Search for the doubly charmed baryon Xi_cc^+

A search for the doubly charmed baryon Xi_cc^+ in the decay mode Xi_cc^+ ->Lambda_c^+ K^- pi^+ is performed with a data sample, corresponding to an integrated luminosity of 0.65/fb, of pp collisions recorded at a centre-of-mass energy of 7 TeV. No significant signal is found in the mass range 3300-3800 MeV/c^2. Upper limits at the 95% confidence level on the ratio of the Xi_cc^+ production cross-section times branching fraction to that of the Lambda_c^+, R, are given as a function of the Xi_cc^+ mass and lifetime. The largest upper limits range from R<1.5 x 10^-2 for a lifetime of 100 fs to R<3.9 x 10^-4 for a lifetime of 400 fs.


Introduction
The constituent quark model [1][2][3] predicts the existence of multiplets of baryon and meson states, with a structure determined by the symmetry properties of the hadron wavefunctions. When considering u, d, s, and c quarks, the states form SU (4) multiplets [4]. The baryon ground states-those with no orbital or radial excitations-consist of a 20-plet with spin-parity J P = 1/2 + and a 20-plet with J P = 3/2 + . All of the ground states with charm quantum number C = 0 or C = 1 have been discovered [5]. Three weakly decaying C = 2 states are expected: a Ξ cc isodoublet (ccu, ccd) and an Ω cc isosinglet (ccs), each with J P = 1/2 + . This paper reports a search for the Ξ + cc baryon. There are numerous predictions for the masses of these states (see, e.g., Ref. [6] and the references therein, as well as Refs. [7][8][9][10][11]) with most estimates for the Ξ + cc mass in the range 3500-3700 MeV/c 2 . Predictions for its lifetime range between 100 and 250 fs [12][13][14].
Signals for the Ξ + cc baryon were reported in the Λ + c K − π + and pD + K − final states by the SELEX collaboration, using a hyperon beam (containing an admixture of p, Σ − , and π − ) on a fixed target [15,16]. The mass was measured to be 3519 ± 2 MeV/c 2 , and the lifetime was found to be compatible with zero within experimental resolution and less than 33 fs at the 90% confidence level (CL). SELEX estimated that 20% of their Λ + c yield originates from Ξ + cc decays, in contrast to theory expectations that the production of doubly charmed baryons would be suppressed by several orders of magnitude with respect to singly charmed baryons [17]. Searches in different production environments at the FOCUS, BaBar, and Belle experiments have not shown evidence for a Ξ + cc state with the properties reported by SELEX [18][19][20].
This paper presents the result of a search for the decay 1 Ξ + cc → Λ + c K − π + with the LHCb detector and an integrated luminosity of 0.65 fb −1 of pp collision data recorded at centre-of-mass energy √ s = 7 TeV. Double charm production has been observed previously at LHCb both in the J/ψ J/ψ final state [21] and in final states including one or two open charm hadrons [22]. Phenomenological estimates of the production cross-section of Ξ cc in pp collisions at √ s = 14 TeV are in the range 60-1800 nb [17,23,24]; the cross-section at √ s = 7 TeV is expected to be roughly a factor of two smaller. As is typical for charmed hadrons, the production is expected to be concentrated in the low transverse momentum (p T ) and forward rapidity (y) kinematic region instrumented by LHCb [24]. For comparison, the prompt Λ + c cross-section in the range 0 < p T < 8000 MeV/c and 2.0 < y < 4.5 at √ s = 7 TeV has been measured to be (233 ± 26 ± 71 ± 14) µb at LHCb [25], where the uncertainties are statistical, systematic, and due to the description of the fragmentation model, respectively. Thus, the cross-section for Ξ + cc production at LHCb is predicted to be smaller than that for Λ + c by a factor of order 10 −4 to 10 −3 . To reduce systematic uncertainties, the Ξ + cc cross-section is measured relative to that of the Λ + c . This has the further advantage that it allows a direct comparison with previous 1 The inclusion of charge-conjugate processes is implied throughout.
