Optimization of selection and data-driven background estimation for the boosted di-higgs to 4b final state search

Transcription

1 Optimization of selection and data-driven background estimation for the boosted di-higgs to 4b final state search Alison Marsh Advisor: Dr. Emily Thompson August 1, 2014 Abstract This paper presents the optimization of a search for a new heavy resonance decaying into two boosted Higgs bosons, with two bb pairs in the final state. This search employs novel track jet b-tagging in order to increase signal efficiency previously limited by b-tagged R = 0.4 calorimeter jets. The analysis also utilizes a data-driven background estimation in order to predict gluon splitting background that previously kept this Higgs decay mode from being observed. Several selections of the search are defined, motivated, and varied in order to increase the significance of signal over background. These selections are then tested using the data-driven background estimation in order to ensure their effectiveness for this analysis. These selections show promising results for future runs of this analysis as the LHC turns on again in 2015 with its nominal energy of 7TeV from each beam. 1

2 Contents 1 Preface CERN and the LHC ATLAS The Standard Model Introduction Calorimeter jets and jet grooming Track jets and b-tagging Analysis Motivation and strategy Analysis selection baselines Data-driven background estimation Selection optimization Track jet p T selection Large-R jet p T selection Conclusions Future work Acknowledgements 13 7 Bibliography 15 A Event charts 16 A.1 Track jet A.2 Leading large-r jet B Background estimation comparison 19

3 Figure 1: The Large Hadron Collider. [1] 1 Preface 1.1 CERN and the Large Hadron Collider CERN, the European Organization for Nuclear Research, was founded in 1954 by the 12 founding Member States: Belgium, Denmark, France, the Federal Republic of Germany, Greece, Italy, the Netherlands, Norway, Sweden, Switzerland, the United Kingdom, and Yugoslavia. Consisting now of 21 member states, CERN straddles the border between France and Switzerland near Geneva. Its main goal is to probe the fundamental structure of the universe and study the fundamental particles. One of CERN s main structures is the Large Hadron Collider (LHC), a 27-kilometre particle accelerator. First started on September 10, 2008, the LHC is the world s largest and most powerful accelerator, consisting of a ring of superconducting magnets that boost particles around a large circle. These particles are brought together in collisions at experiment detectors. The four main particle detectors at the LHC are ATLAS, CMS, ALICE, and LHCb. 1.2 ATLAS ATLAS, A Toroidal LHC ApparatuS, is a general-purpose detector at the LHC and therefore investigates a wide range of physics. Weighing in at about 7000 tons, the ATLAS detector measures 46 meters in length and 25 meters in height. The ATLAS detector consists of four major components: the inner detector, the calorimeters, the muon spectrometer, and a system of magnets. The inner detector tracks particle trajectories and measures particle momenta; the inner detector is especially important for this analysis since we use its silicon tracker for track jet b-tagging. There are two calorimeters in ATLAS: the electromagnetic calorimeter and the hadronic calorimeter. The electromagnetic calorimeter measures the energy of electrons and photons while the hadronic calorimeter measures the energy of hadrons. The muon spectrometer identifies and measures the momenta of muons. The system of magnets, such as the one labeled Barrel Toroid in Figure 2, bend charged particles in order to make momentum 1

4 Figure 2: The ATLAS Detector. [2] measurement possible. Particle interactions in ATLAS are processed as data through a trigger system that selects interesting events and allows them to be transferred to storage through the data acquisition system. Then the computing system analyzes the millions of events and 3200 terabytes of data recorded each year. More than 3000 scientists from 38 different countries collaborate on the ATLAS experiment. 1.3 The Standard Model A cornerstone of particle physics, the Standard Model explains how the basic building blocks of matter interact, governed by four fundamental forces. The Standard Model consists of matter particles, forces, and their carrier particles. Matter particles come in two basic types, quarks and leptons, and are the building blocks of matter. Each of these two groups consists of six particles which are joined in pairs called generations. The six quarks are the up quark and down quark in the first generation, the charm quark and strage quark in the second generation, and the top quark and the bottom quark in the third generation. The six leptons are similarly in three generatons consisting of the electron and the electron neutrino, the muon and the muon neutrino, and the tau and the tau neutrino. The four fundamental forces are the weak force, the strong force, the electromagnetic force, and the gravitational force. All of them except for the gravitational force are described by the Standard Model. Particles of matter transfer energy through the exchanging of these force carrier particles. The strong force is carried by the gluon, the electromagnetic force is carried by the photon, and the weak force is carried by the W and Z bosons. In order to create a complete picture, the Standard Model predicts a fourth carrier particle, the graviton, that carries the gravitational force. It has yet to be discovered, however this generally does not affect particle physics since the effects of gravity on the minuscule scale of particles in negligible. Though the Standard Model creates the best description we have of matter and forces, it is still incomplete. Its inability to describe gravity in the same way as the other forces is not its only short coming. The Standard Model also 2

