Fermilab Modelling detector-specific reconstruction uncertainties in LAr-TPC FERMILAB-TM-2791-STUDENT This manuscript has been authored by Fermi Forward Discovery Group, LLC under Contract No. 89243024CSC000002 with the U.S. Department of Energy, Office of Science, Office of High Energy Physics. Fermi National Accelerator Laboratory Internship Report Modelling detector-specific reconstruction uncertainties in LAr-TPC Lorenzo Vincenzo D’Auria1 *, under the supervision of Harry Hausner, Angela Fava Abstract The Short-Baseline Neutrino (SBN) program features three Liquid Argon Time Projection Chamber (LAr-TPC) detectors positioned along the Booster Neutrino Beam (BNB) axis: the Short Baseline Neutrino Near Detector, MicroBooNE, and the ICARUS T600. As the largest operational LAr-TPC, ICARUS T600 serves as the far detector, located 600 m from the BNB target. While its primary goal is to record neutrino events, it also detects other ionizing events, including cosmic rays. This work focuses on analyzing and modeling detector-specific reconstruction uncertainties in LAr-TPC. These inefficiencies, identified during the Pattern Recognition phase handled by the PANDORA algorithm, impact subsequent Particle Fits and Offline Analysis. Specifically, inaccuracies in track reconstruction can lead to significant physical consequences, such as erroneous particle energy estimates and poor Particle Identification (PID), reducing the efficiency of neutrino event characterization. A key issue addressed is split tracks, caused by missing hits or incomplete track stitching by PANDORA. The aim of this internship is to characterize, model, and quantify the impact of split tracks on track reconstruction. Keywords ICARUS — SBN — Neutrino — SplitTracks 1 Department of Physics and Astronomy, Universiity of Padua, Padua, Italy *Author: lorenzovincenzo.dauria@studenti.unipd.it Contents 1. Introduction 1 Introduction 1 1.1 Physics Program 1.1 Physics Program . . . . . . . . . . . . . . . . . . . . . . . 1 Experimental observation of neutrino oscillations have estab- 1.2 The Short Baseline Neutrino program . . . . . . . . 2 lished a picture consistent with the mixing of three neutrino fla- vors (νe , νµ , ντ ) with three mass eigenstates (ν1 , ν2 , ν3 ) whose 1.3 The ICARUS T600 Detector . . . . . . . . . . . . . . . 3 mass differences turn out to be relatively small. However, sev- 2 Event Reconstruction in ICARUS 4 eral experimental anomalies have been reported which could 2.1 PANDORA reconstruction algorithms . . . . . . . . 5 be hinting at the presence of additional neutrino states with 2.2 LArSoft event scanner . . . . . . . . . . . . . . . . . . . 5 larger mass-squared differences participating in the mixing. Mainly two distinct classes of anomalies pointing at additional 3 Split Tracks Identification 6 physics beyond the Standard Model (SM) in the neutrino sec- 3.1 Plug-in logic and observables calculation . . . . . 6 tor have been reported, namely [1]: 3.2 Determining the selection cuts . . . . . . . . . . . . . 7 3.3 Validating the plug-in capabilities . . . . . . . . . . . 8 (a.) Reactor and Gallium Anomaly: Refers to the deficit of ν̄e observed in numerous detectors a few meters 4 Spatial distribution of split track events 10 away (short baselines) from nuclear reactors compared 4.1 Missed Channel distribution . . . . . . . . . . . . . . 10 to the predicted rates. A similar indication comes from 4.2 HARPS Tool . . . . . . . . . . . . . . . . . . . . . . . . . 11 radioactive νe source in Gallium solar neutrino experi- 5 Conclusion 12 ments (e.g., GALLEX and SAGE). This deficit could be Acknowledgments 12 explaines through νe disappearance due to oscillations at ∆m2 ≥ 1eV 2 . References 13 (b.) LSND/MiniBooNE anomaly: Refers to the evidence for electron-like excesses in interaction with νµ and ν̄µ from particle accelerator in Liquid Scintillator Neutrino Detector (LSND) and MiniBooNE experiments. Modelling detector-specific reconstruction uncertainties in LAr-TPC — 2/13 The most common interpretation of this collection of data is evidence for the existence of one or more additional, mostly “sterile” neutrino states with masses at or below the few eV range. The minimal model consists of a hierarchical 3+1 neutrino mixing and the new sterile neutrino would mainly be composed of a heavy neutrino ν4 with mass m4 such that ∆m241 ≈ [0.1 − 10]eV 2 and m1 , m2 , m3 << m4 [2]. In this 3+1 minimal extension to the SM, the effective νe and νµ disappearance probabilities are described by: Figure 1. Short Baseline Neutrino project layout. ∆m241 L   Pν3+1 α →νβ = δαβ − 4|Uα4 |2 (δαβ − |Uβ 4 |2 )sin2 4Eν where Ui j are elements of the now 4x4 mixing matrix (exten- Finally, the BNB is produced by extracting protons from the sion of the 3x3, UPMNS , mixing matrix) and L is the travel Booster accelerator at 8 GeV kinetic energy and directing distance of the neutrino of energy Eν . them onto a beryllium target to generate a secondary beam of The Short-Baseline Neutrino (SBN) physics program [3] fits hadrons (primarily π). These hadrons are focused and allowed in this framework and it was set up to: to propagate down a 50 m long air-filled tunnel, where the majority decay to produce νµ and νe ; the remaining hadrons 1. Understand the low energy excess found in MiniBooNE are absorbed by a concrete and steel absorber at the tunnel’s by exploiting MicroBooNE; end. The physics program of ICARUS is enhanced by the stand- 2. Search for sterile neutrino with Short-Baseline Near alone studies of neutrino cross-section relevant to the Long Detector (SBND) and ICARUS (both in the appearance Baseline Neutrino Facility program, thanks to the off-axis and disappearance channels); neutrinos from the NuMI beam. Data collected with the NuMI 3. Further develop the LAr-TPC technology and measure beam will also be exploited for searches beyond the SM and the ν − Ar cross section in the GeV region for future dark sector analyses. The two beam-lines are schematized Long Baseline experiments (e.g., DUNE). along with the rest of the accelerator complex in Figure 2. In the following sections, the SBN program will be explored in depth, illustrating the operating principle of the LAr-TPC and paying particular attention to ICARUS. 