Theory uncertainties at HL-LHC

�1

Osamu Jinnouchi(TokyoTech)

Beyond the BSM (The 4th Kavli IPMU-Durham IPPP-KEK-KIAS workshop) 2018 Oct 1 - 4

�2HL-LHC plan & statistical impact

Most of the current papers

x 5 stat.available in next spring

x 10 stat.full LHC

x 100 stat.full HL-LHC

- 4000

0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1b-jet efficiency

1

10

210

310

410

u-je

t rej

ectio

n

|<1η||<2η1<||<3η2<||<4η3<|

|<2.5η||<2.5)ηRun 2 (|

2mμ 50*50

MV2

Inclined Duals

ATLAS Simulation s=14 TeV, tt, <µ>=200

�3Performance of the key objects

▪ Pile-up <μ>(2018) ≃50 → <μ>(HL-LHC)≃>200

▪

▪ threshold tuning, algorithm improvement

▪ keep the performance on tracking

▪ b-tagging: ▪ owing to the Inner tracker replacement,

b-tagging keeps its performance, even better than Run-2

▪ Trigger▪ increase the overall budget 10kHz (Run-2:

1kHz)▪ improve the BG rejection new hard/software

to cope with high rate while keeping the low threshold

ETmiss

major deterioration in performance not foreseen in HL-LHC

[GeV]T EΣ0 500 1000 1500 2000 2500 3000 3500 4000 4500

[G

eV]

mis

sx,

yR

MS

E

0

2040

6080

100

120140

160

|<4.0, ReferencepTR

η|<4.0, |soft track

η||<3.2, Middle

pTRη|<3.2, |

soft trackη|

|<2.7, LowpTR

η|<2.7, |soft track

η|

ATLAS Simulation =190-210µ=14 TeV, stPowhegPythia t

>0.1pTR

����� ! ≃ 25��������������Run-2 tt ⟨μ⟩≃25arXiv: 1802.08168

ITk Pixel TDR CERN-LHCC-2017-021

Phase-II upgrade scoping documentCERN-LHCC-2015-020

|η|<4.0

Run2

HL-LHC

�4

▪ HL-LHC▪ with 100 times statistics, things look different▪ needs for even higher order corrections▪ can discuss on the detailed shape of the distribution tail▪ can deal with very rare events

▪ uncertainties not a big concern so far, become major contributions

▪ “theory uncertainties” could be the one! we should start thinking about reducing them, with the help of huge stats from LHC

Introduction

disclaimer : I consulted with the following documentsATL-PHYS-PUB-2018-010: impact of theory uncertainties related toHiggs boson production in the H→ZZ*→4l and the ttH channels withthe ATLAS detector at the HL-LHC“QCD issues in searches for New Physics with the ATLAS detector”Mattias Saimpert, QCD@LHC2018 presentation JPS 2018 fall : Yuji Yamazaki (Kobe U.) review talk

�5

[1] Higgs physics at HL-LHC▪ Understanding of the Higgs potential

▪ Discovery & measurement of the Higgs trilinear self-coupling to explore the shape of the Higgs potential

▪ promising channels:

▪ Extremely rare, very challenging

▪ Yukawa-couplings for lighter (2nd) generations need to be covered

[2] Searches of the BSM▪ Precisions on the background & signal

estimations are vital

▪ Severe environment at HL-LHC

Theoretical uncertainties are the key for the success of HL-LHC

Important Physics topics at HL-LHC

HH→bb +bb, bb +γγ , bb +τ +τ −

ATL-PHYS-PUB-2017-001

ATL-CONF-2018-031

[1] Theory Uncertainties in Higgs physics

�6

�7

▪ Normalization and modeling ▪ both for signal & backgrounds

Theory uncertainties

topic channel reference dominant theoretical systematicsBSM A/H tautau JHEP 01 (2018) 055 Tau fake estimates, embedding of ZCombination Couplings ATLAS-CONF-2018-031 Signal modelingsDiBoson hWW arXiv: 1808.09054 WW modelingyukawa VHbb arXiv:1808.08238 V+jets modelingyukawa Htautau ATLAS-CONF-2018-021 H pT modelingHH 4b arXiv: 1804.06174 multi-jet shapeμ yukawa Hmumu PRL119(2017)0418002 Drell-yan modelingtop yukawa ttH PRD97 (2018) 072003 tt+V modeling

Ex) Higgs related analysis

▪ For the HL-LHC physics projections, the 1/2 of Run-2 estimation is used ▪ improvements in higher-order corrections, resummation ▪ inputs from experiments would improve the modeling ▪ PDF reduction expected with the help from theory community (?)

