Singular Value Decomposition (SVD)
Isaac Amornortey Yowetu
December 31, 2021
Example 3
Find the matrices U Σ and V for the matrix
A =
"
3 1 1
−1 3 1
#
2/18
Finding the Singular Values
Solution
A =
"
3 1 1
−1 3 1
#
AT
=




3 −1
1 3
1 1




AAT
=
"
11 1
1 11
#
3/18
Characteristics Polynomial
det(AT
A) =
11 − λ 1
1 11 − λ
(1)
P(AT
A) = (11 − λ)(11 − λ) − 1 (2)
= λ2
− 22λ + 120 (3)
= (λ − 12)(λ − 10) (4)
Eigenvalues
λ1 = σ2
1 = 12 (5)
λ2 = σ2
2 = 10 (6)
4/18
Singular Values
σ1 =
p
λ1 =
√
12 (7)
σ2 =
p
λ2 =
√
10 (8)
Singular Values Decompostion of A
Σ =
"√
12 0 0
0
√
10 0
#
5/18
Constructing Matrix U
A = UΣV T
(9)
AAT
= UΣV T
(UΣV T
)T
(10)
= UΣT
V T
V ΣUT
(11)
= UΣ2
UT
(12)
= UDUT
(13)
AAT
=
"
11 1
1 11
#
(14)
6/18
Constructing Matrix U Cont’d
When λ = 12
"
11 − λ 1
1 11 − λ
#
=
"
−1 1
1 −1
#
(15)
By row reduction form, we have:
"
−1 1
1 −1
#
=>
"
−1 1
0 0
#
(16)
Forming equations with some variables:
−x + y = 0 (17)
Our eigenvector becomes: (1, 1)t
7/18
Constructing Matrix U Cont’d
When λ = 10
"
11 − λ 1
1 11 − λ
#
=
"
1 1
1 1
#
(18)
By row reduction form, we have:
"
1 1
1 1
#
=>
"
1 1
0 0
#
(19)
Forming equations with some variables:
x + y = 0 (20)
Our eigenvector becomes: (1, −1)t
8/18

Singular Value Decompostion (SVD): Worked example 3

  • 1.
    Singular Value Decomposition(SVD) Isaac Amornortey Yowetu December 31, 2021
  • 2.
    Example 3 Find thematrices U Σ and V for the matrix A = " 3 1 1 −1 3 1 # 2/18
  • 3.
    Finding the SingularValues Solution A = " 3 1 1 −1 3 1 # AT =     3 −1 1 3 1 1     AAT = " 11 1 1 11 # 3/18
  • 4.
  • 9.
    11 − λ1 1 11 − λ
  • 14.
    (1) P(AT A) = (11− λ)(11 − λ) − 1 (2) = λ2 − 22λ + 120 (3) = (λ − 12)(λ − 10) (4) Eigenvalues λ1 = σ2 1 = 12 (5) λ2 = σ2 2 = 10 (6) 4/18
  • 15.
    Singular Values σ1 = p λ1= √ 12 (7) σ2 = p λ2 = √ 10 (8) Singular Values Decompostion of A Σ = "√ 12 0 0 0 √ 10 0 # 5/18
  • 16.
    Constructing Matrix U A= UΣV T (9) AAT = UΣV T (UΣV T )T (10) = UΣT V T V ΣUT (11) = UΣ2 UT (12) = UDUT (13) AAT = " 11 1 1 11 # (14) 6/18
  • 17.
    Constructing Matrix UCont’d When λ = 12 " 11 − λ 1 1 11 − λ # = " −1 1 1 −1 # (15) By row reduction form, we have: " −1 1 1 −1 # => " −1 1 0 0 # (16) Forming equations with some variables: −x + y = 0 (17) Our eigenvector becomes: (1, 1)t 7/18
  • 18.
    Constructing Matrix UCont’d When λ = 10 " 11 − λ 1 1 11 − λ # = " 1 1 1 1 # (18) By row reduction form, we have: " 1 1 1 1 # => " 1 1 0 0 # (19) Forming equations with some variables: x + y = 0 (20) Our eigenvector becomes: (1, −1)t 8/18
  • 19.
    Constructing Matrix UCont’d Let Z be the egigenvectors of the various eigenvalues. Z = " 1 1 1 −1 # We normalized each column of Z to get U U = "√ 2 2 √ 2 2 √ 2 2 − √ 2 2 # 9/18
  • 20.
    Constructing Matrix V AT A=     3 −1 1 3 1 1     " 3 1 1 −1 3 1 # (21) =     10 0 2 0 10 4 2 4 2     (22) 10/18
  • 21.
    Constructing Matrix VCont’d Characteristics Polynomial det(AAT ) =
  • 29.
    10 − λ0 2 0 10 − λ 4 2 4 2 − λ
  • 37.
    (23) P(AAT ) = λ(λ− 12)(λ − 10) (24) Eigenvalues λ1 = σ2 1 = 12 (25) λ2 = σ2 2 = 10 (26) λ3 = σ2 3 = 0 (27) 11/18
  • 38.
    Constructing Matrix VCont’d When λ = 12     10 − λ 0 2 0 10 − λ 4 2 4 2 − λ     =     −2 0 2 0 −2 4 2 4 −10     (28) By row reduction form, we have:     −2 0 2 0 −2 4 2 4 −10     =>     −2 0 2 0 −2 4 0 4 −8     =>     −2 0 2 0 −2 4 0 0 0     (29) 12/18
  • 39.
    Constructing Matrix VCont’d Forming equations with some variables: −2x + 2z = 0 (30) −2y + 4z = 0 (31) Our eigenvector becomes: (1, 2, 1)t 13/18
  • 40.
    Constructing Matrix VCont’d When λ = 10     10 − λ 0 2 0 10 − λ 4 2 4 2 − λ     =     0 0 2 0 0 4 2 4 −8     (32) By row reduction form, we have:     2 4 −8 0 0 2 0 0 4     =>     2 4 −8 0 0 2 0 0 0     (33) 14/18
  • 41.
    Constructing Matrix VCont’d Forming equations with some variables: 2x + 4y = 0 (34) z = 0 (35) Our eigenvector becomes: (2, −1, 0)t 15/18
  • 42.
    Find the remainingeigenvector AX = 0 (36) " 3 1 1 −1 3 1 #     x y z     = " 0 0 # (37) By row reduction form, we have: " 3 1 1 −1 3 1 # => " 3 1 1 0 10 4 # => " 3 1 1 0 5 2 # (38) Forming equations with the variables: 3x + y + z = 0 (39) 5y + 2z = 0 (40) Our eigenvector becomes: (1, 2, −5)t 16/18
  • 43.
    Constructing Matrix VCont’d Z =     1 2 1 2 −1 2 1 0 −5     (41) We normalize each column vector of Z to obtain matrix V V =     √ 6 6 2 √ 5 5 √ 30 30 √ 6 3 − √ 5 5 √ 30 15 √ 6 6 0 − √ 30 6     (42) 17/18
  • 44.
    Singular Value Decomposition A= "√ 2 2 √ 2 2 √ 2 2 − √ 2 2 # "√ 12 0 0 0 √ 10 0 #     √ 6 6 √ 6 3 √ 6 6 2 √ 5 5 − √ 5 5 0 √ 30 30 √ 30 15 − √ 30 6     (43) 18/18