[firedrake] Problem with Jacobian

[Buesing, Henrik]

> On 06/11/15 14:23, Buesing, Henrik wrote:
> >> On 06/11/15 12:40, Buesing, Henrik wrote:
> >>>>> Are you sure the problem isn't that he needs bounds for his solve?
> >>>>> AFAICS dS/dh=0 for h<hw or h>hn. Otherwise shouldn't rewriting
> >>>>> this with conditionals have worked?
> >>>
> >>> @Stephan: So how can I rewrite the attached saturation.py with
> >> conditionals?
> >>
> >> Sw=conditional(hin>hn, Constant(0.0), conditional(hin<hw,
> >> Constant(1.0),
> >> (hin-hn)/(hw-hn)))
> >>
> >> This gives you a UFL expression that you can use in a larger system,
> >> or project to some function space.
> >>
> >
> > @Stephan: Thank you! Is there also a way to write the attached
> enthn.py in UFL notation? Basically I failed to do the int(dm), if dm is
> a UFL expression. And probably the array access might be a problem?
> 
> Yes, looks like a piecewise linear interpolation from an arbitrary
> lookup table? 

Yes, exactly!

> I don't see how to do this in UFL indeed. You could
> approximate it with a polynomial but that might not be accurate enough
> and have nasty under/over shoots if the data is not smooth.
> 
> So the best way to do this might indeed be to keep your point-wise
> evaluation and add a hook for the evaluation of its derivative

Ok. I will see if I get the "coefficient_derivatives" argument to work. Else I just might pass the whole Jacobian. 
Thank you!

Henrik