[petsc-dev] SIAM PP

[Matthew Knepley]

Since I am representing us in Lois' session, tell me if you want me to
mention anything else. This was all I could think of right now.

Title: Runtime Configurability in PETSc

Abstract:
The ability to dynamically configure a deeply nested hierarchy of objects
is a key feature in the design of PETSc. This allows the user to assemble
an optimal solver tailored to problem characteristics without changing the
application code, and also allows easy comparison among competing methods.
We will demonstrate the efficacy of this approach for both linear and
nonlinear systems. Moreover, we extend this paradigm to encompass residual
evaluation and other key simulation operations.

Outline:

  Why Configure Solvers at Runtime ?
  Why Configure Vectors & Matrices at Runtime ?
  Configuring PETSc
    Never call anything but KSP/SNESSolve()
    Examples from FieldSplit PCs
    Examples from nonlinear solvers paper
  Extending PETSc
    Adding a preconditioner
    Distributing a shared library

Its a lot like the talk I gave at SIAM CS&E but higher level.

   Matt

-- 
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20140208/8ba8f0e3/attachment.html>