Progress in determining the details of the global 3-D seismic velocity stru
cture requires the ability to accurately model seismic wave propagation (e.
g., travel times, waveforms, etc.) through heterogeneous 3-D Earth models.
While for spherically symmetric models (quasi-) analytical solutions are av
ailable for the verification of numerical algorithms, this is not the case
for general heterogeneous models. It is therefore desirable to establish gl
obal 3-D test models and verified reference seismograms, which allow us to
assess the accuracy of numerical algorithms quantitatively. Prior to a work
shop held on this issue at the 1997 IASPEI Meeting, a 3-D test model was ha
nded out to various groups and long-period synthetic seismograms were retur
ned. This workshop was the initiation of the Comparison of global SYnthetic
seismogram techniques (COSY) Project, which aims at establishing a WWW pag
e (http://www.geophysik.uni-muenchen.de/COSY), where the test models and se
ismograms as well as some of the algorithms can be accessed. In this paper,
we study the accuracy of and compare solutions from different numerical me
thods for a spherically symmetric model and the 3-D test model. The algorit
hms compared use the normal-mode method, the Direct Solution Method (DSM),
a direct evaluation of the Greens function for spherically symmetric media
(GEMINI), and the finite-difference (FD) method. Our 3-D test model is a pe
rturbation to the spherically symmetric background model (PREM) based on a
(scaled) temperature field from numerical modeling of mantle convection. Th
e model displays many features in common with recent seismic tomographic im
ages. We suggest that in addition to (future) 3-D reference Earth models, v
erified reference synthetic seismograms should be established for use by th
e seismological community. (C) 2000 Elsevier Science B.V. All rights reserv
ed.