Subject: Review a paper C509 From: John Rice Date: Mon, 07 May 2001 11:21:57 -0500 To: fox@csit.fsu.edu CC: "John R. Rice" I append a review of: C509: Parallel Simulation System for Earthquake Generation: Fault Analysis Modules and Parallel Coupling Analysis http://aspen.csit.fsu.edu/CCPEwebresource/C509iizuka/fox_iizuka.pdf The paper is a good, high level overview of their work. The technical content is not high due to the large size of the project and the amount of material included. The English is definitly poor: understandable but with many misuses. It would be appropriate for a collection giving an overview of an area but not as a regular technical paper. ============================================================================ Notes on the M. Iizuka et al. paper 509 The presentation is readable but there are many minor misuses of English in the paper. Two or three times this results in ambiguity or confusion about the technical content. The paper addresses three topics: Formulation of the model, Description of extending the parallel system GeoFEM, and discussion of some applications. The formulation is fairly brief, just enough for an expert to understand the mathematical model. There are many ways to obtain a numerical model and how this is done is not discussed beyond noting that it is a finite element model. Considerable attention is focused on the contact of two sliding plates. Several diagrams and some discussion are provided but one does not learn much about the actual model of this interaction. An overview is given of GeoFEM and then some discussion is given as how to extend it to handle moving plates in contact. A mathematical model of the contact is presented and then reformulated so as to be more suitable for the parallelization and are within the GeoFEM framework. This presentation is at an intermediate level, one gets the general idea but not the actual numerical model. The key idea is that an agent (called a coupler) is created with the capability of transferring information between two moving meshes. The paper does not use the agent terminology however. The numerical solvers for the equations are named by method but no further details are given. The examples discussed include (1) A 1.3 million degrees of freedom simulation using GeoFEM unmodified. A picture and performance data for one step are given. (2) A 1.2 million degrees of freedom simulation with moving plates. However, the contact model is simplified so this is not a realistic model. A picture is given without any performance data. (3) A materials application which is to benchmark the computation against an independent result. It is not clear to me that the experiment actually validates the methodology; it is only stated that the results are the same using one and two processors. The paper ends with a high level conceptual design of how to combine GeoFEM and LSMearth (another modeling code). This is a plausible design; the issue is how well the implementation works. In summary, this paper is an overview of a large project in simulation of the earth. The parallelization issues are just starting to be addressed, that is, using only a few processors. It is well organized and, though the English is substandard, the presentation is well done. 651 00000 n