1.2 The Short Baseline Neutrino program The Short-Baseline Neutrino (SBN) program [2] includes three Liquid Argon Time Projection Chamber (LAr-TPC) de- tectors located on-axis in the Booster Neutrino Beam (BNB): 1. The Short Baseline Near Detector (SBND, 2024-now) will be located 110 m from the BNB target; 2. The MicroBooNE detector (2015-2021), located in the Liquid Argon Test Facility (LArTF) at 470 m, is a 170- ton total mass (89-ton active mass) LAr-TPC contained within a conventional cryostat. It was developed to measure neutrino interactions in argon for multiple re- action channels and to investigate the source of the Figure 2. Schematic of the Fermilab accelerator complex. currently unexplained excess of low-energy electromag- netic events observed by MiniBooNE. 3. The ICARUS T600 detector (2021-now) serves as the far detector within the SBN program, positioned 600 LAr-TPC Working Principle All the SBN detectors are m from the target along the BNB axis. It contains 760 based on LAr-TPC experimental teechnique [4]: this device tons of LAr (476-ton active mass). For a more in-depth has the potential to reconstruct tracks and showers with high description see section 1.3. level of detail and efficiency, as well as to provide a precise measurement of ionization charge necessary for good particle Figure 1 provides a schematic view of the SBN detectors. identification based on ionization energy loss. The LAr-TPC principle of operation is schematically illustrated in figure 3 Modelling detector-specific reconstruction uncertainties in LAr-TPC — 3/13 ultra purified liquid argon. In each module there are two TPCs separated by a central, vertical, common cathode, in which there is an uniform electric field of ≈ 500V /cm going from the cathode to anode, with a maximum drift time of about 1ms. As said, each TPC has 3 parallel read-out planes, 3mm apart from each other: the first plane (induction 1) has horizontal parallel wires, while the other two (induction 2 and collection) are oriented ±60o with respect to it. Each of the TPCs consists of 13332 wires: the reconstruction of the image of the charged particle traversing the volume is obtained by combining the coordinates in the wires of each of the planes. For a more in-depth description of event reconstruction see the section 2. Figure 3. The LArTPC Principle of Operation. Ionization electrons are produced in the medium by ioniz- ing particles associated with an event; due to an electrostatic field maintained in the chamber, they start drifting to the Figure 4. ICARUS T600 diagram. wires, while positive ions travel towards the cathode. Drift- ing electrons then create a signals on the wires belonging to the induction wire planes “U” and “V”. When the finally Light Collection System (PMT) Ionization in the LAr TPC reach the collection plane “Y” (where they are removed from is accompanied by a scintillation light emission, which is the detector volume) a signal is also produced on its wire. useful for a possible absolute time measurement of an inter- Then the ionization pattern can be reconstructed in the plane action and triggering. A large set of 360 PMTs (90 for each perpendicular to the drift direction by analyzing signals on TPCs) is immersed in the liquid argon, so UV photons from (U,V,Y); every induction and collection wire is then readout the scintillation provide a signal that allows a measurement as an individual channel. Reconstruction along the axis of the of the drift time, and hence of the distance traveled by the drift direction can be done using the information contained in drifted electrons. The photomultipliers selected for each TPC the time evolution of the signals on the (U,V,Y) wire planes were 90 Hamamatsu R5912- MOD of 8 diameter hemispheral through the know value of drift velocity. glass mounted behind the wire chambers and adapted to op- erate at cryogenic temperatures on each TPC. The electronic 1.3 The ICARUS T600 Detector readout system of the PMTs is designed to allow continuous The ultimate goal of ICARUS T600 detector [5] is to record readout, digitization and an independent waveform recording neutrino events, however, in addition to them, the detector of the signals coming from each one of the PMTs of the light records all any kind of ionizing events, also crossing cosmic detection system. In figure 5 it is possible to appreciate the rays. To be able to discriminate interesting events, the analysis arrangement of the PMTs inside the ICARUS T300 module relies on the information provided by the various detector components: TPC, PMT and CRT sub-system. In this section the three subsystems are described in more detail . Cosmic Ray Tagger (CRT) The CRT system is a subdetector external to the cryostats, designed to identify charged particles coming from outside the beam and passing through or near the active volume of the TPC. It covers an area of ≈ 1100m2 and its divided into three subsystems (top, side and bottom CRT), each covering different regions of the TPC. All these subsystems have a time resolution of the order of ns, allowing a filtering of events that rejects those where the primary event trigger was tagged in the subsystem. The CRT is designed to tag ≈ 95% of the cosmic particles entering the TPCs. Time Projection Chamber (TPC) ICARUS consists of two identical adjacent modules (cryostats, named west and east Figure 5. Internal view of an ICARUS T300 module, during modules with respect to the BNB beam direction, see figure the refurbishing activities at CERN. 4) with dimensions of 3.6 × 3.9 × 19.6 m3 with 760 tons of Modelling detector-specific reconstruction uncertainties in LAr-TPC — 4/13 Figure 6. Illustration of the ICARUS TPC signals processing and reconstruction process. 