Post on 25-Aug-2021
1 / 28
The challenge of femtometer scale physics
Nobuo SatoODU/JLab
QCD Evolution 19Argonne National Laboratory
Outline
2 / 28
1. A new perspective for MC event generators
2. Understanding the large pT SIDIS spectrum
3. New developments to identify SIDIS regions
3 / 28
A new perspective for MC event generators
QCD physicists :W. Melnitchouk (JLab-theory)T. Liu (JLab-theory)NS (JLab/ODU-theory)R. E. McCleallan (JLab-theory)
Computer Scientists:Y. Li (ODU)M. Kuchera (Davidson College)
Supported by:JLab LDRD19-13
From detectors to partons
4 / 28
From detectors to partons
4 / 28
Detectorresponse
QEDeffects
QCDeffects
From detectors to partons
4 / 28
Detectorresponse
QEDeffects
QCDeffects
σEXP = wDR ⊗ wQED ⊗ σQCD
A new perspective for MC event generators
5 / 28
An empirical tool such that
σEXP = wDR ⊗ σMCEG
↓wQED ⊗ σQCD
A tool agnostic of theory (partons, factorization, hadronization, etc.)
Machine Learning based MCEG
A new perspective for MC event generators
5 / 28
An empirical tool such that
σEXP = wDR ⊗ σMCEG
↓wQED ⊗ σQCD
A tool agnostic of theory (partons, factorization, hadronization, etc.)
Machine Learning based MCEG
A new perspective for MC event generators
5 / 28
An empirical tool such that
σEXP = wDR ⊗ σMCEG
↓wQED ⊗ σQCD
A tool agnostic of theory (partons, factorization, hadronization, etc.)
Machine Learning based MCEG
Empirically Trained Hadronic Event Regenerator (ETHER)
6 / 28
Nature
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
Empirically Trained Hadronic Event Regenerator (ETHER)
6 / 28
Nature
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
Experimentaldetector
Events:detector level
0.0 0.2 0.4 0.6 0.8 1.00
100
200
300
400distortion
Empirically Trained Hadronic Event Regenerator (ETHER)
6 / 28
Nature
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
Experimentaldetector
Events:detector level
0.0 0.2 0.4 0.6 0.8 1.00
100
200
300
400distortion
Inverse problem → Model dependent
Empirically Trained Hadronic Event Regenerator (ETHER)
6 / 28
Nature
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
Experimentaldetector
Events:detector level
0.0 0.2 0.4 0.6 0.8 1.00
100
200
300
400distortion
ETHER
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
Empirically Trained Hadronic Event Regenerator (ETHER)
6 / 28
Nature
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
Experimentaldetector
Events:detector level
0.0 0.2 0.4 0.6 0.8 1.00
100
200
300
400distortion
ETHER
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
Detectorsimulator
neural netdetector
Empirically Trained Hadronic Event Regenerator (ETHER)
6 / 28
Nature
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
Experimentaldetector
Events:detector level
0.0 0.2 0.4 0.6 0.8 1.00
100
200
300
400distortion
ETHER
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
Detectorsimulator
neural netdetector
Events:detector level
0.0 0.2 0.4 0.6 0.8 1.00
100
200
300
400distortion
Empirically Trained Hadronic Event Regenerator (ETHER)
6 / 28
Nature
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
Experimentaldetector
Events:detector level
0.0 0.2 0.4 0.6 0.8 1.00
100
200
300
400distortion
ETHER
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
Detectorsimulator
neural netdetector
Events:detector level
0.0 0.2 0.4 0.6 0.8 1.00
100
200
300
400distortion
Likelihoodanalysis
Empirically Trained Hadronic Event Regenerator (ETHER)
6 / 28
Nature
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
Experimentaldetector
Events:detector level
0.0 0.2 0.4 0.6 0.8 1.00
100
200
300
400distortion
ETHER
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
Detectorsimulator
neural netdetector
Events:detector level
0.0 0.2 0.4 0.6 0.8 1.00
100
200
300
400distortion
Likelihoodanalysis
datacom
pression
Empirically Trained Hadronic Event Regenerator (ETHER)
7 / 28
Pythia
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
ETHER
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
Likelihoodanalysis
datacom
pression
Empirically Trained Hadronic Event Regenerator (ETHER)
7 / 28
Pythia
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
ETHER
Events:vertex level
0.0 0.2 0.4 0.6 0.8 1.00
50
100
150
200
250
Likelihoodanalysis
datacom
pression
+ ML setup: Generativeadversarial network (GAN’s)
+ Milestones:
! A setup for exclusive l + preaction
! particles multiplicities
' xbj, Q2,...
