A Halloween “Ghost” Story

New Physics at CDF? A Halloween “ghost” story.

Study of multi-muon events produced in pp collisions at !s = 1.96 TeV

arXiv:0810.5357 (31 Oct 08)

Flip Tanedo 4 November 2008

CIHEP Collider Club Hype Caveat Emptor

• This is a controversial result • Only 2/3 of CDF on the author list • Lots of hype from blogs, etc. Latched on to Nima-Neal DM hype • Work in progress. • I don’t have the answers. Maybe you do.

Happy Halloween I’ll tell you a story. Then you tell me what kind of story it was. • True story? Actually a potential signal of new physics • Fairy tale? Signal turns back into a pumpkin at midnight • Morality play? Not literally true, but teaches us lessons

Johannes Pumpkin is burning to ﬁnd out if the CDF anomaly is new physics!

4 bb inconsistencies

• R 2 b =( σ b b ) exp / ( σ b b ) NLO should be " 1 • R = 1.15 ± 0.21 when using 2nd’ry vertex ID • R = 3.0 ± 0.6 when using semileptonic decays

• Invariant mass spectrum doesn’t ﬁt well with simulation of sequential semileptonic b decays

• Time-integrated mixing probability of b hadrons “signiﬁcantly larger” than LEP

5 New result: R ﬁxed

• Phys. Rev. D 77, 072004 (2008) Measurement of correlated bb production in pp collisions at !s = 1960 GeV

• R = 1.15 ± 0.21 • So what gives? • Used tight SVX selection criteria, muon parents decay within 1.5 cm of beam (cf loose SVX) Silicon Vertex Detector The call is coming fromBEAM PIPE OUTSIDEinside the house

• Anomalously large number of muons produced outside the beam pipe • “ghost” muons • Unusual multiplicity of muons in B events (historically how they were discovered) ... and that’s why Neve decided to purchase a cell phone The call is coming fromBEAM PIPE OUTSIDEinside the house

• Anomalously large 6 number of muons 10 produced outside the 105 ghost beam pipe 104

103 • “ghost” muons 102 Muons / (0.008 cm) • Unusual multiplicity 10

of muons in B events 1 (historically how they were qcd discovered) 0 0.5 1 1.5 2 d (cm)

FIG. 7: Impact parameter distribution of muons contributed by ghost ( ) and QCD (histogram) • events. Muon tracks are selected with loose SVX requirements. The detector resolution is 30 µm, Impact parameter distribution ! whereas bins are 80 µm wide.

1/ptrack)2/σ2 ] to measure the diﬀerence between the momentum vectors of the undecayed T 1/pT pions or kaons (h) and that of the closest reconstructed tracks 3. Figure 9 shows the ∆ distribution as a function of R, the decay distance from the beamline. One notes that most of the decays at radial distances R 120 cm yield misreconstructed tracks. The numbers ≤

3 −3 −3 −1 The assumed experimental resolutions are σ [rad] = σ = 10 and σ1 =2 10 [GeV/c] . φ η /pT ·

21 “Ghost” Properties

Type Total Tight SVX Loose SVX All 743,006 143,743 590,970 QCD 589,111 143,743 518,417 Ghost 153,895 0 72,553

• 2x tracks in a 36.8˚cone • 4x real # in a 36.8˚cone • Independent of luminosity, multiplicity of pp interactions

• Impact parameter distribution could ﬁt NP with ! " 20 ps Won 2 Oscars “Ghost” Properties

Type Total Tight SVX Loose SVX All 743,006 143,743 590,970 QCD 589,111 143,743 518,417 Ghost 153,895 0 72,553

• 2x tracks in a 36.8˚cone • 4x real # in a 36.8˚cone • Independent of luminosity, multiplicity of pp interactions

• Impact parameter distribution could ﬁt NP with ! " 20 ps “Lepton Jet” 5 “Ghost” Properties105 10 104 104 ! = 16.8 ± 0.5 ps ! = 21.1 ± 0.5 ps

3 Type Total Tight SVX 103 Loose SVX 10

All 743,006 143,743 102 590,970 102

QCD 589,111 143,743 Muons / (0.008 cm) 518,417 Muons / (0.008 cm) 10 10 Ghost 153,895 0 72,553 0 0.5 1 1.5 2 0 0.5 1 1.5 2

dp (cm) ds (cm)