5 Figure 3: The Standard Model of Particle Physics. [3] fails to answer the questions of dark matter, the disappearance of antimatter following the big bang, the difference in mass scale between the generations of quarks and leptons, and much more. Even so, a recent discovery has given strength to the Standard Model : that of the Higgs boson. On July 4, 2012, the ATLAS and CMS experiments at the LHC announced the observation of a new particle in the mass region predicted for the Higgs (around 126 GeV). Studies of the Higgs, its properties, and its implications are ongoing at CERN. 2 Introduction A few definitions are required for discussion of the analysis. 2.1 Calorimeter jets and jet grooming The jets discussed in this report are a spray of particles and/or energy that occurs as a result of the primary pp collisions at the LHC. The high-luminosity, or high number of events detected, at the LHC causes soft particles unrelated to the hard scattering of the pp collision to contaminate jets and therefore make it more difficult to resolve particles originating from the hard interaction. This is especially true of boosted objects like those discussed in this report. Standard inputs to jet reconstruction for fully reconstructed calorimeter jets are threedimensional topological clusters that have been corrected for noise. The energy of these clusters are then corrected and weighted based on the likeihood that they originated from electromagnetic or hadronic interactions. This analysis makes use of calorimeter jets with distance parameters of R=0.4 ( small-r ) and R=1.0 ( large-r ). Large-R jets are utilized directly in the analysis while small-r jets are mainly for event cleaning and data quality purposes. Due to the soft scattering present at the LHC, jet grooming is neccessary to reduce pile-up from soft particles unrelated to the hard collision that we re interested in studying. ATLAS employs trimming as its standard jet grooming technique. The procedure of trimming [4] uses an algorithm to create subjets 3

6 Figure 4: Jet trimming procedure. [4] from the original calorimeter jet. Any subjets with pi T p jet T < f cut are removed, where p i T is the transverse momentum of the ith subjet, and f cut is a parameter of the method, which is usually a few percent. In this analysis, an f cut of 5% was applied; thus subjets whose p T constituted less than 5% of the p T of the large-r jut were removed. Subjets that pass this cut make up the trimmed jet. While effective, this technique has less of an effect on the mass boosted jets due to the high likeihood that they contain many subjets of high momentum. Trimming jets in this manner removes pile-up from soft-scattering so that the subjets remaining are evidence of hard-scatter interactions. 2.2 Track jets and b-tagging Track jets are defined as jets that contain a track of a charged particle. When reconstructing track jets, track quality selections are imposed in order to ensure good tracks that originate from the primary vertex of pp collision. These selections require tracks to have p T > 500 MeV and η < 2.5. They must also pass hit criteria and impact parameter constraints. Tracks must have at least one pixel hit and at least 6 hits in the silicon tracker. The absolute value of the track θ and the absolute value of θ sin(θ), both with respect to the primary vertex, must be less than 1.5. Finally, for this analysis, track jets must be formed from at least two tracks originating from the primary vertex and must have p T > 20 GeV. Only these track jets are considered for b-tagging. The analysis uses R = 0.3 track jets. After strict track quality selections are imposed to ensure that tracks originate from the primary pp collision, b-tagging is accomplished by searching for a secondary vertex in the area of interest around these strictly selected tracks. This secondary vertex is formed by the decay of the b-hadron and is separated from the primary vertex by the observable lifetime of the b-hadron, as depicted in Figure 5. This analysis makes use of the MV1 b-tagging algorithm, inputting all tracks within R < 0.3 of the track jet axis as inputs. [7] 3 Analysis 3.1 Motivation and strategy Many new physics models predict resonances on a TeV-scale decaying to pairs of electro-weak bosons [6]. This analysis presents a search for a new resonance decaying into a pair of boosted Higgs bosons with a final state four b quarks. 4

