Abstract: The canonical duality theory is developed recently with a
one to one corresponding between a dual and a primal feasible solution.
Initially it is shown computationally efficient using the sufficient conditions
of the strong duality property for manyproblems such as the 0−1quadratic programming, the
multi-integer quadratic programming, and the
sum-of-quadratic-ratios problem. Moreover, we can find that all these models are or
can be transformed to quadratic programs with quadratic constraints.
In this talk, we will outline our recent work on the
modeling and applications of the canonical duality approaches to quadratic programs. Some
well-known problems can be modeled into the quadratic programs, for example, the max-cut, the 0-1
quadratic programming, the sum-of-quadratic-ratios programming, the non-convex quadratic
programming over linear constraints etc.
After a brief introduction of the canonical duality theory,
we provide with a canonical duality optimization model by only considering the positive definite
region. Analytic properties are given for the canonical duality function.
Aquadratic programming with a non-convex quadratic
constraint is a simple case of quadratic programs. With the application of the canonical duality
approaches, it is shown that it is polyno-mially solvable under duality Slater condition and the
global primal minimizer can be easily got. Acanonical duality iteration algorithm is provided for the
quadratic programming over linear constraints, which converges to a Karush-Kuhn-Tucker
point. Some optimality properties are discussed for the canonical duality approach.