Planning Questions

1. Desired outcomes/goals
   1. Enable easy access to mathematical software designed for vector and parallel supercomputers
   2. Assist non-math users in selecting and using high performance math software
   3. Develop and begin using evaluation framework for high performance math software
   4. Target and meet needs of specific HPCC agency customers, such as DoD HPCMP, NASA Center for Computational Sciences, DOE and NSF supercomputer centers.
2. Expected users:
   1. NASA Center for Computational Sciences (NCCS)

      http://sdcd.gsfc.nasa.gov/NCCS/

      Provides computing resources and support services for Earth and space science researchers.

   2. NSF National MetaCenter for Computational Science and Engineering

      http://www.ncsa.uiuc.edu/General/MetaCenter/MetaCenterHome.html

      Offers high performance and scalable parallel architectures for scientists and researchers to attack Grand Challenge problems.

   3. DOD High Performance Computing Modernization Program

      http://www.hpcm.dren.net/

      Common HPC Software Support Initiative (CHSSI)

      Major Shared Resource Centers (MSRCs)

   4. DOE National Energy Research Supercomputer Center

      http://www.nersc.gov/

      Principal supplier of production high-performance computing and networking services to the nationwide energy research community.

   5. HPC providers wishing to satisfy S&ECWG and NCO task force Baseline Development Environment and Priority Capabilities

      http://www.nero.net/ pancake/SSTguidelines/

      Section 3.4 Math Libraries

   6. Computational scientists and applications programmers at academic, government, and industry sites.
3. Funder expectations and success criteria

   Funders of the NHSE are funding HPC-netlib and include NASA and other HPCC agencies. Their expectation is that HPC-netlib provide an Internet resource similar to Netlib, but for high performance math software. Their success criterion is the extent to which high performance math software available from HPC-netlib actually gets used.

4. Contributors

   The main contributors will be academic and government researchers who are developing parallel numerical libraries. Their incentives to contribute are that they want their software available to as wide an audience as possible without taking on undue burden of distributing and supporting it themselves.

5. Maintainers

   The orginal authors will provide software updates.

   The GAMS classfication scheme is maintained by Ron Boisvert at NIST.

6. Relationships with other domains
   1. Numerical libraries for parallel machines perform communication through interface to underlying communications library - e.g., PVM, MPI, or vendor-specific.
   2. Many scientific and engineering applications rely on a small core of linear algebra routines.
   3. Physical simulations in a number of disciplines involve solving systems of PDEs.
   4. Large dense linear systems arise from the solution of boundary integral equations in electromagnetics (radar cross section problem - Helmholtz equation) and fluid mechanics (Laplace equation), and from quantum mechanical scattering (numerical solution of either the time dependent or time independent Schrodinger equations, or of the Lippmann-Schwinger equation). [Edelman, 1993]
   5. Computational chemists solving some form of Schrodinger's equation are the main generators of large dense eigenvalue problems (real symmetric or complex Hermitian. [Edelman, 1993]