Electronics Testing, LArSoft Analysis, and Data Acquisition for MicroBooNE

Victor Genty∗

Columbia University, Nevis Laboratories, Irvington, New York†

(Dated: August 2, 2013)

This paper presents a variety of projects completed for the neutrino group at Nevis Labs.MicroBooNE electronics are tested including gain of phototubes, and a study of splitter reflectionbetween the phototube, signal shaper. LArSoft analysis is presented for cluster reconstructionalgorithms on whole detector simulation. Two clustering algorithms are compared. Visible energyreconstruction is also studied. Data acquisition software and hardware for photomultipliers arecalibrated for the MicroBooNE optical system.

I. INTRODUCTION

A. The Standard Model

The Standard Model of elementary particles and theirinteractions is a quantum field theory of the electromag-netic, weak, and strong nuclear forces. Certified by itsremarkable success at describing a diverse body of ex-perimental observation, the standard model is one of themost successful theories in physics today. According tothis model, all of matter is constructed of 12 spin-1/2particles known as fermions: six leptons and six quarks[1].

Quarks carry fractional charges + 23 and − 1

3 , andare grouped according to flavor: ‘up’, ‘down’, ‘charm’,‘strange’, ‘top’, and ‘bottom’. Participating the electro-magnetic, weak, strong interactions, quarks are boundtogether forming hadronic matter: mesons and baryons.Quarks only exist in bound states, and therefore are notdirectly observable.

Increasing in mass to the right, Leptons are integercharged and interact via the electromagnetic and weakforces. The most familiar particle is the electron, e−.The two heavier leptons are called the muon, µ, and tau,τ . The particles in the bottom lepton row are called neu-trinos and have unknown, but small non-zero mass.

Each force in the Standard Model has associated withit a set of spin 1 carrier particles. The electromagnetic

Particle Flavor Q/e

leptons e µ τ -1

νe νµ ντ 0

quarks u c t + 23

d s b − 13

TABLE I: The fundamental fermions of the Standard Model:particle names, flavor, and ratio of electric to fundamentalcharge.

∗Electronic address: vic.genty@gmail.com†Permanent address: The University of Texas at Austin, Depart-ment of Physics, Austin, Texas

force is mediated by exchange of photons, γ, the weakmediated by W± and Z particles, and the strong inter-action by a set of 8 particles called gluons, g.

The Higgs mechanism is responsible for giving massto the fundamental fermions through spontaneous sym-metry breaking [2]. The Higgs particle is a scalar bosonwith neither color nor electric charge.

Through a description of 12 fundamental particles andthree forces, the Standard Model predicts the richnessof particle dynamics, a multitude of observed compositeparticles, anti-particles, and much more [3].

B. Neutrino Oscillations

In the Standard Model neutrinos are assumed masslessand only interact via the weak force. Massive neutrinoswere first studied by Italian physicist Bruno Pontecorvoin the 1950s as an analogy to neutral kaon mixing [4].According to the theory, neutrinos are created and an-nihilated as flavor eigenstates: νe, νµ, ντ but propagatethrough space in so called mass eigenstates: ν1, ν2, ν3.Each flavor eigenstate can be expressed as a linear com-bination of mass eigenstates,

|να〉 =∑i

Uαi|νi〉,

according to a unitary 3× 3 mixing matrix,

U =

Ue1 Ue2 Ue3Uµ1 Uµ2 Uµ3Uτ1 Uτ2 Uτ3

,

for |να〉 a neutrino of definite flavor with α ∈ [e, µ, τ ]and |νi〉 a neutrino of definite mass with i ∈ [1, 2, 3].Uαi is the so called Pontecorvo-Make-Nakagawa-Sakata(PMNS) matrix analogous to the Cabbibo-Kobayashi-Masakawa (CKM) matrix of quark mixing. The matrixcan be decomposed into four mixing matrices betweenthe eigenstates:

U =

1 0 0

0 c23 s230 −s23 c23

c13 0 s13e

−iδ

0 1 0

−s13eiδ 0 c13

c12 s12 0

−s12 c12 0

0 0 1

,

2

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 5 10 15 20 25 30 35

Pro

babilit

y

L/E (km/GeV)×103

Electron Neutrino Oscillation

νeνµντ

FIG. 1: Vacuum oscillation probability for an initial electronflavor state. The PDG values sin2 2θ12 = 0.857, sin2 2θ23 =0.95, sin2 2θ13 = 0.095, ∆m2

32 = 2.32 × 10−3eV2, ∆m221 =

0.75× 10−5eV2, and δ = 0 are used [4].

where cij = cos θij and sij = sin θij . The angles θijparameterize the relationship between flavor and masseigenstates. The phase δ is only non zero, i.e. U∗ = U , ifneutrinos violate charge-conjugate and parity (CP) sym-metry. Assuming massive neutrino propagating planewaves of the form,

|νi(t)〉 = e−i(Eit−p·r)|νi(0)〉,

for the energy of a mass eigenstate Ei =√|p|2 +m2

i , ttime, p the 3-momentum, and r the neutrino position,the oscillation probability between two flavor states αand β is their inner product in time,

P (να → νβ) = |〈νβ |να〉|2 =

∣∣∣∣∣∑i

U∗αiUβie

−im2iL/2E

∣∣∣∣∣2

,

assuming ultra relativistic neutrinos. In the two neutrinomodel described by a 2× 2 mixing matrix the oscillationprobability simplifies to

P (να → νβ) = sin2 2θ sin2(1.27∆m2L/E).

For L the distance from the flavor eigenstate creation andE its energy. The above equation demonstrates that neu-trino flavors oscillate in time, or distance traveled, andare dependent on mixing angle, energy, and their squaredmass difference. It is the job of experimentalist to mea-sure these parameters.

In FIG 1 shown above, an electron neutrino oscilla-tion probability to muon and tau flavor neutrinos variesas a function of the ratio of propagation distance to en-ergy. The presence of three neutrino mixing producessmall oscillations overlayed on a larger periodic behav-ior. Experimentally measured parameters are used fromthe Particle Data Group (PDG) assuming normal masshierarchy.

Neutrino oscillations are a precise method for study-ing quantum coherence between weak flavor states. Un-fortunately, oscillation experiments are only sensitive todifferences in neutrino masses and not absolute mass. Inaddition, the mass hierarchy of neutrino mass states arecurrently unknown.

C. Neutrino Detection

Neutrinos and their oscillation parameters are the fo-cus of an intense experimental interest in recent years.Due to the nature of the weak interaction, neutrino de-tection is difficult. Experiments must receive and collecta large flux of neutrinos to detect only a few number ofevents. Neutrinos are typically detected through processsuch as inverse beta decay, or charged and neutral cur-rent interactions with hadrons. Modern experiments usea combination of scintillation or cherenkov radiation andcalorimetry coupled to photomultiplier tubes to detectand track neutrino interactions. Many experiments uti-lize novel techniques such as time projection chambers.Experimental detection of neutrinos can be grouped intofour categories: solar, atmospheric, reactor, and beamneutrino detection.

The oldest and first experimental detection of neutrinooscillations came from evidence of deficits in neutrinorates produced in fusion products from the Sun. TheHomestake experiment in Lead, South Dakota measuredonly one-third of theoretical calculations for the electronanti-neutrino flux from the Sun [6]. The experimentersused the process of inverse beta decay between chlorineand argon to count the solar neutrino production rate. Amore modern experiment called SNO (Sudbury NeutrinoObservatory) carried 1,000 tonnes of deuterium oxide ina spherical vessel. SNO measured charged and neutralcurrent interactions [7] for electron neutrino interactions.Solar neutrino experiments are particularly sensitive toθ12 and ∆m2

12 oscillation parameters.Atmospheric oscillation experiments measure neutri-

nos and anti-neutrinos produced in high energy cosmicray decays in Earth’s upper atmosphere. Neutrinos cre-ated from charged pion and muon decay products arestudied. Precise experimental evidence for atmosphericneutrino oscillation was first discovered in 1998 by theSuperKamiokande experiment which used a giant waterdetector to measure atmospheric neutrinos. Atmosphericdetection experiments are sensitive to θ23 and ∆m2

23 pa-rameters.

