Show simple item record

dc.contributor.authorPoulson, Jack
dc.contributor.authorEngquist, Björn
dc.contributor.authorLi, Siwei
dc.contributor.authorYing, Lexing
dc.date.accessioned2016-02-25T12:31:49Z
dc.date.available2016-02-25T12:31:49Z
dc.date.issued2012-10-25
dc.identifier.citationPoulson J, Engquist B, Li S, Ying L (2012) A parallel sweeping preconditioner for frequency-domain seismic wave propagation. SEG Technical Program Expanded Abstracts 2012. Available: http://dx.doi.org/10.1190/segam2012-0765.1.
dc.identifier.doi10.1190/segam2012-0765.1
dc.identifier.urihttp://hdl.handle.net/10754/597371
dc.description.abstractWe present a parallel implementation of Engquist and Ying's sweeping preconditioner, which exploits radiation boundary conditions in order to form an approximate block LDLT factorization of the Helmholtz operator with only O(N4/3) work and an application (and memory) cost of only O(N logN). The approximate factorization is then used as a preconditioner for GMRES, and we show that essentially O(1) iterations are required for convergence, even for the full SEG/EAGE over-thrust model at 30 Hz. In particular, we demonstrate the solution of said problem in a mere 15 minutes on 8192 cores of TACC's Lonestar, which may be the largest-scale 3D heterogeneous Helmholtz calculation to date. Generalizations of our parallel strategy are also briefly discussed for time-harmonic linear elasticity and Maxwell's equations.
dc.description.sponsorshipThe authors thank TACC for access to Lonestar, and particularly Bill Barth for suggesting that we enforce local memory allocations and Tommy Minyard for organizing the large runs. We also thank Sergey Fomel, Anshul Gupta, Edgar Solomonik, and Laura Grigori for helpful discussions. The work is financially supported by KAUST and the TCCS sponsors.
dc.publisherSociety of Exploration Geophysicists
dc.titleA parallel sweeping preconditioner for frequency-domain seismic wave propagation
dc.typeConference Paper
dc.identifier.journalSEG Technical Program Expanded Abstracts 2012
dc.contributor.institutionThe University of Texas at Austin
dc.date.published-online2012-10-25
dc.date.published-print2012-09


This item appears in the following Collection(s)

Show simple item record