Banner
    A 2.5 Sigma Higgs Signal From The Tevatron !
    By Tommaso Dorigo | July 2nd 2012 06:39 AM | 38 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
    Robbing the LHC experiments of media attention for 41 hours, the CDF and DZERO experiments are presenting today the results of their searches of the Higgs boson in the full datasets of proton-antiproton collisions acquired in the course of the last 10 years. You can follow the live streaming of the Tevatron seminar at this link.

    UPDATE: the live streaming is here.

    Below I will give some introductory notions on Higgs physics at the Tevatron; at the bottom of this post I am discussing the actual results.

    Introduction

    In the 2-TeV proton-antiproton collisions produced at the Tevatron in the course of its 2002-2011 run, Higgs bosons were produced at a considerably reduced rate with respect to the 8-TeV proton-proton collisions at the LHC. The small rate makes the searches of Higgs signals in the most striking decay modes -pairs of Z bosons or pairs of photons- much less effective in America than in Europe. CDF and DZERO, however, have a small advantage: they can rely on a production mechanism called "associated WH or ZH production" which is less background-ridden than it is at the LHC. So I guess that in order to explain well what is going on in this final rush to the Higgs boson, I need to take a step back and explain these production modes first.

    Production and Decay

    You can produce a Higgs boson in many ways at a hadron collider: via a Drell-Yan process, by vector boson fusion, by gluon fusion, by bremsstrahlung off a fermion or off a vector boson. Each of these processes has a different rate, and distinguishing features. On the left are shown schematic diagrams of the gluon-fusion production mode (top) and the associated production mode (bottom). In the graphs, time flows from left to right. The Higgs boson is supposed to decay to a pair of W bosons in the top diagram, and into a pair of b-quarks in the bottom diagram.

    Indeed, as I mentioned the Higgs boson has several different decay modes available when it disintegrates: WW or ZZ pairs, b-quark pairs, tau-lepton pairs, or photon pairs (there are many others, but these are not experimentally interesting at the moment). More on this below.

    The rates of the most significant Tevatron production mechanisms are shown in the graph on the right as a function of the (assumed unknown) mass of the Higgs boson; the vertical axis is the signal "cross section", a number proportional to the production rate and expressed here in picobarns.

    [If your dataset amounts to a luminosity L of ten thousand inverse picobarns and the cross section is 0.1 picobarns, you expect to have bagged a thousand Higgs bosons, since N=σL, where σ is the symbol for the cross section, and L is the luminosity].

    You immediately observe that all rates decrease with increasing Higgs mass: that is the result of the increased rarity of quarks and gluons carrying a larger and larger fraction of the proton momentum. The energy they carry in fact is required for the creation of the massive Higgs boson.

    So at the Tevatron we are in the ballpark of hundreds of femtobarns for the signal cross section. Making the exercise again to fix the concept in your brain, let us take a 125 GeV Higgs boson and the associated production processes (blue curve plus green curve): you get to calculate that the Tevatron produced N(WH) = 150 fb * 10 / fb = 1500 Higgs bosons in WH production and N(ZH) = 80 fb * 10 /fb = 800 ZH events in the core of each detector.

    Now let us discuss decay modes for a moment. In the graph on the left you can see the principal decay modes of the Higgs boson, again as a function of the particle mass (note that the axes have logarithmic scales so the information is a bit hard to decrypt). For 125 GeV, the dominant decay (the one with the largest probability) is H->bb (in blue), that is to a bottom-antibottom quark pair. These quarks get imparted with about 60 GeV of momentum each, and thus materialize a stream of hadrons, a energetic jet, which can be easily detected and measured in the detectors. But the Higgs can also decay with significant rates to tau lepton pairs, or WW pairs; less frequent are the decays to ZZ pairs and gamma pairs, and indeed at the Tevatron the sensitivity to these final states is very low.

    In the end, the particular combination of production mode and decay mode you pick will define a final state which will include different observable particles. This final state is of course mimicked by some processes that have nothing to do with Higgs production: the latter are called "background", and the name of the game is to select the data such that backgrounds are minimized while signals are retained in a sufficient amount to make it observable.

