PI:
Jonathan Arons
Co-PIs: Marc Davis, Richard Klein, Christopher McKee
The Astronomy Department at UC Berkeley is an international leader, and, according to recent National Research Council rankings, is a top department nationally as well--tied for first with CalTech and Princeton. The Berkeley Astronomy Department also has the highest ratio of National Academy members to the overall size of the Department faculty among all the units on campus. Astronomy grdauate students continue to fill the ranks of professional astronomy and astrophysics around the world, with 80% of its PhDs going on to careers in astrophysics research, and 50% of these eventually working as professors at PhD granting Universities--a record which has remained steady in the face of recent years' tightening in academic job markets. Theoretical astrophyscis has always been a major component of the Department's research effort, and major computational investigations have been part of the theory program since the 1950s.
Together with faculty from EECS, ME, NE, Physics and Mathematics, Astronomy has decided to target star formation as part of a proposed DOE ASCI Center for Computational Modeling of Complex Systems. Star formation is a central question in understanding the origins of the universe, such as the likelihood of forming stable planetary systems capable of supporting life. This depends on the rate at which young stars are formed, their ultimate sizes, and the rate at which the eject matter and energy back into space. The three big simulation components are the gravitational collapse of giant molecular clouds; the radiation hydrodynamics and Rayleigh-Taylor instabilities governing the formation of massive stars; and the turbulent jets emitted by young stars. These are multi-physics simulations with coupled phenomena over many orders of magnitude of time and length scales. The science and technology needed to solve such problems will enhance our ability to perform many similarly complex simulations. This problem is complex enough to use any available computing power to resolve ever finer details. One ultimate target architecture is the Intel ASCI Red machine at Sandia, for which the campus NOW would be an ideal development platform.
Current investigations include gravitational N-body and gas dynamical investigations of galaxy formation and clustering, against the 3d background of the expnading Universe (Davis); 3D gas dynamic and magnetohydrodynamic studies of collapsing interstellar clouds and their fragmentation into stars (Klein and McKee); one and 2D particle-in-cell (PIC) simulations of relativistic collisionless shock waves and their application to the excitation of synchrotron sources,such as the Crab Nebula (Arons). The simulations are successfully used both to uncover new physical effects which depend on the intrinsic nonlinearities of the equations describing these systems, and to construct quantitatively reliable models which can be confronted with telescopic observations.
These projects all have quite a bit of scientific visibility. The work on the Crab Nebula done to date has revived and redirected the whole subject of the excitation of synchrotron sources by central compact objects (neutron stars, black holes) and has astronomical applications under development ranging from pulsar driven nebulae other than the Crab Nebula to the excitation of synchrotron sources in distant clusters of galaxies.
The work on galaxy and star formation has similar ``luminosity'' within the astrophysics community. The use of tree algorithms appears to be the first time these have been applied to plasma PIC problems. Therefore, the results will have substantial influence on plasma simulation in general, which has many applications to space physics and to the physics of laser-plasma coupling and the production of useful energy from fusion through inertial confinement of plasmas. The adaptive mesh algorithms used in the work on the hydrodynamics of star formation have many applications to fluid dynamcial simulations, ranging from combustion in auto engines to the simulation of the weather. Traditionally, the community working on these problems have used workstations from HP, Sun and IBM. If these projects could show that Pentium hardware could be usefully applied to these ``supercomputer'' projects, a wide market in the scientific community would be opened.
Pentium workstations, some organized into a NOW cluster, some freestanding, will be applied to parallel code development and data analysis phases of the projects described above. To focus on one example, previous work on the properties of relativistic, collisionless shock waves in a plasma composed of heavy ions, electrons and positrons has shown that maser emission of waves by the ions gyrating with a ring momentum distribution, which forms by ion reflection off the magnetic gradient in the shock wave, lead, through cyclotron resonant absorption, to the formation of a nonthermal distribution of electron-positron pairs with properties ideally suited to explain the non-thermal synchrotron emission from the Crab Nebula. In addition, the compressions of the pair plasma created at the turning points of the ions' orbits provide a natural interpretation of the concentric rings of brightening observed one light year from the central pulsar which powers the flow. Indeed, the theory is overdetermined by the data, allowing several observational tests to be made which the theory sucessfully passes. However, investigation of some processs of importance to the flow and the particle acceleration physics require two (and possibly three) space dimensions in the calculations. In particular the existing theory cannot explain the acceleration of electrons and positrons which give rise to the radio synchrotron emission. Magnetic pumping is a likely candidate for the so far unknown process, which requires two space dimensions in order to study the simultaneous magnetic compressions and the pitch angle scattering. To do this with even a semi-realistic ratio of the ion mass to the electron mass requires massively parallel PIC computation. While such capability is available at the National Supercomputer Centers, the difficulty of doing the exploratory code development needed using the finite allocations of computer time available has greatly hindered progress on this project.
Therefore, Astronomy proposes to make use of a Pentium NOW cluster to study the parallelization of the relativistic, electromagnetic, 2D PIC code used to study collisionless shock properties, for eventual use on multi-processor, multi-memory NOW architectures. No local progress has been made to date using this approach, since the existing network of Sun workstations in Astronomy is occupied with the interactive data analysis work and simulations carried out by the many separate research groups which are the workstations' owners. A donation by Intel would allow construction of the NOW cluster, since the new Intel workstations would not have been previously acquired for the purposes of many disparate research projects. This PIC parallelization project will be addressed over the next 12-18 months, with the intial attack using a variant of the Barnes-Hut tree algorithm to speed up the message passing delays in the gather scatter operations that greatly inhibit standard parallelizations of PIC algorithms. The resulting code will be used on the small scale local cluster for preliminary investigations of the shock physics. Deeper investigations will require the large clusters available elsewhere on campus. Free standing workstations will be used for analysis of data coming from the simulations; these analyses will include FFT and correlation function analysis of large data sets, as well as rendering of the phase space distribution functions, often essential as a way of revealing the physical sources of free energy that drive the waves generated in the plasma.
The same cluster
and free standing workstations will be applied to similar tasks required by
the galaxy and star formation projects. In additionn, the resourceswill be available
for analysis of the large data sets generated by the ongoing searches for millisecond
radio pulsars (Prof. D. Backer) and by balloon observations of the microwave
background (Prof. P. Richards and Dr. A. Hanany).
February 1999