Details of the Abstract
| Title of paper |
Evaluation of an iterative solver as the forward operator for three-dimensional inversion of electromagnetic data |
| List of authors | Weiss, M., Rochlitz, R., Günther, T. |
| Affiliation(s) | LIAG Institute for Applied Geophysics, Hanover, Germany |
| Summary | Modern inversion frameworks often employ direct solvers as the forward engine because multiple solutions are easily accessible using inexpensive forward-backward substitution after an initial resource-demanding matrix factorisation step. Iterative techniques require little resources for single forward solutions but are comparatively time consuming if many solutions need computing. Nonetheless, a resource-light iterative solver can be of use for large-scale inversion of controlled-source electromagnetic data. It could further make inversions feasible on smaller-scale computing platforms. In light of this, we integrated an iterative solver into the existing open-source inversion framework based on custEM and pyGIMLi. We implemented a two-level iterative scheme where the outer solver employs a generalised conjugate residual algorithm preconditioned with a highly efficient block-based preconditioner for square blocks also known as PRESB. The inner level solver either is of the same type as the outer solver, but preconditioned with the auxiliary-space Maxwell preconditioner, or may alternatively be a direct solver. We evaluate the iterative forward operator with synthetic and real data inversion. It can lead to comparative time requirements forthe computation of the full sensitivity matrix while reducing the memory by a factor of around two. Beyond that, we compare the iterative and direct operators in terms of computational requirements for a wide range of the data point and model parameter space that span the size of the sensitivity matrix. |
| Session Keyword | 2.0 EM theory, modelling and Inversion |
| File upload |
2.0_evaluation_of_an_iterativ_rochlitz_02.pdf
|