betaFeedback

WorksPhysics and Astronomy

Sensitivity of the SHiP experiment at CERN to sterile-neutrino-induced oscillations

Abstract

The SHiP experiment, approved for the high-intensity beam-dump facility in the ECN3 hall at the CERN SPS in the North Area, will record neutrino interactions both in its scattering and neutrino detector (SND) and, about \SI{70}{m} further downstream, in its calorimeter system. We show that the baseline SHiP detector configuration therefore forms a two-baseline experiment for oscillations induced by a sterile neutrino. From a Geant4 simulation of Genie-generated interactions we derive the energy response of both stations to \nue and \numu charged-current (CC) interactions and the separation of \nue~CC events from the neutral-current \pizero background, and fold them into a near/far disappearance analysis with a 20%20\% flux-shape uncertainty and a 1%1\% far/near normalisation uncertainty. For 6×10206\times10^{20} protons on target, the projected 95%95\%~CL median sensitivities are \Umufour\simeq\UmuLimit at \mfour\simeq\UmuLimitMass~eV and \Uefour\simeq\UeLimit at \mfour\simeq\UeLimitMass~eV. In the muon channel this lies below existing \numu-disappearance limits for \DnineLow\lesssim\mfour\lesssim\DnineHigh~eV.

Introduction

SHiP is a general-purpose beam-dump experiment at the CERN SPS, designed to search for feebly interacting particles and to study the interactions of the intense flux of neutrinos of all flavours produced by 400 GeV protons in a dense tungsten target [1, 2]. Downstream of the target and the muon shield, a scattering and neutrino detector (SND) instrumented for vertexing, tracking and energy measurement records neutrino interactions for the neutrino-physics programme. About 70 m further downstream, after the decay volume and the spectrometer of the hidden-sector search, a calorimeter system provides particle identification for that search, while also recording neutrino interactions (Figure 1).

Figure 1. Layout of the SHiP experiment, with the neutrino interactions in the scattering and neutrino detector (SND) and in the calorimeter system, the near and far stations of this study.

Light sterile neutrinos have been invoked to explain several anomalous results: the excesses of electron-like events in LSND and MiniBooNE [3, 4], the deficit of reactor antineutrinos [5], and the gallium anomaly [6]. These anomalies have been the subject of extensive investigation [7, 8]. Searches for a fourth, mostly sterile mass eigenstate also probe the thermal history of the Universe before Big Bang nucleosynthesis. In particular, decays of light particles into charged leptons and neutrinos can lead to a late reheating, as late as a few hundred keV, while remaining consistent with the cosmic microwave background and the light-element abundances and with a limited impact on structure formation [9]. This impact is now being tested by large-scale-structure and CMB surveys: within Λ\LambdaCDM and in combination with CMB data from Planck [10] and ACT [11], DESI constrains the summed neutrino mass to ∑mν<0.064\sum m_\nu< 0.064 eV and the effective number of relativistic species to Neff=3.23−0.34+0.35N_{\rm eff}=3.23^{+0.35}_{-0.34} at 95% CL [12, 13], constraints that depend on the assumed cosmological model, while Euclid [14, 15], the Rubin Observatory LSST [16, 17] and the Simons Observatory [18, 19] will measure the growth of structure at the percent level and further constrain additional light relics. In the standard cosmological model, a sterile mass eigenstate with Δm412≳10\Delta m^2_{41} \gtrsim10 eV2^2 and mixing at the 10−310^{-3}–10−210^{-2} level would be in strong tension with observations [10]. In this mass range the mixing with νμ\nu_\mu is constrained by short-baseline accelerator experiments [20, 21, 22, 23, 24, 25] and by atmospheric neutrinos [26, 27], and the mixing with νe\nu_e by tritium β\beta decay [28, 29, 30]. It should be noted that the oscillation signature studied here is distinct from the searches for the decay of heavy neutral leptons in the SHiP decay volume, which address masses in the GeV range.

A sterile state with Δm412∼103\Delta m^2_{41} \sim10^3 eV2^2 oscillates over a few tens of metres at the energies of the SHiP neutrino beam (see Sec. II), the scale of the distances between the SHiP target, the SND and the calorimeter system. Comparing the two stations therefore measures the oscillation at two baselines. Here we show that this measurement can be performed with competitive sensitivity in the baseline SHiP detector configuration, with the SND as the near station and the calorimeter system as the far station, and we derive the response of both stations from simulation.

Two earlier studies addressed sterile-neutrino oscillations at SHiP. Reference [31] studied ντ\nu_\tau disappearance, which constrains ∣Uτ4∣2|U_{\tau4}|^2, with the emulsion-based neutrino detector of an earlier SHiP design, in which the τ\tau lepton is identified from its decay topology. In the simulated silicon–tungsten target most τ\tau leptons decay before reaching the next silicon plane, so that their decay topology cannot be reconstructed event by event, and the calorimeter system was not designed for this purpose. The ντ\nu_\tau channel is therefore not studied here and is left to future work. Reference [32] extended the study to all three flavours, in appearance and disappearance, with a near detector at 27 m and an additional far detector at 120 m of the same acceptance, generic detector responses and normalisation uncertainties. The present work uses the baseline SHiP detector configuration, detector responses derived from simulation and a flux-shape uncertainty in addition to the normalisation uncertainties; the sensitivities of the two studies are compared in Sec. VII.

Sterile neutrinos in the 3+1 model

In the 3+1 model the three active flavour states mix with a fourth, sterile state through a unitary 4×44 \times4 matrix UU, να=∑i=14Uαiνi\nu_\alpha= \sum_{i=1}^{4} U_{\alpha i}\nu_i with α=e,μ,τ,s\alpha= e, \mu, \tau, s. The fourth mass eigenstate ν4\nu_4 is mostly sterile, ∣Us4∣2≃1\lvert U_{s4}\rvert^2 \simeq1, and admixed to the active flavours through the small elements ∣Uα4∣2\lvert U_{\alpha4}\rvert^2. For Δm412≫∣Δm312∣\Delta m_{41}^2 \gg\lvert\Delta m_{31}^2\rvert the standard three-flavour oscillations have no time to develop over the baselines of SHiP (∣Δm312∣L/E∼10−6\lvert\Delta m_{31}^2\rvert L/E \sim10^{-6} for L∼100L \sim100 m and E∼40E \sim40 GeV), and the oscillation probabilities in vacuum depend only on the fourth column of UU [8, 33],

P(να→να)=1−sin⁡22θααsin⁡2Δ41,(1)P(\nu_\alpha\to\nu_\alpha) = 1 - \sin^2 2\theta_{\alpha\alpha}\sin^2\Delta_{41}, \tag*{(1)}
P(να→νβ)=4∣Uα4∣2∣Uβ4∣2sin⁡2Δ41,(2)P(\nu_\alpha\to\nu_\beta) = 4\lvert U_{\alpha4}\rvert^2\lvert U_{\beta4}\rvert^2\sin^2\Delta_{41}, \tag*{(2)}

for β≠α\beta\ne\alpha, with Δ41=1.267Δm412L/E\Delta_{41} = 1.267\Delta m_{41}^2L/E for the baseline LL in km, EE in GeV and Δm412\Delta m_{41}^2 in eV2^2, and the disappearance amplitude

sin⁡22θαα=4∣Uα4∣2(1−∣Uα4∣2)=4∣Uα4∣2∑β≠α∣Uβ4∣2,(3)\sin^2 2\theta_{\alpha\alpha} = 4\lvert U_{\alpha4}\rvert^2(1-\lvert U_{\alpha4}\rvert^2) = 4\lvert U_{\alpha4}\rvert^2\sum_{\beta\ne\alpha}\lvert U_{\beta4}\rvert^2, \tag*{(3)}

where the second equality follows from the unitarity of UU. Matter effects are negligible: the potential of the traversed material is many orders of magnitude below Δm412/2E\Delta m_{41}^2/2E. As in the external limits we compare with, each channel is analysed with a single non-zero element, ∣Uμ4∣2\lvert U_{\mu4}\rvert^2 for νμ\nu_\mu and ∣Ue4∣2\lvert U_{e4}\rvert^2 for νe\nu_e. With both non-zero, νμ→νe\nu_\mu\to\nu_e appearance, Eq. (2), would refill about 1% of the νe\nu_e deficit [32], which is neglected; the CPT-conserving model gives the same probabilities for neutrinos and antineutrinos, which are therefore not separated.

