Shedding New Light on Exploding Stars: TeraScale Simulations of Neutrino Driven Supernovae and Their Nucleosynthesis (The TeraScale Supernova Initiative) — Accomplishments

Project Title: Shedding New Light on Exploding Stars: TeraScale Simulations of Neutrino Driven Supernovae and Their Nucleosynthesis (The TeraScale Supernova Initiative)
Project PI: Anthony Mezzacappa (ORNL)
Project Co-Is: Polly Baker (IU), John Blondin (NCSU), Steve Bruenn (FAU), David Dean (ORNL), Jack Dongarra (UTK), George Fuller (UCSD), Wick Haxton (UW), John Hayes (UCSD), Jim Lattimer (SUNYSB), Brad Meyer (Clemson), Madappa Prakash (SUNYSB), Faisal Saied (NCSA), Paul Saylor (UIUC), Mike Strayer (ORNL), Doug Swesty (SUNYSB), and Ross Toedte (ORNL).

Principal Scientific Accomplishments
(1) Three-dimensional hydrodynamics simulations have uncovered a supernova shock wave instability that had here-to-fore not been considered in supernova models. This instability may play a significant role in generating the supernova and in defining its asymmetry, which, in turn, may provide the first explanation of the polarization of the light emitted from a supernova, which most likely results from asymmetries in the explosion. (2) The first stellar core collapse simulations with state-of-the-art neutrino transport and state-of-the-art models of the nuclei in the stellar core have been carried out. These simulations have brought together two fields at their respective frontiers, demonstrated the importance of accurate nuclear physics in supernova models, and motivate planned experiments to measure the nuclear processes occurring in stars. (3) Simulations of the "r-process," the rapid neutron capture process believed to occur in supernovae and responsible for half the heavy elements (above iron), led to a surprising and important result: the r-process can occur in very different environments than here-to fore thought, perhaps providing a way around some of the fundamental difficulties encountered in past supernova models to produce r-process element synthesis. (4) Developments that will enable the first realistic two-dimensional supernova simulations coupling two-dimensional hydrodynamics with two-dimensional, multifrequency neutrino transport have been completed. The first of these simulations will be carried out this year. (5) Extensive calculations to increase the realism of and the number of interactions for the neutrino-stellar core interactions in our neutrino transport models have been completed. (6) Significant progress has been made on performing the first neutrino transport simulations that include coherent quantum mechanical effects associated with neutrino mass that are not manifest with massless neutrinos.

Principal Science-Enabling Accomplishments
(1) We have developed algorithms to solve the underlying algebraic equations for multidimensional radiation transport on TeraScale computers. Without these algorithms, the simulation of neutrino transport in stars would not be possible, and without neutrino transport, realistic supernova models could not be constructed. The algorithms were developed in a joint effort involving the TeraScale Supernova Initiative and the TOPS ISIC. The software has now been incorporated in the software libraries, Hyper and PetSc, used by researchers across applications and disciplines. (2) Managment, analysis, and visualization of TeraBytes of data present a significant challenge. We have used data analysis techniques to significantly reduce our simulation data. These techniques have also provided new views on the data and new ways to quantify and understand the turbulent fluid dynamics of exploding stellar cores. The data analyses were carried out within newly developed "problem solving environments" in which different data analysis tools and different visualization engines can be selected by the application scientist. This effort has brought to the foreground key issues of software integration. We have also explored, with some success, the use of "agent" technology in our data-management--to--visualization pipeline, which has given us an augmented ability to analyse and render distributed simulation data. All of this work was performed by members of the Scientific Data Management (SDM) ISIC and Bahram Parvin's group at LBL, with extensive input from TSI. Progress was made in the area of visualization. Custom, application-specific developments by TSI visualization experts progressed well, while at the same time, off-the-shelf packages (EnSight and the TeraScale Browser) have been explored and implemented with some success. On another front, progress on networking was made, particularly with an eye toward data management and visualization, in a testbed involving ORNL and NCSU. Networking hardware and software was put in place in collaboration with Micah Beck’s group at the University of Tennessee and Nagi Rao at ORNL. The goal will be to use the client-server capabilities of EnSight, together with Logistical Networking, to visualize at NCSU data that resides in "depots" on the network. This remote visualization allows data- and compute-intensive rendering to be carried out on an appropriate computing platform at one location while the end result is viewed at another, with minimal data shipped over the network. (3) The instrumentation of our existing radiation transport and hydrodynamics codes with leading performance analysis tools, SvPablo and Tau, proved instrumental in identifying and overcoming performance bottlenecks. In turn, both tools were improved as a result of this instrumentation, which will now benefit all applications that make use of them. This work was performed by the PERC ISIC with extensive input from TSI.

Plans for FY03 and FY04
This year we plan to carry out the first two-dimensional supernova models with two-dimensional, multifrequency neutrino transport. These will mark a quantum leap in realism relative to the current oversimplified two-dimensional models. In the next year, we plan to follow up with three-dimensional models. Computations to provide state of the art stellar core thermodynamics and neutrino interactions will continue to increase in realism, and we will begin to carry out element synthesis studies in situ that before had been carried out in parameterized models. Preliminary magnetohydrodynamics studies of stellar core collapse, to understand the evolution of stellar core rotation and magnetic fields, will be performed this year, as well, followed next year by a partial integration of the radiation hydrodynamics and magnetohydrodynamics simulations. And we will continue to develop transport models with nonzero neutrino mass to explore the impact of neutrino mass on supernova dynamics and element synthesis. These principal scientific goals for the next two years will require a commensurate leap in data managment, data analysis, and visualization infrastructure, and they will continue to push the development of algebraic solvers, performance tools, and software engineering, as our simulation needs rapidly escalate. As a result, we plan to continue our extensive collaborations with the TOPS, SDM, and PERC ISICs, and work more closely with the CCA and TSTT ISICs. Collaborations with the CCA ISIC will be integral to our software engineering and essential to using solvers developed by the TOPS ISIC, which will adopt CCA-defined interfaces to its solvers, and work with the TSTT ISIC will continue to explore alternative numerical schemes for radiation transport that require far fewer operations and much less memory.

Resource Needs
(1) The two-dimensional supernova models will require millions of processor hours on current DoE platforms at NERSC and CCS. These simulations will require a significant number of processors and significant memory to complete. They will be staged simulations, and will require a supportive queue structure allowing reasonable throughput for each stage in the simulation in order to carry out the complete simulation in a reasonable period of time. In addition, sufficient MPP allocations will be needed to carry out simulations for a variety of presupernova stars and input nuclear and weak interaction physics.
(2) Significant data storage will be needed. Our simulations will typically produce 1 TeraByte of data per simulation per variable for the hydrodynamics. Storage requirements for the multidimensional radiation field data will be much more severe.
(3) TSI is a national effort. Networking hardware and software, and middleware, will be critical in enabling collaboration among TSI’s university partners and ORNL.

back to project page