Short biased minutes of the
4q Meeting held 19 September 00
More systematic studies on the Width Analysis |
J. Thompson |
John has done some more studies on the W width extraction. He noticed that the difference between the W width from the 4q and qqln channels is about 390 ± 261 MeV, i.e. 1.5s . This may be interesting in view of the potential large effects (~ 80 MeV) from FSI on the W Width measurement.
He has also looked at the dependence of the result with the mass window used. There is a clear tendency to reduce the differences found with HERWIG and JETSET associated to the Width (and also to the W mass) when smaller windows are used, confirming the current understanding that most of the differences are localized on the tails of the mass distribution. For instance, reducing the mass window from (60-74-92) to (74-92) the difference between JETSET and HERWIG goes from 203± 23 MeV to 88± 34 MeV for the Width measurement, and from -26± 10 MeV to -17± 10 MeV using (60-74-86) or (74-86) windows for the W mass measurement.
These studies look promising although some inconsistencies with the W width calibration curves need to be sorted out, (John claims a shift of 56 MeV w.r.t. the SM width not seen in the calibration curve).
News from BE studies and the effect on the mass |
H. Ruiz |
Hugo presented an update of the evaluation of the BE systematic error using more MC statistics (250k events) and after the last bug found corrected (BE was not applied to all pions in the event). The new evaluation is:
D Mw = -22 ± 12 MeV
Hugo then reported on his studies to reduce the BE systematic error on the W mass measurement. At the Heidelberg Meeting he had shown that due to phase space accumulation, pions with the lowest Q tend to have very low momentum. Therefore, they suffer from bad resolution and bad reconstruction efficiency. So the original idea to try to reduce BE on the W mass by cutting those events with pairs of pions with low Q does not work. That is, for the events that we expect the maximum BE effect the resolution and efficiency is bad, therefore the region of interest is not those events with lowest values of Q, but the ones in an intermediate region (0.05 < Q < 0.3 GeV). Also, for very low Q the pions are not correctly associated to the right W.
Hugo then tried an alternative approach… for each event define Qclosest computed for all pions in different W's from the combination that is closest in phase space and the two pions have the same sign. So, you have as many Qclosest in each event as pions. Then compute for each pion d p due to BE (i.e., compare the 3-momenta in the two MC samples, with and without BE). Of course, d p is due not only to BE between different W's but mainly due to the interaction with all the surrounding pions. By plotting d p vs Qclosest one can see large effects at low Qclosest (probably due to interactions between different W's) but also that the effect does not cancel for large Qclosest (indicating that the effect of BE inside the same W does not cancel). In order to eliminate the second contribution, Hugo has recomputed Qclosest using pions from different W's but with different sign this time. For large values of Qclosest the two curves converge, so if one subtract both of them the effect is mainly dominated by BE between different W's in the Qclosest region below 0.5 GeV.
Hugo has then propagated d p for each pion to the shift on the invariant mass. The plot of d m vs Qclosest shows that the main effect is in the region 0.05 < Qclosest < 0.3 GeV). The idea now is to build a discriminant variable using this distribution and remove those events with a large BE effect from the W mass reconstruction.
The use of kinematic fit errors on the Mass extraction |
R. Clifft |
Roger has studied the possibility to improve the W mass statistical error on the 4q channel by including the event-by-event kinematic fit error on the p.d.f. as a third dimension. He has found improvements on the expected error at the level of 4% (from 103 MeV to 99 MeV). This is of the same size of improvement that Aafke show using 3D fits and information from Y34 or 5C masses.
It is clear that there is little room for improvement on the statistical error. From the many ideas tried one cannot expect an improvement better than 5% on the statistical error. The systematic errors on the 4q channel amount to 90% of the stat. error, leaving a lot of room for improvement and reducing the significance of any improvement on the stat. error!!!
Effect of a different LEP and MC energy on the W mass |
F. Teubert |
Frederic presented the work done by his summer student, Per-Olof Wallin, to determine what is the effect of not having generated MC with the "right" LEP energy. Assuming that the "right" LEP beam energy is used in the kinematic fit (and rescaling), the only effect is the inability of the reweighting to reshape the MC invariant mass distribution. Using KORALW MC events generated at different c.o.m. Energies, the effect is found to be independent of the W mass and corresponding to a shift of +17 ± 3 MeV per GeV shift on the LEP c.o.m. energy.
First look at the semileptonic W mass "before" Moriond (KINFIT 1998 data) |
A.
