EE5121: Convex Optimization

Assignment 6

1. Consider the problem

maxi=1,...,m aTi x + bi
minimize
min j=1,...,p cTj x + d j
subject to Fx ≤ g,

where x ∈ Rn . Assume that the constraint set is nonempty and bounded and the problem has a finite
optimal value. Convert this problem to an LP by introducing new variables.
Hint: Consider the problem

minimize max aTi y + bit (1)

i=1,...,m

subject to min cTj y + d j t ≥ 1

j=1,...,p

Fy ≤ gt, t ≥ 0,

Show that this problem is equivalent to the problem in question. Then, write the epigraph form of
this problem to get an LP.
2. Consider the LP

minimize cT x
subject to Ax = b, x ≥ 0,

where A ∈ Rm×n and rank of A is r. Assume that the problem is solvable. Prove that there exists an
optimal point x∗ with at most r nonzero entries.
3. Prove that x∗ = (1, 1/2, −1) is optimal for the optimization problem
1 >
minimize x Px + q> x + r
2
subject to − 1 ≤ xi ≤ 1, 1 = 1, 2, 3,

where
   
13 12 −2 −22
P =  12 17 6 , q = −14.5, r = 1.
−2 6 12 13
4. Consider the problem
kAx − bk1
minimize
c> x + d
subject to kxk∞ ≤ 1,

where A ∈ Rm×n , b ∈ Rm , c ∈ Rn , and d ∈ R. We assume that d > kck1 , which implies that
c> x + d > 0 for all feasible x.
(a) Show that this is a quasiconvex optimization problem.
(b) Show that it is equivalent to the convex optimization problem

minimize kAy − btk1

subject to kyk∞ ≤ 1
c> y + dt = 1,

with variables y ∈ Rn , t ∈ R.

1
5. Network flow problem: Consider a network of n nodes, with directed links connecting each pair of
nodes. The variables in the problem are the flows on each link: xi j will denote the flow from node i
to node j. The cost of the flow along the link from node i to node j is given by ci j xi j , where ci j are
given constants. The total cost across the network is
n
C= ∑ ci j xi j
i, j=1

Each link flow xi j is also subject to a given lower bound li j (usually assumed to be non-negative) and
an upper bound ui j .
The external supply at node i is given by bi , where bi > 0 means an external flow enters the network
at node i, and bi < 0 means that at node i, an amount |bi | flows out of the network. We assume that
1T b = 0, i.e the total external supply equals total external demand. At each node we have conservation
of flow: the total flow into node i along links and the external supply, minus the total flow out along
the links, equals zero.
The problem is to minimize the total cost of flow through the network, subject to the constraints
described above. Formulate this problem as an LP.
6. Linear separation of two sets of ellipsoids. Suppose we are given K + L ellipsoids
εi = {Pi u + qi | ||u||2 ≤ 1}, i = 1, ..., K + L
where Pi ∈ Sn . We are interested in finding a hyperplane that strictly separates ε1 , ..., εK from εK+1 , ..., εK+L ,
i.e we want to compute a ∈ Rn , b ∈ R such that
aT x + b > 0 for x ∈ ε1 ∪ · · · ∪ εK ,
aT x + b < 0 for x ∈ εK+1 ∪ · · · ∪ εK+L
or prove that no such hyperplane exists. Express this problem as an SOCP feasibility problem.
7. LPs, QPs, QCQPs, and SOCPs as SDPs. Express the following problems as SDPs.

(a) The LP.

(b) The QP, the QCQP and the SOCP.
Hint: Suppose A ∈ Sr++ , C ∈ Ss , and B ∈ Rr×s . Then
 
A B
≥ 0 ⇔ C − BT A−1 B ≥ 0
BT C

8. Eigenvalue optimization via SDP. Suppose A : Rn → Sm is affine, i.e

A(x) = A0 + x1 A1 + · · · + xn An
where Ai ∈ Sm . Let λ1 (x) ≥ λ2 (x) ≥ · · · ≥ λm (x) denote the eigenvalues of A(x). Show how to pose
the following problems as SDPs.
(a) Minimize the maximum eigenvalue λ1 (x).
(b) Minimize the spread of the eigenvalues, λ1 (x) − λm (x).
(c) Minimize the condition number of A(x), subject to A(x) > 0. The condition number is defined
as κ(A(x)) = λ1 (x)/λm (x), with domain {x | A(x) > 0}. You may assume that A(x) > 0 for
atleast one x.
Hint: You need to minimize λ /γ, subject to 0 < γI ≤ A(x) ≤ λ I. Change variables to y =
x/γ, t = λ /γ, s = 1/γ.
(d) Minimize the sum of absolute values of the eigenvalues |λ1 (x)| + · · · + |λm (x)|.
Hint: Express A(x) as A(x) = A+ − A− , where A+ ≥ 0, A− ≥ 0.
9. Bi-criterion optimization. Figure 1 shows the optimal trade-off curve and the set of achievable values
for the bi-criterion optimization problem
minimize (w.r.t R2+ ) (||Ax − b||2 , ||x||22 )
for some A ∈ R100×10 , b ∈ R100 . Answer the following questions using information from the plot.
We denote by xls the solution of the least-squares problem
minimize ||Ax − b||22

2
Figure 1: Optimal trade-off curve for a regularized least-squares problem. The shaded set is the set of
achievable values. The optimal trade-off curve shown darker, is the lower left part of boundary.

(a) What is ||xls ||2 ?

(b) What is ||Axls − b||2 ?
(c) What is ||b||2 ?
(d) Give the optimal value of the problem

minimize ||Ax − b||22

subject to ||x||22 = 1

(e) Give the optimal value of the problem

minimize ||Ax − b||22

subject to ||x||22 ≤ 1

(f) Give the optimal value of the problem

minimize ||Ax − b||22 + ||x||22

(g) What is the rank of A?

10. [Matrix norm minimization] Let A(x) = A0 + A1 x1 + · · · + An xn , where Ai ∈ R p×q . Consider the
unconstrained problem

minimize kA(x)k2 ,

where kA(x)k2 is the spectral norm (maximum singular value) and x ∈ Rn is the optimization variable.
Pose this problem as an SDP.
11. Consider the following minimization problem (with respect to the 1.5th norm).

minimize kAx − bk 3 ,
2

with optimization variable x ∈ Rd , and given A ∈ Rm×d and b ∈ Rm . Assume m > d and that A is full
rank.
(a) Obtain the optimality conditions for this problem.
(b) Formulate this problem as an SDP.

3
12. [Optimal power assignment in wireless communications] Consider the scenario portrayed in Fig. 2,
wherein K transmitters are communicating with K receivers. Specifically, transmitter i wishes to com-
municate with receiver i by emitting signals with power pi . The received signal power at receiver i
is Gii pi , where Gii is the “path gain” from transmitter i to receiver i. Further, receiver i also receives
power ∑ j6=i Gi j p j , which is termed as “interference” for receiver i. Here, Gi j is the “path gain” from
transmitter j to receiver i. Additionally, the noise power at receiver i is σi2 . Thus, we define the
Signal-to-Noise-plus-Interference-Ratio (SINR) as
Gii pi
γi = .
∑ j6=i Gi j p j + σi2

Note that the SINR at a given receiver measures the “quality” of communication; higher the SINR,
more reliable is the communication. Transmitter i has an additional constraint that it cannot transmit
with power greater than Pi .

(a) First, consider the problem of minimizing the total power subject to the SINR at each receiver
being no less than a threshold SINR γ0 . Pose this problem as a linear program.
(b) Next, consider the problem of maximizing the minimum SINR. Denote the resulting SINR by
γ ∗ . Pose this problem as a quasiconvex program.
Hint: Consider the quantity 1/γ ∗ and relate it to the max-min problem posed for γ ∗ .
(c) Finally, pose the problem of maximizing the minimum SINR as a geometric program.
Hint: Write the epigraph form of the quasiconvex program derived in part (b).

Figure 2: Figure describing the scenario for the optimal power assignment problem.