A Conceptual Sequential Linear Programming Algorithm with Multidimensional Search
Within the field of nonlinear programming, a frequently employed solution
principle is to alternate between the solution of an approximate problem
and a line search with respect to a merit function, that measures the
degree of non-optimality of any tentative solution. The Sequential Linear
Programming (SLP) approach is a method that exploits this principle.
We use the generic column generation scheme to develop a novel SLP type
method for constrained nonlinear optimization. It is based on linear
approximation of both the primal and the dual spaces, which yields a
method which in the primal space combines column and constraints generation.
In the presented algorithm we do not need to find rules to control the move
limits and we can also skip the merit function, but steel get a convergence
to a point that satisfy a Karush-Kuhn-Tucker conditions in the non-convex case.
I am going to visualize the proposed algorithm with a simple example and
present some computational experiments on frequently used nonlinear
Senast uppdaterad: 2014-10-14