Abstract of proposed talk at the 1st Nordic Optimization Symposium Lennart Frimannslund Department of Informatics University of Bergen, Norway lennart@ii.uib.no Title: Separability and sufficient decrease in generating set search methods Abstract: Generating set search (GSS) methods are often one of the few applicable alternatives for optimization of functions which are computed through simulations on computers, since these functions often are subject to numerical noise. It is well known that these methods are slow in their most basic form because of what one can call zig-zagging, a phenomenon caused by the angle between search directions and the negative gradient being large. We present an enhanced version of a previously presented GSS method, and elaborate on what challenges arise when one wishes to exploit partial separability, the benefits of doing so, and issues when it comes to ensuring provable convergence of such a method.