Scribd
Upload
53 views

Indian Institute of Technology, Bombay Chemical Engineering Cl603, Optimization Endsem, 27 April 2018

This document provides an exam for a chemical engineering optimization course. It contains 6 questions testing various optimization concepts: 1) Univariate function minimization using golden section search and interval halving methods 2) Unconstrained multivariate optimization using Newton's method and steepest descent 3) Formulating a machine parts production problem as a linear program 4) Solving a linear program using the simplex method 5) Deriving the KKT conditions for an equality constrained quadratic program 6) Finding the KKT points for a constrained minimization problem and determining which satisfy second order conditions

Uploaded by

Lakshay Chhajer
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
53 views

Indian Institute of Technology, Bombay Chemical Engineering Cl603, Optimization Endsem, 27 April 2018

This document provides an exam for a chemical engineering optimization course. It contains 6 questions testing various optimization concepts: 1) Univariate function minimization using golden section search and interval halving methods 2) Unconstrained multivariate optimization using Newton's method and steepest descent 3) Formulating a machine parts production problem as a linear program 4) Solving a linear program using the simplex method 5) Deriving the KKT conditions for an equality constrained quadratic program 6) Finding the KKT points for a constrained minimization problem and determining which satisfy second order conditions

Uploaded by

Lakshay Chhajer
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Indian Institute of Technology, Bombay Chemical Engineering

CL603, Optimization EndSem, 27 April 2018


Maximum Points: 75. Closed notes, closed book exam. Calculator is allowed. Wherever required,
make reasonable assumptions and state them.
1. Univariate function minimization using search methods: [3+2+2=7 points]
(a) Consider Golden section search method for minimization of a univariate function.
Recall that in Golden section search method the interior points in any interval are
at a distance of γL from either end where γ = 0.382 and L is the interval length.
Derive this value of γ?
(b) In general, if you start with an initial interval of length L, what will be the interval
length after N function evaluations when Golden section search is used. It can be
assumed that for the starting interval, the function values at the interval end points
are available.
(c) Same question as in part (b) but now using interval halving method? Justify/explain
your answers in parts (b) and (c)
2. Unconstrained multivariate optimization: Consider minimization of a function f (x) with
x ∈ Rn . [4+4+2=10 points]
(a) Derive expression for Newton’s search direction at an iterate xk . Also show that if
the Hessian of the function at xk is positive definite, then the Newton’s direction is
a descent direction.
(b) Prove that in the steepest descent method with exact line search, the successive
steps are perpendicular to each other.
(c) Describe what do you understand by a Cauchy step in a trust region method.
3. LP formulation: A manufacturing firm produces two machine parts using lathes, milling
machines, and grinding machines. The different machine times required for each part,
the manufacturing times available on different machines, and the profit on each machine
part are given in Table 1.

Table 1: Machine parts problem


Machining time required (hours) Maximum time available
Type of machine Machine part I Machine part II per week (hours)
Lathes 2 1 5
Milling machines 1 2 5
Grinding machines 1 1 3
Profit per unit $1 $3

The problem is of determining the number of parts I and II to be manufactured per week
to maximize the profit. To solve this problem, do the following [3+3+2+2=10 points]
(a) Write the problem as a linear programming optimization formulation. (There is no
need to add slack/surplus variables or perform any other transformation. Leave the
problem in its original form)

1
(b) Neatly Sketch the feasible region in the x1 − x2 plane with x1 , x2 being the number
of parts I and II manufactured per week, respectively.
(c) Identify the extreme points (vertices) of the feasible region i.e. show these extreme
points in the sketch as well as write down their x1 , x2 coordinates in your answers.
(d) By evaluating the objective function at these extreme points, identify the optimal
solution.

4. Consider the following LP problem: [4+3+4+2+3=16 points]

min z = −3x1 − 2x2 + x3


s.t. x1 − x2 + x3 = 1 (1)
3x1 − 2x2 + x4 = 6 (2)
xi ≥ 0, i = 1, 2, 3, 4

(a) Apply simplex method to solve the above problem. What can you say about the
solution? Give justification.
(b) Consider the same problem as in part (a) but with the objective function being to
minimize z = −3x1 + 4x2 . The constraints remain the same. Use simplex to obtain
the optimal solution.
(c) Obtain the Lagrange multipliers λ∗ and µ∗ corresponding to the optimal solution
in part (c). Show your calculations/justify your answers.
(d) Now again consider problem as in part (b) but with the right hand side of constraint
(1) replaced by 1.1 (in place of 1). Use results from parts (b) and (c) to calculate the
new objective function value without applying simplex from scratch again. What
would be the new objective function value if the right hand side of constraint (2) is
replaced by 6.1 (in place of 6) while keeping the right hand side of constraint (1) as
1.
(e) Write the dual formulation corresponding to the primal problem considered in part
(b). [Just write the formulation. There is no need to solve it.]

5. (a) Consider an equality constrained QP as:


1
min f (x) = xT Gx + dT x
2
s.t. Ax = b

where G is positive definite and matrix [A]m×n has rank m with m ≤ n. Using
KKT conditions show that the vector x∗ and the Lagrange multipliers λ∗ satisfying
the first order necessary KKT conditions can be given as: [4 points]

λ∗ = (AG−1 AT )−1 (AG−1 d + b)


x∗ = G−1 (AT λ∗ − d)

Note: For square A (m=n), these expressions can be simplified to:


λ∗ = (A−1 )T d + (A−1 )T GA−1 b, x∗ = A−1 b.

2
(b) Solve the following QP problem using active set method. Use initial iterate as
x0 = [2 0]T . [15 points]

min f (x) = 0.5(x21 + x22 ) − 3x1 − x2


s.t. − x1 − 0.5x2 + 2 ≥ 0
x1 − x2 ≥ 0
x2 ≥ 0

Consider the largest possible working set W0 to begin with. Show all your steps
neatly to get credit.
Note: In an active set method, at any given iteration, the step direction p is obtained
using a QP subproblem where the constraints are derived from the constraints in
the working set. The step length α in the step direction p is chosen to ensure that
feasibility with respect to all constraints is maintained. Whenever a full step (α = 1)
can be taken, the decision to drop a constraint from the working set is based on the
Lagrange multiplier, otherwise a blocking constraint is added to the working set.

6. Consider the following problem, [7+6=13 points]

min − 0.1(x1 − 4)2 + x22


s.t. x21 + x22 − 1 ≥ 0

(a) Find all KKT points (i.e. x∗ , λ∗ which satisfy the KKT conditions) of the above
problem. To do this consider both possibilities of constraint being active or inactive.
(b) Using second order conditions determine which of the KKT points obtained in part
(a) represent the constrained minimizers.

Misc Information

• Chain rule : When the vector x depends on another vector t that is x = x(t), then the
chain rule for obtaining gradient of a univariate function f (x(t)) is:
n
X ∂f
∇t f (x(t)) = ∇xi (t)
i=1
∂x i

• Inverse of a 2 × 2 matrix is given below:


   
a b −1 1 d −b
A= , A =
c d ad − bc −c a

End of Paper; Best of luck

You might also like