DIMITRI KOMATITSCH
AND
JEROEN TROMP
Department of Earth and Planetary Sciences
Hoffman Building Room 128
Harvard University
Oxford Street, Cambridge, MA 02138, USA
http://www.seismology.harvard.edu/~komatits
komatits@seismology.harvard.edu
Abstract:
In recent years, the use of numerical modeling to understand
the propagation of waves in complex 3-D media has attracted
significant interest. With the increase in speed and memory capacity
of available computers, several techniques have proven to be efficient
in this respect, for instance finite difference,
global pseudo-spectral, and boundary integral methods.
Nonetheless, several aspects have appeared to be difficult from the
point of view of numerical modeling: modeling of media with steep
topography, coupling between acoustic and elastic structures,
modeling of general anisotropy, viscoelasticity/attenuation,
and modeling of real-size 3-D structures (e.g. sedimentary basins)
over a broad frequency range.
To be able to deal with all these problems efficiently,
we use the so-called spectral element method (SEM). This method has been
successfully applied to both acoustic and elastic wave
propagation problems, including a recent extension
to 3-D models. We assess the efficiency of the method, recall its main
advantages,
and solve several problems of high practical interest: modeling of
an acoustic/elastic interface (for instance, to study the propagation
of a Stoneley wave at a water/sea bottom interface with topography),
modeling of any anisotropic 3D material with up to 21 elastic
coefficients (for instance, to study the behavior
of a crystal in the ultrasonic frequency range, or regions with fractured
rocks), and modeling of viscoelasticity. The use of non-structured grids
on a parallel computer, based upon the Message Passing Interface (MPI),
allows us to model wave propagation in real-size 3-D structures
with grids consisting of several tens of millions of points,
and to obtain a very high parallel efficiency.