On 9 Aug 2016, at 13:31, Gregory, Alastair C A <a.gregory14@imperial.ac.uk> wrote:
Hi Onno,
"- has there been an update with a gmsh example in which solid walls and inflow/outflow/specified flow conditions have been imposed at the boundaries"
There has not as of yet. The current set-up conditions are only for firedrake Mesh's and for reflective boundary conditions off of solid walls.
"has the time step been made adaptable based on flood/signal speeds"
Again not as yet, this will definitely be done though. At the moment, we specify a value of dt/dx and this dictates the timestep. This is an argument labelled as Courant in Timestepper class. See help(Timestepper).
"I tried daybreak over uniform beach and lowered the time step (adaptable one preferred) but am not sure whether the slope limiter causes some spurious oscillations. Does the Ern slope limiter approach add an extra time step restriction above and beyond the normal CFL* min_dx/|lambda| with CFL<1 and lambda = |u|+ sort (g h)?"
So this was a bug that came about from not slope limiting the momentum as well as the depth. This is in the latest pull request of the package but hasn't yet been accepted to merge. If you want to try this out, you may just copy straight off of the branch in which this update is on. There isn't an extra condition as I know of.
"Also the speed seems a bit low although I aim to run it against a 2D c-code I once made."
The speed is slow for 2D but this is as fast as I can get it for the time being. I will continue to look further into why this could be.
I suspect that one thing that may be going on is that most of your integrands are non-polynomial. UFL attempts to integrate your forms exactly using numerical quadrature. Since the basis is polynomial this is not possible for your integrands. In this case, UFL resorts to heuristics to come up with a quadrature rule that is "exact enough". These heuristics are rather conservative, and can therefore result in excessively high degree quadrature rules. Probably for convergence you only care about integrating the mass matrix exactly. As such, you should probably specify your quadrature degree "by hand". You do this by __call__ing the measure objects passing in the keyword argument degree. For example: Rather than writing: dot(v, boundary_flux)*ds you write: dot(v, boundary_flux)*ds(degree=4) # Integrate degree 4 polynomials exactly The advice I gave to Justin earlier today about profiling using PETSc tools applies here too. Cheers, Lawrence