7 Figure 5: Jet from a bottom quark. [5] A motivator for this analysis is to study the newly-commissioned track jet b- tagging and the bbbb (or 4b ) final state of two Higgs. This final state, while boasting a very dominant branching ratio, is usually overwhelmed by background, mostly QCD multijet events with a small contribution from tt and Z jets. We address this background issue by imposing p T selections on the large- R jets and a mass window around the Higgs mass, as described in Section 3.3. Though a search for resonant di-higgs production in the 4b final state has been conducted by ATLAS [8], this study only searched up to potential resonance masses of 1.6 TeV and with standard calorimeter b-jets. The motivation for moving into the boosted regime comes from the angular separation of the b hadrons. Since we re interested mainly in potential resonance masses of 1 TeV and above, only part of the new particle s rest mass will be converted into the mass of the Higgs bosons. The rest of it will be converted into the momenta of the Higgs and, eventually, the momenta of the b s. As the momenta of these particles increase, it will result in Lorentz contraction and therefore the angular separation of the decay products will decrease. As they come closer together, the two R = 0.4 b-jets resulting from the two b- hadrons will begin to merge in the detector, thus making them impossible to differentiate. Moving into a boosted regime allows us to contain these two b- jets in a large-r calorimeter jet and use track jets for b-tagging, rather than the standard calorimeter jets. Thus, the use of standard R = 0.4 b-tagged jets will result in a loss of signal efficiency in the RSG mass regime we are interested in exploring, mainly around and above 1 TeV. This motivates our use of track jets for b-tagging, as they allow the use of smaller-r jets of R = 0.2 or 0.3. This enhances signal efficiency in the boosted regime and eliminates the limiting factor that standard R = 0.4 b-tagged calorimeter jets impose. 5

8 3.2 Analysis selection baselines As this report will study varying the selections of our analysis, it is important to start with a definition of our current baseline selections. We currently employ an MV1 > 0.70 selection and a coupling of c = 2.0. We are currently using a track jet p T cut of 20 GeV, a selection that we work to optimize in this report. Our current cuts on the two large-r jets are motivated by the angular separation of the b-hadrons. The angular separation of the 2-body decay products of a heavy particle is approximately R = 2m p T where R = y 2 + φ 2 between the decay products, and p T and m are the transverse momentum and mass, respectively, of the boosted decaying particle, which in this case is the Higgs. Therefore we expect that, in order for the two b-hadrons resulting from the decay of a boosted Higgs to be contained within an R = 1.0 calorimeter jet, the boosted Higgs must have a p T > 250 GeV. Thus our selection on both the leading and subleading large-r jets is p T > 250 GeV. This applies nicely to our analysis in that the leading pt Higgs boson is almost always boosted above p T = 300 GeV at m RSG = 1 TeV. We also employ a selection specifically on the leading large-r jet in order to control the tt background [9]. Referring back to our equation for the angular separation of decay products from a heavy particle, a baseline p T cut of 350 GeV keeps the top quark with its mass of 173 GeV out of our signal elipse. This enhances our sensitivity to signal by cutting out the background in our region of interest. 3.3 Data-driven background estimation The majority of background in the analysis comes from the QCD process of gluon splitting to bb. A very small percentage of the background also comes from tt as previously discussed as well as Z jet background; these types of background can be predicted using Monte Carlo simulation. However due to low statistics, using an analytic fit to predict QCD background isn t effective. Therefore, a data-driven QCD background estimation from a two b-tag sample ( 2b ) is scaled in order to determine the four b-tag ( 4b ) background we can expect. The 2b sample is a collection of dijet events where either the leading or the subleading large-r jet contains two b-tagged track jets. The other large-r jet may have one b-tagged track jet or none at all. All background estimation is preformed before the execution of p T cuts on the jets. A sideband, control, and signal region have been defined in a 2-dimensional mass plane in order to execute this background prediction. In this mass plane, the x-axis corresponds to the mass of the leading large-r jet while the y-axis corresponds to the subleading large-r jet mass, as shown in Figure 6. The control region is defined as a box around the HH signal that includes events with a leading large-r jet mass of 90 GeV < m jet < 160 GeV and a subleading large-r jet mass of 85 GeV < m jet < 155 GeV. Events considered in the control cannot be inside the signal region which is currently blinded. The sideband is defined as all events outside of the control box in this 2-dimensional mass plane. 6

9 (a) Sideband region region Figure 6: Region definitions in the 2-D mass plane. [10] A normalization factor, called µ QCD is determined based on 2b and 4b data in the sideband region. This normalization is then tested by using the 2b data in the control region to predict the control region s 4b data by applying the normalization factor from the sideband. After testing in the control region, this normalization factor can be used to predict the QCD background expected in the signal region. Figures 7, 8, and 9 compare the 2b to 4b data in the sideband and control regions. The histograms in the sideband are fit to each other by construction and therefore test the fitting procedure. The histograms in the control were created by applying the normalization factor, µ QCD, from the sideband region to the 2b data in the control. These histograms are created in order to validate our datadriven background estimation. The three blue sections along with the green section make up the 2b QCD background. The two darker blue sections are events in which the leading large-r jet contains two b-tags and the subleading jet contains one or none. The green and lightest blue sections are events in which the subleading large-r jet contains two b-tags and the leading jet contains one or none. The white and orange sections are tt and Z jet background, respectively. The black dots represent the 4b data. We validate our background prediction by comparing the shape of the 2b data, the multi-colored histogram, to the 4b data, the black histogram, in the control region. As shown by the relatively good fit in Figures 7, 8, and 9, this background estimation method is currently effective for our baseline selections. 4 Selection optimization 4.1 Track jet p T selection In order to enhance our signal efficiency in the analysis, a p T cut is being made on the b-tagged track jets. The lower this cut gets, the more signal we detect therefore increasing our sensitivity. The analysis is currently employing a p T cut of 20 GeV on b-tagged track jets. One suggestion for optimizing our search is to make lower track jet p T cut of 7 GeV. Figure 10 shows a significance plot that 7

10 (a) Sideband Figure 7: Dijet mass spectrum with baseline selections. (a) Sideband Figure 8: Mass of the leading large-r jet with baseline selections. (a) Sideband Figure 9: p T of the leading large-r jet with baseline selections. compares these two selections. Significance is a way to quantize how well we cut out background while maintaining signal by calculating the number of signal events divided by the number of background events in our signal ellipse. It is important to note that this plot is based on signal and background present in varying mass windows around each potential resonance mass point. Therefore, we only take into account the background that is present in the same area as the signal and would thus affect our sensitivity to the signal. With increasing 8

11 Figure 10: Significance plot comparing track jet p T selections of 7 GeV and 20 GeV. resonance mass, the width of the signal also increasing. This spreading of signal causes an inclusion of a wider width of background as well which, along with decreasing statistics, accounts for the decrease of significance at higher potential resonance masses. As the figure displays, a new selection of track jet p T > 7 GeV would increase the significance, and in theory our signal sensitivity, by a nominal amount. The other selection we currently employ are optimized to the track jet p T > 20 GeV cut and thus may constitute a reason why the different in significance is not more pronounced. Unfortunately, track jets with momenta below 20 GeV have not been studied widely up to this point. Thus, cutting at 7 GeV opens the analysis up to greater uncertainties, especially related to charge fluxuation within the jet. Taking this uncertainty into account, the slight increase in significance that a lower p T selection on the track jet would cause is not worth the uncertainties that come with it for this analysis. 4.2 Large-R jet p T selection The analysis is currently employing a p T cut of 350 GeV on the leading large-r jet and 250 GeV on the subleading large-r jet. The general 250 GeV p T cut explained in Section 3.2 is used in order to force the two b-tagged track jets to be fully contained within a R = 1.0 large-r calorimeter jet and the 350 GeV p T cut on the leading large-r jet is currently being made in order to cut out tt background. Optimization of the p T cut on the leading large-r jet has been studied with 350 GeV as a threshold. The motivation for increasing this cut is to cut out as much background as possible while maintaining signal. In order optimize this selection, we wished to make the highest p T cut possible while maintaining 90% signal efficiency in our signal region. Figure 11 shows the leading large-r jet p T 9

12 Figure 11: Momenta of the leading large-r jet for varying potential resonance mass. for signal events at different potential resonance masses between 0.8 TeV and 2.2 TeV. An initial cut of 250 GeV has already been made. Based on this histogram, we integrated over these events until we reached 0.9 and displayed the p T at which this value was reached for each potential resonance mass in Figure 12. As this efficiency plot shows, possible p T cuts increase with increasing potential resonance mass up until about 1.6 TeV. As this point the widening of the signal, as evidenced in Figure 11, pushes the p T cut back down. These calculations led us to study new potential large-r jet p T cuts from our threshold of 350 GeV up to 500 GeV in 50 GeV increments. In order to study the effect of these varying large-r jet selections on our signal sensitivity, we once again calculated the significance. Figure 13 shows the significance of signal over background for leading large-r jet p T cuts of GeV by quantizing the number of signal events over the square root of the number of background events for potential resonance masses. It is again important to note that this plot was also created based on mass windows around each signal histogram and therefore a wider window was required at higher potential resonance masses due to the tendency of the signal to spread with increasing resonance mass. This, along with lower statistics in these regions, causes the significance to fall off at higher potential resonance masses. Our current leading large-r jet p T cut of 350 GeV is oulined in green in Figure 13 while higher cuts are in blue, purple, and light blue. It is apparent that our current 350 GeV selection boasts a higher significance for resonance masses around 0.8 TeV. However, since our analysis is focused on the boosted di- Higgs, our main interest is at resonance masses above 1 TeV. At these potential resonance masses, a leading large-r jet p T cut of 450 GeV or 500 GeV yields stronger significance than our current cut. This conclusion can also be drawn from Table 1, which displays the number of signal and background events in the control region for each potential large- R jet p T cut for a resonance mass of 1 TeV. As evidenced in the table, even 10

13 Figure 12: p T cut on leading large-r jet to maintain 90% of signal events. Figure 13: Significance of signal over background for varying potential resonance mass for pt cuts on the leading large-r jet. Table 1: Events in control for resonance mass 1 TeV p T cut 350 GeV 400 GeV 450 GeV 500 GeV signal background

14 (a) Sideband Figure 14: Dijet mass spectrum with large-r jet p T > 450 GeV. (a) Sideband Figure 15: Mass of the leading large-r jet with leading large-r jet p T > 450 GeV. increasing the leading large-r jet p T cut to 400 GeV cuts our background in half in the control region while maintaining most of the signal. By cutting out more signal than background events, we increase our signal sensitivity. Unfortunately, at high momenta large-r jet selections such as 450 GeV and 500 GeV, we experience a large loss of statistics, which hinders us from validating the background estimation process. The low number of QCD background events in the control region leads to greater errors in the 4b data set. This is evident in the background estimation plots show in Figures 14, 15, and 16. These histograms compare the 2b and 4b data in the sideband and control regions for variables including dijet mass, mass of the leading large-r jet, and p T of the leading large-r jet. Lower statistics make it harder to compare the shapes of our 2b and 4b data and therefore prevent us from validating our µ QCD normalization factor in the control. Since this data-driven QCD background estimation is one of the cornerstones of our analysis strategy, high leading large-r jet p T cuts may not be feasible in this run of the analysis. 12

15 (a) Sideband Figure 16: p T of the leading large-r jet with leading large-r jet p T > 450 GeV. 5 Conclusions This report presented the optimization of our search for a new resonance decaying into a pair of boosted Higgs with a final state four b quarks. This optimization was achieved through studying the significance of different selections as well as their effect on our data-driven background estimation. Based on these studies, our current track jet p T cut of 20 GeV is a more ideal selection than the proposed lower selection of track jets with p T > 7 GeV. Without further low p T track jet performance study, the slight gains in significance at a 7 GeV track jet p T cut are not enough to justify adding the extra uncertainty that low p T track jets carry. Gains in significance are much more pronounced for higher p T cuts to the leading large-r jet than our current selection of 350 GeV. Unfortunately, our data-driven background estimation cannot be validated for these high p T cuts; therefore, our current leading large-r jet selection seems to be the most optimal for this analysis. 5.1 Future work The new selections discussed in this report show considerable promise for increasing sensitivity in future analyses that study this 4-b final state of boosted di-higgs. After further performance work has been done to fully understand low p T track jets, a study could be optimized to use track jets of p T > 7 GeV in order to increase the significance of signal over background. This gain in significance is even more pronounced for leading large-r jet cuts higher than the one we employ in our analysis. I highly recommend leading large-r jet selections of higher than 350 GeV for future runs of this analysis, once the data-driven background estimation can be adjusted to account for issues with shape validation in the control due to low statistics. 6 Acknowledgements I would like to thank the Columbia REU program and Professor Parsons for this incredible opportunity to conduct research at CERN this summer. I would specifically like to extend my thanks the Columbia ATLAS team stationed at CERN for all of their help during my time here. To my advisor, Dr. Emily 13

16 Thompson, for her incredible help, guidance, and mentorship throughout the program. To Lei Zhou for his help with the code that was essential to this study. I would also like to thank the National Science Foundation for their support of REU programs like this one that encourage graduate study by enabling undergraduate students to experience scientific research hands-on. 14

17 7 Bibliography References [1] home.web.cern.ch [2] [3] [4] ATLAS Collaboration, Performance of jet substructure techniques for large- R jets in proton-proton collisions at sqrt(s) = 7 TeV using the ATLAS detector, CERN-PH-EP , Jun [5] [6] K. Agashe, H. Davoudiasl, G. Perez, and A. Soni, Warped Gravitons at the LHC and Beyond, Phys.Rev. D , Aug [7] N. Bruscino, D. Duda, Y. Enari, F. Filthaut, S. Fleischmann, M. Kagan, A. Kobayashi, T. Scanlon, A. Schwartzman, N. Tannoury, E. Thompson, W. Van Den Wollenberg, Q. Zeng, L. Zhou, Flavor Tagging with Track-Jets in Boosted and Resolved Topologies, Internal note, Jun [8] ATLAS Collaboration A search for resonant Higgs-pair production in the bbbb final state in pp collisions at s = 8 TeV, ATLAS-CONF , Mar [9] ATLAS Collaboration, Performance of boosted top quark identification in 2012 ATLAS data, ATLAS-CONF , Aug [10] M. Bellomo, M. Kagan, E.N. Thompson, L. Zhou, Search for a resonance in the boosted di-higgs to 4b final state, Internal note, Jul

18 A Event charts Below are charts listing the number of signal and background events in the sideband and control regions for each selection mentioned in this report as well as the expected number of events in the signal, based on QCD background estimation. A.1 Track jet Figure 17: Current track jet p T > 20 GeV. Figure 18: Track jet p T > 7 GeV. 16

19 A.2 Leading large-r jet Figure 19: Leading large-r jet p T > 350. Figure 20: Leading large-r jet p T >

20 Figure 21: Leading large-r jet p T > 450. Figure 22: Leading large-r jet p T >

21 B Background estimation comparison Below are histograms that compare the 2b to 4b data in the sideband and control regions for the two remaining large-r jet p T selections mentioned in this report: 400 GeV and 500 GeV. (a) Sideband Figure 23: Dijet mass spectrum with leading large-r jet p T > 400 GeV. (a) Sideband Figure 24: Mass of the leading large-r jet with leading large-r jet p T > 400 GeV. (a) Sideband Figure 25: p T of the leading large-r jet with leading large-r jet p T > 400 GeV. 19

22 (a) Sideband Figure 26: Dijet mass spectrum with leading large-r jet p T > 500 GeV. (a) Sideband Figure 27: Mass of the leading large-r jet with leading large-r jet p T > 500 GeV. (a) Sideband Figure 28: p T of the leading large-r jet with leading large-r jet p T > 500 GeV. 20