    At the Tevatron, the associated production of a Higgs boson and a W or Z boson, with the subsequent decay of the Higgs to a pair of b-quarks, is a promising process because W and Z boson decay signals can be collected cleanly when these particles decay to leptons (electrons or muons), and the signature of a W (or a Z) together with a pair of b-quark jets has backgrounds which are manageable (if barely so).

    Note that for a 125 GeV Higgs boson, the decay H->bb is the dominant one, so by concentrating on this channel one is maximizing the number of signal events. On the other hand, the "associated production" mechanism has a rate smaller than the inclusive "gluon fusion" production mode by about a factor of ten (see graph above), so that, too, is a tough choice to make. But perhaps I should not be talking of choices at all here: indeed, the Higgs signal has been sought by CDF and DZERO in virtually all the possible final states -where "possible" should be taken to mean "providing a non-ridiculously-low sensitivity to the signal". Here, however, I am emphasizing the "WH/ZH" production mode and the H->bb decay mode because these are what gives the  bulk of the sensitivity of the Tevatron experiments in their combination.

    Now I should be telling you why, on the other hand, at the LHC the "Wbb" and "Zbb" final states are not as promising (they have been pursued, but are providing only minor contributions to the overall sensitivity). The LHC, having a much higher centre-of-mass energy, produces b-quark pairs of considerable energy associated to W or Z bosons much more frequently than the Tevatron used to do. So, despite the fact that the production rate of Higgs bosons is higher by one order of magnitude as you increase the centre-of-mass energy fourfold, backgrounds are so large in this final state that the LHC sensitivity is actually worse. Here is one instance where the CDF and DZERO experiments have an advantage, other things being equal: and indeed, other things are equal at this time juncture - I am of course talking of the integrated luminosity which the four experiments can analyze right now: 10 inverse femtobarns of data each.

    Enough said about the LHC. So what is the Tevatron showing, after the careful analysis of the full datasets, in search for b-quark pairs associated with a W or Z boson signal ? Well, the title above gave it away already, sort of; but I will wait a few more hours to update this post with the actual results of the experiments... So come back and reload the page, and you will be among the first to know!

    Update 1. As a teaser, here is what I can tell you already about the new Tevatron results:

    - The Tevatron will present a combination of CDF and DZERO search results
    - The updates will be more important for DZERO than CDF: DZERO will update all its results, while CDF will only update their ZH->llbb and VH->vvbb searches (the latter is the signature where you only see two b-jets and there is significant missing energy from the neutrino(s) emitted in the undetected vector boson decay)

    Update 2. There is also going to be a breakdown of the Higgs cross section in the various final states in which the Tevatron experiments have been searching for the particle. The graph will look a bit like the one on the right...

    Update 3. Their observed exclusion for a standard model Higgs boson is Mh<103 and 147<Mh<180 GeV. This is looser than their expected exclusion range (which assumes there is NO Higgs boson of course), which is Mh<120 and 139<Mh<184 GeV. The reason is of course traceable to the broad excess that they already presented at winter conferences. Let me remind you, in fact, that the Higgs mass resolution in the Tevatron experiments, which have most of their sensitivity in the H->bb final state, is only of the order of 15 GeV; so a Higgs-related excess of events will degrade the exclusion power in a broad range.

    Final update: Okay, now the seminar is on, and I can release the results properly. More information is available at the following links:
    - winter 2011 results of Higgs searches
    - CDF Higgs results
    - DZERO Higgs results

    The experiments have made significant progress in the efficiency of b-jet tagging and in the improvement of the mass resolution for pairs of b-quark jets. They both use multivariate discriminants for the selection of the data (Neural Networks and Boosted Decision Trees), and a careful treatment of systematic uncertainties, whose correlation  needs to be assessed across the experiments in the combination of results, as well as across the different channels.