8 / 28
Understanding the large pT SIDIS spectrum
Work based on
o Gonzalez-Hernandez, Rogers, NS, Wang ( PRD98 2018 )o Wang, Gonzalez-Hernandez, Rogers, NS ( arXiv:1903.01529 2019 )
SIDIS FO O(α2S) calculation completed! Wang, Gonzalez, Rogers, NS (’19)
9 / 28
Wµν(P, q, PH) =∫ 1+
x−
dξ
ξ
∫ 1+
z−
dζ
ζ2 Wµνij (q, x/ξ, z/ζ)fi/P (ξ)dH/j(ζ)
Pµνg W (N)µν ; PµνPP W
(N)µν ≡
1(2π)4
∫|M2→N
g |2; |M2→Npp |2 dΠ(N) − Subtractions
Born/Virtual
Real
X Generate all 2→ 2 and 2→ 3 squaredamplitudes
X Evaluate 2→ 2 virtual graphs(Passarino-Veltman)
X Integrate 3-body PS analytically
X Check cancellation of IR poles
X Agreement within 20% with Daleo’s etal(’05)
Why is this calculation relevant?
10 / 28
0.0 0.5 1.0 1.5 2.0 2.5 3.0
qT (GeV)
10−4
10−3
10−2
10−1
100
Γ(1
)(qT,Q
)
Q = 2.0 (GeV)
FO
|AY|Y
|W|W + Y
0 1 2 3 4 5
qT (GeV)
10−4
10−3
10−2
10−1
100
Γ(1
)(qT,Q
)
Q = 5.0 (GeV)
FO
|ASY|Y
|W|W + Y
dσdxdQdzdqT
Why is this calculation relevant?
10 / 28
0.0 0.5 1.0 1.5 2.0 2.5 3.0
qT (GeV)
10−4
10−3
10−2
10−1
100
Γ(1
)(qT,Q
)
Q = 2.0 (GeV)
FO
|AY|Y
|W|W + Y
0 1 2 3 4 5
qT (GeV)
10−4
10−3
10−2
10−1
100
Γ(1
)(qT,Q
)
Q = 5.0 (GeV)
FO
|ASY|Y
|W|W + Y
dσdxdQdzdqT
FO =∑q
e2q
∫ 1
ξmin
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ)
+O(α2S) +O(m2/q2)
Why is this calculation relevant?
10 / 28
0.0 0.5 1.0 1.5 2.0 2.5 3.0
qT (GeV)
10−4
10−3
10−2
10−1
100
Γ(1
)(qT,Q
)
Q = 2.0 (GeV)
FO
|AY|Y
|W|W + Y
0 1 2 3 4 5
qT (GeV)
10−4
10−3
10−2
10−1
100
Γ(1
)(qT,Q
)
Q = 5.0 (GeV)
FO
|ASY|Y
|W|W + Y
dσdxdQdzdqT
FO =∑q
e2q
∫ 1
ξmin
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ)
+O(α2S) +O(m2/q2)
ξmin = q2TQ2
xz
1− z + x
Why is this calculation relevant?
10 / 28
0.0 0.5 1.0 1.5 2.0 2.5 3.0
qT (GeV)
10−4
10−3
10−2
10−1
100
Γ(1
)(qT,Q
)
Q = 2.0 (GeV)
FO
|AY|Y
|W|W + Y
0 1 2 3 4 5
qT (GeV)
10−4
10−3
10−2
10−1
100
Γ(1
)(qT,Q
)
Q = 5.0 (GeV)
FO
|ASY|Y
|W|W + Y
dσdxdQdzdqT
FO =∑q
e2q
∫ 1
ξmin
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ)
+O(α2S) +O(m2/q2)
ξmin = q2TQ2
xz
1− z + x
Does it work?
