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

h(1, 1) + (−1, −1), (1, 1)i = h(0, 0), (1, 1)i

= |0 · 1| + |0 · 1|
= 0.

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

h(1, 1) + (−1, −1), (1, 1)i = h(1, 1), (1, 1)i + h(−1, −1), (1, 1)i
= |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 = 6 {0}. Replace the positivity condition (which
states that hv, vi ≥ 0 for all v ∈ V ) in the definition of an inner product
(6.3) with the condition that hv, vi > 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 h·, ·i1 ∈ A. Since V 6= {0}, there exists v ∈ V −{0}. Then hv, vi1 > 0,
and so h·, ·i1 ∈ B. Thus A ⊆ B.
Conversely, suppose h·, ·i2 ∈ B. Then there exists some v 0 ∈ V such that

1
hv 0 , v 0 i2 > 0. Suppose by way of contradiction there exists u ∈ V is such that
hu, ui2 < 0. Define w = αu + (1 − α)v 0 for α ∈ R. It follows

hw, wi2 = hαu + (1 − α)v 0 , αu + (1 − α)v 0 i2

= hαu, αui2 + 2hαu, (1 − α)v 0 i2 + h(1 − α)v 0 , (1 − α)v 0 i2
= α2 hu, ui2 + 2α(1 − α)hu, v 0 i2 + (1 − α)2 hv 0 , v 0 i2 .

Notice the final expression is a polynomial in the indeterminate α, call it p.

Since p(0) = hv 0 , v 0 i2 > 0 and p(1) = hu, ui2 < 0, by Bolzano’s theorem there
exists α0 ∈ (0, 1) such that p(α0 ) = 0. That is, if w = α0 u + (1 − α0 )v 0 , then
hw, wi2 = 0. In particular, notice α0 = 6 0, for otherwise w = v 0 , a contradiction
0 0
since hv , v i2 > 0. Now, since hw, wi2 = 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

hu, ui2 = htv 0 , tv 0 i2

= t2 hv 0 , v 0 i2
> 0,

where the inequality follows since t ∈ (−1, 0) and hv 0 , v 0 i2 > 0. This contradicts
our assumption that hu, ui2 < 0, and so we have h·, ·i2 ∈ 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 kT vk ≤ kvk
Let V be finite-dimensional. Suppose
for every v ∈ V . Prove that T − 2I is invertible.

√
Proof. Let v ∈ null(T − 6 0.
2I), and suppose by way of contradiction that v =
Then
√ √
Tv − 2v = 0 =⇒ T v = 2v
√
=⇒ k 2vk ≤ kvk
√
=⇒ 2 · kvk ≤ kvk
√
=⇒ 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.

2
 Problem 7
Suppose u, v ∈ V . Prove that kau + bvk = kbu + avk for all a, b ∈ R if
and only if kuk = kvk.

Proof. (⇒) Suppose kau + bvk = kbu + avk for all a, b ∈ R. Then this equation
holds when a = 1 and b = 0. But then we must have kuk = kvk, as desired.
(⇐) Conversely, suppose kuk = kvk. Let a, b ∈ R be arbitrary, and notice
kau + bvk = hau + bv, au + bvi
= hau, aui + hau, bvi + hbv, aui + hbv, bvi
2 2
= a2 kuk + ab hu, vi + hv, ui + b2 kvk .


(1)
Also, we have
kbu + avk = hbu + av, bu + avi
= hbu, bui + hbu, avi + hav, bui + hav, avi
2 2
= b2 kuk + ab hu, vi + hv, ui + a2 kvk .


(2)
Since kuk = kvk, (1) equals (2), and hence kau + bvk = kbu + avk. Since a, b
were arbitrary, the result follows.

Problem 9
Suppose u, v ∈ V and kuk ≤ 1 and kvk ≤ 1. Prove that
q q
2 2
1 − kuk 1 − kvk ≤ 1 − |hu, vi|.

Proof. By the Cauchy-Schwarz Inequality, we have |hu, vi| ≤ kukkvk. Since

kuk ≤ 1 and kvk ≤ 1, this implies
0 ≤ 1 − kukkvk ≤ 1 − |hu, vi|,
and hence it’s enough to show
q q
2 2
1 − kuk 1 − kvk ≤ 1 − kukkvk.
Squaring both sides, it suffices to prove
  
2
2 2
1 − kuk 1 − kvk ≤ 1 − kukkvk . (3)

Notice

2  2

2

2 2
1 − kukkvk − 1 − kuk 1 − kvk = kuk − 2kukkvk + kvk

2
= kuk − kvk
≥ 0,
and hence inequality (3) holds, completing the proof.

