Instructor:
Professor Henry Wolkowicz,
MC 6065, ext. 5589, hwolkowi@orion.math.uwaterloo.ca
Office Hours:
-
Henry Wolkowicz, MC6065: tentative :M 12:30-1:30, R 3-4
Lectures:
-
1:00-2:30 TR, MC 6005 (Perhaps a video link will be used in Chem 2-079,
T 3:30-6:30.)
Texts:
References:
Term Work:
will consist of 5 homework problems .
Final Take-home Exam:
Marking Scheme:
-
Homework.......................... 50%
-
Final Take-home Exam.......... 50%
Semidefinite programming
(SDP) refers to mathematical programs of the type
minimize f(x) subject to g(x) <= 0, x in O,
where f,g are matrix valued functions and A <= B refers to the
Loewner partial order, i.e. the symmetric matrix B-A is positive
semidefinite.
Though these problems have been studied before, it is only recently that
the combination of important applications with implementable algorithms
has resulted in intensive research activity. This area is proving to be
of major importance because of the many applications; in addition, the
high mathematical elegance is attracting many researchers.
This course will provide a thorough treatment of semidefinite
programming.
-
THEORY
-
Convex analysis on symmetric matrices
-
Geometry of SDP
-
First and second order optimality conditions; duality
-
Error analysis and parametric programming
-
ALGORITHMS
-
Potential reduction methods
-
path-following methods
-
implementation
-
APPLICATIONS
-
Combinatorial Optimization problems (e.g.
quadratic assignment, graph partitioning, max-cut, general integer)
-
eigenvalue optimization
-
matrix completion problems
-
nonconvex quadratic optimization
Possible additional speakers: Romesh Saigal, Univ.
of Michigan; Tom Luo and Jose Sturm, McMaster Univ.; Lieven
Vandenberghe, UCLA; Gabor Pataki, DIMACS.
Tentative Time: Tuesday and Thursday 1-2:30 PM.
This course is supported partially by the
Fields Institute