Physics usually progresses by getting new experimental data. Given this data, we refine the theories and eventually we can come up with a picture of how the universe works. However, experimental results can be tricky to interpret. Usually, data is presented as evidence for something, but that depends many times on the model of the noise that is expected.
My most recent encounter with this aspect of physics was the recent paper on dark matter detection by the COGENT collaboration.
The paper states that they see evidence for dark matter in their results. A lot of evidence as a matter of fact. This is from a trial run on a new low noise technology before a full detector is commissioned. Being naturally somewhat skeptic, I raised my right eyebrow a bit more than usual and I hurried one floor down to the High Energy experimentalists to ask how should these new results really be interpreted: is it evidence? Or is it possible that the data reported is a bit too optimistic?
Part of the problem is that when I see the graphs, it is not obvious to me what to look for: this is mostly because I don’t usually deal with this type of data. This is when having colleagues who understand these issues can help a lot. Their expert advice really counts for something. I thought it would be a good idea to share some of this information. So let us begin with the paper in question. Being a theorist, I usually concentrate on the final answer. This is described in plots 3 and 4 of the aforementioned paper.
The figure 3 suggests that they took care of background because they have identified a lot of radioactive decays shown as peaks. They also have some excess at low energies, which is the reported excess in the experiment. The claim is that this excess (if backgrounds were taken into account properly) is due to physics other than nuclear physics and semiconductor physics.
With this data, they can produce plot 4: a likelihood analysis comparing various experiments for where dark matter is located.
This is also a tricky business. The figure four gives the data in terms of a cross section and a mass. But cross sections should depend on what material one is colliding with and are rather model dependent. The coupling can be to atomic number, or to the spin of the nucleus, or to some other combination. A standardized model of dark matter interaction with matter is used for comparing the different experiments. However, the end result can vary a lot depending on the dark matter sector and precisely how it interacts with the visible standard model. This is a topic for another discussion, but it has to be kept in mind in case dark matter has a more exotic form of interaction.
However, I was told that I was looking at the wrong plots. The right plot to look at was the one in figure 2. I reproduce it here for your visual information. I’ve downgraded the quality a bit and I suggest that you go to the original source anyhow.
The plot on the top is a determination of the background using Am decays to produce a fake signal of gammas with the right energy to simulate conditions of the walls of the cave and the inside defects of the materials making the detector. This is their calibration of data.
The plot in the bottom is the true data from the experimental run with the cut line. Above the line the events are treated as background, and below the line they are treated as potential signal.
What I was told to look at was the comparison of the Am ‘background’ data and to overlay it with the plot below. It seems that the Am data underestimates the width of the background in the vertical direction. One also sees that the data is treated very differently in the low energy bins (to the left) than in the high energy bins. As the left panel indicates, the scale is logarithmic, so there is almost a hundred-fold increase in the acceptance of events. If the background data is wider and has more events in the low energy bin section (as the Am setup suggests), the fact that the Am does not seem to capture correctly the background suggests that the low energy bin can be polluted very easily by noise. And since the acceptance is growing at low energy this can produce the spike of events seen in figure 3 of the paper.
So in the end, it seems we might not be there yet and we have to wait until the dust settles a bit more.
What is clear to me is that there is an expectation that we are very close to the point in time where dark matter will be detected directly (if the vanilla flavor model of dark matter is correct). I expect that there is going to be a lot of ‘seeing dark matter’ before it is actually seen.