• 2x tracks in a 36.8˚cone 104 104

! = 19.6 ± 1.1 ps ! = 22.4 ± 0.8 ps 3 3 • 4x real # in a 36.8˚cone 10 10 102 102

Independent of luminosity, Muons / (0.008 cm) Muons / (0.008 cm) • 10 10 multiplicity of pp interactions

0 0.5 1 1.5 2 0 0.5 1 1.5 2 d (cm) d (cm) • Impact parameter distribution p s could fit NP with ! " 20 ps FIG. 25: MuonEvents impact with parameter >2 distributions # for events containing (top) only two muons or (bottom) more than two muons in a cosθ 0.8 cone. We call d and d the impact parameter of ≥ p s initial and additional muons, respectively. The solid lines represent fits to the data distribution with an exponential function. The fit result is shown in each plot.

muons due to ghost events in Fig. 7 is derived using muon tracks that pass the loose SVX selection in order to minimize the possible contribution of interactions in the detector systems surrounding the SVXII detector. As mentioned earlier, this requirement sculpts the impact parameter distribution of muons arising from the decay of particles with a lifetime much larger than that of b quarks. The smaller number of events that contain two or more muons

46 5 105 10

“Ghost” Properties 4 104 10 ! = 16.8 ± 0.5 ps ! = 21.1 ± 0.5 ps

3 Type Total Tight SVX 103 Loose SVX 10

All 743,006 143,743 102 590,970 102 Muons / (0.008 cm) Muons / (0.008 cm) QCD 589,111 143,743 10 518,417 10 Ghost 153,895 0 72,553 0 0.5 1 1.5 2 0 0.5 1 1.5 2

dp (cm) ds (cm)

• 2x tracksMy in 510 a 36.8˚cone Lab Data 104 104 ! = 19.6 ± 1.1 ps ! = 22.4 ± 0.8 ps 103 103 • 4x real # in a 36.8˚cone 102 102 Muons / (0.008 cm) Muons / (0.008 cm) • Independent of luminosity, 10 10 multiplicity of pp interactions 0 0.5 1 1.5 2 0 0.5 1 1.5 2

dp (cm) ds (cm) • Impact parameter distribution could fit NP with ! " 20 ps FIG. 25: MuonEvents impact with parameter >2 distributions # for events containing (top) only two muons or (bottom) more than two muons in a cosθ 0.8 cone. We call d and d the impact parameter of ≥ p s initial and additional muons, respectively. The solid lines represent fits to the data distribution with an exponential function. The fit result is shown in each plot.

46 But is it real? i.e. can we get rid of it? But is it real? Could it be ordinary QCD?

• “QCD” = Drell Yan, !, Z0, heavy ﬂavor, ... 106 0.6 (a) (b) 105 0.5 ) 2

We expect 96% of QCD 4 0.4 • 10

within 1.5 cm of beam 0.3 103

SVXII efficiency 0.2 2 Highly boosted hadrons? Events / (3 GeV/c 10 • 0.1 Don’t see any signature in 10 0 50 100 0 50 100 2 2 the invariant mass Mµµ (GeV/c ) Mµµ (GeV/c ) distribution. FIG. 5: Invariant mass distribution (a) of the dimuon pairs used in the study. The efficiency (b) of the tight SVX requirements as a function of the dimuon invariant mass in the data ( ) is compared • • Can also compare detector to that in the heavy flavor simulation ( ). efficiencies ◦

