FURTHER SPEEDUP

In theory it should also be possible to use sparser sampling in midpoint at low frequencies. Since this would only affect the relatively inexpensive portion of the downward continuation, I would expect at most only a $10-20\%$ speed up. Another method is to do a better job choosing appropriate velocities, thereby reducing the number of velocities needed for accurate downward continuation. Clapp (2002) discussed one method to accomplish this.