fenics team mailing list archive

["Johan Hoffman" <jhoffman@xxxxxxxxxx>]

> I don't mind cG(1), but this sounds like dolfin-modules with a
> restriction (only cG(1) + stabilization allowed). Why not have a
> repository dolfin-modules that may include also other modules?

Think of it more as the automation of modelling in the FEniCS visison. The
basis being a simple and general method (such as cG(1) with adaptivity).
The generality here is with respect to possible applications, not to the
number of numerical methods covered.

> If not, then where should we put those modules? As a separate project
> (dolfin-modules-but-not-cg1-with-stabilization) or should we have
> a separate project for each module?

Since it does not seem that anyone is interested in maintaining a general
collection of solvers, I think the best way is to do as we discussed on
the dolfin-dev list to have a dolfin-modules web-page under Dolfin with
links to modules by individual developers or homepages of groups that
publish Dolfin modules. This is important for visibility of Dolfin I
think.

/Johan


> /Anders
>
> On Tue, Apr 24, 2007 at 04:01:28PM +0200, Johan Hoffman wrote:
>> Hi all!
>>
>> We are now in the process of launching a new FEniCS project, focused on
>> developing a simple and general solver for advanced applications, by
>> identifying and implementing common abstractions. The new project will
>> be
>> based on our adaptive General Galerkin (G2) methodology, which we have
>> developed for incompressible/compressible laminar/turbulent flow, and
>> which we also recently at KTH have extended to fluid-structure
>> interaction
>> (FSI). Thus our initial focus is on developing a general solver for
>> computational continuum mechanics, for which we have a prototype
>> available. Later on we expect to also add electromagnetics, chemistry
>> and
>> physics.
>>
>> So far this development has taken place within Dolfin, but it has shown
>> that solver development inside Dolfin has not been optimal, and recently
>> we in the Dolfin core team decided to move the solver modules outside
>> Dolfin. Dolfin is still the PSE of FEniCS suitable for developing your
>> own
>> solver, whereas this new project is written on top of Dolfin, and is
>> focused on developing a general solver for certain application areas, to
>> start with continuum mechanics (FSI). Since both me and Johan Jansson
>> are
>> main developers of Dolfin, we expect the coupling of the 2 projects to
>> work fine, and we expect Dolfin development to benefit from the new
>> project by involving many new developers working both on Dolfin and the
>> new project. When algorithms are more suitable to implement in Dolfin
>> than
>> in the new project, we will do so. Ideally there will be few lines of
>> code
>> in the new project, taking advantage of the many general PDE tools
>> available in Dolfin-FFC-FIAT.
>>
>> The new project is not a loose collection of solvers. On the other hand,
>> our new project will focus on developing a general solver based on the
>> physics of a certain area, to start with continuum mechanics. Basic
>> ingredients of the methodology are the G2 method, adaptivity and a
>> posteriori error estimation based on duality, and ALE-methods.
>>
>> So far there has been a lot of focus on the basics of CMM, such as
>> assembling forms etc. And as much as I think the bottom-up approach is
>> important to realize the FEniCS vision, I also think it is important to
>> balance this with a top-down project of this type.
>>
>> The project will be maintained by myself and Johan Jansson, and the rest
>> of our group at KTH, but we hope to attract many other developers also
>> outside KTH. Version 0.1 can be expected to more or less consist of
>> dolfin-fsi minus dolfin-dev, together with some Ko.
>>
>> I hope and believe that this new project will generate a lot of
>> interest.
>>
>> A tentative title of this new project is: "FEniCS General Galerkin
>> solver"
>>
>> Any input is appreciated!
>>
>> /Johan
>>
>>
>>
>>
>>
>>
>> _______________________________________________
>> FEniCS-dev mailing list
>> FEniCS-dev@xxxxxxxxxx
>> http://www.fenics.org/mailman/listinfo/fenics-dev
> _______________________________________________
> FEniCS-dev mailing list
> FEniCS-dev@xxxxxxxxxx
> http://www.fenics.org/mailman/listinfo/fenics-dev
>