Scribd
Upload
189 views

Operations On Matrices - Linear Algebra

Matrix notation, operations, row and column vectors, product AB, and transpose are introduced. Key points are: a matrix is a rectangular array of numbers; notation includes dimensions and individual entries; basic operations like addition and scalar multiplication are defined; row and column vectors are introduced; the product AB is defined using row and column vectors; and the transpose operator flips a matrix across its diagonal.

Uploaded by

docsdownforfree
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
189 views

Operations On Matrices - Linear Algebra

Matrix notation, operations, row and column vectors, product AB, and transpose are introduced. Key points are: a matrix is a rectangular array of numbers; notation includes dimensions and individual entries; basic operations like addition and scalar multiplication are defined; row and column vectors are introduced; the product AB is defined using row and column vectors; and the transpose operator flips a matrix across its diagonal.

Uploaded by

docsdownforfree
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 35

Matrix notation, operations, row and column

vectors, product AB(important), transpose

Matrix: a rectangular array of real numbers.


mxn-matrix: m rows and n columns
square matrix: nxn-matrix
Notation A=[a_ij]_mxn, A=[a_ij], (A)_ij=a_ij

" a11 a12 ... a1n


$
$ a21 a22 ... a2n
$M
M O M
$
#am1 am 2 ... amn

%
'
'
'
'
&

A=B if and only if they have same size and the


same entries: a_ij=b_ij for all i,j,
A+B: (A+B)_ij=A_ij+B_ij
A-B: (A-B)_ij=A_ij-B_ij
cA: (cA)_ij=c(A)_ij=ca_ij , -A= (-1)A.
See Ex 1,2,3.

Row vectors: r=[r_1, r_2,,r_n]; i.e., 1xn matrix


Column vectors: c=
" c1 %
$ '
$c 2 '
$M'
$ '
#c m &

Think of mxn matrix as m row n-vectors in a


column.
Think of it as n !
column m-vectors in a row.

[c_1,c_2,,c_n]=

" a11
$
$ a21
$ M
$
#am1

a12
a22

...
...

a1n
a2n

M
O M
am 2 ... amn

%
'
'=
'
'
&

r_i(A)=[a_i1,a_i2,..,a_in]
!
c_j(A)= " a1 j %
$ '
$a2 j '
$M '
$ '
#amj &

" r1 %
$ '
$ r2 '
$M '
$ '
#rm &

a11 x1 +
a21 x1 +
M
am1 x1 +

a12 x 2 + ... +a1n x n


a22 x 2 + ... +a2n x n
M
O
M
am 2 x 2 + ... amn x n

" a11 x1 +
$
$ a21 x1 +
$ M
$
#am1 x1 +

a12 x 2 +
a22 x 2 +

...
...

M
O
am 2 x 2 + ...

+a1n x n
+a2n x n
M
amn x n

= b1
= b2
M
= bm
% " b1 %
' $ '
' = $b2 '
' $M'
' $ '
& #bm &

" a11 %
" a12 %
" a1n % " b1 %
$ '
$ '
$ ' $ '
a
a
a2n ' $ b2 '
21 '
22 '
$
$
$
x1
+ x2
+ ...+ x n
=
$ M '
$ M '
$ M ' $M'
$ '
$ '
$ ' $ '
#am1&
#am 2 &
#amn & #bm &

We define Ax to be:

" a11
$
$ a21
$ M
$
#am1

a12 ... a1n


a22 ... a2n
M
O M
am 2 ... amn

%
'
'
'
'
&

" x1 %
$ '
$x2 '
$M'
$ '
#x n &

" a11 %
" a12 %
" a1n %
$ '
$ '
$ '
a
a
a2n '
21 '
22 '
$
$
$
= x1
+ x2
+ ...+ x n
$ M '
$ M '
$ M '
$ '
$ '
$ '
#am1&
#am 2 &
#amn &

!
Ax=[a_1,a_2,,a_n]=x_1a_1+.+x_na_n
!
where a_i are column vectors.
Thus, we can write a system of linear equations as
Ax=b for b mx1 column vector.

See examples top. Page 83 and Example 4.


