Test Operators

[mloubout]

@ggorman, @mlange05 Now that it is fast enough, I have finalized the adjoint test and gradient script. It passes the adjoint test (exactly) and produce a gradient that looks like what it is supposed to. I am working on the Gradient test now and we should get a 2D acoustic workflow working and running by the end of the week.

[mloubout]

@navjotk , @ggorman , @mlange05 GOOD NEWS I finally fixed everything in the notebook so now we have Adjoint test for the Forward Modelling operator (passes exactly) Adjoint test for Jacobian (passes exactly) Jacobian test (passes exactly) Gradient test (passes exactly)

In the current setup, the gradient needs to be computed inside the boundary layer as well as some receivers are inside this layer, this wouldn't be done in a realistic case but it allows o check that the implementation of the absorbing boundary conditions is correct in the adjoint.

I will start to look at other equations

[mlange05]

Excellent, thank you very much for this, @mloubout. @navjotk, can you please start looking into how to run these tests using opesci-fd, since they provide valuable regression and acceptance tests? Also, I think we can park the SymPy-based cythonization for now, at least until we are doing a more in-depth performance evaluation of these runs.

[navjotk]

@mlange05 I closed the pull request. I am looking into running these tests within opesci-fd Edit: It is taking a while because we here lack a background in physics. Even though the notebook is probably meant to be easy to understand, we need a lot of background reading to be able to understand what is being done. Edit 2: Felipe and I have spent some time trying to understand this together and there are some steps in the analysis that we are not able to understand. Maybe a call with @mloubout might help.