Scribd
Upload
3 views

Advance optimisation_theory part

The document discusses the Kuhn-Tucker conditions for quadratic programming, presenting the formulation of the problem, the introduction of slack variables, and the construction of the Lagrangian function. It also outlines the complementary slackness condition and provides a derivation of the Golden Section Method, explaining the relationship between the Fibonacci sequence and the golden ratio. Additionally, it details Gomory's Constraint, deriving it from the basic variable with the largest fractional value and establishing the necessary conditions for integer solutions.

Uploaded by

djagannath033
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
3 views

Advance optimisation_theory part

The document discusses the Kuhn-Tucker conditions for quadratic programming, presenting the formulation of the problem, the introduction of slack variables, and the construction of the Lagrangian function. It also outlines the complementary slackness condition and provides a derivation of the Golden Section Method, explaining the relationship between the Fibonacci sequence and the golden ratio. Additionally, it details Gomory's Constraint, deriving it from the basic variable with the largest fractional value and establishing the necessary conditions for integer solutions.

Uploaded by

djagannath033
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Q.

Kuhn-Tucker conditions for quadratic programming problem


Ans: Let us consider a quadratic programming problem in the form,
𝑛 𝑛 𝑛
1
Maximize 𝑍 = 𝑓(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = ∑ 𝑐𝑗 𝑥𝑗 + ∑ ∑ 𝑐𝑗𝑘 𝑥𝑗 𝑥𝑘
2
𝑗=1 𝑗=1 𝑘=1
𝑛

subject to, ∑ 𝑎𝑖𝑗 𝑥𝑗 ≤ 𝑏𝑖 ; 𝑖 = 1,2,3, … , 𝑚


𝑗=1
𝑥𝑗 ≥ 0 ; 𝑗 = 1,2,3, … , 𝑛

where, 𝑐𝑗𝑘 = 𝑐𝑘𝑗 ∀ 𝑗 = 1,2, … , 𝑛 and ∀ 𝑘 = 1,2, … , 𝑛, and 𝑏𝑖 ≥ 0, for 𝑖 = 1,2, … , 𝑚.

2
Introducing slack variable 𝑠𝑖 for 𝑖 = 1,2, … , 𝑚 and 𝑠𝑚+𝑗 for 𝑗 = 1,2, … , 𝑛, respectively.

Therefore, the problem becomes,


𝑛 𝑛 𝑛
1
Maximize 𝑍 = 𝑓(𝑥1 , … , 𝑥𝑛 ) = ∑ 𝑐𝑗 𝑥𝑗 + ∑ ∑ 𝑐𝑗𝑘 𝑥𝑗 𝑥𝑘 (1)
2
𝑗=1 𝑗=1 𝑘=1
subject to,
𝑛

∑ 𝑎𝑖𝑗 𝑥𝑗 + 𝑠𝑖 = 𝑏𝑖 ; 𝑖 = 1,2, … , 𝑚 (2)


𝑗=1
2
−𝑥𝑗 + 𝑠𝑚+𝑗 = 0; 𝑗 = 1,2, … , 𝑛 (3)

Now we construct the Lagrangian function:


𝑚 𝑛 𝑛
2
𝐿(𝑥, 𝑠; 𝜆) = 𝑓(𝑥) − ∑ 𝜆𝑖 (∑ 𝑎𝑖𝑗 𝑥𝑗 + 𝑠𝑖 − 𝑏𝑖 ) − ∑ 𝜆𝑚𝑗 (−𝑥𝑗 + 𝑠𝑚+𝑗 )
𝑖=1 𝑗=1 𝑗=1

Thus, Kuhn-Tucker conditions for the problem are given by:


𝑚
𝜕𝐿 𝜕𝑓
= − ∑ 𝜆𝑖 𝑎𝑖𝑗 + 𝜆𝑚+𝑗 = 0; 𝑗 = 1,2, … , 𝑛 (4)
𝜕𝑥𝑗 𝜕𝑥𝑗
𝑖=1
𝑛

𝜆𝑖 (∑ 𝑎𝑖𝑗 𝑥𝑗 − 𝑏𝑖 ) = 0; 𝑖 = 1,2, … , 𝑚 (5)


𝑗=1
𝜆𝑚+𝑗 𝑥𝑗 = 0; 𝑗 = 1,2, … , 𝑚 (6)
𝑛

∑ 𝑎𝑖𝑗 𝑥𝑗 − 𝑏𝑖 ≤ 0; 𝑖 = 1,2, … , 𝑚 (7)


𝑗=1

𝜆𝑖 , 𝜆𝑚+𝑗 , 𝑥𝑗 ≥ 0; 𝑖 = 1,2, … , 𝑚 and 𝑗 = 1,2, … , 𝑛 (8)


𝑛 𝑚

Equation (4)becomes using (1): 𝑐𝑗 + ∑ 𝑐𝑗𝑘 𝑥𝑘 − ∑ 𝜆𝑖 𝑎𝑖𝑗 + 𝜆𝑚+𝑗 = 0; 𝑗 = 1,2, … , 𝑛 (9)


𝑘=1 𝑖=1

Using (2), equation (5)becomes: 𝜆𝑖 𝑠𝑖 = 0 ; 𝑖 = 1,2, … , 𝑚 (10)


𝑚 𝑛

Again, combining (6)and (10): ∑ 𝜆𝑖 𝑠𝑖 + ∑ 𝜆𝑚+𝑗 𝑥𝑗 = 0 (11)


𝑖=1 𝑗=1

This condition is called complementary slackness condition.


