Next: DISCUSSION
Up: Mo: Numerical analysis
Previous: The Crank-Nicolson method
Since the numerical solution is an approximation of the exact solution,
it does not exactly satisfy the continuous partial differential equation at
hand, but it satisfies a modified equation.
In this case, I use the Crank-Nicolson method
to solve the diffraction equation
|  |
(32) |
| ![\begin{displaymath}
{ {P^{(n+1)}-P^{(n)}}\over {\Delta \tau}} = {\alpha \over 2}...
...partial x^2} + {{\partial}^2 P^{(n)}\over \partial x^2} \right]\end{displaymath}](img49.gif) |
(33) |
| ![\begin{displaymath}
{ {P^{(n+1)}_{j}-P^{(n)}_{j}}\over {\Delta \tau}} = {\alpha ...
...)}_{j+1}-2P^{(n)}_{j}+P^{(n)}_{j-1}}\over {\Delta x^2}} \right]\end{displaymath}](img50.gif) |
(34) |
By Taylor series expansion of every term about Pnj, I obtain the
modified wave equation that the numerical solution actually satisfies:
|  |
(35) |
| ![\begin{displaymath}
\epsilon = { {\alpha \Delta x^2}\over 12} {{\partial}^4 P\ov...
...2 {\alpha}^2 \right] {{\partial}^6 P\over \partial x^6}+ \ldots\end{displaymath}](img52.gif) |
(36) |
The error term shows that the trace spacing interval
plays a larger role than the depth step size
of downward continuation in determining the accuracy of the numerical solution.
Next: DISCUSSION
Up: Mo: Numerical analysis
Previous: The Crank-Nicolson method
Stanford Exploration Project
11/17/1997