On the SCD semismooth* Newton method for solving generalized equations

H. Gfrerer, J. Outrata

In this talk we present a novel Newton-type method for solving inclusions, e.g., the first-order optimality condition in nonsmooth programming, that zero belongs to the subdifferential of the objective. The proposed method relies on a new concept of generalized differentiation for set-valued mappings, the so-called Subspace Containing Derivatives (SCD), which is rather simple to compute. Together with the semismooth* property we can derive a superlinearly convergent algorithm. Numerical experiments are also presented.

Keywords: Newton-type methods, generalized differentiation, variational analysis

GT13.OPTCONT3 Invited Session
November 7, 2023  3:30 PM
HC2: Canónigos Room 2

M. J. Cánovas Cánovas, J. Camacho Moro, H. Gfrerer, J. Parra López

J. Rückmann, D. Hernández Escobar

J. Parra López, J. Camacho Moro, M. J. Cánovas Cánovas, J. E. Martínez Legaz