The goal is to :

1. find cuts from simulation/embedding to apply in real data for D0->Kpi decay channel
2. first, try steep cuts, i.e cut on one dimension using significance maximization
3. second, try mutivariable : pca, LDA,TMVA

sample used

1. 5(D0+D0bar) + hijing events : we have embedded 5 D0 and 5D0bar (real pt, |y|<1) in Au+Au hijing central. The same Bfc chain as well as the reco. code have been run over the 5(D0+D0bar) and the mixed files. So files for signal only and background are available. (under the asumption that hijing+pure signal ~ hijing , i.e background. It's relatively true since the Au+Au central events has ~1000 tracks so the combinatorics is enough.
2. embedding files : D0 and D0bar were embedded in real events (%age of the real event multiplicity, flat pt). The advantage of this is that is peak can be seen before any cuts so I can estimate the signal + background under the peak, then take the backgound under the peak (using a fit function) OR take the backgound in side bands.

BACKGROUND REJECTION

I used the first sample. As I have 2 sets of files (from the pure D0/D0bar and the mix), I run a small code to fill in a tuple the same D0 variables, events by event, so that I still compare quantities relative to the same events.

pure d0/d0bar    --> file_a

mixed d0+hijing --> file_b

Then the correlation is to match events in file_a with events in file_b with their id.

--> The idea is to vary some cuts and evaluate the ratio #entries_after_cut/#entries_no cut for the pure d0 and the mixed file ; an efficient cut should have a ratio ~1 (i.e no large suppression) and the mixed file should have a ratio as low as possible (i.e important background rejection)

FIG 1.

The plot shows the inv. mass for the pure d0/d0bar (top) and mixed file(bottom) (unlike sign (left) and K^{-}pi^{+} candidates (right))

• cut : dcaTXY : dca between daughters at seconday vertex

FIG 2.

This plot shows the dcaTXY btw daughter tracks for the pure d0/d0bar (top) and mixed file(bottom) (unlike sign (left) and K^{-}pi^{+} candidates (right))

From this, I choose to vary the cut dcaTXY from 0 to 1

 

results : this plot shows the ratio #entries_after_cut/#entries_no_cut for the pure d0/d0bar and mixed files

First the invariant mass for the pure d0/d0bar and the mixed files, vs. the cut are here and here

(top left is no cut, then from left to right, top to bottom ,the cut changes from 0.005 to 1)

FIG 3.

The difference between the 2 ratios (next plot) shows clearly a maximum for dcaTXY<.02 meaning the rejection is maximal when applying the cut dcaTXY<.02 cm

FIG 4.

same plot but for candidates with 1.81 < M < 1.91 and zoomed

FIG 5.

Finally I plot the signal/background and signal/sqrt(signal+background) where

signal          = #candidates from the pure d0/d0bar with 1.81 <= M <=1.92

background = #candidates from the mixed files where (1.72<= M <=1.77) || (1.97 <=M <=2.02)

FIG 6.

Dependence with Pt : due to the statistic, 3 bins in PT are used ; it seems that the s/b is higher for candidates with high Pt

Pt <1

1 <= Pt < 2

 Pt>=2

• cut : dca3D

root [3] res()
 sTOT = 1229 bTOT = 242594
 cut : 0 sig --> 94 back --> 14877
 cut : 1 sig --> 155 back --> 29462
 cut : 2 sig --> 314 back --> 56823
 cut : 3 sig --> 442 back --> 81248
 cut : 4 sig --> 542 back --> 102110
 cut : 5 sig --> 695 back --> 119580
 cut : 6 sig --> 896 back --> 170507
 cut : 7 sig --> 1052 back --> 205417
 cut : 8 sig --> 1125 back --> 220253
 cut : 9 sig --> 1156 back --> 228581
 cut : 10 sig --> 1172 back --> 233510
 

• cut : slength/dslength (decay length significance)

 

comments :

1. the only observation is for slength/dslength > .6, the difference btw ratio is maximum

• cut  : cosine(theta*) : cosine btw PtD0 and kaon daughter in D0 frame

comments :

1. it seems to have a value around .5;.6 where the signficance reaches a plateau
2. it can be seen also from the difference btw ratios : it has a maximum for the same range

• cut : pt of daughters

 

comments :

1. clearly there is a cut at pt_daughters > .75 where the signifiance is the best : the plateau falls after pt>.75
2. it is also where the difference btw ratio has clearly a maximum

update : 4/24

2 preliminary cuts are tested to reject tracks from primary vertex since it may explain why the significance increases as the cut gets loose :

1. no cut on slength : plots are in this pdf
2. cut on slength slength> 50 microns (positive value) : plots are in this pdf
3. cut on the norm |slength| (50 microns) : plots are in this pdf
4. cut on single dca of track to PV : |dcaXY|>50 microns (to do)
5. I also changed the way of applying the cuts : before I was cutting on the upper limit only, now I'm taking ranges, e.g. dcaTXY>100 microns && dcaTXY<200 microns

comments :

• the new way of plotting results (in range instead of upper limit) has the drawback that there is not a lot of entries for large dcaTXY , so the signal may not be relevant in that case
• we also see (last plot of each pdf) that dcaXYT falls around 500-600 microns for the signal
• applying a cut on slength to remove trakcs from the PV (mainly in the mixed sample) does not really change the significance

 

I also made the distributions (slength vs pt, etc .. ) for both signal and background (was slide 2 of my last presentation)

fig 7 : top: signed decay length for signal(left) and background(right) ; bottom : same but vs. pt D0

 

fig 8 top : error of signed decay length for signal(left) and background(right) ; bottom : same but vs. pt D0

comments : rought cuts would be :

• |slength|< 500 microns
• dslength < 300 microns
• the pt dependence is difficult to see because of the low statistic

fig 9 : top : cosPointing for signal(left) and background(right) ; bottom : same but vs. pt D0

fig 10 : top : cosPointing for signal(left) and background(right) ; bottom : same but vs. pt D0 WITH a cut on slength>50 microns

comments :

• w/o a cut on slength, we can see that cosPointing has more entries for ~ cosPointing>.6
• with a cut slength>50 microns , now cosPointing is almost >0. This is expecting since it's the angle between the momentum vector and the decay vector. Then applying a positive cut makes the 3 vectors pointing in the same directions, thus cosPointing>0. But we still see that the background has some negative entries. The signal also shows entries negative entries but mostly for low pt D0

fig 11 : top : cosine(theta*) for signal(left) and background(right) ; bottom : same but vs. pt D0

 fig 11 : top : cosine(theta*) for signal(left) and background(right) ; bottom : same but vs. pt D0 WITH a cut on slength>50 microns

update : 4/25

I'm looking at another cut to preliminary remove primary track.

Here i'm using the single track dca : sqrt(dcaXY^2 + dcaZ^2)

The following plots show this distribution for kaon candidate :

• the "signal" is from pure D0 sample with a selection 1.81 < Minv < 1.89
• the "background" is from the mixed sample with a selection 1.7<Minv<1.75 U 2 < Minv< 2.04

comments : there is a clear difference in shape btw the signal curve and background curve. Basically, a cut at dca3d > .01 cm for both daughters could be used.

Also I've verified that this cut (dca3D on single track) does not destroy the pt of the track itself ,and the pt of the D0

 

--> More results when this cut increases  : here

page 3 shows that the significance has a maximum around 80-100 microns