NNPDF3.0 parton distributions for the LHC run IInnpdf.mi.infn.it/wp-content/uploads/2017/10/deans.pdf · INTRODUCTION NNPDF3.0 CLOSURE TESTS PRELIMINARY RESULTS CONCLUSIONS PARTON

INTRODUCTION NNPDF3.0 CLOSURE TESTS PRELIMINARY RESULTS CONCLUSIONS

NNPDF3.0 parton distributions for theLHC run II

Christopher S. DeansUniversity of Edinburgh

On behalf of the NNPDF collaboration:R. D. Ball, V. Bertone, S. Carrazza, CD, L. Del Debbio, S. Forte, A. Guffanti,

N. P. Hartland, J. I. Latorre, J. Rojo, M. Ubiali

QCD 14, Montpellier30 June, 2014


OUTLINE

INTRODUCTION

Parton Distribution FunctionsNNPDF ApproachCurrent PDFs

NNPDF3.0New DataMethodology

CLOSURE TESTS

ImplementationResults

PRELIMINARY RESULTS

Parton Distributions

CONCLUSIONS


PARTON DISTRIBUTION FUNCTIONS

Plot from PDG2013 update

I PDF characterize partoncontent of the proton

I Important input intocalculations of LHCobservables

I PDFs are fit from data, andLHC input is important

(H)

[pb]

σ

17

17.5

18

18.5

19

19.5

20

= 0.119 - PDF uncertaintiesSαLHC 8 TeV - iHixs 1.3 NNLO -

NNPDF2.3

MSTW08

CT10

ABM11

HERAPDF1.5


NNPDF APPROACH

I Global Fit, determinedusing a wide range ofobservables (DIS,Drell-Yan, Inclusivejets...)

I Monte Carlo ReplicaPDFs provideuncertainties

I Neural Networks usedas unbiasedinterpolaters

I Genetic Algorthm toefficiently obtain best fitPDFs

x-510 -410 -310 -210 -110

-2

-1

0

1

2

3

4

5

6

7)2xg(x,Q

Replica PDFsMean value

error bandσ168% CL band

)2xg(x,Q


CURRENT PDFS – NNPDF2.3 [ARXIV:1207.1303]

x-510 -410 -310 -210 -110 1

-2

-1

0

1

2

3

4

5

6

7)

02xg(x, Q

NNPDF2.3 NNLO

NNPDF2.1 NNLO

NNPDF2.3 NNLO

NNPDF2.1 NNLO

)02xg(x, Q

x-510 -410 -310 -210 -110 1

0

1

2

3

4

5

6

7)

02(x, QΣx

NNPDF2.3 NNLO

NNPDF2.1 NNLO

NNPDF2.3 NNLO

NNPDF2.1 NNLO

)02(x, QΣx

I First public PDF set use LHCdata in the determination

I Includes all relevant LHC setsavailable with full correlations(at the time)

I ATLAS W and Z leptonrapidity

I ATLAS 7 TeV inclusive jetsI CMS W electron asymmetryI LHCb W rapidity

I Default PDF set inMadgraph5 aMC@NLO,Pythia 8, used in new Monash2013 tune

I Available in LHAPDF5.9 and 6.0


CURRENT PDFS – NNPDF2.3QED [ARXIV:1308.0598]

I PDF set based on NNPDF2.3 with QED corrections andaddition of photon PDF with uncertainty

I Photon-induced LHC processes included via reweightingI LHCb low-mass Z/γ∗I ATLAS inclusive W and Z productionI ATLAS high-mass Z/γ∗

I Both sets are available on LHAPDF at NLO and NNLOwith αS variations

x-510 -410 -310 -210 -110 1

)2(x

,Qγx

-0.02

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16MRST2004QEDNNPDF2.3QED averageNNPDF2.3QED replicas

σNNPDF 1NNPDF 68% c.l.

2 GeV4Photon PDF comparison at 10

x0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

)2(x

,Qγx

-0.01

0

0.01

0.02

0.03

0.04

0.05

2 GeV4Photon PDF comparison at 10

MRST2004QEDNNPDF2.3QED averageNNPDF2.3QED replicas

σNNPDF 1NNPDF 68% c.l.


NNPDF3.0

New PDF set, due for release this summer

I 1000 new datapoints from HERA-II and LHC, for a total of4000 datapoints

I Completely rewritten C++ code, with improved structure

I Extensively upgraded fitting methodology, refined andvalidated using closure tests

I Improved cross-validation to control overfitting

I Extended set of positivity constraints


NNPDF3.0 – NEW DATAI HERA-II

I H1 high Q2 NC and CC data [JHEP 1209 (2012) 061]I H1 low Q2, high y NC data [Eur.Phys.J. C71 (2011) 1579]I ZEUS NC and CC positron beam data

[Phys.Rev. D87 (2013) 5, 052014] [Eur.Phys.J. C70 (2010) 945-963]I HERA combined charm production [Eur.Phys.J. C73 (2013) 2311]

