Saturday, September 24, 2011

Are Neutrinos Superluminal? Judge for Yourself.

Yesterday, I couldn't restrain the physicist inside me. I just had to watch the seminar from CERN where Dario Autiero presented the result from CERN and CNGS of a measurement of the neutrino's velocity. The measurement is a tour de force of modern experimental physics, which harnesses amazing technologies such as the global positioning system, large scale data processing, picosecond lasers, ultrafast digital electronics, billion-volt particle beams, and highway tunnels a mile beneath a mountain in Italy. The scientific communication system is also state-of-the-art. The preprint (with 174 authors) appeared in on thursday night: "Measurement of the neutrino velocity with the OPERA detector in the CNGS beam",  arXiv:1109.4897v1 [hep-ex]. And CERN webcasted the presentation live around the world, with not even a hiccup in the video feed.

What the news reports have failed to convey is that despite the impressive effort, the accuracy of the velocity measurement is teased out of the data with statistical procedures that are sure to come under intense scrutiny.

The way the experiment works is this. Pulses of neutrinos lasting 10.5 microseconds are generated by the accelerator at CERN, pointed at neutrino detectors 730km away under the mountain in Italy. Neutrinos are incredibly hard to detect (they have no difficulty traveling through 450 miles of rock), so only a tiny fraction of them are detected. Over 3 years, 16,111 of the CERN neutrinos were detected in Italy. The shape and timing of each generated pulse is measured and stored to be compared later with the timing of the detected neutrinos. The nub of the matter is shown in this graph, which shows only the leading and trailing edges of the accumulated data:
As you can see, the leading edge of the neutrino pulse is about 500 nsec wide. The red line is the cumulative shape of the generated pulse, the data show the counts and relative timing of the detected neutrinos.

The evidence for superluminal neutrinos is that the red curves at on the bottom, shifted by 60.7 nsec faster than the speed of light, are a better fit to the data than the red curves at the top. The claim is made that the bottom fit is 6 sigma away from the fit at the top. What do you think? Isn't physics fun?
Enhanced by Zemanta


  1. The six-sigma claim is for the whole pulse, not just the leading and trailing edges. Although the edges might unduly influence the fit, there is some detail in-between that might fit better with the 60.7 ns extra shift. Unfortunately, the preprint only shows the full pulse data to a resolution of 150 ns, so it's anyone's guess.

    Also worth noting is that the six-sigma claim seems to depend on an assumption that may or may not be valid: that the shape of the neutrino pulse is exactly the same as the shape of the proton pulse, except for a constant vertical scale factor (of about 1:6,000,000,000,000,000,000).

    As far as I can tell, the statistical modeling assumed there was only one unknown parameter - the time shift. The probability that a proton resulted in a neutrino detection was assumed to be constant across the pulse. If it were higher at the beginning of the pulse than at the end, the best fit might have had a smaller time shift, or even if not, the confidence interval for the time shift might have been wider. I say more about that here:

  2. Steve, one of the questions asked after the presentation concerned the beam profile. The issue is that the proton pulse is measured at the source; the detector samples only a small area of the 2.8 km wide neutrino beam. One possibility is that the the timing of the pulse varies across the beam profile- imagine sweeping a flashlight across an aperture. In response, Autiero remarked that the proton beam profile was remarkably stable and well characterized.