kinematic fitting - desyintroduction: kinematic fitting as a part of data processing > raw data...
TRANSCRIPT
Kinematic fittingA powerful tool of event selection and reconstruction
Sergey
Yaschenko, DESY Zeuthen
University of Glasgow, 12.01.2011
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 2
Outline
> Introduction
> Techniques
> Applications
> Results of use at HERMES
> Summary
> Literature
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 3
Introduction: kinematic fitting as a part of data processing
> Raw data processing
> Detector calibrations
> Track search and reconstruction
> Momentum reconstruction
> Event selection and reconstruction
At this stage 3-momenta of all found tracks are reconstructed
Combine individual tracks to events
Apply certain requirements (cuts) on track correlations to select events of interest and reject the background
Alternatively use kinematic fitting
> Physics analysis
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 4
Introduction: kinematic fitting as a tool of event selection
> Kinematic fitting –
adjustment of measured kinematic parameters under certain assumptions (conditions, constraints)
> Aims
Test if assumptions are true
Improve accuracy of measurements
Check if the knowledge of measurement uncertainties is correct
Check for possible systematic uncertainties
> Has been used for more than 50 years in particle physics, considered as a standard tool of event selection and reconstruction
> But also often considered as too complicated and not really necessary
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 5
Example 1
> Decay of a particle a
to 3
particles a 1+2+3
> Energy of the primary particle is precisely known and equal to Ea = 30 GeV
> Energies of the secondary particles are equal to E1
= 5 GeV, E2
= 10 GeV
and E3
= 15 GeV
and measured with errors distributed by Gaussian with sigmas
σ1
= 1 GeV, σ2
= 1 GeV
and σ3
= 1 GeV
> Minimization of least-squares functional
> under condition
> If measurement errors are the same the result is trivial:
223
1
2 /)( ifit
iii
EE
0321 afitfitfit EEEE
ifit
i EE3/)( 321 aEEEE
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 6
Results of example 1
> Chi square distributed as chi square for one degree of freedom
> Probability distribution
> A measure of the probability that a chi square from the theoretical distribution is greater than chi square obtained from the fit
> Improvement of accuracies of energy measurements by factor of √(2/3)
2χ0 1 2 3 4 5 6 7 8 9 10
10
210
310
410
Probability0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
200
400
600
800
1000
Energy [GeV]0 5 10 15 20 25 300
1000
2000
3000
4000
5000
6000
7000
2
);(
dznzf
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 7
Background process
> Decay of a particle a
to 4
particles a 1+2+3+4
> Energy of the primary particle is precisely known and equal to Ea
=30 GeV
> Energies of the secondary particles are equal to E1
= 3 GeV, E2
= 8 GeV
and E3
= 13 GeV
and measured with errors distributed by Gaussian with sigma σ1
= 1 GeV, σ2
= 1 GeV
and σ3
= 1 GeV
> Particles 4
with energy of 6 GeV
is unmeasured (missed particle)
> Fitting assuming a 1+2+3
hypothesis
2χ0 1 2 3 4 5 6 7 8 9 10
210
310
Energy [GeV]0 5 10 15 20 25 300
1000
2000
3000
4000
5000
6000
7000
Probability0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1
10
210
310
410
510
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 8
Missing energy method vs kinematic fitting
> Process of interest: missing energy is peaked near zero, chi square is distributed as chi square for 1 degree of freedom
