9 Inner product

9.1 Dot product

In calculus, the “dot product” of two vectors ~x = h2, −3i and ~y = h5, 1i is

~x · ~y = h2, −3i · h5, 1i = (2)(5) + (−3)(1) = 7

(multiply corresponding entries and add). In linear algebra we write these same
vectors as    
2 5
x= and y = ,
−3 1
and express the dot product as
 
T
  5  
x y= 2 −3 = 7 (or just 7)
1

y becomes xT y).
(so ~x · ~

Length
The length of a vector x is denoted kxk. A formula for this length comes from
the Pythagorean theorem. For instance, if x = [3, 4]T , then

p
kxk = 32 + 42
√
= 25
= 5.

Note that  
 3
xT x = 3 4 = 32 + 42 ,

4
which is what appears under the square root. In general we have

√
kxk = xT x

1
9 INNER PRODUCT 2

Angle
The angle θ between two vectors x and y is related to the dot product by the
formula

xT y = kxkkyk cos θ

9.1.1 Example Find the angle between x = [2, −3]T and y = [3, 2]T .

Solution We solve the equation above to get

xT y
cos θ =
kxkkyk
 
  3
2 −3
2
= p √
2 + (−3) 32 + 22
2 2

0
= = 0,
13
so θ = cos−1 0 = 90◦ .

This example shows that

x⊥y ⇐⇒ xT y = 0

(x ⊥ y is read “x is orthogonal (or perpendicular) to y”).

9.2 Definition

We have seen that in R2 the length of a vector and the angle between two
vectors can be expressed using the dot product. So in a sense the dot product
is what gives rise to the geometry of vectors. It is certain properties of the dot
product that make this work.
9 INNER PRODUCT 3

The generalization of the dot product to an arbitrary vector space is called an

“inner product.” Just like the dot product, this is a certain way of putting
two vectors together to get a number. The properties it satisfies are enough to
get a geometry that behaves much like the geometry of R2 (for instance, the
Pythagorean theorem holds).

Inner product. Let V be a vector space. An inner product

on V is a rule that assigns to each pair v, w ∈ V a real number
hv, wi such that, for all u, v, w ∈ V and α ∈ R,
(i) hv, vi ≥ 0, with equality if and only if v = 0,
(ii) hv, wi = hw, vi,
(iii) hu + v, wi = hu, wi + hv, wi,
(iv) hαv, wi = αhv, wi.

Note that, combining (iii) and (iv) with (ii), we get the properties

hu, v + wi = hu, vi + hu, wi,

hv, αwi = αhv, wi.

An inner product space is a vector space with an inner product. Each of the
vector spaces Rn , Mm×n , Pn , and FI is an inner product space:

9.3 Example: Euclidean space

We get an inner product on Rn by defining, for x, y ∈ Rn ,

hx, yi = xT y.

To verify that this is an inner product, one needs to show that all four properties
hold. We check only two of them here.

(i) We have
hx, xi = xT x = x21 + x22 + · · · + x2n ≥ 0,
with equality if and only if xi = 0 for all i, that is, x = 0.
(iii) We have

hx + y, zi = (x + y)T z = (xT + yT )z = xT z + yT z = hx, zi + hy, zi,

where we have used that matrix multiplication distributes over addition.

9 INNER PRODUCT 4

9.4 Example: Matrix space

We get an inner product on Mm×n by defining, for A, B ∈ Mm×n ,

m X
X n
hA, Bi = aij bij
i=1 j=1

(multiply corresponding entries and add). For instance,

   
2 −1 3 1 3 8
h , i = (2)(1) + (−1)(3) + (3)(8) + (5)(0) + (0)(1) + (4)(−2)
5 0 4 0 1 −2
= 15.

This inner product is identical to the dot product on Rmn if an m × n matrix

is viewed as an mn × 1 matrix by stacking its columns.

