Scribd
Upload
40 views

School of Computer Science and Applied Mathematics

This document provides instructions for a 4 question, 35 mark online test in Optimization III. It outlines the test duration, submission requirements, and provides the questions and associated mark allocations. Question 1 involves finding the area of a rectangle as a function of x and maximizing this area. Question 2 determines conjugate direction vectors and uses the conjugate gradient method to minimize a quadratic function. Question 3 examines properties of the quasi-Newton algorithm applied to quadratic functions. Question 4 discusses determining a good model-objective function agreement in trust region methods and computes the next iterate using given Cauchy and full step points.

Uploaded by

Fanta Camara
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
40 views

School of Computer Science and Applied Mathematics

This document provides instructions for a 4 question, 35 mark online test in Optimization III. It outlines the test duration, submission requirements, and provides the questions and associated mark allocations. Question 1 involves finding the area of a rectangle as a function of x and maximizing this area. Question 2 determines conjugate direction vectors and uses the conjugate gradient method to minimize a quadratic function. Question 3 examines properties of the quasi-Newton algorithm applied to quadratic functions. Question 4 discusses determining a good model-objective function agreement in trust region methods and computes the next iterate using given Cauchy and full step points.

Uploaded by

Fanta Camara
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

School of Computer Science and Applied Mathematics

APPM 3017: Optimization III


Online Test
Lecturer: Matthews Sejeso Date: 30 September 2021

INSTRUCTIONS
• There are a total of four(4) questions and 35 marks allocated. Answer all questions.
• Show all your calculations. Your writing must be legible.
• The test duration is three(3) hours; submit your work by 17:00, 30 September 2021.
• Upload a single pdf file on Ulwazi. Do not submit anything via email.
• Save your document by student number.

QUESTION 1 [10 MARKS]


1.1 A rectangle has its lower-left corner at the origin (0, 0), and the upper right corner at the point (x, y)
on the graph

x2 + y − 9.

(a) Find the area of such a rectangle as a function of x alone. [2]


(b) Find the dimensions, x and y that maximize the area of the rectangle.
. [4]
1.2 Discuss the convergence rate of the sequence {xk } defined by
( k
1 2
k even
xk = 4
k−1
(x )/k k odd

[4]

QUESTION 2 [10 MARKS]


T T
2.1 Determine whether the following direction vectorsd0 = − 32 , 12 and d1 = − 12 , − 12 are conjugate
 

with respect to the following function at (x, y) = √13 , 1 ,

1 1
f (x, y) = x4 + y 2 − xy + x − y.
4 2
[2]
2.2 What is the main drawback of the conjugate direction method? How does the conjugate
gradient method address this issue? [2]
2.3 Use Fletcher-Reeves with exact line search to minimize the quadratic function
5 1
f (x, y) = x2 + y 2 + 2xy − 3x − y.
2 2
Use the starting point (x, y) = (1/2, −1/2). [4]
2.4 How will you update the stepsize for a non-quadratic function? Give a reason for your answer. [2]

1
QUESTION 3 [8 MARKS]
3.1 Consider the quasi-Newton algorithm xk+1 = xk −si Hk ∇f (xk ), si 6= 0 applied to a quadratic function
1
f (x) = xT Qx + bT x,
2
where Q ∈ Rn×n is a symmetric positive definite matrix, and b ∈ Rn . The approximate Hessian Hk
is updated iterative such that the quasi-Newton condition is satisfied, that is

Hk+1 γ i = δ i , 0 ≤ i ≤ k,

where γ i = ∇f (xi+1 ) − ∇f (xi ), δ i = xi+1 − xi . Show that the search directions d0 , d1 , . . . , dn−1
generated by the quasi-Newton algorithm are Q-conjugate. [5]

3.4 Consider the quasi-Newton algorithm xk+1 = xk − αk Mk ∇f (xk ), where f : R2 → R and Mk ∈ R2×2
is given by
 
1 0
Mk =
0 z

with z ∈ R and
 
αk = arg min f xk − αMk ∇f (xk ) .
α>0

Suppose at some iteration k we have ∇f (xk ) = [1, 1]T . Find the largest range of values of z
that guarantees that αk > 0 for any f . [3]

QUESTION 4 [7 MARKS]
4.1 In the trust region methods, how would you determine a good agreement between a model function
and the objective function? [2]
4.2 What role does the Cauchy point play in the trust region methods? [2]
4.3 At the k-th iteration of the trust region algorithm, the Cauchy point and the full step are respectively
given by
1  1 
k 2 k
dc = and dn = 21 .
0 −2
T
The current approximate solution is xk = 1, 21 . Given that the model function is a good approximation


of the objective function compute the next iterate xk+1 . Use the following parameters: trust region
bound ∆ = 32 , and trust region radius ∆k = 1. How will the trust region radius change? [3]

******* GOOD LUCK! *******

You might also like