NOTE: Four double-sized sheets of notes are allowed. No other notes and books are
allowed. A calculator is allowed. Keep in mind that the number of points does not
necessarily indicate the difficulty level of the question.
Problem 1 (16 points total): Specify whether each of the following sets is a convex
set. Justify your answers.
(a) (4 points) X = x ∈ RN | kxk < 1
(b) (4 points) X = x ∈ RN | kxk = 1
Specify whether each of the following functions is a convex function. Justify your answers.
(c) (4 points) f (x) = x21 x32 , where x ∈ R2 with x1 , x2 ≥ 0.
(d) (4 points) f (x) = x21 x42 , where x ∈ R2 with x1 , x2 ≥ 0.
Problem 2 (12 points total): Consider the following convex optimization problem.
Write down the primal optimal cost of the problem. HINT: Do you need to resolve
the whole problem?
1
(a) (4 points) Consider using the conditional gradient method to solve the problem
numerically. Suppose that the stepsize is 0.1 and the initial point is x0 = (0, 0).
Compute the next point x1 under this method. HINT: Draw the feasible set and
contour lines to help identify the next point.
(b) (4 points) Consider using the gradient projection method to solve the problem
numerically. Suppose that the stepsize before projection is 1, the stepsize after
projection is 0.1, and the initial point is x0 = (4, 0). Compute the next point x1
under this method. HINT: Draw the feasible set to help identify the next point.
Problem 4 (10 points total): Consider the following linear optimization problem.
minimize − x1 − 2x2
subject to x1 + x2 ≤ 4
x2 ≤ 3
x1 , x2 ≥ 0
(a) (2 points) Draw the feasible set for the given problem.
(b) (1 point) Count the number of extreme points in the feasible set.
(c) (2 points) Draw the contour lines for the cost values −2 and −4.
(e) (1 point) Specify the optimal cost associated with the optimal solution in part (d).
(f ) (2 points) Give two example cost functions such that there are infinitely many
optimal solutions in each case.
(a) (1 point) Is the given problem a convex optimization problem? You can simply
answer YES or NO; there is no need to justify.
(b) (6 points) Express the problem as a linear optimization problem in the standard
form. Note that the set of decision variables may have to change.
(c) (1 point) Count the number of decision variables in the standard-form problem in
part (b).
2
Problem 6 (12 points total): Consider the power allocation problem in which each of
the N subchannels is assigned to a distinct user, i.e. N different users. Instead of trying
to maximize the total data rate from all subcarriers, we try to maximize the rate of the
worst subcarrier, i.e. maximizing the minimum rate. It follows that the power allocation
problem becomes
pi
maximize min ln 1 +
i∈{1,...,N } ni
XN
subject to pi ≤ P
i=1
∀i ∈ {1, . . . , N }, pi ≥ 0
Note that p1 , . . . , pN are the decision variables. In addition, assume that n1 , . . . , nN > 0.
(b) (2 points) Write down the Lagrangian for the optimization problem in part (a).
(c) (8 points) Using the KKT conditions to conclude the following for an optimal power
allocation.
1. The subcarriers with nonzero transmit powers have the same data rate equal
to the minimum data rate.
2. All the available power P is allocated.