> Background process: missing energy peak is shifted, chi square distribution is completely different
Missing Energy [GeV]-20 -15 -10 -5 0 5 10 15 200
500
1000
1500
2000
2500
3000
3500
4000
4500
Missing Energy [GeV]-20 -15 -10 -5 0 5 10 15 200
500
1000
1500
2000
2500
3000
3500
4000
4500
2χ0 1 2 3 4 5 6 7 8 9 10
10
210
310
410
2χ0 1 2 3 4 5 6 7 8 9 10
210
310
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 9
Missing energy method vs kinematic fitting
> Apply a cut on the missing energy or to chi-square or probability distribution to select the process of interest and reject the background
Probability0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
200
400
600
800
1000
Probability0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1
10
210
310
410
510
Missing Energy [GeV]-20 -15 -10 -5 0 5 10 15 200
500
1000
1500
2000
2500
3000
3500
4000
4500
Missing Energy [GeV]-20 -15 -10 -5 0 5 10 15 200
500
1000
1500
2000
2500
3000
3500
4000
4500
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 10
Kinematic fitting technique
> Minimization of least squares functional
under constraints
… … …
yi
–
measured kinematic parameters,
ηi
–
fit parameters, σi
–
measurement errors, n –
number of kinematic parameters, m
–
number of constraints
> In case of correlations between kinematic parameters, covariance
matrix Gy
should be used
0),...,,( 212 nf
22
1
2 /)( ii
n
iiy
0),...,,( 211 nf
0),...,,( 21 nmf
n
ijjjiyi
n
ji yGy
1),(
1
2 )()(
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 11
Minimization of least-squares functions with constraints
> Method of elements
Using q
equations of constraints eliminate q
of the n kinematic parameters as functions of n-q
parameters
Not always possible, procedure is not automatic
> Method of Lagrange multiplies
Automatic procedure, widely used
Exact solution if constraints depend linearly on parameters
Simple iterative procedure if constraints are non-linear
> Method of orthogonal transformation
Mentioned in one textbook, not widely used
> Method of penalty functions
Automatic procedure
Can be effectively used with constraints of inequality type
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 12
Method of Lagrange multipliers
> Measurements give instead of true quantities ηi
values yi
> Measurement errors are normally distributed about zero with standard deviations σi
> m
equations of constraints
> If equations are linear write in the matrix form
> Define
> Minimization of Lagrange function
> Total derivative should vanish
iiiy
22 )(
,0)(
ii
i
E
E
mkfk ,...2,1,0
)(2 BcGL T
yT
00 bB
022 BddGdL T
yT
0byBc
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 13
Method of Lagrange multipliers
> Solve system of linear equations
> Obtain
> Can be easily solved
> Estimators of the measurement errors
> Estimators for fit parameters
> Using abbreviation
> Obtain by applying error propagation
01 T
y BBGc
cBBG Ty
11~
~~
y
cBBGBG Ty
Ty
111~
1111~
yBT
yy BGGBGGG
11 TyB BBGG
0
0
BcBG T
yT
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 14
Solution of example 1
> In the example 1
> Vector
> Estimation of fit parameters
> Estimation of error matrix of fit parameters
211121112
311
~G
111B
31
111
000000
111
1
23
22
21
BG
)(31~
321 aii EEEEE
aa EEEEEEEE
c
321
3
2
1
111
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 15
Example 2
> Decay of a particle a
to 3
particles a 1+2+3
> Energy of the primary particle is precisely known and equal to Ea = 30 GeV
> Energies of the secondary particles are equal to E1
= 5 GeV, E2
= 10 GeV
and E3
= 15 GeV
