Jet Recombination Analysis
Using the ISAJET information in the optimized analysis Ntuples a study of different jet combination techniques is performed. The following are some plots for the LQ200 sample of 2000 MC events:
1) Mass of the LQ and LQ-bar reconstructed from parton banks and the cuts used for further analysis steps: Postscript or PDF.
2) Comparison of the different distributions for the identified ISR (solid) and FSR (dashed) jets. Only the events with three or more jets were used in this sample if they passed cut in item 1). Identification is done on proximity basis - if a parton jet matches a RECO jet in (f,h) then this jet is used in the analysis either as a FSR or ISR one, depending on the type of the matching parton jet. It is clear from the plots that
See plots: Postscript or PDF.
The ISR cuts are therefore as follows:
3) At this point we apply the above cuts and also require that all 4 parton jets from electrons and quarks were identified with PELC/PPHO's and RECO jets.
See plots: Postscript or PDF.
First two plots show the DR-match between the parton jets and electrons/RECO jets and the maximum mismatch allowed. The first plot in p.2 shows how the analogous plot from the previous step changed. Now, the third leading jet being a FSR is somewhat more probable than any of the first two leading jets being a FSR. This tells me that the fact some of the first two leading jets were identified as FSR is due to the gluon radiation which is more energetic than the remaining quark. Therefore, the jet type was determined to be a FSR since the FSR gluon is leading and the quark jet was not identified at all (since the quark was masked by the gluon). The next three plots show the minimum dRjj, Mjj, and KTjj for the ISR and FSR jets. The separation is not clear and it looks like the Mjj has more rejection power agains the ISR (Mjj < 100 or so requirement). It's important that all these parameters are minimized by looking only at three leading jets and ignoring the others. Finally, the last two pages show the following correlations for the FSR jets only:
It is clear that Mjj matching is almost 100% correct, i.e. the jet with the lowest Mjj with the FSR candidate is indeed the parent jet. Unfortunately the statistics at this point is really poor.
4) Meanwhile, a PYTHIA LQ200 sample was provided by Sharon:
PRJ$ROOT261:[NEW_PHENOM_6.SLQ]
NEW_SLQ20EE00_PY570_G315TM3_A_10.X_UDST01_1221;1
14269 14-APR-1997 15:18:38.00
This sample looks like it does not have FSR turned on since number of reconstructed jets peaks sharply at 2. The mass spectrum is narrower and has significantly less events on the tail (dashed vs. solid) than the ISAJET sample. The DM/M is also much narrower (first page - PYTHIA; second - ISAJET).
See plots: Postscript or PDF.
04/28/97: Absence of both FSR and ISR was confirmed by Sharon; the new file is being generated.
04/30/97: The new PYTHIA file is now done (courtesy Sharon)
FSU6_1:[TMP67.HAGOPIAN.LQ]
NEW_SLQ20EE00_PY57_G315TM3_A_20.X_UDST01_1221;1
19004 28-APR-1997 20:11:53.00
The agreement between the ISAJET and PYTHIA is much better now. The lower and upper tail looks similar, but the PYTHIA peak is shifted down by about 5 GeV - most likely due to broader jets which PYTHIA tends to produce. The conclusion I derive from thsi comparison is that an ISAJET-based analysis should be sufficient for the LQ searches.
See plots: Postscript or PDF.
5) Back to the jet recombination with the ISAJET one can look at the correlations between Mjj and ETsinq and try to distinguish between the FSR (red circles) and ISR (black triangles). The line shows a cut at Mjj < 80 + 0.4ETsinq which preserves all the FSR events while cutting out about 2/3 of the ISR events. Then this cut is applied to the tagged MC data (second figure: solid - before the cut; dashed - after) for the purpose of combining the third jet with one of the first two. The peak gets a bit sharper, indeed and the low tail moves up giving a more symmetric distribution. I have tried a couple more cuts which are less efficient for the FSR but leaves less ISR in the sample: the one proposed by Mark: Mjj < 32 + 0.82ETsinq; and two more: Mjj < 40 + ETsinq and Mjj < 20 + ETsinq. It looks like Mark's formula and the Mjj < 20 + ETsinq work the best on the clean sample. The next set of plots shows how these and a few more parametrizations work on the entire LQ MC sample. As one can see even though the peaks are visually different and become more symmetric with the cut, the count of the events inside the reasonable window stays the same within a fraction of per cent. Therefore, the conclusion is: FSR/ISR-propeties-based rcombination does not improve the results much. Similar conclusion is true for the dM/<M> peak (see the last figure). Finally, simply to beat this isuue to death, I produced the Mjj/ETsinq correlation plot for the jet 3 only: for ISR case this jet is required to be an ISR, and for FSR case either the FSR jet or its parent is required to be the jet 3. The plot looks pretty much the same as for a general ISR/FSR case, so no additonal ISR rejection can be obtained based on it.
All plots are here in the PDF format.
6) Finally, one can attempt to combine jet 1 and 2 based on the invariant mass distribution shown in the first figure for the FSR jets (red dashed histogram) and real quark jets from the LQ decay (black histogram). The cut which is fully effective for the FSR (statistics is very small though!) is Mjj < 65 GeV; it is very effective for the parent jets as well. The recombination results are shown in the next figure, and again there is no improvement since the small inefficiency for the dominant parent leading jet production kills the advantage which originates from the combination of the two leading jets which are likely to be FSR.
See plots: Postscript or PDF.
Last updated 05/01/97 by Greg Landsberg