Linearity Property: Theorem 3.1.5:
A mxn matrix, u,v column n-vector. Then
(a) A(cu) = c(Au), c a real number
(b) A(u+v)=Au+Av.
Equivalently A(cu+dv)=c(Au)+d(Av), c,d reals

Proof: Simply use definitions and follow

A mxn-matrix, B nxr matrix. Define AB mxr-matrix


so that (AB)x=A(Bx) for any r-vector x.
This is the associativity which is needed.
B=[b_1,b_2,,b_r]
Bx=x_1b_1+x_2b_2++x_rb_r
A(Bx) = A(x_1b_1+x_2b_2++x_rb_r )
=x_1Ab_1+x_2Ab_2++x_rAb_r
(AB)x must equal x_1(Ab_1)+x_2(Ab_2)++x_r(Ab_r)

Definition AB=[Ab_1,Ab_2,,Ab_r].
See Example 5.

A mxs matrix, B sxn matrix -> AB mxn matrix


(AB)_ij = a_i1b_1j+a_i2b_2j++a_isb_sj
= r_i(A)c_j(B) or r_i(A).c_j(B) (second as vectors)
Proof: AB=[Ab_1,Ab_2,,Ab_n]
=

Remove brackets to get a familiar formula

"" a11b11 + a12b21 + ... +a1sbs1


$$
$$ a21b11 + a22b21 + ... +a2sbs1
$$ M
M
O
M
$$
$#am1b11 + am 2b21 + ... amn bs1
$ " a b + a b + ... +a b
12 2n
1s sn
$ $ 11 1n
$..,$ a21b1n + a22b2n + ... +a2sbsn
$ $ M
M
O
M
$ $
# #am1b1n + am 2b2n + ... amn bsn

% " a11b12 +
'$
',$ a21b12 +
'$ M
'$
& #am1b12 +
%
'
'
'
'
&

a12b22 +

...

+a1sbs2

a22b22 + ... +a2sbs2


M
O
M
am 2b22 + ... amn bs2

%%
''
',.'
''
''
&'
'
'
'
'
'
&

Theorem, The ij entry of AB is the product of i-th row


vector of A times the j-th colmn vector of B.
See Example 7.

AB=[Ab_1,,Ab_n]. Thus c_j(AB)=Ac_j(B). This is


the column rule
Ax=x_1c_1(A)++x_sc_s(A) (A mxs x sx1)

r_i(AB)= r_i(A)B. Row rule. (Use Dot. Product rule


t see this.)
" r (A)B %
1

$
'
r
(A)B
'
AB = $ 2
$ M '
$
'
#rm (A)B&

yB=y_1r_1(B)++y_sr_s(B) (B sxn, y 1xs) (To see, this,


yB=[yb_1,,yb_n]=
!

"" y1b11 + % " y1b12 + % " y1b1n + %% " y b , y b ,..., y b


% " y1r1 (B) %
1 11 1 12
1 1n
$$
'$
' $
'' $
' $
'
y
b
+
y
b
+
y
b
+
$$ 2 21 ',$ 2 22 ',...,$ 2 2n '' =$+ y 2b21,+ y 2b22 ,...,+ y 2b2n ' = $+ y 2 r2 (B)'
$$ M ' $ M ' $ M '' $
' $ M
'
M
$$
'$
' $
'' $
' $
'
y
b
y
b
y
b
+
y
b
,+
y
b
,...,+
y
b
+
y
r
(B)
#
&
#
&
#
&
# s s1
&
s s2
s sn & #
s s1
s s2
s sn & #
s s

Theorem 3.1.8.
! of AB is a linear combination of
(a) the j-th column
columns of A with coefficients from j-th column of B.
(b) the i-th row of AB is a linear combinations of rows of
B with coefficients from the i-th row of A.

A mxn AT mxn rows become columns and vice


versa.
(AT)_ij=(A)_ji.
See Example 10.

Given nxn-matrix A (1x1 also), tr(A)= sum of the


diagonal entries A_11,A_22,,A_nn.

# 1 0 3&
%
(
A = %"1 "1 4(,trA =1
%$ 1 1 1('

trA=trAT.

Same size (nx1) column matrix u,v


uTv is 1x1 matrix or a number: matrix inner
product.
uTv = u.v = v.u = vTu

uvT is nxn matrix: matrix outer product


See Example 11
tr(uvT)=tr(vuT)=u.v.

1-14 recognition, computations.


16-22 computations, rules,,, traces

Properties, inverses

Theorem 3.2.1. a,b scalars, A,B,C same size


(a) A+B=B+A commutativity
(b) A+(B+C)=(A+B)+C
(c) (ab)A=a(bA)
(d) (a+b)A=aA+bA
(f) a(A+B)=aA+aB

Proof: simple calculations

AB is not necessarily equal BA. (not commutative)


See Example 1.
Some times AB=BA. Then A and B commute.
See Example *

Theorem 3.2.2. a, A mxn B pxq C rxs


(a) A(BC)=(AB)C (n=p, q=r)
(b)A(B+C)=AB+AC. (n=p=r, q=s)
(c) (B+C)A=BA+CA. (q=s=m, p=r)
(d) A(B-C)=AB-AC (e) (B-C)A=BA-CA
(f) a(BC)=(aB)C=B(aC) (q=r)

Proof (a) The rest is omitted.