Reactor experiments utilize anti-neutrinos produced infission reactors and measure their disappearance over dis-tance. Nearly all reactor experiments including DoubleChooz, Daya Bay, and Reno, implement an identical nearand far detector to count neutrino disappearance. Theexperiments involve liquid scintillator and photomulti-plier tubes surrounded by a veto region to measure in-verse beta decay events. Typically, the liquid scintillatoris doped with a rare-earth metal to en chance neutron

3

capture rates. Rector experiments are sensitive to θ13and currently have set the best limits on the value.

Lastly, neutrino beam experiments employ particle ac-celerators to create a defined neutrino flux and mea-sure neutrino appearance. Experiments such as MINOS,T2K, NOνA are a few of the long-baseline schemes to de-tect neutrinos produced when high energy proton beamsimpinge on a target. The pions and kaons producedfrom strong interactions with the target decay in flightto muon and electron neutrinos. It is the goal of beamexperiments to measure θ23, and ∆m2

23 values.Each neutrino source and detection method is sensi-

tive to the different neutrino oscillation parameters inthe PMNS matrix. Other notable experiments such asNEMO-3 measuring neutrinoless double beta decay, andIceCube measuring high energy astophysical neutrinos,provide important insight into less studied area of con-temporary neutrino physics.

II. MICROBOONE

A. History

In 1995 and 1996, the Liquid Scintillator NeutrinoDetector (LSND) experiment at Los Alamos reportedevidence for νµ → νe oscillations. A low energy excess,∼ 3.8σ [8], of νe events in a pure νµ beam was found.

The allowed values of ∆m2 and sin2 2θ measured fromthe experiment were in conflict with standard modeloscillation theory. High ∆m2 values were not excludedfrom the data analysis. The LSND result could also notbe explained by atmospheric or solar oscillations as theyrequired too low of ∆m2 values. The LSND experimentsuggested the presence of a fourth neutrino with a largemass splitting compared to the three known flavors.One method to accommodate additional mass splittingsin the neutrino sector is the introduction of so called“sterile neutrinos” which act as new flavors that mixwith the known Standard Model flavors. To avoid thelimits on the active neutrinos, sterile neutrinos wouldnot couple to the W or Z bosons [5]. The MiniBooNEexperiment was the first attempt to verify LSND’sobservation. The MiniBooNE detector is an oil-basedcherenkov radiation detector measuring neutrino oscil-lation from the Fermilab Booster beam. The FermilabBooster is a 500 foot diameter, 8 GeV circular protonaccelerator. MiniBooNE found an excess of not only νeas in LSND, but in νe appearance as well from νµ andνµ beams. In addition, these results were measured atan even lower energy that from LSND. Unfortunatelythe evidence was not conclusive enough to declare theexistence of sterile neutrinos but was able to rule out thetwo neutrino oscillation model [9]. The excess could havebeen oscillations, photon radiation, or ν background. Toreconcile the observations of LSND and MiniBooNE, theMicoBooNE experiment was proposed.

FIG. 2: MicroBooNE’s liquid argon time projection chambersurrounded by the 170 ton liquid argon cryostat.

B. Motivation and Goals

The MicroBooNE experiment, located at Fermilab, isa beam neutrino experiment located along the Boosterneutrino beam line. Motivated by the LSND and Mi-croBooNE low energy excess, MicroBoone will search forsterile neutrinos and hopefully elucidate the previous ex-periments results. What sets MicroBooNE apart fromprevious experiments is its high efficiency at reconstruct-ing neutrino interactions, begin nearly free of particlemisidentification background [9]. MicroBooNE improveson the MiniBooNE detector by being able to separatephoton and electron signals in the time project camber.This background was a significant source of uncertaintyfor the MiniBooNE device.

One of MicroBooNE’s secondary goals is be sensitiveto cosmic neutrino flux if a Super Nova event were to oc-cur. The detector will be operating continuously aroundthe clock and along will real time astronomical data willbe able to trigger data read out to possibly collect neu-trinos from such an event.

MicroBooNE will also serve as a physics research anddevelopment for future long baseline neutrino experi-ments such as the Long Baseline Neutrino Experiment(LBNE) which are projected to definitively test for CPviolation in the neutrino sector.

C. Detector

MicroBooNE is currently being installed at a distanceof 469 meters from the beryllium target at Fermilab.Shown in FIG. 2, MicroBooNEs 70 ton fiducial volume iswithin a 2.5×2.4×10.4 meter3 drift region of active liquidargon volume. The total mass of Argon in the detectoris 170 tons. One of the most important aspects of thedetector is the Liquid Argon Time Projection Chamber.

MicroBooNE’s detection apparatus is a Liquid ArgonTime Projection Chamber (TPC). TPC’s are one of themost promising technologies for future large scale neu-

4

FIG. 3: Orientation of wire planes in MicroBooNE TPCchamber. The U (green), and V (red) are induction planesoriented at +60 and −60 with respect to the vertical, andY (blue) is the collection plane. Electrons from ionization arecollected on the collection plane only.

trino detectors [9]. The Booster proton beam will im-pinge on a Beryllium target and the resulting chargedmesons are focused into a decay pipe. In the decay pipes,kaons and pions will decay and an absorber will filter thefinal states for neutrinos. The neutrinos pass through aregion of dirt and will enter into the liquid argon chamberand interact via charged and neutral current interactions,ionizing Argon atoms. An electric field of 500 V/cm pro-vided by a negatively charged cathode plate will drift ion-ization electrons onto three recording planes at a speedof 1.6 mm/µs. The recording planes are shown in FIG.3. The electrons will drift a maximum distance of 2.5meters past the negatively biased U, and V wire planesincluding voltage pulses. The wireplanes are separatedby 3 mm. The electrons will be collected on the collec-tion wires. Together with photomultiplier tubes the threewire planes, separated by 3 mm, will provided enhanced3 dimensional reconstruction ability for the fiducial vol-ume. The wire locations and electron arrival time on thewires enables sub-millimeter position resolution. In ad-dition to the TPC, 30 photomultiplier tubes will be usedto determine event timing from argon scintillation at 128nm. Detector electronics will be able to select events intime with the Booster neutrino beam passage.

III. ELECTRONICS TESTING

A. PMT Gain Study

Photomultiplier tubes (PMTs) are electron tube de-vices which convert light into an electric current throughthe photoelectric effect. The PMTs to be used in theMicroBooNE detector are 8-inch Hamamatsu R5912-02mod. Photomultipliers in MicroBooNE will provideimportant event timing information for triggering on neu-trino interactions, in addition to aiding in the separationof electron and photon signals. A secondary objectiveof the PMTs is to calculate energy deposition from pho-tons produced in Argon scintillation. Photomultiplierswill also aid in distinguishing neutrinos from the cosmicray background. This summer, a 8-inch R5912 PMT wastested. The difference between the Nevis and the Micro-

BooNE PMT is a lack of a UV wavelength shifter to shiftliquid argon scintillation light into the observable spec-trum. The Nevis PMT is also able to transmit signaland high voltage (HV) down separate cables. The R5912PMT gain was studied.

The overall amplification factor, or gain, depends onthe number of dynodes, or electron accelerating stages,inside the phototube. The gain, G, of a PMT is defined asthe ratio of secondary,Ns to primary,Np electrons ejectedfrom the final dynode and photocathode repectively,

G =NsNp

.

