You are on page 1of 6
UNIVERSITY OF PRETORIA DEPT OF MATHEMATICS AND APPLIED MATHEMATICS WTW 750 Mathematical Optimization EXAMINATION 8 June, 2015 ‘TIME: 3 br MARKS: 50 External examiners: Prof Luhandjula (UNISA) Prof R Anguelov (UP) Internal examiners: Prof M.K. Banda. READ THE FOLLOWING INSTRUCTIONS, 1. The paper consists of pages 1 to 7 (questions 1 to 7). Check whether your paper is complete. 2. Give your answers in the answer book provided 3. Start the answer to each question on a new page. 4. No pencil work or any work in red ink will be marked. 5. The use of textbooks or notes is not allowed. Question 1. Using the Simplex Algorithm, solve the following problem fl subject to day +2, — 25 2 +305 <6 3x +2) +31 <9 my +2.— 23 <2 (A) 21, £2, 1320 Question 2. Consider the constrained minimisation problem for f(r)'= x} + (v2 +1) (e2\ o min F(z) subject to \ -x <0 2 H-m<0 -¥a= (P) “ 2.1 Determine the cone of feasible difcctions, the linearising Yone and the cone of tan- gents at (0,0) for this optimisation problem. , 2.2 Use 2.1 to deduce that (0,0) is a KKT point for problem (P2). 23 Use 2.2 to deduce that (0,0) is the global solution of (Ps). Question 3. Consider the optimisation problem: min F(@) subject to (Ps) g(x) <0 for ie Z := {1,...,m} hy(z) =O for je € := {1,...,P}- Prove the Karush-Kuhn-Tucker Theorem for the optimisation problem (P) given above. You may use the Proposition and Lemma provided in the last page without proof. Question 4. Consider the problem 344 min yee (P) 4 subject to dta}<. 4.1 Determine the first-order necessary optimality conditions for (P,) and find all KKT. points 4.2 Using second-order sufficient conditions, determine if any of the KKT points in 4.1 are minimisers. Question 5. +4] 5.1 Consider the function S(@1, 2) Compute the first two iterates 2°) and 2 of the steepest descent method applied to the objective function f (x1,.72) starting with 0 = (2,2). r+ ayaa + 2h. 5.2 Find the minimum of the function 1,52 F@im) = 521+ 50 using the conjugate gradient method starting with 2 = (5, 1) Question 6. Let m,n € N with m < ni. C © R"™™ symmetric positive semi-definite matrix, Suppose that ce R", Ae R™™ and be R™, Consider the following quadratic problem (QP): min deTCn + la e (QP) } subject to - ATr i if ry > 4 and |dy|> = Oy then all); Apai=2A,, else Basi = An end if 2) sm al 4 yd Algorithm 2 Hestenos - Sticfel Conjugate Gradient Algorithm Let go = Vf(2 = Ar + b, do = —go to a given 2 € R". Starting step is steepest descent, for k = 0,1,...,n—1: do if 9, = 0 then STOP; x) is the minimum point of f else ae pC a Rien ks ds a) alt) sm al) + Axde, where 2 as Hence, we have f(x) = D) gen = Vlas daar min f(x + Adg) - Exact Line Search. Gee + Tes idy with loesall Toul he end if « _end for 2 Algorithm 3 QP Active Set Method 0} Let e € F and Jy = A(x), If the points 2,...,2" € F and the corresponding index sets Jo,..., Jy are deter- mined for a k € No, then we obtain 2+) and Ji, as follows: 1) Determine a minimizer y to (a) ae ate = b, for all we J 2) Case 1: ye F Solve Y) yay = ihe If u, 2 0 for all ye Jz, then the point y is a minimiser to (CP) and the algorithm stops. Otherwise choose an index * € J, with u, <0, for example s € Je minimal with us = min{uy|u€ Je}. Set Jezi = A(y)\{s} and x) = y® (deactivation step) 3) Case 2: 6" ¢ F: For d:= y — 2 we determine a maximal a > 0 with Vit) ale +ad)

You might also like