The paper presents a parallel algorithm for n-dimensional unconstrained minimization, based on a second order model of homogeneous form. It iterates a finite indirect procedure, which has been previously proposed for homogeneous functions having one minimum point. The outlined method is proved to be convergent for continuous and almost everywhere derivable objective functions, which may possess non stationary minima and non convex level sets. This procedure does not require line searches; moreover it contains many tasks which may be simultaneously performed, therefore it seems adapt to parallel computers.