9.5 Example: Polynomial space

Let x1 , x2 , . . . , xn be fixed numbers. We get an inner product on Pn by defining,

for p, q ∈ Pn ,
n
X
hp, qi = p(xi )q(xi )
i=1
= p(x1 )q(x1 ) + p(x2 )q(x2 ) + · · · p(xn )q(xn ).

For instance, if x1 = −1, x2 = 0, and x3 = 1, then for p = x2 and q = x + 1, we

get

hp, qi = p(−1)q(−1) + p(0)q(0) + p(1)q(1)

= (1)(0) + (0)(1) + (1)(2)
= 2.

Different choices of the numbers x1 , x2 , . . . , xn produce different inner products.

9.6 Example: Function space

We get an inner product on C[a,b] (= vector space of continuous functions on

the interval [a, b]) by defining, for f, g ∈ C[a,b] ,
Z b
hf, gi = f (x)g(x) dx.
a

We check the first property of inner product:

9 INNER PRODUCT 5

(i) We have
Z b Z b
hf, f i = f (x)f (x) dx = [f (x)]2 dx.
a a

This last integral gives the (signed) area between the graph of y = [f (x)]2
and the x-axis from x = a to x = b. Since [f (x)]2 does not drop below
the x-axis, the integral is ≥ 0 with equality if and only if f (x) = 0 for all
x, that is, f is the zero function.

9.6.1 Example Find hsin x, cos xi using a = −π and b = π.

Solution We have
Z π
hsin x, cos xi = sin x cos x dx
−π
Z 0
= u du (u = sin x, du = cos x dx)
0
=0

Since the inner product generalizes the dot product, it is reasonable to say that
two vectors are “orthogonal” (or “perpendicular”) if their inner product is zero.
With this definition, we see from the preceding example that sin x and cos x are
orthogonal (on the interval [−π, π]).

9.7 Geometry

Let V be an inner product space and let v ∈ V . The norm (or length) of v is
denoted kvk and is defined by

p
kvk = hv, vi

(the square root is defined by property (i) of inner product).

9.7.1 Example With the inner product on M2×2 defined as in Section 9.4
find kAk, where  
1 2
A= .
−1 4
9 INNER PRODUCT 6

Solution We have
s   
p 1 2 1 2
kAk = hA, Ai = h , i
−1 4 −1 4
p √
= (1)(1) + (2)(2) + (−1)(−1) + (4)(4) = 22.

In the preceding example, kAk is called the “Frobenius norm” of the matrix A.
The distance between two vectors in V is the norm of their difference:

dist(v, w) = kv − wk

9.7.2 Example Of the functions x and x3 , which is closer to x2 on the

interval [0, 1] (using the inner product of Section 9.6)?

Solution We compute the two distances:

dist(x, x2 ) = kx − x2 k = hx − x2 , x − x2 i
p
s s
Z 1 Z 1
= (x − x2 )(x − x2 ) dx = (x2 − 2x3 + x4 ) dx
0 0
s 1
x3 2x4 x5 
= − +
3 4 5 0
1
=√
30
9 INNER PRODUCT 7

and

dist(x3 , x2 ) = kx3 − x2 k = hx3 − x2 , x3 − x2 i

p
s s
Z 1 Z 1
= (x3 − x2 )(x3 − x2 ) dx = (x6 − 2x5 + x4 ) dx
0 0
s 1
x7 2x6 x5 
= − +
7 6 5 0
1
= √ .
105

Therefore, x3 is closer to x2 .
The distance between functions is a measure of the space between the graphs
of the functions (although it is not the exact area). One can see that the space
between x3 and x2 (yellow) is less than the space between x and x2 (green).

In order to define the angle between two vectors, we need a theorem:

Cauchy-Schwarz theorem.
For all v, w ∈ V ,
|hv, wi| ≤ kvkkwk.