2. Event Reconstruction in ICARUS induced by a charged particle on a wire and are the primary input to the pattern recognition algorithms. LAr-TPCs design and acquisition parameters lead to a sub- stantial data rate and volume: to deal with it, several strategies Stage 1 The second stage (Stage 1) processing is mainly have been introduced in the ICARUS data processing chain. focused on the reconstruction of TPC, CRT and PMT signals The output data from the detector consists of digitized wave- (and might include calibrations of each subsystem). Stage 1 forms from each TPC readout channel, representing the charged processing for TPC includes: induced by the motion of ionization electrons swept by the drift field from the TPC volume. All the collected raw data 1. Hit filter: Builds tridimensional space points from com- needs to undergo several software phases before it can be bination of 2D Hits across different wire planes. properly analysed; the reconstruction of these events is done in two stages: Stage 0 and Stage 1. 2. Pattern recognition algorithm (PANDORA): Arranges close hits into clusters that are then used to Stage 0 The first stage (Stage 0) translates from raw data identify track or shower candidates and event topology format to LArSoft1 format for offline analysis. It also per- information, including cosmic ray identification. Inside forms signal processing for all three subsystems with the goal each defined interaction, the so-called slices, a hierar- of identifying physical signal (Hits) to be fed to the poat- chy among all reconstructed objects is build to identify tern recognition and event reconstruction algorithms. Stage 0 parent-daughter relationships. A more detailed descrip- processing for TPC includes: tion of PANDORA’s reconstruction algorithms is given in the section 2.1. 1. Decoding: Decompresses and unpacks raw data into a format that can be used in later steps. Includes elec- 3. Particle fits: Applies detailed algorithms to reconstruct tronic noise filtering to remove TPC coherent noise. tracks and showers and obtain calorimetric measure- 2. Deconvolution: Removes effect of the electrostatic field ments of each particle. It also provides information around the wires and electronic response. to allow kinematic reconstruction and to analyze the event. 3. Region Of Interest (ROI) finding: Reduces data volume by selection of ROI around candidate signal. A schematic view of the event reconstruction algorithm in ICARUS TPCs is shown in figure 6. 4. Hit Finding: Builds “Hit” objects from the ROIs in the previous step. A Hit represents the identified signal 1 LArSoft is a toolkit that provides a software infrastructure and algorithms for the simulation, reconstruction and analysis of events in LAr-TPC. Modelling detector-specific reconstruction uncertainties in LAr-TPC — 5/13 2.1 PANDORA reconstruction algorithms on their position relative to the cathode. Channels in each PANDORA has two main chain algorithms for event recon- readout plane are ordered first by increasing z-coordinate and struction in neutrino detectors [6]: PANDORACosmic and then by increasing y-coordinate. The cryostats are labeled PANDORANu, targeting the reconstruction of interactions as C:0 for module E and C:1 for module W. Each TPC set due to cosmic rays and neutrinos respectively. (labeled S) corresponds to an actual TPC and drift volume, and includes two logical TPCs (each half of the physical TPC, PANDORACosmic PANDORACosmic clusters 2D hits on the division of which occurs at the separation of the horizontal each readout plane, splitting them at bifurcations or ambigu- wires), labeled T. The TPC sets are further numbered with ities to ensure high purity. Clusters are then merged based increasing x-coordinate: TPC EE is C:0 S:0, TPC EW is C:0 on proximity or directional alignment to improve complete- S:1, TPC WE is C:1 S:0, and TPC WW is C:1 S:1. Readout ness. These refined 2D clusters undergo 3D reconstruction by planes are groups of channels on the same type of plane within matching consistent 2D groups across all three planes, with a TPC set, with each readout plane potentially containing suitability scored and evaluated using a χ 2 test to resolve one or more logical wire planes. The channel numbering ambiguities. Unassociated 2D clusters, assumed to be delta starts from the first TPC set and the readout plane closest ray fragments, are dissolved. The initial position for cosmic to the cathode, with an order based on z and y coordinates. muons is assigned at the highest vertical coordinate, with Wireless channels are associated only with readout planes, secondary particles linked hierarchically. not wire planes. In total, 55,296 channel IDs are assigned, Tracks are classified as clear cosmic if they cross the detector with 53,248 connected to TPC wires and the rest allocated to boundaries, have hits outside the drift volume, or have incom- wireless, ghost, and virtual channels. LArSoft wire planes are patible time corrections. Once clear cosmics are removed, numbered starting from the plane closest to the cathode (P:0). the remaining clusters are input to PANDORANu, which processes them into slices to isolate neutrino interactions or cosmic remnants, resulting in neutrino candidates. PANDORANu The first step in PANDORANu reconstruc- tion involves using track-oriented clustered hits to generate possible 3D vertex candidates. A quality cut evaluates the proximity of each vertex to hits in all plane views, and a scoring algorithm selects the most likely