8− 6− 4− 2− 0 2 4 6 8

Jet energy scale variation 1

Non-prompt stat. in 5th bin of 2LSS SR

Diboson modeling

Non-prompt stat. in 3rd bin of 3L SR

electron identification

ttZ modeling (shower tune)

ttH modeling (shower tune)

rare top decay cross section

pileup reweighing

Luminosity

ttZ cross section (scale variations)

ttH cross section (PDF)

Non-prompt stat. in 4th bin of 3L SR

four top production cross section

Jet energy scale (pileup subtraction)

ttH modeling (parton shower)

Jet energy scale (flavor comp. 2LSS)

ttZ modeling (generator)

identificationhadτ

ttH cross section (scale variations)

0.1− 0.05− 0 0.05 0.1ttHµ∆

1.5− 1− 0.5− 0 0.5 1 1.5θ∆)/0θ - θ(

Pull1 standard deviation

ttHµPrefit Impact on

ttHµPostfit Impact on

ATLAS SimulationPreliminary

-1 = 14 TeV, 3000 fbs

�8

▪ Due to unexpectedly rapid progress in analyses, e.g. multi-variates, optimizations in analysis/performance, or conservative prospects (?) in 2014, some of them are already reached by the current precisions

▪ More accurate treatments of systematics will be available, after we obtain good amount of statistics from HL-LHC

Uncertainties in Yukawa coupling measurements

ttZ modeling (generator)

ttH modeling (parton shower)

ttH cross section(PDF)

ttZ cross section(scale variation)

theory uncertaintyranking

Projection at 2014 (ATLAS-PHYS-PUB-2014-016)

Run-236 fb-1

Run-125 fb-1

Run-280 fb-1

arXiv:1808.08238

ttH→multi− leptons ttH cross section(scale variation)

ATL-PHYS-PUB-2018-010

�9

▪ signal strength measurement ▪ σtheory has direct impact on the result

σsys projection in ttH(→multi-leptons)

µ =σ signal

σ theory

∝ y f

ttH model (inclusive)affects σ(theory)- missing HO calc.- αs- PDF

Stats no longer an issue at HL-LHC

fake objects

ATL-PHYS-PUB-2018-010

ttH model (acceptance)affects σ(signal)- parton shower model

�10

In HL-LHC, high-x PDF are important▪ high-x gluon PDF▪ jet and top pair production data are inputs▪ NNLO+NNLL for tt▪ high-x quark PDF▪ high mass DY: NNLO pQCD+NLO EW

LHC PDFs & towards HL-LHC

F. Giuli (Oxford)

ATL-PHYS-PUB-2018-017A new ATLAS PDF set: ATLASepWZTop18Based on the NNLO analyses of W/Z, top, ep data from HERA

0.2

0.4

0.6

0.8

1

-310 -210 -110 1

ATLASepWZtop18 NNLO uncertainties: experimentalmodelparameterisation

epWZ

x

xf 2= 10 GeV2Q

vxu

vxd

0.05)!xg (

ATLAS Preliminary

LHC measurements for PDF fits

[2] Theory Uncertainties in BSM searches

(from Run-2 analyses)

�11

�12

1) Uncertainties from fakes (above plots) 2) Uncertainties in Signal & background predictions ▪ matrix element : QCD order, Nparton

▪ parton shower model, ISR/FSR tune▪ PDF/αs variations, flavor scheme, etc

QCD process backgrounds

10 210 310 [GeV]

T p

00.05

0.10.150.2

0.250.3

0.350.4

0.450.5T

/pσ R

elat

ive

reso

lutio

n ATLAS Simulation Preliminary

Jets=0.0η|=2.0η||=4.5η|

=0)φ (µCB =0.0η|=2.0η|

Electrons=0.0η|=2.0η|

γN.C. =0.0η|=2.0η|

τHadr. 1p0n3pXn

|12

ηΔ|0 0.5 1 1.5 2 2.5 3 3.5 4

Frac

tion

of e

ntrie

s

0

0.05

0.1

0.15

0.2

0.25

0.3

DataMultijet (Pythia8)Multijet (Herwig++)Multijet (Sherpa)

= 450 GeVχ∼

= 1000 GeV, mg~m = 450 GeV

χ∼ = 2000 GeV, mg~m

ATLAS-1 = 13 TeV, 36.1 fbs

fake MET through jet resolution effectsATLAS-CONF-2018-038

fake signal topology from combinatoricsPhys. Lett. B 785 (2018) 136

Impact all LHC physics

�13

1) Define control regions (CR)

▪ background dominate regions

▪ corresponds to background source

2) Normalize MC background prediction in the CR

▪ simultaneous fit to data in all CR

▪ normalization uncertainties ~ data stats

3) Cross-check extrapolation quality in validation region (VR)

