next up previous [pdf]

Next: Conclusions and Discussion Up: Maharramov: Efficient elastic modeling Previous: The Method

Implementation and Results

The system of matrix equations 8 is solved only once for each triple of temporal frequency and elastic moduli values, and for each pair of horizontal wave numbers. In our prototype implementation of the one-way extrapolator we compute the matrices $ A,B$ at the beginning of the frequency loop and subsequently use the tabulated matrices in the depth extrapolation loop (inside the frequency loop). A more efficient approach can be employed in a production implementation of the extrapolation method: system 8 can be solved using Newton's method (see e.g. (Higham, 2008)) in a one-off computation for each set of the temporal frequency, elastic moduli and horizontal wave numbers and stored in a look-up table. The symmetry of the extrapolation operators 11, that appear to be multi-component counterparts of the acoustic phase-shift operator (see Claerbout (1985)), can be exploited to achieve a substantial reduction in the size of the precomputed operator tables. Figure 1 is the plot of the maximum of the imaginary parts of the three eigenvalues of operator

$\displaystyle K=-\Delta z E^{-1}\left[A(-i\partial_x,-i \partial_y)+c_{\omega}I\right],$ (15)

within its passband. The operator is the one used later to produce images of Figures 2,3,4,5. The real parts of the eigenvalues of operator 15 are zero within the operator passband and negative outside. The imaginary parts of the other two eigenvalues exhibit similar behavior. Operator $ K$ , of equation 15, is the logarithm of the extrapolation operator 11, and the spectral plot of Figure 1 corresponds to the phase of the phase-shift extrapolator in the acoustic case (see (Biondi, 2005)). The crucial difference in the elastic multicomponent case is that the multicomponent ``phase-shift'' is defined by three such scalar phase-shift operators with phases $ \phi_1,\phi_2,\phi_3$ , and a unitary operator $ Q$ , determined by the eigenvector expansion of $ K$ as follows:

$\displaystyle K=Q\left[ \begin{array}{ccc} i\phi_1 & 0 & 0 \\ 0 & i\phi_2 & 0 \\ 0 & 0 & i\phi_3 \end{array} \right]Q^{\ast}.$ (16)

The pass bands of the three phase shift operators are, generally, different, but the real parts of the eigenvalues of 15 are non-positive across all three pass bands.

Figures 2,3,4,5 demonstrate the result of applying our method to extrapolating displacement waves from a concentrated impulse at the surface. Medium parameters used in this test were $ 316$ m/s shear-wave velocity

$\displaystyle v_{\texttt{S}}=\sqrt{\mu/\rho}$

and $ 632$ m/s pressure-wave velocity

$\displaystyle v_{\texttt{P}}=\sqrt{(\lambda+2\mu)/\rho}.$

The extrapolation grid was $ 128\times 128 \times 128$ with a 5 m step, frequency range 2-12 Hz with 1 Hz step. The values of the elastic parameters used in this test are uncharacteristically low for seismic applications and were chosen solely for the purpose of fast small-scale simulation on a single-core PC using Matlab.

pressure
Figure 2.
Pressure wave extrapolated from an impulse displacement, 2-12 Hz, $ 128\times 128 \times 128$ grid, $ 5$ m step, $ 316$ m/s shear-wave and $ 632$ m/s pressure-wave velocity.
pressure
[pdf] [png]

component3
Figure 3.
Vertical component of a wave extrapolated from an impulse displacement, inline section, 2-12 Hz, $ 128\times 128$ grid, $ 5$ m step, $ 316$ m/s shear-wave and $ 632$ m/s pressure-wave velocity.
component3
[pdf] [png]

Since the impulse at the surface is an asymmetric horizontal displacement but can be assumed to be symmetric in the vertical direction, our waves are effectively a combination pressure and shear waves for the horizontal components, while the vertical displacement wave should kinematically match the pressure wave. And indeed, the pressure wave plot 2 and vertical displacement plot 3 exhibit excellent kinematic agreement.

component1
Figure 4.
Inline component of a wave extrapolated from an impulse displacement, inline section, 2-12 Hz, $ 128\times 128 \times 128$ grid, $ 5$ m step, $ 316$ m/s shear-wave and $ 632$ m/s pressure-wave velocity. Note the slow shear wave.
component1
[pdf] [png]

component2
Figure 5.
Crossline component of a wave extrapolated from an impulse displacement, inline section, 2-12 Hz, $ 128\times 128 \times 128$ grid, $ 5$ m step, $ 316$ m/s shear-wave and $ 632$ m/s pressure-wave velocity. Note the slow shear wave.
component2
[pdf] [png]

The horizontal wave-component plots 4 and 5, on the other hand, show pressure- and shear-wave images, both correctly positioned in agreement with the velocity values used in the simulation. Boundary reflections and low frequency content cause some imaging artifacts that are not related to the method.


next up previous [pdf]

Next: Conclusions and Discussion Up: Maharramov: Efficient elastic modeling Previous: The Method

2012-10-29