Last Name (Print) : First Name:: May 10, 2001 7:15-9:15 A.m., 1221 Computer Sciences & Statistics
Last Name (Print) : First Name:: May 10, 2001 7:15-9:15 A.m., 1221 Computer Sciences & Statistics
Answer all questions: 1, 2 & 3. If any question is missing from your sheets,
please inform the instructor.
2. Question 2:—————
3. Question 3:—————
4. Total:——————–
1. (10 points) Let (xi+1 , ui+1 ) be a KKT point of the IQP iteration:
1
min{∇f (xi )(x − xi ) + (x − xi )Gi (x − xi ) g(xi ) + ∇g(xi )(x − xi ) ≤ 0},
x 2
for the nonlinear program
min {f (x) | g(x) ≤ 0}, (∗)
x
Justification:
Problem 1 Sheet
2. (10 points) Consider the quadratic program
1
min −x21 + x1 x2 − x22 − x1
x1 ,x2 2
s.t. − 1 ≤ x1 ≤ 1, −1 ≤ x2 ≤ 1
h i
1
Start with x = 1
. Compute x1 using the Gradient Projection Algorithm:
x1 = P (x − λ∇f (x)), λ ≥ 0.
Justification:
Problem 2 Sheet
3. (10 points) Let f : Rn → R have a Lipschitz continuous gradient with Lipschitz constant
K > 0. If ∇f (xi ) 6= 0 for some xi in Rn , give an upper bound to minf (x), in terms
x
of xi and K, that is strictly less than f (xi ).
Hint: Use the following inequality, a consequence of the Quadratic Bound Lemma:
K
min f (x) ≤ min (f (xi ) + ∇f (xi )(x − xi ) + kx − xi k2 )
x x 2
Answer:
min
x
f (x) ≤ f (xi ) − . . .
Justification:
Problem 3 Sheet
Extra Sheet