1 experimental results. The production ratio R that is measured is defined as where N sig and N norm refer to the measured yields of the signal (Ξ + cc ) and normalisation (Λ + c ) modes, ε sig and ε norm are the corresponding efficiencies, B indicates a branching fraction, and σ indicates a cross-section. Assuming that B(Ξ + cc → Λ + c K − π + ) ≈ B(Λ + c → pK − π + ) ≈ 5% [5], the expected value of R at LHCb is of order 10 −5 to 10 −4 . By contrast, the SELEX observation [15] reported 15.9 Ξ + cc signal events in a sample of 1630 Λ + c events with an efficiency ratio of 11%, corresponding to R = 9%. For convenience, the single-event sensitivity α is defined as α ≡ ε norm N norm ε sig (2) such that R = αN sig . For each candidate the mass difference δm is computed as where m([pK − π + ] Λc K − π + ) is the measured invariant mass of the Ξ + cc candidate, m([pK − π + ] Λc ) is the measured invariant mass of the pK − π + combination forming the Λ + c candidate, and m(K − ) and m(π + ) are the world-average masses of charged kaons and pions, respectively [5].
Since no assumption is made about the Ξ + cc mass, a wide signal window of 380 < δm < 880 MeV/c 2 is used for this search, corresponding to approximately 3300 < m(Ξ + cc ) < 3800 MeV/c 2 . All aspects of the analysis procedure were fixed before the data in this signal region were examined. Limits on R are quoted as a function of the Ξ + cc mass and lifetime, since the measured yield depends on δm, and ε sig depends on both the mass and lifetime.

Detector and software
The LHCb detector [26] is a single-arm forward spectrometer covering the pseudorapidity range 2 < η < 5, designed for the study of particles containing b or c quarks. The detector includes a high-precision tracking system consisting of a silicon-strip vertex detector (VELO) surrounding the pp interaction region, a large-area silicon-strip detector located upstream of a dipole magnet with a bending power of about 4 Tm, and three stations of silicon-strip detectors and straw drift tubes placed downstream. The combined tracking system provides a momentum measurement with relative uncertainty that varies from 0.4% at 5 GeV/c to 0.6% at 100 GeV/c, and impact parameter (IP) resolution of 20 µm for tracks with large transverse momentum. Charged hadrons are identified using two ring-imaging Cherenkov detectors [27]. Photon, electron, and hadron candidates are identified by a calorimeter system consisting of scintillating-pad and preshower detectors, an electromagnetic calorimeter, and a hadronic calorimeter. Muons are identified by a system composed of alternating layers of iron and multiwire proportional chambers [28].
The trigger [29] consists of a hardware stage, based on information from the calorimeter and muon systems, followed by a software stage, which applies a full event reconstruction.
In the simulation, pp collisions are generated using Pythia 6.4 [30] with a specific LHCb configuration [31]. A dedicated generator, Genxicc v2.0, is used to simulate Ξ + cc baryon production [32]. Decays of hadronic particles are described by EvtGen [33], in which final state radiation is generated using Photos [34]. The interaction of the generated particles with the detector and its response are implemented using the Geant4 toolkit [35] as described in Ref. [36]. Unless otherwise stated, simulated events are generated with m(Ξ + cc ) = 3500 MeV/c 2 , with τ Ξ + cc = 333 fs, and with the Ξ + cc and Λ + c decay products distributed according to phase space.

Triggering, reconstruction, and selection
The procedure to trigger, reconstruct, and select candidates for the signal and normalisation modes is designed to retain signal and to suppress three primary sources of background. These are combinations of unrelated tracks, especially those originating from the same primary interaction vertex (PV); mis-reconstructed charm or beauty hadron decays, which typically occur at a displaced vertex; and combinations of a real Λ + c with other tracks to form a fake Ξ + cc candidate. The first two classes generally have a smooth distribution in both m([pK − π + ] Λc ) and δm; the third peaks in m([pK − π + ] Λc ) but is smooth in δm.