Valassi
|
Andrea presented the status of the interface code which should run on
the common ntuples, perform the kinematic fit of the semileptonic events
by using the KINFIT package and produce the mass distributions to be fitted.
From the technical point of view the code is ready. Andrea managed
to run on the common ntuples (produced with version 1.0 of the ALPHA code)
of the 189 GeV data and MC, to produce calibration curves and to fit the
data.
Physics output is not fully satisfactory and has to be understood.
First look at the semileptonic W mass "after" Moriond (ABCFIT 1999 data) |
G.
Leibenguth
|
Guillaume presented a test of the ABCFIT interface code, which reads
the common ntuples and produces the W mass distributions, performed on
the semileptonic (e and mu) 1999 data. He compared his results with our
preliminary results submitted to Moriond.
He performed the 3D fit on the 1999 data for the first time since not
enough MC events were available at the time of Moriond. To avoid problems
with the energy points with low integrated luminosity he added the likelihood
functions of the mass distributions at each energy and fitted one W mass
per channel. The resulting W masses are compatible with the Moriond results
but to do a more reasonable comparison the 1D fit has been repeated with
the new ntuples and the new code as well. The resulting statistical errors
are worse than the Moriond results:
144 MeV (new) vs 138 MeV (Moriond) for muons and 162 MeV (new) vs 155
MeV (Moriond).
Further investigation will be performed to understand the reason of
this degradation. First attempt will be done when the new common ntuple
code (v1.1) is released.
W mass update using the tau channel |
D.
Boumediene
|
Djamel presented the status of the W mass measurement with the semileptonic
tau events.
He reminded the improvements which brought to an improvement of the
expected statistical error , estimated on the 189 GeV data, from 286 MeV
to 230 MeV:
189 GeV | 212±3 MeV |
196 GeV | 203±3 MeV |
200 GeV | 173±30 MeV |
202 GeV | 188±30 MeV |
Fragmentation studies at KINGAL level |
D.
Boumediene
|
Minutes of the GC session in Aix during the W Jamboree
Updates on the LL method (M. Haarmann for T. Barklow)
Michael presented the work done by Tim after Osaka to improve the LogLikelihood method to deal with non gaussian tails in the detector response. This was tested on the tnqq channel which have these non linearities. After some tests to optimize the correction function with MC events, Tim has found that, even with an infinite amount of MC events, the weight w(g) has an important systematic error. The problem is that the function w(g) appearing in the LL function should be built out of the small number of events that make up the particular LL function in question. For data, it is impossible and , for MC runs, such a construction would be cheating. For each Cn coming for the decomposition of the w function, the systematic error associated will be sn/sqrt(Nev). The propagation of errors from the covariance matrix has been incorporated into the analysis and will be described in a future note. Plots were shown to prove that the final pull distribution has a width of 1. The expected errors on tnqq with 192-202 GeV data are better (Dkg) or of the same order compared to the OO results. The linearity of the method to TGCs will be done and will be applied to qqqq channel. These improvments are very promising.
Prospects in single W analysis ( S. Jezequel for V. Lemaitre)
Vincent forsees the final results (including all LEP2 data) for single W channel ( cross section and TGC ) for Winter conference in 2001. It will include the search for acoplanar two jets and monojet events.
For simulation, there is no problem for TGC ( t channel only) but Reisaburo finds some problems with GRC4f for cross section measurements.The final NN for qq selection and final kinematic information optimisation are still needed. The main systematics ( E scale and trigger efficiency) will be improved.
The reweighting method is now OK but one needs a prescription to combine systematics ( No OO used as WW analysis)
Neutral TGCs from Zg events (B. Trocme)
Benjamin is presenting its first analysis on neutral TGCs using Zg events. His weights are based on U. Baur computation. To compute his limits, he uses the total cross section and the cos q of the photon. His result is slightly worse than OPAL and DELPHI and this will be studied. The L3 results using their own reweighting computation is not comparable to the other experiments. It is asked to include the nng channel for the limit computation.
Neutral TGCs for ZZ events (J. Jousset)
Jacques presents the work of David Fayolle on neutral TGCs using ZZ events. He used the measured and published ZZ cross section. He computes the variation of cross section with TGC using YFSZZ. The result is presented for each LEP energy. Due to the low ZZ cross section, the limits are much smaller than the combined results from the other LEP experiments. It was asked to cross check this result with the measured cross section and limits of the other experiments. The code to reweight ZZ events is ready and will be applied to data.