Let c_j be the j-th column of C.
Question: What is j-th column of DC for some D?
j-th column of (AB)C is (AB)c_j.
j-th column of A(BC) is A(BC)_j. (BC)_j = Bc_j. Thus, A(Bc_j).
We showed A(Bx)=(AB)x for any vector x.

Zero matrix O: all the entries are 0.


A+O=O+A=A; must be of same size
A-A=A+(-A)=O
0A=O
OA=O (may be different size zero matrices)
If cA=O, then either c=0 or A=O.
See Example 3. AB=O but BA is not O.

[1],

"1
"1 0% $0
$
' ,$
#0 1& $0
$
#0

0
1
0
0

0
0
1
0

0%
'
0'
0'
'
1&

I_m denotes mxm identity matrix


A mxn matrix AI_n= A, I_mA= A
!Note that I_m is already in reduced row echelon form.
Conversely, a row reduced mxm-matrix either has
zero rows or equals I_m (Theorem 3.2.4)

A nxn matrix. If B is nxn and satisfy AB=BA=I_n,


then A is invertible and B is an inverse of A.
See Example 4.
A matrix may not have an inverse. (See Example
5)
When a row or a column of it is zero.
When two rows (or two columns) are the same
But there are more than these
We will figure out precisely when

Theorem 3.2.6. A invertible. B,C inverses. Then


B=C.
Thus, we denote the inverse of A as A-1.
The inverse of 2x2-matrix is easy to obtain:

"a b%
"
%
d
(b
1
(1
A = $ ',ad ( bc ) 0,A =
$
'
ad ( bc #(c a &
#c d&

This is a computational result


Example **, Example 8.

(AB)-1 =B-1A-1, A,B nxn matrix


Proof: (AB)(B-1A-1)=A(BB-1)A-1=AA-1=I and
(B-1A-1)AB=B-1(A-1A)B=B-1B=I. Now use
uniqueness

A0=I, An=AAA n-times


A-n= (A-1)n=A-1A-1A-n n-times
Ar+s=ArAs, (Ar)s=ArArAr = Ars

Theorem 3.2.9. A invertible


(a) A-1 is invertible and (A-1)-1=A.
(b) An is invertible and (An)-1=A-n=(A-1)n.
(c) kA is invertible and (kA)-1=k-1A-1

Proof (b): A-nAn=(A-1)nAn. AnA-n=An (A-1)n =I by


cancelation from the middle.
Thus, this operation is very similar to taking
powers in real numbers.

p(x)=a_0+a_1x++a_mxm
Matrix polynomial in A (nxn-matrix):
p(A) = a_0I+ a_1A++a_mAm.

See Example 12.


p_1(A)p_2(A)=(p_1p_2)(A)=(p_2p_1)(A)
=p_2(A)p_1(A)

Proof (e): A mxn, B nxs. AB mxs (AB)T sxm


BTAT sxm also.
ji-th entry of (AB)T is (AB)_ij = r_i(A)c_j(B)
= r_j(BT)c_i(AT) is the ji-th entry of BTAT.
We need to show
r_i(A)c_j(B) = r_j(BT)c_i(AT)

3.2.12 (e) state tr(AB)=tr(BA) for square matrices.


See Example.
Proof: A nxm, B mxn AB nxn
n

tr(AB) = " (AB) ii = " " Aij B ji = " " B ji Aij


i=1
m

= " (BA) jj = tr(BA)


j=1

i=1 j=1

j=1 i=1

Product of row and column vector to be useful


later

Proof: Let u=rT, v=c. uTv=tr(uvT). ((25) sec3.1)


Thus rc=tr(rTcT)=tr((cr)T)=tr(cr)

Au.v=u.ATv and u.Av=ATu.v


Proof: Use u.v= vTu. (*) Why true?
Au.v=vT(Au)=(vTA)u=(Av)Tu=u.ATv.
u.Av=(Av)Tu=(vTAT)u=vT(ATu)=ATu.v

In the dot product, A moves across the dot by


transposing.

1-6 confirmation by direct computations for


specific matrices
7, 8 find unknown
9-12 confirmation
13-18 computations
32-37 a bit harder

You might also like