As I have not received any of my requested input from the rest of the group members, the following progress only discusses work relating to Dan Reynolds.

During May, I finished development of the Enzo self-gravity module interface to MGMPI. Although the mathematical and solver interface was completed months ago, the user-interface required enhancement to allow for the multiplicity of ways that Enzo can be started (new simulations, restart with file inputs for boundary values, and restart with subroutine calls for boundary values). The main work in this area was HDF5 restart write/read capabilities, along with other self-gravity-related restart parameters; these then needed integration within the existing Enzo startup/dump routines, requiring adjustments to a large number of Enzo routines. Development on these capabilities is now finished, with support for read/write by each processor of local boundary conditions to its own file (tested), and read/write by each processor of local boundary conditions to one boundary file (through parallel HDF5, untested).

In addition to the above user-interface improvements, I have extended the self-gravity interface to allow use of cell-centered values (as Enzo uses), instead of interpolation to cell corners (as MGMPI natively supports). The resulting solver is significantly slower, but as the data and solution need no interpolation, the final solution is more accurate. This interface allows simpler extension to FLD enhancements. All of these enhancements have been added to the CVS repository. The new self-gravity interface may be enabled through the ISO_GRAV pre-processor directive, and is activated using parameter file arguments.

I have also continued work on the implicit solver infrastructure for Enzo enhancements. To this end, I have developed a general Enzo vector data class (unigrid only for now), along with a full linesearch-enabled Inexact Newton infrastructure. In this infrastructure, I am first porting the self-gravity problem, so that I can both work out any bugs in the solvers, and to provide a clean, simple example for future implicit solver development within Enzo (such as FLD).

In addition to the above software development, I traveled to LLNL to visit with CASC members regarding the Enzo/LLNL collaboration (visit with Frank Graziani et al. was canceled, due to higher priority visitors to their division). While there, I discussed the equation formulation and time-stepping strategies with Louis Howell, and I discussed multi-level linear solver infrastructure/algorithms/capabilities currently implemented in HYPRE with Rob Falgout and Barry Lee. I also discussed the possible nonlinear solver requirements/approaches with Carol Woodward (SUNDIALS code). While there, I attempted to generate collaboration interest in CASC members for Enzo/LLNL collaboration, and set up preliminary topics of discussion for the upcoming visit with Hayes and Bordner.

In June, I plan to finish the self-gravity incorporation within the new implicit solver infrastructure, which will allow for fast development of single-group, unigrid FLD module. Initial estimates are that the full nonlinear, single-group unigrid FLD solver will require approximately 3 weeks of development once the infrastructure is complete and tested. Of course, this development will be put slightly on hold during the visit to LLNL with Bordner and Hayes.