A parallel, object-oriented framework for frequency-domain wavefield imaging and inversion.

We present a parallel object-oriented matrix-free framework for frequency-domain seismic modeling, imaging and inversion. The key aspects of the framework are its modularity and level of abstraction, which allows us to write code that reflects the underlying mathematical structure and develop unit-tests that guarantee the fidelity of the code. By overloading standard linear-algebra operations, such as matrix-vector multiplications, we can use standard optimization packages to work with our code without any modification. This leads to a scalable testbed on which new methods can be rapidly prototyped and tested on medium-sized 2D problems. Although our current implementation uses (parallel) Matlab, all of these design principles can also be met by using lower-level languages which is important when we want to scale to realistic 3D problems. We present some numerical examples on synthetic data.