and measured with errors distributed by Gaussian with sigma σ1
= 1 GeV, σ2
= 1 GeV
and σ3
= 5 GeV
2χ0 1 2 3 4 5 6 7 8 9 10
10
210
310
410
Probability0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
200
400
600
800
1000
Energy [GeV]0 5 10 15 20 25 300
1000
2000
3000
4000
5000
6000
iifit
i EE )/()( 2
322
21
2321 iai EEEE
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 16
Results of example 2
> Pull distributions defined as
> Should be normally distributed around zero with standard deviation equal to unity
> Help to understand if measurement errors are known correctly
> Check for possible systematic effects
Pull 1-10 -8 -6 -4 -2 0 2 4 6 8 100
500
1000
1500
2000
2500
3000
3500
4000
/ ndf 2χ 68.64 / 81Constant 15.5± 3981 Mean 0.00317± -0.00399 Sigma 0.002± 1.001
Pull 2-10 -8 -6 -4 -2 0 2 4 6 8 100
500
1000
1500
2000
2500
3000
3500
4000
/ ndf 2χ 68.64 / 81Constant 15.5± 3981 Mean 0.00317± -0.00399 Sigma 0.002± 1.001
Pull 3-10 -8 -6 -4 -2 0 2 4 6 8 100
500
1000
1500
2000
2500
3000
3500
4000
/ ndf 2χ 75.82 / 79
Constant 16.1± 4132
Mean 0.003054± -0.004127 Sigma 0.0022± 0.9649
)()()( 22ii
ii
ii
ii
yypull
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 17
Example 3
> Decay of a particle 3
particles a 1+2+3
> Energy of the primary particle is precisely known and equal to Ea
=30 GeV
> Energies of the secondary particles are equal to E1
= 5 GeV, E2
= 10 GeV
and E3
= 15 GeV
and measured with errors distributed by Gaussian with sigma σ1
= 1 GeV, σ2
= 1 GeV
and σ3
= 5 GeV
> Measurement of error of E3
is not known and assumed to be 2 GeV
2χ0 1 2 3 4 5 6 7 8 9 10
210
310
410
Probability0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
22000
Energy [GeV]0 5 10 15 20 25 300
1000
2000
3000
4000
5000
6000
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 18
Results of example 3
> Pull distributions are wider, indication that something is wrong
with the knowledge of measurement errors
Pull 1-10 -8 -6 -4 -2 0 2 4 6 8 100
200
400
600
800
1000
1200
1400
1600
1800
2000 / ndf 2χ 181.6 / 166
Constant 7.3± 1878 Mean 0.00672± -0.00775 Sigma 0.00± 2.12
Pull 2-10 -8 -6 -4 -2 0 2 4 6 8 100
200
400
600
800
1000
1200
1400
1600
1800
2000 / ndf 2χ 181.6 / 166
Constant 7.3± 1878 Mean 0.00672± -0.00775 Sigma 0.00± 2.12
Pull 3-10 -8 -6 -4 -2 0 2 4 6 8 100
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400 / ndf 2χ 163.3 / 142
Constant 8.6± 2202
Mean 0.005731± -0.006458 Sigma 0.004± 1.809
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 19
Example 4
> Decay of a particle 3
particles a 1+2+3
> Energy of the primary particle is precisely known and equal to Ea
=30 GeV
> Energies of the secondary particles are equal to E1
= 5 GeV, E2
= 10 GeV
and E3
= 15 GeV
and measured with errors distributed by Gaussian with sigma σ1
= 1 GeV, σ2
= 1 GeV
and σ3
= 5 GeV
> Bias in the measurement of E3
: 17 GeV
instead of 15 GeV
2χ0 1 2 3 4 5 6 7 8 9 10
10
210
310
410
Probability0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
200
400
600
800
1000
1200
1400
1600
Energy [GeV]0 5 10 15 20 25 300
1000
2000
3000
4000
5000
6000
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 20
Results of example 4
> Pull distributions are shifted
Pull 1-10 -8 -6 -4 -2 0 2 4 6 8 100
500
1000
1500
2000
2500
3000
3500
4000
/ ndf 2χ 75.65 / 81Constant 15.5± 3981 Mean 0.0032± 0.3808 Sigma 0.002± 1.001
Pull 2-10 -8 -6 -4 -2 0 2 4 6 8 100
500
1000
1500
2000
2500
3000
3500
4000
/ ndf 2χ 75.65 / 81Constant 15.5± 3981 Mean 0.0032± 0.3808 Sigma 0.002± 1.001
Pull 3-10 -8 -6 -4 -2 0 2 4 6 8 100
500
1000
1500
2000
2500
3000
3500
4000
