How LArTPC Geometry Affects the Pandora Reconstruction · 2019. 9. 16. · G.Milton & S.Green...

20
G.Milton & S.Green DUNEFD Sim/Reco How LArTPC Geometry Aects the Pandora Reconstruction Gulliver Milton & Steven Green 16th September 2019 1

Transcript of How LArTPC Geometry Affects the Pandora Reconstruction · 2019. 9. 16. · G.Milton & S.Green...

  • G.Milton & S.Green DUNEFD Sim/Reco

    How LArTPC Geometry Affects the Pandora Reconstruction Gulliver Milton & Steven Green 16th September 2019

    !1

  • G.Milton & S.Green DUNEFD Sim/Reco �2

    Introduction

    Pandora is used for the reconstruction in a number of different LArTPC experiments with different detector geometries; the wire angles (to the vertical) for those experiments are:

    θu θv θwProtoDUNE-SP 36 (𝜋/5) -36 (-𝜋/5) 0ProtoDUNE-DP 90 (𝜋/2) - 0

    MicroBooNE 60 (𝜋/3) -60 (-𝜋/3) 0SBND 60 (𝜋/3) -60 (-𝜋/3) 0

    ICARUS 30 (𝜋/6) -30 (-𝜋/6) 90 (𝜋/2)DUNEFD-SP 36 (𝜋/5) -36 (-𝜋/5) 0

    We want to find out, which is ultimately the best geometry to use.

    In order to do this, we need to extract ourselves from all experiment specific software frameworks so that we can compare exactly the same event for different detector geometries.

    To do this, we have built a “Pseudo-LArTPC” that only depends only upon Geant4.

  • G.Milton & S.Green DUNEFD Sim/Reco �3

    Concept:

    Get Geant4 to write out the full 3D information in fine detail (1mm cube bins).

    The “charge” for these hits is the true energy loss in the detector, which isn’t realistic to use in analyses, but we don’t use calorimetry in the reconstruction.

    Introduction

    3D

    x

    y

    z

  • G.Milton & S.Green

    U View

    V View

    W View

    x

    x

    x

    zz*

    cos(θ U

    )-y*s

    in(θ

    U)

    z*co

    s(θ V

    )-y*s

    in(θ

    V)

    3D

    x

    y

    z

    DUNEFD Sim/Reco �4

    Inside (standalone Pandora), project the 3D information into the 2D views (pitch and hit width both of the order of 5mm).

    Then study the reconstruction efficiency as a function of the detector geometry.

    Introduction

    Project

    These functions are the inverse of what is used inside Pandora to

    build 3D points from the 2D images

  • G.Milton & S.Green

    MicroBooNE-Like

    U View

    V View

    W View

    ProtoDUNE-SP-Like ICARUS-Like

    DUNEFD Sim/Reco �5

    Example

    θu [Deg] 60 (𝜋/3) 36 (𝜋/5) 30 (𝜋/6)

    θv [Deg] -60 (-𝜋/3) -36 (-𝜋/5) -30 (-𝜋/6)

    θw [Deg] 0 0 90 (𝜋/2)

    The Pandora reconstruction was initially developed for MicroBooNE

  • G.Milton & S.Green

    0 50 100 150Wire Angle

    0.90

    0.92

    0.94

    0.96

    0.98

    1.00

    1.02

    1.04

    Effic

    ienc

    y

    θu [Deg]

    DUNEFD Sim/Reco �6

    Reconstruction Metric

    Initially we looked at the reconstruction efficiency for events containing a single 1GeV 𝜇- track in a ProtoDUNE-SP-like detector; all daughter particle hits were removed.

    In this (and all other studies presented here) 10,000 isotropic events were considered.

    The Pandora reconstruction efficiency was almost 100% for all detector geometries considered because Pandora could clearly reconstruct single ionisation tracks.

    U View

    V View

    W View

    θu Variedθv -θuθw 0

    Pitch U 4.7mmPitch V 4.7mmPitch W 4.8mm

  • G.Milton & S.Green DUNEFD Sim/Reco �7

    Reconstruction Metric

    U View

    V View

    W View

    Instead we consider the full particle hierarchy.

    In the LArG4 step for the majority of active experiments, the option to suppress electromagnetic shower daughters is active so no MC particles for ẟ-rays appear.

    However, for these studies we disable that to get the full MC particle hierarchy.

    Correctly reconstructed event: Unique reconstructed particle match to all MC particles in the event including the daughter ẟ-rays.

    Cheated Reconstruction To Highlight ẟ-Ray MC Particles

  • G.Milton & S.Green

    0 100 200 300 400Number of MC Hits

    1

    10

    210

    310

    410

    510

    00.02

    0.040.06

    0.080.1

    Mom

    entum [G

    eV]

    0 50

    100

    150

    200

    250

    Number of Hits

    1 10

    210

    310

    410

    DUNEFD Sim/Reco �8

    All ẟ-Rays Momentum > 0.01 GeV

    0 0.02 0.04 0.06 0.08 0.1Momentum [GeV]

    0

    50

    100

    150

    200

    250

    Num

    ber o

    f Hits

    1

    10

    210

    310

    410

    Some ẟ-rays deposit next to no hits in the detector; usually we would not consider these in the Pandora reconstruction metrics.

    However, as later we want to compare detectors with 2 and 3 views, we don’t want to cut on the number of hits because the 3 view detectors will inherently have more than 2 view detectors.

    Instead, we cut on the momentum of the ẟ-rays: Require ẟ-ray momentum > 0.01 GeV for us to consider it.

    Reconstruction Metric

  • G.Milton & S.Green

    0 50 100 150Wire Angle

    0.86

    0.87

    0.88

    0.89

    0.90

    0.91

    0.92

    Effic

    ienc

    y

    Single Phase, Pitch 4.7mm

    θu [Deg]

    Hie

    rarc

    hy R

    econ

    stru

    ctio

    n Ef

    ficie

    ncy

    θu Variedθv -θuθw 0

    Pitch U 4.7mmPitch V 4.7mmPitch W 4.8mm

    3 Views, Pitch 4.7mm

    DUNEFD Sim/Reco �9

    Results

    Hierarchy reconstruction efficiency averages around 90% for all wire angles.

    Efficiency peaks at θu=60 and θu=120 degrees, where the three wire planes are maximally separated.

    There is a dip in efficiency at θu=0 and θu=180 occur at these points the U, V and W planes look identical.

    There is a smaller dip in efficiency at θu=90 where the U and V planes looks identical (but W still distinct).

  • G.Milton & S.Green DUNEFD Sim/Reco �10

    Results

    The majority, ~55%, of events have no high momentum ẟ-ray, but the remainder do have at leat one.

    As the number of ẟ-rays in the event increases, the reconstruction efficiency decreases as it becomes harder to resolve each individual ẟ-ray.

    However, even for complex events (up to 3 ẟ-rays) we have a good reconstruction efficiency (~50%).

    0 2 4 6-RaysδNumber Of

    0.0

    0.1

    0.2

    0.3

    0.4

    0.5

    0.6

    Frac

    tion

    Of E

    vent

    s

    Single Phase, 4.7mm Pitch, 60 DegreesDual Phase, 4.7mm Pitch, 90 DegreesDual Phase, 3.0mm Pitch, 90 Degrees

    0 1 2 3 4 5-RaysδNumber Of

    0.0

    0.5

    1.0

    Effic

    ienc

    y

    Single Phase, 4.7mm Pitch, 60 DegreesDual Phase, 4.7mm Pitch, 90 DegreesDual Phase, 3.0mm Pitch, 90 Degrees

    θu 60θv -60θw 0

    Hie

    rarc

    hy R

    econ

    stru

    ctio

    n Ef

    ficie

    ncy

  • G.Milton & S.Green DUNEFD Sim/Reco �11

    2 View LArTPC

    U View

    V View

    W View

    3 Views 2 Views

    N/A

    The Pandora LArTPC reconstruction was initially developed for a 3 view LArTPC.

    Now we want to compare the reconstruction performance for a 3 view and 2 view LArTPC.

    To do this we remove the V view from the reconstruction*.

    Repeat the same study as before…

    *Note: This is not a direct comparison between single and dual phase detector geometries as the dual phase has a better signal to noise ratio and a different pitch, but we will attempt to address some of these issues later.

  • G.Milton & S.Green DUNEFD Sim/Reco �12

    θu [Deg]

    3 view LArTPCθu Variedθv -θuθw 0

    2 view LArTPCθu Variedθv -θw 0

    The 2 view LArTPC configuration has a lower average reconstruction efficiency by ~6% as a function of wire angle.

    The 2 view LArTPC reconstruction is most efficiency when θu=90 and the wires are maximally separated.

    The efficiency drops at θu=0 and θu=180 when the two views begin to look similar.

    2 View LArTPC

    0 50 100 150Wire Angle

    0.80

    0.82

    0.84

    0.86

    0.88

    0.90

    0.92

    Effic

    ienc

    y

    Single Phase, Pitch 4.7mm

    Dual Phase, Pitch 4.7mm

    3 Views, Pitch 4.7mm2 Views, Pitch 4.7mm

    Hie

    rarc

    hy R

    econ

    stru

    ctio

    n Ef

    ficie

    ncy

  • G.Milton & S.Green DUNEFD Sim/Reco �13

    2 View LArTPC

    3 view LArTPCθu 60θv -60θw 0

    2 view LArTPCθu 90θv -θw 0

    0 1 2 3 4 5-RaysδNumber Of

    0.0

    0.5

    1.0

    Effic

    ienc

    y

    = 60UθSingle Phase, 4.7mm Pitch,

    = 90UθDual Phase, 4.7mm Pitch,

    3 Views, Pitch 4.7mm, θu = 602 Views, Pitch 4.7mm, θu = 90

    Hie

    rarc

    hy R

    econ

    stru

    ctio

    n Ef

    ficie

    ncy

    Hierarchy reconstruction efficiency as a function of the number of delta rays is comparable to 3 view LArTPC in shape, but lower in magnitude.

    Now let’s attempt to account for the other differences between single and dual phase LArTPCs, namely:

    Pitch - In ProtoDUNE single phase pitch ~4.7mm, but dual phase is ~3mm.

    Signal to noise ratio - Dual phase aim is to have a gain roughly 10 times better than single phase.

  • G.Milton & S.Green DUNEFD Sim/Reco �14

    3mm Pitch 4.7mm PitchW View

    Example of the same event with different simulated pitches.

    For illustrative purposes only look at W view here, but same binning is applied to U and V views also.

    Example

  • G.Milton & S.Green DUNEFD Sim/Reco �15

    2 View LArTPC

    Moving the pitch from 4.7mm to 3mm increases the reconstruction efficiency significantly, ~4%, for the 2 view LArTPC model.

    Similar gain observed in reconstruction efficiency as a function of number of ẟ-rays.

    Shape of distributions broadly similar to larger pitch model.

    3 view LArTPCθu Variedθv -θuθw 0

    2 view LArTPCθu Variedθv -θw 0

    0 50 100 150Wire Angle

    0.80

    0.82

    0.84

    0.86

    0.88

    0.90

    0.92

    Effic

    ienc

    y

    Single Phase, Pitch 4.7mmDual Phase, Pitch 4.7mmDual Phase, Pitch 3.0mm

    0 1 2 3 4 5-RaysδNumber Of

    0.0

    0.5

    1.0

    Effic

    ienc

    y

    = 60UθSingle Phase, 4.7mm Pitch, = 90UθDual Phase, 4.7mm Pitch, = 90UθDual Phase, 3.0mm Pitch,

    θu [Deg]

    3 Views, Pitch 4.7mm, θu = 602 Views, Pitch 4.7mm, θu = 902 Views, Pitch 3.0mm, θu = 90

    3 Views, Pitch 4.7mm2 Views, Pitch 4.7mm2 Views, Pitch 3.0mm

    Hie

    rarc

    hy R

    econ

    stru

    ctio

    n Ef

    ficie

    ncy

    Hie

    rarc

    hy R

    econ

    stru

    ctio

    n Ef

    ficie

    ncy

  • G.Milton & S.Green DUNEFD Sim/Reco �16

    In order to mimic a better signal to noise ratio we will place a threshold on the hit energy.

    This is a rough model as it only affects the signal quality and not the background noise level (i.e. fake hits).

    This threshold is applied to raw energy deposit in the detector and will be lower for the 2 view LArTPC than the 3 view LArTPC.

    Hit Threshold

    No Threshold

    1 MeV Threshold

    2 MeV Threshold

    Example : W View Hits

  • G.Milton & S.Green DUNEFD Sim/Reco �17

    2 View LArTPC

    0 1 2Hit Thrshold [MeV]

    0.0

    0.2

    0.4

    0.6

    0.8

    1.0

    Effic

    ienc

    y

    = 60UθSingle Phase, Pitch 4.7mm, = 90UθDual Phase, Pitch 4.7mm, = 90UθDual Phase, Pitch 3.0mm,

    3 Views, Pitch 4.7mm, θu = 602 Views, Pitch 4.7mm, θu = 902 Views, Pitch 3.0mm, θu = 90

    Hie

    rarc

    hy R

    econ

    stru

    ctio

    n Ef

    ficie

    ncy

    Hit Threshold [MeV]

    As the threshold increases the hierarchy reconstruction efficiency drops as fewer hits appear in the detector.

    The 3 view LArTPC is more resilient to large hit thresholds, as the additional view allows Pandora to compensate for information lacking in a single view.

    Significant drop off in 2 view LArTPC reconstruction efficiency around a threshold of 0.6 MeV.

    Using the optimal wire

    angles

  • G.Milton & S.Green DUNEFD Sim/Reco �18

    2 View LArTPC

    *These numbers are very speculative because they approximate a huge number of effects. Do

    not assume they are 100% accurate!

    Using Thresholds*Reconstruction

    Efficiency3 Views

    4.7mm Pitch2 Views

    4.7mm Pitch2 Views

    3mm Pitch1GeV 𝜇- 0.871±0.003 0.848±0.004 0.888±0.0033GeV 𝜇- 0.829±0.004 0.800±0.004 0.850±0.0045GeV 𝜇- 0.821±0.004 0.787±0.004 0.840±0.003

    1 GeV 𝜇-

    0.00 0.05 0.10 0.15 0.20Hit Thrshold [MeV]

    0.80

    0.82

    0.84

    0.86

    0.88

    0.90

    0.92

    0.94

    Effic

    ienc

    y

    = 60UθSingle Phase, Pitch 4.7mm, = 90UθDual Phase, Pitch 4.7mm, = 90UθDual Phase, Pitch 3.0mm,

    3 Views, Pitch 4.7mm, θu = 602 Views, Pitch 4.7mm, θu = 902 Views, Pitch 3.0mm, θu = 90

    Hie

    rarc

    hy R

    econ

    stru

    ctio

    n Ef

    ficie

    ncy

    Hit Threshold [MeV]

    After discussion with TPC calorimeter experts we found that a hit threshold of ~200 KeV is appropriate for single phase and ~20 KeV* for dual phase.

    Using these thresholds the hierarchy reconstruction efficiencies are very similar; the 2 view, 3mm pitch LArTPC is marginally better than the 3 view, 4.7mm pitch LArTPC by ~1.5%.

    A similar trend also appears when the energy of the muon increased.

    These results rely on the hit threshold assumptions, which should always be kept in mind!

    200 KeV 3 View

    20 KeV 2 View

  • G.Milton & S.Green

    3D

    x

    y

    z

    U View

    V View

    W View

    Project

    DUNEFD Sim/Reco �19

    Conclusions

    We have studies the effect on the Pandora cosmic-ray muon reconstruction of varying the detector geometry for 3 and 2 view LArTPCs in a toy simulated detector.

    Both detector technologies yield very similar hierarchy reconstruction efficiencies once the pitch and hit energy thresholds have been accounted for.

    More work is needed to compare how the geometry changes affect the remaining reconstruction algorithm chains (test beam/neutrino) and the accuracy of the 3D reconstruction.

    Thank you for your attention!

    Questions?

  • G.Milton & S.Green DUNEFD Sim/Reco

    Pandora Pattern Recognition

    �20

    Pandora is an open project and new contributors would be extremely welcome.

    We’d love to hear from you and we will always try to answer your questions.

    Pandora SDK Development

    LAr TPC algorithm development

    John Marshall ([email protected])
Mark Thomson ([email protected])

    John Marshall ([email protected])
Andy Blake ([email protected])

    DUNE FD Integration

    MicroBooNE : Joris Jan de Vries, Jack Anthony

    ProtoDUNE : Stefano Vergani

    DUNE : Jhanzeb Ahmed, Mousam Rai, Ryan Cross

    Lorena Escudero ([email protected])

    ProtoDUNE Integration Steven Green ([email protected])

    MicroBooNE Integration Andy Smith ([email protected])

    Graduate Students

    https://github.com/PandoraPFA https://pandorapfa.slack.com

    mailto:[email protected]:[email protected]:[email protected]:[email protected]:[email protected]?subject=mailto:[email protected]:[email protected]://github.com/PandoraPFAhttps://pandorapfa.slack.com