11 / 28
FO =∑q
e2q
∫ 1q2
TQ2
xz1−z
+x
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ) +O(α2S) +O(m2/q2)
data/FO vs qT
11 / 28
FO =∑q
e2q
∫ 1q2
TQ2
xz1−z
+x
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ) +O(α2S) +O(m2/q2)
data/FO vs qT
FO =∑q
e2q
∫ 1q2
TQ2
xz1−z
+x
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ) +O(α2S) +O(m2/q2)
data/FO vs qT
What to do?
12 / 28
FO =∑q
e2q
∫ 1q2
TQ2
xz1−z
+x
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ) +O(α2S) +O(m2/q2)
1) Update FFs: use qT integrated data
2) Add O(α2S)
What to do?
12 / 28
FO =∑q
e2q
∫ 1q2
TQ2
xz1−z
+x
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ) +O(α2S) +O(m2/q2)
1) Update FFs: use qT integrated data
2) Add O(α2S)
What to do?
12 / 28
FO =∑q
e2q
∫ 1q2
TQ2
xz1−z
+x
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ) +O(α2S) +O(m2/q2)
1) Update FFs: use qT integrated data
2) Add O(α2S)
13 / 28
FO =∑q
e2q
∫ 1q2
TQ2
xz1−z
+x
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ) +O(α2S) +O(m2/q2)
data/FO vs qT
FO =∑q
e2q
∫ 1q2
TQ2
xz1−z
+x
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ) +O(α2S) +O(m2/q2)
data/FO vs qT
13 / 28
FO =∑q
e2q
∫ 1q2
TQ2
xz1−z
+x
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ) +O(α2S) +O(m2/q2)
data/FO vs qT
FO =∑q
e2q
∫ 1q2
TQ2
xz1−z
+x
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ) +O(α2S) +O(m2/q2)
data/FO vs qT
13 / 28
FO =∑q
e2q
∫ 1q2
TQ2
xz1−z
+x
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ) +O(α2S) +O(m2/q2)
data/FO vs qT
FO =∑q
e2q
∫ 1q2
TQ2
xz1−z
+x
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ) +O(α2S) +O(m2/q2)
data/FO vs qT
13 / 28
FO =∑q
e2q
∫ 1q2
TQ2
xz1−z
+x
dξ
ξ − x H(ξ) fq(ξ, µ) dq(ζ(ξ), µ) +O(α2S) +O(m2/q2)
data/FO vs qT
Issue at large x remains
14 / 28
2
4
6F
NL
O1
/FL
O1
z = 0.2 qT = Q z = 0.8 qT = Q
0.01 0.1
2
4
6
z = 0.2 qT = 2Q
0.01 0.1 xz = 0.8 qT = 2Q
Q = 2 GeV
Q = 20 GeV
Large threshold corrections
The x at the minimum canindicate where to expect largethreshold corrections
14 / 28
2
4
6F
NL
O1
/FL
O1
z = 0.2 qT = Q z = 0.8 qT = Q
0.01 0.1
2
4
6
z = 0.2 qT = 2Q
0.01 0.1 xz = 0.8 qT = 2Q
Q = 2 GeV
Q = 20 GeV
COMPASS kinematics
0.01 0.1 x1
2
3
4
5
6
7
q T/Q
< z >= 0.24
0.01 0.1 x1
2
3
4
5
6
7
< z >= 0.48
0.01 0.1 x1
2
3
4
5
6
7
< z >= 0.69
x > x0
x ≤ x0
Large threshold corrections
The x at the minimum canindicate where to expect largethreshold corrections
15 / 28
New developments to identify SIDIS regions
Work based on
o Boglione, Collins, Gamberg, Gonzalez-Hernandez, Rogers, NS( PLB 766 2017 )
o Boglione, Gamberg, Gordon, Gonzalez-Hernandez, Prokudin, Rogers, NS( arXiv:1904.12882 )
SIDIS regions (Breit frame kinematics)
16 / 28
p⊥h
yh
Current fragmentationTMD factorization
Current fragmentationCollinear factorization
Soft region????
