Numerical precision in 3D prestack Kirchhoff migration
Stewart A. Levin · 2004
3D prestack Kirchhoff migration will routinely sum together millions of floating point numbers to produce each individual migrated output sample. Such large summations are expected in general to lose at 3 to 4 significant digits of accuracy, which is at least half of the precision of an individual floating point sample. Clearly double precision arithmetic ought to be used to avoid such catastrophic loss of accuracy. Surprisingly, this study finds that single precision is more than adequate for Kirchhoff migration and even a loss of 2 significant figures is unlikely.