The sensitivity of an experiment is set by its ratio L/EL/E. The first oscillation maximum lies at Δm412≃1.24 eV2/(L/E [km/GeV])\Delta m_{41}^2 \simeq1.24\ \mathrm{eV}^2/(L/E\ [\mathrm{km}/\mathrm{GeV}]): for a mean energy of about 40 GeV, at about 5×1025 \times10^2 eV2^2 for the calorimeter system and 2×1032 \times10^3 eV2^2 for the SND. Far below the maximum the oscillation has no effect; around it, the deficit depends on energy and differs between the two stations, which is where a two-baseline comparison is most sensitive. Far above it, the oscillation is averaged over the energy resolution and the interaction depth, leaving a constant deficit 12sin⁡22θαα\frac{1}{2}\sin^2 2\theta_{\alpha\alpha} that is degenerate with the flux normalisation. Experiments at L/E∼0.1L/E \sim0.1–11 km/GeV, such as the Short-Baseline Neutrino programme at Fermilab [25, 34], are in this averaged regime at both of their detectors above Δm412∼102\Delta m_{41}^2 \sim10^2 eV2^2, where the SHiP baselines still resolve the oscillation in energy.

Since the dominant active neutrino eigenstates are much lighter than ν4\nu_4, its mass is m4≃Δm412m_4 \simeq\sqrt{\Delta m_{41}^2}, and the same exclusion can be expressed in the mass–mixing plane (m4,∣Uα4∣2)(m_4,\lvert U_{\alpha4}\rvert^2), the plane used for heavy neutral leptons. At the masses considered here the fourth state is far too light and too long-lived for decay in the apparatus, and the constraint comes only from the disappearance of active flavours.

Simulation, detectors and event samples

Neutrino interactions of all flavours are generated with GENIE 3.6.2, tune G21_11a_00_000 [35], for 4×10194 \times10^{19} protons on target (PoT) at the nominal annual SHiP exposure, and scaled to the 6×10206 \times10^{20} PoT of the full 15-year programme. The final states are transported through a standalone Geant4 simulation [36], injected at the true vertex.

Scattering and neutrino detector

The simulated SND (Figure 2) is a version of the design in which the detector sits inside the last magnet of the muon shield, whose field of 1.9 T magnetises its iron. Downstream of a plastic-scintillator veto (26.96 <z<< z < 27.00 m), it consists of a 40 ×\times 40 cm2^2 tungsten-target (27.02 <z<< z < 28.37 m) of 120 tungsten plates of 3.5 mm, each followed by an xx and a yy plane of silicon strips of 75 μ\mum pitch, and of three magnetised iron sections of 40, 50 and 60 cm transverse size (28.38–29.23, 29.24–30.08 and 30.09–30.93 m). Each section holds 14 iron plates of 5.0 cm, interleaved with sensitive planes that combine, for tracking, two 0.5 mm layers of scintillating fibres of 250 µm diameter and, for calorimetry, a 10 mm layer of 5×55 \times5 cm2^{2} scintillator tiles. The fine segmentation of the target separates νe\nu_e charged-current (CC) interactions from neutral-current (NC) interactions producing a π0\pi^0 (NC-π0\pi^0 events, Sec. IV B), and the magnetised iron measures the muon momentum (§ 0.5 B). The detector layout is still evolving, which may lead to some variation in the exact sensitivity achieved; however, the underlying two-baseline principle and the qualitative conclusions of this study are not expected to be significantly affected.

Simulated geometry of the scattering and neutrino detector

Figure 2. Simulated geometry of the scattering and neutrino detector: the silicon–tungsten target [green] and the magnetised iron [orange] sections. The beam enters from the left.

Calorimeter system

The calorimeter system [37, 38] (Figure 3) consists of a lead electromagnetic section (ECAL) of 20 radiation lengths (X0X_0), sampled every 0.5X00.5 X_0 by scintillator bars in alternating projections, with an air gap after the first 5X05 X_0, followed by an iron hadronic section (HCAL) of five interaction lengths. It spans z=97.02 mz = 97.02\,\mathrm{m}–99.62 m99.62\,\mathrm{m} with a transverse face of 4.32×6.48 m24.32 \times6.48\,\mathrm{m}^{2}.

Geometry of the calorimeter system

Figure 3. Geometry of the calorimeter system: the lead ECAL with scintillator bar layers of alternating orientation, followed by the iron HCAL. The beam enters from the left.

Event samples

The two stations lie at different distances from the target, so a common transverse size would accept a different angular part of the neutrino beam. The samples for both detectors are therefore selected inside a common cone from the target, of half-angle θmax⁡=6.89 mrad\theta_{\max} = 6.89\,\mathrm{mrad}, which keeps the near sample inside the 40×40 cm240 \times40\,\mathrm{cm}^{2} transverse size of the silicon-tungsten target: at the downstream end of the near sample the cone has a half-width of 19.2 cm. The production region is small compared with the distances involved, and the simulated spectra in this matched acceptance agree to within about 2% in mean energy for all flavours (Appendix A).

At the near station both samples are taken from interactions with a vertex at 26.90<z<27.82 m26.90 < z < 27.82\,\mathrm{m}, 19.3% of the SND interactions in the cone: the upstream 71 of the 120 tungsten plates, which hold 59% of the target inter-

actions, and the veto, with 0.6% of the sample. For νe\nu_e the granularity of the target is needed for the separation from NC-π0\pi^0 events, and for νμ\nu_\mu the muons produced there are measured over the full length of the magnetised iron.

At the far station the samples are taken from interactions in the ECAL behind the air gap, to suppress the background from photons entering from upstream. Interactions in the HCAL are not used, since too little instrumented material remains behind such a vertex. The interaction-weighted baselines are Lnear=27.40 mL_{\mathrm{near}} = 27.40\,\mathrm{m} and Lfar=97.09 mL_{\mathrm{far}} = 97.09\,\mathrm{m}, and the far-to-near ratio of CC interactions are RF/N=0.10R_{F/N} = 0.10 (νμ\nu_\mu) and 0.100.10 (νe\nu_e). The struck nuclei are 97.5% tungsten at the near station (the rest: 1.9% silicon and 0.6% carbon), and 71% lead, 20% carbon and 8% aluminium (from the high precision layers) at the far station (the rest: 1.4% hydrogen); the uncertainty this introduces in the far/near normalisation is discussed in § 1 C.

The spectra of interacting neutrinos are shown in Figure 4 and the yields entering the fit in Table I. The νμ\nu_\mu flux originates from pion, kaon and charm decays and the νe\nu_e flux from kaon and charm decays; the ντ\nu_\tau flux comes only from Ds→τντD_s \to\tau\nu_\tau and the subsequent τ\tau decay, and its CC rate is further reduced by the τ\tau production threshold.

Figure 4. Energy spectra of CC-interacting neutrinos, by flavour, in the SND/near station (left) and in the calorimeter system/far station (right), in the matched acceptance, for 6×10206 \times10^{20} PoT.

Electron-neutrino reconstruction

Energy

At the far station the reconstructed energy is a linear calibration of the energy deposited in the ECAL, which contains the electron shower and the part of the hadronic system contained there; it is a proxy for the neutrino energy, not a measurement of the electron energy. Measured in bins of true energy, its resolution is

σEE≃1.06E/GeV⊕0.39,(4)\frac{\sigma_E}{E} \simeq\frac{1.06}{\sqrt{E/\mathrm{GeV}}} \oplus0.39, \tag*{(4)}

i.e. 0.42 at 50 GeV (Figure 5). The large constant term reflects containment: the fluctuation of the escaping fraction does not decrease with energy. The response also carries a median bias, from +0.69 at 2 GeV to 5 GeV to −0.53 above 150 GeV: when σE/E∼0.5\sigma_E/E \sim0.5 a given deposit arises from a wide range of true energies, and an estimator unbiased in bins of reconstructed energy is biased in bins of true energy. The fit needs no unbiased estimator, only the correct response, which the migration matrix built from the 27 521 simulated far νe\nu_e CC interactions provides; building it from the selected events only changes the limit by less than 1%.

Figure 5. Electron-neutrino energy response at the far station. Left: 68% half-width of (Ereco−Eν)/Eν(E_{\mathrm{reco}} - E_{\nu})/E_{\nu} against true energy, with the fit of Eq. (4), and the median bias. Right: the true-to-reconstructed matrix with the analysis bin edges.

At the near station the energy observable is the sum of all deposits in the silicon–tungsten target and in the fibre layers of the iron sections, which collect the leakage of the shower; adding the scintillator layers degrades the resolution, and they are not used. The observable is calibrated to the neutrino energy on the 3 974 simulated νe\nu_e CC interactions of the near sample. It is then unbiased, with σ68/E\sigma_{68}/E between 0.32 and 0.47 from 13 to 134 GeV, and the near migration matrix is built from this response.

eCC versus NC-0 separation

The dominant background to νe\nu_e CC interactions is NC interactions producing a π0\pi^0, whose photons shower like the electron of a νe\nu_e CC event. The electron shower starts at the vertex and carries the ionisation of a single particle in its first layers, whereas the photons convert after a gap, deposit the ionisation of an e+e−e^+e^- pair, and can form two separate showers. A gradient-boosted decision tree (XGBoost [39]) separates the two, using the transverse occupancy image of the shower at the far station and thirteen shower-shape variables at the near station, where an image-based classifier overtrains on the available simulated statistics (Appendix B). All efficiencies and backgrounds in the fit are evaluated on events not used to train the classifier that scores them.

At score ≥0.5\geq0.5 the classifier retains 83.2% (79.2%) of νe\nu_e CC events and rejects 79.9% (81.7%) of NC-π0\pi^0 events at the far (near) station, averaged over the simulated samples. Weighted with the far νe\nu_e spectrum, which is harder than that of the simulated sample, the far efficiency is 90.2%, as in Table I. Varying the threshold between 0.3 and 0.7 at either station changes the νe\nu_e limit by less than 3%. The efficiency rises over the first few tens of GeV (Figure 6), most slowly in the thin near target, where a soft electron develops too little shower to be recognised. Since the oscillation phase scales as 1/E1/E, this low-energy efficiency limits the νe\nu_e reach at Δm412\Delta m^2_{41}.

Figure 6. νe\nu_e CC selection efficiency and NC-π0\pi^0 rejection (upper) and purity of the simulated sample (lower) against true neutrino energy at score ≥0.5\ge0.5, for both stations, on events not used in training.

Muon-neutrino energy reconstruction

The reconstructed energy of a νμ\nu_\mu CC interaction is Eνreco=Eμ+EhadE_\nu^{\mathrm{reco}} = E_\mu+ E_{\mathrm{had}}, the two stations measure the muon and hadronic contributions differently.

Far station

The calorimeter system has no magnetic field, and the muon momentum is estimated from its multiple scattering: the deviations of the track from a straight line, its kink angle and its depth in the HCAL are combined in a regression on the true momentum (Appendix B). Such an estimator is pulled towards the typical momentum of its calibration sample: its median bias ranges from +144% at 2 GeV to 5 GeV to −71% at 50 GeV to 100 GeV, although the core of the residual distribution is narrow for penetrating muons (14–26% between 20 and 100 GeV). The hadronic energy is measured calorimetrically, with a resolution of 36% and a bias in the energy response within ±10%. The resulting far neutrino-energy resolution is 76% (30% at 20 GeV to 50 GeV), and the muon accounts for 96% of its variance. The far migration matrix, built from 17 841 simulated events, carries these biases in full.

Near station

At the near station the muon momentum is measured from its deflection in the magnetised iron. No field map of the muon-shield magnet is available yet: the simulation is field-free, and the deflection is added analytically to the simulated hits, which already carry the multiple scattering. A field uncertainty acts as a momentum-scale uncertainty (§ 1 C). Over the 1.47 m lever arm of a muon produced in the target the displacement from its initial direction is 6.2 cm at 10 GeV and 1.2 cm at 50 GeV, and bent tracks that leave the active area are lost: the field removes 11.1% of the muons at 10 GeV, 3.2% at 50 GeV and 7.6% overall. The fit does not apply this loss separately: the muon acceptance and identification are modelled by a single 90% detection efficiency for every νμ\nu_\mu CC interaction in the target, which is meant to include it, with full identification of detected muons, wherever the muon stops or leaves the detector; the same efficiency is assumed at the far station. Applying the energy-dependent field loss explicitly instead changes the limit by less than 1%. The resolution is limited by multiple scattering in the iron, σp/p≃18\sigma_p/p \simeq18–20% between 10 and 100 GeV [40], and the near response is parametric: the true muon momentum is smeared with this resolution, and the true hadronic energy with an assumed calorimetric resolution of 0.50/E/GeV⊕0.050.50/\sqrt{E/\mathrm{GeV}} \oplus0.05 for the iron–scintillator sampling.

InteractionsSampleNear: In acceptanceNear: SelectedFar: In acceptanceFar: Selected
νe+νˉe\nu_e + \bar{\nu}_e CCνe\nu_e8.18×1058.18 \times10^56.42×1056.42 \times10^58.19×1048.19 \times10^47.39×1047.39 \times10^4
νμ+νˉμ\nu_\mu+ \bar{\nu}_\mu CCνμ\nu_\mu1.85×1061.85 \times10^61.67×1061.67 \times10^61.87×1051.87 \times10^51.68×1051.68 \times10^5
ντ+νˉτ\nu_\tau+ \bar{\nu}_\tau CC, τ→e νe\tau\to e \ \nu_e (bkg.)6.86×1036.86 \times10^35.24×1035.24 \times10^36.94×1026.94 \times10^26.24×1026.24 \times10^2
NC-π0\pi^0νe\nu_e (bkg.)5.59×1055.59 \times10^59.86×1049.86 \times10^45.29×1045.29 \times10^41.04×1041.04 \times10^4

Table 1. Expected yields at 6×10206 \times10^{20} PoT in the matched angular acceptance and the fitted energy range, before and after selection, as they enter the fit. Near: the silicon-tungsten target-vertex fiducial (26.90–27.82 m) for both samples; far: the calorimeter window. Selected νe\nu_e CC and νμ\nu_\mu CC are the signals of the two samples; ντ→e\nu_\tau\to e and NC-π0\pi^0 are the selected backgrounds of the νe\nu_e sample. 1.4–2.5% of the CC interactions lie outside the fitted energy range and are not included.

Oscillation analysis

Two-baseline disappearance

The oscillation probability, Eq. (1), is applied in true neutrino energy before any detector effect. For each station dd the predicted content of reconstructed-energy bin ii is

nid=Md∑jΦjdσjdϵjd⟨P(να→να)⟩j,dRijd(5)n_i^d = M^d \sum_j \Phi_j^d \sigma_j^d \epsilon_j^d \langle P(\nu_\alpha\to\nu_\alpha)\rangle_{j,d} R_{ij}^d \tag*{(5)}

with the fiducial target mass MdM^d, the neutrino fluence Φd\Phi^d through the station integrated over the exposure of 6×10206 \times10^{20} PoT, the CC cross section per unit mass of the station’s target σd\sigma^d, the selection efficiency ϵd\epsilon^d and the migration matrix RijdR_{ij}^d from true bin jj to reconstructed bin ii; the oscillation probability is averaged over each true-energy bin and over the interaction-depth profile of the station. The neutrino source is treated as a point at the start of the target; the production region, mostly within the target, is short compared with both baselines, and folding an exponential production-depth distribution of mean 0.15 m into the baselines changes the limit at its minimum by less than 1% in both channels. At the far station the reconstructed bins are merged until each reaches a purity and efficiency of 0.3, so that the binning follows the resolution; this leaves six bins for νμ\nu_\mu and four for νe\nu_e. Since the near station is itself 27.40 m from the target, oscillations are already appreciable there: it is a second oscillation measurement at a shorter baseline, not an unoscillated flux monitor.

The limits are quoted for one non-zero mixing element at a time (Sec. II), so that only disappearance contributes.

tributes.

Backgrounds

The selected NC-π0\pi^0 and ντ→e\nu_\tau\to e backgrounds are given in Table I. Each selected NC-π0\pi^0 event is reconstructed with the νe\nu_e energy estimator of its station applied to its deposit, so that its response matrix, built from the simulated NC-π0\pi^0 events, maps the energy of the parent neutrino directly to the reconstructed energy. Its component from νe\nu_e parents disappears with P(νe→νe)P(\nu_e \to\nu_e), since a neutrino that has oscillated into the sterile state does not interact; the components from νμ\nu_\mu and ντ\nu_\tau parents do not. The ντ\nu_\tau CC interactions with τ→e\tau\to e cannot be distinguished from the signal and are included as a non-oscillating component.

Systematic uncertainties

Uncertainties in the overall normalisation common to the two stations, from the number of protons on target, the neutrino production in the target and the cross-section scale, cancel in the near/far comparison; this common normalisation is left free in the fit. The remaining uncertainties and their effect on the limits are summarised in Table II.

Variationνμ\nu_{\mu}νe\nu_e
Flux correlation length ℓ=0.05\ell=0.05+9%+4%
Far/near normalisation, 2%×1.02\times1.02×1.55\times1.55
Far/near normalisation, 5%×1.02\times1.02×2.83\times2.83
Energy scales, 0.5% each×1.60\times1.60—
Energy scales, 1% each×1.80\times1.80—
Response shape, 1% per bin×2.4\times2.4×1.1\times1.1
Simulated statistics (r.m.s.)1.2%1.2%

Table II. Effect of variations of the analysis assumptions on the 95% CL median limit at its minimum, as a factor relative to the quoted results (∣Uμ4∣2=5.28×10−3|U_{\mu4}|^{2}=5.28\times10^{-3}, ∣Ue4∣2=1.03×10−2|U_{e4}|^{2}=1.03\times10^{-2}).

