Status of dE/dx Offline Software

WANG Dayongwangdy@mail.ihep.ac.cn

Institute of High Energy PhysicsJan 10,2006

Outline dE/dx software :OO design and development

MdcDedxAlg : Reconstruction DedxCalibAlg : Calibration DedxCorrecSvc : Public service for dE/dx correction

Calibration and systematic corrections Important systematic and enviromental effects Calibration parameteriazation

Reconstruction algorithm studies: Different estimation of most prob Eloss Ionization Curve studies Resolution and residual bias correction

Summary

dE/dx :Particle ID with energy loss measurements

Principle: P = · mImplementation: C++ programming under BOSS

frameworkComponents: MdcDedxAlg, DedxCalibAlg, Dedx

CorrecSvcDesign goal: Resolution 6—7%, good seperation

MDC

tracking

dE/dx~f(v)

Particle type info

Requirements and data flow

MDC Tracking

dE/dx Reconstruction

Global Particle Identification

TransientData Store

(TDS)

MDC digits

Tracks

MDC digitsTracks

Recon dE/dx

Recon dE/dx

partId info

physics analysis Real dataflow

Apparent dataflow

Tracks

Recon dE/dx

MDC digits

。。。

AIM: to give the partID information from the list of pulse heights of hits on the MDC track, and store them into TDS

some corrections are performed to get unbiased dE/dx information.

Some proper dE/dx estimators are constructed

Overview of the software

MdcDedxReconDedxCalibAlg

Calibrationconst

CalibDataSvc

DedxCorrecSvc

EventDataSvc

Transient DataStore

Transient CalibData Store

DST

converter

converter

<<uses>>

<<uses>>

<<uses>>

<<uses>>

<<uses>>

MdcGeomSvc

<<uses>> <<uses>>

dE/dx calibration package

+i ni t i al i se()+execute()+fi nal i se()+BookHi sts()+Fi l l Hi sts()+Anal yseHi sts()+Wri teHi sts()+getChargeOff Corr()+ReadParameters()+Wri teParameters()

DedxCal i b

Al gori thm

DedxCal i bLayerGai n DedxCal i bDri ftDi st DedxCal i bSaturati on DedxCal i bZposDedxCal i bWi reGai n DedxCal i bRunByRun

DedxCal i bParameters<<uses>>

DedxCalibAlg

DedxCorrecSvc+queryI nterface()

I Interface

+setProperty()+getProperty()

IProperty

+StandardCorrec()+Wi reGai nCorrec()+Dri f tDi stCorrec()+SaturCorrec()+ZdepCorrec()+LayerGai nCorrec()+Gl obal Correc()+PathL()

-m_run-m_cal i b_fl ag-m_cal i b_const

DedxCorrecSvc

+name()+i ni t i al i ze()+fi nal i ze()

IServi ce

Servi ce

+StandardCorrec()

IDedxCorrecSvc

MdcGeomSvc

Cal i bSvc

<<uses>>

<<uses>>

Calibration data structure

double m_wireg[6860];double m_ggs[4][43]; double m_ddg[4][43]; double m_zdep[4][43];double m_layerg[43];double m_gain;double m_resol;

Sys. effects and dE/dx corrections

① Gain variations among cells

② Sampling length corrections

③ Drift distance dependence

④ Longitude position(z) dependence

⑤ Space charge effect

⑥ Charge gain non-linearity: from electronics

⑦ Corrections related to particle type

⑧ Run by run pulse height correction:Dependence on the sense wire voltage ， temperature, pressure and other environmental effects…

Parameterizations in calibration

Gas gain: Standard Landau distribution Vavilov distribution Asymmetric Gaussian distribution:

Space charge effect: general form of Q’=Q/(1-k(θ)*Q) BesII: fit with polynomial ： a=F(40°)/F(θ) Q’=Q*a CLEOII formulation: δ:longitude range of avalanche Babar formulation:

Parameterization of other effects: 3 order polynomials (presently implemented) Chebyshev series with the 1st kind of Chebyshev polynomials

)cos

(1

)cos

('

Q

Q

QQ

These parameterizations are to be tested by long-model data analysis

Comparison and choice of dE/dx curve Sternheimer(A) is better at high momentum end Va’vra(B) is relative better at low momentum end Practical global parameterization of curve is prefered

Comparison of Sternheimer and Va’vra formula:

