NNPDFPDFS:

FROM RUN II TO RUN III

STEFANO FORTEUNIVERSITA DI MILANO & INFN

CMS SM MEETING FEBRUARY 25, 2020

PDF PROGRESS: PRECISIONPDF4LHC PDFS (2014) NNPDF3.0 NNLO

GLUON SINGLET FLAVORS

• GLUON BETTER KNOWN AT SMALL x, VALENCE QUARKS AT LARGE x, SEA QUARKS IN BETWEEN

• TYPICAL UNCERTAINTIES IN DATA REGION ∼ 3− 5%

• SWEET SPOT: VALENCE Q - G; DOWN TO 1%

• UP BETTER KNOWN THAN DOWN; FLAVOR SINGLET BETTER THAN INDIVIDUAL FLAVORS

PDF PROGRESS: PRECISIONCURRENT PDFS (2017) NNPDF3.1 NNLO

GLUON SINGLET FLAVORS

• GLUON BETTER KNOWN AT SMALL x, VALENCE QUARKS AT LARGE x, SEA QUARKS IN BETWEEN

• TYPICAL UNCERTAINTIES IN DATA REGION ∼ 1− 3%

• SWEET SPOT: VALENCE Q - G; 1% OR BELOW

• UP BETTER KNOWN THAN DOWN; FLAVOR SINGLET BETTER THAN INDIVIDUAL FLAVORS

PDF PROGRESS: ACCURACYIMPACT OF ATLAS W/Z 7TEV DATA

CT18 NNPDF3.1

• CT18: PDF SETS RELEASED WITH/WITHOUT ATLAS W/Z DATA INCLUDED

• NNPDF3.1: CONSISTENCY OF ALL DATASETS INCLUDED

SUMMARY

TOWARDS NNPDF4.0 (2020)

DATA• THE NNPDF4.0 DATASET

• TOP PAIRS

• SINGLE TOP

• JETS

THEORY• ELECTROWEAK CORRECTIONS

• REGULARIZED COVARIANCE MATRICES

• THEORY UNCERTAINTIES

(METHODOLOGY)

DATA

DATASET WIDENINGNNPDF3.0 VS NNPDF3.1 (CT14 VS. CT18: SIMILAR)

NEW DATA: (BLACK EDGE)

• HERA COMBINED F b2• D0 W LEPTON ASYMMETRY

• ATLAS W,Z 2011, HIGH& LOW MASS DY 2011;CMS W± RAPIDITY 8TEVLHCB W,Z 7TEV & 8TEV

• ATLAS 7TEV JETS 2011,CMS 2.76TEV JETS

• ATLAS & CMS TOPDIFFERENTIAL RAPIDITY

• ATLAS Z pT DIFFERENTIALRAPIDITY & INVARIANT MASS8TEV,CMS Z pT DIFFERENTIALRAPIDITY 8TEV

DATASET WIDENINGNNPDF4.0 SUMMARY (EXPECTED IN 2020)

1. OLD DATASETS WITH IMPROVED TREATMENT

• ASSORTED DEBUGGING

• CORRELATIONS IN ATLAS TOP DISTRIBUTIONS AT 8 TEV• CHOICE OF SCALE AND CORRELATION MODELS FOR SINGLE-INCLUSIVE JET DATA

• MASSIVE CORRECTIONS TO NEUTRINO DIS DIMUON CROSS SECTIONS AT NNLO• NUCLEAR UNCERTAINTIES IN FIXED-TARGET DIS AND DY

2. NEW DATASETS FOR OLD PROCESSES

• DIS c AND b PRODUCTION (HERA COMBINED)• SINGLE JET PRODUCTION (ATLAS, CMS)• TOP PAIR PRODUCTION (ATLAS, CMS)• COLLIDER DY/INCLUSIVE VECTOR BOSON PRODUCTION (ATLAS, CMS, LHCB)• COLLIDER VECTOR BOSON PRODUCTION IS ASSOCIATION WITH CHARM (CMS)