A common way to calculate these parameters is to usea light emitting diode to pulse the PMT and record thevoltage signal it produces on an oscilloscope. The R5912was placed into a light tight box together with a bluelight emitting diode (LED). The LED was pulsed at afrequency of 100 Hz from a digital BNC 565 pulser with astep DC voltage at a width of 30 nanoseconds. The volt-age was stepped between 1.3 and 1.7 volts. Two meth-ods to calculate gain were used, variation in peak height,and width. The LED photon emission is assumed to bepoissonian. Under this assumption, the number of pho-toelectrons produced will follow a poisson distribution.The variation in peak heights method is presented here.The PMT gain and number of photoelectrons was mea-sured as a function of input voltage and time (at constantvoltage).

To calculate the number of primary photoelectronsproduced at the cathode, Np, we measured the meanvariation in peak height over one minute, or N =6000counts, on an oscilloscope. The average peak height, µh,and standard deviation,σh, are related to the number ofprimary photoelectrons by,

µh = CGNp,

σh = CG√Np,

for C some constant. The ratio of mean to standarddeviation gives,

Np =

(µhσh

)2

,

with uncertainties,

µh ±σ√N,

σh ±σh√2N

,

Np ±2µh√Nσh

√1 +

µ2h

2σ2h

.

To calculate the number of secondary electrons, the PMTsignal was integrated over 6000 triggers yielding the

5

0

1

2

3

4

5

6

7

1100 1200 1300 1400 1500 1600 1700 1800

Gain×

107

Voltage (V)

Gain vs. Voltage

(100 mV, 20 ns)/DIV(200 mV, 20 ns)/DIV(300 mV, 20 ns)/DIV

FIG. 4: PMT gain as a function of input voltage at differentoscilloscope divisions. Deviation in gain value is observed atconstant voltage.

charge deposited on the anode. Specifically,

Ns =

∫V dt

eR,

with uncertainty,

Ns ±∆(∫V dt

)eR

,

for e the electron charge and R = 50 Ω the termination inthe oscilloscope. The mean integrated voltage value andstandard deviation were used for the charge calculation.The gain is then,

G =

∫V dt

eR

(σhµh

)2

,

with uncertainty,

G± NsNp

√(∆NsNs

)2

+

(∆NpNp

)2

.

Gain for the R5912 PMT is shown in FIG. 4 plotted atthree different oscilloscope window ranges. Only statis-tical uncertainties are shown. We noticed a deviationin gain value at constant voltage over different oscillo-scope ranges. This has been observed before by bothGeorgia and Leslie. One possible cause of this effect isthat upon changing the number of voltage divisions, thescope’s precision increases and decreases. Large devia-tion in gain values from the specification sheet are ob-served after 1500 volts [13]. After this point the gaindrops off suddenly for certain oscilloscope settings. An-other method to calculate the PMT gain is necessary toilluminate this issue. FIG. 5 shows the number of pri-mary electron varying significantly as a function of inputvoltage. At values starting around 1500 V, significantdeviations in the photoelectron count are observed, the

0

20

40

60

80

100

1100 1200 1300 1400 1500 1600 1700 1800

Np

Voltage (V)

Primary Electrons vs. Voltage

(100 mV, 20 ns)/DIV(200 mV, 20 ns)/DIV(300 mV, 20 ns)/DIV

FIG. 5: The number of primary photoelectrons from the cath-ode as a function of input voltage. We expect the number ofelectrons liberated from the photocathode to be constant as-suming the LED identically pulses the PMT with the samephoton energy profile.

0

1

2

3

4

5

6

7

0 5 10 15 20 25 30 35 40 45

Gain×

107

Time (min)

Gain vs. Time

FIG. 6: PMT gain varies in time at constant 1500 V. This isattributed to both a fluctuation between 30 to 50 photoelec-trons and 600-800 mV output voltage over 40 minutes.

same value at which the gain deviates from expected. Wewould expect the same photoelectron count at each volt-age. In order to understand the behavior of the PMTas a function of time we measured the output voltage,number of primary electrons and the gain for an inputvoltage of 1500 V every 5 minutes for 40 minutes. FIG. 6shows how the gain varies at constant voltage. The gaindoes not significantly deviate over time. A more accurateassessment of the time dependence of gain would requirelonger periods of operation. Finally, we incorporated allmeasurements made in one week from different oscillo-scope settings to estimate the systematic uncertainties.In FIG. 7, red triangle data points show all measurementsmade. Blue points represent the average gain for each in-put voltage. The gain value exhibits large deviation inthe high input voltages of 1600 and 1700 V. The system-atic uncertainties in the gain dominated the statistical

6

0

1

2

3

4

5

6

7

1100 1200 1300 1400 1500 1600 1700 1800

Gain×

107

Voltage (V)

Gain vs. Voltage

MeasurementAverage

FIG. 7: PMT gain as a function of input voltage with allmeasurements taken over a period of a week included. Theblue square represents an average measurement at constantvoltage. The red triangle is a measurement. The systematicuncertainties in the measurement dominated at high voltage.

uncertainty at high input voltages. Further tests at highvoltage are necessary.

While this specific photomultipler tube will not be usedin the MicroBooNE detector, it will be used to test thedata acquisition system for MicroBooNE at Nevis. Aknowledge of the PMT response over input voltage, andtime are essential in calibrating the electronics hardware.

B. Splitter Reflection

A current effort to test MicroBooNE’s optical electron-ics setup is called Bo. Bo is a liquid argon training groundfor the MicroBooNE optical system including cryogenicphotomultipliers, high voltage system, cables and split-ters, read out electronics and much more [16]. In Bo,an issue has arisen in the phototube splitters which aresimilar to those to be used in MicroBooNE.

The output signal from a single PMT in MicroBooNEwill have to exit on the same cable that transports highvoltage to the cathode. Isolating the PMT signal fromthe high voltage requires placing a blocking capacitor, orsplitter, along the cable which connects the PMT to theshaper board. A portion of the signal that reaches thesplitter will be reflected back along the cable. These re-flections cause secondary voltage peaks to be produced inthe splitter output, with time-delays relative to the orig-inal signal on the order of the signal travel-time acrossthe wire. This secondary interference is analyzed by theshaper along with the original signal and can cause dis-tortions in the the final output. A clean shaper signal isimportant to digitization by the front end module analogto digital converters. Reflection is observed in the shaperoutput if one of the two criterion are satisfied.

1. The secondary pulses due to reflections have largeenough amplitude. The amplitude of the reflections

RC1 C2

LVin Vout

FIG. 8: Circuit used at Nevis to test the Bo splitter capaci-tance. Vin is a step function from the pulser, Vout is the signaloutput to the shaper. The values of C1, C2, and L are varied.R = 50 Ω

depends on the value of the capacitance acting asa splitter. Larger capacitors will allow a smallerfraction of the signal from reflecting.

2. The frequency at which these reflections occur islow. This is dependent on the length of the cable.The shaper integrates over a wide enough time win-dow such that fast oscillations in the splitter outputare smoothed out and do not influence the shapedsignal.

Bill Sippach, an electronics engineer at Nevis Labs,simulated the splitter in SPICE, an analog electronic cir-cuit simulator. He found that for a R = 50 Ω impedancecoaxial cable in series with the blocking capacitor C the’ringing’ should not occur in the shaper output as longas the time constant of the circuit is much larger thanthe travel time of the signal down the wire,

τ = RC τtravel =L

vsignal,

for L the length of the cable and vsignal = 1 foot/1.5 nsthe signal speed. Bill’s simulation showed that the timeconstant of the cable and capacitor should be roughly 3-5times larger. We simulated the PMT signal output andsplitter at Nevis using a simple circuit shown in FIG.8. The first portion of the circuit simulates the PMToutput, a few tens of nanosecond exponentially decayingsignal. The signal is fed by 100 µs-wide step potentialfrom a pulser box. The second capacitor, C2 acts as thesplitter. The length of cable connecting the simulatedPMT signal and the splitter can be varied. The outputis then fed to the signal shaper.