For both the Ξ + cc search and the normalisation mode, Λ + c candidates are reconstructed in the final state pK − π + . To minimise systematic differences in efficiency between the signal and normalisation modes, the same trigger requirements are used for both modes, and those requirements ensure that the event was triggered by the Λ + c candidate and its daughter tracks. First, at least one of the three Λ + c daughter tracks must correspond to a calorimeter cluster with a measured transverse energy E T > 3500 MeV in the hardware trigger. Second, at least one of the three Λ + c daughter tracks must be selected by the inclusive software trigger, which requires that the track have p T > 1700 MeV/c and χ 2 IP > 16 with respect to any PV, where χ 2 IP is defined as the difference in χ 2 of a given PV reconstructed with and without the considered track. Third, the Λ + c candidate must be reconstructed and accepted by a dedicated Λ + c → pK − π + selection algorithm in the software trigger. This algorithm makes several geometric and kinematic requirements, the most important of which are as follows. The three daughter tracks are required to have p T > 500 MeV/c 2 , to have a track fit χ 2 /ndf < 3, not to originate at a PV (χ 2 IP > 16), and to meet at a common vertex (χ 2 /ndf < 15, where ndf is the number of degrees of freedom). The Λ + c candidate formed from the three tracks is required to have p T > 2500 MeV/c 2 , to lie within the mass window 2150 < m([pK − π + ] Λc ) < 2430 MeV/c 2 , to be significantly displaced from the PV (vertex separation χ 2 > 16), and to point back towards the PV (momentum and displacement vectors within 1 • ). The software trigger also requires that the proton candidate be inconsistent with the pion and kaon mass hypotheses. The Λ + c trigger algorithm was only enabled for part of the data-taking in 2011, corresponding to an integrated luminosity of 0.65 fb −1 .
For events that pass the trigger, the Λ + c selection proceeds in a similar fashion to that used in the software trigger: three charged tracks are required to form a common vertex that is significantly displaced from the event PV and has invariant mass in the range 2185 < m([pK − π + ] Λc ) < 2385 MeV/c 2 . Particle identification (PID) requirements are imposed on all three tracks to suppress combinatorial background and mis-identified charm meson decays. The same Λ + c selection is used for the signal and normalisation modes.
The Ξ + cc candidates are formed by combining a Λ + c candidate with two tracks, one identified as a K − and one as a π + . These three particles are required to form a common vertex (χ 2 /ndf < 10) that is displaced from the PV (vertex separation χ 2 > 16). The kaon and pion daughter tracks are also required to not originate at the PV (χ 2 IP > 16) and to have p T > 250 MeV/c. The Ξ + cc candidate is required to point back to the PV and to have A multivariate selection is applied only to the signal mode to further improve the purity. The selector used is an artificial neural network (ANN) implemented in the TMVA package [37]. The input variables are chosen to have limited dependence on the Ξ + cc lifetime. To train the selector, simulated Ξ + cc decays are used as the signal sample and 3.5% of the candidates from δm sidebands of width 200 MeV/c 2 adjacent to the signal region are used as the background sample. In order to increase the available statistics, the trigger requirements are relaxed for these samples. In addition to the training samples, disjoint test samples of equal size are taken from the same sources. After training, the response distribution of the ANN is compared between the training and test samples. Good agreement is found for both signal and background, with Kolmogorov-Smirnov test p-values of 80% and 65%, respectively. A selection cut on the ANN response is applied to the data used in the Ξ + cc search. In the test samples, the efficiency of this requirement is 55.7% for signal and 4.2% for background.
The selection has limited efficiency for short-lived Ξ + cc . This is principally due to the requirements that the Ξ + cc decay vertex be significantly displaced from the PV, and that the Ξ + cc daughter kaon and pion have a significant impact parameter with respect to the PV. As a consequence, the analysis is insensitive to Ξ c resonances that decay strongly to the same final state, notably the Ξ c (2980) + , Ξ c (3055) + , and Ξ c (3080) + [20,38].

Yield measurements
To determine the Λ + c yield, N norm , a fit is performed to the pK − π + mass spectrum. The signal shape is described as the sum of two Gaussian functions with a common mean, and the background is parameterised as a first-order polynomial. The fit is shown in Fig. 1. The selected Λ + c yield in the full 0.65 fb −1 sample is N norm = (818 ± 7) × 10 3 , with an invariant mass resolution of around 6 MeV/c 2 .
The Ξ + cc signal yield is measured from the δm distribution under a series of different mass hypotheses. Although the methods used are designed not to require detailed knowledge of the signal shape, it is necessary to know the resolution with sufficient precision to define : Invariant mass spectrum of Λ + c → pK − π + candidates for 5% of the data, with events chosen at random during preselection (due to bandwidth limits for the normalisation mode). The dashed line shows the fitted background contribution, and the solid line the sum of Λ + c signal and background. a signal window. Since the Ξ + cc yield may be small, its resolution cannot be measured from data and is instead estimated with a sample of simulated events, shown in Fig. 2. Fitting the candidates with the sum of two Gaussian functions, the resolution is found to be approximately 4.4 MeV/c 2 .