3. NEW DATASETS FOR NEW PROCESSES

• ISOLATED PHOTON PRODUCTION (ATLAS)• SINGLE TOP PRODUCTION (ATLAS, CMS)• COLLIDER DIJET PRODUCTION (ATLAS, CMS)• DIS+JET(S) PRODUCTION (H1, ZEUS)• COLLIDER VECTOR BOSON PRODUCTION IS ASSOCIATION WITH JETS (ATLAS, CMS)

O(50) NEW/REVISED DATASETS

DISCUSSED IN THIS TALK

NEW DATASETS FOR OLD PROCESSES• DIS c AND b PRODUCTION

HERA COMBINED REDUCED CROSS SECTIONS (REPLACES H1 AND ZEUS DATA SETS)

• SINGLE-INCLUSIVE JETSATLAS 8 TEV DOUBLE DIFFERENTIAL DISTRIBUTIONS (R=0.4; R=0.6) (IMPLEMENT CORRELATION MODELS)CMS 8 TEV DOUBLE DIFFERENTIAL DISTRIBUTIONS (R=0.7)(IMPLEMENT CORRELATION MODELS)

• TOP PAIR PRODUCTIONATLAS 8 TEV DIFFERENTIAL ABS. AND NORM. DISTRIBUTIONS; DILEPTON; mtt; yttCMS 8 TEV DOUBLE DIFFERENTIAL NORM. DISTRIBUTIONS; DILEPTON; (ptT , yt); (mtt, yt); (mtt, ytt)CMS 13 TEV DIFFERENTIAL ABS. AND NORM. DISTRIBUTIONS; DILEPTON; ptT ; yt; ytt; mtt

CMS 13 TEV DIFFERENTIAL ABS. AND NORM. DISTRIBUTIONS; LEPTON+JET; ptT ; yt; ytt; mtt

CMS 5.02 TEV TOTAL CROSS SECTION

• COLLIDER DY/INCLUSIVE VECTOR BOSON PRODUCTION OR IN ASSOCIATION WITH CHARMATLAS 13 TEV W+, W− AND Z TOTAL CROSS SECTIONS (INCLUDING CORRELATIONS)ATLAS 7 TEV W+, W− AND Z RAPIDITY DISTRIBUTIONS (INCLUDIG OFF-PEAK AND FORWARD)ATLAS 8 TEV DOUBLE-DIFFERENTIAL HIGH-MASS DY DISTRIBUTIONS (TO BE CHECKED)ATLAS 8 TEV TRIPLE DIFFERENTIAL DY DISTRIBUTIONATLAS 7 TEV W + c DIFFERENTIAL DISTRIBUTIONSCMS 8 TEV Z + c DIFFERENTIAL DISTRIBUTIONSCMS 8 TEV DOUBLE-DIFFERENTIAL DY DISTRIBUTIONCMS 13 TEV DY DIFFERENTIAL CROSS-SECTIONS (VARIOUS DISTRIBUTIONS, AMONG WHICH ZpT )CMS 13 TEV W + c DIFFERENTIAL DISTRIBUTIONSLHCB 8 TEV W AND Z RAPIDITY DISTRIBUTIONS, ELECTRON CHANNELLHCB 13 TEV Z RAPIDITY DISTRIBUTIONS, DIELECTRON AND DIMUON CHANNELS

ALREADY IMPLEMENTED; ONGOING IMPLEMENTATION; TO DO

NEW DATA SETS FOR NEW PROCESSES• ISOLATED PHOTON PRODUCTION

ATLAS 8 TEV DIFFERENTIAL DISTRIBUTIONS, INCLUDING CORRELATIONSATLAS 13 TEV DIFFERENTIAL DISTRIBUTIONS

• SINGLE TOP PRODUCTION, t-CHANNELATLAS TOP TO ANTITOP RATIO, 7, 8 AND 13 TEVATLAS TOP AND ANTITOP ABS. AND NORM. DIFFERENTIAL DISTRIBUTIONS, 7 TEV, pT ; yATLAS TOP AND ANTITOP ABS. AND NORM. DIFFERENTIAL DISTRIBUTIONS, 8 TEV, pT ; yCMS TOP AND ANTITOP TOTAL CROSS SECTION, 7 TEVCMS TOP TO ANTITOP RATIO, 8 AND 13 TEV

