MATH 304

Linear Algebra
Lecture 17:
Euclidean structure in Rn (continued).
Orthogonal complement.
Orthogonal projection.
Vectors: geometric approach

A B′

A′
• A vector is represented by a directed segment.
• Directed segment is drawn as an arrow.
• Different arrows represent the same vector if
they are of the same length and direction.
−→ −−→
Notation: AB (= A′ B ′ ).
Linear structure: vector addition
Given vectors a and b, their sum a + b is defined
−→ −→ −→
by the rule AB + BC = AC .
−→
That is, choose points A, B, C so that AB = a and
−→ −→
BC = b. Then a + b = AC .

B b
a C
a+b B′ b
A a C′
a+b
A′
Linear structure: scalar multiplication

Let v be a vector and r ∈ R. By definition, r v is a

vector whose magnitude is |r | times the magnitude
of v. The direction of r v coincides with that of v if
r > 0. If r < 0 then the directions of r v and v are
opposite.
v

3v

−2v
Beyond linearity: Euclidean structure

Euclidean structure includes:

• length of a vector: |x|,
• angle between vectors: θ,
• dot product: x · y = |x| |y| cos θ.

C
y

θ x
A B
Vectors: algebraic approach
An n-dimensional coordinate vector is an element of
Rn , i.e., an ordered n-tuple (x1 , x2 , . . . , xn ) of real
numbers.
Let a = (a1 , a2 , . . . , an ) and b = (b1 , b2 , . . . , bn ) be
vectors, and r ∈ R be a scalar. Then, by definition,
a + b = (a1 + b1 , a2 + b2 , . . . , an + bn ),
r a = (ra1 , ra2 , . . . , ran ),
0 = (0, 0, . . . , 0),
−b = (−b1 , −b2 , . . . , −bn ),
a − b = a + (−b) = (a1 − b1 , a2 − b2 , . . . , an − bn ).
Cartesian coordinates: geometric meets algebraic

(−3, 2) (−3, 2)

(2, 1) (2, 1)

Once we specify an origin O, each point A is

−→
associated a position vector OA. Conversely, every
vector has a unique representative with tail at O.
Cartesian coordinates allow us to identify a line, a
plane, and space with R, R2 , and R3 , respectively.
Length and distance

Definition. The length of a vector

v = (v1 , v2 , . . . , vn ) ∈ Rn is
p
kvk = v12 + v22 + · · · + vn2 .

The distance between vectors/points x and y is

ky − xk.

Properties of length:
kxk ≥ 0, kxk = 0 only if x = 0 (positivity)
kr xk = |r | kxk (homogeneity)
kx + yk ≤ kxk + kyk (triangle inequality)
Scalar product
Definition. The scalar product of vectors
x = (x1 , x2 , . . . , xn ) and y = (y1 , y2 , . . . , yn ) is
x · y = x1 y1 + x2 y2 + · · · + xn yn .

Properties of scalar product:

x · x ≥ 0, x · x = 0 only if x = 0 (positivity)
x·y =y·x (symmetry)
(x + y) · z = x · z + y · z (distributive law)
(r x) · y = r (x · y) (homogeneity)
In particular, x · y is a bilinear function (i.e., it is
both a linear function of x and a linear function of y).
Relations between lengths and scalar products:
√
kxk = x · x
|x · y| ≤ kxk kyk (Cauchy-Schwarz inequality)
kx − yk2 = kxk2 + kyk2 − 2 x·y

By the Cauchy-Schwarz inequality, for any nonzero

vectors x, y ∈ Rn we have
x·y
cos θ = for some 0 ≤ θ ≤ π.
kxk kyk
θ is called the angle between the vectors x and y.
The vectors x and y are said to be orthogonal
(denoted x ⊥ y) if x · y = 0 (i.e., if θ = 90o ).
Problem. Find the angle θ between vectors
x = (2, −1) and y = (3, 1).
√ √
x · y = 5, kxk = 5, kyk = 10.
x·y 5 1
cos θ = =√ √ =√ =⇒ θ = 45o
kxk kyk 5 10 2

Problem. Find the angle φ between vectors

v = (−2, 1, 3) and w = (4, 5, 1).
v · w = 0 =⇒ v ⊥ w =⇒ φ = 90o
Orthogonality

Definition 1. Vectors x, y ∈ Rn are said to be

orthogonal (denoted x ⊥ y) if x · y = 0.

Definition 2. A vector x ∈ Rn is said to be

orthogonal to a nonempty set Y ⊂ Rn (denoted
x ⊥ Y ) if x · y = 0 for any y ∈ Y .

Definition 3. Nonempty sets X , Y ⊂ Rn are said

to be orthogonal (denoted X ⊥ Y ) if x · y = 0
for any x ∈ X and y ∈ Y .
Examples in R3 . • The line x = y = 0 is
orthogonal to the line y = z = 0.
Indeed, if v = (0, 0, z) and w = (x, 0, 0) then v · w = 0.

• The line x = y = 0 is orthogonal to the plane

z = 0.
Indeed, if v = (0, 0, z) and w = (x, y , 0) then v · w = 0.

• The line x = y = 0 is not orthogonal to the

plane z = 1.
The vector v = (0, 0, 1) belongs to both the line and the
plane, and v · v = 1 6= 0.

• The plane z = 0 is not orthogonal to the plane

y = 0.
The vector v = (1, 0, 0) belongs to both planes and
v · v = 1 6= 0.
Proposition 1 If X , Y ∈ Rn are orthogonal sets
then either they are disjoint or X ∩ Y = {0}.
Proof: v ∈ X ∩ Y =⇒ v ⊥ v =⇒ v · v = 0 =⇒ v = 0.