neutrino interaction vertex. The score is based on three factors: how well particles point back to the vertex, asymmetry between downstream and upstream hits, and proximity to the beam origin. The candidate with the highest score is chosen, and 2D clusters crossing the vertex are split, creating new clusters on each side. Next, 3D track reconstruction proceeds similarly to PAN- DORACosmic, but with additional reconstruction of primary electromagnetic showers. A Boosted Decision Tree (BDT) then classifies each particle as shower-like or track-like. Sub- sequent steps refine this classification and organize particles into a parent-daughter hierarchy. The final structure starts from the 3D neutrino vertex, adding branches for primary particles and leaves for particles produced by primary parents. The interaction grows until all particles in the slice are asso- ciated, resulting in a single neutrino particle with an internal particle hierarchy that represents the flow of the interaction. Figure 7. Illustation of wire plane layout as referenced in [7]. 2.2 LArSoft event scanner An important tool for the work lies in the possibility of view- ing the events reconstructed by PANDORA using the LArSoft event display: in this way it is possible to directly observe the events, understanding their characteristics and highlighting any anomalies in the reconstruction. LArSoft Wire Geometry LArSoft assigns logical IDs to each TPC readout channel, organizing the ICARUS detector into two cryostats, each with two TPC sets [7]. Each TPC set contains four readout planes, with channels numbered based Modelling detector-specific reconstruction uncertainties in LAr-TPC — 6/13 3. Split Tracks Identification 1. Loading of track pairs and calculation of variables; The purpose of this work is to analyze and model detector- specific reconstruction uncertainties in LAr-TPC. Such re- 2. Selection cuts; construction inefficiencies are identified during the “Pattern Recognition” phase (handled by the PANDORA algorithm 3. Storage of results in .root files. already discussed in the previous section) and significantly affect the subsequent steps of Particle Fits and High Level The following paragraphs will describe the content of the Analysis. In particular, incorrect track reconstruction directly various parts of the code in more detail. leads to significant physical consequences such as erroneous estimation of particle energies or poor Particle Identification (PID), which together result in a substantial reduction in the Importation and calculation of variables The plug-in, start- efficiency of characterizing events associated with neutrino ing from a track identified with an index consistent with that interactions. assigned by PANDORA, iterates over all the tracks, construct- An important class of detector-specific reconstruction uncer- ing a series of possible pairs. For each track in the pair, some tainties in LAr-TPC is represented by “Splitting Tracks”: basic geometric characteristics are obtained: the logical TPC these are caused by missing hits that were not accounted containing the track, its length, and the position and direc- for at the reconstruction level or that PANDORA was not able tion of the start and end points. At this point, the number of to account for when stitching (joining together the segments possible pairs is significantly reduced through a pre-selection belonging to the same particle track) the divided track seg- based on track length, requiring that the tracks be longer than ments. At the detector level, the most common known reasons Lth = 5cm: this threshold, currently purely indicative, is set for track splitting are [8]: considering the physical hypothesis that a track splitting event is more likely to occur for physically longer tracks. After • Malfunctioning channels: as it is known that a set of 32 this pre-selection, the track pairs are significantly reduced, wires is disconnected in one of the TPCs; and two fundamental geometric variables between them are calculated: the gap separating them (calculated as the norm • Presence of the cathode plane; of the segment connecting the endpoint of the first track to • Electric field distortions: produced by the fact that the the start point of the second) and the angle between them. Induction 1 plane is made by two sets of wires joined However, the angle calculation is much more complex and in the middle of the plane, with a few cm wide region requires further explanation. not equipped with them. Internship Aim The aim of the internship lies in character- The angle between the tracks is calculated based on two fun- izing and modeling the split tracks and quantifying this effect damental cases: in track reconstruction. These results were achieved through two main steps: • Case of completely parallel tracks: in this case, the an- gle θ between the two tracks is obtained as • The first step consists in writing a plug-in, based on LArSoft’s ART framework, whose purpose is to identify the presence of split tracks based on their geometric θ = π −α characteristics (length, gap, and angle between them). The calculation of observables will be discussed in more With α being the angle between the direction vector of detail in section 3.1, while the determination of the cuts the endpoint of the first track and the vector connecting to be made to discriminate the split tracks is discussed the endpoint of the first track and the start point of the in section 3.2. Finally, the plug-in’s selection capability second; will be validated in section 3.3; • Once the performance of the plug-in has been validated, • Case of non-parallel tracks: In this case, the two tracks the distribution of lost channels and gap lengths will may not intersect in R 3 , so the direction of the last be recorded (section 4.1): once the behavior of these