P5 Science Drivers: Theory

Stefan Hoche

SLAC National Accelerator Laboratory

DOE Exascale Requirements Review (HEP)

June 10, 2015

P5 recommendations

[http://www.usparticlephysics.org/p5]

From the summary:

I Specific investments in particle accelerator,instrumentation, and computing research and developmentare required to support the program and to ensurethe long-term productivity of the field.

From the report:

I Computing cuts across all activities in particle physics, and these activitiesspur innovation in computing. The field played leading roles in developingand using high-throughput and distributed/grid computing, online [..] dataprocessing, high-performance computing, high-performance networking,large-scale data storage, large-scale data management and analysis, and theWorld Wide Web [..]

Theory computations will continue to increase in importance, ashigher fidelity modeling will be required to understand the data.

1

Traditional HPC: Lattice QCD

I LQCD is only means to extract SMparameters depending onnon-perturbative dynamics of QCD

I Ab-initio calculation in discretespace-time with lattice spacings downto ∼ 0.06 fm in 1443 × 288 hypercube

I Systematic errors from continuumextrapolation and chiral extrapolation,the latter lately reduced and soonto be eliminated

2

Interplay with experiment

I Observables include masses,hadron decay properties andbasic Standard Model parameters

I Direct interpretation of BaBar, CLEO,CDF, DØ, Belle, LHCb, BESIII andBelle II experimental data

I Most precise determination of strongcoupling from LQCD, needed in searchfor new physics through Higgs-bosondecays at ATLAS & CMS

10.5 11.0 11.5 12.0 12.5mc/ms

HPQCD 0910.3102

ETMC 1010.3659

ETMC 1403.4504

MILC 1407.3772

HPQCD 1408 . 4169

Durr 1108.1650

u,d,s,c sea

u,d,s sea

u,d sea

[HPQCD ’14]

I Uncertainty reduction in theory predictions for muon g − 2may come from LQCD → would directly impact FNAL experiment

3

Non-tradtitional HPC: Perturbative QCD

4

Interplay with experiment

jetsN

0 1 2 3 4 5 6 7

) [p

b]je

ts(W

+N

σ

-210

-110

1

10

210

310

410

510

610 ATLAS

jets, R=0.4,tanti-k

| < 4.4j

> 30 GeV, |yj

Tp

) + jetsν l→W(Data,

-1 = 7 TeV, 4.6 fbs+SHERPAATHLACKB

HEJALPGENSHERPAMEPS@NLO jetsN

0 1 2 3 4 5 6 7

Pre

d. /

Dat

a

0.6

0.8

1

1.2

1.4 +SHERPAATHLACKB

ATLAS

jetsN

0 1 2 3 4 5 6 7

Pre

d. /

Dat

a

0.6

0.8

1

1.2

1.4 HEJ

jetsN

0 1 2 3 4 5 6 7

Pre

d. /

Dat

a

0.6

0.8

1

1.2

1.4 ALPGEN

jetsN0 1 2 3 4 5 6 7

Pre

d. /

Dat

a

0.6

0.8

1

1.2

1.4

SHERPA

MEPS@NLO

I Both inclusive and fully differential results needed to scrutinize SM

I Particle-level predictions mandatory for direct comparison and unfolding

5

Perturbative QCD at the LHC

Aspects of the theory

I Perturbative regimeI Hard processesI Radiative corrections

I Non-perturbative regimeI HadronizationI Particle decays

Divide et Impera

I Quantity of interest: Total interaction rate

I Convolution of short & long distance physics

σp1p2→X =∑

i,j∈{q,g}

∫dx1dx2 fp1,i (x1, µ

2F )fp2,j (x2, µ

2F )︸ ︷︷ ︸

long distance

σij→X (x1x2, µ2F )︸ ︷︷ ︸

short distance

6

pQCD Toolkit inventory

Current state of development

I Parton shower Monte Carlo (Herwig,Pythia,Sherpa,...)

I Automated NLO calculations(BlackHat,GoSam,Helac,MadLoop,MadGolem,NJet,OpenLoops,...)

I Matching to parton shower (aMC@NLO,Herwig,POWHEG Box,Sherpa,...)

I Merging of NLO calculations (aMC@NLO,Helac,Pythia,Sherpa,...)

