Efficient Love Wave Modelling via Sobolev Gradient Steepest Descent
McMechan, George A.
MetadataShow full item record
A new method for finding solutions to ordinary differential equation boundary value problems is introduced, in which Sobolev gradient steepest descent is used to determine eigenfunctions and eigenvalues simultaneously in an iterative scheme. The technique is then applied to the 1-D Love wave problem. The algorithm has several advantages when computing dispersion curves. It avoids the problem of mode skipping, and can handle arbitrary Earth structure profiles in depth. For a given frequency range, computation times scale approximately as the square root of the number of frequencies, and the computation of dispersion curves can be implemented in a fully parallel manner over the modes involved. The steepest descent solutions are within a fraction of a per cent of the analytic solutions for the first 25 modes for a two-layer model. Since all corresponding eigenfunctions are computed along with the dispersion curves, the impact on group and phase velocity of the displacement behaviour with depth is thoroughly examined. The dispersion curves are used to compute synthetic Love wave seismograms that include many higher order modes. An example includes addition of attenuation to a model with a low-velocity zone, with values as low as Q = 20. Finally, a confirming comparison is made with a layer matrix method on the upper 700 km of a whole Earth model.