next up previous print clean
Next: Reducing the filter length Up: The Helmholtz equation Previous: Wave extrapolation

Propagating waves with the Wavemovie program

The following pseudo-code provides an algorithm for propagating waves into the Earth with the the new factorization of the wave equation.
Fourier transform input data over time-axis
Loop over frequency {
     Initialize wave at z=0
     Factor wave equation for this w/v
     Recursively divide input data by factor
     Fourier Transform back to time-domain
     Sum into output
}
Incorporating this code into the Wavemovie program Claerbout (1985) provides a laboratory for testing the new algorithm.

Figure [*] compares the results of the new extrapolation procedure with the conventional Crank-Nicolson solution to the 45$^\circ$ equation. The new approach has little dispersion since I use a rational approximation (the `one-sixth trick') to the Laplacian on the vertical and horizontal axes. In addition, the new factorization retains accuracy up to 90$^\circ$. The high dip, evanescent energy in the 45$^\circ$ movie, propagates correctly in the new approach.

 
vs45
Figure 3
Comparison of the 45$^\circ$ wave equation (left) with the helical factorization of the Helmholtz equation (right).
vs45
[*] view burn build edit restore

Figure [*] compares different value of the `one-sixth' parameter, $\beta$. For this application, the optimal value seems to be $\beta=1/12$.

 
sixth
sixth
Figure 4
Helmholtz equation factorization with different values for the `one-sixth' parameter, $\beta$. From left, $\beta =$ 0, 1/12, 1/8 and 1/6.
[*] view burn build edit restore

Figure [*] compares different finite-difference Laplacian operators. In all cases the finite-difference Laplacian was given by the linear sum of 5-point filters,  
 \begin{displaymath}
(1-\gamma) \,
\left(\begin{array}
{ccc}
& 1 & \\ 1 & -4 & 1 ...
 ...
{ccc}
1/2 & & 1/2 \\  & -2 & \\ 1/2 & & 1/2\end{array}\right),\end{displaymath} (33)
where $\gamma$ is an adjustable parameter between 0 and 1. Best results were obtained with $\gamma = 2/3$. The impulse response with $\gamma=0$ only contains energy on every second grid point, since the rotated filter only propagates energy diagonally: as in the game of a chess, if a bishop starts on a white square, it always stays on white.

 
laplac
laplac
Figure 5
Helmholtz equation factorization with different $3 \times 3$ finite-difference representations of the Laplacian. From left, $\gamma =$ 0, 1/2, 2/3 and 1.
[*] view burn build edit restore


next up previous print clean
Next: Reducing the filter length Up: The Helmholtz equation Previous: Wave extrapolation
Stanford Exploration Project
5/27/2001