Proposition 2 Let V be a subspace of Rn and S

be a spanning set for V . Then for any x ∈ Rn
x ⊥ S =⇒ x ⊥ V .
Proof: Any v ∈ V is represented as v = a1 v1 + · · · + ak vk ,
where vi ∈ S and ai ∈ R. If x ⊥ S then
x · v = a1 (x · v1 ) + · · · + ak (x · vk ) = 0 =⇒ x ⊥ v.

Example. The vector v = (1, 1, 1) is orthogonal to

the plane spanned by vectors w1 = (2, −3, 1) and
w2 = (0, 1, −1) (because v · w1 = v · w2 = 0).
Orthogonal complement
Definition. Let S ⊂ Rn . The orthogonal
complement of S, denoted S ⊥ , is the set of all
vectors x ∈ Rn that are orthogonal to S. That is,
S ⊥ is the largest subset of Rn orthogonal to S.
Theorem 1 S ⊥ is a subspace of Rn .
Note that S ⊂ (S ⊥ )⊥ , hence Span(S) ⊂ (S ⊥ )⊥ .
Theorem 2 (S ⊥ )⊥ = Span(S). In particular, for
any subspace V we have (V ⊥ )⊥ = V .
Example. Consider a line L = {(x, 0, 0) | x ∈ R}
and a plane Π = {(0, y , z) | y , z ∈ R} in R3 .
Then L⊥ = Π and Π⊥ = L.
Fundamental subspaces
Definition. Given an m×n matrix A, let
N(A) = {x ∈ Rn | Ax = 0},
R(A) = {b ∈ Rm | b = Ax for some x ∈ Rn }.

R(A) is the range of a linear mapping L : Rn → Rm ,

L(x) = Ax. N(A) is the kernel of L.
Also, N(A) is the nullspace of the matrix A while
R(A) is the column space of A. The row space of
A is R(AT ).
The subspaces N(A), R(AT ) ⊂ Rn and
R(A), N(AT ) ⊂ Rm are fundamental subspaces
associated to the matrix A.
Theorem N(A) = R(AT )⊥ , N(AT ) = R(A)⊥ .
That is, the nullspace of a matrix is the orthogonal
complement of its row space.
Proof: The equality Ax = 0 means that the vector x is
orthogonal to rows of the matrix A. Therefore N(A) = S ⊥ ,
where S is the set of rows of A. It remains to note that
S ⊥ = Span(S)⊥ = R(AT )⊥ .

Corollary Let V be a subspace of Rn . Then

dim V + dim V ⊥ = n.
Proof: Pick a basis v1 , . . . , vk for V . Let A be the k×n
matrix whose rows are vectors v1 , . . . , vk . Then V = R(AT )
and V ⊥ = N(A). Consequently, dim V and dim V ⊥ are rank
and nullity of A. Therefore dim V + dim V ⊥ equals the
number of columns of A, which is n.
Orthogonal projection

Theorem 1 Let V be a subspace of Rn . Then

any vector x ∈ Rn is uniquely represented as
x = p + o, where p ∈ V and o ∈ V ⊥ .

In the above expansion, p is called the orthogonal

projection of the vector x onto the subspace V .

Theorem 2 kx − vk > kx − pk for any v 6= p in V .

Thus kok = kx − pk = min kx − vk is the

v∈V
distance from the vector x to the subspace V .
Orthogonal projection onto a vector
Let x, y ∈ Rn , with y 6= 0.
Then there exists a unique decomposition x = p + o
such that p is parallel to y and o is orthogonal to y.

o x

p
y

p = orthogonal projection of x onto y

Orthogonal projection onto a vector
Let x, y ∈ Rn , with y 6= 0.
Then there exists a unique decomposition x = p + o
such that p is parallel to y and o is orthogonal to y.

We have p = αy for some α ∈ R. Then

0 = o · y = (x − αy) · y = x · y − αy · y.

x·y x·y
=⇒ α = =⇒ p= y
y·y y·y
Problem. Find the distance from the point
x = (3, 1) to the line spanned by y = (2, −1).
Consider the decomposition x = p + o, where p is parallel to
y while o ⊥ y. The required distance is the length of the
orthogonal component o.
x·y 5
p= y = (2, −1) = (2, −1),
y·y 5
√
o = x − p = (3, 1) − (2, −1) = (1, 2), kok = 5.

Problem. Find the point on the line y = −x that

is closest to the point (3, 4).
The required point is the projection p of v = (3, 4) on the
vector w = (1, −1) spanning the line y = −x.
 
v·w −1 1 1
p= w= (1, −1) = − ,
w·w 2 2 2
Problem. Let Π be the plane spanned by vectors
v1 = (1, 1, 0) and v2 = (0, 1, 1).
(i) Find the orthogonal projection of the vector
x = (4, 0, −1) onto the plane Π.
(ii) Find the distance from x to Π.
We have x = p + o, where p ∈ Π and o ⊥ Π.
Then the orthogonal projection of x onto Π is p and
the distance from x to Π is kok.
We have p = αv1 + βv2 for some α, β ∈ R.
Then o = x − p = x − αv1 − βv2 .
 
o · v1 = 0 α(v1 · v1 ) + β(v2 · v1 ) = x · v1
⇐⇒
o · v2 = 0 α(v1 · v2 ) + β(v2 · v2 ) = x · v2
x = (4, 0, −1), v1 = (1, 1, 0), v2 = (0, 1, 1)

α(v1 · v1 ) + β(v2 · v1 ) = x · v1
α(v1 · v2 ) + β(v2 · v2 ) = x · v2
 
2α + β = 4 α=3
⇐⇒ ⇐⇒
α + 2β = −1 β = −2

p = 3v1 − 2v2 = (3, 1, −2)

o = x − p = (1, −1, 1)
√
kok = 3