    The selection is validated in a number of ways. For instance, the standard model production of WW and WZ pairs comes to help. When one W boson decays to leptons and the other boson (W or Z) decays to jet pairs, one gets a final state very similar to the one of WH/ZH production. By selecting the data in search for these signals, one can verify that the observed excesses agree with standard model predictions:


    The graph shows a background-subtracted event count as a function of the dijet mass for this diboson selection. Note that here CDF and DZERO data are combined -a rare instance of combination done at histogram level. The black points show the data; the red and yellow histograms show the expected WW and WZ contribution; and the green histogram shows the WH/ZH contribution. Even disregarding the Higgs signal for a moment, one sees that the diboson yield is very well understood in the sample.

    In the end, combining the data from the two experiments, one can put them in different bins depending on the value of the expected signal-to-noise ratio, so that the Higgs signal will show up in the rightmost bins. The figure on the right shows that the data follows very well the expected backgrounds, and that the two most signal-like bins have a slight excess compatible with what one would expect for a 125 GeV Higgs boson present in the sample.

    In the H->gamma gamma search the progress has been mostly on the DZERO side. The figure below shows the extracted upper limit on the rate, as always expressed in units of the expected Standard Model prediction. This is obtained by combining CDF and DZERO results for the searches, which are only marginally sensitive to the Higgs signal: you can observe that the observed limit curve (in black) is higher, by about one standard deviation, than the expected limit, in the region of mass where the LHC experiments have indicated the Higgs boson evidence last December.



    Finally, by running the machinery that extracts limits on the signal rate and p-values for the background-only hypothesis and by combining the H->bb final state results with those of the other searches (H->WW and H->γγ), the Tevatron experiments achieve a nice sensitivity to the Higgs boson. This can be shown in the two figures below.



    The first one above shows the local p-value of the background-only hypothesis as a function of Higgs boson mass. The black curve shows the local p-value of the combined search, which reaches 3-sigma for a mass of 120 GeV or so; the dashed curve shows the p-value that was predicted if the Higgs boson were actually there (all points of the curve refer to independent hypotheses of Higgs mass). There is compatibility of the result with the expectation, but one notes that the p-value is one-sigma too good. In other words, the Tevatron experiments have been "one-sigma lucky", when their median sensitivity for a 125 GeV Higgs would have been just short of two-sigma.

    The last figure, on the right, shows the best-fit Higgs boson cross section, in units of the standard model prediction, for the three studied decay channels. The green band shows the combination of the three results -hopefully accounting for the fact that the error bars are constrained to lie in the xs>0 half of the plane.

    All in all, the signal seen at the Tevatron has a global significance of 2.5 standard deviations. In truth the experiments today are also saying that the highest local significance in the bb final state alone is of 3.2 standard deviations, but this occurs for a Higgs mass hypothesis of 135 GeV, quite far from the true mass of the Higgs boson. True, the experiments do not have the mass resolution of the LHC experiments; but the searches are optimized separately for each mass point, and that is what makes a difference in the p-values of the various searches.

    In other words, it is not legal for a correct statistical assessment to pick the highest local significance mass point of the highest-significance channel and then argue that the mass resolution is scarce and thus this local significance is also a global one: indeed, one still has to correct for the look-elsewhere effect - the multitude of masses that have been analyzed. By doing that, the 3.2 sigma become 2.9, and these 2.9 are still an overestimate, because they are derived by ad-hoc picking the most significant channel alone. The correct result is a 2.5 standard deviation effect, and to that the Tevatron should stick in their press releases, in my humble opinion.

    In any case, due congratulations to the CDF and DZERO colleagues for this endgame result which, while it will not allow them to eat a part of the discovery cake, does show that the Tevatron was sensitive to the Higgs boson in the end. That is a very nice legacy !

    Comments

    The low WW cross section and the high digamma cross section
    are only consistent in a very small overlap of their error bars.

    Since meson-type things (like technicians and my low-mass T0meson)
    can naturally have higher digamma and very low WW cross sections
    it seems to me that the Tevatron could be telling us that
    the thing around 125 GeV may NOT be Higgs,
    but instead a meson-like thing.