I ATLASI Inclusive jets

√s = 2.76 TeV (correlated with ATLAS 7 TeV

jets) [Eur.Phys.J. C73 (2013) 2509]I High Mass Drell-Yan [Eur.Phys.J. C70 (2010) 945-963]

I CMSI Inclusive jets

√s = 7 TeV [Phys.Rev. D87 (2013) 11, 112002]

I Double differential Drell-Yan [JHEP 1312 (2013) 030]I Muon charge asymmetry [arXiv:1312.6283]I W+ charm [JHEP 1402 (2014) 013]

I LHCbI Large rapidity Z→ ee [JHEP 1302 (2013) 106]

I t̄t total σ from ATLAS and CMS 7 TeV & 8 TeV


NNPDF3.0 DATA

x-610 -510 -410 -310 -210 -110 1

]2

[ G

eV2 T

/ p

2 /

M2

Q

1

10

210

310

410

510

610

710 FT DIS

HERA1

FT DY

TEV EW

TEV JET

ATLAS EW

LHCB EW

LHC JETS

HERA2

ATLAS JETS 2.76TEV

ATLAS HIGH MASS

ATLAS WpT

CMS W ASY

CMS JETS

CMS WC TOT

CMS WC RAT

LHCB Z

TTBAR

NNPDF3.0 NLO dataset


NEW THEORY

Progress in theory are an important part of reducing PDFuncertainties.

NNPDF3.0 makes use of several recent improvements:I Full NNLO top quark production cross section, differential

distribution coming soon. [Czakon, Fiedler, Mitov, Phys.Rev.Lett. 110

(2013) 25, 252004]

I NNLO inclusive jet production in the gg channel.[Gehrmann-De Ridder, Gehrmann, Glover, Pires, Phys.Rev.Lett. 110 (2013) 16,

162003], which we use to determine valid region forapproximate NNLO.

I QED and EW corrections provided by FEWZ3.1NNLO and EW corrections are included in NNPDF3.0 datasetvia c-factors


NNPDF3.0 METHODOLOGY

Major upgradeI Streamlined Genetic Algorithm Several obsolete features

removedI Nodal Mutations Exploit structure of neural networks in

genetic algorithmI Larger number of mutants Explore a larger region of

parameter spaceI Optimized Fits are significantly faster despite larger

datasetI Extended positivity New positivity constraints to ensure

positive definite cross-sections


UPDATED CROSS-VALIDATION

Neural network are very flexible – need to control overfittingIn NNPDF3.0 this is done using updated Cross-Validationstrategy

I Split data into two sets: training and validationI Train networks only on training set, but record validation χ2

I At the end of the fit, rewind to the point with the lowestvalidation χ2

This ‘optimal’ point hasthe best quality of fit tounseen data

Overfitting is prevented

GA generation0 5000 10000 15000 20000 25000 30000

2 χ

2.2

2.25

2.3

2.35

2.4

2.45

2.5

per generation2χReplica

Training

Validation

per generation2χReplica


CLOSURE TESTS

Key Question: How do we know whether our methodologyworks? Or how close is our fit to the ‘true’ value?

Closure Tests can provide an answer: fit pseudo-data generatedusing known theory.

Advantages:I Can compare result of fit directly to (artificial) underlying

lawI Clean environment to test different fitting methodologiesI Possible to investigate overfitting


CLOSURE TESTS – IMPLEMENTATION

Closure test process:

I Generate observables using theory and PDF setI Makes use of existing tools for observables used in fits

(FastKernel tables)I Can use any PDF set as long as it has an LHgrid

I Add statistical noise using experimentaluncertainties and replace data values

I Generates perfectly consistent dataI Also possible create data without noise or to

intentionally add inconsistencies

I Fit with standard methodologyI Runs on same code as fit to experimental dataI Closure test are a good testing ground for

methodological improvements

I Compare fit PDFs to original PDF setI Since underlying law is know can directly compare how

close fit PDFs areI Can also look at overfitting: smaller χ2 but further away

PDFs

GeneratePseudodata

Perform Fit

Compare tooriginal

StartingPDF set


CLOSURE TEST RESULTS: NNPDF2.3

Experiments

NMC SLACBCDMS

CHORUSNTVDMN

HERA1AVZEUSHERA2

H1HERA2HERAF2CHARM

DYE886DYE605

CDF D0ZRAPATLAS

CMS LHCBTOP

2 χ

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

for experiments2χDistribution of

2χClosure_NNPDF23_preliminary 2χNNPDF23_nnlo_as_0118

2χClosure_NNPDF23_preliminary central 2χNNPDF23_nnlo_as_0118 central

for experiments2χDistribution of

I Closure test fit to NNPDF2.3pseudo-data.

