Presentation Information
[250102-02-01]Earthquake simulation enhanced by high-performance computational science
Prof. Tsuyoshi Ichimura (The University of Tokyo)
With the accumulation of measurement data, there is an increasing need for analysis using high fidelity models. The huge analysis cost associated with these models is a common problem in many fields. Such analysis will often not be possible simply by using conventional methods and a supercomputer, so there is a need to develop computer architecture-aware algorithms that can take advantage of massively parallel supercomputers. On the other hand, the integration with data science (e.g. neural networks) is being promoted to improve the efficiency of analysis, but accuracy assurance is often an issue. This talk will present recent computational science algorithms with high performance computing (including integrated algorithms with data science that can guarantee accuracy to further improve the efficiency of analysis) that are suitable for modern computer systems/architectures to efficiently solve large-scale nonlinear/linear, dynamic/static problems. This presentation will show implementations using the finite element method with an application example for earthquake problems, but it is expected to be applied and extended to other similar problems.