ArjunAir: Updating and parallelizing an existing time domain electromagnetic inversion program

Belliveau, Patrick; Farquharson, Colin G.; Haynes, Ronald D.

doi:10.1190/segam2014-1433.1

Cited by 2 publications

(2 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…MUMPS has the ability to automatically adapt to computer load variations during the numerical phase and to solve a wide range of problems including symmetric positive definite, symmetric general, unsymmetric, and rank-deficient systems using either LU or LDL T factorization (Amestoy et al, 2001). Oldenburg et al (2008Oldenburg et al ( , 2013) present a practical formulation for forward modeling and use MUMPS to successfully invert a large airborne data set, whereas Belliveau et al (2014) successfully implement MUMPS and PARDISO within the inversion program ArjunAir. We must point out that direct solvers have large memory requirements, so acceleration using a direct solver can be achieved only when parallel computation is implemented; otherwise, the direct solver is invariably limited to small problems.…”

Section: Acceleration Techniquesmentioning

confidence: 98%

“…Moreover, the inversion of ArjunAir in P223F is based on an iterative GaussNewton method. Belliveau et al (2014) update and parallelize the ArjunAir program with a hybrid MPI/OpenMP forward solver. The parallel inversion of large-scale airborne time-domain EM data is discussed by Haber and Schwarzbach (2014) with stochastic Gauss-Newton methods in which a stochastic approximation to the gradient, Hessian, or both is used.…”

Section: Minimization Of Equation 19 Yieldsmentioning

confidence: 99%

See 1 more Smart Citation

Review on airborne electromagnetic inverse theory and applications

et al. 2015

View full text Add to dashboard Cite

Airborne electromagnetic (AEM) data processing and inversion has progressed from the early conductivity-depth imaging, to 1D inversion for a layered earth model, and most recently to 2D/3D inversions. The new processes have used methods for searching for solutions, applying constraints, and focusing images, such as Occam's, laterally constrained, and holistic inversions. For airborne data imaging and 1D inversion, some algorithms are fast, whereas others have control over data misfit and the vertical or lateral smoothness. Beyond 1D algorithms, 2D/3D inversions are based on currently used methods such as isolated conductor models (which are good for highly conductive orebodies in resistive background) and techniques that discretized regions that are used for more complex structures and backgrounds. In the latter case, the computations are slow, so research is focusing on time-efficient computer algorithms for solving the equations such as Gauss-Newton, quasi-Newton, and nonlinear conjugate gradient algorithms. For the electromagnetic (EM) problem, the solution can be obtained in a more practical time frame if material that has minimal impact is ignored. Two approaches can be used to speedup the EM inversion process -the moving footprint and direct solver methods. We hope our work will to some extent help stimulate and focus the research in AEM inversion.

show abstract

Section: Acceleration Techniquesmentioning

confidence: 98%

Section: Minimization Of Equation 19 Yieldsmentioning

confidence: 99%