    Tony

    PS - For a concrete example of such a meson-type thing that does not carry the baggage of being something in a model that I (with the reputation of an ostracized blacklisted crackpot) thought up,
    look at the paper by Eichten et al in arXiv 1206.0186
    which describes technipions.
    Although Eichten's paper is in the context of the CDF Wjj bump
    similar reasoning might apply to the 125 GeV peak.

    Interested non-particle physicist here. My first time commenting here, so first, thanks for all your updates and in-depth discussions. I've learned a lot!

    Now my question.

    Am I right to note that the best fit cross-section graph is consistent with the ATLAS and CMS results from December, with an excess in H->gamma_gamma and a dearth of H->WW events?

    dorigo
    Hello Dan,
    thanks for the encouragement.

    Yes you are right, the WW is low and the gg is high, as the LHC results show. However, the
    statistical power of the Tevatron data is almost null in these two channels, so it is an observation with very little value - same as picking which among gg or WW has more observed xs at the Tevatron with the flip of a coin.

    Cheers,
    T.
    Typo in my previous message:
    "technicians" should be "technipions".

    My computer autocorrects spelling but does not know that "technipions" is a word.

    Tony

    OK, congrats and all, but only 2.5 sigma? I thought I'd have to retire my "Dirty Higgsy" take-off as out of date, and that level is lower than the four I thought was rather in the bag (even without "metastudy", or did I miss something?) This is still like saying "we know those are the 125 GeV gorilla footprints, but we haven't seen him yet."

    I know what you’re thinking: "Did we find five sigma, or only four?" Well, to tell you the truth, in all this excitement, I’ve kinda lost track myself. But being this is the LHC, the most powerful collider in the world, and would blow your mind clean off, you’ve got to ask yourself one question: "Do I feel lucky?" Well do ya, punk?

    Hank
    2.5 is quite good, especially compared to other sciences.  Heck, climate scientists would be over the moon if they had that accuracy.  Biology too.
    T - for how long would the Tevatron have needed to run to gather the additional statistics to push to a 4 or 5 sigma confidence? So from 10 to ?? inv fb? Cheers

    Heh, well I gather the Tevatron just can't push past around 3 sigma anyway, so this is more about corroboration from another source. More to the point, I was itching to flash around my "Dirty Higgsy" joke again before it's obsolete (when the "punks" can claim their luck, presumably on July 4.) It looks like various studies will be combined for "meta-analysis" as sometimes done in statistics (after all, no difference in principle between which "study" a given event belongs to, to make the sample size. But caution is advised, are the cases really all comparable?) Cheers.

    dorigo
    I think this question is hard to answer precisely, but I remember old forecasts done in 2003, and can actually fish out the graph:


    Now, as you can see the prediction of the 2003 update (the shorter bands) is optimistic, because it was based on a scenario where the detectors would be upgraded in their silicon detectors (this did not happen). However, the old 1999 study (kudos to John Conway for directing that effort!) is not too incorrect, since the Tevatron combination has a sensitivity of 2 sigma for 10/fb/experiment at 125 GeV, while the graph predicts 3.2 sigma or so.

    If we have to extrapolate to 5 sigma... I would say the graph says of the order of 40/fb would be needed. In terms of running time, that means some twelve more years or so!

    Cheers,
    T.
    Thanks Mr. T and quite the memory to be able to pull out those graphs! Theres a bit of me that wonders whether than the expense of LHC (call it $10B) was worth it to save a decade, though I guess something like it would have been built by someone eventually anyways.

    Of course there are hopes for many other discoveries too in the future, with LHCb in particular, though it does seem that at least in the heavy ion/qcd realm RHIC is at least as well position as LHC in most areas.

    Also, Tommaso, you refer to 135 GeV as far from the "true mass" of the Higgs, so are you satisfied it is surely close to 125 GeV?

    dorigo
    Yes I am Neil, but that's no discovery - I have said I was convinced by the December 2011 data already.

