Spectral Modelling Lanczos on Magnetotelluric

Abstract

A recurring problem in solution modeling approximate Maxwell's equations for fields that propagates in a diffusion regime is the cost computational (Boyse et al., 1992; Liu & Chew, 1990; Madden and Mackie, 1989; Oristaglio & Hohmann, 1984; Weiland, 1986). In the frequency domain, modeling is generated independently for each frequency, source and receiver position. Furthermore, every frequency, it is necessary to factorize the global matrix of the problem, which is large and sparse. One solution is to use accelerators convergence. Various techniques are used to Convergence contribution (Saad & Schultz, 1986; Manteuffel et al., 2005; Van der Vorst, 1992; Freund, 1993). In this work, with the aim of reducing the cost computational, is applied to data modeling magnetotelluric (MT) method of the invention Lanczos spectral (LSDM) (Druskin & Knizhnerman, 1994a, 1994b; Zunoubi et al., 1998, 1999) as convergence accelerator. Through it, it is possible estimate response functions for various frequencies from of estimates for a single frequency, avoiding thus, the reproduction of the modeling for each frequency. These estimates are discovered in the subspaces of Krylov through the Lanczos iteration method (Lanczos, 1950)