XAIENCE 2019 - Seoul National University · 2019-11-09 · BDTB : Bagging BDTD : decorrelation + Adaptive Boost BDTF : allow usage of fisher discriminant for node splitting Various

Probing Trilinear Higgs Self–coupling at the HL-LHC

with Machine Learning

Collaborators : Prof. Kingman Cheung, Prof. Jae Sik Lee, Dr. Chih-Ting Lu, Dr. Jung Chang

XAIENCE 2019

Jubin ParkIUEP and Chonnam National University

2019. 11. 8. 서울대, 자연대 25-1

Standard Model(표준모형)

https://www.mpoweruk.com/standard_model.htm

http://factmyth.com/the-standard-model-of-particle-physics-explained/

Higgs boson

• Particle physics1. Validation of the Standard Model2. Symmetry breaking of the electroweak interaction3. Particle mass acquisition4. Scalar fields and extension of the Standard Model

• Cosmology1. Inflaton2. Nature of the universe, and its possible fates3. Vacuum energy and the cosmological constant

https://en.wikipedia.org/wiki/Higgs_boson

LHC (Large Hadron Collider)https://youtu.be/pQhbhpU9Wrg

Simulated Large Hadron Collider CMS particle detector data depicting a Higgs boson produced by colliding protons decaying into hadron jets and electron

https://en.wikipedia.org/wiki/Large_Hadron_Collider

The boson's existence was confirmed in 2012 by the ATLAS and CMS collaborations based on

collisions in the LHC at CERN.

1. Higgs couplings to fermions and gauge bosonshave been compatible with SM so far.

2. What happens for the Higgs self-couplings ?

?

Higgs In the SM

• Higgs field (ℎ) : responsible for ① the spontaneous EW symmetry breaking② the generation of masses of all the SM particle

• The potential is characterized by only two parameters :① vacuum expectation value 𝑣② the Higgs mass 𝑚𝐻

𝑣 =1

2 𝐺𝐹

≈ 246 GeV

Trilinear and Quartic Higgs boson couplingin the SM

New Physics can affect the Higgs potential form

Sizeable departures from the SM form

λ3 = λ3𝑆𝑀 + 𝛿λ3

𝑆𝑀 , λ4 = λ4𝑆𝑀 + 𝛿λ4

𝑆𝑀

Trilinear Higgs boson coupling𝝀𝟑 = λ3

𝑆𝑀 + 𝛿λ3𝑆𝑀

Therefore, it is really important to check (or directly measure) this trilinear Higgs coupling to verify that SM is correct.

What do we actually do ?

Outline of Simulations

2015 MadGraph school on Collider Phenomenology November 23-27 @ Shanghai

Parton Showering and Hadronization

Hard Scattering

Detector Simulation b-jet efficiency

c-jet efficiency

ATL-PHYS-PUB-2016-026

Outline of Simulations

2015 MadGraph school on Collider Phenomenology November 23-27 @ Shanghai

Parton Showering and Hadronization

Hard Scattering

Detector Simulation b-jet efficiency

c-jet efficiency

ATL-PHYS-PUB-2016-026

After Detector Simulation (DELPHES)

Analysis

Event selection criteria applied in the analysis

• Cut Based Analysis

• Machine Learning

Multi-Variate Analysis : BDT (Boosted Decision Tree)

(work in progress)

TMVA (Toolkit for Multivariate Data Analysis with ROOT)

Analysis Methods

A. Hoecker, P. Speckmayer, J. Stelzer, J. Therhaag, E. von Toerne, and H. Voss,

TMVA - Toolkit for Multivariate Data Analysis,

PoS ACAT 040 (2007), arXiv:physics/0703039

TMVA Inputs : 8 (kinematic) variables

𝑷𝑻_𝜸𝜸 , 𝑷𝑻_𝒃𝒃 ,𝑴𝜸𝜸 ,𝑴𝒃𝒃 ,𝑴𝜸𝜸𝒃𝒃 , ∆𝑹𝜸𝜸 , ∆𝑹𝒃𝒃 , ∆𝑹𝜸𝒃

BDT : Adaptive BoostBDTG : Gradient BoostBDTB : BaggingBDTD : decorrelation + Adaptive BoostBDTF : allow usage of fisher

discriminant for node splitting

Various ML methods in TMVAIdeal line of ROC curve

AUC = 1

A better direction

BDT-related methods give the best results

BDT machine optimized with

Signal

BDT (response)

cut

Results : Improved

Expected Yieldsand

Significance(Z)

significance Z = 1.05

∴ BDT-improved Z = 1.55

# of signal = 9.63

# of bg = 80.26

# of signal = 6.19

# of bg = 14.04

48% enhancement on Z

Significance of the signal over the background versus λ3𝐻 at the HL-LHC

+0.2 +7.1

95% CL

Conclusion

HL-LHC : constraint the

1. Cut-Based Analysis : -1.1 < λ3𝐻< 7.0 at 95% CL,

Z = 1.05 (λ3𝐻=1)

2. BDT Analysis + NLO_dist : 0.5 < λ3𝐻< 6.8 at 95% CL,

Z = 1.55 (λ3𝐻=1)48%

enhancement on Z

With 3000 fb−1

3. We find a bulk region of 0.5 ≤ λ3𝐻 ≤ 4.5, where it is hard to pin down the trilinear coupling.

• Combined analysis : 𝑏 ത𝑏𝑏ത𝑏 + 𝑏ത𝑏𝛾𝛾 + 𝑏 ത𝑏𝜏𝜏 → Z (≥ 3𝜎 )

• Advanced technology in the future …Increased luminosity, improved tagging efficiency, improved resolution and so on….

• More precise simulation, higher order QCD correction ..Improved MC Event generators (at the NLO, NNLO QCD order),QCD NLO, NNLO, NNNLO corrections …….

Last, but not least

ATL-PHYS-PUB-2018-053

Thank you for your attention !