A Quantum Diaries Survivor

2007 has been a rather bumpy year for blogging about Higgs searches, and in particular those for a MSSM Higgs decay to third-generation fermions. It all started when John Conway discussed at Cosmic Variance his brand new result on the decay to tau lepton pairs, which showed an intriguing excess attributable to a 160 GeV boson. In a series of posts here and there, the discussion on whether a corresponding signal was observable in data containing two b-quark jets in CDF and D0 samples used to extract the decay led the press to become intrigued. In March, New Scientist produced a slightly incorrect account of the whole picture, and the Economist pitched in, adding more inaccuracies. Many in the CDF and D0 experiments resented that.

Just as the buzz level on the internet was getting back down to a physiological level, rumors started again about a possible large signal found by D0 in a search for MSSM H->bb  decays in events with three or four b-quark jets. Since that is another good sample where to search for that particle, people were again very interested in the issue, and a new wave of interest arose everywhere, and got picked up by slate and other high-traffic web sites, and still more newspapers.

The rumor was not based on a public result, so there was not much to do but sit and wait for the moment when D0 would be ready to discuss their achievements. But D0 did not produce anything, only making the curiosity grow.

I am too lazy to replicate all the links here, so if you want to dig in the issue please find all the required addresses in this more comprehensive summary .

Anyway, that is the past.  The news now are that, for the Lepton-Photon 2007 conference, CDF has produced a public result on the same search. Since CDF and D0 have similarly sized datasets, and similar sensitivities, it is quite interesting to see what they find! So let us look at the new result, whose main authors are Thomas Wright and Dante Amidei - two knowledgeable and skilled physicists who usually (and this is no exception) produce very accurate results. 

The search is based on a trigger collecting multi-jet events enriched with b-quarks. The trigger is made possible in CDF by the Silicon Vertex Tracker, a wonderfully complex set of custom-designed hardware boards organized in a highly parallel architecture. The parallelism allows speed of execution: in less than 20 microseconds, the hardware collects information on hits in the silicon layers, compares hit patterns to those stored in associative memory boards, and performs a linearized fit, achieving a measurement of charged track transverse momentum, azimuthal angle, and -crucially- impact parameter with respect to the beam position with a precision quite similar to the best achievable offline with a glass of brandy in one’s hands. A sketch of the operation flow of SVT is shown below, where you also get to see the actual boards performing the operation.

The multijet trigger collects events with at least three jets, two of GeV and a third with looser  GeV. Each of the two leading jets is required to be matched in azimuth to a track with . The latter is the impact parameter (I.P.) cut, which allows to spot tracks which are likely originated in the decay point of a B hadron. The drawing on the left shows what happens to tracks originated from a decay in flight (of the B particle): they produce a I.P. (blue segment) with respect to the interaction point (in red). Forget the yellow arrows please.

The sample of data collected by the SVT multijet trigger is enriched in b-quark jets, but a more stringent selection is needed. Such is provided by the SecVtX algorithm, which explicitly searches inside jet cones for sets of charged tracks which can be fit to a common origin displaced from the interaction point: a secondary vertex. CDF selects events with all three jets tagged by SecVtX, to select a sample enriched with the process .

Tom and Dan found that to search for MSSM production of Higgs and one b-jet, the ”smoking gun” distribution was the invariant mass of the two highest- b-jets: they are most likely to be those coming from Higgs decay. However, even after triple SecVtX b-tagging, CDF data contains a mixture of several different QCD processes, not just a single background of “multiple b-jet production”: events with three b-jets, but also events with two b-jets and a light-quark jet, or a charm jet.  These backgrounds have to be modeled with percent accuracy in order to achieve a reasonable sensitivity on Higgs decay.

Monte Carlo generators do not predict with the necessary accuracy the correct mixture of processes with b-jets and c-jets in multijet events.  So one has to rely on the data to model them. Events with only two b-tagged jets provide a good handle, but the key to understanding how the different backgrounds contribute to the 3-tag sample is given by a variable called , constructed with the invariant mass of tracks fitted to the secondary vertex in each jet:

where  is the “vertex mass” of tracks in jet . By fitting simultaneously the invariant mass distribution of the leading jets and , backgrounds get constrained much better, allowing a much more precise modeling and a better result for the Higgs search.

So here is the result of the fit: The dijet mass distribution with the best fit overlaid is shown in the plot on the left, for a hypothesized MSSM Higgs of 150 GeV. The black points are CDF data with three b-tags, the four main backgrounds are shown in light blue (bbb= three b-jets), dark blue (bb+light-quark jet), violet (bcb= events where a charm jet produced along with two b-jets is one of the two leading jets), and yellow (bqb=events where a mistagged light-quark jet is one of the leading jets). The best fit allows some small fraction of signal (in red), but it is also compatible with the no signal hypothesis, such that a limit can be set to the cross section of  production, and an exclusion obtained in the - tan(beta)  plane. The limit is shown below.

The limit above, shown with a black line, is compared -as has become customary lately- with expectations from pseudo-experiments (dashed black line, and red 1-sigma and purple 2-sigma bands). You can see that between 140 and 160 GeV CDF obtained a worse limit than the one they expected to set, mainly because of the small fluctuation which is shown in the dijet mass fit above. In all cases, the fluctuation is at most a 1.5-sigma excess, which again, is one more piece of non-evidence for MSSM Higgs at the Tevatron. Nonetheless, let me say that this analysis is a very beautiful attempt at something which I know out of experience to be a quite difficult task: understanding the composition of multijet datasets in terms of their flavor composition, and modeling correctly the different nuisances of the mass distribution of the leading jet pairs, is no small feat.

For the more technically inclined, I should also mention that the plot above was derived by assuming no width effects in the Higgs mass templates. That is to say, the increased bbH coupling due to a large value of tan(beta) will increase the width of the breit-wigner shape. The cross section also gets modified by the change, and so one has better compute a limit taking those effects into account. Of course, both including and neglecting the effect is useful. You can find the width-included limit in  the public note describing the result.