fenics team mailing list archive

[Martin Sandve Alnæs <martinal@xxxxxxxxx>]

A note to developers on this: You can no longer use
ufc::form::num_cell_domains() etc to check if a form has integrals of that
type. Instead use ufc::form::has_cell_integrals(). See the assemblers for
how to use these, and the UFC.cpp file. I've added some inline helper
functions to UFC.h, get_cell_integral(domain), which handle proper fallback
to default and removes some boilerplate code from the assemblers.

Martin


On 30 January 2013 12:32, Anders Logg <logg@xxxxxxxxx> wrote:

> ok.
>
> --
> Anders
>
> On Wed, Jan 30, 2013 at 12:31:25PM +0100, Martin Sandve Alnæs wrote:
> >    Yes and yes. The dx() notation is just to be more explicit and
> >    distinguish it from the old dx=dx(0) notation. I might add an optional
> >    warning for uses like f*dx+g*dx(1), which could be forms intending
> >    dx(0) or dx().
> >    Martin
> >
> >    On 30 January 2013 12:15, Anders Logg <[1]logg@xxxxxxxxx> wrote:
> >
> >    On Wed, Jan 30, 2013 at 09:54:39AM +0100, Martin Sandve Alnæs wrote:
> >    >    In addition to integration over the entire domain, explained in
> >    another
> >    >    post, we have introduced a concept called 'regions' in ufl. For
> >    this
> >    >    post, assume a single mesh. Support for multiple meshes will get
> >    some
> >    >    attention in the future.
> >    >    The ufl language changes for expressing domain relations is work
> >    in
> >    >    progress, but some useful features are already possible in trunk.
> >    We
> >    >    keep the concept of disjoint numbered subdomains, which can be
> >    easily
> >    >    represented as a single meshfunction in dolfin, having a single
> >    integer
> >    >    label for each cell. In addition we introduce the concept of a
> >    region,
> >    >    which is the union of any number of subdomains. Thus regions of
> >    one
> >    >    domain can be overlapping by including some of the same
> >    subdomain(s).
> >    >    With some automatic symbolic manipulation in ufl, we can now
> >    express
> >    >    integrals over overlapping regions with zero additional cost at
> >    >    assembly time.
> >
> >      Nice!
> >
> >    >    Here's the currently simplest syntax for integrals over regions:
> >    >    # Define regions as tuples of subdomain labels
> >    >    DL, DM, DR = (1,2), (2,), (2,3) # ***
> >
> >      I guess 'DM = 2' will also work fine here? It will be passed to
> >      dx(DM)
> >      so that will result in an integral over subdomain 2.
> >
> >    >    # Define new measures associated with the interior domains
> >    >    dx = Measure("dx")[domains]
> >    >    # Make forms for equation
> >    >    a = u*v*dx() + alpha*dot(grad(u), grad(v))*dx() # Integrals over
> >    entire
> >    >    domain
> >
> >      Is the '()' necessary here in 'dx()'? Will it work with just dx?
> >
> >    >    L = f*v*dx(DR) + g*v*dx(DL) - (f+g)/2*v*dx(DM) # Integrals over
> >    regions
> >    >    For the full running code, see this demo in the latest dolfin:
> >    >
> >    >
> >
>  demo/undocumented/overlapping-regions/python/demo_overlapping-regions.p
> >    >    y
> >    >    Martin
> >
> >      > _______________________________________________
> >      > Mailing list: [2]https://launchpad.net/~fenics
> >      > Post to     : [3]fenics@xxxxxxxxxxxxxxxxxxx
> >      > Unsubscribe : [4]https://launchpad.net/~fenics
> >      > More help   : [5]https://help.launchpad.net/ListHelp
> >
> > Referenser
> >
> >    1. mailto:logg@xxxxxxxxx
> >    2. https://launchpad.net/~fenics
> >    3. mailto:fenics@xxxxxxxxxxxxxxxxxxx
> >    4. https://launchpad.net/~fenics
> >    5. https://help.launchpad.net/ListHelp
>