Flux shape. The flux uncertainty is a covariance in true neutrino energy,

Cjj′=σΦ2exp⁡[−ln⁡2(Ej/Ej′)2ℓ2](6)C_{jj'} = \sigma_{\Phi}^{2} \exp\left[-\frac{\ln^{2}(E_j/E_{j'})}{2\ell^{2}}\right] \tag*{(6)}

propagated through the response of both stations and diagonalised, each principal component becoming one nuisance parameter [41, 42]. The near spectrum constrains this common distortion, and only the part the two stations cannot resolve remains degenerate with an oscillation. We take σΦ=20%\sigma_{\Phi}=20\%, the level of the discrepancy between the measured muon flux from a replica SHiP target and the simulation [43], and ℓ=0.5\ell=0.5 in ln⁡E\ln E; both are assumptions, since that measurement does not determine the neutrino-energy covariance. Below Δm412≃103 eV2\Delta m_{41}^{2}\simeq10^{3}\ \mathrm{eV}^{2} the sensitivity changes by about 10% for correlation lengths down to 0.05; above it only the near station contributes and the reach depends on ℓ\ell. The distortion is assumed common to the two stations; a component that differs between them, for instance from the extended production region, is not evaluated here. The covariance acts on the summed νμ+νˉμ\nu_{\mu}+\bar{\nu}_{\mu} flux, so the shapes of the two are taken as fully correlated, as expected from their common hadroproduction origin; if they were independent, the νμ\nu_{\mu} limit would weaken by up to about 21%, a dependence that the muon-charge measurement at the near station removes, reproducing the quoted limit to within 3%.

Far/near normalisation. One nuisance parameter with a Gaussian prior σrel\sigma_{\mathrm{rel}} scales the far rate relative to the near rate. It covers the ratio of cross sections per unit mass in the two targets, the target masses and the relative selection efficiency of the two detector technologies. The target composition is part of the simulated yields; σrel\sigma_{\mathrm{rel}} covers the uncertainty on the ratio. Between tungsten and lead, charged-lepton data show nuclear modifications scaling as AαA^{\alpha} with ∣α∣<0.05|\alpha|<0.05 [44], i.e. differences below 0.6% at any Bjorken xx; about 30% of the far interactions occur on carbon and aluminium, for which model predictions of cross-section ratios to heavy nuclei can fail at the 10% level [45]. We adopt σrel=1%\sigma_{\mathrm{rel}}=1\% as a benchmark performance requirement for the relative near/far normalisation, not as a precision already demonstrated; it is tighter than what experiments with dissimilar near and far detectors have achieved so far [42]. With σrel=2%\sigma_{\mathrm{rel}}=2\% the limits become ∣Uμ4∣2=5.38×10−3|U_{\mu4}|^{2}=5.38\times10^{-3} and ∣Ue4∣2=1.60×10−2|U_{e4}|^{2}=1.60\times10^{-2} (Table II). The νe\nu_e channel, whose far spectrum has only four energy bins, draws its sensitivity mostly from the ratio of the event rates at the two stations, which a far/near normalisation change mimics directly, and so depends strongly on this assumption. The νμ\nu_{\mu} channel draws its sensitivity from the energy dependence of the deficit, which a normalisation change cannot reproduce, and is almost unaffected.

Backgrounds. The NC-π0\pi^{0} and ντ→e\nu_{\tau}\to e backgrounds have independent normalisations at each station, with Gaussian priors of 30% and 50%, the latter reflecting the uncertainty on DsD_s production.

Detector response. The limits assume that the energy

scales and responses of the two stations are known. Independent uncertainties on the muon and hadronic energy scales of each station, or an uncorrelated bin-to-bin uncertainty on the response shape, weaken the νμ\nu_{\mu} sensitivity (Table II), the largest contribution coming from the far-station muon energy. The finite simulated statistics contribute about 1%.

Statistical procedure

The near and far spectra are compared with the no-oscillation expectation in a Gaussian approximation to the Poisson likelihood, with the nuisance parameters profiled analytically. The 95% CL median expected limit and its bands follow from a Feldman–Cousins construction [51] for an amplitude estimator bounded at sin⁡22θ≥0\sin^{2}2\theta\geq0. Pseudo-experiments reproduce the analytic limit to within 2% (νμ\nu_{\mu}) and 4% (νe\nu_e), and injected signals are recovered in both channels (Appendix C).

Neither station is redundant: the calorimeter system alone reaches at best ∣Uμ4∣2≃1.56×10−1|U_{\mu4}|^{2}\simeq1.56\times10^{-1}, a factor of about 30 above the combined result (about 17 for νe\nu_e). With the flux shape assumed known, the near station alone would nearly reproduce the combined result; with the flux-shape uncertainty, a deficit at one station is equally well explained by a mismodelled spectrum, and only the comparison of the two baselines separates the two.

Sensitivity

The projected 95% CL median sensitivities are shown in Figure 7. The νμ\nu_{\mu} channel is strongest where the oscillation phases at the two stations differ most, reaching ∣Uμ4∣2≃5.28×10−3|U_{\mu4}|^{2}\simeq5.28\times10^{-3} at Δm412≃598 eV2\Delta m_{41}^{2}\simeq598\ \mathrm{eV}^{2} (m4≃24.4 eVm_4\simeq24.4\ \mathrm{eV}). Its median sensitivity lies below the existing νμ\nu_{\mu}-disappearance limits for 12.7≲m4≲31.6 eV12.7\lesssim m_4\lesssim31.6\ \mathrm{eV}, where both stations contribute, and remains below them up to m4≃81 eVm_4\simeq81\ \mathrm{eV} with the near station alone in every flux-correlation model considered; it is also below them in a narrow window at 4.7–8.7 eV. With an additional 1% response-shape uncertainty it lies below them over 13.7–79.7 eV, except in two narrow intervals near 17 and 29 eV, where CCFR is stronger by up to 9%. These statements compare a median expected sensitivity at 95% CL with observed limits, most of which are given at 90% CL;

Figure 7. Projected SHiP 95% CL median expected sensitivity, with the ±1σ\pm1\sigma and ±2σ\pm2\sigma bands, for $6 \times 10^{20}$ PoT. Left: sin⁡22θμμ\sin^2 2\theta_{\mu\mu}, with accelerator limits [20, 21, 22, 23, 24, 25] and atmospheric limits [26, 27], each drawn over the Δm412\Delta m^2_{41} range of its publication; the dashed line is the median sensitivity with an idealised far muon spectrometer (Sec. VII). Right: sin⁡22θee\sin^2 2\theta_{ee}, with reactor (PROSPECT [46], DANSS [47]), β\beta-decay (KATRIN [30, 48], Troitsk [29, 49], Mainz [28], {}^{187}\mathrm{Re} {citep}[50]) and atmospheric {citep}[7]` constraints. Below \Delta m^2_{41} \approx 10^3\ \mathrm{eV}^2$ both stations contribute and the curves are stable to ≈10%\approx10\% under changes of the flux-correlation model; above it only the near station contributes, the curves use the most conservative flux model (ℓ=0.05\ell= 0.05), and the reach depends on that model; the narrow features of the curves there, a few per cent wide in Δm412\Delta m^2_{41}, are numerically converged and arise where the near-station oscillation pattern can be absorbed by the flux-shape model at the near-station energy resolution, so they are not resolvable features of the sensitivity. External limits are shown at their published confidence level: 95% where labelled, 90% otherwise. The SHiP curves start at Δm412=1 eV2\Delta m^2_{41} = 1\ \mathrm{eV}^2.

the comparison is therefore conservative.

Far muon spectrometer

The least precise ingredient of the muon channel is the far muon momentum (§ 0.5 A). In an idealised study, a magnetised spectrometer behind the calorimeter system measuring the momentum of the muons that cross it to 1% improves the far νμ\nu_\mu energy resolution from 76% to 30% and the sensitivity to ∣Uμ4∣2≃3.02×10−3|U_{\mu4}|^2 \simeq3.02 \times10^{-3} (dashed line in Figure 7). It also reduces the cost of the energy-scale uncertainties from a factor 1.80 to 1.23 and lets the far station contribute above Δm412≈103 eV2\Delta m^2_{41} \approx10^3\ \mathrm{eV}^2. Beyond this point the sensitivity is limited by the flux-shape and far/near normalisation uncertainties rather than by resolution. With the muon charge measured at both stations, the spectrometer would also allow the disappearance amplitudes of neutrinos and antineutrinos to be fitted independently, without assuming CPT symmetry; the projected sensitivities to ∣Uμ4∣2|U_{\mu4}|^2 are then 3.08×10−33.08 \times10^{-3} from the νμ\nu_\mu amplitude and 4.07×10−34.07 \times10^{-3} from the νˉμ\bar{\nu}_\mu amplitude. Such a measurement could use a spectrometer of the NA62 type: the NA62 spectrometer [52] has been dismantled and is kept in the ECN3 hall.

