AMR Workshop Summary

Wolfgang Bangerth and Shijie Zhong helped organize the workshop. There was only one session with four invited talks that took place in the morning of Oct 24 and was chaired by Scott King. The rest of the workshop was mainly for deal.II tutorials run by Wolfgang Bangerth. Wolfgang showed a large number of examples in which the deal.II package was used to construct AMR finite element software. On the morning of Oct 26, Shijie Zhong chaired a discussion session on how our community could benefit from using AMR software and how we could proceed to build it.

From invited talks and tutorials, most of the workshop participants clearly recognized the potential benefits of AMR software and the feasibility of building the software. However, different communities are at different stages in terms of their capability for building AMR software.

The short-term tectonics community, as part of CIG activities, had developed the PyLith package. PyLith is quite compatible with AMR. PyLith has most of the components to move to the next step of using AMR, if the community decides to move to that direction. What is needed are additional modules.

The existing codes for mantle dynamics (CitcomS and CitcomCU) and long-term tectonics (Gale) are not easily extended to incorporate AMR capabilities. There are two ways to realize AMR in mantle dynamics and long-term tectonics. The first is to use the deal.II package to build some sample codes for mantle convection and plasticity problems. Although deal.II-based software can only be scaled to ~100 cores/processors, it can still be very valuable both in terms of research and teaching. For example, deal.II can help sort out better algorithms for mantle convection problems, given its flexibility in using different element types and solvers. The second is to take advantage of the recent effort by Omar Ghattas’ group from UT-Austin in building a scalable AMR-based mantle convection code. Mike Gurnis from Caltech/CIG and Shijie Zhong’s group from University of Colorado are already in collaboration with Omar Ghattas’ group in this effort. The purpose is not only to use AMR software for mantle convection but also to make a code scalable up to 100,000 processors to achieve petascale performance. However, we also recognize that the particle tracing which is important for geophysical applications in parallel and with AMR is a challenging book-keeping problem.

It was suggested that CIG should spend some effort in creating a suite of deal.II example codes for potential geodynamics applications. It was also suggested that CIG needs to maintain the existing codes such as CitcomCU, CitcomS and its latest version with compressible convection. We also need to examine to what extent these existing codes can be scaled up to a large number of processors. This is important both because there is a large user group and because it does some things very well.

Submitted by Shijie Zhong