Q. Derive Golden Section Method.
In the Fibonacci method,
𝐹𝑛−1
the interval of uncertainty at the end of the experiments is given by 𝐿2 = 𝐿
𝐹𝑛 0
𝐹𝑛−1
In the golden section method as n is very large, becomes 𝐿2 = lim ( ⋅ 𝐿0 )
𝑛→∞ 𝐹𝑛

Also in Fibonacci method 𝐿3 is given by,

𝐹𝑛−2 𝐹𝑛−2 𝐹𝑛−1 𝐹𝑛−1 2


𝐿3 = lim ( ⋅ 𝐿0 ) = 𝐿0 lim ( ⋅ ) = 𝐿0 lim ( )
𝑛→∞ 𝐹𝑛 𝑛→∞ 𝐹𝑛−1 𝐹𝑛 𝑛→∞ 𝐹𝑛

𝐹𝑛−1 3 𝐹𝑛−1 𝑘−1


Similarly, we get 𝐿4 = 𝐿0 ( lim ) In general, 𝐿𝑘 = 𝐿0 ( lim )
𝑛→∞ 𝐹𝑛 𝑛→∞ 𝐹𝑛

We know that the Fibonacci sequence is given by, 𝐹𝑛 = 𝐹𝑛−1 + 𝐹𝑛−2

𝐹𝑛 𝐹𝑛−2
or, =1+
𝐹𝑛−1 𝐹𝑛−1
(Taking both sides' limit as 𝑛 → ∞)
𝐹𝑛 𝐹𝑛−2
lim = 1 + lim
𝑛→∞ 𝐹𝑛−1 𝑛→∞ 𝐹𝑛−1

𝐹𝑛 1
or, lim =1+
𝑛→∞ 𝐹𝑛−1 𝐹
lim 𝑛−1
𝑛→∞ 𝐹𝑛−2

𝐹𝑛 1
or, lim =1+
𝑛→∞ 𝐹𝑛−1 𝐹
lim 𝑛
𝑛→∞ 𝐹𝑛−1

1 𝐹𝑛
𝑜𝑟, 𝑟 = 1 + Let, 𝑟 = lim
𝑟 𝑛→∞ 𝐹𝑛−1
𝑜𝑟, 𝑟 2 − 𝑟 − 1 = 0

1 ± √5 1 + √5
∴𝑟= =
2 2

1 + √5
Since r is a positive number, 𝑟= = 1.618
2
Which is the golden ratio or golden section.

1 𝑘−1
Hence, we have in general 𝐿𝑘 = 𝐿0 ( ) = (0.618)𝑘−1 𝐿0
𝑟

Therefore, in the golden section method, the interval of uncertainty, at the end of k th experiment is given by
𝐿𝑘 = (0.618)𝑘−1 𝐿0
Q. Derive Gomory's Constraint.

From table 1, choose the basic variable with the largest fractional value.
Let this basic variable be 𝑥𝑖 . This variable can be expressed from the 𝑖𝑡ℎ equation of the table 1 as,
𝑛

𝑥𝑖 = 𝑏‾𝑖 − ∑ 𝑎‾𝑖𝑗 𝑦𝑗 (1)


𝑗=1
where 𝑏‾𝑖 is a noninteger. Let us write,
𝑏‾𝑖 = 𝑏ˆ𝑖 + 𝛽𝑖 (2)

𝑎‾𝑖𝑗 = 𝑎ˆ𝑖𝑗 + 𝛼𝑖𝑗 (3)


where 𝑏ˆ𝑖 and 𝑎ˆ𝑖𝑗 denote integers. Thus, 0 < 𝛽𝑖 < 1 and 0 ≤ 𝛼𝑖𝑗 ≤ 1
𝑛 𝑛

Putting (2)& (3)in (1)we get, 𝑥𝑖 − 𝑏ˆ𝑖 + ∑ 𝑎ˆ𝑖𝑗 𝑦𝑗 = 𝛽𝑖 − ∑ αij yj (4)


𝑗=1 𝑗=1
Since 𝑥𝑖 and 𝑦𝑗 must be integers at an optimal integer solution, the RHS of (4) must be an integer.
𝑛

Thus, 𝛽𝑖 − ∑ 𝛼𝑖𝑗 𝑦𝑗 = integer (5)


𝑗=1

Since 𝛼𝑖𝑗 are nonnegative fractions and 𝑦𝑗 are nonnegative integers,


𝑛

Hence, 𝛽𝑖 − ∑ 𝛼𝑖𝑗 𝑦𝑗 ≤ 𝛽𝑖 < 1 (6)


𝑗=1
𝑛

Since, (𝛽𝑖 − ∑ 𝛼𝑖𝑗 𝑦𝑗 ) is integer, so it can be either a zero or a negative integer.


𝑗=1
𝑛

Hence, we obtain the desired constraint as, +𝛽𝑖 − ∑ 𝛼𝑖𝑗 𝑦𝑗 ≤ 0 (7)


𝑗=1

By adding a nonnegative slack variable 𝑠𝑖 , the Gomory constraint equation becomes,


𝑛

𝑠𝑖 − ∑ 𝛼𝑖𝑗 𝑦𝑗 = −𝛽𝑖 (8)


𝑗=1

where 𝑠𝑖 must also be an integer by definition.

You might also like