Total number of participants: 23
Total number of gueste outside of CBC: 5
Number of different nationalities represented: 6
Total number of speakers: 3
Total number of talks: 6

Program

• Monday (August 4, 13-15)

    Overview (Matt)

    Sieve Concepts (Dmitry)

    Discussion

• Tuesday (August 5, 13-15)
  

    FEM Transformations (Rob)

    Discussion

• Wednesday (August 6, 13-15)
  

    Sieve Design (Dmitry)

    Discussion

• Thursday (August 7, 13-15)
  

    Sieve Implementation (Matt)

    FMM in Sieve (Matt)

    Discussion

• Friday (August 8, 13-15)
  

    Open

Abstracts

Programmability remains the main challenge in high performance

scientific software today. The lack of common conceptual pieces

among various methods and application domains clearly contributes

to this problem. We will present an abstract formulation of the finite

element paradigm: restrict--compute--complete, and discuss

implementations of this principle, some undergoing active research,

that conform to a common compact interface. The implementations

will be specialized to the different application domains. Not only does

this approach significantly reduce the complexity of common FEM usage,

but also the developed components can be reused in many other domains,

such as optimal direct solvers of the Fast Multipole Method type.

We discuss a conceptualization of finite element computing in the usual

terms, breaking it into locally homogeneous computational kernels, and

also the methodology for systematic assembly of local computation into

a global whole. We will discuss contemporary techniques that take advantage

of the ubiquity of the local kernels and optimize their use. We also show that

the same paradigm of global assembly of local pieces can be used to

conceptualize FEM computation, parallel data management and hierarchical

solvers.

Robert C. Kirby

Traditionally, special-purpose finite elements are created for different PDE types,

and special purpose programming implements particular basis functions for them.

However, there are mathematical foundations that unify our understanding of a

very wide class of finite elements and guide the implementation of general-purpose

code for computing all of the basis functions of any order on a reference element.

Second, finite element codes typically map these reference basis functions to each

element in a mesh to evaluate stiffness matrices and load vectors. While the classic

Lagrange element may be transformed very easily, many other elements require more

complex transformations. A new theory of transforming equivalent, interpolation

equivalent, and not interpolation equivalent elements will be presented, with applications

to scalar elements and H(div) elements as well.

Participant list

Tom David Atkinson  <tomat@simula.no>  CBC@SIMULA

Joakim Sundnes  <sundnes@simula.no> CBC@SIMULA

Robert Artebrant  <ra@simula.no>  CBC@SIMULA

Oddrun Christin Myklebust  <oddrun@simula.no>  CBC@SIMULA

Peter Burne  <burne@uchicagp.edu>  University of Chicago

Dmitry Karpeev  <karpeev@mcs.anl.gov>  Argonne National Laboratory

Samuel  Wall   <sam.wall@simula.no>  CBC@SIMULA

Anna Blechingberg  <annable@simula.no>  CBC@SIMULA

Kristoffer Selim <selim@simula.no>  CBC@SIMULA

Kristian Valen-Sendstad  <kvs@simula.no>  CBC@SIMULA

Martin Alnæs  <martinal@simula.no>  CBC@SIMULA

Glenn Terje Lines <glennli@simula.no>  CBC@SIMULA

Kirsten ten Tusscher  <tentusch@simula.no>  CBC@SIMULA

Wenjie Wei  <wenjie@simula.no>  CBC@SIMULA

Tomas Ruud  <tomassru@simula.no>  CBC@SIMULA

Harish Narayanan <harish@simula.no> CBC@SIMULA

Johan Hake <hake@simula.no>  CBC@SIMULA

Joachim Haga <jobh@simula.no>  CBC@SIMULA

Kent-Andre Mardal  <kent-and@simula.no> CBC@SIMULA

Robert Kirby  <robert.c.kirby@ttu.edu>  Texas Tech

 Matt Knepley   <knepley@mcs.anl.gov>  Argonne National Laboratory

Anders Logg  <logg@simula.no>  CBC@SIMULA

Marie Rognes  <meg@math.uio.no>  CMA@UIO