• COLLIDER DI-JET PRODUCTIONATLAS 7 TEV DIJET DOUBLE DIFFERENTIAL DISTRIBUTIONS (R=0.4; R=0.6)CMS 5.02 TEV DIJET DOUBLE DIFFERENTIAL DISTRIBUTIONSCMS 7 TEV DIJET DOUBLE DIFFERENTIAL DISTRIBUTIONS (R=0.7)CMS 8 TEV DIJET TRIPLE DIFFERENTIAL DISTRIBUTIONS (R=0.7)

• DIS+JET(S) PRODUCTIONH1 SINGLE- AND DI-JET DIFFERENTIAL DISTRIBUTIONSZEUS SINGLE- AND DI-JET DIFFERENTIAL DISTRIBUTIONS

• COLLIDER VECTOR BOSON PRODUCTION IN ASSOCIATION WITH JETSATLAS 8 TEV W+JET DIFFERENTIAL DISTRIBUTIONS

ALREADY IMPLEMENTED; ONGOING IMPLEMENTATION; TO DO

TOP PAIR PRODUCTION(PAST RESULTS)

CONSISTENCY OF DIFFERENT OBSERVABLESTOP PRODUCTION AND THE GLUON

INCLUSION OF ATLAS TOP DATA IN HERA+TOP FIT (XFITTER)HQ PAIR RAPIDITY DISTN INVARIANT MASS DISTN.

INCONSISTENCY?

TOP PAIR PRODUCTION(PAST RESULTS)

TOP PRODUCTION AND THE GLUONINCLUSION OF ATLAS TOP DATA IN NNPDF3.1-LIKE FIT

ATLAS ONLY

0.1 0.2 0.3 0.4 0.5 0.6 0.7 x

0.6

0.7

0.8

0.9

1

1.1

1.2

1.3

1.4

1.5

g (

x, Q

) /

g (

x, Q

) [r

ef]

Baseline no-top+ ATLAS ytt+ ATLAS mtt+ ATLAS yt

det), Q = 100 GeVSαNNPDF3.1NNLO (ATLAS+CMS

0.1 0.2 0.3 0.4 0.5 0.6 0.7 x

0.6

0.7

0.8

0.9

1

1.1

1.2

1.3

1.4

1.5

g (

x, Q

) /

g (

x, Q

) [r

ef]

Baseline no-top

+ ATLAS+CMS ytt

+ ATLAS+CMS mtt

+ ATLAS+CMS yt

det), Q = 100 GeVSαNNPDF3.1NNLO (

• FOR ATLAS mtt & y DISTRIBUTIONS PULL IN OPPOSITE DIRECTION⇒ COMPATIBLE WITHIN UNCERTAINTIES

• mtt HAS MUCH LESS PULL

• FOR CMS, BOTH mtt & y PULL IN THE SAME DIRECTION

CONSISTENCY!

ASIDE: THE CHARM PDF FROM DATAIMPACT ON LIGHT QUARK PDFS

• TRADITIONALLY CHARM PERTURBATIVELY GENERATEDFROM (LOW-ORDER) MATCHING CONDITIONS

• NNPDF3.1: CHARM FITTED

FITTED VS. PERTURBATIVE CHARMQQBAR LUMI

( GeV )XM10 210 310

Qua

rk -

Ant

iqua

rk L

umin

osity

0.85

0.9

0.95

1

1.05

1.1

1.15

1.2 fitted charm

perturbative charm

LHC 13 TeV, NNLOANTIDOWN PDF

x 4−10 3−10 2−10 1−10

) [r

ef]

2 (

x, Q

d)