3
 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.

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

hu, vi = kukkvk 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 kuk, B has length kvk and C has length ku − vk. 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
2 2 2
ku − vk = kuk + kvk − 2kukkvk cos θ.
It follows
2 2 2
2kukkvk cos θ = kuk + kvk − ku − vk
= hu, ui + hv, vi − hu − v, u − vi


= hu, ui + hv, vi − hu, ui − 2hu, vi + hv, vi
= 2hu, vi.
Dividing both sides by 2 gives the desired result.

4
 Problem 15
Prove that  2   
n n n 2
X X X b j 
 a j bj  ≤  jaj 2  
j=1 j=1 j=1
j

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

Proof. Let
√ √
 
  1 1
u = a1 , 2a2 , . . . , nan and v = b1 , √ b2 , . . . , √ bn .
2 n
Pn
Since hu, vi = k=1 ak bk , the Cauchy-Schwarz Inequality yields
2 2 2
(a1 b1 + · · · + an bn ) ≤ kuk kvk
!

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
k(x, y)k = max{|x|, |y|}
for all (x, y) ∈ R2 .

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

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

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
2 2
ku + vk − ku − vk
hu, vi =
4
for all u, v ∈ V .

5
Proof. Suppose V is a real inner product space and let u, v ∈ V . It follows
   
2 2 2 2
2
ku + vk − ku − vk
2 kuk + 2hu, vi + kvk − kuk − 2hu, vi + kvk
=
4 4
4hu, vi
=
4
= hu, vi,

as desired.

Problem 20
Suppose V is a complex inner product space. Prove that
2 2 2 2
ku + vk − ku − vk + ku + ivk i − ku − ivk i
hu, vi =
4
for all u, v ∈ V .

Proof. Notice we have

2
ku + vk = hu + v, u + vi
2 2
= kuk + hu, vi + hv, ui + kvk

and
2
−ku − vk = −hu − v, u − vi
2 2
= −kuk + hu, vi + hv, ui − kvk .

Also, we have
2

ku + ivk i = i hu + iv, u + ivi
 
2
= i kuk + hu, ivi + hiv, ui + hiv, ivi
 
2 2
= i kuk − ihu, vi + ihv, ui + kvk
2 2
= ikuk + hu, vi − hv, ui + ikvk

and
2

−ku − ivk i = −i hu − iv, u − ivi
 
2
= −i kuk − hu, ivi − hiv, ui + hiv, ivi
 
2 2
= −i kuk + ihu, vi − ihv, ui + kvk
2 2
= −ikuk + hu, vi − hv, ui − ikvk .

6
Thus it follows
2 2 2 2
ku + vk − ku − vk + ku + ivk i − ku − ivk i = 4hu, vi.
Dividing both sides by 4 yields the desired result.

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

h(u1 , . . . , um ), (v1 , . . . , vm )i = hu1 , v1 i + · · · + hum , vm i

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

Proof. We prove that this definition satisfies each property of an inner product
in turn.
Positivity: Let (v1 , . . . , vm ) ∈ V1 × . . . Vm . Since hvk , vk i is an inner product
on Vk for k = 1, . . . , m, we have hvk , vk i ≥ 0. Thus
h(v1 , . . . , vm ), (v1 , . . . , vm )i = hv1 , v1 i + · · · + hvm , vm i ≥ 0.
Definiteness: First suppose h(v1 , . . . , vm ), (v1 , . . . , vm )i = 0 for (v1 , . . . , vm ) ∈
V1 × · · · × Vm . Then
hv1 , v1 i + · · · + hvm , vm i = 0.
By positivity of each inner product on Vk (for k = 1, . . . , m), we must have
hvk , vk i ≥ 0. Thus the equation above holds only if hvk , vk i = 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
h(v1 , . . . , vm ), (v1 , . . . , vm )i = hv1 , v1 i + · · · + hvm , vm i
= h0, 0i + · · · + h0, 0i
= 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
h(u1 , . . . , um )+(v1 , . . . , vm )), (w1 , . . . , wm )i
= h(u1 + v1 , . . . , um + vm ), (w1 , . . . , wm )i
= hu1 + v1 , w1 i + · · · + hum + vm , wm i
= hu1 , w1 i + hv1 , w1 i + · · · + hum , wm i + hvm , wm i
= h(u1 , . . . , um ), (w1 , . . . , wm )i + h(v1 , . . . , vm ), (w1 , . . . , wm )i,