The νe\nu_e channel reaches ∣Ue4∣2≃1.03×10−2|U_{e4}|^2 \simeq1.03 \times10^{-2} at Δm412≃716 eV2\Delta m^2_{41} \simeq716\ \mathrm{eV}^2 (m4≃26.8 eVm_4 \simeq26.8\ \mathrm{eV}). In this mass range the tritium β\beta-decay searches constrain the electron mixing more strongly wherever they apply: KATRIN [30, 48] below m4≈35 eVm_4 \approx35\ \mathrm{eV}, and Troitsk up to m4≈90 eVm_4 \approx90\ \mathrm{eV} (2014 [29]) and ≈400 eV\approx400\ \mathrm{eV} (2017 [49]); 187Re{}^{187}\mathrm{Re} decay [50] covers higher masses. The SHiP measurement is obtained by a different method, an oscillation over two known baselines rather than a distortion of a β\beta spectrum, with different systematic uncertainties.

Figure 8 shows the same sensitivities in the mass–mixing plane (Sec. II), complementary to the decay searches of the SHiP programme at the GeV scale.

Figure 8. The sensitivities of Figure 7 in the (m4,∣Uμ4∣2)(m_4, |U_{\mu4}|^2) (left) and (m4,∣Ue4∣2)(m_4, |U_{e4}|^2) (right) planes, with the ±1σ\pm1\sigma and ±2σ\pm2\sigma bands, existing limits and the seesaw line. The remarks of Figure 7 on Δm412≳103 eV2\Delta m^2_{41} \gtrsim10^3\ \mathrm{eV}^2 (m4≳30 eVm_4 \gtrsim30\ \mathrm{eV}) and on the confidence levels apply; in particular, the SHiP curves start at m4=1 eVm_4 = 1\ \mathrm{eV} (Δm412=1 eV2\Delta m^2_{41} = 1\ \mathrm{eV}^2).

Comparison with earlier studies. Reference [32] reaches ∣Uα4∣2≈0.1|U_{\alpha4}|^2 \approx0.1 near Δm412≈103 eV2\Delta m^2_{41} \approx10^3\ \mathrm{eV}^2 with a near detector alone and the 10−210^{-2} level with an additional far detector, at 90% CL for 2×10202 \times10^{20} PoT. The two studies differ in assumptions that act in opposite directions: that study assumes a far detector with the same acceptance as the near one, whereas the calorimeter system collects about ten times fewer interactions than the SND target (RF/NR_{F/N}, Sec. III C); it uses a three times smaller exposure; and it includes normalisation uncertainties only, whereas the present analysis also includes an energy-dependent flux-shape uncertainty, its leading systematic effect (without it the νμ\nu_\mu limit would be ∣Uμ4∣2≃1.35×10−3|U_{\mu4}|^2 \simeq1.35 \times10^{-3}), and quotes 95% CL. With these differences, the two studies reach mixings of the same order.

Conclusions and outlook

The SHiP SND and calorimeter system form a near/far pair at 27.40 and 97.09 m from the target. With responses derived from simulation, νe\nu_e CC events are reconstructed with a containment-limited resolution of 1.06/E/GeV⊕0.391.06/\sqrt{E/\mathrm{GeV}} \oplus0.39 in the calorimeter system and separated from NC–π0\pi^0 events with 83.2% efficiency at 79.9% rejection, and with 79.2% at 81.7% in the SND target; the muon momentum is measured from magnetic deflection in the SND with a resolution of about 20%, and estimated from multiple scattering in the calorimeter system. The projected 95% CL sensitivities are ∣Uμ4∣2≃5.28×10−3|U_{\mu4}|^2 \simeq5.28 \times10^{-3} and ∣Ue4∣2≃1.03×10−2|U_{e4}|^2 \simeq1.03 \times10^{-2} near m4≃24.4m_4 \simeq24.4 and 26.8 eV; in the muon channel this lies below existing limits for 12.7≲m4≲31.612.7 \lesssim m_4 \lesssim31.6 eV. The result assumes known energy scales and responses of the two stations and a 1% far/near normalisation; the relative calibration of the two stations' energy scales is the main experimental requirement for the muon channel.

The SHiP detector design can still be optimised for this measurement: its least precise ingredient is the far muon momentum, and a muon spectrometer behind the calorimeter system would strengthen the muon channel and reduce its dependence on the energy scales. Other detector options have been discussed for the facility, among them a liquid-argon time-projection chamber [53]. This work demonstrates that the baseline SHiP detector configuration can conduct competitive studies of sterile-neutrino-induced oscillations.

Acknowledgments

We thank Nick van Remortel and Alfons Weber for their comments. M. Climescu has received support from the FWO, under grant no. 12A4026N (Belgium). This work is supported by the Swiss National Science Foundation through an Ambizione grant No. PZ00-2_240997.

Event samples

Figure 9 shows the transverse distribution of the CC interaction vertices in the two fiducial volumes: 68, 90, 95 and 99% of the interactions lie within 17, 20, 22 and 25 cm of the axis at the near station and within 59, 72, 78 and 87 cm at the far station. Figure 10 compares the spectra at the two stations with a common transverse acceptance and with the matched cone.

Figure 9. Transverse distribution of CC interaction vertices in the near (silicon–tungsten target) and far (ECAL) fiducial volumes, with the matched cone and the 68, 90, 95 and 99% containment contours.

Figure 10. CC interaction spectra of νμ\nu_\mu, νe\nu_e and ντ\nu_\tau at the two stations, normalised to unit area, with a common transverse acceptance, for which the two stations sample different parts of the beam, and with the matched cone, for which the spectra coincide.

Reconstruction details

Neutrino-energy response. Figure 12 summarises the neutrino-energy response of the four samples as they enter the fit. The far responses of both channels are strongly biased at the ends of the spectrum, through containment for νe\nu_e and through the multiple-scattering estimate of the muon momentum for νμ\nu_\mu; the migration matrices carry these biases in full (Secs. IV A and V A).

Figure 12. Neutrino-energy response of the four samples as used in the fit: σ68/E\sigma_{68}/E of the reconstructed neutrino energy against the true EνE_\nu (a) and against 1/Eν/GeV1/\sqrt{E_\nu/\mathrm{GeV}} (b), and the median bias (c). Near νμ\nu_\mu: parametric response (Gluckstern muon ⊕\oplus 0.50/E⊕0.050.50/\sqrt{E} \oplus0.05 hadrons); far νμ\nu_\mu and νe\nu_e: simulated events, with statistical errors; near νe\nu_e: the simulated response of the estimator calibrated to the neutrino energy (markers, with bootstrap errors) and the migration matrix built from it (dashed), drawn above 10 GeV, where its 7.5 GeV reconstructed-energy bins resolve the response.

Muon momentum. At the far station the muon momentum is estimated by a linear least-squares fit of ln⁡pμ\ln p_\mu to seven standardised features of the muon track, whose points are digitised to the bar pitch of each layer: the kink angle between straight-line fits to the first and last thirds of the track, the r.m.s. residual of a straight-line fit to all points and to the HCAL points alone, the depth reached in the HCAL, and the numbers of ECAL, HCAL and total points. Multiple scattering in the lead [33, 54, 55] makes the kink angle and the residuals scale as 1/p1/p. Muons that do not reach the HCAL (1.6% of the sample) are assigned 0.5 GeV. The hadronic energy is a linear combination of the visible ECAL and HCAL energies. Both estimators are calibrated on a random half of the simulated events, and the far migration matrix is built from the remaining, independent events. The quoted bias and core are the median and the 68% half-width of Ereco/Etrue−1E_{\mathrm{reco}}/E_{\mathrm{true}} - 1 in bins of true energy. At the near station the resolution follows the Gluckstern form [40],

(σpp)2=(σxp0.3BL2720N+4)2+(0.054BLX0)2(7)\left(\frac{\sigma_p}{p}\right)^2 = \left(\frac{\sigma_x p}{0.3 B L^2}\sqrt{\frac{720}{N+4}}\right)^2 + \left(\frac{0.054}{B\sqrt{L X_0}}\right)^2 \tag*{(7)}

with σx\sigma_x, LL and X0X_0 in metres, BB in tesla and pp in GeV/cc; with B=1.9 TB = 1.9\,\mathrm{T}, σx=250 μm\sigma_x = 250\,\mu\mathrm{m} and L=1.47 mL = 1.47\,\mathrm{m} of iron the multiple-scattering term dominates at all momenta. The charge is determined from the sign of the sagitta, correctly above about 10 GeV.

