Comparison between Neighborhood and Genetic Algorithms on two Analytical Objective Functions and on a 2.5D Synthetic Seismic Inverse Problem

A geophysical inverse problem consists in obtaining the earth model for which the predicted data best fit the observed one. The problem is often non-linear and can be solved using a local linearization method (such as Gauss-Newton, steepest descent or conjugate gradient) or using a global optimization method (such as Grid Search, Simulated Annealing, Genetic Algorithms, Particle Swarm and Neighborhood Algorithm). In this work we compared and evaluated the efficiency and the limits of methods varying the dimensions of the model space. We first tested these methods on two analytical objective functions, a multidimensional convex parabola and a more complex egg-box functional. Lastly we performed an acoustic full waveform inversion considering a small and smoothed portion of the Marmousi model