2
$\begingroup$

What is the best way to impose a "constant constraint" for a PDE? Specifically, I want to solve an eigenvalue problem $Au=\lambda u$ on the rectangle $(0,2\pi)\times(-\pi/2,\pi/2)$ with periodicity constraints horizontally and constant constraint on the lines $y=\pi/2, y=-\pi/2$. Here a constant constraint means that $u(x,\pi/2)$ is constant and the constant is not fixed. The same in $-\pi/2$ with possibly a different constant. I use a finite difference grid.

Is there any way to impose the constant constraint in the matrix $M$ which is the discretized of $A$?

$\endgroup$
4
$\begingroup$

Let's say you have numbered your vertices along the line $y=-\pi/2$ as $u_1,u_2,\ldots,u_n$ (so that you'll have a total of $N=nm$ vertices, with $m$ rows of vertices covering your domain). Then your constraint can be written as $$ u_2=u_1, \\ u_3=u_1, \\ \vdots \\ u_n=u_1. $$ (And the same, obviously, for the vertices at the upper boundary.) This can be written as a homogenous constraint $CU=0$ where $C$ is a rectangular matrix where the constraints above will yield $m-1$ rows and $N$ columns.

Solving linear systems $AU=F$ where $CU=0$ is exactly the kind of operation we always do when we deal with hanging nodes, for example. Typically, we would eliminate $u_2, \ldots, u_n$ from the linear system.

You can find a number of descriptions of this in the literature. The earliest, I believe, is in Babuska and Reinboldt from the 1970s. I'm quite sure it's in Graham Carey "Computational Grids" book. I know it's in mine and Oliver Kayser-Herold's 2009 paper on hp, see # 27 at http://www.math.tamu.edu/~bangerth/publications.html#x-reviewed .

| cite | improve this answer | |
$\endgroup$
2
$\begingroup$

$$ \left.\frac{\partial u}{\partial x}\right|_{y=\pi/2} = 0 $$ would be one strategy, though it's a little unusual to impose a tangential derivative boundary condition, and your system will likely be under-constrained if you don't fix the value of $u$ at at least one point.

If your $A$ is a decent operator, then your combination of periodic and fixed constant conditions will likely remove the $y$ dimension from the problem all together, allowing you to solve a 1-D problem instead. Can you say a little more about what $A$ is?

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ $A$ is the Laplace-Beltrami operator on the sphere written in polar coordinates. My goal is to compute the Laplace-Beltrami eigenvalues of a piece of a sphere $\Omega$ by considering the auxiliary problem $-\Delta_{LB} u + \mu u = \lambda u$ where $\mu$ is $+\infty$ on the complementary of $\Omega$ (numerically, something large). This works reasonably well in 2D on a rectangle, so I thought if I could "flatten" the sphere to a rectangle I could get it to work like this. $\endgroup$ – Beni Bogosel Dec 4 '13 at 11:27
  • $\begingroup$ You said $x$ and $y$ in your original question. Are you including the curvature terms in your operator, or are your variables really $\theta$ and $\phi$? I suspect that if you analyze your problem from a theoretical perspective, you'll find that it's 1D or quasi-1D and amenable to analytic solution. $\endgroup$ – Bill Barth Dec 4 '13 at 22:38
  • $\begingroup$ Yes, I included the boundary terms in the operator, and the variables are the angles $\theta,\phi$. I will look into what you are saying. $\endgroup$ – Beni Bogosel Dec 4 '13 at 23:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.