CE 771 Lecture 4 Interaction PDF

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

9/2/20

CE 771: Optimization in Civil


Engineering

Multi variable optimization: Unconstrained


Problems and solutions

AvijitMaji©

HESSIAN MATRIX

AvijitMaji©

1
9/2/20

Positive, negative Definite


• If A is 𝑛×𝑛 diagonal matrix

𝑑! ⋯ 0
A= ⋮ ⋱ ⋮
0 ⋯ 𝑑"
Then A is :
• Positive definite if and only if 𝑑# > 0 for i=1,2,……,n
• Negative definite if and only if 𝑑# < 0 for i=1,2,……,n

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 3

Positive, negative Definite


• Positive semidefinite if and only if 𝑑# ≥ 0 for 𝑖 =1,2,……,n
• Negative semidefinite if and only if 𝑑# ≤ 0 for 𝑖 =1,2,……,n
• Indefinite if and only if 𝑑# > 0 for some indices 𝑖 , 1 ≤ 𝑖 ≤ 𝑛 and
negative for other indices

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 4

2
9/2/20

Converting matrix into diagonal matrix


−14 −28 −44
• Hessian matrix = −7 −14 −23
9 18 29

• Corresponding diagonal matrix after transformation-

−1 0 0
0 0 0
0 0 2
• Thus eigen values are -1,0,2 which clearly states the hessian matrix is
indefinite that the particular solution would be a saddle point

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 5

Positive, negative Definite


• Let A is 𝑛×𝑛 symmetric matrix, and let A𝑘 be the submatrix of A
obtained be taking the upper left-hand corner k×k submatrix of A
and ∆$ = det 𝐴$ , the kth principal minor of A. Then
• Positive definite if and only if ∆$ > 0 for k=1,2,……,n
• Negative definite if and only if (−1)$ ∆$ < 0 for k=1,2,……,n
• Positive semidefinite if ∆$ > 0 for k=1,2,……,n-1 and ∆" = 0
• Negative semidefinite if and only if (−1)$ ∆$ > 0 for k=1,2,……,n-1
and ∆" = 0

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 6

3
9/2/20

Example 1
Find the solution of the multivariable unconstrained optimization
problem :

𝑓 𝑥, 𝑦 = 𝑥2 – 2𝑥𝑦 + 4𝑦2

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 7

Solution
Necessary condition for the optimum point
#$
#%
= 2𝑥 − 2y = 0 ∴𝑥=y
#$
#&
= −2𝑥 + 8y = 0 ∴ x = 4y
Possible decision variable value 𝑥 ∗ , 𝑦 ∗ = (0,0)
To find whether the solution found corresponds to a maximum or a minimum we
apply the sufficiency conditions to𝑓 𝑥, 𝑦 . The second-order partial derivatives of f
at 𝑥 ∗ , 𝑦 ∗ are given by :
#! $ #! $
= 2, 𝑎𝑡 (𝑥 ∗ , 𝑦 ∗ ) = 8, 𝑎𝑡 (𝑥 ∗ , 𝑦 ∗ )
#% ! #&!
#! $ #! $
#%#&
= −2, 𝑎𝑡 (𝑥 ∗ , 𝑦 ∗ ) #(#%
= −2, 𝑎𝑡 (𝑥 ∗ , 𝑦 ∗ )

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 8

4
9/2/20

Solution Contd.
Sufficiency conditions:
2 −2
• Hessian matrix :
−2 8
• Principal determinants:
)
#! $ #! $ #! $ #! $
#% !
=2 >0 𝑎𝑛𝑑 #% ! #&!
− #%#&
= 12 (> 0)

the Hessian matrix of f is Positive definite at 𝑥 ∗ , 𝑦 ∗ . Hence the point 𝑥 ∗ , 𝑦 ∗


corresponds to the minimum of f and 𝑓*+% = 0

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 9

Example 2
Find the solution of the multivariable unconstrained optimization
problem

𝑓 𝑥, 𝑦 = 𝑥 % − 3𝑥 & + 𝑦 % − 3𝑦 &

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 10

10

5
9/2/20

Solution
Necessary condition for the optimum point
𝜕𝑓
= 3𝑥 ) − 6𝑥 = 0
𝜕𝑥
#$
#&
= 3𝑦 ) − 6y = 0
Possible decision variable value 𝑥 ∗ , 𝑦 ∗ = 0,0 , 0,2 , 2,0 , (2,2)
To find whether the solution found corresponds to a maximum or a
minimum we apply the sufficiency conditions to𝑓 𝑥, 𝑦 . The second-order
partial derivatives of f at 𝑥 ∗ , 𝑦 ∗ are given by :
# !$ # !$
! = 6x − 6, 𝑎𝑡 (𝑥 ∗ , 𝑦 ∗ ) = 6y − 6, 𝑎𝑡 (𝑥 ∗ , 𝑦 ∗ )
#% #& !
# !$ # !$
#%#&
= 0, 𝑎𝑡 (𝑥 ∗ , 𝑦 ∗ ) #(#%
= 0, 𝑎𝑡 (𝑥 ∗ , 𝑦 ∗ )

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 11

11

Sufficiency conditions:
6𝑥 − 6 0 −6 0
q Hessian matrix for (0,0): 0 6𝑦 − 6
=
0 −6
• Principal determinants:
)
#! $ #! $ #! $ #! $
= −6 < 0 𝑎𝑛𝑑 − = 36 (> 0)
#% ! #% ! #&! #%#&
the Hessian matrix of f is negative definite at 𝑥 , 𝑦 . Hence the point 𝑥 ∗ , 𝑦 ∗ ∗ ∗
= (0,0) corresponds to maximum of f and fmax = 0

