Banner
    The Jet Energy Scale As An Explanation Of The CDF Signal
    By Tommaso Dorigo | April 7th 2011 04:02 PM | 31 comments | Print | E-mail | Track Comments
    About Tommaso

    I am an experimental particle physicist working with the CMS experiment at CERN. In my spare time I play chess, abuse the piano, and aim my dobson...

    View Tommaso's Profile
    In my post about the new CDF signal of a mysterious new resonance decaying to jet pairs, there is an active comments thread. I posted there a graph crafted by Tommaso Tabarelli de Fatis, a CMS collaborator, who picked the CDF data and simulation and scaled the energy scale of the latter up by a few percent, showing that the agreement of simulation and data was better, and that the bump at 145 GeV could be explained away this way. Below you can see the result of scaling the jet energy scale up by 4% (the jet energy scale is bumped up by just scaling the dijet mass; this is in principle approximate, but it is a good one at that).



    I mention this because Tommaso has now produced an animated GIF of the tweaking up of the MC/data Jet energy scale in the CDF analysis, showing how the residues of data minus simulation behave for JES shifts of 0% to 7%. To put this in perspective, you should know that in 2006 CDF published (actually a few physicists from CDF, a dozen of them to be exact, and among them yours truly) an analysis of their Jet energy scale, assessing at 3% the uncertainty on that quantity.

    One should also remember that CDF can now verify the agreement of their jet energy scale with simulations to a much better precision than 3%, by fitting the W->jj decay signal in top pair decays when they fit the top quark mass. The current measurements have a precision of 1% on the jet energy scale, in fact. However, one should also consider that the background in the new CDF analysis may be due to gluon-originated jets, which might be subjected to a different jet energy scale with respect to quark-jets that originate in hadronic W boson decays measured in top mass measurements.

    In any case, I suggest you to give a look at Tommaso's animated GIF.

    Comments

    I'm confused. Surely this is something that the CDF review process would cover?

    An off topic, but topical, question: what is the point of a collaboration quoting a 2% confidence level?
    See new MINOS paper.

    dorigo
    Ciao Kea,

    the internal review process did consider these effects, of course. I cannot write more about it here, but if  the paper is out it means that CDF considered the JES not a valid explanation of the signal.

    About 2% CLs, the point is probably to give something like a p-value for their observation. I do find it kind of twisted myself, but I have not checked the paper yet (I'm on a slow connection right now).

    Cheers,
    T.
    The point is that fitting mixing parameters for the neutrino and anti-neutrino oscillation data independently (4 parameters) yields a better description of the data than a joint fit (2 parameters). The question is how often does this happen by chance when the underlying oscillation parameters are the same for neutrinos and anti-neutrinos? The procedure is to do the 2 parameter fit to the data, then use MC pseudo experiments at that best fit to figure out how often your 4 parameter fit yields a description that is at least as improved (chi2 difference) as what you see when fitting the data with 4 parameters. The answer is that this happens 2% of the time. Hard to boil this down into one pithy sentence.

    Sweet lord, that may make the bump go away, but it breaks havoc elsewhere at higher and lower mass: if the purpose of that animated gif was to convince us that this is an artifact of the energy scale, I am afraid the effect on me has been the exact opposite. Besides, it was emphasized several times yesterday that they have a good grip on this scale.

    dorigo
    Be careful Anon, the two leftmost bins cannot be trusted, because they do not benefit from a shift upward of data that was removed from the first plot before crafting it.

    In other words, if I place a cut on jet energies at 20 GeV, say, on data and MC, and then scale up all MC energies by 10%, I do not expect that there will be any events at 20 GeV left in the MC after the procedure. Clear ?

    I am more impressed by the better agreement of the W/Z peak after a shift.

    Cheers,
    T.
    Hi Tommaso, thank you for your answer to my question on how this plot was made that I asked at your original post. I wrote a reply there. You confirmed that Tommaso Tabarelli de Fatis made these very nice plots using only the Mjj plot given on the left. For a JES scale increase by 4%, the energy of each jet increases by 4% and the Mjj value increases by 4%. That means that on the graph the MC-based background shape is shifting to the right. However, I have the impression that this works only when you have the values of Mjj on an event by event basis, which again Tommaso Tabarelli de Fatis does not have. Otherwise, he has to assume all the events in the first bin have exactly one value of Mjj and thus the entire bin content should move to another value of Mjj that lives in another bin. Is this what he did? If so, then either all content of the first bin remained in the first bin, or all moved to the second bin. But in his plot only some events from the first bin moved to the second bin. That makes me think that he assumes a distribution of events inside each bin, either a Gaussian distribution or a flat distribution or linear distribution from the value of previous bin to the value of the next bin. If he does that, he can estimate the value of Mjj on an event by event basis. Then he can shift the values of Mjj by 4%. Then he would see each value in which bin it is. And then he would get the new plot. Is this how he did? Is there another way, short of knowing the real values on an event by event basis? I ask as as a student I would like to understand and learn this technique. It can be helpful in the future.

