The result in question is a nice new search, performed on the full Run 2 data sample of proton-antiproton collisions delivered by the Tevatron collider last decade. The searched process is a very rare one, which involves the production of a Upsilon meson and a vector boson (W or Z). This associated production has a cross section in the Standard Model that makes it inobservable by CDF, but new physics mechanisms could enhance the production rate quite significantly. Seeing a few such events would thus be a very clean way to spot new physics.
Now, my interest for this paper on VY production does not stop there. It so happens that I was the first one to search for it, in CDF Run 1 data - a data sample 113 times smaller than the one available now. In 2003 I published a paper on Physical Review Letters, which was the result of almost two years of work. To get there I was helped by my colleague Alberto Ribon, who produced the simulation samples of the signal, and by Giovanni Busetto, who stimulated my interest in the analysis. The originator of the idea was however Kwan Lai, a Berkeley physicist who was not a member of CDF but was in contact with our CDF group in Padova.
Having spent two years of my life in that analysis, and having put several nice touches in it, I had great expectations as I found the preprint of the new search in the Cornell Arxiv this evening... Did they see a signal ? Did they use my selection strategy ? Did they employ my data-driven background estimate method ? And most of all, did they search for ALL possible decay modes of W and Z bosons, as I did in 2003 ?
How to search for VY pairs
Let me open a parenthesis. I must first explain what happens if a proton-antiproton collision produces a WY or a ZY pair. The Y meson decays only 2.5% of the time in a muon pair, but experimentally one may neglect the other decay modes (2.5% in electrons, taus, and then a multitude of unreconstructable hadronic states). Muon pairs are just so clean in comparison with everything else that one just does not bother trying the Y->ee final state, which is the next cleanest.
For W and Z bosons the matter is apparently quite similar. Usually one still looks for W and Z decays in the leptonic final states involving electrons and muons (the W->ev and W->μv decays for W's, and the Z->ee and Z->μμ decays for Z's). The reason is the same: the other decay modes are background-ridden and offer a lot of trouble in exchange for dubious improvements.
My "discovery" in 2002 was however that it is not at all unreasonable to search for all other main decay modes of the vector bosons in this particular case, as the process is a rare one to begin with. In other words, one does not need to be scared by backgrounds, if backgrounds are non-existent or very small!
In my 2003 article (okay, I should say "in the CDF 2003 article", but it really is a paper I feel I own) in fact I demonstrated that by searching for jet pairs produced by W and Z decays, as well as for large missing transverse energy alone (which again can be produced by W->tau nu decays, Z->tau tau decays, as well as by the one-in-five Z decay of Z bosons into neutrino pairs), the sensitivity to the VY production process was significantly larger. In fact, these channels turned out to be the highest sensitivity ones!
Diatriba mode on
The new analysis, unfortunately, does not bother looking for the jet-jet and the missing ET signatures. I think that is a mistake, which degrades the potential of the search. The article mentions in the introduction that they restrict to the leptonic signatures:
"We use only the electron and muon decay channels of the W and Z bosons, which give the best sensitivity for this search".
I find the sentence puzzling. Or rather, slightly misleading. In fact, they are probably right that, if the jet-jet and missing ET final states are not exploited with care, the leptonic channels are more sensitive. But the sentence appears to say that their choice of restricting to the leptonic final states yields them the best sensitivity, which is obviously impossible - added channels would by definition improve the sensitivity, as there is more information there to exploit.
Now, things are not so clear-cut, and the choice of the authors can be understood as a way to avoid a very unrewarding work with low-pT muon triggered data. This is because by looking for leptonic W and Z decays they can restrict to data triggered by a high-pT muon or electron: those datasets are the "golden" ones at the Tevatron, as they are the starting point for extraction of top quark datasets and all vector boson physics measurements.