/ 2

( x

, Qd

0.9

0.95

1

1.05

1.1

1.15 Fitted charm

Perturbative charm

NNPDF3.1 NNLO, Q = 100 GeV W+/W− XSECT RATIO

1.26 1.27 1.28 1.29 1.30 1.31 1.32 1.33σW+/σW−

ATLAS 13 TeV

Heavy: NNLO QCD + NLO EWLight: NNLO QCD

Ratio of W+ to W− boson

NNPDF3.1

NNPDF3.0

CT14

MMHT14

ABMP16

data ± total uncertainty

• QUARK (ESPECIALLY QUARK-ANTIQUARK) LUMI AFFECTED

• FLAVOR DECOMPOSITION ALTERED

• NECESSARY FOR AGREEMENT OF 13TeV PREDICTED CROSS-SECTIONS

• NECESSARY FOR AGREEMENT WITH ATLAS W/Z 7 TeV

TOP PAIR PRODUCTION(NEW RESULTS)

(S.F., E. Nocera, J. Rojo, subm. to Les Houches 2019)

• WHAT IF WE FIT ALL DISTRIBUTIONS AT ONCE?CORRELATIONS ⇒ NOW POSSIBLE

• NORMALIZED VS. ABSOLUTE DISTRIBUTIONS?

• FITTING CHARM? (NEEDED FOR GOOD FIT TO ATLAS W/Z IN NNPDF3.1)

t RAPIDITY VS. ALL ALL NORM. VS. ABSOLUTE ALL NORM FIT. VS. PERT. CHARM

TOP PAIR PRODUCTION(NEW RESULTS)

(S.F., E. Nocera, J. Rojo, subm. to Les Houches 2019)

• WHAT IF WE FIT ALL DISTRIBUTIONS AT ONCE? ⇒ NO EFFECT!CORRELATIONS ⇒ NOW POSSIBLE

• NORMALIZED VS. ABSOLUTE DISTRIBUTIONS? ⇒ BIG EFFECT!

• FITTING CHARM? ⇒ SIZABLE EFFECT!t RAPIDITY VS. ALL ALL NORM. VS. ABSOLUTE ALL NORM FIT. VS. PERT. CHARM

• DIFFERENT DISTRIBUTIONS CONSISTENT (BUT pt BAD FIT)

• ABSOLUTE PATHOLOGICAL: BAD FIT, INCONSISTENT GLUON

• FITTED CHARM NEEDED FOR CONSISTENCY WITH GLOBAL DATASET

SINGLE TOP PRODUCTION (t-CHANNEL)

(E. Nocera, M. Ubiali, C. Voisey, arXiv:1912.09543)

• CANNOT USE MULTIPLE OBSERVABLES (NO CORRELATION)

• OPTIMAL SET⇒ XSECT RATIOS (ATLAS, CMS), TOTAL XSECT (CMS) + NORMALIZED RAPIDITYDISTRIBUTION (ATLAS)

• SEVERAL PROBLEMATIC DATA: pT DISTRIBUTIONS, ATLAS XSECT RATIOS

UP DOWN U/D RATIO

• CENTRAL VALUES UNCHANGED

• SLIGHT REDUCTION OF UNCERTAINTY ON LIGHT QUARKS

SINGLE TOP PRODUCTION(t-CHANNEL)IMPACT ON PHENOMENOLOGY

• MOST PRECISE REGION FOR qq LUMI SOMEWHAT EXTENDED

• UNCERTAINTY ON W+/W− HARDLY REDUCED

QUARK-ANTIQUARK LUMINOSITYBASELINE

4 2 0 2 4y

101

102

103

104

MX (

GeV)

Relative uncertainty for qq-luminosityBaseline (Fit 1) - s = 13000 GeV

1

5

10

25

50

Rela

tive

unce

rtain

ty (%

)

BASELINE+SINGLE TOP

4 2 0 2 4y

101

102

103

104

MX (

GeV)

Relative uncertainty for qq-luminosityOptimal fit - s = 13000 GeV

1

5

10

25

50

Rela

tive

unce

rtain

ty (%

)

W+/W− RATIO

3.3 3.4 3.5 3.6 3.7W [fb] ×106

4.1

4.2

4.3

4.4

4.5

4.6

4.7

4.8

4.9

5.0

W+

[fb]

×106 W± (NNLO)BaselineOptimal fit

TO BE UNDERSTOOD:• WHAT’S WRONG WITH THE ATLAS 8TEV XSECT RATIO?