    Thanks,
    Adrian

    dorigo
    Hi Adrian,

    I do not know the details of what Tommaso did, but I trust he did things the best way allowed by the data he had. Which could be something of the kind you describe above. I would personally fit the total mass distribution to derive a functional form to get distributions within each bin, but I think these effects are small and beside the point.

    In general, there is a better way, which is called "template morphing", to handle these variations. One works with the cumulative distributions instead than with the differential ones, and interpolates those.

    Cheers,
    T.
    Tommaso, something is wrong with your RSS feed. I never get updates anymore, just an error that says your live bookmark failed to load while all my others work fine.

    same problem here! I forget to check the blog without the RSS feed ;)

    dorigo
    Hi Andre,

    1% is a reasonable value for W-->jj decays, the ones one can fit in top pair decays (see e.g.
    http://www-cdf.fnal.gov/physics/new/top/2009/mass/meatv3_p19_public/index.html , where the fit returns a DJes=0.3+-0.3, which is 1+-1% in the scale).

    But there is another measurement we did with earlier data, using Z-->bb decays. Of course b-jets are special (and that's part of the interest of that particular measurement). That early measurement determined the scale to better than 2% precisio, see e.g.

    http://arxiv.org/abs/0801.3906 (a paper I edited)...

    I am confident that the scale can be determined to better than 1% in the long run with the Tevatron data.

    Cheers,
    T.
    Hmm... The result you pointed out (DJes=0.3+-0.3) implies E/E_meas ~ (1 - DJes * 3%) ~ 0.99+-0.01. So, as you argued, the JES seems to be known to 1% from this analysis.

    But the fact that E/E_meas < 1 would mean that the disagreement in the Wjj analysis would be even worse (unless the 0.99 is corrected out, which I could not figure out).

    The main motivation to scale Mjj is not really to make the excess go away, but rather to better describe the peaking part. Let me put it in another way.

    Which one is more likely:
    1 - A 3 sigma excess due to a BSM physical source, or
    2 - A 3 sigma excursion of JES uncertainty?

    Cheers,

    Andre

    dorigo
    There's no question about that Andre. We all reason in terms of priors, and our priors for BSM are very, very low... So those three sigmas do very little to make them higher as posteriors than the posterior of a mere statistical fluke.

    Cheers,
    T.
    2x more data + D0 will tell :-)
    (The LHC's gonna take some time...)

    dorigo
    Well, DZERO might do it rather quickly, but not 2x the data... It takes time to process and calibrate these large datasets.
    I believe CMS and ATLAS will want to say something on this by the summer. 100-200/pb will be enough IMO (cross sections are five to ten times larger for diboson production).

    Cheers,
    T.
    Hi Adrian,

    I think you can produce a set of random N numbers which are distributed accordingt the distribution given in the paper.
    Then you can easily multiply each number by a the factor you like most and re-make the plot.
    this is the way I should do it.

    cheers,
    D

    Thank you, D. It is a nice idea. Do you know of a technical (C++/ROOT) way to produce random numbers distributed to a desired distribution binned histogram? That would be indeed very useful.

    Thanks,
    Adrian

    dorigo
    One line of code: x= TH1->GetRandom();

    Cheers,
    T.
    Let us assume that this bump is a Z' with mass around 150 GeV.
    Is it possible that Tevatron missed the p pbar --> Z' -->jj signal?

    Tevatron measured sigma(p pbar --> jj) with a precision of 1% at M_jj > 200 GeV
    (fig.2 of http://arxiv.org/abs/0812.4036v1). But no data have been published at lower M_jj.
    Do such data exist??

    There is only one such search from UA2 but apparently those limits can be evaded. Please look at this paper for a nice review of the possibility a Z' with 150GeV exists and we just missed with current and past experiments: arXiv:1103.6035

    dorigo
    Wait AS, no electroweak signal can be extracted like that. The cross section is suppressed by the ratio of squares of couplings... One can extract a non-strong-produced dijet resonance only in special cases, or -like in this case- when it is produced together with some other particle which "selects out" some of the background.

    Cheers,
    T.
    What JES shift gives the best description of the first peak, coming from W+V ?

    Thank you. Just to be sure: could you see the single Z or W production, followed by hadronic decay, at TeVatron?

    dorigo
    Hi AS,

    as far as I know, the detection of inclusive W or Z bosons going to dijets has been done only twice: once by UA2 at sqrt(s)=630 GeV, where they could collect data with a very, very low energy jet trigger (something which we cannot do nowadays because there is more interesting data to collect and these triggers have huge rates), and once by CDF - that is, by yours truly single-handedly. That was in fact my PhD thesis in 1999. I extracted a signal of Z->bb decays from inclusive lepton triggers in Run 1, and then in Run II with a small group we found a larger signal in 6 times more data and produced a paper on the calibration of the b-jet energy scale with it. The Z->bb decay makes it easier to find the hadronic vector boson  decay signal, because the QCD background is poor in b-jets at that energy.