We tested the shaper output of this circuit for differ-ent cable lengths and splitter capacitances. Our initialtest was for a short cable L = 4 m, and long cable L = 20m, C1 =5 pF, C2 = 1 nF shown in FIG. 9. The short

7

-505

1015202530

-0.5 0 0.5 1 1.5

Volt

age

(mV

)

Time (µs)

Short Cable - Shaper

-505

1015202530

-0.5 0 0.5 1 1.5

Volt

age

(mV

)

Time (µs)

Long Cable - Shaper

FIG. 9: Shaper output for short cable of length L = 4 m,C2 = 1 nF and for long cable of L = 20 m. Significant ringingis observed in the shaper output of the long cable. The shortand long cable input to the shaper resembles the left handcolumn in FIG. 10. In this test, the reflection in the longcable is 10 times larger in amplitude than in FIG. 10.

0

10

20

30

40

-0.5 0 0.5 1 1.5

Volt

age

(mV

)

Time (×100ns)

Short Cable - Scope - 10 nF

0

10

20

30

40

-0.5 0 0.5 1 1.5

Volt

age

(mV

)

Time (µs)

Short Cable - Shaper - 10 nF

-505

1015202530

-0.5 0 0.5 1 1.5 2 2.5

Volt

age

(mV

)

Time (×100ns)

Long Cable - Scope - 10 nF

0

10

20

30

40

-0.5 0 0.5 1 1.5

Volt

age

(mV

)

Time (µs)

Long Cable - Shaper - 10 nF

FIG. 10: The left column depicts raw shaper input from thesplitter. The right column is shaper output. The splitter ca-pacitance is increased to C2 = 10 nF. No ringing is observedin the long cable despite the low frequency reflection at ∼ 200ns. This is because of a decrease in amplitude from previousFIG. 9. The short cable is presented as an example raw split-ter output on the oscilloscope before shaping. The shapertakes this signal as input.

cable sees lots of reflection but because of the shaperslong integration time the shaper output is clean and noringing is observed. The long cable shows ringing in theshaper output, the dip in voltage after the rise, becausefor this circuit has time constantτ = 50 ns and the sig-nals τtravel = 100 ns. The ringing comes much later intime with respect to the short cable which is amplifiedby the shaper.

We then replaced the splitter capacitor with C2 = 10nF. The effect is shown in FIG. 10. The ringing in theoscilloscope output decreased in amplitude and was nolonger found in the shaper output as expected by Bill’ssimulations since τ = 500 ns. We then replaced thelength of the cable with a the same length and electronicspecifications as the Bo cable. The length increased fromL=20 m to L=30m. The results are shown in FIG. 11.We saw no ringing in the shaper output for the long ca-ble since τtravel = 150 ns is still much less than τ = 500ns. With these specifications we expected to see ringing

-505

1015202530

-2 -1 0 1 2 3 4

Volt

age

(mV

)

Time (µs)

Long Cable - Scope - 10 nF

-505

1015202530

-0.4 -0.2 0 0.2 0.4 0.6 0.8 1

Volt

age

(mV

)

Time (µs)

Long Cable - Shaper - L=30 m

FIG. 11: The length of cable was increased to the same lengthused in Bo, L=30. Reflection is seen in the oscilloscope outputat the 300 ns mark. No ringing is produced in the shaperoutput as expected.

PMT

+HV in

Anode x1

Anode x0.1

10nF (2kV) 10k 500

room temperature

10nF 16k (2kV)

GND

10M 10k

GND

450

10nF| |

FIG. 12: Circuit schematic for the Bo splitter. 10 nF ca-pacitors channel the high voltage input to the PMT and thecentral capacitor allows signal to pass through

in the shaper output as the Bo splitter was simulated asbest as possible.

In situations in which the time-constant for the cir-cuit is several times larger than that for reflections inthe wire, ringing does not present itself. Ringing in theshaper does not appear even if reflections in the split-ter output are still present (as seen in FIG. 10, where asmall reflection signal can be seen). However, our testfor a splitting capacitance of 10 nF and a cable lengthof 30m which shows no ringing does not agree with themeasurements from Bo. For the same parameters, ring-ing is observed.

We then received the Bo splitter from Fermilab. Thecircuit diagram is shown in FIG. 12. Upon inspectionwith Bill we deduced that the capacitor which acts as thesplitter in this circuit is the one connecting the PMT in-put/output directly to the “Anode x1” output. The split-ting capacitance is a 10 nF capacitor creating a 500 nstime-constant for RC (as mentioned above). This wouldbe sufficiently larger than the 150 ns travel-time for thesignal in the long wire, and would mean no ringing in theshaper output. There are however two reasons why thisdescription is not correct.

1. The capacitor in the top left of the circuit diagram,connecting the ground of the output anodes and theground of the PMT can be thought of as in series

8

16K

R2

10nF3KV

C1

P3LEMO

P4LEMO

P1

SHV

P2

SHV

10K

R1

500R3

10nF3KV

C2

0

R4

0

0

R8

450R5

openR9

10KR7

0.1uF75V

C3

1GR61G

X1 Signal

x0.1 Signal

HV Input

PMT

ShaperGnd

HVGnd

Det

ecto

rGnd

Shield

FIG. 13: PMT high voltage splitter for the MicroBooNE de-tector.

with the main splitter capacitor. This means thatoverall we have a reduction in effective capacitance.Because the two capacitors are identical the overallcapacitance goes down by a factor of two.

2. From the specification sheet, the 10nF capacitor israted at 3kV (even though the diagram indicates2kV). When the applied voltage is at the same or-der of magnitude as that for which the capacitoris rated, the effective capacitance decreases. Thishappens because the high voltage biases the dielec-tric in the capacitor, reducing the capacitance. Thecapacitors used in the Bo splitter are Z5U and at anapplied voltage of 1.4 kV, the same voltage whichused for the PMT. For a 3 kV rated capacitor thisinduces a further reduction of the capacitance ofaround ∼ 12%. One can also find the more detailedspec sheets for the individual capacitors.

FIG. 13 shows a diagram for the MicroBooNE splittersthat are now being manufactured. The overall layout ofthe circuit is identical to that of Bo. First the secondcapacitor, C2 = 10 nF can be thought of as in serieswith the splitting capacitor now at C3 = 100 nF. Thismakes the equivalent capacitance lower. Furthermore,even though the splitting capacitor is rated for 3 kV andwill have a similar voltage of 1.4 kV applied to it, itis a different type of capacitor: Z7R [14], which losesless capacitance as high voltage is applied. The drop incapacitance is only expected to be roughly ∼ 5%. Witha crude estimation, we can predict the 10 nF capacitorto result in an actual capacitance of 10 nF → 9.5 nF →Ceq = 1/(1/100+1/9.5)) = 8.67 nF. This corresponds toa time-constant of 433 ns. This is not quite 3 times asmuch as the signal travel-time in the wire. Ideally, BillSippach mentioned that the RC time-constant should beat least 3, and preferably 5 times larger than that ofthe cable. Currently the MicroBooNE splitter still seesreflections. The splitter capacitance in the MicroBooNEcircuit is to be doubled in value according to [15] as aresult of these tests.

IV. LARSOFT ANALYSIS

A. LArSoft Framework

LArSoft, or Liquid Argon Software, is an event genera-tion, simulation, and reconstruction package for liquid ar-gon based particle detectors built on the ART framework.The ART event processing framework coordinates eventprocessing via modules such analysis and event recon-struction, separating the algorithms from raw data. To-gether with LArSoft specific GEANT4, GEometry ANdTracking, simulation code, LArSoft is a playground totest the liquid argon time projection chamber based onmonte-carlo particle interactions. The essential stepsin the monte-carloing of MicroBooNE are event gener-ation, whole detector simulation; particle simulation, re-construction, identification; and event reconstruction.

