 2, 3, 4), e1 ⟩e1 + ⟨(1, 2, 3, 4), e2 ⟩e2
   
3 1 1 11 1 2
=√ √ , √ , 0, 0 + √ 0, 0, √ , √
2 2 2 2 5 5
   
3 3 11 22
= , , 0, 0 + 0, 0, ,
2 2 5 5
 
3 3 11 22
= , , , ,
2 2 5 5

completing the proof.

Problem 13
Find p ∈ P5 (R) that makes
Z π
2
sin x − p(x) dx
−π

as small as possible.

Proof. Let CR [−π, π] denote the real inner product space of continuous real-
valued functions on [−π, π] with inner product
Z π
⟨f, g⟩ = f (x)g(x)dx,
−π

and let U denote the subspace of CR [−π, π] consisting of the polynomials with
real coefficients and degree at most 5. In this inner product space, observe that
sZ sZ
π
2 π
2
sin x − p(x) = sin x − p(x) dx = sin x − p(x) dx.
−π −π

qR
π 2
Notice also that −π
sin x − p(x) dx is minimized if and only if
Rπ 2
−π
sin x − p(x) dx is minimized. Thus by 6.56, we may conclude p(x) =
Rπ 2
PU (sin x) minimizes −π sin x − p(x) dx. To compute PU (sin x), we first find
an orthonormal basis of CR [−π, π] by applying the Gram-Schmidt Procedure to
the basis 1, x, x2 , x3 , x4 , x5 of U . A lengthy computation yields the orthonormal

26
basis
1
e1 = √
2π
q
3
2x
e2 = 3/2
xq
5


2 π 2 − 3x2
e3 = −
q 2π 5/2
7


2 3π 2 x − 5x3
e4 = −
2π 7/2

3 3π − 30π 2 x2 + 35x4
4
e5 = √
8 2π 9/2
q
11 4 2 3 5


2 15π x − 70π x + 63x
e6 = − .
8π 11/2
Now we compute PU (sin x) using 6.55(i), yielding



105 1485 − 153π 2 + π 4 315 1155 − 125π 2 + π 4 3
PU (sin x) = x− x
8π 6
4π 8
693 945 − 105π 2 + π 4 5
+ x ,
8π 10
which is the desired polynomial.

27