Please use this identifier to cite or link to this item:
http://hdl.handle.net/2122/8028
Authors: | Tondi, R.* Cavazzoni, C.* Danecek, P.* Morelli, A.* |
Title: | Parallel ‘large’ dense matrix problems: application to 3D joint inversion of seismological and gravity data | Journal: | Computers & geosciences | Series/Report no.: | /48 (2012) | Publisher: | Elsevier Science Limited | Issue Date: | Nov-2012 | DOI: | 10.1016/j.cageo.2012.05.026 | Keywords: | Parallel Dense matrix Block-cyclic distribution Inverse problem Probability density function ScaLAPACK Gravity field Shear-wave velocity structure Density structure |
Subject Classification: | 04. Solid Earth::04.01. Earth Interior::04.01.02. Geological and geophysical evidences of deep processes 04. Solid Earth::04.03. Geodesy::04.03.04. Gravity anomalies 04. Solid Earth::04.06. Seismology::04.06.07. Tomography and anisotropy 05. General::05.01. Computational geophysics::05.01.03. Inverse methods 05. General::05.01. Computational geophysics::05.01.05. Algorithms and implementation |
Abstract: | To obtain accurate and reliable estimations of the major lithological properties of the rock within a studied volume, geophysics uses the joint information provided by different geophysical datasets (e.g. gravimetric, magnetic, seismic). Representation of the different types of information entering the problem using probability density functions can provide the mathematical framework to formulate their combination. The maximum likelihood estimator of the resulting joint posterior probability density functions leads to the solution of the problem. However, one key problem appears to limit the use of this solver to an extensive range of real applications: information coming from potential fields that implies the presence of dense matrices in the resolving estimator. It is well known that dense matrix systems rapidly challenge both the algorithms and the computing platforms, and are not suited to high-resolution 3D geophysical analysis. In this study, we propose a procedure that allows us to obtain fast and reliable solutions of the joint posterior probability density functions in the presence of large gravity datasets and using sophisticated model parametrization. As it is particularly CPUconsuming, this 3D problem makes use of parallel computing to improve the performance and the accuracy of the simulations. Analysis of the correctness of the results, and the performance on different parallel environments, shows the portability and the efficiency of the code. This code is applied to a real experiment, where we succeed in recovering a 3D shear-wave velocity and density distribution within the upper mantle of the European continent, satisfying both the seismological and gravity data. On a multiprocessor machine, we have been able to handle forward and inverse calculations with a dense matrix of 215.66 Gb in 18 min, 20 s and 20 min, 54 s, respectively. |
Appears in Collections: | Article published / in press |
Files in This Item:
WEB OF SCIENCETM
Citations
20
10
checked on Feb 10, 2021
Page view(s) 5
740
checked on Mar 16, 2024
Download(s) 50
104
checked on Mar 16, 2024