Reconstruction software takes the calibrated hits and reconstructs the neu- trino interaction. In ND280 the reconstruction is made harder than for a collider detector because the position of the event vertex is unknown; indeed, it need not even be in the detector. Past neutrino experiments such as MINOS and Super-Kamiokande used monolithic detectors, which made reconstruction simpler. ND280 contains a number of subdetectors with different detection methods making the global reconstruction challenging. In ND280 a package called RECPACK [79] carries out a global likelihood fit to associate clusters and tracks in separate clusters.
oaRecon
While global reconstruction algorithms are contained within the RECPACK software, oaRecon manages the order in which the data from each subdetector is processed. Each subdetector has a dedicated reconstruction package which can be run via oaRecon. Subdetector reconstruction algorithms return objects inheriting the TReconObject base class. Ultimately a subdetector will return a TReconPID class, containing a likelihood for different particle hypotheses which is used for the global likelihood fit. oaRecon then attempts to connect together clusters from the different subdetectors to reconstruct full events. At the time of writing this was implemented and tested for the TPC and FGD. The ECal is also implemented but untested.
TPC and FGD Reconstruction
The TPC and FGD detectors share a common reconstruction algorithm. It begins with the TPC as it possesses the highest spatial resolution of any part of ND280. Reconstruction looks for hits close in time, and on neighbouring MICROMEGAS pads in the y axis, giving a position in the XY plane. Recon- struction starts here because high momentum leptons travel along the z axis. A cellular automata algorithm [64] clusters segments of tracks that lie close to each other in time as well as in position. The longest chain of connected segments is then made into a track.
Tracker particle identification carried out by the TPC is discussed in detail in Chapter 6. Here it is sufficient to note that the TPC allows an accurate measurement of the energy loss of the particle to be made, as well as its momentum. By measuring the energy loss at a known momentum the species of particle can be identified.
Once tracks in the TPCs have been identified, a Kalman filter is used to extend the tracks into the FGDs. Tracks are propagated into a FGD one hit at a time, updating the position, direction, length and curvature of the track after each step. The Kalman filter is implemented in the RECPACK software. RECPACK is also capable of connecting together tracks that have passed through multiple TPC and FGD modules. The FGDs also have a standalone cellular automata based clustering algorithm. This is because some tracks will travel in or close to the XY plane and so not pass into a TPC.
ECal Reconstruction
Particle Identification in the ECal will be discussed extensively in Chapter 4. This section describes the structure of the package. The ecalRecon package runs a series of algorithms in a fixed order, each adding a layer of reconstruc- tion before storing reconstructed clusters. A cluster is a set of hits that the reconstruction treats as being from the same particle. Attached to each cluster is an estimate of the energy and kinematic properties of the particle, as well as an identification of particle type. Before reconstruction begins, the hits from an event are ‘prepared’ by performing basic checks, such as a cut on charge deposit to remove detector noise. Hits passing this cut are clustered based on which ECal module the hit was found in and the time bucket the cluster was found in.
Charge released by the photosensors is the basic quantity measured by the ECal electronics. It is this charge that is digitised by the ADCs. Reconstruc- tion however uses MEUs (MIP Equivalent Unit), a unit proportional to the energy of the particle. The MIP energy is defined as the energy deposited by a muon passing at normal incidence through 1 cm of scintillator. Currently the calibration to convert from a charge deposit to a number of MIPs is carried out in two stages. The first is to convert the measured charge in femtocoulombs to the number of photons on the face of the photosensors. The second recon- structs the position of the hit and uses that position to calculate how many photons were created by the original energy deposit.
A clustering algorithm is applied to associate hits due to the same particle. This step is split into four steps and is described fully in [64]. Basic clustering is the first level of reconstruction, using a nearest neighbour algorithm to separate hits into clusters representing individual particles. The second stage
of clustering, ‘combi clustering’, seeks to combine clusters split by the effects of layer efficiency or a gamma not converting straight away. A final stage of clustering expands the existing clusters by determining whether any of the un-clustered hits are on the periphery of a cluster. When the hits are grouped together, they are placed in separate lists, one for each view of the detector. For example, hits in the DsECal are arranged into two lists; one for hits in the XZ plane and another for hits in the YZ plane. After clustering, a set of clusters exists for each view of a given ECal module. The next step, matching, associates clusters in each view due to the same particle.
Once all the hits due to a common particle have been grouped together, then the energy of the particle can be reconstructed. Particles showering electro- magnetically in the ECal, primarily electrons and gammas, currently have their energy measured by the ECal reconstruction software. The energy re- construction algorithm is trained using Monte Carlo gamma rays passing into the ECal at a range of energies. Gammas are used instead of electrons to reduce the effects of pre-showering in the electronics and metalwork surround- ing the ECal. Variables used in the training are the distributions of the mean, sigma and skew of the energy deposit, all of which are fit to a skewed Gaussian at each energy used in the training. A maximum likelihood fit is then used to find the most likely energy of the particle.
The energy of muons is measured by observing how far they travel before being ranged out either in the ECal or the SMRD. Low energy muons, up to about 250 MeV, will stop in the ECal. Higher energy muons will range out in the SMRD. The highest energy muons, of a few GeV or more, can escape the detector completely; although for T2K this is in the high energy tail of the neutrino energy spectrum. The visible energy of muons and other track
like particles passing through the ECal can be measured, providing a very useful measurement for calibrating the absolute energy scale in the detector. Hadronic showers also occur in the ECal, due mostly to pions. Currently there is no hadronic shower fitter in the ECal reconstruction, as hadronic showers are not as well understood as electromagnetic showers. On top of this, a hadronic shower loses a large fraction of its energy in neutrons which the ECal is not designed to detect.
Particle identification (PID) is the final stage of the ECal reconstruction. De- velopment of the PID algorithm is the subject of Chapter 4. The ECal recon- struction uses a mixture of topological features and charged based quantities to decide what type of particle has caused a cluster. Each cluster is fit as both a track and a shower. Track fitting currently fits a track as a straight line using a chi-squared fit. Shower fitting uses PCA (Principal Component Analysis), as discussed in Section 4, to find a direction, centre and opening angle for the shower. Both track and shower fitters find start and ends points for a cluster. As with the clustering, the PID assumes that the starting point of a track or shower is the end on the inside of the detector. Outputs from the reconstruction are used as the discriminating variables in the PID and are discussed at length in the Chapter 4.