Cutting edge technology & future directions

I Inclusive NNNLO (gg→H)

I Differential NNLO (V/H(+jet),γγ,VV,. . . )

I NNLO+NxLL resummation (gg→H,tt,. . . )

I NNLO+parton shower (W,Z,gg→H)

120

140

160

180

200

220

240

260

280

σto

t [p

b]

Scale variation

LO

NLONNLO

LL

NLLNNLL

LL NLLNNLL

LHC 8 TeV; mtop=173.3 GeV; A=0MSTW2008 LO; NLO; NNLO

Fixed OrderNLO+res

NNLO+res

tt cross section [Czakon et al.]

7

Ultra-high precision calculations

10-1

100

101

dσ/dpjet

T[pb/GeV

]

W+ (→ lν) +1j, @ 8 TeV

LO

NLO

NNLO

40 60 80 100 120 140 160 180

p jetT [ GeV ]

0.6

1.0

1.4

1.8

2.2

K

NLO/LO

NNLO/NLO

[Boughezal,Focke,Liu,Petriello ’15]

I New method for regularizingdivergences (jettiness subtraction)

I Calculation performed usinghybrid MPI+OpenMP approach

LO NLO NNLO NNNLO

0

10

20

30

40

50

σ/pb

LHC

pp→h+X gluon fusion

MSTW08 68cl

μ=μR=μF ∈ [mH /4,mH ]

Central scale: μ = mH /2

2 4 6 8 10 12 14-0.2

-0.1

0.0

0.1

0.2

0.3

S /TeV

[Anastasiou,Duhr,Dulat,Herzog,Mistlberger ’15]

I First complete N3LO calculationat a hadron collider

I Total scale variation 3%, reducingtheory uncertainty by factor 3

8

HPC usage projections from Snowmass 2013

Type of calculation CPU hours per project projects per year

NLO parton level 300,000 10-12Matrix Element Method 200,000 3-5

NNLO parton level 250,000 5-6Precision event generation 200,000 3Exclusive jet cross sections 300,000 1-2

Parton Distributions 50,000 5-6MSSM phenomenology 500,000 10

BSM constraints 150,000 2Model building 100,000 1-2

I Projected total of ≥6M CPUh for pQCD and ≥5.45M CPUh for BSM

I Prone to rapid changes depending on theory and technology developments

9

Actual HPC usage

Job Count by Number of Nodes

m1758 & m1738, Jun 2014 - Jun 2015

1024

512

256

128

64

32

16

8

4

2

1

Wall Hours by Number of Nodes

m1758 & m1738, Jun 2014 - Jun 2015

1024

512

256

128

64

32

16

8

4

2

1

I NERSC usage during past year ∼6.07M CPUh (pQCD only)

I Still at small-scale, but large potential for growth

I 7 publications in 2014, 3 in 2015 (as of Jun 8)

I First (full) calculation of W /H+jet at NNLO arXiv:1504.02131, arXiv:1505.03893

I First NNLO+PS matched simulation for Drell-Yan arXiv:1405.3607

10

Technology development

threads used1 10 210

run

time

(sec

onds

)

1

10

PP-> H(->bb)+ 2 jets @ LO

Processor:

Intel Psi 5110P

AMD 6128 HE

Intel Xeon X5650

Intel Core i7-4770

PP-> H(->bb)+ 2 jets @ LO

threads used1 10 210

run

time

(sec

onds

)

210

310

PP-> H(->bb)+ 2 jets @ NLO

Processor:

Intel Psi 5110P

AMD 6128 HE

Intel Xeon X5650

Intel Core i7-4770

PP-> H(->bb)+ 2 jets @ NLO

I MCFM generator has been a workhorse in NLO calculations for years

I Recently thread- (OpenMP) and MPI-parallelized arXiv:1503.06182

I Used for jettiness subtraction at NNLO arXiv:1505.03893 and arXiv:1504.02131

11

Summary

I Theory computing includes traditional and non-tradtional cases

I Lattice QCD has larger needs and well-developed technology

I Perturbative QCD is exploiting a large potential for growth

I BSM phenomenology still to join at larger scale (LBNL only so far)

I Small investments on computing can yield large returns on theory side(example pQCD NNLO W /Z+jet, computed at NERSC)