Next: Gradient comparison Up: Shen and Clapp: Boundary Previous: Introduction

algorithm description

FWI as used here is a gradient-based time domain implementation (Shen, 2010). Different boundary conditions are used for the gradient calculation, leading to slight modifications of the gradient calculation algorithm. However, the steplength search uses absorbing boundary conditions in all cases. The pseudo-code of the gradient calculation using the absorbing boundary condition is as follows:
$\begin{algorithm} % latex2html id marker 29\caption{Pseudo code of gradient ca... ...ource wavefield to generate gradient; \ENDFOR \end{algorithmic} \end{algorithm}$

The pseudo-code of the gradient calculation using the time-reversible boundary condition (random boundary condition and continuation of velocity) is as follows:
$\begin{algorithm} % latex2html id marker 36\caption{Pseudo code of gradient ca... ...undary, generate gradient on the fly; \ENDFOR \end{algorithmic} \end{algorithm}$

It is worth mentioning that for the MPI version of the code, wavefield propagation on different computational nodes uses different random boundary realizations, which further reduces artifacts.

2012-05-10