/ ndf 2χ 80.15 / 78Constant 16.1± 4134 Mean 0.0031± 0.3668 Sigma 0.0022± 0.9643
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 21
Applications of kinematic fitting
> Event selection
> Background rejection
> Improvement of resolution
> Better understanding of measurement errors, search for possible sources of systematic uncertainties
Analysis of chi square (probability) and pull distributions
> Detector calibration
If number of constraints is enough, some measurements can be considered as unknown parameters, reconstructed by kinematic fitting and then used for calibration
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 22
Possible problems
> Measurement errors are not known precisely
> Systematic uncertainties
> Measurement errors are not Gaussian
> If different particles are measured in different coordinate systems –
misalignment between these systems
> Multiple scattering and radiative
effects usually lead to tails in chi-
square and pull distributions
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 23
Outline
> Introduction
> Techniques
> Applications
> Results of use at HERMES
> Literature
> Summary
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 24
Deeply virtual Compton scattering (DVCS) at HERMES
p p’
e
e’
*γ γ
,t)ξGPDs(x,
ξx+ ξx-
t
p p’
e e’
*γ
γ
p p’
e e’
*γ
γ
> DVCS and Bethe-Heitler: the same initial and final state
> Bethe-Heitler
dominates at HERMES kinematics
> Generalized Parton Distributions (GPDs) accessible through cross section differences and azimuthal
asymmetries via interference term
Bethe-HeitlerDVCS
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 25
The HERMES experiment
1
0
2
−1
−2
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����
�����m
LUMINOSITY
CHAMBERSDRIFT
FC 1/2
TARGETCELL
DVC
MC 1−3
HODOSCOPE H0
MONITOR
BC 1/2
BC 3/4 TRD
PROP.CHAMBERS
FIELD CLAMPS
PRESHOWER (H2)
STEEL PLATE CALORIMETER
DRIFT CHAMBERS
TRIGGER HODOSCOPE H1
0 1 2 3 4 5 6 7 8 9 10
RICH270 mrad
270 mrad
MUON HODOSCOPEWIDE ANGLE
FRONTMUONHODO
MAGNET
m
IRON WALL
e+
27.5 GeV
140 mrad
170 mrad
170 mrad
140 mrad
MUON HODOSCOPES
SILICON e'
p'
Gas targets: • Longitudinally polarized H, D• Unpolarized
H, D, 4He, N, Ne, Kr, Xe• Transversely polarized H
Beam:• Longitudinally polarized e+
and e−
with both helicities• Energy 27.6 GeV
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 26
DVCS/BH measurement with the Recoil Detector
> Pre-Recoil data
Scattered lepton and photon were detected in the forward spectrometer
Recoil proton was not detected
Exclusivity achieved via missing mass technique
Associated processes (epe∆γ) were not resolved (12%
contribution in the signal)
> Recoil data
Detection of recoil proton
Suppression of the background to <1%
level
1 Tesla superconducting solenoid
Photon Detector (PD)
Scintillating Fiber Tracker
Lepton beam
Silicon Strip Detector
Target cell of unpolarized
target
)2
(GeV2xM
0 10 20 30
DIS
1000
N/N
0
0.1
0.2
0.3
)2
(GeV2xM
0 10 20 30
DIS
1000
N/N
0
0.1
0.2
0.3 data+e data-e
MC sumelastic BHassociated BHsemi-inclusive
e
p
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 27
Elastic DVCS/BH and associated background
> Elastic DVCS/BH: ep
epγ
Beam 3-momentum is known precisely (in comparison with 3-momenta of secondary particles)
Target proton is at rest
For e, p,
and γ,
3-momenta are measured with certain precision
> Background from associated process epe∆+γ
with ∆+pπ0
> Not possible to separate by missing mass only
> Low efficiency of selection if one-dimensional cuts (coplanarity, momentum-momentum correlations) are applied
