Gluon Polarisation Measurements @ COMPASS LIP Lisbon lsilva@lip.pt Outline: Brief Motivation High pt analysis Open Charm (LO and NLO) analyses G/G results Summary and Conclusion On behalf of the COMPASS Collaboration
The Nucleon Spin SN = ½ = ½ ΔΣ + ΔG + L Quarks Gluons Partons Angular Orbital Momenta Future! GPDs Well known! Poorly known Exploratory and discovery stage. Some experiments and data might give hints. In 1988 EMC measured = 0.12 ± 0.17 (Phys.Lett.B206,364) A recent result, including COMPASS, gives: = 0.30 ± 0.01(stat.) ± 0.02(evol.) Phys.Lett.B647,8 COMPASS, HERMES, CLAS, STAR, PHOENIX 2/15
Direct measurement of ΔG/G 2002 2004 2006 A PGF = Photon-gluon fusion process (PGF) μ beam * * g q q N N PGF PGF N PGF N PGF G/G Experiments with polarised beam and target are sensitive to gluon helicity 3/15
Direct measurement of ΔG/G To select this process there are two methods : High transverse momentum hadrons (Q2<1 and Q2>1 (GeVc)2) * g q q * Much more statistics. Physical background: strongly model dependent, requires a very good agreement between MC and Data. Open-charm meson (D mesons) Provides the purest sample of PGF events, almost free from background contamination. Small dependence on MC. Low statistics. Photon-gluon fusion process (PGF) 4/15
High pt Analysis PGF C LO ΔG LO C LO A x = x g a PGF A x a A x D LL 1 C LL 1 Bj G Tot Tot Tot 2h LL A1LO : estimated by an inclusive sample QCDCompton PGF LO DIS Final formula for the gluon polarisation ΔG av 1 x g = G β [A PGF, incl incl β=a PGF R PGF LL R PGF a LL 2h LL R LO Rincl LO x Bj A corr, incl a PGF LL ] incl C PGF incl LO R R R A2hLL : measured from the two a D R LO Acorr = A1 x Bj D A1 x C β1 A1 x C ' β 2 R LO C LL hadron sample. a ill and R i : estimated from MC and parametrised using a Neural Network. incl 5/15
MC Simulation and Neural Network Full chain of MC has been used: Generator (LEPTO) + Apparatus Simulation (GEANT) + Reconstruction Program. PDF: MSTW2008LO. High pt sample: Data MC comparison: Q2, pt and Hadron Multiplicities. MC with parton shower ON. A new tuning was performed to improve the hadron description. Data Training and Parametrisation Event By Event NN MC G/G Extraction 6/15
Results ΔG =0.125±0.060±0.063 G 0.08 x =0.09 0.04 G 2 The whole statistics was divided, for the first time, in 3 independent samples, having each one its own xg distribution. 1st point ΔG/G 0.147 ± 0.091 ± 0.088 <xg> 2 =3.4 (GeV/c) 0.07 +0.05-0.03 2nd point 0.079 ± 0.096 ± 0.082 0.10 +0.07-0.04 3rd point 0.185 ± 0.165 ± 0.143 0.17 +0.10-0.06 Within the errors the 3 points show no xg dependence 7/15
High pt Analysis, Q2 < 1 (GeV/c)2 ~90 % of our statistics in this sample 0 0.1 0.2 0.3 PGF QCD-C LO DIS direct resolv. γ qq qq qg qg gg gg 10% of statistics low pt neglect LO DIS and low pt 2002 2004 Preliminary: G/G = 0.016 ± 0.058(stat) ± 0.055(syst) 2002 2003 Published: G/G = 0.024 ± 0.089(stat) ± 0.057(syst) Phys. Lett. B 633,25 8/15
Open Charm The relation between the number of reconstructed D0 (for each target cell configuration) and G/G is given by: [ S G B bg N t = a n S B 1 f P T P a L L D A S B G S B acceptance, muon flux, number of target nucleons ], t= u, d, u', d ' Open Charm event probability Each equation is weighted with a signal weight ws = f Pm all S/(S+B) and also with a background weight wb = f Pm D B/(S+B): 8 equations with 7 unknowns: G/G, Abg + 5 independent α = (a ϕ n) factors The system is solved by a χ2 minimisation 9/15
