← Back to team overview

fenics team mailing list archive

Re: Application domains

 

On Tue, Sep 27, 2005 at 12:28:05PM -0400, tomtzigt wrote:
> 
> Johan:
> 
>   I love your concept of 'hiding' the complexity of the FEM. In this
> context, what in your mind would be required to hook Fenics up in an
> environment like FEMLAB where you would be able to write the equations,
> handle the boundary conditions, and automate design studies? In my taste,
> there are a little bit too many clicks in FEMLAB to set parameters of the
> equations but it does bundle enough components to create a usable FEM
> environment for normal people.
> 
> Theo
> 

We already have what FEMLAB has sans the GUI in DOLFIN (this also
applies to Sundance). I haven't used FEMLAB much, but from what I've
tested and seen, you select a pre-defined equation from a library and
then specify coefficients. DOLFIN has modules which correspond to an
equation, and where you can set coefficients. What is missing is some
polish (GUI, more coefficients, more equations).

Fenics can go much further though. Instead of selecting an equation,
you can simply write an arbitrary equation yourself, with any
coefficients you like. Fenics then generates the assembly routine for
you. This already exists in Fenics, in the FIAT/FFC/DOLFIN
chain. Sundance also does this, but generates the assembly routine at
run-time, where FFC generates code which needs to be compiled (and
will likely always give higher performance).

Again, to challenge FEMLAB, some polish is needed (GUI, some system
for JIT, i.e. the assembly code needs to be compiled and loaded at
run-time).

What would be needed is also a way to generate a solver. If an
equation is pre-defined, then the solver can also be pre-defined to go
with that equation. But if an arbitrary equation is input, then the
solver will look different depending on if the problem is
time-dependent, non-linear etc. I made a suggestion how to handle that
in an automatic way in a previous post, I don't think it should be too
hard to make it work.

  Johan



Follow ups

References