Conversely, if one wants to also look for jet-jet or missing ET decays of the vector bosons, one must trigger the event with the Y->μμ signature. Dimuon samples are much harder to deal with, for a number of reasons which yours truly unfortunately knows very well, having had to deal with them at length. Further, in the latter part of Run 2 the collection of data by low-pT muon triggers was reduced in order to contain the output data rate of the trigger selection. All these are details and I cannot explain more of that here.
There are further reasons why in Run 1 the jet-jet and missing ET signatures may have been more attractive than in Run 2. In Run 2, the high luminosity reached by the Tevatron caused a significant amount of "pile-up" - multiple proton-antiproton collisions occurring simultaneously. When one looks for the Y->mm plus jet-jet signature these pile-up interactions increase the hadronic background significantly; and when one looks for the Y->mm plus missing ET signature one has to be careful that the missing ET may be originating from a different collision than the one that produces the dimuon pair.
All in all, it is difficult to say from the outset that the restriction to leptonic W and Z decays was a significant degradation of sensitivity with respect to what could have been obtained by exploiting all possible datasets and selections. What remains to be seen is whether the new search has overall a better sensitivity than the old one, or not. So let us look at the final numbers of the searches and compare them.
In the 2003 analysis, with 83.4 inverse picobarns of collisions, my analysis placed limits on YW and YZ production at 93 and 101 pb, respectively. The new analysis places limits at 5.6 and 21 pb using 113 times more data. However to be fair we should take the "expected" limits of the new search, which are better in the case of the ZY search (13 pb) -CDF in Run 2 saw one event candidate in the ZY search while expecting a background of 0.1 events, so the observed limit is worse. The 2003 search saw a total of 3 events with 3.5 expected, so we can take the observed limits without changing any conclusion (I do not have the "expected" limits for the old search handy).
Above, a transverse view of the ZY event candidate found by the Run 2 search. The four muon tracks are shown in green.
In a statistics-dominated search like this one, when backgrounds are large the sensitivity scales with the square root of the data increase factor - so a factor 10.5 in this case. Hence the Run 2 limits should be 10.5 times stricter if the sensitivity of the analysis were the same, correct ?
Wrong. In fact, when the number of background events expected is small, the sensitivity scales more like a linear function of data size, not with the square root.
To show why that is so, let us simplify a bit the situation. Consider that you are searching for ZY events in 83.4/pb of data and you expect 0.001 events from backgrounds. Your upper limit on the signal, upon observing indeed 0 events, is N<3 events at 95% confidence level. Now imagine you collect 113 times more data. Your background expectation is now 0.113 events, and you still see zero events. Your upper limit still is N<3 events. When you convert that <3 events of the 83.4/pb search into a cross section limit, you get some value, σ<X; when you convert the N<3 events of the 9400/pb search into a cross section limit, you get a new number which is 113 times smaller, not 10.5 times smaller.
All the above is to say that the limits of the new search (respectively, 16.4 times and 7.7 times tighter than the old one, with 113 times more data) appear to confirm that its sensitivity is worse than the old one. And I find this a bit nagging, for two reasons. One, that this is almost certainly the final word of CDF on this search - it is a "legacy paper", so to speak, and standards for it should be higher. Two, that the road for a better measurement had been shown 11 years ago, but was neglected. (One could add a third point concerning the fact that the Run 2 detector was significantly better with respect to muon efficiency, but let's not diverge).
Of course, with the thinning resources and manpower of an experiment that is closing down, there are reasons to excuse the shortcuts that were taken; but that is what it is - an excuse. I would have appreciated it if in the article the authors had given more explanations of why they restricted to the leptonic W and Z decays, instead of saying that "it gives the best sensitivity".
Okay, rant off. I still congratulate the authors for their result, and I am glad they took the time to repeat the search in Run 2. Physics-wise, the new result is quite interesting. While the Standard Model predictions for the cross section remain 100 times smaller than the placed upper limit, there are a number of SUSY models that could have produced observable VY signals but have not. SUSY remains at large, once more.
Comments