The primal-dual active set strategy as a semi-smooth Newton method
 
 
Description:  We focus on the primal-dual active set strategy, a versatile algorithmic tool in solving constrained optimal control problems, obstacle problems or, more generally, (linear) complementarity problems in function space settings. After reviewing its motivation by Moreau-Yosida type approximations, discussing some of its main properties and conditional global convergence results, we establish the equivalence of the primal dual active set strategy to semi-smooth Newton methods in function spaces. This latter aspect relies on a suitable generalization of the differential calculus (i.e. slant differentiability) for non-smooth operators to infinite dimensions and offers a different view on local convergence properties, like locally superlinear convergence rates, well known from semi-smooth Newton methods in finite dimensions. Under appropriate growth conditions one is able to establish ``convergence-with-a-rate'' results. Also, the connection to Newton type methods in function spaces allows to prove the important property of mesh independence of semi-smooth Newton methods for operator equations in general, and the primal-dual active set strategy in particular. Finally, numerical results shall emphasize our theoretical findings.
Area(s):
Date:  2002-04-18
Start Time:   15:00
Speaker:  Michael Hintermuller (University of Graz, Áustria)
Place:  Room 5.5
Research Groups: -Numerical Analysis and Optimization
See more:   <Main>  
 
© Centre for Mathematics, University of Coimbra, funded by
Science and Technology Foundation
Powered by: rdOnWeb v1.4 | technical support