Two complementary procedures are used to estimate the signal yield given a mass hypothesis δm 0 . Both follow the same general approach, but use different methods to estimate the background. In both cases, a narrow signal window is defined as 2273 < m([pK − π + ] Λc ) < 2303 MeV/c 2 and |δm − δm 0 | < 10 MeV/c 2 , and the number of candidates inside that window is taken as N S+B . Candidates outside the narrow window are used to estimate the expected background N B inside the window. The signal yield is then N S = N S+B − N B . This avoids any need to model the signal shape beyond an efficiency correction for the estimated signal fraction lost outside the window of width 20 MeV/c 2 .
The first method is an analytic, two-dimensional sideband subtraction in m([pK − π + ] Λc ) and δm. A two-dimensional region of width 80 MeV/c 2 in m([pK − π + ] Λc ) and width 200 MeV/c 2 in δm is centred around the narrow signal window. A 5 × 5 array of nonoverlapping bins is defined within this region, with the central bin identical to the narrow signal window. It is assumed that the background consists of a combinatorial component, which is described by a two-dimensional quadratic function, and a Λ + c component, which is described by the product of a signal peak in m([pK − π + ] Λc ) and a quadratic function in δm. Under this assumption, the background distribution can be fully determined from the 24 sideband bins and hence its integral within the signal box calculated. In this way the value of N B and the associated statistical uncertainty are determined. This method has the advantage that it requires only minor assumptions about the background distribution, given that part of that distribution cannot be studied prior to unblinding. It is adopted as the baseline approach for this reason.
The second method, used as a cross-check, imposes a narrow window on all candidates of 2273 < m([pK − π + ] Λc ) < 2303 MeV/c 2 to reduce the problem to a one-dimensional distribution in δm. Based on studies of the m([pK − π + ] Λc ) and δm sidebands, it is found that the background can be described by a function of the form where L(δm; µ, σ) is a Landau distribution, a is chosen such that L(µ; µ, σ L ) = aL(µ; µ, σ R ), and µ, σ L , and σ R are free parameters. The data are fitted with this function across the full range, 0 < δm < 1500 MeV/c 2 , excluding the signal window of width 20 MeV/c 2 . The fit function is then integrated across the signal window to give the expected background N B .

Efficiency ratio
To measure R, it is necessary to evaluate the ratio of efficiencies for the normalisation and signal modes, ε norm /ε sig . The method used to evaluate this ratio is described below. The signal efficiency depends upon the mass and lifetime of the Ξ + cc , neither of which is known. To handle this, simulated events are generated with m(Ξ + cc ) = 3500 MeV/c 2 and τ Ξ + cc = 333 fs and the efficiency ratio is evaluated at this working point. The variation of the efficiency ratio as a function of δm and τ Ξ + cc relative to the working point is then determined with a reweighting technique as discussed in Sec. 7. The kinematic distribution of Ξ + cc produced at the LHC is also unknown, but unlike the mass and lifetime it cannot be described in a model-independent way with a single additional parameter. Instead, the upper limits are evaluated assuming the distributions produced by the Genxicc model.
The efficiency ratio may be factorised into several components as where efficiencies are evaluated for the acceptance (acc), the reconstruction and selection excluding PID and the ANN (sel), the particle identification cuts (PID), the ANN selector (ANN) for the signal mode only, and the trigger (trig). Each element is the efficiency relative to all previous steps in the order given above. In most cases the individual ratios are evaluated with simulated Ξ + cc and Λ + c decays, taking the fraction of candidates that passed the requirement in question. However, in some cases the efficiencies need to be corrected for known differences between simulation and data. This applies to the efficiencies for tracking, for passing PID requirements, and for passing the calorimeter hardware trigger. Control samples of data are used to determine these corrections as a function of track kinematics and event charged track multiplicity, and the simulated events are weighted accordingly. The data samples used are J/ψ → µ + µ − for the tracking efficiency, and D * + → D 0 (→ K − π + )π + and Λ → pπ − for both the PID and calorimeter hardware trigger requirements. The track multiplicity distribution is taken from data for the Λ + c sample, but for Ξ + cc events it is not known. It is modelled by taking a sample of events containing a reconstructed B 0 s decay, on the grounds that B 0 s production also requires two non-light quark-antiquark pairs. The efficiency ratio obtained at this working point is ε norm /ε sig = 20.4. Together with the value for N norm obtained in Sec. 4 and the definition in eq. 2, this implies the single-event sensitivity α is 2.5 × 10 −5 at m(Ξ + cc ) = 3500 MeV/c 2 , τ Ξ + cc = 333 fs.