point of the first track is projected forward and the di- distributions is understood, they are implemented in rection of the first point of the second track is projected the HARPS tool, with the aim of generating a dataset backward. The midpoint M of the perpendicular line where hits are removed from the tracks according to the minimizing the distance between them is then found, gap distribution (section 4.2). and the angle θ formed by the end of the first track, M, and the start of the second track is stored. To streamline 3.1 Plug-in logic and observables calculation the discussion, the detailed calculations are reported in The plug-in written for the identification of split tracks, and Appendix. A graphical representation of the process is used on Stage 1 files, is based on three main steps: shown in Figure 8. Modelling detector-specific reconstruction uncertainties in LAr-TPC — 7/13 information about the tracks (such as the IDs of the tracks identified as a pair) is also printed to the screen, allowing for cross-checks with what is visualized using the event scanner (this function will be used to validate the plug-in’s ability to select split track events and will allow for the visual identifi- cation of any anomalies, see Section 3.3 for more details). 3.2 Determining the selection cuts This section discusses the results of applying the plug-in, particularly validating the plug-in’s ability to satisfactorily reconstruct the observables of interest. The primary goal here is not to evaluate the plug-in’s selection capability but rather the consistency of the reconstructed variables’ distribution with the reference values. The run under study (hereafter re- Figure 8. (Top figure) Example of 3D projection; (Bottom ferred to as ”Run 9435”) has previously been studied using an figure) Example of the stored angle. event scanner, through which information about the presence or absence of split tracks has been obtained, and the relevant variables were calculated. In the following, “reconstructed A major issue that emerged when applying the methods pro- variables” refers to those calculated by the plug-in, while “true posed above is associated with tracks crossing the cathode: variables” refer to those reported in the archive. these tracks exhibit a distorted profile at the ends (as shown in Figure 9), which alters the angle estimation when consid- To validate the reconstruction capability, the search was lim- ering only the extremal points of the tracks. To address this ited to events where the presence of split tracks was certain, issue, the direction of the first and second tracks is obtained and, in addition to the pre-cut on the threshold length, a cut on as the average of the last and first N = 20 hits, respectively. the gap between two split tracks was applied, selecting only Although simplistic, this solution provides a satisfactory es- pairs of tracks with a gap smaller than 25cm. The results are timate of the angle between the tracks (as confirmed in the shown in Figure 10. analysis presented in Section 3.2). Figure 9. Example of anomalous behavior in the angle calculation: case of a track crossing the cathode. Selection cuts Once the observable reconstruction capa- bilities have been validated, it is necessary to determine the optimal cut values to apply to the variables in order to dis- criminate split track pairs from regular track pairs. For this Figure 10. Comparison between reconstructed and true purpose, a run whose split track information was previously variables. Light green represents reference variables, red obtained from the event display is used, and for a more de- represents reconstructed variables, and dark green shows the tailed description of the selection process, refer to Section overlap of the histograms. 3.2. Storage of results The output results from the plug-in lead It can be seen that there is satisfactory agreement both in the to the production of a .root file consisting of different TTrees selection degree and, especially, in the reconstruction of the depending on the cuts applied: only on the gap, on gap and variables. Specifically, for the length calculation, the agree- angle, on gap-angle-length. For each track pair that passes ment is almost total (fewer than three events where the plug-in the pre-selection on the length threshold, the observables are does not detect due to the too-tight gap cut). also computed and saved, so that comparisons with the data After confirming the quality of the reconstruction, the distribu- sets after the cuts can be made. For each program execution, tion of the same variables is analyzed for all track pairs present Modelling detector-specific reconstruction uncertainties in LAr-TPC — 8/13 in the run (which, let us remember, pass the pre-selection). At this point, it was first decided to construct the histograms The trend is shown in figure 11. of the variables of interest for the entire dataset to understand the overall trend and compare it with what was obtained in the previous section. Figure 12 shows the trend of the variables for the 7xxx runs. Figure 12. Distribution of the variables of interest for other Figure 11. Distribution of variables for all track pairs testing runs. passing the pre-selection. Based on the event scanner visualization of the split tracks, A trend similar to that obtained for run 9435 is identified (al- some recurring geometric characteristics can be observed. In though with greater statistics). Starting from an initial number fact, they must be: of N = 4, 362, 526 events, the cuts are applied sequentially to understand the reduction in statistics. Specifically: • Parallel and close: We expect the splitting to occur in a reduced portion of space and likely along a straight (a.) With the Gap selection only, the number of tracks that trajectory; pass this selection