Tuesday, May 10, 2016

Is The 750 GeV Resonance Real? I Don't Think So

Hundreds of papers have been written about an apparent 750 GeV diphoton resonance seen at some moderate significance by both of the LHC experiments which Lubos likes to call the Cernette.

These bare facts would suggest a boson rather than a fermion (since the spin of the decay products is even), probably of spin-0 or spin-2 (because spin-1 bosons can't decay to two photons), and an intermediate electrically charged state of an otherwise electrically neutral particle, so it can couple to photons.

A second round of experimental data earlier this year confirmed initial indications of a "bump" that was too mild to be sure evidence of a new particle, but could conceivable be one, with a possible secondary Z boson-photon bump at 375 GeV.

The data are inconclusive at this point on the question of whether this resonance is narrow or wide (resonances are graphed as bell curves with a peak at the mass of the particle and a width measured half way up the peak at nearby masses that corresponds to the mean lifetime of the particle).

This paper from January which was updated last week, makes the case for a 4 sigma local significance, before considering look elsewhere effects, resonance with a narrow width and spin-0 (that also couples to gluons and hence must have strong force color charge as well).  It notes that:
1. The required cross section to fit the anomaly reported by ATLAS is in tension with the 8 TeV results, as well as the required cross section to fit the CMS anomaly. 
2. Combining all data sets yields a local significance of ∼ 4.0σ for a 750 GeV spin-0 resonance produced through couplings to gluons or heavy quarks. While quoted statistical significances must be taken with a grain of salt, as they are obtained using binned data without inclusion of systematic errors, I find the combination yields a net increase in the statistical significance as compared to the ATLAS data alone.

3. The spin-2 interpretation is mildly disfavored compared to a spin-0 mediator. This is due to correlations in the photon momenta which results in a relative decrease in the ATLAS acceptance compared to CMS.

4. The combination of ATLAS and CMS 13 TeV data has a slight statistical preference for a spin-0 mediator with a natural width much smaller than the experimental resolution, as compared to the Γ = 45 GeV preferred by ATLAS alone. When the 8 TeV data is added, there is a slight statistical preference for a wide resonance over the narrow option, as it is easier to hide a wide resonance in the 8 TeV background. . . .

When considering the 750 GeV diphoton excess, the theoretical community must balance its natural exuberance with the recognition that the statistical size of the anomalies are very small. As a result, any further slicing of data will yield at best modest statistical preferences for the phenomenological questions that we in the community want answers to. That said, given that this excess is the most significant seen at the LHC since the discovery of the 125 GeV Higgs, and the resulting avalanche of theoretical papers which shows no sign of slowing, it is still a useful exercise to carefully analyze the available data and determine what we do – and do not – know at this stage. While there is some useful information to be gleaned from this exercise, we are fortunate that the continuation of Run-II will be upon us shortly.

From the existing data, we can conclude the following:

1. Explaining the anomaly through a spin-0 resonance is preferred over a spin-2 mediator, though this preference is less than 1σ in most cases.

2. Combining the 8 and 13 TeV data from ATLAS and CMS sets yields a ∼ 4.0σ statistical preference for a signal of ∼ 4(10) fb, assuming a narrow (wide) spin-0 resonance. This ignores the look-elsewhere effect, as discussed. Given that the significance of my fits to individual ATLAS and CMS data-sets are underestimates when compared to the full experimental results, it is possible that the actual statistical preferences are larger than these quoted values. However this would require a combined analysis performed by the ATLAS and CMS Collaborations.

3. The cross sections needed for the Atlas13, Cms13, and Cms13/0T data sets are incompatible at the two sigma level, though they agree in mass. The most straightforward reading of this (while maintaining a new physics explanation for the anomalies) is that the larger Atlas13 cross section constitutes a modest upward fluctuation from the “true” cross section, which is more in line with the Cms13 value. The reverse is also possible of course, but would bring the diphoton excess in the 13 TeV data in greater tension with the 8 TeV null results.

4. When considering only the 13 TeV data, the Cms13 data does not share the Atlas13 preference for a 45 GeV width. I find that the “wide” interpretation of the resonance has a statistical significance in the combo13 data set which is approximately 0.5σ less likely than the “narrow” interpretation. The corresponding likelihood ratio shows no preference for either width. Thus, while the theoretical challenge of a wide resonance may be appealing, the data in no way requires any new physics explanation to have the unusually large width of Γ ∼ 45 GeV.

