C++ API reference

The true bulk fo this package is written in the class fdModel. This class takes care of the wavefield simulations. Additionally, it is able to backpropagate a wavefield. By storing the forward propagation and correlating it during a backward (or adjoint) simulation, we can compute sensitivity kernels.

Typical usage should be done in Python. However, if you want to interface directly with the finite difference code, you can work with the C++ API. It gives direct access to the dynamical fields, sensitivity kernels, etc. This C++ API is exposed in Python using PyBind11.

Here and there we need to do some type casting between Python and C++ objects. This additional layer is not documented in this reference (as it is not part of the C++ api). It can be found in src/psvWave.cpp

Using the C++ API also requires the configuration files to be present at time of instance construction. In other words, everytime we make an fdModel object, we need to have a properly formatted conf.ini file. How to create and use these is described in the Python API reference.

Important parts

• fdModel class
• Miscellaneous methods and definitions

Type definitions

Throughout the C++ code, we have a few important type definitions. This makes working with varying precision and eigen types a little more concise.

using real_simulation = double

Typedef that determines simulation precision. On x86_64 systems it is typically fastest to use double, as this is the native precision. Double has about a 10 % performance gain over floats as tested on Ubuntu 18.04 x86_64.

using dynamic_vector = Eigen::Matrix<real_simulation, Eigen::Dynamic, 1>

Typedef that is a shorthand for the correct precision column vector. This vector has the right precision and shape to be used in matrix equations. It is of dynamic size.