search

daniel-koehn / DENISE-Black-Edition

2D time-domain isotropic (visco)elastic FD modeling and full waveform inversion (FWI) code for P/SV-waves
GNU General Public License v2.0
121 stars 66 forks source link

QUELLTYPB=4 fails to do FWI in AC problem #39

Closed ADharaUTEXAS123007 closed 1 year ago

ADharaUTEXAS123007 commented 1 year ago

When I do SEISMO=2, QUELLTYPB=4, The LBFGS method fails to find the step length.

daniel-koehn commented 1 year ago

Does this happen in case of the Marmousi-2 benchmark problem or did you use another model? I tested the parameter combination (SEISMO=2,QUELLTYPB=4, LBFGS) for the Marmousi-2 model extensively, as discussed in this thread:

https://github.com/daniel-koehn/DENISE-Black-Edition/issues/23#issuecomment-667046246

Additionally, you have to set GRAD_FORM=2 in the DENISE parameter file to use a specific type of acoustic FWI gradients. In this case, the updates of the density model also seem to be quite crappy. So I would recommend to update only the P-wave velocity model as discussed in the thread.

Best regards,

Daniel

ADharaUTEXAS123007 commented 1 year ago

Thanks Daniel. This worked as expected.

But I want to bring to your attention in file psv.c, line 180 / adjoint explosive source / if ((QUELLTYPB >= 4) && (mode == 1)) { psource(nt, (wavePSV).psxx, (wavePSV).psyy, (acq).srcpos_loc_back, (seisPSVfwi).sectionpdiff, nsrc_loc, 1); }

it is still nsrc_loc instead of ntr. Is this as expected?

ADharaUTEXAS123007 commented 1 year ago

closing this issue.