SPECFEM / specfem2d

SPECFEM2D simulates forward and adjoint seismic wave propagation in two-dimensional acoustic, (an)elastic, poroelastic or coupled acoustic-(an)elastic-poroelastic media, with Convolution PML absorbing conditions.
https://specfem.org
GNU General Public License v3.0
203 stars 147 forks source link

strange loop in the forward wavefield #1218

Open zuoxuan-de opened 7 months ago

zuoxuan-de commented 7 months ago

When I perform forward modeling of various homogeneous elastic media, after the source is excited for a period of time, a small circle with weak energy will appear again, independent of the initial wave. forward_image000000700 forward_image000000800 forward_image000000900 forward_image000001000 forward_image000001100 无限均匀 By the way, my source is the moment tensor source.

mnagaso commented 7 months ago

Hi,

Please remind that the color level in the output images are relative to the max/min value at each time step. So after several steps, the main wave has been absorbed by PML, then a very week energy for example a very small reflection from PML becomes visible. So what we see on the image is it.

zuoxuan-de commented 7 months ago

Hi,

Please remind that the color level in the output images are relative to the max/min value at each time step. So after several steps, the main wave has been absorbed by PML, then a very week energy for example a very small reflection from PML becomes visible. So what we see on the image is it.

Sorry, I still feel confused. In this picture, we can see that the faint energy reflected by the absorption boundary is exactly the opposite direction of the source propagation, but the small circle of weak energy that I am talking about is exactly the same as the direction of the source propagation. I don't know enough about this; can you give us an example to illustrate that?

mnagaso commented 7 months ago

Those are the ripples. Making the mesh size or time step size half or so will reduce it I think.

zuoxuan-de commented 7 months ago

Those are the ripples. Making the mesh size or time step size half or so will reduce it I think.

I tried to cut the time step in half, but I didn't get the result we wanted!It doesn't seem to be any different from before. On the other hand, it's not weak, and in my other example it's very obvious. tttt

mnagaso commented 7 months ago

I recreate the same situation as below: movie

You can see the same ripples after the main lobe. forward_image000001400

Here is the recorded signal at the 3rd station from the bottom Screenshot from 2024-04-18 09-48-50

Then if I limit the y axis range, you can see the source of the ripples, which is actually very small and thus can be regarded as a numerical artifact (as I explained at first). Screenshot from 2024-04-18 09-50-53

If you don't want to see this on the image, you can check the parameter USE_CONSTANT_MAX_AMPLITUDE in Par_file.

mnagaso commented 7 months ago

If we check the other artifacts (which is happening just after the main lobe) at the 3rd station from the left of this image,

forward_image000001400

then the recorded signal is

8 8zoom

This is caused by the high frequency component of source time function at the end of the main lobe is a bit too high for the current mesh design. If we make the mesh size half (use doubled number of nx etc.) this becomes a bit better.

8 8zoom

zuoxuan-de commented 7 months ago

Thank you very much for your detailed reply, it helped me a lot.

mnagaso commented 7 months ago

You're welcome!