This is a read only copy of the old FEniCS QA forum. Please visit the new QA forum to ask questions

Negative viscosity stabalized by fourth order terms

+2 votes

I am trying to solve a "Navier-Stokes"-type problem where the viscosity is negative. Of course this renders the equation unstable and thus I add a fourth order term, so the entire equation becomes:

dv/dt + (v*nap)v = - nap p + nu * n^2 v - n2 * nap^4 v

where n < 0 and n2 > 0 and nap = del operator.

Now the fourth order term should stabalize the solution with negative viscosity - I can show this by stability analysis - but I can't get fenics to solve it. The solutions just blow up.

I followed this guide: http://fenicsproject.org/documentation/dolfin/dev/python/demo/pde/navier-stokes/python/documentation.html to get fenics to solve the basic Navier Stokes. I then added the term

n2*inner( div(grad(u0)), div(grad(v)))*dx

and I also tried with

n2*inner( div(grad(div(grad(u0)))),v)*dx

Neither works - and surprisingly they also give different results. As you can probably infer from my questions, I am new to this CFM, but would really like some help.

Thanks in advance.

asked Sep 30, 2013 by JesperJuul FEniCS Novice (440 points)

2 Answers

+1 vote

Interesting question and I only can give some hints.

Because of the pressure-velocity interaction, the time integration of the Navier-Stokes Equations has its own stability issues. So I suggest you first try your stabilization for a convection-diffusion equation for a scalar quantity $\rho$.

Then you should give some more thoughts on what is the right weak formulation of the fourth order term and what are the right finite-elements to use. E.g., if your ansatz functions are not smooth enough, you need to take a discontinuous Galerkin framework.

Regarding the weak formulation: The commonly used identity $(\nabla \cdot v, q ) = -(v,\nabla q)$, where $v$...vector and $q$...scalar, is only valid under particular conditions. In your case, the '2nd' application is not justified, because $\nabla \cdot (\nabla \rho)$ will not be zero at the boundary. That's also the reason, why you get different results depending on your formulation.

Furthermore, $\nabla v$ is actually a tensor and the relation of $\nabla \cdot (\nabla v) = \Delta v$ includes some assumptions (like $\nabla \cdot v=0$) that may not hold in a repeated application.

All in all, I do think it is a question of numerical analysis rather than of fenics implementation. If you have the equations ready, you may try to get help at http://scicomp.stackexchange.com/

answered Oct 1, 2013 by Jan FEniCS User (8,290 points)
edited Oct 1, 2013 by Jan
+2 votes

The vanilla finite element method will not provide the regularity that you need for this problem.

For some ideas on how to solve problems with fourth-order spatial derivatives, for scalar equations take a look at the demo/documented/cahn-hilliard for a operator splitting approach and at demo/documented/biharmonic for a $C^{0}$ discontinuous Galerkin (DG) method. You can find a $C^{0}$ discontinuous Galerkin for fourth-order, nonlinear elasticity equations in the supporting material to the paper http://dx.doi.org/10.1016/j.jmps.2011.04, which is at https://www.repository.cam.ac.uk/handle/1810/236975. The model in the paper can have negative stiffness, which is very similar to what you're looking at.

answered Oct 2, 2013 by Garth N. Wells FEniCS Expert (35,930 points)
...