UNIVERSITY of NEW SOUTH WALES SCHOOL of MATHEMATICS AND STATISTICS SESSION 1, JUNE 2007 optimization TIME ALLOWED - Two hours. TOTAL NUMBER OF QUESTIONS - three.
UNIVERSITY of NEW SOUTH WALES SCHOOL of MATHEMATICS AND STATISTICS SESSION 1, JUNE 2007 optimization TIME ALLOWED - Two hours. TOTAL NUMBER OF QUESTIONS - three.
UNIVERSITY of NEW SOUTH WALES SCHOOL of MATHEMATICS AND STATISTICS SESSION 1, JUNE 2007 optimization TIME ALLOWED - Two hours. TOTAL NUMBER OF QUESTIONS - three.
Optimization (1) TIME ALLOWED - Two hours (2) TOTAL NUMBER OF QUESTIONS - 3 (3) ANSWER ALL QUESTIONS (4) THE QUESTIONS ARE NOT OF EQUAL VALUE (5) ALL STUDENTS MAY ATTEMPT ALL QUESTIONS. MARKS GAINED ON ANY QUESTION WILL BE COUNTED. GRADES OF PASS AND CREDIT CAN BE GAINED BY SATISFACTORY PERFORMANCE ON UNSTARRED QUESTIONS. GRADES OF DISTINCTION AND HIGH DISTINCTION WILL REQUIRE SATISFACTORY PERFORMANCE ON ALL QUESTIONS, INCLUDING STARRED QUESTIONS (6) THIS PAPER MAY BE RETAINED BY THE CANDIDATE (7) ONLY THE PROVIDED ELECTRONIC CALCULATORS MAY BE USED All answers must be written in ink. Except where they are expressly required pencils may only be used for drawing, sketching or graphical work. SESSION 1, JUNE 2007 MATH3161/MATH5165 Page 2 1. [23 marks] Minimize subject to (NLP) i) Consider the following nonlinear programming problem 2 2 9 Xl +x 2 - 2" X I - 4X2 X2 - xi 2 0, 3XI +2X2 - 9 = O. a) Sketch the feasible region of (NLP). b) Is the problem (NLP) a convex programming problem? Give reasons for your answer. c) Show that the point x* = [ ~ ~ f is a regular point for (NLP). d) Verify that the point x* = [ ~ ~ J T i s a constrained stationary point for (NLP). e) Determine whether or not the point x* = [ ~ ~ f is global minimizer of (NLP). ii) Consider the following optimization problem (P) Minimize xi +x ~ +x ~ subject to Xl +2X2 +X3 = 1. a) Can you guarantee that a global minimizer exists for this problem? Give reasons for your answer. b) Formulate the (quadratic) penalty function problem for (P) with parameter /-L > O. c) Let x(/-L) = /-L [1 2 If. Verify that, for each /-L > 0, x(/-L) is a 6/-L +1 global minimizer of the penalty function problem of part b). d) Find the limit point x* of the vectors x(/-L) as /-L -7 00. e) Using second-order sufficient optimality conditions, show that the limit point x* of part cl) is a strict local minimizer of (P). Please see over ... SESSION 1, JUNE 2007 MATH3161/MATH5165 Page 3 2. [22 marks] Consider the quadratic function on lR. n , where d is a constant n x 1 vector, G is a constant n x n symmetric matrix and c is a scalar. i) a) Write down the gradient \7q(x) and the Hessian \7 2 q(x) of q. b) Suppose that q(x) is a convex function. State what can then be deduced about G. c) Stating clearly any theorems that you use, show that x* is a mini- mizer for q(x) if and only if G is positive semi-definite and Gx* + d = O. d)* If q(x) is a strictly convex function then show that G is positive definite. e)* Show that x* is the unique minimizer for q(x) if and only if G is positive definite and Gx* + d = O. ii) Consider applying Newton's method to q(x) starting at x(1) when G is positive definite. Let x* be the minimizer of q(x) and x(1) # x*. a) Write down the Newton direction at x(1) and show that it is a descent direction. b) How many iterations will Newton's method take to reach the mini- mizer x* of q(x). Give reasons for your answer. iii) Consider applying the method of steepest descent with exact line searches to q(x) starting at x(1) when G is positive definite. Let x* be the minimizer of q(x) and x(1) # x*. a) Show that the initial search direction is S(l) = G(x* - x(l)). b) If the next iterate X(2) = x* then show that the step length 0'(1) is an eigenvalue of G- 1 . c) Let Find the minimizer x* of q(x) and estimate the least number of iterations that would be sufficient to reach a point within a distance of 0.5 x 10- 6 of the minimizer x*. Please see over ... SESSION 1, JUNE 2007 MATH3161/MATH5165 Page 4 3. [20 marks] Consider the following optimal control problem Minimize f 0 6 x 2 dt uEU In subject to x= u x(O) = 1, x(6) = 4 with bounded controls U := {u I 0 ::; u(t) ::; 1, Vt E [0, 6]}. i) Write down the Hamiltonian function H for this problem. [You may assume the problem is normal and set Zo = -1.] ii) Write down the differential equation for the costate variable z. iii) State clearly the Pontryagin Maximum Principle conditions that an op- timal solution satisfies. iv) From the conditions in part iii) find the optimal control u* as a function of z. v) Using a phase-space diagram find the optimal control u*(t) as a function oft. vi) Find the formula for the optimal trajectory x* (t).