dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Sat, Feb 28, 2009 at 10:40:24AM +0200, Evan Lezar wrote:
> 
> 
> On Fri, Feb 27, 2009 at 5:20 PM, Garth N. Wells <gnw20@xxxxxxxxx> wrote:
> 
> 
> 
>     Martin Sandve Aln s wrote:
> 
>         On Fri, Feb 27, 2009 at 4:08 PM, Garth N. Wells <gnw20@xxxxxxxxx>
>         wrote:
> 
> 
>             Martin Sandve Aln s wrote:
> 
>                 On Fri, Feb 27, 2009 at 4:03 PM, Garth N. Wells <
>                 gnw20@xxxxxxxxx> wrote:
> 
>                     Evan Lezar wrote:
> 
>                         Hi
> 
>                         I am working on some electromagnetic problems that
>                         require I model the
>                         electric field as a complex phasor.  When compiling
>                         user-defined
>                         functions I get an error message.
> 
>                         For example, consider the following simple function:
> 
>                         f = Function(V, '1.0j')
> 
> 
>                         when calling the JIT compiler, the log shows the
>                         following error (in the
>                         last couple of lines):
> 
>                         dolfin_compile_function_934f16b21b1210d9244eba2ba7985a9a_wrap.cxx:
>                         In
>                         member function  virtual void
>                         dolfin::function_91ee55df622466fafbf3ba887ffb745e::eval
>                         (double*, const
>                         double*) const :
>                         dolfin_compile_function_934f16b21b1210d9244eba2ba7985a9a_wrap.cxx:3221:
>                         error: cannot convert  double __complex__  to  double 
>                         in assignment
>                         error: command 'gcc' failed with exit status 1
> 
>                         Does dolfin support complex valued functions?
> 
>                     No.
> 
>                     If so, how do I go about
> 
>                         getting them working?  And if not, how do I go about
>                         adding the
>                         functionality.
> 
> 
>                     It would be a big task to make it seamless. The starting
>                     point would be
>                     having FFC (or soon UFL) support complex functions. I
>                     thought about this
>                     a few years ago but never got around to it.
> 
> 
>             Thinking about, it shouldn't be too hard to patch something
>             together using
>             mixed elements.
> 
> 
>         Won't be as user-friendly as it could be though, but perhaps you
>         could define e.g. complex multiplication as python functions on top.
> 
> 
> 
>     Agree, it wouldn't be user friendly. Making it user friendly would require
>     a lot of work on UFL/FFC/DOLFIN.
> 
>     Garth
> 
> 
> 
> 
> 
> 
>                     UFL question: Does or will UFL support complex functions?
> 
>                 No. And I don't think making such an addition before it has
>                 stabilised
>                 is an option, we can't change everything everywhere
>                 simultaneously.
> 
> 
>             OK. Would be good for the long-term goals list.
> 
> 
>         Sure. I'll add it.
> 
>         Martin
> 
> 
> 
> 
> 
> Thanks for the feedback.  I have got some ideas for an interim solution, will
> give them a go and let you know what I come up with.
> 
> Evan

I solved a complex-valued Stokes problem a while back. I suggest just
writing out your equation in terms of real and imaginary parts and
separate it as a system of two real-valued equations.

Here's the code I used:

  P2 = VectorElement("Lagrange", "triangle", 2)
  P1 = FiniteElement("Lagrange", "triangle", 1)
  TH = MixedElement([P2, P2, P1, P1])

  (vr, vi, qr, qi) = TestFunctions(TH)
  (wr, wi, pr, pi) = TrialFunctions(TH)

  fr = Function(P2)
  fi = Function(P2)

  mu = Constant("triangle")
  rho = Constant("triangle")

  a = (-dot(vr, wr) - mu*dot(grad(vr), grad(wi)) -
      (1/rho)*div(vr)*pr)*dx + \
      (-dot(vi, wi) + mu*dot(grad(vi), grad(wr)) -
      (1/rho)*div(vi)*pi)*dx + \
      qr*div(wr)*dx + \
      qi*div(wi)*dx

  L = dot(vr, fr)*dx + dot(vi, fi)*dx

For details, see

  http://home.simula.no/~logg/pub/papers/SikJen2008.pdf

-- 
Anders

Attachment: signature.asc
Description: Digital signature