Target regionFracture functions
SIDIS current region
17 / 28
P
PB
q
ki
kf
kX
ki =(
Q
xN√
2,xN (k2
i + k2i,T)√
2Q,ki,T
)(1)
kf =(k2f + k2
f,T√2xNQ
,zNQ√
2,ki,T
)
SIDIS in the current region
18 / 28
For “current region” we must have
R1 ≡PB · kfPB · ki
→ small
Deviation from 2→ 1 kinematics
R2 ≡|(kf − q)2|
Q2 ' (1− zN) + zNq2
TQ2
Deviation from 2→ 2 kinematics
R3 ≡|k2X |Q2
q
ki
kf
SIDIS in the current region
18 / 28
For “current region” we must have
R1 ≡PB · kfPB · ki
→ small
Deviation from 2→ 1 kinematics
R2 ≡|(kf − q)2|
Q2 ' (1− zN) + zNq2
TQ2
Deviation from 2→ 2 kinematics
R3 ≡|k2X |Q2
q
ki
kf
q
ki
kf
k2
SIDIS in the current region
18 / 28
For “current region” we must have
R1 ≡PB · kfPB · ki
→ small
Deviation from 2→ 1 kinematics
R2 ≡|(kf − q)2|
Q2 ' (1− zN) + zNq2
TQ2
Deviation from 2→ 2 kinematics
R3 ≡|k2X |Q2
q
ki
kf
q
ki
kf
k2
q
p
N
k1
SIDIS regions web app
19 / 28
o https://sidis.herokuapp.com/
o feedback/questions arewelcomed
o it might take few seconds toload be patient
example for pions
20 / 28
p⊥h
yh
Current fragmentationTMD factorization
Current fragmentationCollinear factorization
Soft region????
Target regionFracture functions
R1 ≡ PB·kfPB·ki
qTQ
y
R1
example for pions
21 / 28
p⊥h
yh
Current fragmentationTMD factorization
Current fragmentationCollinear factorization
Soft region????
Target regionFracture functions
R2 ≡ |(kf−q)2|
Q2
qTQ
y
R2
example for pions
22 / 28
p⊥h
yh
Current fragmentationTMD factorization
Current fragmentationCollinear factorization
Soft region????
Target regionFracture functions
R3 ≡ |k2X |Q2
qTQ
y
R3
example for kaons
23 / 28
p⊥h
yh
Current fragmentationTMD factorization
Current fragmentationCollinear factorization
Soft region????
Target regionFracture functions
R1 ≡ PB·kfPB·ki
qTQ
y
R1
example for kaons
24 / 28
p⊥h
yh
Current fragmentationTMD factorization
Current fragmentationCollinear factorization
Soft region????
Target regionFracture functions
R2 ≡ |(kf−q)2|
Q2
qTQ
y
R2
example for kaons
25 / 28
p⊥h
yh
Current fragmentationTMD factorization
Current fragmentationCollinear factorization
Soft region????
Target regionFracture functions
R3 ≡ |k2X |Q2
qTQ
y
R3
Using the ratios in phenomenology
26 / 28
Recall the Bayesian regression paradigm
P(a|data) = L(a, data)π(a)
The likelihood
L(a,data) = exp(−1
2∑i
(datai − theoryi(a)δdatai
)2)
The priors
π(a) ∝ Πiθ(amini < ai < amax
i )
E[O] =∫dna P(a|data)O(a) ,
V[O] =∫dna P(a|data) (O(a)− E[O])2
Using the ratios in phenomenology
27 / 28
IDEA: use Ri as priors
π(Rk) ∝ exp (−|Rk|p)
The full prior becomes
π(a) ∝ Πiθ(amini < ai < amax
i )×Πj exp
− ∑k=1,2,3
|Rk(a,b,Ωj)|p× π(b)
→ parameters a enter directly in TMD factorization
→ parameters b are other parameters that characterize additionalpartonic d.o.f. (i.e. virtualities)
Summary and outlook
28 / 28
A new perspective for MC event generatorso New ML based MCEG are getting builto ETHER at theory agnostic MCEGo Gaps between theory, experiment and computing are getting narrower
Understanding the large pT SIDIS spectrumo O(α2
S) corrections are important to describe SIDIS at COMPASSo The large x region receives large threshold corrections which can
explain the difficulty to describe the datao Inclusion of SIDIS large pT data in PDFs/FFs analysis is required
New developments to identify SIDIS regionso New tools to map SIDIS regions (web-app)o The indicators can be used as Bayesian priors for the regression in
TMD phenomenology