fenics team mailing list archive

["Johan Hoffman" <jhoffman@xxxxxxxxxx>]

> On Thu, Apr 26, 2007 at 10:32:38AM +0200, Johan Hoffman wrote:
>> > On Wed, Apr 25, 2007 at 08:27:40PM +0200, Johan Hoffman wrote:
>> >
>> >> Yes, boosting interest is good, but why can't the two be combined? I
>> >> would prefer a single project dolfin-modules containing a collection
>> of
>> >> DOLFIN solvers, including your solvers and others.
>> >> >
>> >> > /Anders
>> >> >
>> >>
>> >> I do not understand. What other solvers are you referring to?
>> >> I talk about to constitute our new project is the Navier-Stokes/Euler
>> >> solvers (compressible + incompressible, the ALE-FSI solver, and the
>> >> elasticity solver. In the list at
>> >> www.fenics.org/wiki/DOLFIN_solver_modules, I can only see Stokes,
>> >> Poisson,
>> >> heat eqn. and convection-diffusion, apart from these. Are these the
>> >> solvers you refer to? I consider these solvers rather being simple
>> >> demos,
>> >> to either stay in the Dolfin-kernel for demo/test, or to be dropped,
>> not
>> >> being part of a solver project for advanced applications.
>> >>
>> >> If the motive of such a dolfin-module project would only be a
>> showcase
>> >> of
>> >> a collection of Dolfin solvers, I think we might as well host a
>> webpage
>> >> at
>> >> the wiki with links to download the solvers from developers together
>> >> with
>> >> info on what version of Dolfin it works for, and couple this to the
>> >> gallery. This would serve the purpose and would minimize maintainment
>> >> work. And here our new project would also be present for showcasing
>> >> Dolfin.
>> >>
>> >> The project I suggest is not just a showcase of Dolfin solvers, it is
>> >> aimed at developing solvers for automating/generalizing modelling for
>> >> advanced applications in different areas. For starters, our NS-solver
>> >> for
>> >> turbulent flow would fit here as an automated turbulence solver,
>> using
>> >> one
>> >> simple method (G2 including adaptivity) for "all" turbulent flows
>> >> (contrary to common practise of using specific turbulence models for
>> >> different flows). We are now working on generalizing this solver to
>> >> compressible flow and fluid-structure interaction. Thus there would
>> then
>> >> be one solver for turbulent fluid-structure interaction, which would
>> >> correspond to automated modeling of FSI.
>> >>
>> >> So far we have no FEniCS project on advanced solvers or automation of
>> >> modeling, so this project would fill that gap, and would not
>> interfere
>> >> with any other FEniCS project as far as I can see.
>> >
>> > True, but why must the set of solvers be limited to NSE with cG(1)?
>>
>> It is not, initially it would include incompressible/compressible flow,
>> fluid-structure interaction, including visco-elasto-plastic material and
>> contact. I am not talking about cG(1), I am talking about G2 which
>> refers
>> to Galerkins method together with L2-control of part of the residual
>> (that
>> is, it includes about all FEM, FDM, FVM methods) together with a
>> posteriori error adaptivity. Which is not that restricted.
>>
>> > Examples of other solvers that would be nice to see in a
>> > dolfin-modules collection would be the plasticity solver and the
>> > other modules we just removed along with the NSE solver. Indeed some
>> > of these were very simple, but they don't need to be. The goal has
>> > always been that the modules would offer more than the demos. Just
>> > having an adaptive and very efficient Poisson solver in such a
>> > collection of modules would be very useful.
>>
>> They are included, see above.
>
> Is the DOLFIN plasticity solver (that Garth and Kristian wrote)
> included? And Poisson?
>
>> >> If the name is disturbing, signaling exclusivity, we can use another
>> >> name,
>> >> not containing FEniCS or Dolfin. If then someone else want to publish
>> an
>> >> advanced solver project at FEniCS that is of course also very
>> welcome.
>> >> So
>> >> far I have heard nothing about such a project.
>> >>
>> >> /Johan
>> >
>> > I don't mind the name, I just think here is a chance to take the
>> > DOLFIN modules and make something very good and useful of them, which
>> > would be to offer a collection of very efficient, adaptive and
>> > polished solvers for a range of equations.
>>
>> We have indeed the goal to make this good and useful, and adaptive. And
>> sooner or later it will be polished as well.
>>
>> I do not see any argument against what we are proposing.
>>
>> We could of course publish our solver somewhere else, but that would
>> make
>> little sense. I expect this project to be of much benefit for FEniCS,
>> and
>> by publishing our development at www.fenics.org we invite to a
>> collaborative effort.
>>
>> /Johan
>
> I just suggest that you do exactly what you intend + a little more,
> which would be to maintain the entire collection of DOLFIN solvers,
> not just your own modules.

This is not in line with the project we propose, since we are aiming for
as few solvers as possible, ideally just one, applicable to as many
applications as possible. Whereas the dolfin_module project you propose
rather aims for many solvers for a set of chosen PDEs.

We push for our project because this is in line with what we want to
achieve, and as much as I would have liked consensus I do not see any
strong reason not to launch such a project based on the discussion here,
or the parallel discussions off the fenics-dev list. Therefore we will
launch our project as soon as we are ready. About the name, it appears
that it may be best to go with a neutral name, not referring to FEniCS or
Dolfin, to open for similar efforts that are very welcome. Let us remember
that the existing Ko project is of this nature, and the present project
may be seen as an extension of that one. One main reason for publishing
the project at www.fenics.org is to open for a collaborative effort, so
naturally everyone interested is welcome to join.

As for the dolfin_modules, I think we should go with what Garth suggested
on the dolfin-dev list: set up a DOLFIN_MODULES wikipage with links to
dolfin modules that are actively devloped and maintained (including the
new project). But this discussion rather belongs to the dolfin-dev list I
guess.

/Johan


>
> /Anders
> _______________________________________________
> FEniCS-dev mailing list
> FEniCS-dev@xxxxxxxxxx
> http://www.fenics.org/mailman/listinfo/fenics-dev
>