An important component of LArSoft is GENIE, Gener-ates Events for Neutrino Interaction Experiments. GE-NIE takes neutrino information such as 4-momentum,flavor, and interaction-medium and simulates neutrinointeractions in all parts of a detector based on geom-etry. GENIE is the primary neutrino-nucleus interac-tion simulator which is then passed on to GEANT4 tobe propagated. With “flux-files” the user can control awealth of initial neutrino physics information, mirroringthe current neutrino beam specifications such as the Fer-milab Booster. The MicroBooNE detector is simulatedin GEANT4, a physics package to simulate the passageof particles through matter. GEANT4 models the largescale detector objects such as the cryostat, TPC, opticalelectronics and their response to particle interactions inthe liquid argon. Particle simulation is also processedby GEANT4, including drifting ionization charge to theTPC wireplanes and liquid argon scintillation capturedby photomultiplier tubes

After a simulated particle has interacted inside Mi-croBooNE, a simulated response on the wireplanes andphotomultipliers occurs. From the simulated detector re-sponse, a reconstruction chain is implemented to identifyimportant event information. Information such as en-ergy deposition, particle track length, vertex, and finalstate topology are used to reconstruct the true particleinformation such as particle ID, track, and energy. Thisinformation is critical to ultimately reconstructing neu-trino interactions in MicroBooNE. To execute particlereconstruction, a chain of algorithms has been devised[16]. The first step in the chain is calibrating raw datasuch as signals on the TPC wire planes as a function oftime. The signal vs. time information is reconstructedabove a certain threshold indicating true energy deposi-tion has occurred. This information is called a hit. Next,a cluster finding algorithm groups the hits which as re-lated to one another spatially and temporally. So far thereconstructed information is only two dimensional. Twodimensional reconstruction is unable to determine trackinformation including distance of interaction vertex andtrack from a wire. This information is critical to energy

9

Fuzzy

DBSCAN

FIG. 14: Example of clustering algorithm on single electronevents generated in the center of the MicroBooNE detector.The top panel is Fuzzy cluster and the bottom DBSCAN.DBSCAN contains many small clusters, all different colors,despite the hits being generated from the same electron. Afew of the clusters are circled.

reconstruction as electron recombination between ioniza-tion location and wire plane, hindering energy depositionaccording to Birk [11]. Three dimensional reconstructionis carried out by track finding algorithms. These algo-rithms utilizing timing and hit locations on TPC wiresover the three views to build a track. The alternate caseis shower reconstruction which analyzes γ → e+e− elec-tromagnetic showers resulting from pair producing pho-tons. Once the tracks or showers have been identified avertex finding algorithm locates tracks originating at acommon location. The complete three and two dimen-sional information is stored in an event module. Fromreconstructed final state topologies, information aboutincoming particles can be reconstructed.

The philosophy behind LArSoft is that code it to beshared between all experiments and written for a broadliquid argon community. The code is shared between anumber of experiments such as ArgoNueT, and LBNE.LArSoft is an important tool to study particle interac-tions with liquid argon before the experiment is com-pleted.

B. Cluster Reconstruction

A signal on a wire can be reconstructed typicallyusing a gaussian fitting algorithm called FFTHit orGausHitFinder. The hit information can be used by a va-riety of cluster finding algorithms to group them accord-ing to some specified criterion. Two mainstream cluster-ing algorithms are DBSCAN and Fuzzy clustering. Thissummer I studied current clustering algorithms against aspecific final state of 1 electron and 1 proton topologiesfrom a charged current interaction inside the detector. Icalibrated the data against single electron showers. Theevent displays for electron and neutrino interactions areshown in FIGs. 14 and 15.

DBSCAN is a density based clustering algorithm pro-posed by M. Ester et al. in 1996 [12]. The main ideabehind the algorithm is for each hit in a cluster the neigh-borhood of radius epsilon in time and distance has to con-

Fuzzy

DBSCAN

FIG. 15: Example of clustering algorithm on νe + Ar →1e−+ 1p charged current interactions. The top panel is fuzzycluster and the bottom is DBSCAN. The top shower is anelectron shower and the bottom track is the proton. Fuzzyclustering clusters these two object separate, indicated by thetwo different red and tan colors. DBSCAN does not.

tain at least a minimum number of other hits. DBSCANuses a distance function to determine these parameters.With a minimum number of hits and epsilon input, acollection of hits can be scanned to create a cluster. To-gether with a line finding algorithm, DBSCAN is excep-tional at identifying straight track like clusters which 3Dreconstruction algorithms capitalize on. Fuzzy cluster-ing is different. Unlike DBSCAN, Fuzzy cluster assignsonly “degrees of belonging” to a hit to map clusters. Ahit initially belongs to all clusters until the algorithm de-termines its membership level to a specific cluster afterminimizing some objective function. Fuzzy clustering ispowerful at clustering particle showers. The algorithmis especially powerful at clustering separately one non-showing particle from another if they share the same in-teraction vertex as in FIG. 15.

To study the clustering algorithms I generated 100 sin-gle electron events at 1.5 GeV and 5000 charged currentevents filtering for 1e− + 1p final states. I wrote a LAr-Soft analysis module called MCHitter to calculate thepurity and efficiency of reconstructed clusters. Usingthe BackTracking algorithms in LArSoft the module ex-ecutes the following:

1. Identify all reconstructed clusters on this three wireplanes and use the backtracker object to map thereconstructed hits to the clusters.

2. Identify each reconstructed hit and store the truthlevel information for the particle which contribut-ing ionization energy.

3. Map all true particles to their hits and calculate thepurity and efficiency of each cluster defined below.

Cluster purity is defined as,

Purity =# of hits from trackID in cluster

total # of hits in cluster.

As an example, consider a collection of 50 hits producedfrom two particle tracks, an electron and a charged piongrouped into one cluster. At truth level the electron pro-duced 15 of the hits while the pion produced 35. The

10

Entries 11642

Purity0 0.2 0.4 0.6 0.8 1

Fre

quency

00.10.20.30.40.50.60.70.80.9

1Entries 11642Entries 11642

Electron Purity ­ Fuzzy

Entries 7799

Purity0 0.2 0.4 0.6 0.8 1

Fre

quency

00.10.20.30.40.50.60.70.80.9

1Entries 7799Entries 7799

Electron Purity ­ DB

Entries 11642

Efficiency0 0.2 0.4 0.6 0.8 1

Fre

quency

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4Entries 11642Entries 11642

Electron Efficiency ­ Fuzzy

Entries 7799

Efficiency0 0.2 0.4 0.6 0.8 1

Fre

quency

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4Entries 7799Entries 7799

Electron Efficiency ­ DB

FIG. 16: Single electron events. Fuzzy clustering applied onthe left and DBSCAN on the right.

purity of the electron in this cluster is 15/50 = 0.3. Thepurity of the pion in this cluster is 35/50 = 0.7. Purityis a measure of how much a single cluster is composedof a true particle. If the purity value is less than 1 theclustering algorithm failed to group the true particle hitsseparate from one another. The reconstructed quantityof the total number of hits in the cluster and the numer-ator is extracted from monte-carlo truth information.

Cluster efficiency is defined as,

Efficiency =# of hits from trackID in cluster

total # of hits for that trackID.

Consider the previous example except the charged pionproduced a collection of hits at another location in thedetector such that the total number of hits produced bythe pion is 100. In the first cluster the pion produced35 hits and in the second cluster 65. The pion efficiencyin cluster one is 35/100 = 0.35 and 65/100 = 0.65 inthe second. Efficiency is a measure of how many of allparticle hits lie in a specific cluster. If efficiency is lessthan unity, the clustering algorithm did not group all theparticle hits into a single cluster. The only reconstructedparameter is the grouping of hits into a cluster, all otherinformation is truth level.

For both single electrons and 1e− + 1p final states,clusters with extremely low efficiency made the resultsdifficult to discern between the two algorithms. There-fore, I have cut on efficiencies greater than 10% sincethese clusters containing only one or two particle hits areare not representative of the cluster group as a whole.

The first group is single electron events. The resultsare shown in FIG. 16. Each entry corresponds to a clus-ter. Frequency is the fraction of all clusters analyzed.As expected, the DBSCAN and Fuzzy clustering algo-rithms produced nearly identical purity plots since onlysingle electrons were generated. In the efficiency graphson the second row of FIG. 16 the efficiency plots have very

Entries 2897

Purity0 0.2 0.4 0.6 0.8 1

Fre

quency

0

0.05

0.1

0.15

0.2

0.25

0.3Entries 2897

+1p Purity ­ Fuzzy­

1e

Entries 2897 Entries 2747

Purity0 0.2 0.4 0.6 0.8 1

Fre

quency

0

0.05

0.1

0.15

0.2

0.25

0.3Entries 2747

+1p Purity ­ DB­

1e

Entries 2747

Entries 2897

Efficiency0 0.2 0.4 0.6 0.8 1

Fre

quency

0

0.05

0.1

0.15

0.2

0.25

0.3Entries 2897Entries 2897

+1p Efficiency ­ Fuzzy­

1e

Entries 2747

Efficiency0 0.2 0.4 0.6 0.8 1

Fre

quency

0

0.05

0.1

0.15

0.2

0.25

0.3Entries 2747

+1p Efficiency ­ DB­

1e

Entries 2747

FIG. 17: 1e− + 1p final states. Fuzzy clustering applied onthe left and DBSCAN on the right.

different features. Fuzzy clustering grouped the result-ing electron shower into larger clusters containing morehits than DBSCAN. This is expected from the event dis-play in FIG. 14. DBSCAN on the other hand, producedmuch less efficient clusters. These clusters did not con-tain much of the large electron shower and have lowerefficiencies compared to Fuzzy cluster. From this cali-bration sample both Fuzzy cluster and DBSCAN sufferfrom low efficiency clusters containing only 1 or two hitsverified by the event displays. Fuzzy clustering has aqualitatively better efficiency spectrum. The secondgroup of events is the one electron and one proton finalstates showing in FIG. 17. The goal of this study wasto determine whether fuzzy clustering on average couldidentity the electron shower from the proton track at theinteraction vertex. Distinguishing the two true tracks atthe two dimensional level would allow three dimensionaltracking algorithms to capitalize on their separation. Thepurity plots between the two algorithms in the top paneshow that Fuzzy cluster is more likely to cluster the twotracks separately. The number of high purity clusters forfuzzy cluster is greater than that of DBSCAN for thesame data set. As shown in FIG. 15 DBSCAN is unableto distinguish the two true particle tracks leading to lowpurity clusters. The Fuzzy clustering algorithm does notgroup the electron and proton hits separately 100% of thetime. This is why the distribution of purities is not unity.The efficiency plots reveal something more striking. Ofthe clusters studied, Fuzzy clustering was able to groupthe entirety of true particle tracks into one cluster muchbetter than DBSCAN. In the bottom pane of FIG. 17the small peak shifts left between DBSCAN and Fuzzycluster.

Overall Fuzzy cluster was found to better cluster show-ering particles than DBSCAN. Not shown here, a smallsample of 100 1.5 GeV muons was studied. Muons areminimally ionizing particles with very straight tracks

11

passing through the length of the detector. It was foundthat while fuzzy cluster was unable to produce com-pletely pure clusters, it was much more effective at con-taining most of the muon hits along its trajectory, nearlytripling efficiency over DBSCAN.

Cluster reconstruction is an important step in the re-construction chain. Current clustering algorithms areable to provide physics based hit clustering to trackingfinding algorithms later in the chain. Fuzzy clusteringshows improved purity and efficiency distributions forfiltered 1e− + 1p filtered final states. Both fuzzy clus-tering and DBSCAN suffer from low efficiency clustersdominating from showering particles.

C. Energy Reconstruction

One of the primary goals of any reconstruction codeis to determine the energy of incoming particles in thedetector. Typically, calorimetry analysis uses 3D tracksto reconstruct particle energy. Complete event energyreconstruction is currently in its infancy. Instead I stud-ied the visible energy deposition on the TPC wireplanes.The reconstructed hits provided this information. I wrotean analysis module called MCCounter to study the energydeposition for the collection plane. I generated 1000 sin-gle electrons at 1 GeV and and 1000 electrons between auniform range of 0.5-5 GeV in LArSoft. Electrons werechosen because of their ability to produce showers in thedetector, depositing a large amount of ionization chargeover a broad area. I then reconstructed the hits and clus-ters with FFTHit and Fuzzy cluster.

FIG. 18 shows the number of reconstructed clustersfor the two samples. The mean number of reconstructedclusters had similar distributions with the 1 GeV havinga higher mean. In the 0.5-5.0 GeV electron sample thenumber of reconstructed clusters is energy independent.

Energy deposition on the TPC collection planes isin the form of electrons propagated through the wires.FIG. 20 shows the fraction of true energy deposition totrue electron energy. This plots are useful at under-standing how much of the electrons true energy could,at maximum, be deposited on the wires planes after driftrecombination effects. On average about 40-45% of anelectrons true energy is deposited on the wires. Thisis because not all energy depositions are reconstructed.There is a threshold electron count which must be satis-fied before the hit is considered for reconstruction. Thiscriterion eliminates a large fraction of energy depositedon the wires leading to an ionization to true energy ra-tio far from unity. For the 1 GeV electron in the toppanel of FIG. 20 the distribution is much narrower thanin the bottom pane. The uniform energy electrons havea higher RMS because lower and higher electron will pro-duce different amounts of ionization energy. The higherthe incoming electron, the more likely it’s hits are abovethe threshold and counted as energy deposition. Thelower energy electrons do not meet this threshold as fre-

Entries 1000

Mean 7.938

RMS 2.671

0 5 10 15 20 25 30 35 40

Eve

nts

0

20

40

60

80

100

120

140

160Entries 1000

Mean 7.938

RMS 2.671

Cluster Count

1 GeV Electron

Entries 1000

Mean 6.271

RMS 2.652

Number of Clusters

0 5 10 15 20 25 30 35 40E

ve

nts

0

20

40

60

80

100

120

140Entries 1000

Mean 6.271

RMS 2.652

0.5 ­ 5.0 GeV Electron

FIG. 18: Number of reconstructed clusters with Fuzzy Clus-tering algorithm. The top plot is for 1 GeV electrons and thebottom 0.5 - 5.0 GeV. Cluster count appears independent ofenergy range

0

1

2

3

4

5

0 1 2 3 4 5

AD

CC

ounts×

106

True e− Energy (GeV)

ADC Counts vs. Energy

FIG. 19: Reconstructed ADC count on collection plane wiresas a function of true electron energy. ADC counts from wiresmaps linearly with true electron energy.

quently.For the uniform distribution of electron energies, the

reconstructed ADC counts are plotted versus true energyin FIG. 19. Reconstructed ADC counts are the analog todigital conversion units specific to the detector electron-ics. Higher ADC values correspond to more ionization

12

Entries 1000

Mean 0.431

RMS 0.01129

0 0.2 0.4 0.6 0.8 1

Eve

nts

0

50

100

150

200

250

300

350

400

450 Entries 1000

Mean 0.431

RMS 0.01129

Energy Fraction

1 GeV Electron

Entries 1000

Mean 0.411

RMS 0.03992

Ionization/True Energy0 0.2 0.4 0.6 0.8 1

Eve

nts

0

20

40

60

80

100

120

140 Entries 1000

Mean 0.411

RMS 0.03992

0.5 ­ 5.0 GeV Electron

FIG. 20: Energy deposition on TPC collection plane as func-tion of true electron energy. The top plot is for 1 GeV elec-trons and the bottom 0.5 - 5.0 GeV. Expected energy depo-sition, or visible energy, is 40% of true electron energy.

electrons deposited in the detector. The scatter plot in-dicates that for low energy electrons the spread in ADCis less than at higher energies. Despite a few scatteredpoints relationship is roughly linear by eye.

The energy studies done here are important for cali-brating the MicroBooNE detector against simulated par-ticle events. Ultimately, reconstruction algorithms mustbecome sophisticated enough to distinguish, track, andanalyze multiple particle events. Neutrino interactionsmust be isolated to probe the low energy excess.