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 28
Selection of elastic DVCS/BH using simple cuts
> One-dimensional cuts
Momentum-momentum correlations
Coplanarity
condition
> Efficiency of 70%
at background contamination of 5%
> background contamination of 2%
can be achieved but efficiency drops below 50%
> No improvement of resolutions
Hermes 2007 data
0
500
1000
1500
-2 0 2 4 6 8 10 12 14
Mx2 [ (GeV/c)2 ]
cou
nts Traditional DVCS analysis
(Eγ >5 GeV)
Recoil protonin acceptance
with Coplanarity cut
Coplanarity cutturned around
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 29
Kinematic fitting for elastic DVCS/BH
> Nine
kinematic parameters: Scattered electron Photon Proton
> Four
equations of constraints
13
112
111
/1)/tan()/tan(
pyppyppy
zy
zx
)sin/(1 339
38
37
pyyy
00
00
3214
3213
3212
3211
pbeam
beamzzz
yyy
xxx
meeeefppppf
pppfpppf
25
225
224
)/tan()/tan(
eyppyppy
zy
zx
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 30
Constraints as functions of parameters
> Constraints are non-linear
> Derivatives of constraints can be calculated analytically or numerically
pbeampe
beam
memyyymyf
pyyyyyyyyf
yyyyyyyyyyf
yyyyyyyyyyf
))(sin/(1/1
))tan(/(11/1//1
/)sin(1/1//
/)cos(1/1//
822
9622
34
8925
246
22
2133
9725
2465
22
21322
9725
2464
22
21311
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 31
Fitting procedure
> Minimization of chi square with constraints using penalty term
> If T is large enough constraints are satisfied automatically during the fitting procedure
> In order to equalize penalty contributions from different constraints errors of constraints are formally calculated using error propagation
> Non-linear constraints of inequality type can be included
m
kckk
n
ijjjiyi
n
ji fTyGy
0
22
1),(
1
2 /)()(
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 32
Results for elastic DVCS/BH and associated background
> For elastic DVCS/BH (Monte Carlo)
> For associated backgroundProbablility
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
210
310
2χ0 1000 2000 3000 4000 5000 6000 7000 8000 9000 100000
50
100
150
200
250
2χ0 10 20 30 40 50 60 70 80 90 1000
500
1000
1500
2000
2500
3000
2χ0 10 20 30 40 50 60 70 80 90 1000
1
2
3
4
5
Probablility0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1
10
210
310
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 33
Efficiency of elastic event selection
cut2χ0 10 20 30 40 50 60 70 80 90 100
Eff
icie
ncy
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
cut2χ0 10 20 30 40 50 60 70 80 90 100
Bac
kgro
un
d c
on
tam
inat
ion
0
0.001
0.002
0.003
0.004
0.005
0.006
> Efficiency of event selection of elastic DVCS/BH events is high at a reasonable cut on chi-square
> Contamination of associated process is well below 1%
> In addition, improvement of resolutions
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 34
Improvement of kinematic parameter reconstruction
Momentum [GeV/c]8 10 12 14 16 18 20
p/p
Δ
0
0.005
0.01
0.015
0.02
0.025
Momentum [GeV/c]8 10 12 14 16 18 20
p/p
Δ
0
0.01
0.02
0.03
0.04
0.05
0.06
Momentum [GeV/c]0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6
[ra
d]
θ Δ
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
Momentum [GeV/c]0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6
p/p
Δ
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
Momentum of electron Energy of photon
Polar angle of proton Momentum of proton
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 35
t resolution
> Mandelstam t=(p-p´)2
, important
physical
observable
> Blue squares
–
reconstruction
using
measured
kinematic
