2.13 Tips for troubleshooting problems

When CAMFR produces results that are clearly nonphysical, here are number of things you can do to improve convergence:

• If CAMFR is missing modes, increase the precision with set_precision(p), where p is an integer number that defaults to 100. If you expect missing radiation modes, increase set_precision_rad() beyond the default of 100.
• Increase or decrease the absorption in the PML, the real distance between the metal walls and the number of modes. Using too few modes will cause convergence problems. A PML absorption that is too low will not damp out all parasitic reflections. On the other hand, if the absorption is excessively high, some modes can be missed. Increasing the real distance between the walls and lowering the PML absorption is a viable option in this case, but also requires that you use more modes.
• If the PML absorption is high, try set_chunk_tracing(0) at the expense of longer run times.
• If there are problems due to inverting numerically unstable matrices, you can change the way these situations are handled. Use set_stability(extra) to use extra row and column equilibration. With set_stability(SVD) a pseudo inverse is calculated using singular value decomposition.
• If your waveguides don't have multiple cores, try using set_degenerate(0).
• If the field profiles seem to be calculated incorrectly, try set_orthogonal(0). This will cause CAMFR to treat modes which are not exactly orthogonal due to rounding errors as non-orthogonal.
• If you are dealing with very lossy materials, using a different solver might help: set_solver(series) and set_mode_surplus(n). This will first construct an initial estimate of the modes based on a plane wave expansion. The number of plane waves used in this expansion is n times get_N(). These estimates are subsequently refined using the full dispersion relation.
• For trouble with uncoupled waveguides, try experimenting with set_unstable_exp_threshold. E.g. setting it to 1e-6 will mean waveguides will be treated sooner as decoupled than when the parameter was 1e-12.
• If you're working with circular structures in the regime of backward and complex modes (e.g. when the metal wall is close to the last interface), you can use set_backward_modes(1) for extra stability.

If CAMFR crashes, make sure you haven't called free_tmps() too soon, i.e. before you're done with the objects. Another common cause of crashes to look out for is defining a wall or an InfStack inside an expression. Also, if you're upgrading from a pre 1.0 version of CAMFR, make sure all your coordinates are purely real.

Finally, we want to remind the user that in Python 1/3 = 0 rather than .3333. This will change in Python 3.0, but if you already want such behaviour now, include from __future__ import division at the top of your scripts.

This document was generated by root on August, 27 2008 using texi2html 1.76.