2/fb CDF results for New Physics search

A paper describing a model-independent search for new physics, based on 2/fb of 1.96 TeV proton-antiproton collisions, has been submitted by CDF last week. In this short post -not meant for outsiders, sorry- I will just explain the basic idea behind the search, and show the distributions which are found to be the ones most clearly indicating that indeed, there is new physics out there. Just kidding: no signal is found for anything beyond the Standard Model, but indeed, some strange features of CDF data have been unearthed with a quite original method.

The method is actually a triad of different tools. Vista, Bump Hunter, and Sleuth. Vista checks 399 independent final states including high-Pt electrons and muons, photon candidates, high-energy jets, and missing transverse energy (which signals the escape of an energetic neutrino): these are signals of objects that, in a way or another, may conspire to put in evidence the presence of some new process in the data. The Standard Model background processes contributing to each of the 399 combinations of those objects are evaluated with tuned Monte Carlo simulations, and a prediction for the number of events collected in each final state is computed. As shown by the graph below, the comparison with the data shows positive and negative discrepancies: these distribute nicely around zero, with a Gaussian-like shape and no real outlier, indicating that there is no smoking gun of some new physics process contributing to the data.

The histogram has one entry per each final state checked. Each entry sits at the value of standard deviations on the difference between number of events observed minus number of events predicted: the positive tail should contain final states with an excess of events, which are possible indications of new physics. Unfortunately, there is nothing particularly strange about the distribution of the green histogram, which is well modeled by the black curve, showing the expected shape of the 399 tests.

Vista also checks for kinematical shapes: variables obtained by combining the momenta and the angles of the objects found in the 399 categories of events. For instance, if you have three jets, you can produce a histogram of the angle between the two less energetic jets: and this indeed is something that Monte Carlo simulations have a hard time modeling, as shown below.

The one above is the most significant disagreement observed by Vista, which examined 19,650 different distributions. The interpretation of the disagreement is not of new physics, but indeed, just a bad modeling of soft gluon emission -the production of a jet with low energy by the emission of a gluon off another quark or gluon, at small angle. The black points are CDF data, the red histogram is the Standard Model background expected by Monte Carlo simulations. The two sketches in grey show that at large the three jets have a typical “Mercedes-like” appearance, while at low there is one jet recoiling against the two less-energetic ones, which stay close in angle: well, it appears that nature has a smoother transition between the two topologies than what simulations indicate.

Bump Hunter is instead a search for excesses with a bump-like appearance in a number of invariant mass distributions, again constructed by taking into account all possible meaningful combinations in the data. 5036 distributions are checked, with no significant surprise, except a few which are driven by the bad modeling of the distribution already discussed above.

Finally, Sleuth performs a search for numerical excesses of events at high-energy, under the prejudice that new physics will most likely start showing up at the high-end of those spectra (notice the green color of the statement: green is the color of hope). The variable studied by Sleuth is the sum of transverse momentum of all the bodies available in the event among those mentioned above: jet energies, lepton momenta, and missing transverse energy are summed into one single scalar variable. Different final states, of course, end up in different histograms; Sleuth searches those histograms for the most striking disagreement of the tails with Standard Model backgrounds.

Sleuth finds some remarkable excesses in CDF data. The three most significant excesses are found in the high-transverse momentum distribution of final states including two leptons of the same electrical charge! Here are the distributions:

Above, the Sum Pt of events with an electron and a muon of the same sign. This is the most discrepant distribution, with the “optimal” cut to discriminate the oddity at . The P-value is 0.00055.

The one above is instead the sum of transverse momenta for events which have an electron and a muon of the same sign, plus two jets. The optimal cut is at 342 GeV this time, isolating thre events with scarce backgrounds. The P-value for this distribution is 0.0021.

Finally, the one above is obtained from events with an electron and a muon of the same sign, plus significant missing transverse energy. The cut at 169 GeV isolates 16 events, with a background of 5.7. The P-value in this case is 0.0042.

The P-values (probability of the excess) quoted above must not mislead you: they do not account for the “trial factor”, that is the fact that we had to search in many different distributions to find some that were in such disagreement. The probability to observe what is shown above is evaluated at about 8%, so there is no need to get hyper… Yet. If you want my own interpretation of the excesses seen in the three distributions (which, I stress it for clarity, are mutually exclusive, i.e. they do not contain shared events), I can give it to you, but beware, it is quite mundane.

I have the feeling that the muons in those distributions contain a sizable fraction of fakes, which the simulation does not account properly. Fake muons may be caused by hadrons, which manage to traverse the detector and get detected in the muon chambers. In case that happens, there is no real reason for the muon and the electron to be of opposite sign, since their production is uncorrelated. Beware, this implies that also opposite-sign electron-muon pairs should contain a similar excess: well noted, but in that case, such an excess would have to fight much larger backgrounds from correlated electron-muon production by physical processes such as top-antitop production, WW production, etcetera. If that is what really happened, poor Sleuth was stuck with finding the same-sign excess, since opposite-sign data had too large statistics to make a few odd events stick out.

Ok, now that I have said it (and I stress it is my own personal opinion and not my experiment’s), take a ticket. I will be proven right one day, but until then you can shoot the sitting duck (me).

More information about the study described above is available in the public web page of the Global Search analysis.