search

slimgroup / JUDI.jl

Julia Devito inversion.
https://slimgroup.github.io/JUDI.jl
MIT License
96 stars 30 forks source link

MFE: JUDI's boundary function fails in test_adjoint_J from Devito #99

Closed nogueirapeterson closed 2 years ago

nogueirapeterson commented 2 years ago

This MFE is related with PR #95 . I could notice that the boundary from JUDI fails in test_adjoint_J in Devito. I only brought the initialize_damp from JUDI to Devito and ran the test. We create a MFE mfe.txt to show that issue. You can run the MFE as: python mfe.py -d devito
to run the test_adjoint_J with initialize_damp from Devito. And python mfe.py -d judi to run the test_adjoint_J with initialize_damp from JUDI.

So considering this fact, could we create a damp_kernel so that we could keep the two functions? I suggest this because I would like to have the viscoacoustic operators pass the tests like in Devito.

mloubout commented 2 years ago

All the operators including J pass the adjoint test in JUDI test suite so I don't think external tests are necessarily relevant.

I'll have a quick look though, by how much does it fail

mloubout commented 2 years ago

After some thoughts, since you add Q in a much better way, the current implementation isn't really needed. So feel free to replace the current one with the "devito" one. I'll check on one of my PR still works for long modeling times

nogueirapeterson commented 2 years ago

thanks for considering this.