Systematic uncertainties
The statistical uncertainty on the measured signal yield is the dominant uncertainty in this analysis, and the systematic uncertainties on α have very limited effect on the expected upper limits. As in the previous section, they will be evaluated at the working point of m(Ξ + cc ) = 3500 MeV/c 2 and τ Ξ + cc = 333 fs, and their variation with mass and lifetime Beyond this, there are several instances where the simulation may not describe the signal accurately in data. These are corrected with control samples of data, with systematic uncertainties, outlined below, assigned to reflect uncertainties in these corrections.
The IP resolution of tracks in the VELO is found to be worse in data than in simulated events. To estimate the impact of this effect on the signal efficiency, a test is performed with simulated events in which the VELO resolution is artificially degraded to the same level. This is found to change the efficiency of the reconstruction and non-ANN selection by 6.6%, and that of the ANN by 6.7%. Taking these effects to be fully correlated, a systematic uncertainty of 13.3% is assigned.
A track-by-track correction is applied to the PID efficiency based on control samples of data. There are several systematic uncertainties associated with this correction. The first is due to the limited size of the control samples, notably for high-p T protons from the Λ sample. The second is due to the assumption that the corrections factorise between the tracks, whereas in practice there are kinematic correlations. The third is due to the dependence on the event track multiplicity. The fourth is due to limitations in the method (e.g. the finite kinematic binning used) and is assessed by applying it to samples of simulated events. The sum in quadrature of the above gives an uncertainty of 11.8%.
Systematic uncertainties also arise from the tracking efficiency (4.7%) and from the hardware trigger efficiency (3.3%). Additional systematic uncertainties associated with candidate multiplicity, yield measurement, and the decay model of Ξ + cc → Λ + c K − π + , which may proceed through intermediate resonances, were considered but found to be negligible in comparison with the total systematic uncertainty. The systematic uncertainties are summarised in Table 1. Taking their sum in quadrature, the total systematic uncertainty is 26%.

Variation of efficiency with mass and lifetime
The efficiency to trigger on, reconstruct, and select Ξ + cc candidates has a strong dependence upon the Ξ + cc lifetime. The efficiency also depends upon the Ξ + cc mass, since this affects the opening angles and the p T of the daughters.
The simulated Ξ + cc events are generated with a proper decay time distribution given by an exponential function of average lifetime τ Ξ + cc = 333 fs. To test other lifetime hypotheses, the simulated events are reweighted to follow a different exponential distribution and the efficiency is recomputed. Most systematic uncertainties are unaffected, but those associated with the limited simulated sample size and with the hardware trigger efficiency increase at shorter lifetimes (the latter due to kinematic correlations rather than direct dependence on the decay time distribution). The values and uncertainties of the single-event sensitivity α are given for several lifetime hypotheses in Table 2.
To assess the effect of varying the Ξ + cc mass hypothesis, large samples of simulated events are generated for two other mass hypotheses, m(Ξ + cc ) = 3300 MeV/c 2 and 3700 MeV/c 2 , without running the Geant4 detector simulation. Two tests are carried out with these samples. First, the detector acceptance efficiency is recalculated. Second, the p T distributions of the three daughters of the Ξ + cc in the main m(Ξ + cc ) = 3500 MeV/c 2 sample are reweighted to match those seen at the other mass hypotheses and the remainder of the efficiency is recalculated. In both cases the systematic uncertainties are also recalculated, though very little change is found. Significant variations in individual components of the efficiency are seen-notably in the acceptance, reconstruction, non-ANN selection, and hardware trigger efficiencies-but when combined cancel almost entirely. This is shown in Table 3, which gives the value of α including the mass-dependent effects discussed above but excluding the correction for the efficiency of the δm signal window described in Sec. 4 (α u ), the correction for the variation in resolution, and the combined value of α. Because the variation of α u with mass is extremely small, a simple first-order correction is sufficient. A straight line is fitted to the three points in the table and used to interpolate the fractional variation in α u between the mass hypotheses. The resolution correction is then applied separately. Due to the smallness of the mass-dependence, correlations Table 3: Variation in single-event sensitivity for different mass hypotheses m(Ξ + cc ), assuming τ = 333 fs. The uncertainties quoted include statistical and systematic effects, and are correlated between different mass hypotheses. The variation is shown separately for all effects other than the efficiency of the δm window (α u ), for the correction due to the mass-dependent resolution, and for the combination (α).   As explained in Sec. 1, the value of R at LHCb is not well known but is expected to be of the order 10 −5 to 10 −4 , while the SELEX observation corresponds to R = 9%. Table 4 shows the expected signal yield, calculated according to eq. 1, for various values of R and lifetime hypotheses. From studies of the sidebands in m([pK − π + ] Λc ) and δm, the expected background in the narrow signal window is between 10 and 20 events. Thus, no significant signal excess is expected if the value of R at LHCb is in the range suggested by theory. However, if production is greatly enhanced for baryon-baryon collisions at high rapidity, as reported at SELEX, a large signal may be visible. The procedure for determining the significance of a signal, or for establishing limits on R, is discussed in the following section.
