Solutions to the wave equation on non orientable surfaces like a mobius strip

Given a mobius strip, what do the solutions of the wave equation look like qualitatively? How do they differ from solutions on the equivalent strip glued together as a cylinder? Any refs, particularly to symmetry?

-
What are the boundary conditions? –  S. Carnahan Jun 23 '10 at 1:00
Well, say I'm trying to excite it with vibrations near the fundamental frequency of the corresponding orientable cylindrical surface. Is the resonant frequency (eigenvalue) of the mobius strip shifted? I know this doesn't exactly answer your question, but you see what I'm imagining. –  sigoldberg1 Jun 23 '10 at 1:09
I guess to more specifically respond, the boundary conditions are free, with forcing at theta = 0, on a line orthogonal to the midline circle. –  sigoldberg1 Jun 23 '10 at 1:19
Force is harmonic (sinusoidal), to be applied normal to the surface. –  sigoldberg1 Jun 23 '10 at 1:24

Any solution to the wave equation on a Möbius strip lifts to a solution on its orientation double cover, which is a cylinder of equal width but twice the circumference. In order for a solution on the cylinder to descend to the Möbius strip, it is necessary and sufficient that it be invariant under a certain order two symmetry. If the cylinder has coordinates given by $[0,\pi a]$ in the free direction and $[0, 2\pi L]$ in the loop direction, then the solutions are linear combinations of products $AB$, where $A$ has the form $\cos (kn_a t)\cos(\frac{n_a}a x)$ or $\sin (kn_a t)\cos(\frac{n_a}a x)$, and $B$ has the form $\cos(kn_Lt)\cos(\frac{n_L}{L}y)$, $\sin(kn_Lt)\cos(\frac{n_L}{L}y)$, $\cos(kn_Lt)\sin(\frac{n_L}{L}y)$, or $\sin(kn_Lt)\sin(\frac{n_L}{L}y)$. Here, $n_a$ and $n_L$ are nonnegative integers, and $k$ is a constant. Invariance under the symmetry is equivalent to $n_a + n_L$ being an even number. In contrast, solutions on the cylinder of the same dimensions correspond to solutions on the double cover such that only $n_L$ is even.
The order two symmetry is defined by $(x,y) \mapsto (\pi a - x, \pi L + y)$, where $x$ is the coordinate in the free direction, $y$ is the coordinate in the loop direction, and the $y$ coordinates are considered modulo $2 \pi L$. The proof of invariance comes from examining the eigenfunctions I listed above, and seeing which ones are unchanged under the transformation. This reduces to the fact that cosine and sine get multiplied by $-1$ when the domain is translated by $\pi$. I do not know a reference. –  S. Carnahan Jun 29 '10 at 21:38