Classifiers. Table III summarises the classifiers. The near-station variables are the gap between the vertex and the shower start, the number of track hits at the vertex, the energy fraction and the ionisation (in units of a minimum-ionising particle) at the shower start, the number of electromagnetic clusters and the invariant mass of the two leading ones, the transverse r.m.s. and the longitudinal length of the shower, its core energy fraction, the electromagnetic and hadronic energies, the electromagnetic fraction and the number of hits. All classifiers use trees of depth 4, a learning rate of 0.05, a minimum child weight of 10 and L2L_2 regularisation of 2, with the classes balanced by weight; the image classifiers in addition subsample events (0.8) and features (0.3). The efficiencies used in the fit are five-fold out-of-fold predictions. The near-station image classifier, on events not used in its training, reaches 73.5% efficiency at 79.9% rejection and would not improve the νe\nu_e limit. The classifier outputs are shown in Figure 11.

FarNearNear, alt.
Inputimagevariablesimage
Inputs45 0241312 003
νe\nu_e CC events27 57823 19919 880
NC-π0\pi^0 events17 80016 46214 178

Table III. Classifiers for the νe\nu_e CC versus NC-π0\pi^0 separation.

Figure 11. Classifier output for νe\nu_e CC and NC-π0\pi^0 events at the far (left) and near (right) station, on events not used in training, with the working point, and the efficiency and rejection against threshold.

Likelihood and validation

The near and far reconstructed spectra are stacked into one vector. For νe\nu_e the likelihood is evaluated on a common grid of 40 bins of 3.75 GeV per station, onto which each response matrix is mapped by interval overlap, so that the energy information is that of the matrices’ own bins: 7.5 GeV wide at the near station and four bins at the far station. The prediction is ni(s,θ)=ni0−shi+∑kθkAikn_i(s,\theta) = n_i^0 - s h_i + \sum_k \theta_k A_{ik}, with s=sin⁡22θααs = \sin^2 2\theta_{\alpha\alpha} the null spectrum, hh its response to ss at s=0s = 0, and AA the nuisance responses. In the Gaussian approximation on the Asimov data set, with W=diag⁡(1/ni0)W = \operatorname{diag}(1/n_i^0) and the nuisances profiled analytically, the information on ss is

I=hTWh−hTWA(ATWA+P−1)−1ATWh(8)I = h^{\mathsf{T}}Wh - h^{\mathsf{T}}WA\left(A^{\mathsf{T}}WA + P^{-1}\right)^{-1}A^{\mathsf{T}}Wh \tag*{(8)}

where P−1=diag⁡(1/σk2)P^{-1} = \operatorname{diag}(1/\sigma_k^2) is the prior precision (zero for a free parameter) and σ(s)=1/I\sigma(s) = 1/\sqrt{I}. The median expected limit is s95=1.96 σ(s)s_{95} = 1.96\,\sigma(s), from the Feldman–Cousins belt of a Gaussian estimator bounded at s≥0s \ge0, evaluated at each Δm412\Delta m_{41}^2 separately with Δm412\Delta m_{41}^2 fixed (a raster scan); the ±1σ\pm1\sigma and ±2σ\pm2\sigma bands use factors 1.10, 2.96 and 0.62, 3.96. The nuisance parameters are listed in Table IV. With 2×1042 \times10^4 pseudo-experiments per amplitude at five values of Δm412\Delta m_{41}^2 between 10210^2 and 104 eV210^4\ \mathrm{eV}^2, Poisson-fluctuated and with the constrained nuisances drawn from their priors, the factor 1.96 is reproduced to within 2% (νμ\nu_\mu) and 4% (νe\nu_e), never on the optimistic side.

NuisancePriorνμ\nu_\muνe\nu_e
Common normalisationfree11
Far/near normalisation1%11
Flux shape, Eq. (6)unit/mode1611
NC-π0\pi^0, per station30%–2
ντ→e\nu_\tau\to e, per station50%–2
Total1817

Table IV. Nuisance parameters of the fits.

A signal injected at Δm412=400 eV2\Delta m_{41}^2 = 400\ \mathrm{eV}^2 with sin⁡22θμμ=3.87×10−2\sin^2 2\theta_{\mu\mu} = 3.87 \times10^{-2} is recovered at 321–528 eV2\mathrm{eV}^2 with amplitude 3.87×10−2±1.18×10−23.87 \times10^{-2} \pm1.18 \times10^{-2}; in the νe\nu_e channel an injection of sin⁡22θee=1.25×10−1\sin^2 2\theta_{ee} = 1.25 \times10^{-1} at the same Δm412\Delta m_{41}^2 is recovered at 280–579 eV2\mathrm{eV}^2 with amplitude 1.25×10−1±3.31×10−21.25 \times10^{-1} \pm3.31 \times10^{-2}. In Asimov data the recovery is exact by construction; the intervals show the resolution in Δm412\Delta m_{41}^2 (Figure 13).

Figure 13. Validation of the muon (left) and electron (right) channels: (a) closure of the quoted limit against the recomputed belt, (b) recovery of an injected signal (Δχ2\Delta\chi^2 profile in Δm412\Delta m^2_{41}), (c) the two asymptotic regimes, and (d) the sensitivity with and without the near station.