4) Extrapolate to signal region (SR) normalized by Transfer Factor (TF) Big impact from QCD modeling uncertainties on MC

• highest order MC available is used for prediction• evaluated by varying ME, PS, ISR/FSR tune, PDF/αs, flavor scheme

Background estimation strategy in non-resonant searches

Eur. Phys. J. C75 (2015) 153

�14

▪ [ fundamental assumption ] high MET arise from fluctuations in the detector response to jets

▪ Method:1) multi-jet event selection with good jets

(low MET events)2) generate pT response function (RF) based

on MC as a starter 3) jet pT smeared according to expected pT

response, generate pseudo-events, compared to data, modify RF, repeat this

4) generate variables based on final RF

▪ Method limitations in future use

▪ non-gaussian tail (large uncertainties)

▪ fake MET from pileup jet removal algorithm inefficiency, not implemented

(I) Jet smearing method

0 0.5 1 1.5 2 2.5 3

/32

πEv

ents

/

-110

1

10

210

310

410 -1L dt ~4.7 fb∫ ATLAS

SR-C Loose

= 7 TeV)s Data 2011 ( Pseudo-data + non-QCD MC Pseudo-data W+jets Z+jets

and single topt t multijet

[rad.] min

)TmissP(jet,φΔ

0 0.5 1 1.5 2 2.5 3

Dat

a/Es

timat

e

00.5

11.5

22.5

Multijet validation regionSUSY strong 0-lepton analysisPhys. Rev. D 87, 012008 (2013)

200 300 400 500 600 [GeV]

TpLarge-R jet

0.00

0.01

0.02

0.03

0.04

0.05

Frac

tiona

l JES

unc

erta

inty

Syst.⊕Stat. -resolutione E -scalee EPile-up (JVT) MC modelling -resolution ID Eµ

-resolution MS Eµ -scale (charge) residual Eµ -scale (charge) Eµ-scale Eµ Pile-up (NPV shift) Statistical

Sub-leading jet veto φΔ

+jetZ, -1 = 13 TeV, 36.2 fbs=1.0 (LCW+JES+JMS)R tkTrimmed anti-

| < 0.8Jet

η|

ATLAS

�15(II) large-R jet substructure uncertainties

• R=1.0 anti-kT jet to reconstruct high pT W, top• QCD modeling crucial : energy scale, substructure, etc

L. Skinnari

[GeV]Tp Jet RLeading large-

500 1000 1500 2000 2500

)bk

g∈

Back

grou

nd re

ject

ion

(1/

5

10

15

20

25

30

35

40

45

[GeV]Tp Jet RLeading large-

500 1000 1500 2000 2500

MC

Dat

a

0.50.75

11.251.5

Data - Sig.Pythia8 dijetHerwig++ dijetStat. uncert.

syst. uncert.⊕Stat.

Data - Sig.Pythia8 dijetHerwig++ dijetStat. uncert.

syst. uncert.⊕Stat.

ATLAS Preliminary-1 = 13 TeV, 36.7 fbs

=1.0R tkTrimmed anti-Dijet Selection

23d + 32τ = 80%): sig∈Top tagger (

ATLAS-CONF-2017-064 arXiv:1807.09477

QCD jet rejection in Top tagger

sys. > stat.

large-R jet Energy scale calibration uncertainties

�16

▪ dominant systematics from ▪ ISR modeling in tt events→ difference btw MCs▪ multi-jet background

a) Impact on Stop 0L search (low Δm)

[GeV]1t

~m200 400 600 800 1000 1200

[GeV

]0 1χ∼

m

0

100

200

300

400

500

600

700

800

900

SRA+SRB+SRC+SRD+SRE

01χ∼

+ m

t

< m

1t~m01χ∼

+ m

b

+ m

W

< m

1t~m) = 100%0

1χ∼ (*) t→ 1t

~Top squark pair production, B(

-1=13 TeV, 36.1 fbs

ATLAS )theorySUSYσ1 ±Observed limit (

)expσ1 ±Expected limit (

=8 TeVs, -1ATLAS 20 fbLimits at 95% CL

compressed region: top mass cannot be reconstructedhard ISR jet required

direct stop pair productionJHEP 12 (2017) 085

RISR : 0.3-0.4 0.4-0.5

dominant systematics from theory

�17

▪ dominant systematics from ▪ tt events modeling ▪ JER, JES, b-tagging

a) Impact on Stop 0L search (high Δm)