parameters
of electron
and photon
> Magenta triangles
-
reconstruction
using
measured
kinematic
parameters
of recoil
proton
> Light blue
triangles
-
reconstruction
using
measured
kinematic
parameters
of electron
and photon
(excluding
photon
energy) assuming
proton
mass
> Red squares
–
reconstruction
using
kinematic
fitting
]2-t [GeV0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
]2 t
[G
eVΔ
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 36
Pull distributions
Pull 1-5 -4 -3 -2 -1 0 1 2 3 4 50
50
100
150
200
250
300
350
400
/ ndf 2χ 206.9 / 139
Constant 3.4± 370.7
Mean 0.007389± -0.009738 Sigma 0.006± 1.007
Pull 2-5 -4 -3 -2 -1 0 1 2 3 4 50
50
100
150
200
250
300
350
400
/ ndf 2χ 244.6 / 131Constant 3.5± 372.2 Mean 0.00737± -0.01557 Sigma 0.006± 1.001
Pull 3-5 -4 -3 -2 -1 0 1 2 3 4 50
50
100
150
200
250
300
350
400
/ ndf 2χ 439.1 / 140Constant 3.5± 361 Mean 0.00776± -0.01269 Sigma 0.006± 1.021
Pull 4-5 -4 -3 -2 -1 0 1 2 3 4 50
50
100
150
200
250
300
350
400
/ ndf 2χ 202.4 / 138Constant 3.4± 367.8 Mean 0.00745± -0.01282 Sigma 0.006± 1.016
Pull 5-5 -4 -3 -2 -1 0 1 2 3 4 50
50
100
150
200
250
300
350
400
/ ndf 2χ 224.6 / 133Constant 3.5± 376.1
Mean 0.007288± -0.006971 Sigma 0.006± 0.992
Pull 6-5 -4 -3 -2 -1 0 1 2 3 4 50
50
100
150
200
250
300
350
400
/ ndf 2χ 338.9 / 137Constant 3.4± 369.7
Mean 0.007607± 0.008947 Sigma 0.005± 1.003
Pull 7-5 -4 -3 -2 -1 0 1 2 3 4 50
50
100
150
200
250
300
350
400
450 / ndf 2χ 287.1 / 135
Constant 3.5± 378.7
Mean 0.007220± -0.005882 Sigma 0.0056± 0.9816
Pull 8-5 -4 -3 -2 -1 0 1 2 3 4 50
50
100
150
200
250
300
350
400
/ ndf 2χ 330.2 / 143Constant 3.5± 369.5
Mean 0.007460± 0.003169 Sigma 0.006± 1.003
Pull 9-5 -4 -3 -2 -1 0 1 2 3 4 50
50
100
150
200
250
300
350
400
/ ndf 2χ 319.6 / 139Constant 3.6± 367.9 Mean 0.0074± -0.0108 Sigma 0.006± 1.009
Electron Photon Proton
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 37
Selection of elastic DVCS/BH events (HERMES data)
> Kinematic fitting is developed and tested on Monte-Carlo
3 particles detected 4 constraints from energy-momentum conservation
Allows to suppress the associated processes and semi-inclusive background to negligible level
> Applied for data for physics analysis
Systematic studies in progress
First physics results expected soon
> Missing mass distribution
No requirement for Recoil
Positively charged Recoil track
Kinematic fit probability > 1%
Kinematic fit probability < 1%]4/c2 [GeVX
2M-4 -2 0 2 4 6 8 10 12 140
500
1000
1500
2000
2500
3000
Data
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 38
Summary
> Kinematic fitting is a powerful tool of event selection and reconstruction
> Many applications
Improvement of resolution
Better understanding of measurement errors, search for possible sources of systematic uncertainties
Detector calibration
> Kinematic fitting technique is well developed and described in literature
> Different problems could appear in different experiments, solutions are not always straightforward
Sergey Yaschenko | Kinematic fitting | 12.10.2011 | Page 39
Literature (personal choice)
> W.E.Deming, Statistical Adjustment of Data
> S.Brandt, Data Analysis
> R.Frühwirth, M.Regler, R.K.Bock, H. Grote, D. Notz, Data Analysis Techniques for High-Energy Physics
> J.P.Berge, F.T.Solmitz, H.D.Taft, Rev. Scient. Instr. 32 N5 (1961) 538
> R.Boeck
CERN 60-30
> W.G.Davis, Nucl. Instr. Meth A 386 (1997) 498
> S.N.Dymov
et al, Nucl. Instr. Meth A 440 (2000) 431
> M.Williams, C.A.Meyer, CLAS-NOTE 2003-017
> P.Avery, http://www.phys.ufl.edu/~avery
> V.Bloebel, http://www.desy.de/~blobel/