S1600525 has been coated in Fort Collins with 480nm of pure tantala. I used the emasured loss angles (after deposition, before annealing) to estimate the shear and bulk loss angles.
Model
First, my COMSOL simulation shows that even if I don’t include the drumlike modes, I still have a significant scatter of shear/bulk energy ratio. The top panel shows indeed the ratio shear/bulk for all the modes I can measure, and the variation is quite large. So, contrary to my expectation, there is some room for fitting here. The bottom panel just shows the usual dilution factors.
Then I tried to fit the total losses in my sample (the substrate is negligible) using four different models:
1) one single loss angle for both bulk and shear, constant in frequency
2) one single loss angle for both bulk and shear, linear in frequency
3) separate bulk and shear loss angles, constant
4) separate bulk and shear loss angels, linear in frequency
Instead of using Gregg harry's technique (taking pairs of losses together), I simply fit the whole datasets with the assumptions above. I derived the 95% confidence intervals for all parameters. I also weighed each data point with the experimental uncertainty. I’m not sure yet how to compare the performance of the various models and decide which is the best one, since clearly the more parameters I plug into the model, the better the fit gets.
If I use two different loss angles, but constant, I get numbers similar to what Gregg presented at the last Amaldi conference ( G1701225), but inverted in bulk and shear. I cross checked that I didn’t do any mistake. Instead, if I allow linear dependency on frequency of bulk and shear, I get a trend similar to the one in Gregg's slides.
My plan is to have this sample annealed today or tomorrow and measure it again before the end of the week.
Results
One loss angle  constant
One loss angle  linear in frequency
Bulk and shear  constant
Bulk and shear  linear in frequency