Proof. If w = 0, then both sides are zero and the inequality holds. Assume that
w 6= 0 and put
hv, wi
α= .
hw, wi
9 INNER PRODUCT 8

Using the properties of inner product we get

0 ≤ kv − αwk2 = hv − αw, v − αwi

= hv, vi − hv, αwi − hαw, vi + hαw, αwi (iii) and (iv)
2
= hv, vi − 2αhv, wi + α hw, wi (ii) and (iv)
2 2
hv, wi hv, wi
= hv, vi − 2 +
hw, wi hw, wi
2
hv, wi
= hv, vi − ,
hw, wi
so
hv, wi2 ≤ hv, vihw, wi
and taking the square root of both sides gives
|hv, wi| ≤ kvkkwk.

If v, w 6= 0, then the theorem implies that

hv, wi
−1 ≤ ≤ 1,
kvkkwk
so we can define the angle θ between v and w by

hv, wi
θ = cos−1
kvkkwk

(This gives hv, wi = kvkkwk cos θ which generalizes the dot product formula
in Section 9.1.) Since θ = 90◦ if and only if hv, wi = 0, we say that v is
orthogonal (or perpendicular) to w if and only if hv, wi = 0:

v⊥w ⇐⇒ hv, wi = 0

9.7.3 Example Let x1 = −1, x2 = 0, and x3 = 1 in the definition of the

inner product on P3 given in Section 9.5. Find the angle θ between p = x2 and
q = x + 1.

Solution In Section 9.5 we found that hp, qi = 2. We also have

p p √ √
kpk = hp, pi = p(−1)2 + p(0)2 + p(1)2 = 1 + 0 + 1 = 2,
p p √ √
kqk = hq, qi = q(−1)2 + q(0)2 + q(1)2 = 0 + 1 + 4 = 5,
9 INNER PRODUCT 9

so
hp, qi 2 2
θ = cos−1 = cos−1 √ √ = cos−1 √ ≈ 51◦
kpkkqk 2 5 10

Pythagorean theorem.
Let v, w ∈ V . If v ⊥ w, then kv + wk2 = kvk2 + kwk2 .

Proof. Assume that v ⊥ w. Using a property of inner product, we get

kv + wk2 = hv + w, v + wi
= hv, vi + hv, wi + hw, vi + hw, wi (iii)
= hv, vi + 0 + 0 + hw, wi v⊥w
= kvk2 + kwk2 .

9.7.4 Example Verify the Pythagorean theorem for sin x, cos x ∈ C[−π,π] .

Solution In Example 9.6.1 we found that hsin x, cos xi = 0 so sin x ⊥ cos x and
the Pythagorean theorem applies. We have
Z π
k sin xk2 = hsin x, sin xi = sin2 x dx.
−π

Now, using the identity sin2 x = 21 (1 − cos 2x) we get

π π
1
Z Z
sin2 x dx = (1 − cos 2x) dx
−π 2 −π
Z 2π
1
= (1 − cos u) du (u = 2x, du = 2dx)
4 −2π
1 2π
= (u − sin u) −2π
4
= π,
9 INNER PRODUCT 10

so k sin xk2 = π. Similarly, using the identity cos2 x = 12 (1 + cos 2x) we get
k cos xk2 = π. On the other hand,
Z π
k sin x + cos xk2 = hsin x + cos x, sin x + cos xi = (sin x + cos x)2 dx
−π
Z π Z π Z π
= sin2 x dx + 2 sin x cos x dx + cos2 x dx
−π −π −π
= π + 0 + π = 2π,

where we have used Example 9.6.1 to see that the second integral is 0. Therefore,

k sin xk2 + k cos xk2 = π + π = 2π = k sin x + cos xk2

so the Pythagorean theorem is verified.

9 – Exercises

9–1 Find the angle between x = [1, 2, −1, 2]T and y = [2, −1, 1, 3]T using the
inner product of Section 9.3.

9–2 Show that x ⊥ x2 in P5 using the inner product of Section 9.5 with
x1 = −2, x2 = −1, x3 = 0, x4 = 1, and x5 = 2.

9–3 Verify the Pythagorean theorem for the functions 1 and x in C[−1,1] using
the inner product of Section 9.6. (The theorem requires that 1 ⊥ x so this needs
to be checked first.)