Meeting
=======
date:   Fr 29.10.2004
time:   10:00
place:  MA, EPFL
participants:
  from MOX:
    Vincent Martin (VM)
    Alexandra Moura (AM)
    Martin Prosi (MP)
  from EPFL:
    Erik Burman (EB)
    Christophe Prud'homme (CP)
    Gilles Fourestey (GF)
    Christoph Winkelmann (CW)

I) Coupling
===========
Who does what, where are connections?

* multiphysics: MP, GF, CW
* multiscale: VM, AM, CV, MP
* multielement/multidiscretization: MP, VM

CV = Christian Vergara

current and planned work, as well as connections, per person, in more detail:

Martin Prosi (MP)
-----------------
current:     mass transport with given (analytic) velocity field
             * in 1 domain, with Galerkin elements
             * in 2 domains
planned:     couple to fluid solver (Navier-Stokes)
connections: needs IP stabilized NS solver, DG, Nitsche, Raviart-Thomas

Gilles Fourestey (GF)
---------------------
current:     fluid structure interaction with Steklov-Poincaré algorithm
             now there are 3 algorithms:
             * Steklov-Poincaré
             * rootfinding with Newton
             * fixed point iteration
             Steklov-Poincaré solved with nonlinear Richardson iteration
             * stepsize from Aitken
             * preconditioned with
               - solid (working)
               - fluid (in progress)
               - linear combination of solid and fluid (in progress)
planned:     * domain decomposition
             * nonlinear structure model (not necessarily LifeV)
             * instationary structure model (currently quasistationary)
             * definition of clear interfaces (drivers and solvers) in
               order to be able to couple to something else than LifeV:

                         o--------o
                         | Driver |
                         o--------o
exchange                /  LifeV   \ exchange
- displacement         /            \ - displacement
- solid stress tensor /              \ - fluid stress tensor
                     /                \
 o------------------o                  o--------------o
 | Structure Solver |                  | Fluid Solver |
 o------------------o                  o--------------o
    LifeV or other                          LifeV

connections: * domain decomposition works by VM in LifeV
             * structure works by Marcela (Module F, PVM)
             * nonlinear structure works by Miguel Fernandez in LifeV

Christoph Winkelmann (CW)
-------------------------
current:     time dependent version of IP stabilized NS solver
planned:     * two-phase flow with free surface
             * treatment of free surface by level-set method
             * IP stabilization for NS and level-set advection
connections: * MP needs IP stabilized NS solver
             * MP needs NS solver for inhomogeneous fluids
               - separate solver for nonconstant density,
                 treatment of its advection problem dependent
                 (discont. -> level set, cont. -> std. density advection)
               - keep it separate from solver with constant density (efficiency)

Vincent Martin (VM)
-------------------
current:     one dimensional blood flow models with 2 tubes
planned:     couple to three dimensional fluid structure interaction model
connections: FSI

Christophe Prud'homme (CP)
--------------------------
planned:     * require simple, clear and clean interfaces
             * externalize temporal loop (for coupling)
               -> "A solver does one timestep"

Alexandra Moura (AM)
--------------------
current:     coupling 3D NS with OD model, 2 strategies
planned:     * coupling to 1D model
             * coupling to fluid structure interaction
             * put coupling interfaces (drivers and solvers) into LifeV lib
connections: * 1D model by CV
             * FSI by GF

Christian Vergara (CV)  reported by Christophe Prud'homme (CP)
----------------------
current:     NS solver with flux BC, one flux,
             implemented using current NS solver
planned:     2 fluxes. Big issue there: flow division ratio
connections: flux BC is important for coupling to lower dimensional models,
             because they only provide mean values as boundary conditions

Christophe Prud'homme (CP) (continued)
--------------------------
planned:     * coupling infrastructure: example:

                       3D Navier-Stokes solver
                       ---->------>------>----
                      /                       \
                     / u(x)                    \ p(x)
                    /                           \
               o--------o ^                 o--------o |
               | OD->3D | | distribute      | 3D->0D | | average
               o--------o |                 o--------o v
                    \                           /
                     \ mean flux               / mean pressure
                      \                       /
                       ----<------<------<----
                              0D solver

             * externalize data: e.g. 
               - 0D network (coefficents and structure, data are available,
                 CV has them from Vuk Milisic)
               - 3D meshes

II) LifeV
=========
Goals:
* "We want to have simple code."
* Solvers should do only processing, no postprocessing
  possible solutions:
  -> provide access to data
  -> observer pattern
* Linear solvers should be called only through the encapsulated solver classes