I made a plaster cast of the cavity to measure the actual inner diameter of the cavity at the small plate. I had trouble getting the plaster out. I tried chipping it out and heating it for 15 minutes at 450o in a toaster and quenching it to shock it out. It did not come out until I heated it with the oxyacetylene torch. I took care not to anneal it by accident, but the flange is more bendable now. The finish is ruined.

The measured OD of the mandrel is 73.20 mm at the small plate. The measured OD of the plaster cast is 73.90 mm. This implies I lost approx. 0.70 mm between forming imperfections and polishing. This is an interesting benchmark for future fabrication.

I am interested that the diameter at the small plate is lower than dc @ the lower resonant frequency, 2351.999 MHz. This should have cut off that frequency. Feynman II suggests that, after the cutoff diameter, the signal does not die immediately, but dies out exponentially quickly so that it is a source of losses and is essentially gone within a distance of a (the radius). The chamber may hhave been past cut-off for such little distance that it wasn't visibly affected, but I would have expected it to affect the peak more than it seems to. This may speak to my theory that the cutoff frequency equation may be affected by the slope of the walls of the chamber.

I was thinking last night about the use of Raspberry Pis in a Hadoop cluster and I found that it is "difficult" to translate FFT to a MapReduce paradigm. However, I read more about DFT and found that it is possible to see whether a specific frequency is present in a signal with a technique called "correlation." This technique translates very well to MapReduce. Considering tracking the resonant frequency and feature recognition to determine if the resonant frequency is near the fc of the small plate radius (especially as the small plate expands in the heat) is essentially a search of the frequency domain, and a full characterization of the frequency domain probably isn't necessary, this could be a useful insight. Large transistor, slow speed processors running in a distributed MapReduce fashion may overcome processing limitations we could face.