• WHAT’S WRONG WITH THE pT DISTRIBUTIONS?

• CORRELATIONS BETWEEN OBSERVABLES?

JETS:SINGLE-INCLUSIVE: NNPDF3.1 DATASET

NNPDF+NNLOJET++, in preparation

• IMPACT ON GLUON UNCERTAINTY AT LARGE x

• SERIOUS ISSUES WITH THE ATLAS 7 TEV COVARIANCE MATRIX

IMPACT ON THE GLUONNLO

0.1 0.2 0.3 0.4 0.5 x

0.8

0.9

1

1.1

1.2

1.3

) [r

ef]

2)

/ g (

x, Q

2g

( x,

Q

Baseline (no jet data)

)jet

T=pµ+ 7 TeV InclJets (

)T

H=µ+ 7 TeV InclJets (

NNPDF3.1 NLO, Q = 100 GeVNNLO UNCERTAINTY

0.1 0.2 0.3 0.4 0.5 x

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

)2R

elE

rr o

n g

( x,

Q

Baseline (no jet data)

)jet

={T}µ+ 7 TeV InclJets (

)T

H=µ+ 7 TeV InclJets (

NNPDF3.1 NNLO, Q = 100 GeV

NNLO

0.1 0.2 0.3 0.4 0.5 x

0.8

0.9

1

1.1

1.2

1.3

) [r

ef]

2)

/ g (

x, Q

2g

( x,

Q

Baseline (no jet data)

)jet

={T}µ+ 7 TeV InclJets (

)T

H=µ+ 7 TeV InclJets (

NNPDF3.1 NNLO, Q = 100 GeVEW CORRECTIONS

0.1 0.2 0.3 0.4 0.5 x

0.8

0.9

1

1.1

1.2

1.3

) [r

ef]

2)

/ g (

x, Q

2g

( x,

Q

Baseline (no jet data)

+ 7 TeV InclJets (QCD)

+ 7 TeV InclJets (QCD+EW)

NNPDF3.1 NNLO, Q = 100 GeV

• HT BETTER FIT AT NLO VS pT , NO DIFFERENCE AT NNLO

• MODERATE IMPACT OF EW CORRNS ON PDFS, BUT BETTER FIT

JETSSINGLE-INCLUSIVE: FULL DATASET

NNPDF+NNLOJET++, in preparation

• CMS, ATLAS, 7, 8 TEV DATSETS CONSISTENT ⇒ UNCERTAINTY REDUCTION

• 8 TEV DATA ⇒ LARGE x GLUON AFFECTED

IMPACT ON THE NNLO GLUONPDF

10 2 10 1

x

0.85

0.90

0.95

1.00

1.05

1.10

1.15

Ratio

to B

asel

ine

(no

jet d

ata)

g at 100.0 GeVBaseline (no jet data) (68 c.l.+1 )Baseline + ATLAS and CMS single jets 7 TeV (68 c.l.+1 )Baseline + all single-jet data (68 c.l.+1 )

UNCERTAINTY

10 2 10 1

x

0.02

0.04

0.06

0.08

0.10

(Rat

io to

Bas

elin

e (n

o je

t dat

a))

g at 100.0 GeVBaseline (no jet data)Baseline + ATLAS and CMS single jets 7 TeVBaseline + all single-jet data