5. Combining the 13 TeV data with the 8 TeV, I find that gluon-initiated mediators are preferred, due to having the largest ratio of relevant p.d.f.s. In particular, the combination of all six data sets for a gluon-initiated narrow resonance has the same statistical preference for a signal as the Combo13 data alone does, though the best-fit cross section decreases slightly when the 8 TeV data is added (∼ 4.0σ for a ∼ 4 fb signal). In the narrow width assumption, heavy quark-initiated mediators have slightly smaller statistical preference, and a light-quark coupling has a fairly significant decrease in statistical preference, indicating a more serious conflict between the 13 and 8 TeV data.

6. Combining the 13 and 8 TeV data sets under the Γ = 45 GeV spin-0 model increases the statistical preference for signal as compared to the Combo13 result, as the excess can be more easily absorbed by the background model here. Combining all the data sets in this way results in a ∼ 3.5σ preference for a ∼ 10 fb signal (a 0.5σ increase over the Combo13 wide-resonance fit), with a likelihood ratio of ∼ 20 rejecting the narrow interpretation. Again, these statistical preferences are relatively small, thus theorists are free to explore the options, but should keep in mind that the experimental results are inconclusive.

The conclusions of this paper are perhaps not a surprise. There is clear tension between the Atlas13 and Cms13 results, as well as with the non-observation in 8 TeV data. The question of the width is especially puzzling; but further slicing of the data, as I have demonstrated, leads to somewhat conflicting results which do not have a clear statistical preference towards any one solution. I note that if the ATLAS excess is indeed an upward fluctuation from a signal which is more in line with the Cms13 value, then perhaps this could also give a spurious signal of large width. However, the true answers will only come with more data, though I note that, if the signal is indeed real, but on the order of 4 fb, then we may need 10-20 fb−1 for a single experiment to have 5σ discovery.
Some theorists have also hypothesized that this is really a quad-photon resonance with pairs of photons so close to each other that they register as single photons in the detectors.

Despite the superficial similarities of a spin-0 750 GeV particle to a Higgs boson, the resonance is very un-Higgs like.  For example, the primary decay of the Higgs boson is to b quark pairs, even though the diphoton channel is easier to see.  And, pairs of b quarks with 371 GeV of momentum (plus 4 GeV or so of rest mass) would be hard to miss.

Lubos who is a bit generous about fitting the data to numbers because he thinks this is probably real (he gives it a 50% chance of being real), also asks his readers to:
Recall that the ATLAS diphoton graph also shows an excess near 1.6TeV, close to two times 750GeV, and 375GeV is just a bit higher than twice the top quark mass, 173GeV.
I'm not as impressed with "almost" twice the mass, particularly when the 750 GeV v. 375 GeV pairing seems to be exact (although he also points to a possible 340 GeV mass for the lower bump). Twice the top quark mass is about 346 GeV, and twice that is about 692 GeV which is pretty much impossible to reconcile with 750 GeV unless you have a composite particle with some a lot of strong force binding energy to make toponium and very little to go further and make a toponium molecule or top tetraquark at 750 GeV that in turn almost instantly annihilates into photons.

But, of course, if you had a top quark-top antiquark composite particle, you ought to be seeing decays predominantly to pairs of b quarks and not to photons and Z bosons the way you do when pairs to top quarks that aren't bound to each other in hadrons are product.  To get those decays you'd need toponium to experience a matter-antimatter annihilation to photons before the constituent top quarks could experience weak force decay.  And, you'd probably want a process that produces the toponium from extremely high energy gluon fusion or something like that.  (Glueball annihilation is also naively attractive, but the predicted glue ball masses aren't anywhere near a 750 GeV resonance; they're much too light.)

The diphoton channel at that energy level doesn't have a lot of potential background noise, so even a pretty modest number of diphoton detections at that energy scale could be significant.

But, there are hundreds of papers, rather than just a few leading ones, because a resonance at this mass, with the characteristics it has and the lack of strong signals in other channels at the same energy level, is not very well motivated in any of the leading extension of the Standard Model of particle physics.  Also, the different data sets used to infer its existence are in some tension with each other.

For the most part, models that can accommodate a particle that decays to a 750 GeV diphoton resonance while having few other decay modes, require the existence of whole classes of other new particles to go with it (even if it is composite, rather than fundamental).  This gives rise to a very baroque new beyond the Standard Model theory.

For that reason, despite the notability of the bump statistically, my money is on this turning out to be a statistical fluke or systemic experimental error, rather than an actual new particle.

We'll learn if this prediction is right or not within a year or two, depending on how many more weasel, baguettes and other mishaps interfere with the LHC experimental schedule.

No comments: