Optimization of non-linear gravity models through generalized simulated annealing

In this paper we apply the generalized simulated annealing (GSA) approach to the inversion of gravity data for 2-D and 3-D density distributions. We consider a modeling process where the input is the vector of model parameters m (that can be density contrast, mass or some coordinates) and the output is described by the transformation h(m)=d, where d is the vector of data parameters, we generally have access in practical problems. If the vector d describes the observed actual output of the system, the problem is to “choose”, or estimate, the parameters mest in order to minimize, in some sense, in our case in the least-squares sense, the difference between the observed vector d and the prescribed output of the system h(mest). The tests with synthetic data show the promising application of GSA in gravity inversion. The results obtained suggest us that the GSA approach enables to find quickest optimization machines than the two conventional approaches (Boltzmann and Cauchy machines).