➩use of KFParticle code instead of TCFIT for fitting

Tests are done using single D0 and some comparison with TCFIT (+ some benchmark stuffs) is done

A. preliminary  stuff

I found 2 cuts to remove wrong KFParticle variables (KFpT and KFChi2) :

• KFDecayLength and its error are sometimes wrong ( error ~ 10e4 cm) : due to low pt ; a cut KFpT>.1 is then applied and safe
• KFchi2 also has some large values : a cut at KFProb > 0.0005 removes this effect(similar as TCFIT) : plot is here

Some plots : png or pdf

• top       : from left to right : decaylength, its error, ratio decaylength/error for KFParticle
• middle : from left to right : decaylength, its error, ratio decaylength/error for TCFIT
• bottom :  from left to right : differences

comments :

• from the bottom row, we see that  :  decaylength(TCFIT) - decaylength(KFParticle) ~ 15μm

B. Benchmark stuff

Ultimately the idea will be to compare a real event (#tracks ~500-1000) and check the memory usage and processed time compared to TCFIT

• for 50kevents : memory (kBytes) vs event processed : here

       (note : most events have ~2 or 3 tracks)

C. Single Track Cuts

The idea here is to apply cut on track significance S_t = DCA / σ ,as I've done for the D⁺ and L_c ; the argument is that track from displaced vertex should have larger DCA. Track significance is a better variable than the DCA only since it takes into account the error of the track.

The 2 next plots show S_t for kaon and pion  : we see that the distribution for signal is under the background, whereas it should be above ( = broader DCA)

The reason is because of large DCA values for the background, as seen in these 2 plots (kaon, pion).

Then applying a cut at |DCA|< .1 cm (which is relatively safe as we've seen from real data) makes S_t broader for the signal, which is the expected expected : pion , kaon

Additionnal cuts :

1. |DCA|<.05 cm  : pion , kaon
2. SVT>1 : pion , kaon

comments : both above cuts improved the track significance distribution

D. Comparison (Kπ) pairs from single D0 sample with hijing files

[KFProb>.0005 is used]

The next plots , here, show :

• top : from left to right : the KFChi2 for (Kπ) pairs from single D0, from hijing, both
• top : from left to right : the KFProb for (Kπ) pairs from single D0, from hijing, both
• top : from left to right : the KFChi2/NDF for (Kπ) pairs from single D0, from hijing, both

comments :

• KFChi2 seems to be a good varaible to discriminate btw signal and hijing
• a possible cut would be KFChi2 > 2 to be safe

E. DCA to SV of daughters

cuts are :

• KFchi2 >2 (see above)
• SSd+SVT >2 

For pion, kaon

We see (also from the stats box), that the DCA of daughters to the found secondary vertex, is broader for the background, allowing then a cut on these variables.

This is the main difference with TCFIT : KFParticle returns the coordinates of the secondary vertex, not only a signed distance.

Then the distance of daughters tracks to this point is more precise.

F. proposed cuts

1. KFprob> 0.0005
2. KFchi2 > 2
3. |S_t| > 1.5
4. |DCA_XY|<.1 cm
5. DCA to SV < 500(400) μm for kaon (pion)