    Cheers,
    T.
    BDOA
    Looking at that that Higg to gamma gamma in the Tevatron fine data, is it that a confirmation of the
    gamma gamma excess over a standard model Higgs found in the Dec 2011 LHC data (approx 1.7 times the excepted value). To what extend those the H->gamma gamma measurement at the Tevatron represent an standard model or above standard model figure.

    This is a very interesting figure as its the only non standard model figure, coming out of an otherwise standard model Higgs.
    BDOA Adams, Axitronics
    dorigo
    Hello Barry,

    hell, yes, we can freely speculate. While I think this IS the SM Higgs, and nothing else,
    the inspiring deviations from the picture-perfect "Asimov" data all expect to come out of the LHC (and the Tevatron to some extent) must keep our head spinning.

    (By the way, an "Asimov" dataset is one which behaves EXACTLY as the prediction. That is, observed and expected event counts coincide in each bin. The name of the famous writer Isaac Asimov is used to describe it because of a short story he wrote, when the US elections are decided by the vote of just one "average" person chosen wisely by a computer program).

    Cheers,
    T.
    Very interesting article. Thanks Tommaso for keeping us all updated. Cheers

    So next Wednesday, are you going to title CMS and ATLAS Higgs search significances based on their best channels or combination of all channels? While you say Tevatron should stay in papers in combined results....

    dorigo
    The title on Wednesday will be updated constantly, because I will be blogging live from the main auditorium, and I intend to update the post at 5' intervals.

    Cheers,
    T.
    If anyone was ever in doubt why the CMS and ATLAS experiments at the LHC needs to be blind to one another, here is the answer. No Higgs found at TeVatron for years of running and analysis. A 3 sigma evidence shows up at the LHC around 125 GeV, and suddenly TeVatron sees it too.

    dorigo
    This is quite false and demonstrably so. The Tevatron was seeing 1.3 sigma in the very same region well before we learned something from the 2011 LHC run.

    Cheers,
    T.
    I did not mean to say that TeVatron stole anything or behaved dishonest in any way. I think they have behaved just as they should, and I'm quite impressed that the deadline of prior to this coming Wednesday is actually met.

    All I'm saying is, that the different analysis are not blind to one another anymore. All we're looking for is excesses around 125 GeV. And if one desires to make a claim for New Physics -- which a SM Higgs still is! -- you cannot go in with an a priori knowledge of where to look. And surely the former analysis is compatible with this one. Anything else would be outrageous. But the road to hell is paved with 1.3 sigma deviations.

    Hank
    Sure, I don't think it is opportunism.  I said in a comment on one of your articles three years  Tevatron likely had this (and that's how common knowledge it was to interested people; what do I know?) and you said it again in a 2010 article. It was based on expert intuition, it just took time to sift through it all.

    It isn't like the LHC scheduled a press conference and Tevatron could suddenly rush through billions of collisions and find this.
    Though I do like your informative posts (this one at least as much as previous ones), the american vs european comment is Motl-esque.

    dorigo
    Most unfortunately, you are quite right. Bad companies, bad companies.

    Cheers,
    T.
    Thanks again, Tommaso, for very interesting and informative updates! Tomorrow will be an exciting day, I'll stay tuned! Btw, I also really enjoyed your "Pot-Pourri Of Particle Searches"!

    Hi T,

    This is more a question for the LHC rather than the Tevatron -- to claim a HIGGS discovery (or finding) they would either have to show consistency of BRs with the standard model Higgs (not have 95% exclusion of std model Higgs in any channel -- like diphoton BR being too high or WW BR being too low) or they'd have to establish the spin of the particle to be zero in their measurements. Otherwise how do we know its a Higgs?

