Comparing the ATLAS and CMS SUSY Searches
Neither paper reports evidence for supersymmetry, a fact that has been discussed on several other blogs, such as Not Even Wrong, Quantum Diaries Survivor and Backreaction among others. I’ll avoid the tedious debates about whether SUSY is in trouble or is a bad idea in the first place.
My interest is much more in what the experiments did with their data.
Someone made the pithy statement: CMS optimized for background rejection, while ATLAS optimized for the SUSY signal. There is some real truth in this statement. The CMS analysis is extremely robust, clever and safe. There is little chance of a false positive and a very conservative, even skeptical attitude toward the detector and reconstruction algorithms has been adopted. The ATLAS analysis follows the paths laid out at the Tevatron, grasping the SUSY signatures by the horns and simply going for it as if they had years of experience with multi-jet event topologies.
Keep in mind that both papers are looking at events with multiple hadronic jets and missing transverse energy (“MET”). They both veto events with leptons and the jet reconstruction is quite similar. The initial event samples from ordinary pp collisions are huge and looking for a signal for new physics means picking a promising corner of the haystack and hoping to find the needle there.
The CMS analysis works with a kinematic variable called αT, given as the ratio of the ET of the second jet to MT, the transverse mass of the event reduced to a di-jet system. The αT distribution is the crux of the analysis, because the QCD background should fall below 0.5 but a SUSY signal will have a (small) tail above 0.5. Look at these plots from the CMS paper:
Notice the log scale. There is a tremendous amount of background below the cut αT<0.55 which is clearly very well controlled by the CMS reconstruction; almost nothing leaks above this cut. Unfortunately, most of the likely SUSY signal also falls below the cut so the CMS search is really trying to catch the cat by its tail.
The ATLAS analysis takes samples of 2-jet and 3-jet events and applies hard kinematic cuts on the MET and on an effective mass variable, meff which is just the sum of MET and the ET of the two most energetic jets. For the most part they end up counting Z+jets events in which the Z had decayed to neutrinos, as you can see from these two plots:
Of course, the agreement of the data with the simulation is excellent, as in the case of the CMS plot above. One should not under-appreciate how important such agreement is, given the coming challenges of finding (or not) supersymmetry in larger data samples.
One point of the CMS analysis which I particularly like is the way the backgrounds are estimated through several data-driven methods. You must read the paper for the details. There are two ways of extrapolating from one kinematic region dominated by background into the signal region. There is a direct measurement based on a W→μν + jets signal which is exploited in two different ways. Finally, Z→νν backgrounds are checked with γ+jets. One has no doubt that the background really is pinned at something like 10±3 events, somewhat below the 13 events observed; the CMS exclusion contour is weaker than one would have predicted in advance.
The ATLAS analysis, in contrast, relies on rates and shapes taken from Monte Carlo simulations. To be sure, they have made many tests and studies to convince themselves that the simulations are reliable. But one wonders whether they would be willing to claim a discovery with this kind of analysis. Out of four subsamples of events, three of them have fluctuated downward, so their limits are slightly better than predicted.
ATLAS published a good plot which includes the CMS contour for comparison:
The red line shows the exclusion the ATLAS, which is far more impressive than the black line showing the CMS exclusion. As already pointed out, it is slightly better than the expected exclusion contour, given by the blue dashed line. Note that 1σ fluctuations given by the dotted lines are quite far off showing that small ordinary fluctuations will have a big impact on the contour obtained from the data. The CMS expected exclusion is only slightly better than the one achieved; overall the ATLAS expected exclusion is better than the CMS expected exclusion.
Thus, we have to return to that earlier pithy statement: CMS optimized for background rejection, while ATLAS optimized for the SUSY signal. As a result, CMS has complete supreme control of its background, while ATLAS achieved a more stringent exclusion in the (m0,m1/2) plane.
Of course, this is only the beginning, and both collaborations have more results in the pipeline.
(Notice: I am a member of the CMS Collaboration.)
Entry filed under: Particle Physics.