Institute for Operations Research

ETH Zurich HG G21-23

Sebastiano Columbano [email protected]
Vania Dos Santos Eleuterio [email protected]

Spring Term 2008

Convex Optimization — Assignment 5

http://www.ifor.math.ethz.ch/teaching/lectures/convexopt/Uebungen

Exercise 1: Saddle point of Lagrangian

Let (P ) be an optimization problem in standard form and (D) the Lagrangian dual problem. Denote
f ∗ and d∗ the optimal value of (P ) and (D) respectively. Prove the following theorem

(x̄, λ∗ , µ∗ ) is a saddle point of the Lagrangian L ⇐⇒ (i) x̄ solves (P )

(ii) (λ∗ , µ∗ ) solves (D)
(iii) f ∗ = d∗

Exercise 2: Duality Example ([BV04], Ex. 5.1)

Consider the optimization problem

min f0 (x) := x2 + 1
s.t. f1 (x) := (x − 2)(x − 4) ≤ 0

where x ∈ R.
(a) Analysis of primal problem. Give the feasible set, the optimal value, and the optimal solution.
(b) Lagrangian and the dual function. Plot f0 (x). On the same plot, show the feasible set, optimal
point and value, and plot the Lagrangian L(x, λ) versus x for a few positive values of λ. Verify
the lower bound property (p⋆ ≥ inf x L(x, λ) ∀ λ ≥ 0). Derive and sketch the dual function g.
(c) Lagrange dual problem. State the dual problem, and verify that it is a concave maximization
problem. Find the dual optimal value and dual optimal solution λ⋆ . Does strong duality hold?
(d) Sensitivity Analysis. Let p⋆ (u) denote the optimal value of the problem

min f0 (x) := x2 + 1
s.t. f1 (x) := (x − 2)(x − 4) ≤ u

as a function of the parameter u. Plot p⋆ (u). Verify that dp⋆ /du = −λ⋆ .

Exercise 3: A convex problem in which strong duality fails ([BV04], Ex. 5.21)
Consider the optimization problem

minimize e−x
subject to x2 /y ≤ 0

with variables x and y, and domain D = {(x, y)|y > 0}.

(a) Verify that this is a convex optimization problem and find the optimal value.
(b) Give the Lagrange dual problem, and find the optimal solution λ∗ and optimal value d∗ of the
dual problem. What is the optimal duality gap?
(c) Does Slater’s condition hold for this problem?

1
(d) What is the optimal value p∗ (u) of the perturbed problem

minimize e−x
subject to x2 /y ≤ u

as a function of u? Verify that the global sensitivity inequality

p∗ (u) ≥ p∗ (0) − λ∗ u

does not hold.

Exercise 4: Geometric Program and its Dual

Consider the following geometric program in convex form.
K0
!
X T
minimize log ea0k x+b0k
k=1
Ki
!
aT
X
subject to log e ik x+bik ≤ 0, i = 1, · · · , m
k=1

a) Write the conjugate function of the function

f i : RK i −→ R
Ki
!
X
yk
y 7−→ log e
k=1

b) Using the conjugate function computed previously, write the dual of the geometric program.

Exercise 5: SDP Relaxation of Two-Way Partitioning Problem ([BV04], Ex. 5.39)

We consider the two-way partitioning problem described in [BV04] on page 219 (5.7),

minimize xT W x (1)
subject to x2i = 1, i = 1, · · · , n

with variable x ∈ Rn . The Langrange dual of this non convex problem is given by the SDP

maximize −1T ν (2)

subject to W + diag(ν)  0

with variable ν ∈ Rn .
(a) Two-way partitioning problem in matrix form. Show that the two-way partitioning problem ca be
cast as

minimize tr(W X)
subject to X 0
rankX = 1
Xii = 1, i = 1, · · · , n

with variable X ∈ Sn . Hint: show that if X is feasible, then it has the form X = xxT , where
x ∈ Rn satisfies xi ∈ {−1, 1} and vice versa.

2
(b) SDP relaxation of the two-way partitioning problem. Using the formulation in part (a) we can
form the relaxation

minimize tr(W X) (3)

subject to X 0
Xii = 1, i = 1, · · · , n

with variable X ∈ Sn . Explain why its optimal value gives a lower bound on the optimal value of
the two-way partitioning problem 1. What can you say if an optimal point X ∗ for this SDP has
rank one?
(c) We now have two SDP’s that give a lower bound on the optimal value of the two-way partitioning
problem 1: the SDP relaxation 3 and the Langrange dual 2. What is the relation between the two
SDP’s? What can you say about the lower bounds found by them? Hint: relate the two SDP’s
via duality.
(d) Using any SDP solver, for example SPDT3 for Matlab, compute the different lower bounds for
a random instance of the two-way partitioning problem. For example generate the matrix W as
follows in matlab:
randn(’state’,0); n = 10; W = randn(n); W = 0.5*(W + W’);,

References
[BV04] S. Boyd, L. Vandenberghe. Convex Optimization. Cambridge University Press, 2004.

Please hand in your assignment no later than Tuesday, May 20th 2008, 13:00 at HG G21 (“Convex Optimization” tray) or
during the class.
Certificate condition: 50% of the total number of exercises have to be solved (each assignment will contain approx. 4-6
exercises).