JETS:SINGLE-INCLUSIVE VS. DIJETS

NNPDF+NNLOJET++, in preparation

• CANNOT USE BOTH BECAUSE OF MISSING CORRELATIONS

• DIJETS BETTER FIT QUALITY

IMPACT ON THE GLUONPDF

10 2 10 1

x

0.85

0.90

0.95

1.00

1.05

1.10

1.15

Ratio

to B

asel

ine

(no

jet d

ata)

g at 100.0 GeVBaseline (no jet data) (68 c.l.+1 )Baseline + all single-jet data (68 c.l.+1 )Baseline + all di-jet data (68 c.l.+1 )

UNCERTAINTY

10 2 10 1

x

0.02

0.04

0.06

0.08

0.10

0.12

0.14

(Rat

io to

Bas

elin

e (n

o je

t dat

a))

g at 100.0 GeVBaseline (no jet data)Baseline + all single-jet dataBaseline + all di-jet data

• SMILAR IMPACT OF EITHER OBSERVABLE

• SINGLE-INCLUSIVE ⇒ SMALLER UNCERTAINTY (BUT MORE DATA)

• DIJETS ⇒ (MORE) LARGE x ENHANCEMENT

• NEED FULL DATASET TO FIGURE OUT THE FINAL PRECISE PDF SHAPE

THEORY

ELECTROWEAK CORRECTIONSEXAMPLE: TOP PAIR PRODUCTION

NNPDF3.1

ELECTROWEAK CORRECTIONSEXAMPLE: TOP PAIR PRODUCTION

NNPDF4.X

S. Carrazza, E. Nocera, C. Schwan, M. Zaro, in preparation

ELECTROWEAK CORRECTIONSEXAMPLE: TOP PAIR PRODUCTION

NNPDF4.X

S. Carrazza, E. Nocera, C. Schwan, M. Zaro, in preparation

• TOOLCHAIN BASED ON mg5 aMC, aMCblast, APPLgridEW INTERFACED TO NNPDFTHROUGH APFELcomb

• GRIDS COMPUTED AND TEST FITS PERFORMED FOR SELECTED PROCESSES

EXAMPLE: GLUON-GLUON GRID FOR tt 8 TeV

010203040ya

0

10

20

30

40

y bδgg→tt(xa(ya), xb(yb), Q

2 = m2t)

−10.0

−7.5

−5.0

−2.5

0.0

2.5

5.0

7.5

10.0

• δ = O(ααs )/O(α2s); yi(x) = − lnxi + 5(1− xi)

• CORRECTION TYPICALLY ∼ 5%

ELECTROWEAK CORRECTIONSTHE DOUBLE-COUNTING PROBLEM

CONTRIBUTIONS SOMETIMES SUBTRACTED BY EXPERIMENTEXAMPLE: ATLAS 7 TeV HIGH-MASS DY IN Z, γ → e+e− CHANNEL

• LO PHOTON-INDUCED SUBTRACTED (BUT NOT PHOTON-QUARK, NLO)

• THREE DATASETS DEFINED:– BORN LEPTONS: NO FINAL-STATE RADIATION (OK FOR QCD)– BARE LEPTONS: FINAL-STATE RADIATION

– DRESSED LEPTONS: FINAL-STATE RADIATION + RECOMBINATION WITHINGIVEN RADIUS (OK FOR QCD+EW)

DESIDERATA

• NEED DATASETS BOTH AT THE BARE & DRESSED LEVEL 4• PLEASE DO NOT SUBTRACT PHOTON INDUCED! (WHY WOULD YOU?) 7

THE COVARIANCE MATRIX PROBLEM• FOR SEVERAL DATSETS, FULL CORRELATED χ2 POOR

EXAMPLES:

• ATLAS 8 TeV tt ABSOLUTE DISTRIBUTIONS: χ2/Ndat ∼ 6,DECORRELATING OBSERVABLES χ2/Ndat ∼ 2

• ATLAS 2011 7 TeV SINGLE-INCLUSIVE JETS: χ2/Ndat ∼ 2.5,DECORRELATING RAPIDITY BINS χ2/Ndat ∼ 1 ⇔ IF ONLY ONE BIN FITTED, ALL BINS

PREDICTED WITH χ2/Ndat ∼ 1

• ATLAS W/Z 2011 7 TEV: χ2/Ndat ∼ 2

THE COVARIANCE MATRIX PROBLEMEXAMPLE: THE ATLAS JETS

• ONLY CENTRAL RAPIDITY BIN ⇒ χ2 ∼ 1, ALL BINS χ2 ∼ 2.5

• BUT PDFS UNCHANGED!