    Do you know if they have taken data that can establish the spin to be zero or non-zero and if this will be analysed in time for tomorrow's seminar?

    dorigo
    Hi,
    I think my posts of today answer the question, in particular the one about the CMS searches.
    Cheers,
    T.
    They see exactly three events where the integral of the expected background is 2. (the previous bin contains nothing). So for me it looks like they optimised their cuts on the data.

    dorigo
    You can't be serious!
    Cheers,
    T.
    Well, yes and no. I do trust multivariate techniques, and if the significance is 2.5 than I believe that.
    But what are the systematical errors? To produce the plot one must select different phase space regions with different signal/background content. These contours are then tagged from 0 to 1, or anything that is reasonable. Typically the highest signal/background region is small (or it might be an open area), and I don't doubt that this is the case here, since there is not much kinematical difference between the Higgs decay and the background, only the mass scale preference, so at the it will probably just find the mass peak. In some cases this area is large, like when it is open ended. Lets say it is a closed, small area. How do you determine the qcd->bb there exactly? It is small phase space, you need to generate a lot of QCD for that, which is probably not the case, as everything is data driven and this case it was probably estimated with some q->b tag extrapolation.
    I have to admit, I have no time to read all of their articles. My main problem is that it is hard to believe something that is based on 3 events, because the systematics is never simple.

    dorigo
    You are wrong, the signal is not based on three events, but on a combination of many different channels, with different sensitivities. The human eye indeed cannot appreciate the statistical power of the data by just looking at that summary graph.
    Cheers,
    T.
    http://dkue3ufa3e1f8.cloudfront.net/files/images/sbordered_tev2012.jpg
    This is based on three events. It might have a statistical power of 3-4 sigma, since the expected background is 1/10 here. My question was about the systematics of the background, since the MVA is basically learning the mass (probably just a small phase space region with the same mass). And well, I hope I'm wrong.

    Not only D0 and CDF updated their results
    http://arxiv.org/abs/1207.0449

    but also ATLAS updated their 2011 numbers yesterday
    http://arxiv.org/abs/1207.0319

    Both are reporting 2.9sigma significance. I assume that ATLAS and CMS include some 2012 data into their analysis tomorrow, which should further increase the significance.

    Tommaso,

    a stupid question: how can a Higgs decay into WW, even when the mass of the Higgs is 125 GeV and the two W have a combined mass of 160 GeV?

    dorigo
    Hi Clara,
    how does the neutron decay ? By emitting a W. But how can a neutron, with a mass of 939 MeV, emit a 80.4 GeV particle ? It can't: the W is off mass shell. The same happens in the Higgs decay, where one of the W's is off mass shell, and carries an effective mass of just 125-80 GeV.
    Cheers,
    T.
    The discovery is great, but I heard through the grapevine, uncertain reliability: Joseph Incandela and Fabiola Gianotti, said what was found is not that standard model Higgs boson, but more likely a new boson and still may not explain the origin of mass. True, and if so, what are the implications? Are the presenters being candid enough about how certain they are that this is the classic Higgs (qualitatively, not to be confused with certainty there is really a particle per se, or re it's own mass) or not? tx

    dorigo
    Quite false Neil, nobody in their right mind can make such a statement. Indeed, the signals fit well the Higgs hypothesis. Whether it is a SM Higgs or something else will need much more data to be ascertained, as was known to us since ten years ago.
    Cheers,
    T.
    I have a question regarding this discovery.

    To my understanding, the Higgs boson is the particle that gives all matter its mass. Some particles "interact" with the Higgs energy field full of Higgs bosons, "attracting" Higgs bosons in the process and gaining mass. Please correct me if this is wrong.

    Question is, how did the Higgs bosons themselves gain mass? Do they have it all along? Shouldn't the quest have been about how Higgs Bosons have mass instead of how some of the fundamental particles of the standard model have mass?

    In other words: particle A has mass. The proposition since 50 odd years is that particle A's mass is due to it interacting with Higgs field and "gaining" mass in the process. Why wasn't it taken one step further and why was the question "how do Higgs bosons themselves gain mass" never posed? If it's already posed, or if that's the direction in which the quest is headed, I apologize.

    dorigo
    Dear Kishore,
    the question you ask is a fundamental one, to which there is no answer yet. What is the mass of particles ?
    You might equivalently ask where do the different yukawa couplings of fermions come from. In any case, that is still a mystery.
    Cheers,
    T: