Research

• Massively parallel fluid simulation
• Communication library for exa-computing
• Low power optimization of HPC applications

• Massively parallel fluid simulation

We develop the massively parallel numerical simulation codes of Terrestrial, Jovian and Kronian magnetospheres. If you have questions, please ask me.

• Characters of Jupiter and Saturn

  
  Jupiter's magnetospheric configuration is determined by a combination of the solar wind, rapid atmospherically driven rotation, a huge magnetic field and plasma from the moon Io while the overall configuration and dynamics of the Earth's magnetosphere are largely controlled by the solar wind and interplanetary magnetic field (IMF). Saturn was thought to have aspects similar to both Jupiter and the Earth with internally driven and solar wind driven dynamics.
  
  Table 1.　Characters of Earth, Jupiter and Saturn
  

  	Earth	Jupiter	Saturn
  Equatorial radius [km]	6,378	71,492	60,268
  Mass [Earth=1]	1	317.8	95.16
  Dipole magnetic field [nT]	31,000	420,000	21,000
  Rotation period [days]	0.997	0.414	0.444
  Density [kg/m3]	5.52	1.33	0.69

  
  



• Magnetosphere

  
  No less than 99.9% of the matter in the visible Universe (except for the dark matter distribution which is visible through gravitational lensing) is in the plasma state, that is, in the state of ionized gases. The solar/stellar system is filled with plasmas ejected from the upper atmosphere of the sun or stars. The stream of plasma is called as the solar wind or stellar wind, which carries the magnetic field of the sun or stars. Various phenomena, such as space storms or geomagnetic storms, are caused by the interactions between the solar/stellar wind and an intrinsic planetary magnetic field. Thus studies of space plasma are important to understand our Universe and to forecast the space weather.
  
  

  
  Figure 1.　Jovian magnetosphere [Bagenal et al., 2004]




• MHD simulation

  
  To understand the global configuration and dynamics of space plasma, MHD equation is often used. The MHD equation is derived from the moments of Vlasov equation and Maxwell equation and is as follows.
  
  

  

  We use a three-dimensional MHD code, which has been used for studies on global structures and dynamics of planetary magnetospheres [Ogino et al., 1992, Fukazawa et al., 2012]. The MHD code uses the “Modified Leapfrog” method (MLF), in which partial difference equations are solved by the two-step Lax-Wendroff method for one time step and then by the Leapfrog scheme for (l-1) time steps and the procedure is repeated. Detailed explanation of MLF is in our previous paper [Fukazawa et al., 2014].
  
  

  
  Figure 2.　Simulation result of Terrestrial magnetosphere

  
  



• MHD simulation of Jovian magnetosphere

  
  The scale of Jovian magnetosphere is huge due to the intrinsic magnetic field. we use the supercomputer system to simulate the such a huge Jovian magnetosphere, however we need 1000 times higher spatial resolution to see the appropriate configuration of magnetosphere.
  
  

  
  Figure 3.　Direction of plasma flow in the equatorial plane [Woch et al., 2002]

  The periodic plasma flow ejection in the Jovian magnetosphere is observed by the Galileo spacecraft (Figure 3). From the observation results it is hard to understand the ejection mechanism, however, our simulation can reproduce the similar periodic plasma ejection (Figure 4 and 5) to the observation and find out the ejection is controlled by the plasma corotation region and tail reconnection position.
  
  

  
  Figure 4.　Evolution of magnetic plasmoid in the meridian plane [Fukazawa et al., 2005]

  
  

  
  Figure 5.　Periodic plasmoid ejection in the equaotrial plane [Fukazawa et al., 2005]

  
  



• MHD simulation of Kronia magnetosphere

  
  The movie of Kronian magnetosphere simulation using 300 million grid points is uploaded here.
  We have reported that many vortices are formed in the Kronian magnetosphere by the simulation [Fukazawa et al., 2007, Figure 6] then the Cassini spacecraft find the evidence of vortex configuration in the Kronian magnetosphere [Walker et al., 2010]. Recently we have performed the high resolution of Kronian magnetospheric simulation and suggested the relationship between the vortex and emission of aurora [Fukazawa et al., 2012, Figure 7].
  

  
  Figure 6.　The temperature and flow vectors in the equatorial plane for the simulation with no IMF (a), southward (b) and northward IMF (c) [Fukazawa et al., 2005]

  
  

  
  Figure 7.　The magnetic field connection from the polar region to the magnetosphere [Fukazawa et al., 2012]

  
  



• High efficient and massive parallel computing

  
  The size of magnetosphere is very large and the 10th order difference phenomena are occurred. These phenomena are related with each other thus we need to calculate the all of system at once to understand their dynamics. In addition to simulate the plasma behavior we have to calculate the magnetic field including the neutral fluid. From these circumstances we use the supercomputer system to simulate the global magnetosphere. Using supercomputer system, it is important to maximize the performance of application thus we study the optimization and parallelization of application. For example, we investigate which type of domain decomposition (Figure 7) is suitable for the various supercomputer systems.
  

  
  Figure 7.　Domain decomposition [Fukazawa et al., 2010]

  Figure 8 shows the evaluation results using MHD code on the SR16000/L2. In Table 2 we compare the performance of MHD code on various super computer systems.
  
  

  
  
  Figure 8.　Effective performance of SR16000/L2 [Fukazawa et al., 2010]

  
  Table 2.　Performances of MHD code on various computer systmes
  

  
  

△Return to the top

• Communication library for exa-computing

ACE project（CREST, PI: Dr. Nanri@Kyushu Univ.）

• ACP library
  Communication library developped by the Nanri's CREST
• Halo communication
  Optomization of Halo communocation (in preparing)

△Return to the top

• Low power optimization of HPC applications

PomPP project (CREST, PI: Dr. Kondo@Univ. of Tokyo）

• Power Management Framework for Post-Petascale Supercomputers (PomPP)
  Introduction of PomPP
• Power budget optimization
  Maximization of Power performance in HPC application with controling the power budget of CPU and DRAM (in preparing)

△Return to the top

Return to the web page of Iwashita/Fukazawa Labo.

Return to the web page of Academic Center for Computing and Media Studies

Academic Center for Computing and Media Studies, Kyoto University