Leaving aside any ffc issues, this formulation is a problem mathematically -- second derivatives of piecewise Lagrange polynomials don't make sense because they are only C^0. There are several workarounds: As you've observed, Hermite or other non-Lagrange elements are needed to resolve the higher derivatives in plate bending, biharmonic, Cahn-Hilliard, and other 4th order problems.
- Hermite is a nonconforming C^1 (it's really C^1 in 1d, but not on triangles or tets). - The Morley triangle is also a suitable nonconforming element - Argyris is a fully C^1 triangle The bug problem with all these element is that we don't support them yet. They are on the to-do list, but will require some modifications to ffc because of technicalities in how they transform from a reference element. Anders and I have a plan :) - You can also use a mixed formulation, writing the 4th order problem as a system of 2nd order equations. In strong form, Delta Delta u = f --> v = Delta u Delta v = f Delta is the Laplacian, so you can multiply each equation by a test function and integrate by parts to get a weak form where you never need more than one derivative - You can also use the "continuous/discontinuous Galerkin method, where you add penalty terms on the first derivatives. Garth: you've used the C/DG method, can you contribute an example? On Thu, Jun 25, 2009 at 5:08 PM, <n...@ptmat.fc.ul.pt> wrote: > First I have to say that I'm not sure that this should work... > > 1- Probably I'm missing/ignoring something (mathematically speaking) > 2- I didn't find yet, in the literature, a simple example > (but I've started this week :) ) > If anyone knows such an example, > I would by happy to make a demo with it. > 3- all the cases I've seen are related with plate bending and requires > hermite elements or some other "non-standard elements" > 4-Concerning the code: I'm not sure on the Boundary conditions syntax. > > I was thinking in a simple "4th order like poisson " > for instance: > > -lap lap u(x,y)=-sin(x) in the [0,2pi]x[0,2pi] > lap u= -sin(x) and grad(u).n=(cos(x),0).n in the Boundary > > with solution u(x,y)=sin(x) > > LapLap.ufl > ------------------------------------------------------------------------ > element = FiniteElement("Lagrange", triangle, 2) > v = TestFunction(element) > u = TrialFunction(element) > f = Function(element) > g = Function(element) > n = VectorConstant("triangle") > a = -div(grad(v))*div(grad(u))*dx > L = v*f*dx-f*inner(grad(v),n)*ds+div(grad(g*n[0]))*v*ds > ---------------------------------------------------------------------- > the main.cpp is modified from the poisson demo and follows in attachment. > (I'm using 0.9.2 dolfin version) > It really fails. > > > This should work, but I don't know of a demo testing it. > > > > See if you could create a very simple test case that fails. > > -- > Nuno David Lopes > > e-mail:n...@ptmat.fc.ul.pt <e-mail%3a...@ptmat.fc.ul.pt> (FCUL/CMAF) > nlo...@dec.isel.ipl.pt (ISEL) > http://ptmat.ptmat.fc.ul.pt/%7Endl/ > > Thu Jun 25 22:19:44 WEST 2009 > > > _______________________________________________ > FFC-dev mailing list > ffc-...@fenics.org > http://www.fenics.org/mailman/listinfo/ffc-dev > >
_______________________________________________ DOLFIN-dev mailing list DOLFIN-dev@fenics.org http://www.fenics.org/mailman/listinfo/dolfin-dev