OUTLINE

D ORTHOGONAL VECTORS

D PROJECTIONS

D PROJECTION MATRICES

0 APPLICATION → LEAST SQUARES

^ 't x

x
x
T

-
ORTHOGONAL VECTORS -

Ijdotprodnet
×
, y are
orthogonal if
→
yT.x=o
aid

ieg/x=[ ! 11×11=1424325
'
,
7
If 11×11=514
11×112+14112=11 112 if
xdyxt.x-yt.y-cxtygtfxtyjxt.gr
x

-
O

=
xtxtytytxytytx
EE
''
"" " →
" "

Cy
 subspace s is orthogonal to subspace T
vector in S is 1 to
means
every every
vector in T (
they only meet at the
zero
.

neuter )

row
space

{
c. CAT )
dim -_ rank CA ) Row 1>
space
diner
Rn >
.
orthogonal
null
to the

NCA )
space
dim -
n - ,

NCA )
A- x = 0
if x E

- '

:
-

÷÷:3 -
.

( row 1
of A) . X = 0 so xd row 1

( row m
if A) . X = 0 so x d row 2

eat rank CA ) I

]
-
= r =

* =
23, Yo n
-
-
s

dim [ NCA )] =3 -
I = 2
117=2
n =3

4,2551¥;) ;¥µy
o
a- x
-

¥7
is
NCA ) plane perpendicular 3×1
to vector

fig) ← NCA )

(2)
 -
PROJECTIONS -

start with 2- dimensions

as if >a b
i.e e =
p
-

p :
projection of b
p
-
x. a

onto a

>n
p is a

multiple of
e -1 a vector a

ad e
e = b -
X .
oh

at .
e = 0

at.(b -
X a) = O

atb = at x. a xata =aTb

X =
Atb
ate

a. X
p x. a
= =

aa¥p
P
-
-

projection
.

T
matrix
.

p = ace → matrix
→ scalar

pfP
qaif4.fi
""

1K¥
'

'
a' "
P,
space of
Column "

•
p=
Ccp ) is
.

line
through a .

a
a.az

•
rank P )
( =L .ae

•
symmetric ,
PT =p
pn =P
e.g Pp's Pp
•
, .

@
 .ae/-sL-aOjagyOronu--s
a

aaa:
a .az 922 dim[ Ccp )J=r=1

P2=p
Try
Let's dimensional
try
3
space
-

"
Zarb e- b -

p AX=b no X
"
ie assume exists
etch)
Pg:> are that
there is
no solution

fee .

79
Lin >
)n

① ②

.ve#sine.iE--LaiiaI)
there any
'

n
Eats:S
solution
b
# CA)
p= Iga -15202
,
Find the right
some combination of combination , i.e .

the column vectors -

f- A a vector so
that elect)

p= A. I
AI ICCA )
find I key b
: -

La ,

aifb-AXI-oazqb-AII-ota2-aaiica.mn "

;
2×3

µ
 £ b- AI is
At (b A- 5) } erector 2

t,m
-

e in the
/ CCA )=p=z
Ate -0
nuohfyf.eu

aim
3-2=1 Atb -

ATA I = O

ATA I = Atb
' '

CATAI A' AREA AI Atb

''

ne -

'

I -
CATAJ Atb
A. I
p
-

'

p
-

-
ACATAI Atb
-

⑧ projection
matrix
'

F- ACATAJ AT pT=p
→ p'=P

.aatiI¥
'
'
'

a'

Finna anemia '

"¥¥¥¥?÷¥
\
.

is it invertible ?

(5)
 -
Ax = b

eny
1
A
f
=/!;)
=

's
fi
ar
i

a' A
HI!
go ] -12717:L see:#Ais
Ata
,

always inutile

:÷÷÷÷÷÷÷÷÷÷t
n ::: ::c
i:÷
by only looking at matrix A ,

: ::÷÷÷÷÷÷÷.
NC ATA) = NLA) → ATA is
invertible
"
÷
both sides
multiply
XT
by
XT ATA .
x = 0

Taxi

(6)
→x÷¥¥En
 "

Tf A-
-
enisted ,
meaning full"
rank
,
r =m=n

spans all
of IR
'
A- steel
if
'

Note : em

Entree
D=
ACA AT
at #I' AT
'
'
cases A .
A →
Egnatia
BECCA ), P.b=b
if bibe.EE#t
CAT )^ AT
'
A A' = I Battin's is

tf b ICCA ) ,
P b
.
- O
WITI Easter
because m
>n
b

-
LEAST SQUARES -

b
a
'
+ It objective is to find
E- a linear relationship
2 between two variables
b & t
*
objective
-
i
,

µ ' ' '

> t → b = Ct D. t
i 2 3
tib
Ch D= 1

¥÷÷÷i÷¥¥
' ) c +

Sometimes we carrot
¥ :# z

.no#ion-i:3-ks-ifz1 '

because
none

drat b
# CCA) A X b
b

ff
case
A- X=b
find
a
want to
we "
:*

:t;mzgm
tube

b onto CCA)

(7)
 A- TAI Atb
f.) ¥)
-

a-
* →
T T

=L: III. It : :]
'

a' a-
:

.b=¥;
At =

% ! :] iPod ! :]
-

E- in
2=213

I=[g) [2,1g ]
=

Try to find
I using
f=CAtATAtb
as an enexise

(8)