next up previous [pdf]

Next: Impulse Response Tests Up: 3D Implicit Finite-difference Propagation Previous: 3D Implicit Finite-difference Propagation

Extrapolation Algorithm

Using the 4th-order approximation is equivalent to solving a cascade of partial differential equations (, )
$\displaystyle \frac{\partial}{\partial \alpha} U =$   $\displaystyle {\rm i} \omega s U,$  
$\displaystyle \frac{\partial}{\partial \alpha} U =$   $\displaystyle {\rm i} \omega s
\left[
\frac{ \frac{a_1B^2}{\omega^2s_A^2} \frac...
...+\frac{b_1C^2}{\omega^2s_A^2} \frac{\partial^2}{\partial \gamma^2}
}
\right] U,$ (A-13)
$\displaystyle \frac{\partial}{\partial \alpha} U =$   $\displaystyle {\rm i} \omega s
\left[
\frac{ \frac{a_2 B^2}{\omega^2s_A^2} \fra...
...1+\frac{b_2C^2}{\omega^2s_A^2}\frac{\partial^2}{\partial \gamma^2} }
\right] U.$  

We solve these equations implicitly at each $ \Delta\alpha$ extrapolation step by a finite-difference splitting method that alternately advances the wavefield in the $ \beta$ and $ \gamma$ directions. Splitting methods allow us to apply the $ B$ and $ C$ scaling factors directly by introducing rescaled effective slowness models: $ s^B_{eff}=\frac{s_A}{B}= \frac{A}{B}s$ and $ s^C_{eff}=\frac{s_A}{C}=
\frac{A}{C}s$.

One drawback to splitting methods is that they generate numerical anisotropy. To minimize these effects, we apply a Fourier-domain phase-correction filter (, )

$\displaystyle U = U{\rm e}^{ {\rm i} \Delta \alpha k_L},$ (A-14)

where

$\displaystyle k_L = \sqrt{ 1-\frac{k_\beta^2}{(\omega s_B^r)^2} -\frac{k_\gamma...
...a}{\omega s_A^r})^2}{1-b_i ( \frac{k_\gamma}{\omega s_A^r})^2} \right) \right],$ (A-15)

and $ s_B^r$ and $ s_C^r$ are reference slownesses chosen to be the mean value of $ s^B_{eff}$ and $ s^C_{eff}$ defined above.
next up previous [pdf]

Next: Impulse Response Tests Up: 3D Implicit Finite-difference Propagation Previous: 3D Implicit Finite-difference Propagation

2009-04-13