7
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

hλ(u1 , . . . , um ), (v1 , . . . , vm )i = h(λu1 , . . . , λum ), (v1 , . . . , vm )i

= hλu1 , v1 i + · · · + hλum , vm i
= λhu1 , v1 i + · · · + λhum , vm i
= λ(hu1 , v1 i + · · · + hum , vm i)
= λh(u1 , . . . , um ), (v1 , . . . , vm )i,

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

h(u1 , . . . , um ), (v1 , . . . , vm )i = hu1 , v1 i + · · · + hum , vm i

= hv1 , u1 i + · · · + hvm , um i
= hu1 , v1 i + · · · + hum , vm i
= h(v1 , . . . , vm ), (u1 , . . . , um )i,

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 h·, ·i1 by

hu, vi1 = hSu, Svi

for u, v ∈ V . Show that h·, ·i1 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 hv, vi1 = hSv, Svi ≥ 0.
Definiteness: Suppose hv, vi = 0 for some v ∈ V . This is true iff hSv, Svi = 0
(by definition) which is true iff Sv = 0 (by definiteness of h·, ·i), which is true iff

8
v = 0 (since S is injective).
Additivity in first slot: Let u, v, w ∈ V . Then

hu + v, wi1 = hS(u + v), Swi

= hSu + Sv, Swi
= hSu, Swi + hSv, Swi
= hu, wi1 + hv, wi1 .

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

hλu, vi1 = hS(λu), Svi

= hλSu, Svi
= λhSu, Svi
= λhu, vi1 .

Conjugate symmetry Let u, v ∈ V . Then

hu, vi1 = hSu, Svi

= hSv, Sui
= hv, ui1 .

Problem 25
Suppose S ∈ L(V ) is not injective. Define h·, ·i1 as in the exercise above.
Explain why h·, ·i1 is not an inner product on V .

Proof. If S is not injective, then h·, ·i1 fails the definiteness requirement in the
definition of an inner product. In particular, there exists v 6= 0 such that Sv = 0.
Hence hv, vi1 = hSv, Svi = 0 for a nonzero v.

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

9
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
kw − uk +kw − vk −u + v
= −
2 2
2 2 2
kw − uk +kw − vk ku − vk
= − ,
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 k·k1 and k·k2 are norms on vector space V . We say k·k1
and k·k2 are equivalent if there exist 0 < C1 ≤ C2 such that
C1 kvk1 ≤ kvk2 ≤ C2 kvk1
for all v ∈ V .
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 `1 -style norm k·k1 defined by
kvk1 = |α1 | + · · · +|αn |
for all v = α1 e1 + · · · + αn en ∈ V .
Let k·k be a norm on V . We wish to show C1 kvk1 ≤ kvk ≤ C2 kvk1 for all
v ∈ V and some choice of C1 , C2 . Since this is trivially true for v = 0, we need
only consider v 6= 0, in which case we have
C1 ≤ kuk ≤ C2 , (*)
where u = v/kvk1 . Thus it suffices to consider only vectors v ∈ V such that
kvk1 = 1.
We will now show that k·k is continuous under k·k1 and apply the Extreme
Value Theorem to deduce the desired result. So let  > 0 and define M =
max{ke1 k , . . . ,ken k} and

δ= .
M
It follows that if u, v ∈ V are such that ku − vk1 < δ, then
kuk −kvk ≤ ku − vk
≤ M ku − vk1
≤ Mδ
= ,

10
and k·k is indeed continuous under the topology induced by k·k1 . Let S = {u ∈
V | kuk1 = 1} (the unit sphere with respect to k·k1 ). Since S is compact and k·k
is continuous on it, by the Extreme Value Theorem we may define

C1 = minkuk and C2 = maxkuk .

u∈S u∈S

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

Problem 29
For u, v ∈ V , define d(u, v) = ku − vk.

(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).

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 ⇐⇒ hu − v, u − vi = 0
⇐⇒ hu − v, u − v, =i0
⇐⇒ u − v = 0
⇐⇒ u = v.

Symmetry: Let u, v ∈ V . We have

d(u, v) = ku − vk
= (−1)(u − v)
= kv − uk
= d(v, u).

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

d(u, v) + d(v, w) = ku − vk +kv − wk

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

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

{vk }∞
k=1 is Cauchy. Then for  > 0, there exists N ∈ Z
+
such that

11
 kvm − vn k <  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, k·k is equivalent to k·k1 (where k·k1 is defined in

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

ckvm − vn k1 ≤ kvm − vn k < ,

and hence  
Xp
c αm,i − αn,i  < .
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 ke1 k + · · · + αj,p − αp ep .