D0 invariant mass spectra: 2002-2007 data D0 Kπ Ksubπ Kππ0 Kπππ Number of D0: N(D0) = 13000 Total = 90600 D0 Kπ π 0 6LiD = 65600 NH3 = 25000 10/15
Neural Network parametrisation Two real data samples (with the same cuts applied) are compared by a Neural Network (using some kinematic variables as a learning vector): D0Kp tagged spectrum in bins of = S/(S+B)NN Signal model gcc = K+ s + K + s+ (D0 spectrum: signal + background) Background model wcc = K+ + s + K s+ (no D0 is allowed) If the background model is good enough: The Neural Network is able to distinguish the signal from the combinatorial background on a event by event basis (inside gcc) G G 11 = = G FOM G FOM 11/15
G/G Results (LO) Preliminary 12/15
G/G Results (LO) Preliminary G = 0.08 ± 0.21 stat ± 0.08 syst G x g = 0.11 0.11 0.05, 2 = 13 GeV /c 2 13/15
NLO corrections for Open Charm analysis NLO corrections to the analysing power all NLO: bg LO: PGF virtual corrections NLO: PGF gluon bremsstrahlung corrections 14/15
G/G new Result (NLO) G = 0.20 ± 0.21 ± 0.08 syst G 2 2 @ x g = 0.28 0.19 0.10, = 13 GeV /c Preliminary: theoretical uncertainties still under study (all) 15/15
Summary and Conclusion Summary: Preliminary The importance of the gluon polarisation measurement concerning the nucleon spin structure was emphasised. The direct measurement methods were explained. The gluon polarisations results are presented. Conclusion: The ΔG seems to be small contribution. The missing contribution could be in Lpartons. All measurements of ΔG/G are compatible with zero, around Xg~0.1 COMPASS-II program foresees to measure Lpartons via GPDs. 16/15
Spares 17/15
THE COMPASS EXPERIMENT Beam: 2. 108 µ+/ spill Luminosity: ~ 5. 1032 cm-2 s-1 Beam polarisation: 80% Beam momentum: 160 GeV/c Data taken: 2002-2010,... LHC SPS COMPASS ~250 physicists 25 institutes 11 countries 18/15
The COMPASS Spectrometer Common Muon and Proton Apparatus for Structure and Spectroscopy Two staged spectrometer: Trackers Magnets RICH Electromagnetic Calorimeters LAS and SAS Polarised beam and target m 50 Hadronic Calorimeters Absorbers Target SM2 NIM A577 (2007) 455 Ge 0 16 SM1 + μ V Acceptance: 70 mrad (2002 04) 180 mrad (2006) About 350 detector planes Track reconstruction p > 0.5 GeV/c 19/15
Monte Carlo Simulation This analysis uses information from the MC, thus a strong effort and care to ensure that the MC simulation describes as good as possible the data was undertaken. Two MC samples were used in the analysis: high pt and inclusive samples. Full chain of MC has been used: Generator (LEPTO) + Apparatus Simulation (GEANT) + Reconstruction Program. PDF: MSTW2008LO. High pt sample: MC with parton shower ON has been used in the analysis. A new tuning was performed to improve the hadron description. 20/15
MC Tuning The purpose of the MC tuning is to correct the shapes of the hadron variables (momenta) and fragmentation (multiplicity). In LEPTO this can be achieved by changing JETSET parameters: PARJ(21) PARJ(23) PARJ(24) Transverse momentum of the hadron fragmentation PARJ(41) PARJ(42) Fragmentation function These parameters can be divided into two sets regarding the component of the trajectory of the particles: Transverse and longitudinal variable components. The sets can be tuned independently. The tuning improves substantially the Data-MC agreement. 21/15