PDFS: CENTRAL BIN VS ALL BINSGLUON QUARK SINGLET

10 5 10 4 10 3 10 2 10 1 100

x

0.900

0.925

0.950

0.975

1.000

1.025

1.050

1.075

1.100

Ratio

to C

urre

nt F

it

at 1.7 GeVCurrent Fit (68 c.l.+1 )Reference Fit (68 c.l.+1 )

THE COVARIANCE MATRIX PROBLEMATLAS W/Z 2011 7 TEV CORRELATION MATRIX

• POORLY CONDITIONED MATRIX: ONE (OR MORE) VERY SMALL EIGENVALUES

• GENERIC SITUATION WITH SMALL UNCORRELATED UNCERTAINTIES AND ONE ALMOSTFULLY CORRELATED UNCERTAINTY

REGULARIZING ⇔ DECORRELATINGZ. Kassabov, E. Nocera, M. Wilson, in preparation

• PERFORM SINGULAR-VALUE DECOMPOSITION

• REPLACE SMALLEST EIGENVALUES WITH CUTOFF 1/δ

ATLAS W/Z 2011 7 TEV CORRELATION MATRIX

• EFFECTIVELY ADDS 1/δ TO DIAGONAL OF COVARIANCE MATRIX

• CORRELATION MATRIX ALMOST UNCHANGED

REGULARIZED FITZ. Kassabov, E. Nocera, M. Wilson, in preparation

• APPLY REGULARIZATION TO FULL NNPDF3.1 DATASET

• REFIT

RATIO OF REGULARIZED TO DEFAULT

• χ2 DOWN FROM 1.16 TO 1.0 FOR ∼ 4000 DATAPOINTS

• PDFS UNCHANGED

THE MISSING HIGHER ORDER UNCERTAINTY ON PDFS

NLO-NNLO SHIFT VS. NLO PDF UNCERTAINTY (NNPDF3.1)ANTIDOWN

x 4−10 3−10 2−10 1−10

)2 (

x, Q

d ∆

0.9

0.95

1

1.05

1.1

1.15

NLO PDF uncertainties

TH error (NLO => NNLO shift)

NNPDF3.1, Q = 100 GeVGLUON

x 4−10 3−10 2−10 1−10

)2 g

( x

, Q∆

0.9

0.95

1

1.05

1.1

1.15

NLO PDF uncertainties

TH error (NLO => NNLO shift)

NNPDF3.1, Q = 100 GeV

• NEVER INCLUDED IN CURRENT PDF SETS

• TODAY: NLO PDF & MHOU UNCERTAINTIES COMPARABLE

• NEAR FUTURE: SHOULD WE WORRY ABOUT NNLO MHOU?

THE THEORY COVARIANCE MATRIX FROM SCALE VARIATION

NNPDF, 2019

• INDEPENDENT NUISANCE PARAMETERS ⇒ TH. AND EXP. ERRORS COMBINE IN QUADRATURE

χ2 =∑Ndat

i,j=1

(Di − T

(0)i

)[S + C]−1

ij

(Di − T

(0)i

)• REN. SCALE ⇒ CORRELATIONS INDUCED BETWEEN

EXPERIMENTALLY UNRELATED MEASUREMENTS OF SAME PROCESS

• FACT. SCALE ⇒ CORRELATIONS INDUCED BETWEEN DIFFERENT PROCESSES

THE COVARIANCE MATRIXEXPERIMENT

DIS NCDIS CC DY JETSTOP

DIS NC

DIS CC

DY

JETSTOP

Experimental Covariance Matrix

102101

10010 1

10 2

0

10 210 1

100101

102

% o

f dat

a

THEORY (9 PT)

