Is there an analog of perfectly matched layers for finite element methods? References or small examples are much appreciated.
-
$\begingroup$ I have seen plenty of papers with PML for FEM. E.g., ftp.lstc.com/anonymous/outgoing/ubasu/website/papers/… $\endgroup$ – nicoguaro♦ Sep 6 '14 at 17:16
-
$\begingroup$ That's great for wave equations, but what about something like a Poisson equation? Any good references? $\endgroup$ – James Nagel Sep 23 '14 at 18:33
A perfectly matched layer (PML) is generally introduced as a continuous concept without reference to the discretization method. A damping term is multiplied by the advection operator: $$ \frac{\partial}{\partial x} \rightarrow \frac{\partial}{\partial x}\left(\frac{1}{1 + i\sigma/\omega}\right) $$ If $\sigma$ is positive, the imaginary component in the above equation drives the standard oscillatory solution of a wave equation to be exponentially decaying. A simple way of implementing this would just modify the discretized equation everywhere in space and then assign the value of $\sigma$ to be zero everywhere except near the boundaries.
So, a PML could be implemented in any volumetric discretization like finite volumes, finite elements or finite differences.
For an introduction, I like these notes. Also, try this paper.
Take a look at some of the papers by my colleague Joe Pasciak on the PML and its approximation: http://www.math.tamu.edu/~joe.pasciak/