Liebe Mitarbeiter des INS, im Rahmen unseres Forschungsseminars "Mathematics of Computation" wird Lutz Gross (University of Queensland) zum Thema "On the Solution of Large-scale Geophysical Inversion using Finite Element Methods on Parallel Computers" vortragen. Der Vortrag findet am Donnerstag, den 23.4. um 13 Uhr in 6.020 statt. Eine Kurzzusammenfassung des Vortrags gibt es unten in dieser Nachricht oder unter http://ullrich.ins.uni-bonn.de/seminar/ . Mit den besten Grüßen Tino Ullrich und Daniel Peterseim ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Abstract: On the Solution of Large-scale Geophysical Inversion using Finite Element Methods on Parallel Computers (Lutz Gross, University of Queensland) The inversion of geophysical data is the attempt to build a three-dimensional model of the Earth's subsurface from data collected on or near the surface. The problem can be formulated as an optimization problem minimizing the defect of measurements and corresponding predictions subject to constraints from physical models in the form of partial differential equations. It is common practice to solve the problem using a 'first discretize then optimize' approach.. Although being very efficient for small to medium size problems the approach has strong limitations for large-scale data sets and non-quadratic inversion problems for instance for electro-magnetic data sets and joint inversion problems. In the talk we will present an alternative framework tailored for using the finite element method following the 'first optimize then discretize' approach. As this approach maintains the problem sparsity throughout the inversion the calculations can easily be applied to large spatial grids in parallel across thousands of processor cores with good scalability. We will discuss mathematical and computational aspects of the approach in the context of the inversion of magnetic and gravity anomaly data. As an application scenario we will present the inversion of satellite gravity data over the entire Australian continent. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ _______________________________________________ Ins-mitarbeiter mailing list Ins-mitarbeiter@ins.uni-bonn.de https://mail.ins.uni-bonn.de/mailman/listinfo/ins-mitarbeiter