the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Calibration of Absorbing Boundary Layers for Geoacoustic Wave Modeling in Pseudo-Spectral Time-Domain Methods
Carlos Spa
Oilio Rojas
Josep de la Puente
Abstract. This paper develops a calibration methodology of the artificial absorbing techniques typically used by Fourier pseudo-spectral time-domain (PSTD) methods for geoacoustic wave simulations. Specifically, we consider the damped wave equation (DWE) that results from adding a dissipation term to the original wave equation, the sponge boundary layers (SBL) that apply an exponentially decaying factor directly to the wavefields, and finally, a classical split formulation of the Perfectly Matched Layers (PML). These three techniques belong to the same family of absorbing boundary layers (ABL), where outgoing waves and edge reflections are progressively damped across a grid zone of NABL consecutive layers. The ABLs used are compatible with a pure Fourier formulation of PSTD. We first characterize the three ABL with respect to multiple sets of NABL and their respective absorption parameter for homogeneous media. Next, we validate our findings in heterogeneous media of increasing complexity, starting with a layered medium and finishing with the SEG/EAGE 3-D Salt model. Finally, we algorithmically compare the three PSTD-ABL methods in terms of memory demands and computational cost. An interesting result is that PML, despite outperforming the absorption of the other two ABLs for a given NABL value, requires approximately double the computational time. The parameter configurations reported in this article, can be readily used for PSTD simulations in other test cases, and the ABL calibration methodology may be applied to other wave propagation schemes.
- Preprint
(1578 KB) - Metadata XML
- BibTeX
- EndNote
Carlos Spa et al.
Status: open (extended)
-
RC1: 'Comment on gmd-2023-76', Anonymous Referee #1, 16 Aug 2023
reply
This paper is interesting and very relevant for computational seismologists. Although it does not develop new methods, it contributes a great comparison of different strategies for modeling absorbing boundary conditions. The extensive tests regarding both accuracy and computational performance are very useful and something I have not seen in the literature before. I especially appreciate the honest assessment of unpredictability when it comes to modern FFT libraries.
I really don’t have much to criticize, but I would primarily ask the authors to elaborate a bit more in the discussion on these three points:
- If I understand it correctly, the error measure considers the wavefield in the entire domain. In many applications, however, one would only be interested in a few measurement points at the surface. Have you done any analysis using such a restricted error criterion, and would you expect a qualitative change of the results?
- You mention in section 2.1 that by setting the ABL parameter to zero, the same wave equation can be used in the physical domain and in the exterior. Do you distinguish those domains in the implementation? This would potentially affect computational performance considerably (with potential implications on load balancing)?
- I understand that absorbing boundaries for elastic media are a totally different can of worms, but I would be interested in getting your take on the validity / transferability of the results to VTI acoustic (and maybe elastic) media.
There just a few more small points I noted, which I am listing below.
Line 100:
Do you assume differentiability of the source in time or in time and space? Just wondering about point sources here.
Line 125:
What is the reasoning behind adding the additional point with a pressure of zero? Is it correct that this essentially gives a homogeneous Dirichlet boundary condition? Why don’t you use some first-order condition at the outer boundary instead?
Line 160:
Have you analyzed the effect of the non-differentiability of sigma, i.e., the kink at the transition from the inner to the outer domain? I can imagine this might lead to artificial reflections.
Eq. 11:
Is there any physical intuition behind the exponential decay? I assume artificial reflections can occur similar to DWE if the slope is too steep? Intuitively, I would have assumed that N_ABL also enters into the formula, but it seems the values at the outer boundary will differ depending on the distance only?
Just as a personal preference, I probably would have used other symbols than $\sigma$ and $\mu$ for the ABL-related coefficients, as I would associate those with stress and shear modulus, respectively.
Line 265:
Typo: expressions
Line 288:
Out of curiosity: Is there a reason for not using a more recent version of g++?
Line 397:
Just to double-check: Are you using a free surface condition at the top? I don’t think this is the case, but I somehow would have expected this for the realistic SEG/EAGE model.
Line 445:
Consistency when referring to Fig., Figure, figure.
Line 480:
Typo: Such -> such
Citation: https://doi.org/10.5194/gmd-2023-76-RC1 -
AC1: 'Reply on RC1', Carlos Spa, 27 Aug 2023
reply
-
AC2: 'Reply on AC1', Carlos Spa, 27 Aug 2023
reply
-
AC2: 'Reply on AC1', Carlos Spa, 27 Aug 2023
reply
-
EC1: 'Comment on gmd-2023-76', James Kelly, 05 Sep 2023
reply
Review of Spa, Rojas, and de la Puente
This paper compares and analyzes three popular absorbing boundary layer (ABL) methods for solving linear wave propagation problems with the pseudo-spectral time-domain methods. A novel metric is presented to compare the accuracy of the candidate methods, and some recommendations are given. The numerical results/analysis should be useful to people who use PSTD codes, and may help researchers who use other methods as well (see below). While the writing is acceptable, there are some problems (see minor comments), and the manuscript could benefit from additional editing. Hence, I recommend the paper for publication in GMD subject to minor revisions.
Major Comments
- Can the authors discuss the applicability of this ABL analysis to wider classes of wave equations. For example, dispersive and/or nonlinear wave equations? Besides geoacoustics, fields such as biomedical acoustics and electromagnetics, often use dispersive and nonlinear models.
- While the PSTD is widely used in wave propagation problems, other numerical methods are also available, such as spectral finite element methods and finite volume methods (amongst others). Are any of these results applicable to such methods. For example, is the efficiency of the SBL relative to the PML restricted to PSTD, or is it a more general result? While I don’t expect the authors to implement ABLs in such solvers, a short discussion of the wider applicability would give the paper a wider scope.
Minor Comments
Abstract, title of Sec. 2.4: Berenger’s paper uses the term “Perfectly Matched Layer”, not “Layers”. Recommend sticking to the singular.
Line 100: What does “finite in space and time” mean? Does this mean “bounded”?
Line 245: Gao et al. is repeated.
Line 251: Replace “less” with “the least”.
Line 255: Eq. (18) is proportional to the (discrete) L^2 norm.
Figures 3 and 5: The vertical axis is label “energy” which is not precise since the quantity given by Eq. (19) does not have the units of energy. Recommend using the notation defined in Eq. (19) on vertical axis.
Figure 7: What are the units on the x and z axes?
Line 478: “Such” should not be capitalized. Also, “put to the test” could be replaced with “tested”.
Line 492: Need a space after “of”.
Citation: https://doi.org/10.5194/gmd-2023-76-EC1 -
AC3: 'Reply on EC1', Carlos Spa, 15 Sep 2023
reply
-
AC4: 'Reply on EC1', Carlos Spa, 15 Sep 2023
reply
-
AC5: 'Reply on EC1', Carlos Spa, 16 Sep 2023
reply
Dear Editor,
In the document https://doi.org/10.5194/gmd-2023-76-AC3 (in A2) and in the marked document https://doi.org/10.5194/gmd-2023-76-AC4 (in Sec. 3.1.) we wrote :
"Moreover, It is important to highlight that the methodology for calibration of ABLs presented in
this work is based upon three main components. Firstly, using representative models, secondly,
establishing suitable metrics for absorption and finally, reducing the calibration to two parameters.
We are not adding any assumptions regarding the underlaying PDEs used (linear acoustic waves, in
our case). Similarly there are no assumptions tied to the numerical method (pseudospectral time-
domain, in our case). Nevertheless two modifications are foreseen for broadening the applicability
of the method. On one hand, in the case of using other physical models, we would need to modify
Eq. (18) with an alternate energy proxy. On the other hand, in the case..."Eq. (18) should be Eq. (21). Note that we have fixed into the final version of the tex document.
best regards,
the authors
Citation: https://doi.org/10.5194/gmd-2023-76-AC5
Carlos Spa et al.
Carlos Spa et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
207 | 52 | 64 | 323 | 3 | 4 |
- HTML: 207
- PDF: 52
- XML: 64
- Total: 323
- BibTeX: 3
- EndNote: 4
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1