Now that I have had some coffee, I would like to add a few comments and corrections to my earlier post.

First, I realize that RMS isn't peak, and that the translation isn't 100%, since the RMS values are proportional to the sample values squared. An error of 1% at the peak translates to a larger error, since the error term is squared. Of course there are also errors on the rising and falling parts of the waveform as well, and they can be large (since the rate of change is higher than on the peaks), nonetheless, they contribute less to the overall value, since they are of lower amplitude.

Also, I mentioned that the sine of the sine of both 78 and 102 degrees are within 2% of the peak (> .98). It is actually 79 and 101. But to get a sample in that range, you only have to sample every 101-79 = 22 degrees. That amounts to a little over 8 samples per half-cycle.

Revisiting this issue has piqued my curiosity enough that I'm going to dig into it a bit more.