A

B

Landau formula X P2~0 4-par fit X

BESIII Simulation Preliminary

Global 5-parameter fit for phmp_nml vs

5

44

13ln2

1p

pp

ppp

dx

dE

binning with nearly the same statisticsat each point to reduce the errorUsing garbage events in order to fastly calibrate this curve for BESIII in futureA uniform formula to avoid discrete expression for density effect The curve fit the BESII data OK

Beam-gas proton

Cosmic rays

Radiative bb

BESII data

1. In whole BesIII momentum range: 0.15—2GeV/c, good uniformity is seen with different particles and with momentum overlap;

2. Quality of curve fitting is good in the whole range

3. The fitting results is quite stable

The best dE/dx curve obtained

5

44

13ln2

1p

pp

ppp

dx

dE

BESIII Simulation Preliminary

Algorithm studies: different estimation of most probable energy

lossLandau distribution has no definite mean. The algorithm use

d must estimate the most probable energy loss Truncated mean Double truncated mean: truncate at both ends Median Geometric mean

Harmonic mean

Transformation:

Logorithm truncated mean: studies based on BESII data

idea:these methods give less bias to large values,then the satured hits have less effect to give better shape and better seperation

Different estimation of most probable energy loss: resolution

5.51% 5.34%

6.06% 5.09%

5.75% 5.44%

5.71% 2.61%

BOOST MC, MIP muon

Truncation rate: 0.7

Different estimation of most probable energy loss: seperation power

Pi/K Pi/P

0.7GeV 1.2GeV

Pi/K Pi/P 0.7GeV 1.3GeV

Pi/K Pi/P

0.7GeV 1.3GeV

Pi/K Pi/P

0.75GeV 1.3GeV

BOOST MC, MIP muon

Pi/K Pi/P

0.7GeV 1.2GeV

Pi/K Pi/P

0.7GeV 1.2GeV

Pi/K Pi/P

0.6GeV 1.1GeV

Pi/K Pi/P

0.75GeV 1.3GeV

Comparison of linear&logorithm TM

Cosmic rays Radiative Bhabha

Pull width: 1.020 0.9995 Pull width: 0.8477 0.9304

shape is more Gaussian-like shape is more Gaussian-like

Logorithm TM(right figure),compared to plain TM(left figure):

Suppress high-end residual Landau tail

The distribution more Gaussian likeBESII DATA, J/Psi hadrons

Study of truncated mean method

Well established method of dE/dx estimation

Simple and robust

Rejection of lower end hits to remove contributions from noise and background fluctuation

Truncation of higher tail to remove Landau tail due to hard collisions

Just cooresponding to ~5% lower cut

After truncation, distribution just Gaussian-like

Landau tail

BOOST MC, 1GeV electrons

Resolution curve with different truncation rates

70% truncation ratio is adopted for the algorthm

Number of good hits is required to no less than 10 for each track

Resolution from perfect MC consistent with empirical formula

BOOST MC, 1GeV electrons

Calibration of σdE/dx

35.2

0.8132.0

46.0

In

n

Empirical formula :

2

1153.0

t

A

Z

Q dependence of σdE/dx

σ /Q= p0+p1*ln （ Q ） ，p0,p1 is fitting parameters

Hits number and polar angle dependence

32.0-0.46 sin,n

σdE/dx~ polar angle relationship

0p1norm sinp

σdE/dx~ hits number relationship

0phit1norm Np

Present performance(I)

Software are robust

Basic calibration and correction,and need more

dE/dx resolution can reach design requirements: 6-7%

5.96%

χ distribution for Kaon sample Prob （ K ） distribution for Kaon sample

Present performance(II)

dxdE

mea dEdxdxdE

/

exp )/(

Distribution of is nearly a normal N(0,1)distribution

Distributions of probability function are flat

Our estimation is unbiased and can provide good partId info

dE/dx seperation for 5 particles(MC) seperation power with dE/dx

Present performance(III)

Good particle seperation in a wide range for different particles

The important π/K seperation(3 σ )can reach nearly 800MeV/c

Particle identification efficiency is more than 90% with MC samples

summary OO designed dE/dx software is developed unde

r BOSS, released and used for physics Calibration algorithms and service are develop

ed and many corrections performed to get unbiased estimation of dE/dx

Different reconstruction algorithms are explored to get best performance

Particle id is tested with MC samples, dE/dx resolution, distributions, pid efficiency is satisfactory.

To reach BESIII design goals, there are still much to understand and deal with

Thank you谢谢！

Backed -up slides…