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 k·k.
(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
kwk = a, kvk = c, and kw − vk = b. The identity follows by applying Problem 27
with u = 0.

12
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

h(cos θ, sin θ), (− sin θ, cos θ)i = − sin θ cos θ + sin θ cos θ = 0
and
h(cos θ, sin θ), (sin θ, − cos θ)i = 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 ke1 k = ke2 k = 1, there
exist θ, ϕ ∈ [0, 2π) such that
e1 = (cos θ, sin θ) and e2 = (cos ϕ, sin ϕ).
Next, since he1 , e2 i = 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
3π
and thus ϕ = θ + − nπ, for n ∈ Z. Since θ, ϕ ∈ [0, 2π), this implies
2
ϕ = θ ± π2 . If ϕ = θ + π2 , then
   !
π π
e2 = cos θ + , sin θ +
2 2
= (− sin θ, cos θ),
π
and if ϕ = θ − 2, 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).

13
 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 kv1 k = 1, e1 = v1 .
Next, we have

v2 − hv2 , e1 ie1
e2 =
v2 − hv2 , e1 ie1
(1, 1, 1) − h(1, 1, 1), (1, 0, 0)i(1, 0, 0)
=
(1, 1, 1) − h(1, 1, 1), (1, 0, 0)i(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
and
v3 − hv3 , e1 ie1 − hv3 , e2 ie2
e3 =
v3 − hv3 , e1 ie1 − hv3 , e2 ie2
    
(1, 1, 2) − h(1, 1, 2), (1, 0, 0)i(1, 0, 0) − (1, 1, 2), 0, √12 , √12 0, √12 , √12
=     
(1, 1, 2) − h(1, 1, 2), (1, 0, 0)i(1, 0, 0) − (1, 1, 2), 0, √1 , √1 0, √12 , √12
2 2
 
(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.

14
 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
hf, gi = 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.

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,

15
so indeed all vectors have norm 1. Now we show them to be pairwise orthogonal.
Suppose j, k ∈ Z are such that j 6= 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,

  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,

16
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
hp, qi = 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).

Proof. First notice k1k = 1, hence e1 = 1. Next notice

v2 − hv1 , e1 ie1 = x − hx, 1i

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

17
and therefore we have
√
 
1
e2 = 2 3 x − .
2
To compute e3 , first notice
"  #
1 √ Z 1
Z 
2 2 2 1
v3 − hv3 , e1 ie1 − hv3 , e2 ie2 = x − x dx − 2 3 x x− dx e2
0 0 2
 ! " #
√ Z 1 2 √

1 x 1
= x2 − − 2 3 x3 − dx 2 3 x −
3 0 2 2
  
1 1 1 1
= x2 − − 12 − x−
3 4 6 2
 
1 1
= x2 − − x −
3 2
1
= x2 − x +
6
and
s 
2 1 1 1
x −x+ = x2 − x + , x2 − 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.

18
 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 hp, qi = 0
p(x)q(x) dx on P2 (R). Define
1


ϕ ∈ L(P2 (R)) by ϕ(p) = p 2 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) = hp, qi 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
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 − hvj , e1 ie1 − · · · − hvj , ej−1 iej−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.

19
 Problem 11
Suppose h·, ·i1 and h·, ·i2 are inner products on V such that hv, wi1 = 0
if and only if hv, wi2 = 0. Prove that there is a positive number c such
that hv, wi1 = chv, wi2 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 hv, wi1 = chv, wi2 . 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 hv, vi1 , hw, wi1 ,
hv, vi2 , and hw, wi2 are all nonzero as well. It now follows

hv, wi1
0 = hv, wi1 − hv, vi1
hv, vi1
* ! +
hv, wi1
= hv, wi1 − v, v
hv, vi1
1
* ! +
hv, wi1
= v, w − v
hv, vi1
1
* ! +
hv, wi1
= v, w − v
hv, vi1
2
* ! +
hv, wi1
= hv, wi2 − v, v
hv, vi1
2
hv, wi1
= hv, wi2 − hv, vi2
hv, vi1
hv, vi2
= hv, wi2 − hv, wi1 ,
hv, vi1