V. DATA ACQUISITION

A. Shaper and FEM Set Up

The MicroBooNE detector will employ 30 photomul-tiplier tubes to record timing for particle interactionsinside the detector. A PMT measures incoming photonsby generating few nanosecond wide pulse which is sentto the read out electronics. The pulse height and areaare correlated to the amount of energy deposited inthe phototube. The Nevis group is constructing andcalibrating the read out electronics for MicroBooNE.This summer I helped calibrate the shaper and FEM,

Trigger Module

Pulse

Generator

RC

Circuit

Shaper

PC

Beam Gate

Ch. X

FEM

FIG. 21: Shaper and FEM data acquisition set up.

Front End Modules, for the detector. The calibrationset up is shown in FIG. 21. A trigger module issues twosignals. One pulse is sent to a trigger a pulse generatorwhich outputs a step function to an RC circuit. Thevoltage amplitude of the pulse generator is steppedbetween nine levels. The RC circuit produces an narrowPMT-like pulse to the signal shaper. The shaper has16 channels. The second trigger module pulse triggersthe opening of the beam gate, the readout windowfor data collection. The delay between the beam gateopening and the pulse arrival from the RC circuit canbe adjusted. The reason they are not simultaneous is toretrieve information on the ADC pedestal for the firstfew time samples to estimate the baseline. The purposeof the shaper is to modify the input signal bandwidth toshape the signal before being digitized by the FEM. Theshaper effectively smooths high frequency interferenceand amplifies low frequency interference. The FEM thensends the digitized signal to a PC to be recorded.

The shaper and FEM response is important for energyreconstruction of phototube signals. The shaper andFEM response should be a linear function of inputvoltage amplitude. Linearity is related to how well onecan reconstruct the energy of the PMT pulse. If the re-sponse is not linear, it can be corrected by some function.

B. Data Processing

This summer I studied the ADC pedestal, PMT pulsereconstruction, and the shaper-FEM linearity. I wrote asimple pulse reconstruction album with Kazuhiro Terao’sAnalysis and Decoder package called pmtbaseline.

The goal is to correlate the input step function ampli-tude from the pulser box with the reconstructed chargeof the PMT pulse through the electronics system. Foreach channel on the shaper, a known PMT like pulse isinputted, and reconstructed on a PC after digitization inthe FEM. Data was recorded for 10,000 pulses. Estimatesfor the pedestal height, peak pulse height, start, max am-plitude, and end time and charge sum for each pulse were

13

Channel0 5 10 15 20 25 30 35 40

AD

C

2040

2045

2050

2055

2060Entries 400000

Mean x 20

Mean y 2049

RMS x 11.54

RMS y 3.248

0

1000

2000

3000

4000

5000

6000

7000

8000

Entries 400000

Mean x 20

Mean y 2049

RMS x 11.54

RMS y 3.248

Pedestal Mean

Channel0 5 10 15 20 25 30 35 40

AD

C

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6Entries 400000

Mean x 20

Mean y 0.3674

RMS x 11.54

RMS y 0.1782

0

1000

2000

3000

4000

5000

6000

7000

8000

Entries 400000

Mean x 20

Mean y 0.3674

RMS x 11.54

RMS y 0.1782

Pedestal RMS

FIG. 22: Pedestal mean and standard deviation for each FEMchannel. The spread in ADC mean is roughly 10 ADC countwith a primarily 0.5 ADC count variation in one channel.

calculated using the pmtbaseline module. This processis repeated for all 16 shaper channels.

C. Results

First, a good understanding of the ADC pedestal isimportant for pulse reconstruction. The analog to digi-tal converter or ADC measures the total charge sent toit in a specified period of time. When triggered, a gateis opened and the signal is allowed into the device. Thecharge from the signal is collected on a capacitor, mea-sured and then a digital signal is sent onward, containingthe amount of charge collected. A slight complication tothis process is that a small quantity of charge is sent tothe ADC capacitor whether there is a signal or not. Thismeans that if there was no signal, the ADC will still mea-sure a small quantity of charge called the pedestal. Sincethis pedestal is always the same during a given run, it canbe measured and subtracted. For 10,000 beam gatesamples, the first time samples are averaged to estimatethe pedestal. The results are plotted in FIG. 22. TheADC pedestal has a mean values between FEM channelsof 2049 with a standard deviation of approximately 0.36ADC counts.

Next, Kazu, Dan, and I used the setup described inFIG. 21 to test the shaper-FEM linearity. For 9 ampli-tude settings on the pulser box, data was collected overthe 16 shaper channels. The raw data was processedwith pmtbaseline. The integrated charge and pulse am-plitude are two estimators of pulse energy. The distribu-tions are plotted in FIG. 23. In the right pane of FIG. 23the distribution of pulse amplitudes is highly asymmetricdue to a digitization effect of the FEM. The digitizationsplits the otherwise continuous distribution into discretedistributions based on the ADC resolution. Values nearthe resolution threshold of the ADC either get placed onthe left or right of the bit value. A similar effect has beenreported on by [17]. Due to this effect it is preferable toseparate the distribution into three smaller ones. Threecuts were made on the pulse height distribution becausemany of the peak height distributions show three peakfeatures. The procedure involves finding the pulse ampli-

Charge100 150 200 250 300 350 400 450 500 550 600

Pu

lse

Co

un

t

200

400

600

800

1000Integrated Charge Ch. 4

Peak Amplitude70 72 74 76 78 80 82

Eve

nts

0

200

400

600

800

1000

1200

1400

1600

1800

2000

2200

Peak Heights Ch. 4

FIG. 23: Left pane plots integrated pulse charge over 9 am-plitudes increasing in amplitude by one step. Right pane iszoomed in distribution of two peak height distributions. Thepeak heights distributions are highly asymmetric.

Pulse Amplitude24 25 26 27 28 29 30

Puls

e C

ount

0

500

1000

1500

2000

2500

3000

3500

Pulse Height Division

Charge140 145 150 155 160 165 170 175 180 185

Puls

e C

ount

0

100

200

300

400

500

600

700

800

900

Charge: Amp=4.0

FIG. 24: Cuts on the pulse height distribution in the top panereveal three distinct Gaussian like charge sums in the bottompane.

tude found in the maximum number of pulses and cut-ting left and right ±0.5 in pulse amplitude. The resultsare shown in the top pane of FIG 24. Cutting on thepeak height distribution reveals three corresponding low,middle, and high energy components in the charge distri-bution: the bottom panel in FIG. 24. For data analysiswe choose the middle, blue, distribution as we believethe middle peak heights are the most representative ofthe true PMT pulse height and therefore the best esti-mator of the energy value in the charge sum. Scanningthe input amplitude between 4 and 12, the middle chargedistribution is fit to a gaussian curve and the results forone channel are presented in FIG. 25.

14

Amplitude4 5 6 7 8 9 10 11 12

Mid

dle

Mean F

itte

d C

harg

e

150

200

250

300

350

400

450

500 / ndf 2χ 255.1 / 7p0 0.2447± ­12.28 p1 0.02926± 43.64

/ ndf 2χ 255.1 / 7p0 0.2447± ­12.28 p1 0.02926± 43.64

Middle Charge Sum Ch. 4

Amplitude4 5 6 7 8 9 10 11 12

2χ

0

1

2

3

4

5

6 vs. Amplitude Ch. 42χ

All

Middle

FIG. 25: Linear fit to mean charge from gaussian fit of mid-dle charge distribution as a function of input amplitude frompulser box. The bottom panel is a plot of χ2 value for eachgaussian fit over the amplitude range.