    Your question is interesting and I will make a stand-alone post with it, with links and pictures, so come back if you are interested. I hope to publish it later today or tomorrow at the latest.

    Cheers,
    T.

    Thank you. Just to be sure: could you see at TeVatron, despite the QCD background, the single Z or W production followed by hadronic decay?

    CDF studies the jet energy scale (JES) in gamma+jet, Z+jets, and ttbar->WbWbar events. In the former two samples the jets come both from quarks and gluons; the jet energy scale is seen to be controlled to the claimed 3% level. Ttbar events allow JES calibration from the W->qq decays, i.e. a pure quark sample. The latest CDF data sum up to about 5000 Ws analyzed in three different top-antitop decay modes
    (see http://www-cdf.fnal.gov/physics/new/top/public_mass.html). An average of these three measurements would bring the CDF jet energy scale for light quark jets to be roughly 0.5%+-1%, i.e. perfectly consistent with zero and with very good precision. That means that apparent shift in the diboson peak CANNOT be produced by a jet energy scale shift in the diboson itself. It can come on other hand from a data-MC shift in the W+jets background modeling of some quantity that affects Mjj, but it does not have to be necessarily the JES! JES is just the most trivial you can think here...

    Indeed, is it nice that the authors of the paper mention deltaR. While they have difficult choice to re-weight it without underlying model, maybe it was now a good time to publish this paper so theorists can also participate on studying that shift. That deltaR seems to be a hint from the CDF, maybe they know more than what they publish. And that might be too difficult for the JES attackers, they are not that far understanding their data yet. What it comes to gluon and quark jets, there are papers out how to select 90% contamination of gluon jets and 90% contamination of quarks jets, sound maybe a bit bold, but if it is even close to true, that eats JES attackers one more bullet.

    Hi Tommaso-

    "To put this in perspective, you should know that in 2006 CDF published (actually a few physicists from CDF, a dozen of them to be exact, and among them yours truly) an analysis of their Jet energy scale, assessing at 3% the uncertainty on that quantity."

    Tangential to the post, certainly - but this caught my eye. I assume you're talking about the JES NIM?
    Nucl.Instrum.Meth.A566:375-412,2006 http://arxiv.org/abs/hep-ex/0510047
    It certainly had more than a dozen authors. Looks like about 42 names (including both of ours).
    But, certainly a "small" author list by CDF and experimental HEP standards.

    Cheers,
    Adam

    dorigo
    Hi Adam,

    you're right - I confused myself with the other paper that's been quoted in this thread, the one on the JES with Z->bb. _That_ one had a dozen authors.

    Cheers,
    T.
    Tommaso (both Tommasos actually),
    to what extent (if any) are your feelings about jet energy scale at CDF
    influenced
    by the CMS circumstances described by Sertac Ozturk
    in his thesis fermilab-thesis-2011-05
    (especially chapter 5 Measurement of Dijet Mass Spectrum):
    "... A dijet mass preselection of corrected Mjj greater than 100 GeV was performed
    because jet energy correction where dijet mass was less than 100 GeV was not reliable
    ...
    The HTL_50U trigger becomes fully efficient when the dijet mass is equal to 220 GeV ... Thus, dijet mass spectrum was started from 220 GeV ...".

    A couple of naive outsider questions:

    1 - Does the unreliability of dijets less than 100 GeV apply to both CDF and CMS, or only to CMS ?

    2 - Does the fact that, due to its trigger, the CMS starts at 220 GeV indicate that CMS might have diffficulty in replicating the CDF bump results of Cavaliere et al?

    Tony

    dorigo
    Hi Tony,

    CMS has a better calorimeter than CDF, but it is still in its infancy with respect to its older uncle. I have not read in detail the analysis you are referring to, but the high threshold may be due to the use of jets measured in the forward part of the detector, where jets with transverse energy below 30 GeV or so cannot be easily calibrated (and there are no charged tracks to use for a tuning).

    The jets in the Wjj CDF analysis are central, and the CDF energy corrections work down to 15 GeV of transverse energies. So the two things cannot be easily compared.

    Further, the turn-on of the trigger is hard to model to the precision required by the measurement that the guy is making in the analysis you mention. He is interested in the high-mass end of the spectrum, so he has no problem in cutting away the low mass part. If he were interested in the lower energy part, he would have a different trigger (note that the jets in the CDF analysis are not used in the trigger selection of the events).

    Finally, there is no question that CMS can and will do this analysis.

    Cheers,
    T.