is approximately 0.12%. An evident • Long: Avoiding considering isolated particles as be- problem related to the spatial distribution of the tracks longing to the same track, while also imposing a con- is identified, identifying all nearby tracks as split tracks siderable track length in relation to the requested gap leads to the error of evaluating positive match associated length. with different particles (e.g., tracks 22-29, Figure 13). It is evident that it is necessary to limit the observation While the first characteristic is easily understandable, we can to only parallel tracks (which most likely refer to the convince ourselves of the second by comparing the length passage of the same particle): from the distribution of distributions for all track pairs and for known split tracks. It is the angles, it is possible to highlight the consistency of possible to highlight that most of the counts (in the lower plots the angle cut with what was found in run 9435.. of figure 11) occur for lengths shorter than 20cm; whereas, in the split track distribution, evident percentages are also present at higher lengths. Therefore, a cut around the length of Lcut = 20cm was chosen (which will later be refined to Lcut = 40cm by analyzing data samples from other runs), significantly reducing the total number of tracks at the cost of losing a small fraction of split tracks. 3.3 Validating the plug-in capabilities After validating the plug-in’s ability to reconstruct the vari- ables of interest, in order to optimize the cuts to select split track events and validate the selection capability, a dataset comprising eleven different testing runs was used. The choice to use all these runs is motivated by the desire to have suffi- cient statistics to perform a thorough analysis and validate the reconstruction capability on a larger sample. Figure 13. Example of an event where the Gap-only selection leads to the incorrect identification of split track events. Modelling detector-specific reconstruction uncertainties in LAr-TPC — 9/13 (b.) By making a further selection on the angle, the number of tracks that pass the selection is about 0.06%. It is possible to show how the tracks identified as split tracks are now parallel and close to each other, but the selec- tion is still not satisfactory; as can be observed (e.g., tracks 30-34, Figure 14), there are many tracks that are too short compared to the gap separating them, leading to an incorrect estimation of their characteristics: this suggests implementing an additional cut on the length (consistent with the one performed for run 9435). Figure 16. Distribution of the observables of interest as the applied selection varies. Some observations can be made: first of all, by applying a selection only on the gap, the angle distribution flattens significantly in the central region and shows peaks at the extremes, reflecting the actual characteristic of split track events being close and parallel (thus with angles close to 0 or π). As the selection step increases, the statistics are significantly reduced, but the trend from the previous cuts is maintained: particularly in the case of track lengths, the final cut removes a large portion of events with short lengths that Figure 14. Example of event where the Gap+Angle selection were not related to split tracks (as already justified in previous leads to the incorrect identification of split track events. sections). It can thus be concluded that the plug-in has the ability to retrieve the variables of interest and identify split track events with a high degree of accuracy, allowing further analysis in the spatial distribution of split track events (Section (c.) Following the subsequent selection on the length of the 4). tracks, pairs are identified that are indeed split tracks (e.g., tracks 3-4, Figure 15). The number of tracks that pass the selection is about 0.007%, perfectly in agreement with the percentage of split tracks present in run 9435. Figure 15. Example of an event where the identification of a split track phenomenon occurs successfully. For an overview, Figure 16 shows a comparison of the variable trends at each stage of the selection Modelling detector-specific reconstruction uncertainties in LAr-TPC — 10/13 Figure 17. Distribution of the number of lost channels as a function of the first channel that loses the event. The different colors indicate the different planes according to the correspondence channels-physical planes of the wires implemented in LArSoft. 4. Spatial distribution of split track events the second track and the closest to the end of the first). An important note is that the plugin applies a veto before stor- In this section, the spatial distribution of the split track events, ing these variables: to avoid storing channels from different identified with the plug-in written and presented in the previ- cryostats, only the channels from the cryostat where there is ous sections, will be studied. The spatial distribution will be correspondence with the passage of the track are recorded. investigated by identifying the missed channels in the pairs Once these two variables are stored, it is possible to produce of tracks identified as split tracks, and analyzing the distri- a two-dimensional plot, shown in Figure 17, to verify its trend. bution across the different wire planes (Section 4.1). This study will further serve as the basis for the implementation of missed channels in the HARPS tool (Hit Activity Removal from Particles for Systematics), developed by Bruce Howard The highlighted zones in the figure represent, according to and Harry Hausner. The goal of HARPS is to modify the the wire geometry described earlier and the correspondence events by removing certain hits, according to the channel lost with the channels, the three wire planes present in ICARUS: distribution, and thus generate modified samples to study how two Induction 1 planes (IND1-S and IND1-N, which are split reconstruction changes depending on the applied modification. at z = 0), the Induction 2 plane (IND2), and the Collection It can work in two ways: drop context (i.e., keeping only a plane (COLL). The division naturally follows from knowing specific selected particle) or keep only context (i.e., selecting the correspondence between planes and wires (as described only a specific particle to perform systematic studies). In in [7]): knowing the first missed channel, it is possible to Section 4.2, the form of the plug-in, the changes made, and an construct, for each TPC, four independent sets of Gap data application example will be presented. (in terms of the number of missed channels) reflecting the division in the different planes. Then, to increase the statistics, 4.1 Missed Channel distribution the counts are summed for each TPC and cryostat. The search for missed channels is implemented in the plugin described in the previous sections. Specifically, once a split track phenomenon is identified based on geometric character- istics, the nearest channels to the endpoint of the first track From the distribution presented in Figure 17, a uniform distri- and the starting point of the second track are searched. At bution is evident for the first missed channel for each plane: this point, two key pieces of information are stored: the first this reflects the fact that there is no privileged location where missed channel (i.e. the channel closest to the end of the first the split track phenomenon occurs. The behavior of the missed track) and the number of missed channels starting from it (i.e. channels, however, is more challenging to consider and is the number of channels between the closest to the start of shown in Figure 18. Modelling detector-specific reconstruction uncertainties in LAr-TPC — 11/13 Figure 18. Distribution of the number of lost channels, the distributions are divided according to the geometric plane of wires. Figure 19. Landau+Uniform fit on the distribution of the number of lost channels divided by geometric plane of wires. It is evident that the trend is significantly different for the IND1 planes compared to the IND2 and COLL planes. The model used to parameterize the distributions must necessarily 4.2 HARPS Tool consist of two parts: As introduced in the previous section, the HARPS module • Physical: a component that takes into account the phe- aims to modify the events by removing hits from the tracks nomena that cause the split track effect (e.g., a track based on known probability distributions. The selection of crossing the cathode). This shape is parameterized as tracks from which hits are removed is not random: the code a Landau distribution, centered at x1 = 3 for the IND1 requires a .txt file as input, which contains information about planes and centered at x1 = 1 for IND2 and COLL. As the particle IDs for which we want to identify the tracks and expected, the gaps are small and there is no signifi- remove the hits based on the aforementioned probabilities. In cant difference in the distribution between the differ- this way, the code allows for the separate study of particles ent planes, as the physical phenomenon occurs almost interacting with the detector and the quantification, by varying equally on each plane; parameters, of the effect of the split track phenomenon on the reconstruction of neutrino physics events. The plug-in • Reconstruction effect: a component that accounts for presented in the previous sections, aimed at identifying and the actual coherent noise experienced by the different characterizing split track phenomena in a sample, can there- planes. The wire planes, during the event reconstruc- fore be used in synergy with the HARPS tool: by applying tion process, undergo an initial phase where coherent the module to a simulated dataset with Monte Carlo, a dataset noise is removed; however, this removal could result with a known abundance of split tracks is generated, and by in the loss of physical signal. For this reason, it is not using the plug-in, it is possible to derive the characteristics necessary for the gap to be small, and moreover, this of the events and quantify the impact of split tracks on the noise tends to occur when a track is aligned with the reconstruction. wires. This component is parameterized by a uniform distribution, which is found to have significantly vary- The first modification made is related to the management of ing levels between the planes. In particular, the IND1 the .txt input file, where if the file is empty or non-existent, all planes have a much higher noise component than the the tracks in the sample are considered. The second modifica- IND2 and COLL planes. This is predictable consider- tion, not fully implemented and tested on a sample, concerns ing that the IND1 planes are the outermost planes and the way hits are selected for removal from a given track: the are therefore subject to noise from other events much first hit is chosen randomly, and starting from this hit, the next more noticeably than the subsequent planes. N hits are removed following a Landau distribution with a Once the data trend is parameterized, a fit is performed by uniform background. The characteristics of these two distri- summing the two distributions. The obtained result is shown butions were described in the previous section, and the code in Figure 19. At this point, all the necessary data are available must take into account that the initial parameters vary sig- to implement the hit removal with HARPS in a simulated nificantly depending on the wire plane that sees the track. sample (implemented in section 4.2): we know that each plane Additionally, and very importantly, after the random