≻≻≻≻—≻≻≻≻—Lecture Nine —≺≺≺≺—≺≺≺≺

Chapter Four: The Euclidean Space

Review. A n × 1 matrix of R is called an n-vector of R.

 
u1
u 
 2
Given an n-vector u =  . , We call ui the i-th component of u.
.
un
The set of all n-vectors of R is called an n-dimensional Euclidean space, denoted by Rn.

Linear Algebra lecture 9−1

    
u1 v1
u  v 
 2  2
Review. Let u =  .  and v =  .  be vectors in Rn and c be a scalar (c ∈ R).
. .
 
un vn u1 + v 1
u + v 
 2 2
Then the sum of u and v is again an n-vector defined by u + v =  .  and
 . 
  un + v n
cu1
 cu 
 2
the scalar multiple of u by c is an n-vector defined by cu =  . . Note
 . 
cun
(α) Rn is closed under the operation of vector addition and
(β) Rn is closed under the operation of scalar multiplication.

Linear Algebra lecture 9−2

Furthermore,
(a) for all u, v ∈ Rn, we have u + v = v + u.
(b) for all u, v, w ∈ Rn, we have (u + v) + w = u + (v + w).
(c) 0 is the (unique) vector in Rn satisfying u + 0 = u for all u ∈ Rn.
(d) For each u ∈ Rn, −u ≡ (−1)u is the (unique) vector in Rn satisfying u + (−u) = 0.
(e) for all u, v ∈ Rn and c ∈ R, we have c(u + v) = cu + cv.
(f) for all u ∈ Rn and c, d ∈ R, we have (c + d)u = cu + du.
(g) for all u ∈ Rn and c, d ∈ R, we have (cd)u = c(du).
(h) for all u ∈ Rn, we have 1u = u.

Linear Algebra lecture 9−3

Definition. An inner product on Rn is a function h·, ·i from Rn × Rn to R satisfying
(a) (Positive-definiteness) hu, ui ≥ 0 for all u ∈ Rn and hu, ui = 0 if and only if u = 0.
(b) (symmetry) hu, vi = hv, ui for all u, v ∈ Rn;
(c) (linearity) hcu + w, vi = chu, vi + hw, vi for all u, v, w ∈ Rn and c ∈ R;

Example. The dot product (Euclidean inner product) of the n-vectors u and v defined by
hu, vi = u · v = u1 × v1 + u2 × v2 + . . . + un × vn is an inner product.
Proof. Skip.

Theorem 4.2.4. (Cauchy-Schwarz inequality) hu, vi2 ≤ hu, uihv, vi for all u, v ∈
Rn .

Proof. Skip.
Linear Algebra lecture 9−4
Definition. A function || · || from Rn to R is called a norm (length) provided that
(a) ||u|| ≥ 0 for all u ∈ Rn and ||u|| = 0 if and only if u = 0.
(b) ||cu|| = |c| ||u|| for all u ∈ Rn and c ∈ R.
(c) (triangular inequality) ||u + v|| ≤ ||u|| + ||v|| for all u, v ∈ Rn.

p
Example. The Euclidean norm (L2 norm) defined by ||u|| = ||u||2 = u21 + · · · + u2n is a
norm in Rn.
Proof. Skip.

Example. The taxicab norm or Manhattan norm (L1 norm) defined by ||u|| = ||u||1 =
|u1| + |u2| + · · · + |un| is a norm in Rn.
Proof. Skip.

Linear Algebra lecture 9−5

 p
Fact. Given an inner product h·, ·i, we can define a norm by ||u|| = hu, ui. This
norm || · || is said to be induced by the inner product h·, ·i.
◦
p
Proof. 1 ”well-defined” Since hu, ui ≥ 0 for all u, ||u|| = hu, ui is well-defined.
2◦ ”||u|| ≥ 0 and ||u|| = 0 if and only if u = 0”
First it is obvious that ||u|| ≥ 0.
Furthermore, since h·, ·i is an inner product,
p
it follows immediately ||u|| = hu, ui = 0 if and only if u = 0.
3◦ ”||cu|| = |c| ||u||”
p
By definition, we have ||cu|| = hcu, cui.
By symmetry and linearity of the inner product, we have
p p p
2
hcu, cui = c hu, ui = |c| hu, ui = |c|||u||.
4◦ ”||u + v|| ≤ ||u|| + ||v||”
By symmetry and linearity of the inner product, we have
||u + v||2 = hu + v, u + vi = hu, ui + 2hu, vi + hv, vi.
It follows the Cauchy-Schwarz inequality that
p
hu, ui + 2hu, vi + hv, vi ≤ hu, ui + 2 hu, uihv, vi + hv, vi.
p p
We obtain ||u + v|| ≤ ( hu, ui + hv, vi)2 = (||u|| + ||v||)2.
2

Linear Algebra lecture 9−6

 p
Example. The Euclidean norm ||u||2 = u21 + u22 + · · · + u2n is induced from the dot
product u · v = u1v1 + u2v2 + . . . + unvn.

Fact. (Parallelogram law) If || · || is a norm induced from an inner product h·, ·i, then
one has ||u + v||2 + ||u − v||2 = 2(||u||2 + ||v||2).

Proof. Skip.

Linear Algebra lecture 9−7

Definition. A function d : Rn × Rn −→ R is called a distance (metric) if d satisfies
(a) d(u, v) ≥ 0 for all u ∈ Rn and d(u, v) = 0 if and only if u = v.
(b) d(u, v) = d(v, u) for all u, v ∈ Rn
(c) (triangular inequality) d(u, v) ≤ d(u, w) + d(w, v) for all u, v, w ∈ Rn
Also, we call (Rn, d) a metric space.

Example. The Euclidean distance (L2 distance, L2 metric) is defined as d(u, v)

p
= (u1 − v1)2 + (u2 − v2)2 + · · · + (un − vn)2.

Example. The taxicab metric (Manhattan distance, L1 distance) is d(u, v) = |u1 − v1| +
|u2 − v2| + · · · + |un − vn|.

Linear Algebra lecture 9−8

Fact. Given a norm || · ||, we can define a distance by d(u, v) = ||u − v||. The metric
d is said to be induced by the norm || · ||.
Proof. Skip.

Example. The Euclidean distance is induced by the Euclidean norm and the taxicab metric
is induced by the taxicab norm.

Fact. If a distance d(·, ·) is induced by a norm || · ||, then we have

(1) (translation invariance) d(u, v) = d(u + w, v + w);
(2) (homogeneity) d(cu, cv) = |c|d(u, v).

Proof. Skip.

Linear Algebra lecture 9−9

Definition. Two nonzero vectors u and v in Rn are said to be parallel if u = tv for some
t ∈ R and they are parallel in the same direction if t > 0 and in the opposite direction if
t < 0.

Definition. A vector u in Rn is said to be a unit vector provided that ||u|| = 1.

u
Note. Given u 6= 0 ∈ Rn, ||u||
is the only unit vector parallel to u in the same direction.

Definition. Two vectors u and v in Rn are said to be orthogonal if u · v = u1v1 + · · · +

unvn = 0 (hu, vi = 0).

Linear Algebra lecture 9−10

Definition. A function of Rn into Rm is a map assigning a unique vector (point) f (v) in
Rm for each vector (point) v in Rn. The vector f (v) is called the image of v. The set of all
images of the vectors in Rn is simply called the image (range) of f , Rf = {f (v) : v ∈ Rn}.
Given V ⊆ Rn, the image of V under f is defined as f (V ) = {f (v) : v ∈ V }. Given
W ⊆ Rm, the inverse image of W under f is f −1(W ) = {v : f (v) ∈ W }.

Definition. A linear transformation L of Rn into Rm is a function L : Rn −→ Rm

satisfying (a) L(v + u) = L(v) + L(u) for all v, u ∈ Rn.
(b) L(cv) = cL(v) for all v ∈ Rn and c ∈ R.

Remark 4.3.ET4. The above conditions (a) and (b) is equivalent to the condition (ab)
L(cv + u) = cL(v) + L(u) for for all v, u ∈ Rn and c ∈ R
′
(or (ab ) L(cv + du) = cL(v) + dL(u) for for all v, u ∈ Rn and c, d ∈ R).

Linear Algebra lecture 9−11

 " #!
v1
Example. (Reflection with respect to x-axis in R2) L : R2 −→ R2 defined by L =
v2
" #
v1
is linear.
−v2

Example. (Projection of points in R3 into the xy-plane) L : R3 −→ R2 defined by

 
v1 " #
v1
L  v2  = is linear.
 
v2
v3
Example. (Scalar multiplication, dilation, contraction) Given r ∈ R, L : Rn −→ Rn
defined by L(v) = rv is linear.

Example. Given r ∈ R and b 6= 0m ∈ Rn, L : Rn −→ Rn defined by L(v) = rv + b is

not linear because L(2v) = r(2v) + b 6= 2(rv + b) = 2L(v).
Linear Algebra lecture 9−12
Fact. Given an m × n matrix, the function L : Rn −→ Rm defined by L(v) = Av is a
linear transformation.
Proof. 1◦ From matrix multiplication properties,
we immediately have L(v + u) = A(v + u) = Av + Au = L(v) + L(u)
and L(cv) = A(cv) = c(Av) = cL(v) for all v, u ∈ Rn and c ∈ R.

Theorem 4.3.6&7. If L : Rn −→ Rm is a linear transformation, then

(a) L(c1v1 + c2v2 + · · · + ck vk ) = c1L(v1) + c2L(v2) + · · · + ck L(vk ) for any vectors
v1, v2, . . . , vk in Rn and any scalars c1, c2, . . . , ck .
(b) L(v − u) = L(v) − L(u) for all u, v in Rn.
(c) L(0n) = 0m.

Proof. Skip.

Linear Algebra lecture 9−13

Corollary 4.2.1. Let L : Rn −→ Rm be a function. If L(0n) 6= 0m, then L is not a
linear transformation.

Example. Given r ∈ R and b 6= 0m ∈ Rn, L : Rn −→ Rn defined by L(v) = rv + b is

not linear because L(0n) = b 6= 0m.

Notation. In Rn, ei, for i = 1, . . . , n, denotes the vector that has ith component to be 1
and 0 otherwise.

Corollary. If L : Rn −→ Rm is a linear transformation, then L(v) = v1L(e1) +

v2L(e2) + · · · + vnL(en)

Linear Algebra lecture 9−14

Example 4.3.2. Let L : R3 −→ R2 be linear transformation for which we know that
     
1 " # 0 " # 0 " #
2 3 −1
L  0  = , L  1  = , and L  0  = . Please (a) find
     
−1 1 2
0 0 1
   
−3 v1
L  4  and (b) express L  v2 
   

2 v3
        
−3 1 0 0
(a) L  4  = L (−3)  0  + 4  1  + 2  0 
        

2 0 0 1
     
1 0 0
= (−3)L  0  + 4L  1  + 2L  0 
     

0 0 1
" # " # " # " #
2 3 −1 4
= (−3) +4 +2 = .
−1 1 2 11
Linear Algebra lecture 9−15
         
v1 1 0 0
(b) L  v2  = L v1  0  + v2  1  + v3  0 
        

v3 0 0 1
     
1 0 0
= v1L  0  + v2L  1  + v3L  0 
     

0 0 1
" # " # " # " #
2 3 −1 2v1 + 3v2 − v3
= v1 + v2 + v3 =
−1 1 2 −v1 + v2 + 2v3
 
" # v1
2 3 −1  
=  v2  .
−1 1 2
v3

Linear Algebra lecture 9−16

 Theorem4.3.8. Let L : Rn −→ Rm be a linear transformation. Then there exists
a unique m × n matrix A such that L(v) = Av for all v ∈ Rn. Furthermore A has
columns L(e1), L(e2), . . . , L(en).

Proof. 1◦ ”existence”
Given any v ∈ Rn, note v = v1e1 + v2e2 + · · · + vnen .
Hence by linearity of L, we have L(v) = hL(v1e1 + v2e2 + · · · + vnein )
= v1L(e1) + v2L(e2) + · · · + vnL(en) = L(e1) L(e2) · · · L(en) v = Av,
h i
where A = L(e1) L(e2) · · · L(en) , whose size is m × n.

2◦ ”uniqueness”
Suppose we also have L(v) = Bv for all v ∈ Rn.
Then (A − B)v = 0m for all v ∈ Rn.
Particular, letting v = ei, i = 1, 2, . . . , n,
we obtain the ith column of A − B is (A − B)ei = 0m.
Hence A and B agree.
h i
Definition. The matrix A = L(e1) L(e2) · · · L(en) is called the standard matrix
representing L.

Linear Algebra lecture 9−17