Was yesterday the day when a crack appeared in the Standard Model that will lead to its demise? Maybe. It was a very interesting day, that’s for sure. [Here’s yesterday’s article on the results as they appeared.]
I find the following plot useful… it shows the results on photon pairs from ATLAS and CMS superposed for comparison. [I take only the central events from CMS because the events that have a photon in the endcap don’t show much (there are excesses and deficits in the interesting region) and because it makes the plot too cluttered; suffice it to say that the endcap photons show nothing unusual.] The challenge is that ATLAS uses a linear horizontal axis while CMS uses a logarithmic one, but in the interesting region of 600-800 GeV you can more or less line them up. Notice that CMS’s bins are narrower than ATLAS’s by a factor of 2.
Both plots definitely show a bump. The two experiments have rather similar amounts of data, so we might have hoped for something more similar in the bumps, but the number of events in each bump is small and statistical flukes can play all sorts of tricks.
Of course your eye can play tricks too. A bump of a low significance with a small number of events looks much more impressive on a logarithmic plot than a bump of equal significance with a larger number of events — so beware that bias, which makes the curves to the left of the bump appear smoother and more featureless than they actually are. [For instance, in the lower register of CMS’s plot, notice the bump around 350.]
We’re in that interesting moment when all we can say is that there might be something real and new in this data, and we have to take it very seriously. We also have to take the statistical analyses of these bumps seriously, and they’re not as promising as these bumps look by eye. If I hadn’t seen the statistical significances that ATLAS and CMS quoted, I’d have been more optimistic.
Also disappointing is that ATLAS’s new search is not very different from their Run 1 search of the same type, and only uses 3.2 inverse femtobarns of data, less than the 3.5 that they can use in a few other cases… and CMS uses 2.6 inverse femtobarns. So this makes ATLAS less sensitive and CMS more sensitive than I was originally estimating… and makes it even less clear why ATLAS would be more sensitive in Run 2 to this signal than they were in Run 1, given the small amount of Run 2 data. [One can check that if the events really have 750 GeV of energy and come from gluon collisions, the sensitivity of the Run 1 and Run 2 searches are comparable, so one should consider combining them, which would reduce the significance of the ATLAS excess. Not to combine them is to “cherry pick”.]
By the way, we heard that the excess events do not look very different from the events seen on either side of the bump; they don’t, for instance, have much higher total energy. That means that a higher-energy process, one that produces a new particle at 750 GeV indirectly, can’t be a cause of big jump in the 13 TeV production rate relative to 8 TeV. So one can’t hide behind this possible explanation for why a putative signal is seen brightly in Run 2 and was barely seen, if at all, in Run 1.
Of course the number of events is small and so these oddities could just be due to statistical flukes doing funny things with a real signal. The question is whether it could just be statistical flukes doing funny things with the known background, which also has a small number of events.
And we should also, in tempering our enthusiasm, remember this plot: the diboson excess that so many were excited about this summer. Bumps often appear, and they usually go away. R.I.P.
Nevertheless, there’s nothing about this diphoton excess which makes it obvious that one should be pessimistic about it. It’s inconclusive: depending on the statistical questions you ask (whether you combine ATLAS and CMS Run 2, whether you try to combine ATLAS Run 1 and Run 2, whether you worry about whether the resonance is wide or narrow), you can draw positive or agnostic conclusions. It’s hard to draw entirely negative conclusions… and that’s a reason for optimism.
Six months or so from now — or less, if we can use this excess as a clue to find something more convincing within the existing data — we’ll likely say “R.I.P.” again. Will we bury this little excess, or the Standard Model itself?