References

The SHiP Collaboration

M. Akincy39^{39}, A. Aksoy13^{13}, R. Albanese20,d^{20,d}, K. Albrecht9^{9}, F. Alessio35^{35}, A. Alexandrov20,d^{20,d}, F. Aloschi20,d^{20,d}, A. Anokhina27^{27}, T. Arndt14^{14}, D. Arutinov13^{13}, A. Ballarino35^{35}, N. Bangaru20,d^{20,d}, L. Baudin35^{35}, V. Bautin33^{33}, A. Bay36^{36}, I. Bekman13^{13}, C. Betancourt24^{24}, I. Bezshyiko37^{37}, O. Bezshyyko44^{44}, D. Bick12^{12}, M. Birch42^{42}, A. Blanco Castro32^{32}, M. Bogomilov2^{2}, P. Borges De Sousa35^{35}, K. Bondarenko31^{31}, W.M. Bonivento19^{19}, M. Borisyak10^{10}, N. Bourcey35^{35}, T. Bowcock41^{41}, A. Boyarsky31,44^{31,44}, D. Breton7^{7}, A. Brignoli9^{9}, V. Bücher15^{15}, S. Buontempo20^{20}, C.S. Caillot35^{35}, M. Campanelli43^{43}, D. Centanni20^{20}, A. Cervelli18^{18}, S. Charity41^{41}, K.-Y. Choi30^{30}, D. Chokheli8^{8}, A. Chukanov33^{33}, M. Climescu(CA)1^{1}, M. Cristinziani16^{16}, G.M. Dallavalle18^{18}, M. Dam14^{14}, N. D’Ambrosio17^{17}, G. D’Appollonio19,b^{19,b}, D. Davino20,i^{20,i}, P. de Bryas36^{36}, J. De Carvalho Saraiva32^{32}, G. De Lellis20,d^{20,d}, M. de Magistris20,h^{20,h}, A. De Roeck42^{42}, M. De Serio17,b^{17,b}, G. Del Giudice20,d^{20,d}, P. Deucher15^{15}, A. Devred35^{35}, A. Di Crescenzo20,d^{20,d}, C. Di Cristo20,d^{20,d}, H. Dijkstra35^{35}, D. Dobur1^{1}, D. Duarte Ramos35^{35}, N.C. Duren39^{39}, T. Enik33^{33}, O. Fecarotta20,d^{20,d}, F. Fedotov43^{43}, T. Ferber14^{14}, M. Ferrillo37^{37}, M. Ferro-Luzzi35^{35}, A. Fiorillo20,d^{20,d}, H. Fischer11^{11}, R. Fresa20,g^{20,g}, G. Frisella35^{35}, S.J. Fuenzalida Garrido45^{45}, T. Fukuda24^{24}, G. Galati17,b^{17,b}, E. Gamberini35^{35}, B. Goddard35^{35}, L. Golinka-Bezshyyko37,44^{37,44}, A. Golutvin42^{42}, A. Gonnin7^{7}, D. Gorbunov27^{27}, V. Gorkavenko44^{44}, E. Graverini36,m^{36,m}, A. M. Guler38^{38}, V. Guliaeva36^{36}, C. Gwilliam41^{41}, G.J. Haefeli36^{36}, C. Hagner12^{12}, J.C. Helo Herrera3,5^{3,5}, M.J. Henriquez Zerené4^{4}, E. van Herwijnen42^{42}, A. Hollnagel15^{15}, C. Issever9^{9}, S. Izquierdo Bermudez35^{35}, A. Iuliano(CA)20,d^{20,d}, R. Jacobsson35^{35}, G. Japaridze8^{8}, D. Jokovic34^{34}, C. Kamiscioglu39^{39}, D. Khidasheli8^{8}, V. Kholoimov36^{36}, S.H. Kim28^{28}, Y.G. Kim29^{29}, V. Koch13^{13}, K. Kodama22^{22}, T. Koettig35^{35}, D.I. Kolev2^{2}, L. Kolupaeva33^{33}, M. Komatsu24^{24}, P.H. Kooij37^{37}, V. Kostyukhin16^{16}, I. Krasilnikova10^{10}, K. Kuznetsova27^{27}, S. Kuleshov3,4^{3,4}, H.M. Lacker9^{9}, M. Lamont35^{35}, O. Lantwin16^{16}, A. Lauria20,d^{20,d}, K.Y. Lee28^{28}, W.C. Lee12^{12}, N. Leonardo32,o^{32,o}, V.P. Loschiavo20,i^{20,i}, I. Lomidze8^{8}, F. Lyons11^{11}, J. Maalmi7^{7}, A.M. Magnan42^{42}, F.J. Mangiarotti35^{35}, J.P. Marquez Hernandez7^{7}, S.N. Medina Figueroa5^{5}, G.H. Mendizabal37^{37}, A. Mianon^{n}, S. Mikado25^{25}, A. Mikulenko31^{31}, A. Milanese35^{35}, J. A. Molins i Bertram15^{15}, T.A. Molzberger11^{11}, M.C. Montesi20,d^{20,d}, K. Morishima24^{24}, Y. Mukhamejanov27^{27}, A. Mukhamejanova27^{27}, N. Naganawa24^{24}, T. Nakano24^{24}, S.L. Ochoa Guaman11^{11}, S. Ogawa26^{26}, O. Olshevskiy33^{33}, M. Ovchynnikov35^{35}, P. Owen37^{37}, U. Parzefall11^{11}, A. Pastore17^{17}, M. Patel42^{42}, K. Petridis40^{40}, O. Pirotte35^{35}, N. Polukhina33^{33}, L.F. Prates Cattelan37^{37}, J.D. Price41^{41}, A. Prota5^{5}, S.R. Qasim37^{37}, A. Quercia36^{36}, A. Rademakers35^{35}, F. Ratnikov3^{3}, F. Redl1^{1}, A. Reghunath9^{9}, S. Ritter15^{15}, T.J. Rock11^{11}, E. Rodriquez41^{41}, S. Romakhov33^{33}, H. Rokujo24^{24}, O. Ruchayskiy6^{6}, T. Ruf35^{35}, O.M. Ruud35^{35}, A. Saba42^{42}, S. Sakhiyev27^{27}, K. Salamatin33^{33}, O. Samoylov33^{33}, P. Santos Díaz35^{35}, O. Sato24^{24}, F. Savary35^{35}, V. Scalera20,h^{20,h}, C. Scharf9^{9}, W. Schmidt-Parzefall12^{12}, E. Schopf6^{6}, M. Schumann11^{11}, P. Schupp10^{10}, N. Serra6^{6}, M. Shaposhnikov36^{36}, L. Shchuska36^{36}, H. Shibuyа26^{26}, A. Sidoti18^{18}, C. Silano20,i^{20,i}, P. Simone17,b^{17,b}, K. Skovpen1^{1}, M. Smith42^{42}, G. Soares32^{32}, J.Y. Sohn28^{28}, O. Soto Sandoval3,4^{3,4}, F. Steeg15^{15}, O. Steinkamp37^{37}, W. Sutcliffe37^{37}, S. Tapia45^{45}, H. Tilquin42^{42}, I. Timiryasov6^{6}, V. Tioukov20^{20}, C. Touramanis Douramanis41^{41}, D. Treille35^{35}, Z. Tsamalaidze6^{6}, N. Tsverava8^{8}, P. Ulloa Poblete3,5^{3,5}, E. Ursov9^{9}, A. Ustyuzhanin10^{10}, C. Valdivieso5^{5}, G. Vasquez35^{35}, N. Viaux45^{45}, D.L. A. Viera Lopes32^{32}, S. Vilchinskii44^{44}, C. Visone20,d^{20,d}, S. van Waasen13^{13}, R. Wankewicz15^{15}, J.M. Webb11^{11}, C. Weiser11^{11}, M. Wensing46^{46}, I.M. Wöstheinrich9^{9}, M. Wurm15^{15}, S. Yamamoto24^{24}, D. Yilmaz39^{39}, S.M. Yoo30^{30}, C.S. Yoon28^{28}, A. Zaitsev33^{33}, J. Zamora Saa3,4^{3,4}, T. Zholdybayev27^{27}

1^{1} Ghent University, Ghent, Belgium

2^{2} Faculty of Physics, Sofia University, Sofia, Bulgaria

3^{3} Millennium Institute For Subatomic Physics At High-Energy Frontier - SAPPHIR, Chile

4^{4} Universidad Andrés Bello (UNAB), Santiago, Chile

5^{5} Universidad De La Serena (ULS), La Serena, Chile

6^{6} Niels Bohr Institute, University of Copenhagen, Copenhagen, Denmark

7^{7} IJCLab, CNRS, Université Paris-Saclay, Orsay, France

8^{8} Georgian Technical University, Tbilisi, Georgia

9^{9} Institut für Physik, Humboldt-Universität zu Berlin, Berlin, Germany

10^{10} Constructor University, Bremen, Germany

11^{11} Physikalisches Institut, Universität Freiburg, Freiburg, Germany

12^{12} Universität Hamburg, Hamburg, Germany

13^{13} Integrated Computing Architectures (ICA — PGI-4), Forschungszentrum Jülich GmbH (KFA), Jülich, Germany

14^{14} Karlsruhe Institute of Technology, Karlsruhe, Germany

15^{15} Institut für Physik and PRISMA Cluster of Excellence, Johannes Gutenberg Universität Mainz, Mainz, Germany

16^{16} Universität Siegen, Siegen, Germany

17^{17} Sezione INFN di Bari, Bari, Italy

18^{18} Sezione INFN di Bologna, Bologna, Italy 19 Sezione INFN di Cagliari, Cagliari, Italy

20 Sezione INFN di Napoli, Napoli, Italy

21 Laboratori Nazionali dell’INFN di Gran Sasso, L’Aquila, Italyk

22 Aichi University of Education, Kariya, Japana

23 Kobe University, Kobe, Japana

24 Nagoya University, Nagoya, Japan

25 College of Industrial Technology, Nihon University, Narashino, Japana

26 Toho University, Funabashi, Chiba, Japana

27 Institute of Nuclear Physics, Almaty, Kazakhstan

28 Physics Education Department & RINS, Gyeongsang National University, Jinju, Korea

29 Gwangju National University of Education, Gwangju, Koreae

30 Sungkyunkwan University, Suwon-si, Gyeong Gi-do, Koreae

31 University of Leiden, Leiden, The Netherlands

32 Laboratory of Instrumentation and Experimental Particle Physics(LIP),Portugal

33 Joint Institute for Nuclear Research, Dubna, Russia

34 Institute of Physics, University of Belgrade, Serbia

35 European Organization for Nuclear Research (CERN), Geneva, Switzerland

36 Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland

37 Physik-Institut, Universität Zürich, Zürich, Switzerland

38 Middle East Technical University (METU), Ankara, Turkey

39 Ankara University, Ankara, Turkey

40 H.H. Wills Physics Laboratory, University of Bristol, Bristol, United Kingdom

41 University of Liverpool, Liverpool, United Kingdom

42 Imperial College London, London, United Kingdom

43 University College London, London, United Kingdom

44 Taras Shevchenko National University of Kyiv, Kyiv, Ukraine

45 Universidad Tecnica Federico Santa Maria (UTFSM), Valparaiso, Chilej

46 Bergische Universitaet Wuppertal, Wuppertal, Germany

a Associated to Nagoya University, Nagoya, Japan

b Università di Bari, Bari, Italy

c Università di Cagliari, Cagliari, Italy

d Università di Napoli “Federico II“, Napoli, Italy

e Associated to Gyeongsang National University, Jinju, Korea

g Università della Basilicata, Potenza, Italyk

h Università di Napoli Parthenope, Napoli, Italyk

i Università degli Studi del Sannio di Benevento, Benevento, Italyk

j Associated to SAPHIR, Chile

k Associated to Università di Napoli “Federico II“, Napoli, Italy

l Currently at the University of Bergamo, Bergamo, Italy

m Also at the University of Pisa, Pisa, Italy

n Pegaso University, Napoli, Italy

o Departamento de Física, Instituto Superior Técnico, Universidade de Lisboa, Lisbon, 1049-001, Portugal