DIS NCDIS CC DY JETSTOP

DIS NC

DIS CC

DY

JETSTOP

Theory Covariance matrix (9 pt)

101

100

10 1

10 2

0

10 2

10 1

100

101

% o

f dat

a

PDFS WITH THEORY UNCERTAINTIESGLUON

5−10 4−10 3−10 2−10 1−10 x

0.8

0.9

1

1.1

1.2

1.3

) [r

ef]

2)

/ g (

x, Q

2g

( x,

Q

NNPDF3.1 Global, Q = 10 GeV

NLO, CNLO, C+S(9pt)NNLO, C

NNPDF3.1 Global, Q = 10 GeVANTIDOWN

5−10 4−10 3−10 2−10 1−10 x

0.8

0.9

1

1.1

1.2

1.3

) [r

ef]

2 (

x, Q

d)

/ 2

( x

, Qd

NNPDF3.1 Global, Q = 10 GeV

NLO, CNLO, C+S(9pt)NNLO, C

NNPDF3.1 Global, Q = 10 GeV

C C + S(9pt)

χ2 1.139 1.109φ 0.314 0.415

• FIT QUALITY χ2 IMPROVES

• RELATIVE ERROR φ ON PREDICTION MILDLY INCREASED

• CENTRAL VALUE MOVES TOWARDS KNOWN NNLO

EQUALLY PRECISE BUT MORE ACCURATE RESULT!

PHENOMENOLOGY (LHC 13)HIGGS GG

NLO NNLO N3LO

30

35

40

45

50

σ[p

b]

light: scale uncertainty

dark: PDF uncertainty

left: C

right: C+S (9pt)

Higgs production: gluon fusion

NNLO PDFs

HIGGS VBF

NLO NNLO N3LO

3.88

3.90

3.92

3.94

3.96

3.98

4.00

4.02

4.04

σ[p

b]

light: scale uncertainty

dark: PDF uncertainty

left: C

right: C+S (9pt)

Higgs production: Vector Boson Fusion

NNLO PDFs

TOP PAIRS

NLO NNLO

650

700

750

800

850

σ[p

b]

light: scale uncertainty

dark: PDF uncertainty

left: C

right: C+S

pp→ tt, LHC 13 TeV, global NLO PDFs at all orders

NNLO PDFs

USAGE: JUST COMPUTE PDF ERROR AS USUAL &COMBINE WITH MHOU ON HARD MATRIX ELEMENT COMPUTED WITH YOUR PREFERRED RECIPE

• GLUON FUSION, TOP

– NO EFFECT ON CENTRAL VALUE

– VISIBLE INCREASE OF UNCERTAINTIE

• VBF– MODERATE EFFECT ON UNCERTAINTIES

– VISIBLE SHIFT OF CENTRAL VALUES

• ALL: TRUE NNLO NOW WITHIN NLO ERROR BAND

METHODOLOGY

NNPDF4.0: N3FITS. Carrazza, J. Cruz-Martinez, NNPDF, 2020

• FIRST PDF SET BASED ON DEEP LEARNING MODELS:HYPEROPTIMIZED METHODOLOGY

• CLOSURE TESTED

• “FUTURE” TESTED

PREDICTED VS TRUE GLUONONLY PRE-HERA DATA USED

NNPDF4.X HIGHLIGHTS

• BASED ON DEEP-LEARNING & CLOSURE TESTED

• VERY LARGE DATASET: ∼ 50 NEW DATASETS, SEVERAL NEW PROCESSES

• THEORY UNCERTAINTIES ACCOUNTED FOR

• ELECTROWEAK CORRECTIONS INCLUDED UP TO NLO

THE WORK OF MANY PEOPLE

NNPDF collaboration and N3PDF team meeting,

Varenna, Italy, September 2019