Randomized Newton-Raphson and Animal Search,
Abstract
Adding systematic noise to the step term of the Newton-Raphson (NR) root finding algorithm permits expected q-linear convergence and convergence almost surely to the root for a larger class of functions and larger starting sets than those for which NR converges deterministically. These results have application not only to a wide range of optimization problems but also to understanding the behavioral repertory of animals undertaking pheromone induced search. It is shown that the search reduces in many cases to finding the root of a function of two or three dimensions. In cases (as in the search of the gypsy moth for its mate) where the animal cannot simply travel in the direction of increasing signal (scent) randomized NR gives insight into the search behavior required to discover the signal source.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1992
- Accession Number
- ADP007158
Entities
People
- A. Levin
- J. Liukkonen
Organizations
- Tulane University of Louisiana