References

  1. [1]SHiP Collaboration, BDF/SHιP at the ECN3 high-intensity beam facility, Tech. Rep. CERN-SPSC-2023-033, SPSCP-369 (CERN, Geneva, 2023).
  2. [2]M. Climescu (SHiP), The SHiP/NA67 experiment at the ECN3 high-intensity beam facility at the CERN SPS (2026), Rencontres de Blois 2026 proceedings, arXiv:2606.11039 [hep-ex].arxiv.org/abs/2606.11039
  3. [3]A. Aguilar et al. (LSND), Phys. Rev. D 64, 112007 (2001), arXiv:hep-ex/0104049.DOI
  4. [4]A. A. Aguilar-Arevalo et al. (MiniBooNE), Phys. Rev. D 103, 052002 (2021), arXiv:2006.16883.DOI
  5. [5]G. Mention, M. Fechner, T. Lasserre, T. A. Mueller, D. Lhuillier, M. Cribier, and A. Letourneau, Phys. Rev. D 83, 073006 (2011), arXiv:1101.2755.DOI
  6. [6]V. V. Barinov et al., Phys. Rev. C 105, 065502 (2022), arXiv:2201.07364.DOI
  7. [7]M. Dentler, A. Hernández-Cabezudo, J. Kopp, P. A. N. Machado, M. Maltoni, I. Martinez-Soler, and T. Schwetz, JHEP 08 (2018), 010, arXiv:1803.10661 [hep-ph].DOI
  8. [8]M. A. Acero et al., J. Phys. G 51, 120501 (2024), arXiv:2203.07323 [hep-ex].DOI
  9. [9]K. Akita, O. Inhatenko, and M. Ovchynnikov, Late reheating in the era of precision cosmology (2026), arXiv:2610.36062 [hep-ph].arxiv.org/abs/2610.36062
  10. [10]N. Aghanim et al. (Planck), Astron. Astrophys. 641, A6 (2020), arXiv:1807.06209.DOI
  11. [11]M. S. Madhavacheril et al. (ACT), Astrophys. J. 962, 113 (2024), arXiv:2304.05203 [astro-ph.CO].DOI
  12. [12]M. Abdul Karim et al. (DESI), Phys. Rev. D 112, 083515 (2025), arXiv:2503.14738 [astro-ph.CO].
  13. [13]W. Elbers et al. (DESI), Phys. Rev. D 112, 083513 (2025), arXiv:2503.14744 [astro-ph.CO].arxiv.org/abs/2503.14744
  14. [14]Y. Mellier et al. (Euclid), Astron. Astrophys. 697, A1 (2025), arXiv:2405.13491 [astro-ph.CO].DOI
  15. [15]M. Archidiacono et al. (Euclid), Astron. Astrophys. 693, A58 (2025), arXiv:2405.06047 [astro-ph.CO].DOI
  16. [16]Ž. Ivezić et al. (LSST), Astrophys. J. 873, 111 (2019), arXiv:0805.2366 [astro-ph].
  17. [17]S. Mishra-Sharma, D. Alonso, and J. Dunkley, Phys. Rev. D 97, 123544 (2018), arXiv:1803.07561 [astro-ph.CO].
  18. [18]P. Ade et al. (Simons Observatory), JCAP 02 (2019), 056, arXiv:1808.07445 [astro-ph.CO].DOI
  19. [19]M. Abitbol et al. (Simons Observatory), JCAP 08 (2025), 034, arXiv:2503.00636 [astro-ph.CO].
  20. [20]F. Dydak et al., Phys. Lett. B 134, 281 (1984).
  21. [21]I. E. Stockdale et al., Phys. Rev. Lett. 52, 1384 (1984).
  22. [22]MINOS+ Collaboration, Phys. Rev. Lett. 122, 091803 (2019), arXiv:1710.06488.
  23. [23]G. Cheng et al. (MiniBooNE, SciBooNE), Phys. Rev. D 86, 052009 (2012), arXiv:1208.0322.DOI
  24. [24]NOvA Collaboration, NOvA Dual-Baseline Search for Active-to-Sterile Neutrino Oscillations using Neutrino- and Antineutrino-Enriched Samples (2026), arXiv:2609.08900 [hep-ex].
  25. [25]ICARUS Collaboration, First search for sterile neutrino oscillation leading to νμ disappearance in the Booster Neutrino Beam at ICARUS (2026), arXiv:2603.22557 [hep-ex].arxiv.org/abs/2603.22557
  26. [26]K. Abe et al. (Super-Kamiokande), Phys. Rev. D 91, 052019 (2015), arXiv:1410.2008.
  27. [27]R. Abbasi et al. (IceCube), Phys. Rev. D 110, 072007 (2024), arXiv:2407.01314 [hep-ex].DOI
  28. [28]C. Kraus, A. Singer, K. Valerius, and C. Weinheimer, Eur. Phys. J. C 73, 2323 (2013), arXiv:1210.4194.DOI
  29. [29]A. I. Belesev et al., J. Phys. G 41, 025001 (2014), arXiv:1307.5687.
  30. [30]H. Acharya et al. (KATRIN), Nature 648, 70 (2025), arXiv:2503.18667 [hep-ex].arxiv.org/abs/2503.18667
  31. [31]K.-Y. Choi, S. H. Kim, Y. G. Kim, K. Y. Lee, K. S. Lee, B. D. Park, J. Y. Sohn, S. M. Yoo, and C. S. Yoon, JHEP 06 (2024), 166, [Erratum: JHEP 01 (2025) 036], arXiv:2403.04191 [hep-ph].DOI
  32. [32]K.-Y. Choi, Y. S. Jeong, S. H. Kim, Y. G. Kim, K. Y. Lee, K. S. Lee, B. D. Park, J. Y. Sohn, S. M. Yoo, and C. S. Yoon, Phys. Rev. D 112, 075040 (2025), arXiv:2505.12785 [hep-ph].arxiv.org/abs/2505.12785
  33. [33]F. Takahashi et al. (Particle Data Group), Int. J. Mod. Phys. A 41, 2630011 (2026).
  34. [34]D. Cianci, A. Furmanski, G. Karagiorgi, and M. Ross-Lonergan, Phys. Rev. D 96, 055001 (2017), arXiv:1702.01758 [hep-ph].DOI
  35. [35]C. Andreopoulos et al., Nucl. Instrum. Meth. A 614, 87 (2010), arXiv:0905.2517 [hep-ph].DOI
  36. [36]S. Agostinelli et al. (GEANT4), Nucl. Instrum. Meth. A 506, 250 (2003).
  37. [37]W. M. Bonivento (SHiP), JINST 13 (2018), C02041.
  38. [38]M. A. V. Climescu, Design and development of calorimetry at SHiP and SND@LHC, Ph.D. thesis, Mainz U., Johannes Gutenberg University Mainz (2025).DOI
  39. [39]T. Chen and C. Guestrin, in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (2016) pp. 785–794, arXiv:1603.02754 [cs.LG].
  40. [40]R. L. Gluckstern, Nucl. Instrum. Meth. 24, 381 (1963).
  41. [41]DUNE Collaboration, Deep Underground Neutrino Experiment (DUNE), Far Detector Technical Design Report, Volume II: DUNE Physics (2020), arXiv:2002.03005 [hep-ex].
  42. [42]T2K Collaboration, Phys. Rev. D 96, 092006 (2017), arXiv:1707.01048 [hep-ex].
  43. [43]C. Ahdida et al. (SHiP), Eur. Phys. J. C 80, 284 (2020), arXiv:2001.04784 [physics.ins-det].
  44. [44]J. Gomez et al., Phys. Rev. D 49, 4348 (1994).
  45. [45]B. G. Tice et al. (MINERvA), Phys. Rev. Lett. 112, 231801 (2014), arXiv:1403.2103.DOI
  46. [46]PROSPECT Collaboration, Phys. Rev. D 103, 032001 (2021), arXiv:2106.11210.
  47. [47]I. G. Alekseev et al. (DANSS), Phys. Atom. Nucl. 88, 406 (2025).
  48. [48]M. Aker et al. (KATRIN), Phys. Rev. D 105, 072004 (2022), arXiv:2201.11593 [hep-ex].DOI
  49. [49]J. N. Abdurashitov et al., Pisma Zh. Eksp. Teor. Fiz. 105, 723 (2017), arXiv:1703.10779.DOI
  50. [50]M. Galeazzi, F. Fontanelli, F. Gatti, and S. Vitale, Phys. Rev. Lett. 86, 1978 (2001).
  51. [51]G. J. Feldman and R. D. Cousins, Phys. Rev. D 57, 3873 (1998), arXiv:physics/9711021.DOI
  52. [52]E. Cortina Gil et al. (NA62), JINST 12 (2017), P05025, arXiv:1703.08501 [physics.ins-det].
  53. [53]M. Ferrillo, M. Ovchynnikov, F. Resnati, and A. De Roeck, JHEP 02 (2024), 196, arXiv:2312.14868 [hep-ph].DOI
  54. [54]V. L. Highland, Nucl. Instrum. Meth. 129, 497 (1975).
  55. [55]T. Bolton, High-energy muon momentum estimation from multiple Coulomb scattering in dense detectors (1997), arXiv:hep-ex/9705007.DOI

Paper details

vlnc:W_k7ys6vhe
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 6

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents