Searching for Clues for a Matter Dominated Universe in LArTPCs

Searching for Clues for a Matter Dominated Universe in LArTPCs

SLAC Experimental Seminar Series

Yeon-jae JwaColumbia University

1

Yeon-jae Jwa | SLAC experimental seminar series (Nov. 24th 2020)

Matter dominated universe &Sakharov’s conditions for baryogenesis

Andrei Sakharov made an argument for a matter dominated universe in 1967.Three conditions to generate matter-antimatter asymmetry in the early universe.

1. Baryon number violating process

2. CP violating process

3. Interactions out of thermal equilibrium

2


Matter dominated universe &Sakharov’s conditions for baryogenesis

Andrei Sakharov made an argument for a matter dominated universe in 1967.Three conditions to generate matter-antimatter asymmetry in the early universe.

1. Baryon number violating process

2. CP violating process

3. Interactions out of thermal equilibrium

3

Investigate in neutrino experiments (LArTPCs)


Contents

● Introduction to current and next generation LArTPC detectors

● Leptonic CP sensitivity in DUNE with a light sterile neutrino

● Neutron-antineutron oscillation search in LArTPCs

● ML-based based rare physics triggering in DUNE and FPGA implementation

4


Liquid Argon Time Projection Chambers (LArTPCs)

5

● LArTPC technology is shared by the SBN program and DUNE.

● Ionization electrons (~dE/dx) are drifted to anode wire planes by the electric field.

● LArTPCs can capture 3D event with ~mm resolution for ionizing particle.

● Excellent calorimetry and particle ionization.


Short-Baseline Neutrino (SBN) program @ Fermilab

6

ICARUS600m

476 tons

MICROBOONE470m

89 tons

SBND110m

112 tons

arXiv:1503.01520


Testing / Installing / Commissioning TPC electronics for SBND

7

arXiv:1503.01520

We were here!

Front-end modules (FEMs) test @ Nevis lab.

Crate installation @ SBND


Physics in SBN● SBN program’s physics objectives :

○ Light sterile neutrino search

○ Neutrino-argon interactions measurement

○ Beyond Standard Model searches

● Development platform for Deep Underground Neutrino Experiment (DUNE)

8


Standard 3-neutrino oscillation ● PMNS matrix : Standard 3-neutrino mixing

between the flavor states and mass eigenstates

● Neutrino oscillations can be characterized by

Δm221 (~ 7.5 × 10-5 eV2), Δm2

31 (~ 2.5 × 10-3 eV2),

each accounts for slow “solar”, fast “atmospheric”

oscillations.

9

﹣𝝂e ﹣𝝂𝜇 ﹣𝝂𝛕


Neutrino Oscillation with a ‘sterile’ neutrino● Anomalies* in short-baseline neutrino experiments in

the last decades motivated to hypothesize a fast oscillation Δm2

41~1eV2.

10

Electroweak boson

PMNS mixing matrix extended with 1 sterile flavor

LSND anomaly : data excess in reconstructed 𝝂𝜇 p → e+ n candidates.MiniBooNE anomaly : data excess in reconstructed e- (e+) -like events in 𝝂𝜇 (𝝂𝜇) mode.

● “Sterile”: non-weakly-interacting, however participating in the non-trivial mixing through PMNS matrix.


Sterile neutrino search in SBN program

● SBN will be the most sensitive experiment for a light sterile neutrino search.

● Current globally allowed region is significantly covered by 5𝜎 curve.

- Set a strong exclusion limit on the parameter space

- Possibly discover a light sterile neutrino

11

doi:10.1146/annurev-nucl-101917-020949


Physics in MicroBooNE● LSND/MiniBooNE low energy excess

(LEE) investigations○ Electron contribution○ Photon contribution

● Cross-section measurements

● Exotic physics searches (e.g. heavy neutral lepton, neutron-antineutron oscillation)

● Higgs Portal scalar bosons search using NuMI beam

● Supernova burst detection using continuous readout

12

Reconstructed neutrino energy distribution in MiniBooNE https://doi.org/10.1103/PhysRevLett.121.221801

SN1987A

Heavy neutral lepton

Neutron-antineutron oscillation


Deep Underground Neutrino Experiment (DUNE)

● Next generation LArTPC.

● Near detector located @ Fermilab, far detector located @ SURF, South Dakota.

● Far detector, 1.5km underground, will have 4 modules with 10 kton fiducial volume each.

13


Physics in DUNE

14

Crab nebula

● Beam (neutrino oscillation physics) :

- Testing CP violation in lepton sector (δCP=0?)

- Determination of neutrino mass ordering, θ23 octant

● Off-beam :

- Investigating supernova neutrinos, solar neutrinos

- Atmospheric neutrinos, cosmic rays

- Exotic physics searches (proton decay,

neutron-antineutrino osc., etc.)

arXiv:2007.14792

https://arxiv.org/abs/2007.14792


Leptonic CP sensitivity in DUNE in the presence of light sterile neutrino

15



16

[DUNE Technical Design Report]

● CP violation

○ One of the Sakharov’s conditions

○ Main physics objective in DUNE

https://arxiv.org/pdf/2002.03005.pdf



17


● CP violation

○ One of the Sakharov’s conditions

○ Main physics objective in DUNE

● What if there is a sterile neutrino?

○ Does it have an impact on DUNE’s CP

violation sensitivity?



Sterile mixing parameters and CP phases

● 3 neutrino PMNS matrix can be parametrized with 3 mixing angles and one CP phase.

● One sterile neutrino brings in 3 extra mixing angles and 2 extra CP phases.

18

PMNS mixing matrix extended with 1 sterile flavor


Ambiguities in oscillation probability

19

3+1ν, 0<δCP, δ14<2𝛑3+0ν, 0<δCP<2𝛑

⭐

● Bi-probability plot of oscillation probabilities at DUNE far detector● P(anti-νμ → anti-νe) vs. P(νμ → νe) at E=3GeV, L=1300km● Additional mixing parameters (on the right) bring out ambiguities in oscillation probability


Toy MC simulation for DUNE

20

● Toy MC was simulated for the cases with○ 3𝝂 and CP conversing / violating ○ 3+1𝝂 and CP conserving / violating

● 3𝝂 (δCP = 𝛑/2), 3+1𝝂 (δCP = 0) both have impact towards the excess in the distributions.

δCP = 𝛑/2

νe CC far detector, 3𝝂 (δCP = 𝛑/2)

3+1 ν, δCP = 0

νe CC far detector, 3+1𝝂 (δCP = 0)


DUNE’s CP violation, mass ordering sensitivities

21

● The simulated results of 3 neutrino model in DUNE show CP violation sensitivity over 5𝜎 for maximally violating phases, sensitivity of mass ordering over 5𝜎 over all δCP.

● Comparable to DUNE’s Conceptual Design Report (CDR).

CP Violation Sensitivity Mass Ordering Sensitivity



22

● If we evaluate CP sensitivity using globally allowed 3+1 parameters, under the 3 neutrino hypothesis, the sensitivity can vary considerably.




23

● If SBN sets a limit on the sterile neutrino parameter, these uncertainties in CP violation and mass ordering sensitivity can be shrunk to the level of retainment of CP violation sensitivity over 5𝜎 at maximally violating δCP.



If SBN determines a sterile neutrino with a precision

24

Dm2 = 1.7eV2,δ14 = [-pi, pi],

δ34 = 0, 𝜃23 = [38, 53],𝜃14 = 8.3°,𝜃24 = 6.9°,𝜃34 = 10°

CP Violation Sensitivity ● If SBN discovers a sterile neutrino, leptonic CP violation sensitivity varies as a function of δ14 ,δ34.

● Precision measurements on 𝜃14, 𝜃24, 𝜃34 will be prerequisites.


Leptonic CP sensitivity in DUNE in the presence of light sterile neutrino: Summary

25

● Next steps are updating 3+1 global region and doing more rigorous exploration in

3+1 parameter space using the correlated 𝛘2 fit for SBN and DUNE.


Neutron-antineutron (n-n̅) oscillation search in DUNE and MicroBooNE

26


Theoretical motivation of searching for n-n̅

● n-n̅ oscillation: n spontaneously converts itself to n̅

● Baryon number violation (BNV)

○ n-n̅ oscillation violates baryon number by 2

units (|ΔB|=2); this process could account for

the baryon dominance in the universe.

○ n-n̅ oscillation is predicted by a number of

beyond-Standard Model theories.

27


● Neutral particle-antiparticle oscillations through electroweak interaction

○ , oscillations

● Neutral fermion oscillation through non-trivial mixing○ Neutrino flavor oscillations

● BSM physics○ n-n̅ oscillation

Neutral particle oscillations

28

doi:10.1103/PhysRevD.87.075004

https://arxiv.org/ct?url=https%3A%2F%2Fdx.doi.org%2F10.1103%2FPhysRevD.87.075004&v=55a42ada


n-n̅ oscillation searches with free neutrons

● n-n̅ search in free neutrons

- Look for an antineutron in propagated neutron beam

- Current best n-n̅ oscillation lifetime limit using free neutrons: 8.6 × 107 s by ILL/Grenoble

experiment.

- Future project HIBEAM/NNBAR is planned at the European Spallation Source (ESS).

29


n-n̅ oscillation searches with bound neutrons

n-n̅ search in bound neutrons

- Look for antineutron’s annihilation signature in a nucleus.

- The bound n-n̅ oscillation lifetime ( ) is related to free n-n̅ oscillation lifetime ( ) by

intranuclear suppression factor R, as follows:

- Suppression factor for 40Ar was recently calculated ( R = 5.6 × 1022 s-1 ). [PRD.101.036008]

- Super-K holds the current best n-n̅ oscillation lifetime limit for oxygen-bound neutron:

3.6 × 1032 yrs, equivalent to 4.7 × 108 s for free n-n̅ oscillation lifetime. [New result at Neutrino 2020]

30

https://link.aps.org/doi/10.1103/PhysRevD.101.036008

https://nusoft.fnal.gov/nova/nu2020postersession/pdf/posterPDF-43.pdf


n-n̅ oscillation signature

● Multiple-pion final state:

- In 40Ar, n-n̅ oscillation is followed by the annihilation with a nearby nucleon (p or n).

- (n̅, n), (n̅, p) annihilation generates multiple pions.

- Momentum sum ~0 GeV, total energy release ~2 GeV

- The topology of decay products has spherical symmetry with high pion multiplicity:

n-n̅ ‘star-like’ signature topology.

31

Discovery of the antiproton, 1955Photographic emulsion image at Bevatron.

Example neutron-antineutron oscillation signature. Fermilab-thesis-2017-27

http://old.inspirehep.net/record/1662648


n-n̅ oscillation search in the DUNE or MicroBooNE

32

● Deep underground, well-shielded from cosmogenic activity.

● Anticipated backgrounds: atmospheric neutrino interactions

● Very large detector mass and long exposure: 4 × 10 kton TPC volume, 10 years of exposure.

● Assumes self-triggering on rare events like proton-decay, n-n̅ oscillation, atmospherics.

● On-surface operation● Background-dominated: constant

stream of cosmic ray muon tracks● Very small detector mass (500x smaller

than DUNE), and shorter exposure: 80 t TPC volume, ~hrs of exposure.

● Random trigger for off-beam readout.When the cosmic ray muons rain, they pour … in the MicroBooNE liquid-argon time projection chamber.source: https://news.fnal.gov/2018/03/when-it-rains-2/

DUNE MicroBooNE

MicroBooNE DetectorAtmospheric

neutrino production DUNE far detector Cosmogenic activity typical

of any recorded MicroBooNEevent


Analysis overview: ML based Method

33

● Analysis combines traditional reconstruction with image classification: A Boosted Decision Tree (BDT) combines:

○ traditionally reconstructed kinematic information, with

○ image classification score obtained from a Convolutional Neural Network (CNN)

● Alternatively:○ CNN-only analysis○ BDT-only analysis

* This analysis scheme can be shared between DUNE and MicroBooNE which gives a unique opportunity of validation on MicroBooNE data.

Example input wire plane image



34







● Image classification using Convolutional Neural Networks○ Image inputs are mapped to feature maps after every convolution layer○ Final layer shows the scores of the input image



35









36







Number of tracks > N1

Number of showers > N2

Number of showers > N3

n-n̅ background n-n̅ background

YES NO

YES YESNO NO

● Decision tree is a flow of criteria to make a decision.● Boosted decision tree builds strong classifier using optimizing the

weights of ensemble of decision trees.



37








BDT input variables in DUNE

38

Input BDT variables include:- CNN score - Number of tracks- Number of showers- Track-like visible Energy- EM-like visible Energy- Visible energy- Longest track particle ID- Longest track Momentum- EM-like visible Energy fraction- Most energetic shower dE/dx

(atmospherics)( )




Selection cut and preliminary sensitivity in DUNE

39

● Using an optimized combined BDT cut, 8% signal efficiency with 0.02% background efficiency can be achieved.

● For 400 kt-years exposure, ○ ~20 background events are expected.○ ~30 signal events are expected,

when = 3.6 × 1032 yrs (current best limit at Super-K) is assumed.

( )

(atmospherics)







40

● The 90% C.L. sensitivity for argon-bound n-n̅ oscillation lifetime corresponds to 6.45 × 1032 yrs, for an exposure of 1.33 × 1035 neutron-years in DUNE.

○ This translates to a free n-n̅ osc. lifetime of 5.53 × 108 s.○ The result is comparable with the current best limit at Super-K (3.6 × 1032 yrs for

bound, 4.7 × 108 s for free n-n̅ osc. lifetime).○ The effort on further improving the analysis and sensitivity continues.

● 25%, 25%, 3% of uncertainties for the signal efficiency, background efficiency, and exposure were assumed.

● A systematic error evaluation due to the choice of generators and intranuclear modeling is being established.


Yeon-jae Jwa | SLAC experimental seminar series (Nov. 24th 2020) 41

● With relatively small LArTPC and shorter exposure, MicroBooNE will not push the current existing limit. Still, it can serve as a testbed to carry out this analysis method on the real LArTPC data, while the study in DUNE is only available through simulation.

● The analysis can be developed using○ MC prediction [MicroBooNE-PUB-1093]○ Data-driven prediction (on-going)

n-n̅ oscillation search analysis in MicroBooNE

https://microboone.fnal.gov/wp-content/uploads/MICROBOONE-NOTE-1093-PUB.pdf


MC analysis on MicroBooNE Simulations

● The CNN-only, the BDT-only, the Combined methods have been demonstrated on MC events: cosmogenic background with signal overlays, and cosmogenic background only.

42

CNN-only BDT-only

Signal efficiencies are62.3%, 90.5%, 90.7% for BDT, CNN, Combined at 95% background rejection

[MICROBOONE-NOTE-1093-PUB]

https://microboone.fnal.gov/wp-content/uploads/MICROBOONE-NOTE-1093-PUB.pdf


● A small subset of MicroBooNE Run1 off-beam data was used to validate the CNN-only and BDT-only methods.

○ The off-beam data is expected to follow the CNN/BDT responses of cosmic MC, since negligible signal is expected in the data given existing limit.

○ Data-MC disagreement is evident, understood to be due to imperfect modeling of cosmic flux and/or detector response.1

○ This illustrates challenges in applying traditional reconstruction & ML techniques developed only on simulation.

43

100% MC analysis validation on MicroBooNE Off-beam Data

CNN-only

BDT-only

1Other MicroBooNE analyses currently make use of data overlays to emulate cosmogenic backgrounds and better approximate detector effects such as electronics noise.



Data-driven cosmic simulation for MicroBooNE analysis

● Data-driven comic simulation has shown to improve data-to-MC agreement in other MicroBooNE analyses.

● Using data-driven cosmic simulation, ○ Take reconstructed 3D clusters from cosmics

only and cosmics with simulated n-n̅ oscillation○ Label the 2D projections of cosmics and n-n̅

oscillation clusters○ Train CNN classifier to tell n-n̅ oscillation from

cosmic cluster

Y (vertical)

X (drift)

Z (beam)

n-n̅cosmics

MicroBooNE active TPC


Neutron-antineutron (n-n̅) oscillation search in DUNE and MicroBooNE: Summary

● The sensitivity study for a search for baryon number violating n-n̅ oscillation was

studied in DUNE, projecting a world’s best sensitivity.

● In MicroBooNE, the first ever search for n-n̅ oscillation in LArTPC is on its way

aiming at a finalized analysis in Spring 2021.


Machine learning based rare physics triggering in DUNE and FPGA implementation

46


DAQ challenge in large scale LArTPCs

47

● DUNE far detector is much larger in scale than current LArTPCs.

- 384,000 channels per 1 far detector module (e.g. 8,256 channels in MicroBooNE)

● Continuous self-triggering on supernova neutrinos, rare physics (p-decay, n-n̅ oscillation, atmospheric neutrinos, cosmics)

- ~3-4 TB/s raw data rate before triggering

- Total allowance: 30 PB/year for offline permanent storage

- Rate reduction factor of 104 is required.

● Readout unit:

- FELIX(Front-End LInk eXchange) board with FPGA

- CPU for data processingFELIX-712 arXiv:1806.10667

https://arxiv.org/abs/1806.10667


Machine-learning based self-triggering on supernova burst and rare physics

48

● ML-based data selection with raw readout processing on GPUs - Meets the data rate requirement of DUNE, achieving ~70% efficiency for single supernova neutrino (LE),

~90% for high-energy neutrino (HE).● This result can be passed to higher level supernova triggering scheme.

- Galactic supernova coverage ~100% was achieved.

Example raw readout for high-energy neutrino (HE).

Example raw readout for low-energy supernova neutrino (LE).

Example raw readout for noise backgroun only (NB).


Downsizing for FPGA implementation

49

*Y. Jwa, G. di Guglielmo, L. Carloni, G. Karagiorgi;

DOI: 10.1109/NYSDS.2019.8909784

● Using down-sized input & smaller network, the inference time on GPU reached target inference time < 2~3 ms.

● FPGA inference showed 4 times higher energy efficiency than CPU.

Classification performance shown with GPU support with 64x64 input, single convolution network.

(NB: noise background only, LE: low-energy supernova neutrinos,

HE: high-energy neutrinos)

FPGA power consumption and timing for single convolution network (CNN_s)

showing x4 energy efficiency than CPU.

https://doi.org/10.1109/NYSDS.2019.8909784


● High Level Synthesis: a process where an algorithmic description for a task is implemented on a hardware.

● hls4ml: Designed tool developed by particle physicists and computer scientists to enable implementations of ML algorithms on FPGAs.

- Highly configurable to manage resource usage- Sparsifying networks- Quantization-aware training using QKeras

● Testing the Python API using DUNE simulated input images.- Performance evaluation on FPGA is on-going.

50

CNN implementation on FPGA using HLS4ML:High Level Synthesis (HLS) for Machine Learning


Conclusion● Exciting time for neutrino physicists!

○ Investigating anomalies, searching for new physics, R&D opportunities

Clues for a Matter Dominated Universe?○ Phenomenological evaluation of CP violation sensitivity for DUNE.○ ML-based neutron-antineutron oscillation analysis in DUNE and

MicroBooNE (ongoing).○ Development of a ML-based rare physics trigger for LArTPCs, FPGA

implementation (ongoing).

51


Thank you!


