Week 6 update: double peaking mystery

A while back, one of my friends observed that there is a finite amount of time in which an undergrad is allowed to work at JLab before everybody else realizes that they’re there and subsequently approaches them to work on their own parts of the experiment. For me, that amount of time was 5 weeks. At the beginning of my sixth week, one of the postdocs asked me to look at some data that just didn’t look right. Generally, when we look at the data we get from the beam current monitors and their differences, we expect Gaussian distributions that look something like this:

Screen Shot 2019-08-05 at 1.13.41 PM

Instead, for some of the runs, we were getting double peaks like this:

Screen Shot 2019-08-05 at 1.17.48 PM

We weren’t sure if there was something wrong with our beam current monitors or if the current actually was changing in unexpected ways throughout the run. The success of the experiment depends on us keeping the beam as steady as possible and knowing exactly how it does change, so it was absolutely essential that we figure this out. As I’ve mentioned, the experiment works by very quickly flipping the beam fromĀ spin-up to spin-down and back. At first, it might make sense to just switch back and forth in a 01010101 pattern, where 0 and 1 represent opposing spin states. However, any sort of predictable pattern means that statistical biasĀ could creep into the measurements and affect the final result. Instead, we have set two possible blocks: 0110 and 1001, where the order of the blocks is randomized. That way, even if the previous or future spin state has an effect on the measurement of the current state, the effects should cancel each other out over time.

One possible explanation for the double peaking was that the previous spin state was having a larger effect than we expected. We hypothesized that one of the two peaks could represent all measurements where the previous spin was 0, while the other peak showed events with previous spin of 1. To test that, I took the data and broke it into 4 chunks: events where the previous spin was 0 and the current was 0, previous 0 and current 1, previous 1 and current 0, and previous 1 and current 1 (in short, 00, 01, 10, and 11). I plotted all four chunks and got plots that looked like this:

Screen Shot 2019-08-05 at 1.56.03 PM

All 4 of the chunks showed the double peaking, which indicated to us that our hypothesis was incorrect. The combinations of previous and current spin did not skew the data one way or the other.