Scribd
Upload
43 views7 pages

Constrained Optimization 1

This document discusses solving optimization problems using the Karush-Kuhn-Tucker (KKT) conditions. It provides an example of formulating and solving an optimization problem with constraints in multiple iterations. It also discusses determining if an optimum is global or not by analyzing the Hessian matrix and eigenvalues. Additional problems examine using KKT conditions to check if a point is optimal, solve an optimization problem with no active inequality constraints, and determine if a problem is convex or not based on eigenvalues of the Hessian.

Uploaded by

Erzhan Omarbekov
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
Download as docx, pdf, or txt
43 views7 pages

Constrained Optimization 1

This document discusses solving optimization problems using the Karush-Kuhn-Tucker (KKT) conditions. It provides an example of formulating and solving an optimization problem with constraints in multiple iterations. It also discusses determining if an optimum is global or not by analyzing the Hessian matrix and eigenvalues. Additional problems examine using KKT conditions to check if a point is optimal, solve an optimization problem with no active inequality constraints, and determine if a problem is convex or not based on eigenvalues of the Hessian.

Uploaded by

Erzhan Omarbekov
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1/ 7

Assignment 1 Problem 1

a) For the given annual costs for a centrifugal compressor, a reactor, and a recycle unit the optimization problem can be formulated as the following:

Now, we can assign functions for every constraints: ( ) ( ) ( ) b) Once the formulation is ready, the KKT point for the optimization problem can be determined: ( )

The values we have just found is considered as the first iteration. Now, we check our first iteration results for optimality by substituting the values to our constraints: ( ) ( ) ( ) Thus, we should include the violated constraints into active constraint in the problem formulation with corresponding multipliers present in the equation: ( )

{ ( ) ( ) {

That was the second iteration which gave us a negative value for one of the multipliers: It means that and calculated value for r can be excluded from the system of equations, since the multipliers should satisfy non-negativity condition. Stating that, a third iteration takes place:

{ { {

Here we have our optimal solution, because there is no violation of non-negativity of a multiplier:

Thus to minimize our objective function, there should be:

c) To determine whether the optimum is global or not, the following procedure is followed: Hessian matrix of the following form is to be found:

( ) [ ]

Now, we should find eigenvalues in the following form: ( Thus, |[ ] [ ]| | | )

( (

)( )

) ( )

Here we have obtained a quadratic equation in terms of eigenvalue . To determine whether it is global optimum or not it should be positive definite, i.e. the both eigenvalues should be nonnegative. To achieve this, lets find the roots of the quadratic equation: ( )

To make the roots non-negative, the first terms should be larger than the second term both in the numerator, i.e.:

should be more than ( Lets compare: This is our two terms to compare: Raising to 2nd degree both terms will not change the inequality sign: If we take P and r as non-negative, the whole term can be added to both sides of an inequality: Now we know the unknown inequality: ( ) )

Hence, both eigenvalues are non-negative, which means our optimum is global minimum.

Problem 2 Part A

( )

We have only inequality constraints, so the Lagrange conditions will look like follows, ( ) ( ) ( )

To check if (4, 2) is an optimal solution, we can make one of constraints (g1) active, and check whether the value of multiplier will be the same

This gives solution

1/3 and

4/9. The values differ which implies that (4, 2) is not an optimal

Part B ( Assume no inequality constraints active ) ( ) ( )

x1=0, and x2 has no solution. Thus, constraints having x2 should be considered

This yields x2=0, x1=2,

=1,

=1, and

=0.

Thus, KKT conditions are satisfied and the optimal solution is (2, 0).

Part C The convexity of a function is defined from the eigenvalues. ( ( ( ( ) [


( ) ( ) ( ( ) )

) ) ) ) ( ( ) ) ]
( )

Next, it is necessary to check if eigenvalues are positive, ( ( ( ( )( ( ( ( ) ( ) ) ( ) ) ) ) )

To satisfy the above condition, the term under square root must be less than the first term. To check the following technique can be used:

( ( ( So, ( (

) )

( ( (

( (

Thus, the problem is not a convex programming problem.

Part D ( )

The function is maximized when x1 takes its maximum values and x2 takes least possible value. From constraints it is seen that x1=2 and x2=0.

You might also like