The fitted mean value is plotted against input ampli-tude. The relationship is linear. The size of the fittedmean error bar is smaller than the size of the dot. Unfor-tunately the statistical uncertainty in the gaussian fit isextraordinarily small leading to a large χ2 value on thelinear fit. Systematic uncertainties in measurement havenot been included. One way to incorporate this is to giveeach charge integral measurements a default systematicuncertainty. Despite this, the linearity of the shaper andthe FEM is good. The bottom panel in FIG. 25 shows theχ2 goodness of fit value for the gaussian curve of the mid-dle charge distribution. The blue dots are the χ2 valuesfor a naive gaussian fit to the charge distribution with-out cuts. Magenta dots are the χ2 values for the chargedistribution with cuts. The cut distribution fits have amuch lower χ2 value over the range of input amplitudesindicating a better fit.

The slope and y-intercept of mean fitted charge versusamplitude is plotted over the 16 shaper channels. Datafor channels 13 and 14 have not been collected yet. Thedata is shown in FIG. 26. The slope varies between 42and 46 with errors bars smaller than the dot size. Thisshows the shaper response is the same between each in-put channel. The fitted y-intercept value is nonzero andperhaps reveals a non-linearity at low input amplitudes.We would expect a zero gaussian mean at zero input am-

40

42

44

46

48

50

Slo

pe

of

Fit

ted

Mea

nC

harg

e

Mean & Y-inter Fitted Charge vs. Channel

-30

-25

-20

-15

-10

-5

0

0 2 4 6 8 10 12 14F

itte

dY

-inte

rcep

tC

harg

e

Shaper Channel Number

FIG. 26: Fitted Mean and Y-intercept values for the linearityplot in the top panel of FIG. 25. The slope varies between 42and 46. The non-zero value of the y-intercept values indicatedsome non-linearity at low amplitudes.

plitude. Further investigation is necessary.The signal shaper and FEM response is linear in out-

put with pulser amplitude. The next step is to test therest of the signal shapers for MicroBooNE and expandon the reconstruction algorithms to account for late lightproduced from Argon scintillation.

Acknowledgments

I would first like to thank Professor Mike Shaveitz forproviding the opportunity to gain experience in Micro-BooNE project as well as his helpful criticism duringweekly presentations. I would also like to thank Profes-sor John Parson for his administrative finesse throughoutthe summer.

A big thanks to Kazu for teaching me PyROOT andgetting me excited about real physics analysis. Your en-thusiasm for software has taught me that anything canbe programmed.

I am thankful to Georgia Karagiorgi for getting mestarted with LArSoft analysis. Her exuberance for neu-trino physics is inspiring.

Thanks to David Caratelli for allowing me to be hisPMT and Bo ringing partner.

15

I acknowledge the revisions of Douglas Davis.Finally, thanks to the National Science Foundation for

making this summer possible.

16

[1] Donald H. Perkins, Introduction to High Energy Physics(Cambridge University Press, United Kingdom, 2001),4th ed.

[2] David Griffiths, Introduction to Elementary Particles(Wiley-VCH, 2008) 2nd ed.

[3] B.R. Martin, G. Shaw, Particle Physics (John Wiley &Sons, United Kingdom, 2008) 3rd ed.

[4] B. Pontecorvo, “Mesonium and Antimesonium” (J. Ex-ptl. Theoret. Phys, U.S.S.R) 33, 549 (1957).

[5] J. Beringer et al. “Particle Data Group”, PR D86, 010001(2012)

[6] B. T. Cleveland et al (1998). ”Measurement of the So-lar Electron Neutrino Flux with the Homestake ChlorineDetector”. Astrophysical Journal 496: 505526

[7] SNO Collaboration, “Direct Evidence for Neutrino Fla-vor Transformation from Neutral-Current Interactions inthe Sudbury Neutrino Observatory” Phys. Rev. Lett. 89,011301 (2002)

[8] L. Camilleri, “MicroBoone”, Nuclear Physics B Proceed-ings Supplement 00 (2012) 13

[9] B. Flemming, S. Zeller et al. “A Proposal for a New Ex-periment Using the Booster and NuMI Neutrino Beam-lines: MicroBooNE” (2007)

[10] Ben Jones, “Results from the Bo Liquid Argon Scintilla-tion Test Stand at Fermilab” Light Detection In NobleElements (LIDINE) at Fermilab, May 29th 2013

[11] Birks, J.B. The Theory and Practice of ScintillationCounting. London: Pergamon (1964)

[12] Martin Ester, Hans-Peter Kriegel, Jrg Sander, XiaoweiXu, ”A density-based algorithm for discovering clustersin large spatial databases with noise” (1996)

[13] R5912 Specification Sheet retrieved fromhttp://www.alldatasheet.com/datasheet-pdf/pdf/62744/HAMAMATSU/R5912.html on July28, 2013.

[14] X7R Specification Sheet retrieved fromhttp://www.avx.com/docs/catalogs/cx7r.pdf on July28, 2013.

[15] G. Karagiorgi “Bo readout update” - MicroBooNE Doc-ument 2700-v1

[16] B. Jones “Flowcart for larsoft reconstruction” - Micro-BooNE Document 2572-v1

[17] W. E. Cleland, E.G. Stem “Signal processing considerations for liquid ionizat ion calorimeters in a high rate en-vironment” Nucl. Instr. and Meth. in Phys . Res. A338(1994) 467-497

17

FIG. 27: PMT GUI

Appendix A: Document

Document produced in LATEXusing REVTeX 4. Allgraphs were generated by the author in gnuplot usingthe tikz terminal, and ROOT.

Appendix B: PMT GUI

Toward the end of the summer, I wrote a simple ROOTGUI to interpret the status word output by PMT elec-tronics. WinDriver is software for the PCIE controller onUBELEC in the electronics room used to access the elec-tronics. The GUI takes the two 32-bit status words andprints out the relevant bit information in screen. Cur-rently, Mike Phipps has taken over the project and ex-panded it to query the TPC crates. A picture of the GUIis shown in FIG. 27. The GUI generates two fake 32-bitwords and tests the relevant bits.

Appendix C: Vic’s Files

I have a lot of files.

/a/home/houston/vic.genty

• lardev

– EnergyStudy - Input a reconstructed root filewith spacepoint and track information, out-put to stdout the reco track coordinates forparticle. Cool for plotting in 3D

– MCCalo - Failed attempt at true calorimetry

– MCCounter - True energy studies for hit recon-struction, energy deposition on wireplanes

– MCFilter - Input is ROOT file with clusteringalgorithm and outputs nothing useful. Firstattempt at getting information out about hitsin clusters

– MCFinder - Input is a ROOT file with clus-tering algorithm, output to stdout which finalstates are 1e+1p. Used to look a event display.

– MCHitter - Output is ROOT file with purityand efficiency plots for single electron and pro-ton final states sorted by wireplane. Use withROOT macro in ~/larscripts

• larwork

– Lots of LArSoft generated ROOT files listed(mostly) by date. Most have been generatedand reconstructed with an outdated LArSoftversion so they will become useless.

• larscripts

– coolplots_MCCounter - Before I learned Py-ROOT I used CINT. Takes ROOT files fromMCCounter and displays the tree contents. Use—readroots.C—.

– loops_MCHitter - Takes ROOT filesfrom MCHitter and plots the tree con-tents. Use readroots.C for 1e− + 1p andreadroots_muonselectrons.C for singleelectrons and muons.

If deleted: a full backup (∼ 14 GB) can be found ina/data/morningside/vic.genty/houston

/a/data/westside/vic.genty

• 0617release- 5000 CC events generated with GE-NIE for the 0617 release of LArSoft. There aresome muons here too.

• 0625release- 5000 CC events generated with GE-NIE for the 0617 release of LArSoft. There aresome electrons and muons here too.

/a/data/morningside/vic.genty

• calo_tests - Nothing useful.

• houston - Back up of home folder.

• runs -

– nue_cc_uniform - At one point during thesummer, after LArSoft updated, it was ter-ribly hard to generate ν events with GENIE.The program lar would eat up gigs and gigsof virtual memory eventually erroring out.These are 100 CC events each I was able togenerate and test with the June 25 release ofLArSoft.

18

– runs - Single electrons I generated with June25 release to study energy deposition. 10,1000, 5000 at mono- and uniformly-energetic

electrons produced.