Searching for Unusual Hay in a Haystack: The Case of CDMS

Over the past two weeks, rumors have swirled around the web that the CDMS collaboration had discovered particles of "dark matter". [I have not yet written a promised post on dark matter, but there is this.] It all started with a single blog post which contained "facts", such as the statement that there was a paper in press at the journal Nature, which turned out to be false. One very connected person tweeted about the post, and it spread like wildfire. Soon the Nature editor sent the blogger a snarky letter denying the claim, which the blogger posted. Others speculated that the Nature editor was just trying to throw them off track. The next day the Nature editor posted a comment on the blog apologizing for the snarky nature of the letter, but again refuting the claims. Still rumors shot around the net about what result there might be.

So there was much anticipation Thursday when the CDMS collaboration gave two simultaneous talks announcing their results.

I watched a live stream of one of them. It proceeded in a halting fashion from the strain of the web traffic. Then, when the speaker got to the point of announcing their results, the stream froze for ten solid minutes. When it recovered, it zipped straight to her conclusions (how many of you were assuming the speaker was male--come on admit it), and I was left to guess a number of the details. But the bottom line is this: they saw 2 events with a background of 0.8. What does that mean, you ask?

The experiment looked for a very rare signal: that a particle of dark matter, which rarely interacts with anything, leaves a small ripple in the detector. The detector is located at the bottom of a mine to shield it from most cosmic rays. But there are still background events: interactions in the detector from particles which come from radioactive elements in the rock or particles which somehow survive going through hundreds of meters of rock. There are telltale signatures of dark matter particles (such as the energy and timing of the event) which help distinguish them from background particles, but occasionally a background particle mimics those signatures by chance. In the CDMS experiment, they calculate that over two years of running, that happened on average 0.8 times ( it took heroic efforts to keep it this small) . Maybe this helps: if they ran for 20 years with the same detector, i.e. 10 times longer, then they'd expect it to happen 8 times.

Now they saw 2 events. So what is the chance that those events are really signals of dark matter particles? Well, it is easier to ask "what is the chance they are background events?". If you ran for 20 years, what is the chance that 2 of the background events happened in the first two years. Using something called the Poisson distribution, they find that there is about a 1/4 chance those 2 events are both just background events. That's not a strong signal. As good as their efforts were at reducing backgrounds, it was not enough. If there were no dark matter particles and you ran the experiment for 20 years and divided them into ten two year periods, about two or three of those ten periods would happen to have 2 background events in them.

Still, if the events do turn out to be really from dark matter, it will begin to explain one of the great mysteries of science. So we await future experiments with more signal and less background.

UPDATE: Sometimes It's Just Hay