Scribd
Upload
27 views6 pages

MIT15 093J F09 Final 2003

Uploaded by

benjalobosm1999.blm
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
27 views6 pages

MIT15 093J F09 Final 2003

Uploaded by

benjalobosm1999.blm
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/ 6

15.

093 Optimization Methods


Final Examination
Instructions:
1. There are 5 problems each with 20 points for a maximum of 100 points.
2. You are allowed to use class notes, your homeworks, solutions to home-
work exercises, and the book by Bertsimas and Tsitsiklis. You are not
allowed to use any other book.
3. You have 3 hours to work in the examination.
4. Please explain your work carefully.
5. Good luck!

1
Problem 1 (20 points)
Please answer true or false. No explanation is needed. A correct answer is
worth 2 points, no answer 0 points, a wrong answer -1.
1. A problem of maximizing a convex, piecewise-linear function over a
polyhedron can be formulated as a linear programming problem.
2. The dual of the problem min x1 subject to x1 = 0, x1 x2 0 has a
nondegenerate optimal solution.
3. If there is a nondegenerate optimal basis, then there is a unique optimal
basis.
4. An optimal basic feasible solution is strictly complementary.
5. The convergence of the primal-dual barrier interior point algorithm is
aected by degeneracy.
6. Given a local optimum x for a nonlinear optimization problem it always
satises the Kuhn-Tucker conditions when the gradients of the tight
constraints and the gradients of the equality constraints at the point x
are linearly independent.
7. In a linear optimization problem with multiple solutions, the primal-
dual barrier algorithm always nds an optimal basic feasible solution.
8. In the minimum cost ow problem with integral data all basic feasible
solutions have integral coordinates.
9. The convergence of the steepest descent method for quadratic problems
min f (x) = x Qx highly depends on the condition number of the matrix
0

Q. The larger the condition number the slower the convergence.


10. The convergence of the steepest descent method highly depends on the
starting point.

2
Problem 2 (20 points)
Let f () be a concave function. Consider the problem

minimize
Xn f (xj )
j �1
subject to Ax = b
x 0:
Let P = fx 2 Rn j Ax = b x 0g:
(a) Prove that if there exists an optimal solution, there exists an optimal
solution which is an extreme point of P .
(b) Suppose we now add the constraints that xj 2 f0 1g for all j , i.e., the
problem becomes a 0-1 nonlinear integer optimization problem. Show
that the problem can be reformulated as a linear integer optimization
problem.
Problem 3 (20 points) Let Q be nn matrices. The matrix is positive
semidenite.
minimize c x + 21 x Qx
0 0

subject to d x + 21 x x  a:
0 0

(a) Write the Kuhn-Tucker conditions.


(b) Propose an algorithm for the problem based on the Kuhn-Tucker con-
ditions.
(c) Suppose the matrix Q is also positive semidenite. Reformulate the
problem as a semidenite optimization problem.

3
Problem 4 (20 points)
(a) You are given points (xi ai), i = 1 : : : m where xi are vectors in <n
and ai are either 0 or 1. The interpretation here is that point xi is
of category 0 or 1. We would like to decide whether it is possible to
separate the points xi by a hyperplane f x = 1 such that all points of
0

category 0 satisfy f x  1 and all points of category 1 satisfy f x > 1.


0 0

Propose a linear optimization problem to nd the vector f .


(b) You are given points (xi yi), i = 1 : : : m, where xi are vectors in <n
and yi 2 < are response variables. We would like to nd a hyperplane
a x = 1, such that for all points a xi  1, yi  1 xi, while if a xi > 1,
0 0 0 0

yi  2 xi. More formally, for all those points xi with a xi  1, we


0 0

will choose 1 in order to minimize


X jyi ; 1
0
xij
i: a xi
0
�1

while for all those points xi with a xi > 1, we will choose


0
2 in order
to minimize
X jyi ; 2
0
xij:
i: a0 xi >1
Propose an integer programming problem to nd the vectors a 1 2.

4
Problem 5 (20 points) Consider the problem
Z = min c x
� 0

s:t: a1x b1
0

a2x b2
0

x 2 f0 1gn
with a1 a2 c 0: We denote by ZLP the value of the LP relaxation.
(a) Consider the relaxation:
Z1 = max 0 min c x ; (b2 ; a2x)
�
0 0

s:t: a1x b1
0

x 2 f0 1gn
Indicate how you can compute the value of Z1 . What is the relation
among Z Z1 ZLP . Please justify your answer.
�

(b) Propose a dynamic programming algorithm for computing Z : �

5
MIT OpenCourseWare
http://ocw.mit.edu

15.093J / 6.255J Optimization Methods


Fall 2009

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.

You might also like