• 1.1 The Physics Kernel (PROTEUS)
  • - Explicit Alfven waves are being tested.
  • - Specific problems with the FE formulation in general coordinates on a non-uniform grid have been identified.
  • - Ideas for solving these non-uniform grid problems have been formulated, but not tested.
  • - The "red-black" problem still exists.
  • - The "3rd rail" problem still exists (CAN SOMEONE PLEASE BE MORE SPECIFIC ABOUT WHAT THIS IS???)
• 1.2 The Implicit Solver
  • - Dan Barnes has written "version 0" (ITER.F) of a block CG solver that works (ie, gives the right answers). It is written in F77 with array syntax extensions, allocate, deallocate, and automatic storage allocation.
  • - ITER.F has been run only serially.
  • - ITERT.F might be too slow for final production version.
• 1.3 The Pre-Processor
  • - Grid/Block generators exist as separate codes.
• 1.4 The Post-Processor
  • - XDRAW:
    • - Needs improvement to contour plots.
    • - Needs ability to reead NETCDF files.
    • - Needs menus to be added for "point & click" capability.
• 1.5 GUI
  • - Proto-screens designed (can be seen on the web).
  • - Nobody in charge of development.

The following were agreed to as tasks that could proceed in parallel:

• - Physics Kernel (cold plasma; T = 0, n = const.)
• - Time advance of Thermodynamic Quantities (n, T)
• - GUI
• - Pre-Processor (including Grid Generator)
• - Post-Processor
• - Improvements to Implicit Solver/Parallel Computing Issues.

Specific plans for the advancement of these tasks prior to the APS meeting in November were developed. These are summarized below.

• 2.1 Kernel Development Plan:

  DATE MILESTONE

  8/15:
  • - Written document describing the NIMROD data structure complete (Alan G.).
  • - Written list of NIMROD subprograms, including function and I/O interface definition, complete.
  • - Flowchart of NIMROD code complete.
  • - Ongoing explicit studies of cold Alfven waves with PROTEUS completed.
  • - Decision made on spatial discretization for NIMROD.
  • - Interface to implicit solve defined and finalized.
  8/29:
  • - Incorporation of full grid structure complete.
  • - Implcit solve incorporated.
  9/12:
  • - Implicit field solve with susceptability tensor S incorporated. (Initially S will not contain the full plasma response.) This code will be compiled by this date, but not necessarily checked out. (Cold plasma only.)
  • - Validation plan prepared.
  • - Begin testing with different forms of S.
  10/24: - Prepare APS posters.

  11/4: - APS meeting, Denver.

• 2.2 Thermodynamic Advance

  It was decided to delay the development of the algorithm for the advancement of (n, T) until after the APS meeting.

• 2.3 GUI/Pre-/Post-Processor Development

  DATE MILESTONE

  8/5:
  • - Updated GUI functionality documents complete and on the Web (Ming, Olivier) 8/10:
  • - Define role of LLNL and Tom G.in GUI development. 8/31:
  • - Mike P. to report on feasability of HTML/Java and HTML/Python for GUI.
  • - Alan G. to report on interfacing XDRAW with Netscape. 9/15:
  • - Create screens using Tk/Python (Mike P., Alan G., Daniel). 11/1:
  • - Limited functionality and integration with Kernel successfully demonstrated (Mike P., Alan G., Daniel). 11/7:
  • - APS meeting, Denver.
• 2.4 Improvements to Implicit Solver/Parallel Computing Issues.

  It was decided to use Dan's version 0 (ITER.F) for the APS prototype. Discussions were held on longer term options for the implicit solve. These are summarized below:

• 1. Dan's existing code (ITER.F)
• 2. AZTEC with modifications to NIMROD data structure.
• 3. Modify AZTEC to deal with structured grid and symmetric, positive definite matrix.
• 4. Domain decomposition version of AZTEC.
• 5. ISIS (as is!) (Options 2-5 require storing the matrix)
• 6. ITER.F enhanced with domain decomposition iterative method (modified multigrid approach).

• 1. Portability. Must run on: IBM cluster, HP/SMP, T3E, C90.
• 2. Must plug into NIMROD F90 interface (firewall!).
• 3. Must have F90 callable interface.
• 4. Must be faster than version 0 of ITER.F.
• 5. Must have technical support.
• 6. Selectable options (eg, choice of preconditioner).
• 7. Scalable.

• 1. Dan to put ITER.F on web and update (immediately). Others to study and give feedback to Dan.
• 2. LLNL/Alice to set up e-mail reflector for CG working group: SNL, LLNL, Dan B., Rick N. Rick is the Point of Contact for this group.
• 3. CG working group to decide on site and date for group meeting (Rick N.)
• 4. CG group to formulate development plan by 8/31.
• 5. Curt to check on getting new allocations at NERSC on T3E for NIMROD.