[GeV]1t

~m200 400 600 800 1000 1200

[GeV

]0 1χ∼

m

0

100

200

300

400

500

600

700

800

900

SRA+SRB+SRC+SRD+SRE

01χ∼

+ m

t

< m

1t~m01χ∼

+ m

b

+ m

W

< m

1t~m

) = 100%01

χ∼ (*) t→ 1t~Top squark pair production, B(

-1=13 TeV, 36.1 fbs

ATLAS )theorySUSYσ1 ±Observed limit (

)expσ1 ±Expected limit (

=8 TeVs, -1ATLAS 20 fbLimits at 95% CL

t̃

t̃

t

tp

p

χ̃01

b

W

χ̃01

b

W direct stop pair productionJHEP 12 (2017) 085

�18b) multi large-R jetsSearch for massive supersymmetric particles in multi-jet final states

g̃

g̃

�̃01

�̃01

p

p

q q

�00

q

qq

q q

�00

q

qq

RPV SUSY gluino scenarios

fully hadronic, no MET requirement many jets (≥ 4/5 large-R jets, b-tag categories)

total jet mass as main discriminant

[TeV]ΣJM

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

Frac

tion

of e

ntrie

s

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

DataMultijet (Pythia8)Multijet (Herwig++)Multijet (Sherpa)

= 450 GeVχ∼

= 1000 GeV, mg~m = 450 GeV

χ∼ = 2000 GeV, mg~m

ATLAS-1 = 13 TeV, 36.1 fbs

Phys. Lett. B 785 (2018) 136

main background : multi-jet, combinatorics → huge uncertainties

→ fully data-driven estimates (next page)

�19

▪ the SM mass template obtained in CR ( JHEP 05 (2014) 005 ) ▪ estimate uncertainties by varying CR definition▪ background estimate uncertainties, ~20%, have a large impact

on the final results

b) multi large-R jets

7− 6− 5− 4− 3− 2− 1− 0)

Tp/m(

10log

5−10

4−10

3−10

2−10

1−10

Frac

tion

of E

ntrie

s

ATLAS = 13 TeV,s

-1 36.1 fb

< 644 GeVTp600 GeV <

| < 1.0η0.5 < |

Pythia8 non-matchedPythia8 b-matchedData non-matchedData b-matched

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2Ev

ents

1

10

210

310

410 ATLAS-1 = 13 TeV, 36.1 fbs

| < 1.4ηΔ| 5≥ jetN

5jSR

dataσ1 ±bkg pred

σ1 ±Syst. = 2000 GeVg~m = 50 GeV0

1χ∼m = 1900 GeVg~m = 1050 GeV0

1χ∼m

SR

[TeV]ΣJM

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2Dat

a/bk

g pr

ed

00.5

11.5

2

jet mass template data vs. MC

background estimates in SRPhys. Lett. B 785 (2018) 136

�20

▪ Search for the long-lived gluinos (Split-SUSY scenarios)

c) displaced vertex search

Cut and count analysis- Missing ET > 250 GeV- >= 1 Displaced vertex

- mDV>10GeV- ntrack>=5

- No SM irreducible background expected - Experimental (detector) backgrounds

Phys. Rev. D97 (2018) 052012

detailed material map needs to be knownto suppress hadronic interaction BG non-standard reconstruction

�21

▪ data driven background estimate

c) displaced vertex search

merged vertex

CR

VR

SR 5-track(4+1)-trackSRBackground prediction

random crossing BG estimation

however dominant bg comes from the hadronic interactionthere are gases in the detector gap regionsprediction in SR is extrapolated from low mass region, exposhape assumed → -100 ~ 300%

need to revisit BG modelings

random crossing(high mass region)

hadronic interaction(low mass region)

�22

▪ Towards HL-LHC, people in experimentalists currently work very hard on the hardware upgrades and physics prospects analyses

▪ Due to the lack of hints/evidences of the new physics yet, we would foresee a hard time in keeping the strong momentum during the lifetime of HL-LHC (budget, attracting younger generation…)

▪ Precision measurements of Higgs couplings, searches of SUSY in even higher mass regime (>2TeV) are important and just fine, but we need something more

▪ I think, we (experimentalists) are desperate for new theoretical framework (like “SUSY” before the start of LHC) which guide us to a new paradigm in HL-LHC regime

Personal Remark (request for theory community)

�23

▪ Theory uncertainties will be major systematics in HL-LHC▪ systematics on signal/background predictions▪ estimation of the fake backgrounds

▪ These affect measurements (eg. Higgs) type analyses, as well as search (eg. BSM) type analyses

▪ Need to tackle this: looking for the new higher order predictions, new methods/modelings to have better descriptions of data with the simulations

▪ Cooperation between experiments and theorists is vital

Summary