2.098/6.255/15.093 Optimization Methods Practice True/False Questions

2.098/6.255/15.
093 Optimization Methods
Practice True/False Questions
December 11, 2009
Part I
For each one of the statements below, state whether it is true or false. Include a 1-3 line
supporting sentence or drawing, enough to convince us that you are not guessing the
answer, but not a comprehensive, rigorous formal justification.
a) There exist nonempty and bounded polyhedra of the form {x ∈ Rn : Ax ≤ b}, in

which every basic solution is also a vertex.
b) Consider the problem of minimizing c′ x subject to Ax ≤ b. If we increase some

component of b, then the optimal cost cannot increase.
c) A feasible linear program of minimizing c′ x subject to Ax ≥ b has finite cost if and

only if c is a nonnegative combination of the rows of A.
For the next three statements, assume that we are dealing with a linear programming
(minimization) problem in standard form, with the matrix A having full row rank.
d) If a tie occurs while choosing the pivot row during the primal simplex method,
then the next basic feasible solution will be degenerate.
e) If the optimal cost is −∞, then the right hand side vector b can be adjusted in
some way to make the optimal cost finite.
f) The dual of the Phase I problem will never have an infinite optimal cost.
Part II
g) On very degenerate LP problems, the simplex method performs better than interior
point methods.
h) Given a local optimum x̄ for a nonlinear optimization problem it always satisfies

the Kuhn-Tucker conditions when the gradients of the tight constraints and the
gradients of the equality constraints at the point x̄ are linearly independent.
i) In a linear optimization problem with multiple solutions, the primal-dual barrier

algorithm always finds an optimal basic feasible solution.
j) The larger the condition number of the Hessian, the slower the convergence of
Newton’s method.
k) The convergence of the steepest descent method depends highly on the starting
point.
l) For the problem min x s.t. y ≤ x3 , y ≥ 0, the gradients of the constraints satisfy
the constraint qualification condition (CQC).
m) For a constrained optimization problem, if a point x̄ is feasible, satisfies the CQC,

and satisfies the KKT conditions, then it is a local minimum.
Solutions
Part I
a) True. Consider P = {x ∈ R : −x ≤ 0, x ≤ 1}. Then x = 0 and x = 1 are the only
basic solutions, but they are also vertices of P .
b) True. Increasing a component of b gives a feasible set which is no smaller, so the
optimal cost cannot increase. (A more complicated way to see this is to take the
dual and invoke a shadow price argument using the sign of the dual variables).
c) True. The dual problem is written as
maximize b′ p
subject to A′ p =c
p ≥ 0.
If c is a nonnegative combination of the rows of A, then the dual is feasible. Since

we are given that the primal is feasible, it follows that both problems have equal
and bounded optimal costs. Conversely, if the primal has a bounded optimal cost,
the dual must also. In particular, the dual must be feasible, which implies that c
is a nonnegative combination of the rows of A.
d) True. Let θ∗ be the value of the min-ratio test. Then all basic variables xj which
x
have this value of θj = ujj , where uj is the j th component of the basic direction
become xj − θ∗ uj = xj − θj uj = 0. Since there are more than one of these such
variables, at least one stays in the basis, which implies that the resulting BFS is
degenerate.
e) False. If an LP is unbounded, its dual must be infeasible. But dual feasibility
depends in no way on the vector b.
f) True. If the dual of the phase I formulation has an infinite optimal cost, then the
phase I problem must be infeasible. This cannot be true, since (x, y) = (0, b) is
always used as a starting BFS for the phase I problem.
Part II
g) False. Barrier interior-point methods are unaffected by degeneracy; see BT p. 439.
h) True. KKT conditions hold for a local minimum under the linearly independent
constraint qualification condition (CQC).
i) False. Barrier interior-point methods find an interior point of the face of optimal
solutions. See BT p. 423.
j) False. For example, Newton’s method converges in one iteration for quadratic
problems, regardless of condition number.
3
k) True. Recall the zig-zag phenomenon shown in lecture.
l) False. The only local minimum is (0, 0). At this point, the gradient of the con
straints are linearly dependent.
m) False. Since the KKT are essentially first-order conditions, they can only “see” the
linear part of the objective and constraints, and are thus unable to ensure even
local optimality (unless further conditions, such as convexity, are imposed). As
a simple example, consider the objective function f (x) := −x2 , and the feasible
set defined by g(x) := x ≤ 0. Clearly, the origin x = 0 is a KKT point (since
∇f (x) + u1 ∇g(x) = 0 + 0 · 1 = 0), but it is a local maximum (not a minimum).
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.

2.098/6.255/15.093 Optimization Methods Practice True/False Questions

Uploaded by

Copyright:

Available Formats

2.098/6.255/15.093 Optimization Methods Practice True/False Questions

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2.098/6.255/15.093 Optimization Methods Practice True/False Questions

Uploaded by

Copyright:

Available Formats

2.098/6.255/15.

093 Optimization Methods

Practice True/False Questions

December 11, 2009

a) There exist nonempty and bounded polyhedra of the form {x ∈ Rn : Ax ≤ b}, in

b) Consider the problem of minimizing c′ x subject to Ax ≤ b. If we increase some

c) A feasible linear program of minimizing c′ x subject to Ax ≥ b has ﬁnite cost if and

h) Given a local optimum x̄ for a nonlinear optimization problem it always satisﬁes

i) In a linear optimization problem with multiple solutions, the primal-dual barrier

m) For a constrained optimization problem, if a point x̄ is feasible, satisﬁes the CQC,

If c is a nonnegative combination of the rows of A, then the dual is feasible. Since

15.093J / 6.255J Optimization Methods

You might also like