selection has a uniform probability distribution for losing a channel and of the channel and the subsequent removal of the N following that starting from it, depending on which plane loses the event, hits, a veto condition is applied to ensure that the final length N channels will be lost, following a probability distribution of the pair of segments is greater than 40cm, as previously given by the sum of a Landau and a uniform background. presented. A visual example is shown in Figure 20. Modelling detector-specific reconstruction uncertainties in LAr-TPC — 12/13 Appendix Calculation of the angle between non-parallel tracks The angle between the two tracks is obtained through three main steps: 1. Searching the points of closest approach: These points are obtained as Closest1 = endPos1 + avgDir1 · t1 Figure 20. Example of hits removal using the HARPS tool on an event, displayed with the LArSoft event viewer. Closest2 = startPos2 + avgDir2 · t2 where t1 and t2 are obtained by minimizing the distance In this figure, an example of a track recorded on the three wire between the two tracks (for the explicit calculations, planes is shown, along with the removal of a cluster of hits on see [9]). one of the planes. By subsequently applying the plug-in to the 2. Searching for midpoint M: The midpoint is calculated dataset following the removal, it is possible, in addition to fur- as ther validating its selection capabilities, to obtain information 1 M = closest1 + · closestSep regarding the artificially generated Split Track and compare 2 its characteristics with the dataset before the HARPS tool was where closestSep = closest2 − closest1 is the vector applied: the ability to select which particle to focus on and connecting the points of minimum distance. vary the probability distribution according to which tracks are removed represents a highly important tool for understanding 3. Searching for the angle: The angle is then obtained and quantifying the impact of the Split Track phenomenon on starting from the dot product of the vectors pointing neutrino event reconstruction. from the end of the first track to the midpoint and from the midpoint to the beginning of the second track, in formulas: 5. Conclusion   startToMid · endToMid In conclusion, the work conducted in this study can be sum- θ = arccos ||startToMid|| · ||endToMid|| marized as follows: • The importance of accurate event reconstruction was Acknowledgments understood, and several potential issues were high- lighted, with particular focus on the phenomenon of I would like to extend my profound thanks to the Organizing Split Tracks; Committee of the Italian Student Program at Fermilab, espe- cially Giorgio Bellettini, Simone Donati, Emanuela Barzi and • A versatile and optimized plug-in in ART was devel- Marco Mambelli for granting me the exceptional opportunity oped, which can be applied to any dataset to identify to participate. Being selected was a true honor and a defining Split Track events based on geometric characteristics moment in my academic path. and store their main observables: angle between tracks, gap, missed channels, and segment lengths; I am deeply grateful to my supervisor, Harry Hausner, for his continuous guidance and support at all hours of the day. • The geometric distribution of Split Track events in the Your invaluable advice has been instrumental in the success detector was studied, identifying the distribution fol- of this project, and your expertise has been a constant source lowed by the number of missed channels. Using this, of inspiration during challenging moments. the result was implemented into the HARPS module. I would also like to express my heartfelt thanks to Angela Possible future developments of this work include the im- Fava for her numerous suggestions and for the important plementation of additional methods for selecting split track discussions that have helped guide me towards achieving events (e.g., energy-based selections) and the validation of the the goals of the project. My sincere thanks also go to the plug-in’s performance on further runs. Moreover, a more in- ICARUS Collaboration for fostering a collaborative research depth study of the missed channel distribution, followed by its environment. implementation into the HARPS module, could pave the way for more precise quantifications and direct interventions to ad- dress the systematic issue of split tracks in the reconstruction of neutrino events in ICARUS. Modelling detector-specific reconstruction uncertainties in LAr-TPC — 13/13 References [1] P. Abratenko et Al. ICARUS at the Fermilab Short-Baseline Neutrino Program - Initial Operation. arXiv:2301.08634v1, 2023. [2] The ICARUS-WA104, LAr1-ND, MicroBooNE Collab- orations. A Proposal for a Three Detector Short-Baseline Neutrino Oscillation Program in the Fermilab Booster Neutrino Beam. arXiv:1503.01520v1, 2015. [3] R. Triozzi. Studies of the trigger performance of the ICARUS T600 detector at Fermilab. Fermi National Ac- celerator Laboratory Internship Report, 2022. [4] M. Diwan et Al. A novel method for event reconstruction in Liquid Argon Time Projection Chamber. Journal of Physics: Conference Series 762, 012033, 2016. [5] M. Artero Pons. Study of the reconstruction of ν µCC QE events from the Booster Neutrino Beam with the ICARUS detector. Philosophiae Doctoral Course Thesis in Physics (Università degli studi di Padova), 2022/2023. [6] R. A. Jaimes Campos. Cosmic Ray Tagging in ICARUS: a combined analysis. Master Thesis in Physics (Alma Mater Studiorum), 2023/2024. [7] B. Behera et Al. ICARUS TPC Channel Mapping. SBN DocDB 25069, 2022. [8] M. Scarnera. Improving track reconstruction in the ICARUS detector. Master Thesis in Physics (Alma Mater Studiorum), 2023/2024. [9] Ronald Goldman. Intersection of Two Lines in Three Space. Graphics Gems (p. 304), 1995.