of the dimuon invariant mass. The efficiency of tight SVX requirements in the data is below that in the simulation only for dimuon invariant masses smaller than 40 GeV/c2, and then rises to the expected value of 0.257 where events are mostly contributed by prompt Z0 decays. This feature does not favor the first hypothesis (a). A long-lived particle contribution is suggested by the comparison of the impact parameter distribution of dimuons that pass the loose and tight SVX requirements. The request that muons pass loose SVX requirements is momentarily used to reduce the possible contribution of muons from secondary interactions occurring beyond the SVXII detector. We note that loose SVX requirements sculpt the impact parameter distribution of muons arising from the decay of objects with a lifetime much longer than that of b hadrons, such as π, K, or 0 KS mesons. Two-dimensional impact parameter distributions are shown in Fig. 6. One- dimensional distributions are shown in Fig. 7. The impact parameter distribution of muons in ghost events differs from that of the QCD contribution. According to the heavy flavor simulation [6], dimuons with impact parameter larger than 0.12 cm only arise from b¯b production. We fit the impact parameter distribution in Fig. 8 with the function A exp( d/(cτ)) − in the range 0.12 0.4 cm. The best fit returns cτ = 469.7 1.3 µm in agreement with the − ± value 470.1 2.7 µm expected for the b-hadron mixture at the Tevatron [5]. We conclude ±

19 But is it real? Could it be in-ﬂight decays of K, $?

150 150 • Particles with ! > heavy QCD ﬂavor lifetime (a) (b)

100 100 • In-ﬂight decay to #’s lead to R (cm) R (cm) misreconstructed50 tracks 50 • Simulated with HERWIG 0 0 1 10 102 1 10 102 • Can account for 35% of! ! ghosts, butFIG. only 9: Distribution 10% of ∆ of(see text) as a functionMeasure of the distance ofR $of, theK (a)momentumK and (b) π decay vs. those withvertices d % from 0.5 the beamline.cm For comparison, the analogousclosest distribution reconstructed for real muons from tracks heavy ﬂavor decays does not extend beyond ∆ = 9.

h track h track 2 TABLE III: Number of pions and kaons1 η correspondingη to( aφ misreconstructedφ ) track (∆1 5) with1 ∆2 = − + − +( ≥ )/σ2 2 2 h track 1/pT pT 3 GeV/c and η 0.7, that decay3 insideσ the tracking volume,σ produce CMUPp muons− pwith ≥ | | ≤ η φ T T a L1 primitive, and pass diﬀerent SVX selections.

Selection π K Tracks 2667199 1574610 In-ﬂight-decays 14677 40561 CMUP+L1 1940 5430 Loose SVX 897 3032 Tight SVX 319 1135

CMUP pair due to b¯b production with the same kinematic acceptance (p 3 GeV/c and T ≥ η 0.7). The ratio of the number of pions to that of kaons is approximately 5/1. Each | | ≤ simulated track in the kinematic acceptance is weighted with the corresponding in-ﬂight-

partons with transverse momentum larger than 3 GeV/c [6].

23 But is it real? 0 Could it be punchthough of K s or hyperons?

106 0 0 (a) (b) 3000 K s 105 ) • Hadrons from K s and 2 104

hyperon decay can mimic 3 2000 10 muons 102 1000 Muons / (0.008 cm) 10 Events / (0.002 GeV/c 1 0 + - • e.g. K s "$ $ 0.4 0.45 0.5 0.55 0.6 0 0.5 1 1.5 2 2 M (GeV/c ) d (cm)

FIG. 11: Distributions of (a) the invariant mass of pairs of initial muons and opposite sign tracks Explains about 8% of ghost 106 • 0 (a) and of (b) the impact parameter of initial(b) muons, produced by KS decays, that pass the loose SVX 3000 105 ) events 2 selection. The solid line represents a ﬁt described in the text. In the impact parameter distribution, 104 the combinatorial background is removed with a sideband subtraction method. For comparison, 3 2000 the vertical10 scale in (b) is kept the same as in Fig. 7.

102

1000 0 + − dataMuons / (0.008 cm) 10 set for KS π π decays in which a pion punchthrough mimics the muon signal.

Events / (0.002 GeV/c → We combine1 all initial muon tracks with all opposite sign tracks with pT 0.5 GeV/c ◦ ≥ contained in a 40 cone around the direction of the initial muons. Muon-track combinations Invariant mass and impact0.4 0.45 parameter 0.5 0.55 0.6 0 0.5 1 1.5 2 2 are constrained to arise from a common space point. They are discarded if the three- distribution for opposite-signM (GeV/c muons.) d (cm) dimensional vertex fit returns a χ2 larger than 10. Figure 11 (a) shows the invariant mass 0 FIG. 11: Distributions of (a) the invariant mass ofdistribution pairs of initial of the muonsKS andcandidates opposite reconstructed sign tracks assuming that both tracks are due to pions. A fit of the data0 with a Gaussian function to model the signal plus a second order polynomial and of (b) the impact parameter of initial muons, produced by KS decays, that pass the loose SVX to model the background yields a signal of 5348 225 K0 mesons. The impact parameter selection. The solid line represents a fit described in the text. In the impact parameter distribution, ± S distribution of initial muons produced by K0 decays is shown in Fig. 11 (b). The data also the combinatorial background is removed with a sideband subtraction method. For comparison,S contain a smaller number of cases in which the initial muon is mimicked by the products of the vertical scale in (b) is kept the same as in Fig. 7. hyperon decays. Using a similar technique, we have searched the data for Λ pπ− decays → and we find a signal of 678 60 Λ baryons (see Fig. 12). Since in both case the kinematic 0 + − ± data set for KS π π decays in which a pionacceptance punchthrough times reconstruction mimics the e muonfficiency signal. is approximately 50%, source (c) ( 12000 events) → # We combine all initial muon tracks with all oppositeexplains sign8% tra of thecks ghost with events.pT 0.5 GeV/c ◦ # ≥ contained in a 40 cone around the direction of theThe initial final muons. source Muon-track (d) of ghost combinations events, secondary interactions in the detector volume, is are constrained to arise from a common space point. They are discarded if the three- 25 dimensional vertex fit returns a χ2 larger than 10. Figure 11 (a) shows the invariant mass 0 distribution of the KS candidates reconstructed assuming that both tracks are due to pions. A fit of the data with a Gaussian function to model the signal plus a second order polynomial to model the background yields a signal of 5348 225 K0 mesons. The impact parameter ± S 0 distribution of initial muons produced by KS decays is shown in Fig. 11 (b). The data also contain a smaller number of cases in which the initial muon is mimicked by the products of hyperon decays. Using a similar technique, we have searched the data for Λ pπ− decays → and we find a signal of 678 60 Λ baryons (see Fig. 12). Since in both case the kinematic ± acceptance times reconstruction efficiency is approximately 50%, source (c) ( 12000 events) # explains 8% of the ghost events. # The final source (d) of ghost events, secondary interactions in the detector volume, is

25 But is it real? Could it be secondary interactions?

• Interactions with the

detector volume106 106 (a) (b) 105 105 e.g. detector support • 104 104 structure 103 103

2 2 • Signature: spikesMuons / (1 cm) 10 in Muons / (1 cm) 10 distance to reconstructed10 10 1 1 -20 0 20 40 60 -20 0 20 40 60 vertex R (cm) R (cm)

• Appears to FIG.be 13: negligible Distributions of R, the. signed distance of muon-track vertices from the nominal beam line for (a) QCD and (b) ghost events (see text).

transverse to the beam line, is shown in Fig. 13. The distance R is negative when the secondary vertex is in the hemisphere opposite to that containing the momentum of the muon-track system. Secondary interactions are characterized by spikes at R values where the detector material is concentrated, such as SVX supports or the COT inner support cylinder. From the absence of visible spikes, we conclude that the contribution of multi- prong secondary interactions to initial muons in ghost events is negligible. At the same time, we cannot exclude some contribution to ghost events from elastic or quasi-elastic nuclear scattering of hadronic tracks in the detector material. Our estimate of the size of possible sources of ghost events underpredicts the observed number of ghost events by approximately a factor of two (154000 observed and 69000 accounted for). However, given the possible large uncertainty of the in-ﬂight-decay prediction and the possible contribution of elastic or quasi-elastic nuclear scattering in the detector material, at this point of our study we cannot exclude that the ghost sample can be com- pletely accounted for by a combination of all the above-studied background sources. Were ghost events all due to these ordinary sources, they would not contain a signiﬁcant number of additional real muons. Therefore, these sources are unlikely to be the origin of the excess of low-mass dileptons reported in Ref. [3]. That study is repeated in the next section.

27 But is it real? Does anything else fake multiple muons?

• All of the previous points could plausibly be a conspiracy of Monte Carlos, experiment • But these backgrounds would not account for the additional number of real muons (2x QCD) • Apply tight SVX to multi-# sample, • efﬁciency drops from 0.193 to 0.166 • QCD fakes have much fewer additional muons, would expect efﬁciency to increase to 0.244 (QCD expectation) But is it real? Does anything else fake multiple muons?

8000 )

2 Ghost Ghost • In a cone with cos " < 0.8 8000

6000 2x as many tracks, 4x # 6000

4000 4000

• “Surprisingly large number Muon pairs / (0.01) 2000 2000 of tracks with pT % 2Muon pairs / (0.5 GeV/c GeV 0 5 10 15 20 -1 -0.5 0 0.5 1 2 Mµµ (GeV/c ) cos!

• Shapes of impact parameter 15000 nd’ry8000

) 5 15 distribution for 2 2 differsQCD QCD (a) (b) from QCD; large tail,60004 like 10000 10 ) primary #s (correlated)2 40003 5000

2 Muon pairs / (0.01)

M (GeV/c 2000 5 Muon pairs / (0.5 GeV/c 0 Number of tracks • Estimate fakes with 1D 0 5 10 15 20 -1 -0.5 0 0.5 1 decays; doesn’t match high 2 0 Mµµ (GeV/c ) 0 cos! 0 1 2 3 4 5 0 5 10 15 2 multiplicity tail of ghosts M (GeV/c ) Number of tracks FIG. 18: Events with OS initial muon pairs and an additional muon combined with the opposite-

FIG.charge 33: initial Two-dimensional muon. We show distributions the invariant of mass, (a) theMµµ invarian, and openingt mass, angle,M, ofθ, all distributions muons and of (b) these the combinations for the QCD and ghost contributions.◦ total number of tracks contained in a 36.8 cone when both cones contain at least two muons. The QCD and fake muon contributions have been subtracted. tribution of additional muons in ghost events extends to much larger values and is consistent with that of the initial muons. However, the impact parameters of the additional and initial of all muons contained in the 27990 cones containing at least two muons is consistent with

muons are loosely correlated (the correlation factor is ρdpds =0.03). that of the 3016 events in which both cones contain at least two muons. Figure 35 shows The contribution of fake muons is evaluated by weighting all tracks with pT 2 GeV/c, the invariant mass distribution of all muons and all tracks with pT 2 GeV/c≥in events in η 1.1, and contained in a cos θ 0.8 cone, with the fake probabilities≥ shown in Fig. 14. which| | ≤ both cones contains two or more≥ muons. Table VII lists the number of these tracks for QCD and ghost events. The QCD and Following the procedure outlined in Sec. VII A, we count the number of secondary vertices of two-track systems in events with two cones36 containing at least two muons. Figure 36 shows the average number of secondary vertices in one cone as a function of the number of secondary vertices in the other cone.

VIII. CONCLUSIONS

We have studied a sample of events containing at least two central muons with p T ≥ 3 GeV/c and invariant mass 5 m 80 GeV/c2. The data sets were collected with ≤ µµ ≤ the CDF II detector at the Fermilab Tevatron collider, and correspond to integrated lu- minosities up to 2100 pb−1. Similar data samples have been previously used by the CDF

and DØ collaborations to derive measurements of the correlated σb→µ,¯b→µ cross section that are inconsistent with the NLO theoretical prediction. A similar data set was used by the

55 Caveat Emptor II

• When only “best” muon tracks selected, signal signiﬁcance decreases

• Estimating the fake rate is hard

• Tagging events is hard; pT cut only below 3-5 GeV ... muons are relatively soft • Analysis is still ongoing... still a few kinks What’s next?

• Does D0 see the same thing? • B analysis is subtle, don’t expect much from the LHC in early years • B factories? • Continue CDF analysis • 1st priority: study electrons In one slide...

• CDF may have found an excess of high-multiplicity, high-impact parameter muons • Doesn’t seem to be background, but this has not yet been ruled out. • Analysis is ongoing. • Model-building “circus” has already begun (prematurely?) Paper trail

• Phys. Rev. D 77 072004 (2008) Main study using tight vs loose SVX • arXiv:0810.5730 Model-building paper...