5:15 PM - 6:45 PM
[STT39-P03] Accelerated numerical simulation of slow earthquakes by HPC –Application to the Nankai region-
Keywords:Slow earthquakes, Numerical simulation, High Performance Computing
The seismic moment of fast and slow earthquakes ranges in more than 10 orders. Slip velocities also can be more than 10 orders of magnitude, ranging from almost locked state to coseismic slip. In addition, typical slip duration of M1-class fast earthquake is about 0.01 s. On the other hand, typical duration of long-term slow slip events (SSEs) sometimes reaches to several years. The recurrence interval of the phenomena ranges over time scales of several hundred years for large earthquakes on plate boundaries and several- to ten-thousand years for inland earthquakes. Fast and slow earthquakes have been shown to follow different scaling laws. The magnitudes of slow and fast earthquakes are proportional to the characteristic time and the cube of the source duration, respectively. In addition, fast earthquakes obey a power law, which is well known as the Gutenberg-Richter law, while the functional shape of frequency distribution of magnitude of slow earthquakes is ongoing discussion (e.g., power law, or exp-type).
In order to understand fast and slow earthquakes through models, it is necessary to reproduce and examine these statistical properties for events that occur over a large-scale range. In addition, occurrences of slow earthquakes are sometimes found before the occurrence of fast earthquakes. As a recent example, aseismic slip is suggested during swarm activity in the Noto region. Understanding such multiscale characteristics, and slow-to-fast process is a main scientific question in our research project, "Spatio-temporal multiscale modeling and forecast of slow and fast earthquakes" in the MEXT Grant-in-Aid for Transformative Research Areas (A), entitled “The Science of Slow-to-Fast Earthquakes”. To numerically reproduce such multiscale phenomena directly, large scale computing is essential. Therefore, the use of HPC is also a major theme of our project.
The program, which is presented by this paper, aims to reproduce slow slip events (SSEs), which has the duration from 1 day to several years, in the time scale of seismic cycles of megathrust earthquakes. Plate interface is modeled by small triangular elements, at which frictional stress is given by the rate- and state-dependent friction law with cutoff velocities. Interaction between elements is given as the stress change, assuming quasi-static response of semi-infinite elastic medium. As the Green's function of stress change for unit slip can be obtained as an analytical solution, temporal evolution of slip velocity and stress is calculated by a boundary element method, adopting an adaptive time step Runge-Kutta method. A main bottleneck of this simulation is the evaluation of the product of a large dense matrix and a vector to calculate stress change. I presented the development of a GPU-compatible numerical code in this session last year. In this paper, I present the result of application of the developed numerical code.
As a computation result, slip history in the Nankai region of about 1500 years are now obtained. Large earthquakes at the plate boundary occur repeatedly with the interval of about 100 years. Slow earthquakes with the duration of several days (i.e., short-term SSEs) are also reproduced, and repeats at the intervals of several months. As very large storage device is necessary to store the result at all time steps, output of the result to a file is limited to selected time steps in our numerical code. However, the result of full timesteps is sometimes necessary, for example, in the spatial-temporal discussion of peak slip velocity. Therefore, peak slip velocity during output time steps is also saved to a file, in the current numerical code. However, the total size of output files for about 1500 years is still about 4TB, while it depends on the parameter setting. It takes much time to extract specific data, and visualize the result. In this presentation, I also discuss such problems in processing the obtained result.
In order to understand fast and slow earthquakes through models, it is necessary to reproduce and examine these statistical properties for events that occur over a large-scale range. In addition, occurrences of slow earthquakes are sometimes found before the occurrence of fast earthquakes. As a recent example, aseismic slip is suggested during swarm activity in the Noto region. Understanding such multiscale characteristics, and slow-to-fast process is a main scientific question in our research project, "Spatio-temporal multiscale modeling and forecast of slow and fast earthquakes" in the MEXT Grant-in-Aid for Transformative Research Areas (A), entitled “The Science of Slow-to-Fast Earthquakes”. To numerically reproduce such multiscale phenomena directly, large scale computing is essential. Therefore, the use of HPC is also a major theme of our project.
The program, which is presented by this paper, aims to reproduce slow slip events (SSEs), which has the duration from 1 day to several years, in the time scale of seismic cycles of megathrust earthquakes. Plate interface is modeled by small triangular elements, at which frictional stress is given by the rate- and state-dependent friction law with cutoff velocities. Interaction between elements is given as the stress change, assuming quasi-static response of semi-infinite elastic medium. As the Green's function of stress change for unit slip can be obtained as an analytical solution, temporal evolution of slip velocity and stress is calculated by a boundary element method, adopting an adaptive time step Runge-Kutta method. A main bottleneck of this simulation is the evaluation of the product of a large dense matrix and a vector to calculate stress change. I presented the development of a GPU-compatible numerical code in this session last year. In this paper, I present the result of application of the developed numerical code.
As a computation result, slip history in the Nankai region of about 1500 years are now obtained. Large earthquakes at the plate boundary occur repeatedly with the interval of about 100 years. Slow earthquakes with the duration of several days (i.e., short-term SSEs) are also reproduced, and repeats at the intervals of several months. As very large storage device is necessary to store the result at all time steps, output of the result to a file is limited to selected time steps in our numerical code. However, the result of full timesteps is sometimes necessary, for example, in the spatial-temporal discussion of peak slip velocity. Therefore, peak slip velocity during output time steps is also saved to a file, in the current numerical code. However, the total size of output files for about 1500 years is still about 4TB, while it depends on the parameter setting. It takes much time to extract specific data, and visualize the result. In this presentation, I also discuss such problems in processing the obtained result.