6𝑥 − 6 0 −6 0
qHessian matrix for (0,2): =
0 6𝑦 − 6 0 6
• Principal determinants:
)
#! $ #! $ #! $ #! $
= −6 < 0 𝑎𝑛𝑑 − = −36 (< 0)
#% ! !
#% #& ! #%#&
the Hessian matrix of f is indefinite at 𝑥 , 𝑦 . Hence the point 𝑥 ∗ , 𝑦 ∗ = (0,2)
∗ ∗
corresponds to the saddle point

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 12

12

6
9/2/20

6𝑥 − 6 0 6 0
qHessian matrix for (2,0): =
0 6𝑦 − 6 0 −6
• Principal determinants:
)
#! $ #! $ #! $ #! $
=6 >0 𝑎𝑛𝑑 − = −36 (< 0)
#% ! #% ! #&! #%#&
the Hessian matrix of f is indefinite at 𝑥 ∗ , 𝑦 ∗ . Hence the point 𝑥 ∗ , 𝑦 ∗ = (2,0)
corresponds to the saddle point

6𝑥 − 6 0 6 0
qHessian matrix for (2,2): =
0 6𝑦 − 6 0 6
• Principal determinants:
)
#! $ #! $ #! $ #! $
#% !
=6 >0 𝑎𝑛𝑑 #% ! #&!
− #%#&
= 36 (> 0)
the Hessian matrix of f is positive definite at 𝑥 ∗ , 𝑦 ∗ . Hence the point 𝑥 ∗, 𝑦 ∗ =
(0,2) corresponds to minimum of f and fmin = -4
1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 13

13

Example 3
Example: Find the solution of the multivariable unconstrained
optimization problem
𝑓 𝑥, 𝑦 = 𝑥 % − 12𝑥𝑦 + 8𝑦 %

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 14

14

7
9/2/20

Solution
Necessary condition for the optimum point
𝜕𝑓
= 3𝑥 ) − 12𝑦 = 0
𝜕𝑥
𝜕𝑓
= −12x + 24𝑦 ) = 0
𝜕𝑦
Possible decision variable value 𝑥 ∗ , 𝑦 ∗ = 0,0 , (2,1)
To find whether the solution found corresponds to a maximum or a minimum we
apply the sufficiency conditions to 𝑓 𝑥, 𝑦 . The second-order partial derivatives of f
at 𝑥 ∗ , 𝑦 ∗ are given by
#! $ #! $
#% !
= 6x, 𝑎𝑡 (𝑥 ∗ , 𝑦 ∗ ) #&!
= 48y, 𝑎𝑡 (𝑥 ∗ , 𝑦 ∗ )
#! $ #! $
= −12, 𝑎𝑡 (𝑥 ∗ , 𝑦 ∗ ) = −12, 𝑎𝑡 (𝑥 ∗ , 𝑦 ∗ )
#%#& #(#%

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 15

15

Solution Contd.
Sufficiency conditions:
0 −12
• Hessian matrix for (0,0):
−12 0
• Principal determinants:
&
" !# " !# " !# " !#
"$ ! = 0 𝑎𝑛𝑑
"$ ! "% !− "$"%
<0
the Hessian matrix of f is indefinite at 𝑥 ∗ , 𝑦 . Hence the point 𝑥 ∗ , 𝑦 ∗ = (0,0)

corresponds to the saddle point
12 −12
• Hessian matrix for (2,1):
−12 48
• Principal determinants:
&
" !# " !# " !# " !#
>0 𝑎𝑛𝑑 − >0
"$ ! !
"$ "% ! "$"%
the Hessian matrix of f is positive definite at 𝑥 ∗ , 𝑦 ∗ . Hence the point 𝑥 ∗ , 𝑦 ∗ = (2,1)
corresponds to the minimum of f with fmin = -8

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 16

16

8
9/2/20

Example 4
• Find the solution of the multivariable unconstrained optimization
problem :
𝑓 𝑥, 𝑦 = 4𝑥 & − 4𝑦 %

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 17

17

Solution
• Necessary conditions for the optimum point
#$
#%
= 8𝑥 = 0
#$
#&
= −12𝑦 ) = 0
Decision variable values: (x, y) = (0, 0)
• Sufficiency conditions:
8 0
Hessian matrix 𝐴 =
0 −24𝑦
)
#! $ #! $ #! $ #! $
Principal determinants 𝐴, = #%! > 0 and 𝐴) = #%! #&! − #%#&
= 0 at (0,0)
• Hence this is a case of Positive Semidefinite

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 18

18

9
9/2/20

Example 5
• Find the solution of the multivariable unconstrained optimization
problem :
𝑓 𝑥, 𝑦 = 3𝑦 % − 3𝑥 &

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 19

19

Solution
• Necessary conditions for the optimum point
"#
= −9𝑥 = 0
"$
"#
= 9𝑦 & = 0
"%
Decision variable values: (x, y) = (0, 0)
• Sufficiency conditions:
−9 0
Hessian matrix 𝐴 = It is not a positive definite nor positive semidefinite
0 18𝑦
Lets check for Negative definite and negative semidefinite
9 0
Hessian matrix −𝐴 =
0 −18𝑦

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 20

20

10
9/2/20

Solution Contd.
)
#! $ #! $ #! $ #! $
Principal determinants 𝐴, = #%! > 0 and 𝐴) = #%! #&! − #%#&
= 0 at (0,0)
• Hence this is a case of Negative Semidefinite

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 21

21

Thank you

1st Semester of 2020-21 Academic year CE 771: Optimization in Civil Engineering (Online course) AvijitMaji© 22

22

11

You might also like