Remark 11.

Although it may not be of particular seismic interest, it is surprising that the Acousmod2d algorithm can have large errors for values of grid points per median wavelength that are well beyond the $12.5$ that is used in seismic simulations. We recall that the model used for these calculations is a periodic box of $2,000$ feet on a side, nonreflecting interface conditions with a wave speed of $5,000$ ft/sec in the exterior region, and $20,000$ ft/sec in the interior region. The large amplitude errors are caused by the focusing of the wave and might be expected when caustics form.