I χ2 of 1.053 compared with1.058 for NNPDF2.3 itself

I Distances (see right)compatible with PDFuncertainties

x-510 -410 -310 -210 -110 1

0

1

2

3

4

5

6

7)2(x,QΣx

Closure_NNPDF23_preliminary

NNPDF23_nnlo_as_0118_1gev

)2(x,QΣx

x-5

10 -410-3

10 -210 -110

Dis

tanc

e (in

uni

ts o

f sig

ma)

0

1

2

3

4

5Distances for NNPDF2.3 closure test

gΣV

3Ts∆+s-s

Distances for NNPDF2.3 closure test

x0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Dis

tanc

e (in

uni

ts o

f sig

ma)

0

1

2

3

4

5Distances for NNPDF2.3 closure test

gΣV

3Ts∆+s-s

Distances for NNPDF2.3 closure test

x-5

10 -410-3

10 -210 -110

Dis

tanc

e (in

uni

ts o

f sig

ma)

0

1

2

3

4

5Uncertainty

gΣV

3Ts∆+s-s

Uncertainty

x0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Dis

tanc

e (in

uni

ts o

f sig

ma)

0

1

2

3

4

5Uncertainty

gΣV

3Ts∆+s-s

Uncertainty

Closure test fit vs Generating PDF


CLOSURE TEST RESULTS: OTHER SETSMSTW2008

x-510 -410 -310 -210 -110 1

0

1

2

3

4

5

6

7)2(x,QΣx

Closure_MSTW_preliminary

MSTW2008nlo68cl

)2(x,QΣx

x-5

10 -410-3

10 -210 -110

Dis

tanc

e (in

uni

ts o

f sig

ma)

0

1

2

3

4

5Distances for MSTW2008 closure test

gΣV

3Ts∆+s-s

Distances for MSTW2008 closure test

x0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Dis

tanc

e (in

uni

ts o

f sig

ma)

0

1

2

3

4

5Distances for MSTW2008 closure test

gΣV

3Ts∆+s-s

Distances for MSTW2008 closure test

x-5

10 -410-3

10 -210 -110

Dis

tanc

e (in

uni

ts o

f sig

ma)

0

1

2

3

4

5Uncertainty

gΣV

3Ts∆+s-s

Uncertainty

x0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Dis

tanc

e (in

uni

ts o

f sig

ma)

0

1

2

3

4

5Uncertainty

gΣV

3Ts∆+s-s

Uncertainty


CT10

x-510 -410 -310 -210 -110 1

0

1

2

3

4

5

6

7)2(x,QΣx

Closure_CT10_preliminary

CT10nnlo

)2(x,QΣx

x-5

10 -410-3

10 -210 -110

Dis

tanc

e (in

uni

ts o

f sig

ma)

0

1

2

3

4

5Distances for CT10 closure test

gΣV

3Ts∆+s-s

Distances for CT10 closure test

x0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Dis

tanc

e (in

uni

ts o

f sig

ma)

0

1

2

3

4

5Distances for CT10 closure test

gΣV

3Ts∆+s-s

Distances for CT10 closure test

x-5

10 -410-3

10 -210 -110

Dis

tanc

e (in

uni

ts o

f sig

ma)

0

1

2

3

4

5Uncertainty

gΣV

3Ts∆+s-s

Uncertainty

x0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Dis

tanc

e (in

uni

ts o

f sig

ma)

0

1

2

3

4

5Uncertainty

gΣV

3Ts∆+s-s

Uncertainty



PRELIMINARY NNPDF3.0 RESULTS

x-510 -410 -310 -210 -110 1

-2

-1

0

1

2

3

4

5

6

7)2xg(x,Q

NNPDF30_nnlo_preliminary

NNPDF23_nnlo_as_0118

)2xg(x,Q

x-510 -410 -310 -210 -110 1

0

1

2

3

4

5

6

7)2(x,QΣx



)2(x,QΣx

x-510 -410 -310 -210 -110 1

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6)2xV(x,Q



)2xV(x,Q

x-510 -410 -310 -210 -110 1

-0.2

0

0.2

0.4

0.6

0.8

1)2(x,Q+xs

140618-r1796-003-jr

NNPDF23_nnlo_as_0118_1gev

)2(x,Q+xs

Preliminary Preliminary

Preliminary Preliminary


CONCLUSIONS

I Current NNPDF releases:I NNPDF2.3: First public set with LHC dataI NNPDF2.3QED: Determination with QED corrections and

photon pdf with uncertaintiesI All of our fits are available on LHAPDF and on our

webpage nnpdf.hepforge.org

I New pdf set, NNPDF3.0I New data from LHC, HERAIII Closure test demonstrate success of updated methodologyI Available soon, followed by NNPDF3.0QED

I Additional work within NNPDF collaboration:I Intrinsic charmI Polarized NNPDFs, see arXiv:1303.7236

nnpdf.hepforge.org

Documents

NNPDF3.0 parton distributions for the LHC run IInnpdf.mi.infn.it/wp-content/uploads/2017/10/deans.pdf · INTRODUCTION NNPDF3.0 CLOSURE TESTS PRELIMINARY RESULTS CONCLUSIONS PARTON