Back up


doi:10.1146/annurev-nucl-101917-020949


3+1 Parameter space

55

Δm412 = 1.7eV2,

δ14 = [-𝛑, 𝛑],δ34= 0,

𝛉23 = [38°, 53°],𝛉14 = [5.7°, 12.3°],𝛉24 = [3.65°,12.93°]

𝛉34 = [0°, 25°]3+0 fit


CP violation sensitivity calculation

For 3+0 ν, our CPV sensitivity is close to DUNE CDR.

56

DUNE Collaboration CDR, arXiv:1512.06148

Yeon-jae Jwa Columbia University


DUNE far detector spectra (3+0 ν, δCP = 𝛑/2)

● Maximally violating CP

● CPV has large effect on signal event rate in νe CC and anti-νe CC

57Yeon-jae Jwa Columbia University

δCP = 𝛑/2 δCP = 𝛑/2

δCP = 𝛑/2δCP = 𝛑/2

νe CC far detector

anti-νe CC far detector

νμ CC far detector

anti-νμ CC far detector


3+0 δCP = 𝛑/2, near detector

58Yeon-jae Jwa Columbia University


Super-fast MC simulation for DUNE

● Neutrino events for DUNE are generated by Genie MC generator for the given beam flux distribution ranging 0 to 40 GeV.

● Signatures of neutrino oscillation can be found in the appearance channel (νμ → νe), the disappearance channel (νμ → νμ) in DUNE.

● For both channels, detector effects and event selections are applied.

59

T. Alion, et al. arXiv:1606.09550


νe CC far detector

Yeon-jae Jwa | SLAC experimental seminar series (Nov. 24th 2020)60

Mass OrderingCP Violation (Inverse Ordering)

Within globally allowed 3+1 region, DUNE’s sensitivities in CPV and mass ordering can be ranging with large uncertainty.

CPV, mass ordering sensitivities in DUNE Inverse ordering, 3+1 global

Δm412 = 1.7eV2,

δ14 = [-𝛑, 𝛑],δ34= 0,

𝛉23 = [38°, 53°],𝛉14 = [5.7°, 12.3°],𝛉24 = [3.65°,12.93°]

𝛉34 = [0°, 25°]3+0 fit


SBN sensitivity on light sterile neutrinos

● SBN program is estimated to achieve 5σ sensitivity over a large percentage of parameter spaces in 3+1 scenario.

● If SBN’s result favors 3+0 neutrino model, it will rule out large 3+1 parameter space.

61

D. Cianci et al., Phys. Rev. D 96, 055001 (2017)


SBN SENSITIVITY COVERAGE OVER 3+1 PARAMETER SPACE


● Codes, frameworks are in place to fit SBN and DUNE spectra side-by-side and perform a combined 3+1 sensitivity study.

● What is completed.- 𝛘2 fit was performed using coarse grid of 3+1 oscillation parameters, with global

region provided by MIT neutrino group.- 3+1 hypotheses with 99% global region were fit to 3 neutrino ground truth.- 3+1 hypotheses with 99% global region were fit to 3+1 neutrino best-fit ground truth.

● What is next.- Updating 3+1 global region including current experimental results.- 3+1 hypotheses fitting 3+1 hypotheses:

‣ This requires more computation power. (O~n2)‣ Nevis condor nodes will be utilized for this computation.

62

Phenomenology Study:CPV/MO sensitivity evaluation with 3+1 hypothesis for DUNE



63

● Using an optimized combined BDT cut, 8% signal efficiency with 0.02% background efficiency can be achieved.

● For 400 kt-years exposure, ○ ~20 background events are expected.○ ~30 signal events are expected, when = 3.6 × 1032 yrs

(current best limit at Super-K) is assumed.● In calculating the 90% C.L. sensitivity in DUNE,

○ Given exposure (E), signal efficiency ( ), background counts (b)○ Bayesian posterior distribution of decay width (Γ):

Poisson distribution in combination with Gaussian priors yields

( )

(atmospherics)






DUNE upstream DAQ

64

Documents

Searching for Clues for a Matter Dominated Universe in LArTPCs