In this study, a new finite-difference technique is designed to reduce
the number of grid points needed in frequency-space domain modeling.
The new algorithm uses optimal nine-point operators for the approximat
ion of the Laplacian and the mass acceleration terms. The coefficients
can be found by using the steepest descent method so that the best no
rmalized phase curves can be obtained. This method reduces the number
of grid points per wavelength to 4 or less, with consequent reductions
of computer memory and CPU time that are factors of tens less than th
ose involved in the conventional second order approximation formula wh
en a band type solver is used on a scalar machine.