Monte Carlo Simulation b = PARJ(42) a = PARJ(41) 2 b m 1 T f z 1 z a exp 2 z COMPASS new tuning LEPTO default tuning PARJ(21) PARJ(23) PARJ(24) PARJ(41) PARJ(42) 0.34 0.04 2.8 0.025 0.075 0.36 0.01 2.0 0.3 0.58 Transverse momentum of the hadron fragmentation Fragmentation function 22/15
Data Monte Carlo comparison 23/15
Data Monte Carlo comparison high pt sample: hadron variables (pt1, pt2 and pt2) 24/15
Data Monte Carlo comparison high pt sample: hadron variables (p1, p2 and multiplicity) 25/15
Weighted method A Neural Network is used to assign to each event a probability to be originated from one of the three processes (LO, PGF or Compton). MC Data training A MC sample is used to train the Neural Network (NN). A parametrisation is constructed for all variables involved in the weight. NN parametrisatio n Event by event calculatio n A Data sample is weighted on an event-by-event basis. G/ G Optimal usage of the data sample statistics 26/15
Weighted method A weight is applied on event by event basis: W = fdpbβ, where β is a factor depending on a ill and R i Therefore for every event we have to know: cl in c l in c l R P G F, R C,R L O, Rin, R P G F C, RL O, PGF C P G F, in c l C,in c l a L L,aL L,aL L,aL L, xc, xg, f, D,P b f,d,pb are directly obtained from data. The all the others variables have to be estimated/parametrised. 27/15
Example: Stability plots for NN We parametrise the R i fractions as probabilities. 28/15
Results ΔG =0.125±0.060±0.063 G 0.08 0.04 x =0.09 G 2 =3.4 (GeV/c) 29/15 2
Systematic Uncertainties Sources of Systematic Uncertainties δ(δg/g) High pt MC Simulation 0.05 Formula Simplification 0.04 False Asymmetries 0.02 A1 Parametrisation 0.02 NN Parametrisation 0.01 PB, PT, f 0,004 Open Charm 0.08 0.01 all 0.01 s/(s+b) 0.01 Total 0.06 0.08 30/15
S/(S+B): Obtaining final probabilities for a D0 candidate D0Kp tagged spectrum in bins of Σ = S/(S+B)NN Events with small S/(S+B)NN Mostly combinatorial background is selected S/(S+B) is obtained from a fit inside this bins (correcting with the NN parameterisation) Events with large S/(S+B)NN G G 11events = = G FOM G FOM Mostly Open Charm are selected 31/15
Neural Network qualification of events Two real data samples (with the same cuts applied) are compared by a Neural Network (using some kinematic variables as a learning vector): Signal model gcc = K+ s + K + s+ (D0 spectrum: signal + background) Background model wcc = K+ + s + K s+ (no D0 is allowed) If the background model is good enough: The Neural Network is able to distinguish the signal from the combinatorial background on a event by event basis (inside gcc) Example of a good learning variable 32/15
Analysing power (muon-gluon asymmetry all) all is dependent on the full knowledge of the partonic kinematics: PGF 2, al L= y, Q x g, zc, P G F Can't be experimentally obtained: οnly one charmed meson is reconstructed all is obtained from Monte-Carlo (in LO), to serve as input for a Neural Network parameterisation on some reconstructed kinematical variables: y, xbj, Q2, zd and pt Parameterised all, shows a strong correlation with the generated one AROMA) 33/15 (using
Comparison of all(lo) with all(nlo) The AROMA generator is used to simulate the fase space for the NLO (PS on) / LO (PS off) calculations of all. The resulting D0 mesons are reconstructed in the COMPASS spectrometer like real events. The respective all distributions are: 34/15