1 MeV/c 2 steps for a total of 501 tests. For a given value of δm, the signal and background yields and their associated statistical uncertainties are estimated as described in Sec. 4. From these the local significance S(δm) is calculated, where S(δm) is defined as and σ S+B and σ B are the estimated statistical uncertainties on the yield in the signal window and on the expected background, respectively. Since multiple points are sampled, the look elsewhere effect (LEE) [39] must be taken into account. The procedure used is to generate a large number of pseudo-experiments containing only background events, with the amount and distribution of background chosen to match the data (as estimated from sidebands). For each pseudo-experiment, the full analysis procedure is applied in the same way as for data, and the local significance is measured at all 501 values of δm. The LEE-corrected p-value for a given S is then taken to be the fraction of the pseudo-experiments that contain an equal or larger local significance at any point in the δm range. The procedure established before unblinding is that if no signal with an LEE-corrected significance of at least 3σ is seen, upper limits on R will be quoted. The CL s method [40,41] is applied to determine upper limits on R for a particular δm and lifetime hypothesis, given the observed yield N S+B and expected background N B in the signal window obtained as described in Sec. 4. The statistical uncertainty on N B and systematic uncertainties on α are taken into account. The 95% CL upper limit is then taken as the value of R for which CL s = 0.05. Upper limits are calculated at each of the 501 δm hypotheses, and for five lifetime hypotheses (100, 150, 250, 333, 400 fs).

Results
The δm spectrum in data is shown in Fig. 3, and the estimated signal yield in Fig. 4. No clear signal is found with either background subtraction method. In both cases the largest local significance occurs at δm = 513 MeV/c 2 , with S = 1.5σ in the baseline method and S = 2.2σ in the cross-check. Applying the LEE correction described in Sec. 8, these correspond to p-values of 99% and 53%, respectively. Thus, with no significant excess found above background, upper limits are set on R at the 95% CL, shown in Fig. 5 for the first method. These limits are tabulated in Table 5 for blocks of δm and the five lifetime hypotheses. The blocks are 50 MeV/c 2 wide, and for each block the largest (worst) upper limit seen for a δm point in that block is given. Similarly, the largest upper limit seen in the entire 500 MeV/c 2 mass range is also given. A strong dependence in sensitivity on the lifetime hypothesis is seen.
The decay Ξ + cc → Λ + c K − π + may proceed through an intermediate Σ ++ c resonance. Such decays would be included in the yields and limits already shown. Nonetheless, further checks are made with an explicit requirement that the Λ + c π + invariant mass be consistent with that of a Σ ++ c , since this substantially reduces the combinatorial background. For

Conclusions
A search for the decay Ξ + cc → Λ + c K − π + is performed at LHCb with a data sample of pp collisions, corresponding to an integrated luminosity of 0.65 fb −1 , recorded at a centre-ofmass energy of 7 TeV. No significant signal is found. Upper limits on the Ξ + cc cross-section times branching fraction relative to the Λ + c cross-section are obtained for a range of mass and lifetime hypotheses, assuming that the kinematic distributions of the Ξ + cc follow those of the Genxicc model. The upper limit depends strongly on the lifetime, varying from 1.5 × 10 −2 for 100 fs to 3.  of R found at SELEX. This may be explained by the different production environment, or if the Ξ + cc lifetime is indeed very short ( 100 fs). Future searches at LHCb with improved trigger conditions, additional Ξ cc decay modes, and larger data samples should improve the sensitivity significantly, especially at short lifetimes.