where the fifth equality follows by our hypothesis. Thus

2
kvk1
hv, wi1 = 2 hv, wi2 . (4)
kvk2

20
By a similar computation, notice

hv, wi1
0 = hv, wi1 − hw, wi1
hw, wi1
* +
hv, wi1
= hv, wi1 − w, w
hw, wi1
1
* +
hv, wi1
= v− w, w
hw, wi1
* +1
hv, wi1
= v− w, w
hw, wi1
2
* +
hv, wi1
= hv, wi2 − w, w
hw, wi2
2
hv, wi1
= hv, wi2 − hw, wi2
hw, wi2
hw, wi2
= hv, wi2 − hv, wi1 ,
hw, wi1

and thus
2
kwk1
hv, wi1 = 2 hv, wi2 (5)
kwk2
as well. By combining Equations (4) and (5), we conclude

hv, vi1 hw, wi1

= .
hv, vi2 hw, wi2
2 2
Since v and w were arbitrary nonzero vectors in V , choosing c = kuk1 /kuk2
for any u 6= 0 guarantees hv, wi1 = chv, wi2 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 hw, vj i > 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) =

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

21
 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
hf, gi = 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 ) = hf, gi
for every f ∈ CR ([−1, 1]).

Proof. Suppose not. Then there exists g ∈ CR ([−1, 1]) such that
ϕ(f ) = hf, gi
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) 6= 0 for all x ∈ [a, b]. By the Extreme Value
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) = hv, ui

for v ∈ V .
(a) Show that if F = R, then Φ is a linear map from V to V 0 .
(b) Show that if F = C and V 6= {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 0 .

22
Proof. (a) Suppose F = R. Let u, u0 ∈ V and α ∈ R. Then, for all v ∈ V , we
have

Φu+u0 (v) = hv, u + u0 i = hv, ui + hv, u0 i = Φu (v) + Φu0 (v)

and
Φαu (v) = hv, αui = αhv, ui = αhv, ui = αΦu (v).
Thus Φ is indeed a linear map.

(b) Suppose F = C and V 6= {0}. Let u ∈ V . Then, given v ∈ V , we have

Φiu (v) = hv, iui = ihv, ui,

whereas
iΦu (v) = ihv, ui.
Thus Φiu 6= iΦu , and indeed Φ is not a linear map, since is is not homoge-
neous.

(c) Suppose u, u0 ∈ V are such that Φu = Φu0 . Then, for all v ∈ V , we have

Φu (v) = Φu0 (v)

=⇒ hv, ui = hv, u0 i
=⇒ hv, ui − hv, u0 i = 0
=⇒ hv, u − u0 i = 0.

In particular, choosing v = u − u0 , the above implies hu − u0 , u − u0 i = 0,

which is true iff u − u0 = 0. Thus we conclude u = u0 , so that Φ is indeed
injective.

(d) Suppose F = R and dim V = n. Notice that since Φ : V ,→ V 0 , we have

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

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

= V 0 , as was to be shown.

C: Orthogonal Complements and Minimization

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

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

23
Proof. Suppose v ∈ {v1 , . . . , vm }⊥ . Then hv, vk i = 0 for k = 1, . . . , m. Let
u ∈ span(v1 , . . . , vm ) be arbitrary. We want to show hv, ui = 0, since this implies
v ∈ (span(v1 , . . . , vm ))⊥ and hence {v1 , . . . , vm }⊥ ⊆ (span(v1 , . . . , vm ))⊥ . To
see this, notice

hv, ui = hv, α1 v1 + · · · + αm vm i
= α1 hv, v1 i + · · · + αm hv, vm i
= 0,

as desired. Next suppose v 0 ∈ (span(v1 , . . . , vm ))⊥ . Since v1 , . . . , vm are all

clearly elements of span(v1 , . . . , vm ), clearly v 0 ∈ {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
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.

24
 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.

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.

25
Proof. We first apply the Gram-Schmidt Procedure to v1 = (1, 1, 0, 0) and
v2 = (1, 1, 1, 2). This yields
v1
e1 =
kv1 k
(1, 1, 0, 0)
=
(1, 1, 0, 0)
 
1 1
= √ , √ , 0, 0
2 2
and
v2 − hv2 , e1 ie1
e2 =
v2 − hv2 , e1 ie1
    
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
u − (1, 2, 3, 4) is minimized by the vector

u = PU (1, 2, 3, 4)
= h(1, 2, 3, 4), e1 ie1 + h(1, 2, 3, 4), e2 ie2
   
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.

26
 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
hf, gi = 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
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

27
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
2 4
693 945 − 105π + π
+ x5 ,
8π 10
which is the desired polynomial.

28